Research ArticlePrediction of Tropical Cyclonesrsquo Characteristic Factors onHainan Island Using Data Mining Technology

Ruixu Zhou1 Wensheng Gao1 Bowen Zhang2 Xianggan Fu3 Qinzhu Chen3

Song Huang3 and Yafeng Liang3

1 Department of Electrical Engineering Tsinghua University Beijing 100084 China2 China Electric Power Research Institute Beijing 100192 China3Hainan Power Grid Corporation Haikou Hainan 570203 China

Correspondence should be addressed to Ruixu Zhou zrx13mailstsinghuaeducn

Received 18 August 2014 Revised 20 October 2014 Accepted 28 October 2014 Published 20 November 2014

Academic Editor Luis Gimeno

Copyright copy 2014 Ruixu Zhou et al This is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

A newmethodology combining datamining technology with statistical methods is proposed for the prediction of tropical cyclonesrsquocharacteristic factors which contain latitude longitude the lowest center pressure and wind speed In the proposed method thebest track datasets in the years 1949sim2012 are used for prediction Using themethod effective criterions are formed to judgewhethertropical cyclones land on Hainan Island or not The highest probability of accurate judgment can reach above 79 With regard toTCswhich are judged to land onHainan Island related prediction equations are established to effectively predict their characteristicfactors Results show that the average distance error is improved compared with the National Meteorological Centre of China

1 Introduction

Typhoon is a kind of tropical cyclones (TCs) the center-sus-tained wind speed of which arrives at level 12 to level 13(typhoon is not distinguished from TC in this paper unlessspecially emphasized) Hainan Island (108∘371015840Esim111∘051015840E18∘101015840Nsim20∘101015840N) in China is well known as ldquotyphoon corri-dorrdquo According to the historical data analysis of TCs landingon Hainan Island the yearly and the monthly statisticalresults are shown in Figures 1 and 2 respectively (Note in thispaper the condition to determine whether a typhoon lands ornot is that theminimumdistance between the typhoon centerand Hainan Island is no more than the preset influencingradius which is 300 km herein) Thus the frequency of TCslanding onHainan Island is very high Besides typhoon ranksat the top among all kinds of disasters on Hainan IslandTaking the typhoon ldquoDamreyrdquo as an example in 2005 itdestroyed 18 cities of Hainan and affected up to 6305 millionpeople among whom 21 persons were killed The direct eco-nomic loss reached 121 billion RMB [1]Therefore the timelyand accurate forecast of TCs is very important for disaster

prevention onHainan Island It can also effectively reduce thedamage and loss caused by the TC when it happens

The main methods for traditional TC forecast containstatistical methods and dynamic methods most of whichare along with complicated processes or lower precision Thestatistical methods use the historical TCsrsquo positions intensityand so on to predict TCrsquos characteristic factors such as fuzzymulticriteria decision support model [2] conditional non-linear optimal perturbation first singular vector ensembletransform Kalman filter [3] back propagation-neural net-work [4] adaptive neural network classifier using a two-layerfeature selector [5] and a support vector machine using datareduction methods [6] Dynamic methods are mainly basedon numerical forecast such as a simplified dynamical systembased on a logistic growth equation (LGE) [7] a regionalcoupled atmosphere-oceanmodel [8] the PSU-NCARMeso-scale Model version 5 [9] and the GFDL 25-km-ResolutionGlobal Atmospheric Model [10] Taking three main predic-tion centers for example the average distance error of 2448hoursrsquo forecast by theUSANationalHurricaneCenter (NHC)is 106187 km which is 125243 km for Japan Meteorological

Hindawi Publishing CorporationAdvances in MeteorologyVolume 2014 Article ID 735491 13 pageshttpdxdoiorg1011552014735491

2 Advances in Meteorology

0

2

4

6

8

9

Year

1949

1954

1959

1964

1969

1974

1979

1984

1989

1994

1999

2004

2009

2012

Num

ber o

f TCs

Number of TCs landing on Hainan Island

Figure 1 The yearly statistical result

1 2 3 4 5 6 7 8 9 10 11 120

5

10

15

20

25

30

35

Month

Num

ber o

f TCs

Number of TCs landing on Hainan Island

Tropical depressionTropical stormSevere tropical storm

TyphoonSevere typhoonSuper typhoon

Figure 2 The monthly statistical result

Agency (JMA) and 120215 km for National MeteorologicalCentre of China (NMCC) [11] Zhang et al comparedthe monitoring data from HY-2 and QuikSCATrsquos satellitescatterometers with the actual typhoon data from groundobservation The result shows that the deviations of typhoonpath and intensity are large and their standard deviationsare also very big [12] Therefore although there are manytyphoon forecast methods in use at present their precisionstill cannot meet the need for real-time typhoon warning

By using data mining technology in combination withstatistical methods a new TC forecast method based on thehistorical data is proposed in this paper Firstly the regionwhere typhoon centers were located 48 (or 72 as a compar-ative experiment) hours before landing on Hainan Island is

divided into five (or a number of 1 3 7 as a comparativeexperiment) areas using119870-means clustering algorithmThenthe TC landing criterion of each area is formed by classifi-cation and regression trees (CART) Further prediction sumof squares (PRESS) algorithm and its progressive optimalalgorithm are applied to optimize forecast factor sets Finallypart of the historical data is used to establish predictionequations by multiple linear regression model (MLRM) andthe accuracy of these equations is examined by the remaininghistorical data All results show that thismethodology ismoreaccurate compared with present existing forecast methods

2 Data and Methodology

21 Data Data used in this research is based on TCsrsquo besttrack datasets of the years 1949sim2012 in the northwesternPacificwaters (including the SouthChina Sea northern of theequator and western of 180∘E) [13] which are derived fromthe TC information center of China Meteorological Admin-istration (CMA) (httptcdatatyphoongovcn) CMA besttrack datasets contain 2172 TCs which in total have 62663observation points Every observation point may provideinformation as follows the observation time strength gradelatitude longitude the lowest center pressure (hereinafterreferred to as air pressure) 2-minute-average-near-center-maximumwind speed (hereinafter referred to aswind speed)and average wind speed in 2 minutes Because the aver-age wind speed in 2 minutes of most observation pointscannot be obtained strength grade (SG) latitude (LAT)longitude (LON) air pressure (AR) wind speed (WS) lati-tude migration velocity (LATMV) and longitude migrationvelocity (LONMV) are selected as seven predictors (here-inafter referred to as observation point information) CurrentLATMV and LONMV can be calculated using the followingmethod

Set the moments of current observation point and pre-vious two observation points as (119864

119905 119873

119905) (119864119905minus1

119873

119905minus1) and

(119864

119905minus2 119873

119905minus2) respectively (where 119864 is on behalf of longitude

and 119873 is on behalf of latitude) Then the LONMV andLATMV of the current observation time are LONMV

119905and

LATMV119905 which are calculated as

LONMV119905=

1

2

119877 sdot cosminus1 [cos2119873119905sdot cos (119864

119905minus 119864

119905minus1) + sin2119873

119905]

sdot sgn (119864119905minus 119864

119905minus1)

+ cosminus1 [cos2119873119905minus1

sdot cos (119864119905minus1

minus 119864

119905minus2)

+ sin2119873119905minus1

]

sdot sgn (119864119905minus1

minus 119864

119905minus2) times 6

minus1

(1)

LATMV119905=

1

2

119877 sdot

(119873

119905minus 119873

119905minus1) + (119873

119905minus1minus 119873

119905minus2)

6

=

119877 sdot (119873

119905minus 119873

119905minus2)

12

(2)

Advances in Meteorology 3

1085 109 1095 110 1105 111 111518

185

19

195

20

205

The fitting curve of Hainan IslandThe actual outer boundary of Hainan Island

Latit

ude (

∘N

)

Longitude (∘E)

Figure 3 The curve fitting of the outer boundary of Hainan Island

where 119877 is the mean radius of the Earth with the value of6370856 km sgn(119909) represents the sign function the unit ofLONMV

119905and LATMV

119905is kmh

22 Methodology As mentioned in the Introduction a land-ing TC is defined as a TC that theminimumdistance betweenthe typhoon center and Hainan Island is no more than thepreset influencing radius Hence in order to distinguish TCsbetween landing and not landing onHainan Island themini-mum distance between each TCrsquos track and the outer bound-ary ofHainan Islandneeds to be calculated according toCMAbest track datasets Due to the variety of TCsrsquo tracks applyinggeneral curve-fitting directly does not provide a good resultTherefore polynomial fitting [14] is applied in this paperwhere an intermediate variable is introduced to conductcurve-fitting with the latitude and longitude respectivelyTaking an arbitrary TC for example the specific fitting effectsare shown in Figures 3 and 4 Using the fitting polynomials ofeach TCrsquos and the outer boundary of Hainan Islandrsquos latitudeand longitudewith respect to the corresponding intermediatevariables the distance between any point of each TCrsquos trackand any point on the outer boundary of Hainan Island canbe calculated from which the minimum distance can beselected The great circle distance (GCD) between any twopoints on the Earth can be calculated using formula (3) TheGCD is the shortest distance between any two points on theEarth Set any two points on the Earth as 119889

1(119864

1 119873

1) and

119889

2(119864

2 119873

2) and the GCD is

1003816

1003816

1003816

1003816

119889

1119889

2

1003816

1003816

1003816

1003816

= 119877 sdot arccos [cos1198731sdot cos119873

2sdot cos (119864

1minus 119864

2)

+ sin1198731sdot sin119873

2]

(3)

The time intervals used to forecast TCs by three mainprediction centers (NHC JMA and NMCC) are 24 48 and72 hours In order to forecast TCs in a timely manner andcompare forecast accuracy among different methods here

100 110 120 130 1404

6

8

10

12

14

16

18

20

22

The actual trackThe fitting trackHainan Island

Longitude (∘E)

Latit

ude (

∘N

)

The actual valueThe fitting value

0 20 40 605

10

15

20

Intermediate variable

0 20 40 60110

120

130

140

Intermediate variable

The actual valueThe fitting value

The l

ongi

tude

of T

C ce

nter

(∘E)

The l

atitu

de o

f TC

cent

er(∘

N)

Figure 4 The curve fitting of an arbitrary TC

110 115 120 125

12

14

16

18

20

22

24

26

Hainan IslandThe landing positions

The positions where TCsrsquo centerswere located 48 hours before theylanded on Hainan Island

Longitude (∘E)

Latit

ude (

∘N

)

Figure 5 The region selected as research object

the region where TCsrsquo centers were located 48 hours beforethey landed on Hainan Island (shown in Figure 5) is selectedas research object In order to narrow the research scope 119870-means clustering algorithm [15 16] is applied to divide theregion where TCsrsquo centers were located 48 hours before theylanded on Hainan Island into five areas In this section thesituation inwhich the regionwhere TCsrsquo centers were located48 hours before they landed is selected as research object andthe research object is divided into five areas is taken as

4 Advances in Meteorology

Screening out the landing TCs and not landing TCs

Filtrating all the observation points of both landing and not landing TCs entering into each area

Forming the landing criterion of each area using CART algorithm

Obtaining the TCsrsquo records aroundHainan Island

Finding out the region where TCsrsquo centers were located48 (or 72) hours before they landed

Dividing the research region into 5 (or a number of 1 3 7) areas

Figure 6 Flow diagram of forming landing criterions

an example for a convenient statement Other situations ofa comparative experiment are also conducted in Section 33

For each of the five areas all the observation points ofboth landing and not landing TCs which entered into thisarea are filtrated With strength grade latitude longitude airpressure wind speed latitude migration velocity and longi-tude migration velocity as classification properties the TClanding criterion of each area is formed by using CART algo-rithm [17 18]The flow diagram of forming landing criterionsis shown in Figure 6

For TCs which are judged to be landing on HainanIsland PRESS and its progressive optimal algorithm andMLRM can be used to forecast TCsrsquo characteristic factors(including latitude longitude the lowest center pressureand wind speed) Forecasts in this paper contain landingprediction pattern and dynamic prediction pattern Landingprediction pattern is defined as employing the observationpoint information of those points which first enter into anyarea to predict the characteristic factors when TC landsDynamic prediction pattern is defined as 24 hoursrsquo and 48hoursrsquo prediction with respect to the observation point whichenters into any area The flow diagrams of landing forecastpattern and dynamic forecast pattern are shown in Figures 7and 8 respectively Here PRESS [19] and its progressiveoptimal algorithm [20 21] are used to select the best forecastfactor set from seven predictors which will be used toforecast corresponding characteristic factor MLRM [22] isused to establish corresponding forecast equations MLRM isexpressed as [23]

119910

119894= 120573

0+ 120573

1119909

1198941+ 120573

2119909

1198942+ sdot sdot sdot + 120573

119898119909

119894119898+ 120576

119894

(119894 = 1 2 sdot sdot sdot 119899)

(4)

where119910119894is the estimated value120573

0sim 120573

119898is the regression coef-

ficients 120576119894is the random error and 119909

1198941sim 119909

119894119898are the forecast

factors of the observation point

Table 1 The geometric centers and scopes of five areas

Area number LongitudeUnit ∘E

LatitudeUnit ∘N

RadiusUnit degree

1 1184511 152845 279022 1131435 149623 317063 1169517 208023 330484 1239859 144951 270985 1226830 185854 28346

Table 2 The number of OPs in five areas

Area numberNumber oflanding(1198851

)

Number of notlanding(1198852

)

The ratio oflanding

119885

1

(119885

1

+ 119885

2

)

1 537 1031 34252 855 1389 38103 788 1514 34234 378 1070 26105 340 1419 1933

3 Results and Discussions

In this section the situation in which the region where TCsrsquocenters were located 48 hours before they landed is selectedas research object and the research object is divided into fiveareas is firstly researched Other situations of a comparativeexperiment in which the research object may be the regionwhere TCsrsquo centers were located 72 hours before they landedand the number of areas of divided research object may beany number of 1 3 5 7 are also discussed at the end of thesection

31 Dividing the Research Region into Five Areas 119870-meansclustering algorithm is used to divide our research region intofive areas as described in Section 22 of which the geometriccenters and scopes are shown in Table 1 With respect to eacharea all the observation points of both landing and not land-ing TCs which entered into this area are filtrated The posi-tions of these observation points are shown in Figure 9 andare used to form TCsrsquo landing criterions The numbers ofthese observation points (OPs) for both landing and notlanding TCs are shown in Table 2 The division of five areasfurther narrows the research scope and makes the selectionof the observation points more pertinent so as to form theeffective landing criterions which will be illustrated furtherin Section 33

32 The Formation of Landing Criterions in Five AreasAccording to the CART algorithm the landing criterionsin five areas are shown in Figure 10 (refer to Section 21for the meaning of seven predictors) The correspondingprobability of accurate judgment (119875AJ) probability of falsealarm (119875FA) and probability of false dismissal (119875FD) are shown

Advances in Meteorology 5

Finding out the area in which the observation point first enters into

Is the TC landing on Hainan Island

Using the corresponding landing criterion

Need not forecast

Screening out all the observation points of historical landing TCs when they firstly

entered into the corresponding area

Finding out the observation information of these observation points when they landed

Using PRESS and CART to establish corresponding forecast equation for each

characteristic factor

Obtaining four characteristic factors when forecasted TC lands

on Hainan Island

Yes

No

Figure 7 Flow diagram of landing prediction pattern

Finding out the area in which the observation point first enters into

Is the TC landing onHainan Island

Using the landing criterion of

corresponding area

Need not forecast

Screening out all the observation points of

historical landing TCs when they firstly entered into the

corresponding area

Finding out the observation information of these

observation points 24 (or 48) hours later

Using PRESS and CART to establish corresponding forecast equation for each

characteristic factor

Finding out the area in which the observation point appears (it is not

necessary that the observation point enters into this area for the first time)

Screening out all the observation points of

historical landing TCs when they appeared in the corresponding area

1 Forecast model 12 Forecast model 2

Finding out the observation information of these

observation points 24 (or 48) hours later

Using PRESS and CART to establish corresponding

forecast equation for each characteristic factor

Obtaining four characteristic factors of the TC 24 (or 48)

hours later

Yes

No

12

Figure 8 Flow diagram of dynamic prediction pattern

in Table 3 Set the numbers ofOPs for landing and not landingTCs in any area to be 119885

1and 119885

2 respectively the number

of OPs which are judged to be landing according to landing

criterions when they landed truly is denoted as 1198721and the

number of OPs which are judged to be not landing accordingto landing criterions when they did not land truly is denoted

6 Advances in Meteorology

110 115 120 12510

12

14

16

18

20

22

24

26

The positions of OPs for landing and not landing TCs in area 1

OPs of not landing TCsOPs of landing TCsHainan Island

The positions when TCs landedThe boundary of area 1

OPs of not landing TCsOPs of landing TCsHainan Island

The positions when TCs landedThe boundary of area 2

OPs of not landing TCsOPs of landing TCsHainan Island

The positions when TCs landedThe boundary of area 3

OPs of not landing TCsOPs of landing TCsHainan Island

The positions when TCs landedThe boundary of area 4

OPs of not landing TCsOPs of landing TCsHainan Island

The positions when TCs landedThe boundary of area 5

110 115 120 12510

12

14

16

18

20

22

24

26

The positions of OPs for landing and not landing TCs in area 5

Longitude (∘E)

Longitude (∘E)

110 115 120 12510

12

14

16

18

20

22

24

26

The positions of OPs for landing and not landing TCs in area 2

Longitude (∘E)

Latit

ude (

∘N

)

Latit

ude (

∘N

)

Latit

ude (

∘N

)

110 115 120 12510

12

14

16

18

20

22

24

26

The positions of OPs for landing and not landing TCs in area 3

Longitude (∘E)

Latit

ude (

∘N

)

110 115 120 12510

12

14

16

18

20

22

24

26

The positions of OPs for landing and not landing TCs in area 4

Longitude (∘E)

Latit

ude (

∘N

)

Figure 9 The positions of OPs for landing and not landing TCs which entered into each area

Advances in Meteorology 7

Calculate the seven predictors

of any OP in area 1

Calculate the seven predictors

of any OP in area 2

LONMVYes No

LAT

LONMV

LONMV

LAT

LON

No

NO

Yes

Yes

No

No

No

Yes

Yes

Yes

Landing

Landing

Landing

Not landing

Not landing Not

landingNot

landing

Not landing

Not landing

Not landing

Not landing

Not landing

Not landing Not

landing

Not landing

Not landing

Not landing

Not landing

Not landing

Not landing

LATYes

LONMVLAT

WS

AP

No

Yes

Yes

Yes

No

Yes

No

No

No

Landing

Landing

Calculate the seven predictors

of any OP in area 3

LONMV

LAT

LON

LONMV

LATMV

Landing

Landing

Landing

Yes

Yes

No

No

Yes

Yes

Yes No

No

No

Calculate the seven predictors

of any OP in area 4

LONMV

LAT

LONMV

SG

AP

YesYes

Yes

Yes

Yes

No

No

No

NoNo

Landing

Landing

Landing

Calculate the seven predictors

of any OP in area 5

LONMV

LONMV

LATMV LAT

LATMV

AP

AP

No

Yes

Yes No

Yes NoYes

NoNo No

No

YesYes

Yes

Landing Landing

le14250

le116850

le minus27918

leminus5358

leminus16795le15950

leminus0892gt14450

le11

le10055

leminus6058

le20650

le116650

le22310

le2317

le minus13357

le12550

le1500

le989000

le minus9666

le minus22048

le10656

le19450le10656

le968500

le1002500

leminus21621

gt14950

Figure 10 The landing criterions in five areas

Table 3 119875AJ 119875FA and 119875FD of each criterion

Area number 119875AJ 119875FA 119875FD

1 7468 1794 39482 7531 1555 39533 7750 1176 43154 6623 3654 25935 7976 1649 3588

as 1198722 Then 119875AJ 119875FA and 119875FD of this area are calculated as

follows

119875AJ =119872

1+119872

2

119885

1+ 119885

2

119875FA =

119885

2minus119872

2

119885

2

119875FD =

119885

1minus119872

1

119885

1

(5)

Table 4 The labels of different situations

Ti Nu Label of corresponding situation48 1 FE148 3 FE348 5 FE548 7 FE772 1 ST172 3 ST372 5 ST572 7 ST7

33 Other Situations as a Comparative Experiment In Sec-tions 31 and 32 the region where TCsrsquo centers were located48 hours before they landed is selected as research object andthe research object is divided into five areas In this sectionother situations as a comparative experiment are researched

8 Advances in Meteorology

Table 5119875AJ119875FA and119875FD for each of the remaining seven situations

(a) 119875AJ 119875FA and 119875FD of each area for FE1

Area number 119875AJ 119875FA 119875FD

1 07884 00696 05279

(b) 119875AJ 119875FA and 119875FD of each area for FE3

Area number 119875AJ 119875FA 119875FD

1 07454 01139 051912 07636 01875 030543 07829 00446 07814

(c) 119875AJ 119875FA and 119875FD of each area for FE7

Area number 119875AJ 119875FA 119875FD

1 08522 00154 084142 07486 01991 039593 07302 01808 041764 07719 01165 043985 07236 00740 067356 07834 01512 029757 07631 0 1

(d) 119875AJ 119875FA and 119875FD of each area for ST1

Area number 119875AJ 119875FA 119875FD

1 08164 00412 06998

(e) 119875AJ 119875FA and 119875FD of each area for ST3

Area number 119875AJ 119875FA 119875FD

1 08320 00677 058772 07396 02062 034623 08270 0 1

(f) 119875AJ 119875FA and 119875FD of each area for ST5

Area number 119875AJ 119875FA 119875FD

1 08284 0 12 07613 0 13 07566 01590 035644 08596 0 15 08281 00678 05813

(g) 119875AJ 119875FA and 119875FD of each area for ST7

Area number 119875AJ 119875FA 119875FD

1 07874 0 12 07020 01664 047233 08799 00118 087774 08358 0 15 08537 0 16 07761 00955 050477 07407 01028 06720

and compared with each other Finally we select the situationwhich produces the best result

In order to distinguish different situations the labels ofthemare denoted inTable 4 where the parameter Ti is used to

Table 6 The Index for each of eight situations

Label of corresponding situation IndexFE1 06750FE3 06659FE5 07026FE7 06625ST1 06297ST3 06572ST5 06391ST7 06231

illustrate the research object is the region where TCsrsquo centersare located Ti hours before they landed onHainan Island Nudenotes the number of areas of divided research object

It can be seen from Table 4 that FE5 is the situation whichhas been researched in Sections 31 and 32 The remainingseven situations are researched as follows using the methodswhich are identical with FE5

The 119875AJ 119875FA and 119875FD of each area for each of the remain-ing seven situations can be calculated according to formula(5) the results of which are shown in Table 5

In order to select the best of these eight situations inTable 4 an evaluation method is introduced with which theIndex of each situation is calculated where Index is definedaccording to formula (6) For any situation 119873 denotes thenumber of areas of divided research object and119875AJ 119894119875FA 119894 and119875FD 119894 denote the 119875AJ 119875FA and 119875FD of the 119894th area respectivelyConsider the following

Index =

1

119873

119873

sum

119894=1

radic[

1

10

119875

2

AJ 119894 +3

10

(1 minus 119875FA 119894)2

+

3

5

(1 minus 119875FD 119894)2

]

(6)

It is obvious that the higher the Index is the better theresult of the landing criterions on the whole is The Indexfor each of eight situations is shown in Table 6 The situationFE5 shows the best result which also illustrates that theresearch scheme in Sections 31 and 32 is better comparedwith other situations Finally situation FE5 is selected to formthe landing criterions

34 Forecast of TCsrsquo Characteristic Factors

341 Landing Prediction Pattern The landing forecast pat-tern is defined as follows obtaining the observation pointinformation (seven predictors) when landing TCsrsquo centersfirst enter into any area which can be used to forecast thecharacteristic factors (LAT LONAP andWS)whenTCs landon Hainan Island The flow diagram is shown in Figure 6Taking area 1 for example the OPs of historical landing TCsrsquocenters when they first entered into area 1 are shown inFigure 11 and the tracks of historical landing TCs whichpassed through area 1 are shown in Figure 12

Dividing the historical landing TCs passing through eacharea into two groups with the same number one group of TCsis used to establish prediction equations and the other group

Advances in Meteorology 9

110 115 120 125

10

12

14

16

18

20

22

24

26

Hainan IslandThe boundary of area 1

The LPs of TCsThe OPs of landing TCs enteringinto area 1 for the first time

Longitude (∘E)

Latit

ude (

∘N

)

Figure 11 The OPs of historical landing TCsrsquo centers when theyfirstly entered into area 1 (LP landing position)

Hainan IslandThe boundary of area 1

The LPs of TCs

110 115 120 1258

10

12

14

16

18

20

22

24

Latit

ude (

∘N

)

Longitude (∘E)

The TCs which passedthrough area 1

Figure 12 The tracks of historical landing TCs which passedthrough area 1

is used to test the accuracy of these equations The results oftesting of these prediction equations for TCs which passedthrough each area are shown in Table 7 Making use of theactual and predicted longitude and latitude of TCsrsquo centersin combination with the formula (3) the calculated meanand standard deviation of GCD in the landing predictionpattern are shown in Figures 13 and 14 respectively Averagingthe results of five areas it can be obtained that the averageof the meanstandard deviation (SD) of GCD is 1446382978740 km In [24] Yu et al analyze the averageGCD error of48 hoursrsquo forecast in the South China Sea which is 2226 kmTherefore the landing prediction pattern proposed in this

1 2 3 4 50

20

40

60

80

100

120

140

160

Area number

The m

ean

of G

CD (k

m)

The mean of GCD for each area in landing prediction pattern

Figure 13 The mean of GCD for each area in landing predictionpattern

1 2 3 4 50

20

40

60

80

100

Area number

The s

tand

ard

devi

atio

n of

GCD

(km

)The standard deviation of GCD for each area in

landing prediction pattern

Figure 14 The standard deviation of GCD for each area in landingprediction pattern

paper shows good prediction accuracy For TCs which arejudged to be landing on Hainan Island as long as the obser-vation point information when their centers first enter intoany area are obtained the corresponding forecast equationscan be used to predict characteristic factors when they land

342 Dynamic Prediction Pattern The dynamic predictionpattern is using the current observation point informationto conduct 24 hoursrsquo and 48 hoursrsquo forecast which is alsoillustrated in Figure 8There are two different forecast modelsin dynamic prediction pattern that are described as follows

Forecast Model 1 It is to obtain the current observation pointinformation (seven predictors) when landing TCsrsquo centersenter into any area for the first time making use of which toconduct 24 hoursrsquo and 48 hoursrsquo forecast

10 Advances in Meteorology

1 2 3 4 50

50

100

150

200

Area number

The m

ean

of G

CD (k

m)

The mean of GCD for 24 hoursrsquo forecast in each areausing forecast models 1 and 2

Forecast model 1Forecast model 2

(a) The mean of GCD for 24 hoursrsquo forecast of two forecast models

Forecast model 1Forecast model 2

1 2 3 4 50

20

40

60

80

100

Area number

The s

tand

ard

devi

atio

n of

GCD

(km

)

The standard deviation of GCD for 24 hoursrsquo forecastin each area using forecast models 1 and 2

(b) The standard deviation of GCD for 24 hoursrsquo forecast of two forecastmodels

Figure 15 The meanstandard deviation of GCD for 24 hoursrsquo forecast of two forecast models

Forecast model 1Forecast model 2

1 2 3 4 50

50

100

150

200

250

300

Area number

The m

ean

of G

CD (k

m)

The mean of GCD for 48 hoursrsquo forecast in each areausing forecast models 1 and 2

(a) The mean of GCD for 48 hoursrsquo forecast of two forecast models

Forecast model 1Forecast model 2

1 2 3 4 50

50

100

150

200

Area number

The s

tand

ard

devi

atio

n of

GCD

(km

)

The standard deviation of GCD for 48 hoursrsquo forecastin each area using forecast models 1 and 2

(b) The standard deviation of GCD for 48 hoursrsquo forecast of two forecastmodels

Figure 16 The meanstandard deviation of GCD for 48 hoursrsquo forecast of two forecast models

Forecast Model 2 It is to obtain the current observation pointinformation (seven predictors) when landing TCsrsquo centers arein any area (not necessarily enter into any area for the firsttime)making use ofwhich to conduct 24 hoursrsquo and 48hoursrsquoforecast

In the process of actual prediction for TCs which arejudged to be landing on Hainan Island the observation pointinformationwhen their centers enter into any area for the first

time and the established equations in forecast model 1 areused to conduct dynamic prediction Furthermore the obser-vation point information when TCsrsquo centers are in area (it isnot necessary that TCsrsquo centers enters into this area for thefirst time) any area and the established equations in forecastmodel 2 can be used to conduct dynamic prediction

Similar to Section 341 the historical observation pointsthat meet the corresponding requirements in corresponding

Advances in Meteorology 11

Table 7 Prediction results in landing prediction pattern

Area numberThe meanSD of LAT

deviationUnit ∘N

The meanSD of LATdeviationUnit ∘E

The meanSD of APdeviationUnit hpa

The meanSD of WSdeviationUnit ms

1 0984908671 0726406404 9288967882 59540438392 0526806009 0867708973 7994665339 53036389973 0971007805 0634007711 10694183758 70915526374 1153708640 0715006162 10346361761 68151474985 1115206941 0672807103 12113885744 8040353759

Table 8 The results of testing forecast model 1 and forecast model 2

Forecast hour Area number Forecast model(1 or 2)

The meanSD ofLAT deviation

Unit ∘N

The meanSD ofLON deviation

Unit ∘E

The meanSD ofAP deviationUnit hpa

The meanSD ofWS deviationUnit ms

24 h

1 1 0812806515 1029507600 4927839733 37787288582 0700405153 0880207011 5411443187 3963729843

2 1 0913105832 0986407794 6341244247 38268278572 0753806538 1005008670 4917549673 3848730261

3 1 0755505945 1052809093 7814969316 47342417822 0772706183 0894907255 7348966512 5171843368

4 1 0658205365 0879706251 6637355333 36128264862 0640404911 0950006746 6228056617 3752125645

5 1 0689904989 0861806124 6221863820 29503258232 0699605634 1001707915 6320361131 4186732785

48 h

1 1 1219808343 1773514304 9778257246 68402402892 1259808949 1694112488 9455772439 6503044361

2 1 1225009604 2196016746 10977584764 79032593072 1285710131 2033015552 8143463813 6529343273

3 1 1240707253 2165517786 9658069204 75541529892 1325009565 1841415666 8102472309 6245253772

4 1 0916208120 1761813047 7900955968 57899453712 0989607848 1502011732 7097848208 5282637708

5 1 0965007833 1696014327 9110774706 87411640062 1131510845 1973115516 9244968041 6805748502

forecast model (1 or 2) are divided into two groups with thesame number of observation points One group of observa-tion points is used to establish prediction equations and theother group of points is used to test the accuracies of theseequations The results of testing these prediction equationsin forecast model 1 and forecast model 2 are shown inTable 8 In combinationwith formula (3) the calculatedmeanand standard deviation of GCD in two forecast models areshown in Figures 15 and 16 respectively Averaging theresults of five areas it can be obtained that the averagesof the meanstandard deviation of GCD under forecastmodel 1 and forecast model 2 for 24 hoursrsquo forecast fare1505192846156 km and 1415464812509 km respectivelyFor 48 hoursrsquo forecast the averages of the meanstandarddeviation of GCD under two different forecast modelsare 26175171456345 km and 25671091452903 km respec-tively Even though the mean of GCD in dynamic prediction

pattern is no less than three main prediction centers (NHCJMA and NMCC) it is much less than the numerical pre-diction model in [25] the means of which are 18633195 kmbased on System T106 and 16182958 km based on T213 bothfor 2448 hoursrsquo forecast Besides forecast models 1 and 2 areall more accurate than the forecast using satellite scatterome-terrsquos monitoring data in the sense of the standard deviationof GCD and the mean of weed speed error which are1496002 km and 119618ms in [12] It can be seen fromTable 8 and Figures 15 and 16 that the accuracies of forecastmodel 1 and forecast model 2 vary from different areas anddifferent characteristic factors The more accurate forecastmodel can be selected from forecast models 1 and 2 accordingto actual conditionsThe results of statistical significance testsfor each equation used to forecast corresponding charac-teristic factor in forecast model 1 and forecast model 2 areshown in Table 9 which show that 119875 value is much less than

12 Advances in Meteorology

Table 9 The results of statistical significance tests for each equation used to forecast corresponding characteristic factor in forecast model 1and forecast model 2

Forecast hour Area numberForecastmodel(1 or 2)

119877-square119875 valuefor LAT

119877-square119875 valuefor LON

119877-square119875 valuefor AP

119877-square119875 valuefor WS

24 h

1 1 05883952 times 10minus11 06011791 times 10minus9 06222446 times 10minus10 07615547 times 10minus15

2 04493226 times 10minus32 05809593 times 10minus50 07030131 times 10minus68 07731101 times 10minus81

2 1 07177186 times 10minus14 04391366 times 10minus5 07556503 times 10minus16 07995357 times 10minus18

2 06808517 times 10minus101 06753177 times 10minus99 07481213 times 10minus122 07277371 times 10minus114

3 1 06471636 times 10minus11 06559348 times 10minus11 06310183 times 10minus10 06495324 times 10minus10

2 06802646 times 10minus94 07506119 times 10minus114 05509140 times 10minus62 05573984 times 10minus65

4 1 07572105 times 10minus12 05348332 times 10minus8 07594115 times 10minus13 08174247 times 10minus15

2 06405483 times 10minus42 06266164 times 10minus40 06903101 times 10minus44 07652925 times 10minus57

5 1 05031240 times 10minus6 05154151 times 10minus6 08732273 times 10minus14 08524503 times 10minus15

2 06239284 times 10minus35 06464167 times 10minus37 07835698 times 10minus52 07550152 times 10minus48

48 h

1 1 0133700240 03943391 times 10minus5 03208177 times 10minus4 04725534 times 10minus6

2 0217700049 0039903398 0253200016 04275792 times 10minus6

2 1 04515905 times 10minus6 0266800218 06720172 times 10minus10 04622599 times 10minus6

2 04022164 times 10minus36 03095651 times 10minus26 02960505 times 10minus23 03332627 times 10minus27

3 1 0225100218 03935457 times 10minus5 0002607270 01362015962 03073618 times 10minus25 04369213 times 10minus39 01768251 times 10minus12 02479891 times 10minus18

4 1 04992960 times 10minus7 03055274 times 10minus4 04303153 times 10minus5 06249184 times 10minus9

2 03323377 times 10minus16 03270102 times 10minus16 02471170 times 10minus9 03580365 times 10minus16

5 1 04546619 times 10minus5 0255400043 03705191 times 10minus4 03535000122 04219109 times 10minus18 03825344 times 10minus17 02120218 times 10minus7 03322246 times 10minus12

005 in almost every case and prove that the correspondingprediction equation is significant

4 Summary

In this paper the CMA best track datasets from 1949 to 2012are used in combination with data mining technology andstatistical methods to put forward a new methodology toforecast TCsrsquo characteristic factors This methodology canaccurately judge whether TCs land on Hainan Island or notand forecast their characteristic factors (including longitudelatitude the lowest center pressure and wind speed) Theaverage of the probabilities of accurate judgment for landingcriterions is 7470 and the highest accuracy can reach7976 For the forecast of landing TCsrsquo characteristic factorslanding prediction pattern and dynamic prediction patternare proposed which not only can accurately forecast thecharacteristic factors when TCs land but also realize dynam-ically 24 hoursrsquo and 48 hoursrsquo forecast The effect of thelanding prediction pattern is better of which the mean ofGCD is 1446382 km compared with the current 48 hoursrsquoforecast in the South China Sea which is 2226 km Eventhough the mean of GCD in dynamic prediction pattern isno less than three main prediction centers (NHC JMA andNMCC) it is much less than the numerical prediction modelin [25] and the method using satellite scatterometerrsquos mon-itoring data in [12] The forecast methodology proposed inthis paper provides a new method for typhoon warning on

Hainan Islandwithout getting toomuch knowledge ofmeteo-rology involved and thus simplifies the implementation of theprediction process andmeanwhile guarantees the accuracy ofprediction

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

Acknowledgments

This work is in part supported by the National Science-Technology Support Plan in the domain of advanced energytechnology Hainan Power Grid Corporation provided sup-port for this research through ldquoIntegrated DemonstrationProject of Regional Smart Gridrdquo no 2013BAA01B03 DrXinhong Huang a Research Engineer in the Department ofElectrical amp Computer Engineering from the University ofWestern Ontario also puts forward valuable revision com-ments on this paper

References

[1] National Climate Center of China Typhoon ldquoDamreyrdquo HasCaused Serious Damages to Hainan Province 2005 httpncccmagovcnWebsiteindexphpNewsID=1454

Advances in Meteorology 13

[2] J S Pedro F Burstein and A Sharp ldquoA case-based fuzzy mul-ticriteria decision support model for tropical cyclone forecast-ingrdquo European Journal of Operational Research vol 160 no 2pp 308ndash324 2005

[3] X Qin and M Mu ldquoA Study on the reduction of forecast errorvariance by three adaptive observation approaches for tropicalcyclone predictionrdquoMonthlyWeather Review vol 139 no 7 pp2218ndash2232 2011

[4] Y Wang W Zhang and W Fu ldquoBack Propogation(BP)-neuralnetwork for tropical cyclone track forecastrdquo in Proceedings of the19th International Conference on Geoinformatics pp 1ndash4 IEEEShanghai China June 2011

[5] B Feng and J N K Liu ldquoAn adaptive neural network classifierfor tropical cyclone prediction using a two-layer feature selec-torrdquo in Advances in Neural NetworksmdashISNN 2005 vol 3497of Lecture Notes in Computer Science pp 399ndash404 SpringerBerlin Germany 2005

[6] H-J Song S-H Huh J-H Kim C-H Ho and S-K ParkldquoTyphoon track prediction by a support vector machine usingdata reduction methodsrdquo in Computational Intelligence andSecurity vol 3801 of Lecture Notes in Computer Science pp 503ndash511 Springer Berlin Germany 2005

[7] M DeMaria ldquoA simplified dynamical system for tropicalcyclone intensity predictionrdquoMonthly Weather Review vol 137no 1 pp 68ndash82 2009

[8] H R Winterbottom E W Uhlhorn and E P Chassignet ldquoAdesign and an application of a regional coupled atmosphere-ocean model for tropical cyclone predictionrdquo Journal ofAdvances in Modeling Earth Systems vol 4 no 10 Article IDM10002 2012

[9] L M Ma and Z M Tan ldquoImproving the behavior of the cumu-lus parameterization for tropical cyclone prediction convectiontriggerrdquo Atmospheric Research vol 92 no 2 pp 190ndash211 2009

[10] J S Gall I Ginis S-J Lin T P Marchok and J-H ChenldquoExperimental tropical cyclone prediction using the GFDL 25-km-resolution global atmospheric modelrdquo Weather and Fore-casting vol 26 no 6 pp 1008ndash1019 2011

[11] G Lv ldquoThe Northwestern Pacific typhoon track forecast in2005rdquo in Proceedings of the Meteorological Science and Technol-ogy Symposium on Taiwan Strait p 7 Chinese MeteorologicalSociety 2006 (Chinese)

[12] D Zhang Y Zhang T Hu B Xie and J Xu ldquoA comparison ofHY-2 and QuikSCAT vector wind products for tropical cyclonetrack and intensity development monitoringrdquo IEEE Geoscienceand Remote Sensing Letters vol 11 no 8 pp 1365ndash1369 2014

[13] MYingWZhangHYu et al ldquoAnoverviewof theChinamete-orological administration tropical cyclone databaserdquo Journal ofAtmospheric and Oceanic Technology vol 31 no 2 pp 287ndash3012014

[14] G E Forsythe ldquoGeneration and use of orthogonal polynomialsfor data-fittingwith a digital computerrdquo Journal of the Society forIndustrial amp Applied Mathematics vol 5 no 2 pp 74ndash88 1957

[15] KWagstaff C Cardie S Rogers and S Schroedl ldquoConstrainedk-means clustering with background knowledgerdquo in Proceed-ings of the 18th International Conference on Machine Learningvol 1 pp 577ndash584 2001

[16] Y Ye ldquoNeighborhood density method for selecting initialcluster centers in K-mean clusteringrdquo in Proceedings of theWorkshop on Data Mining for Biomedical Applications (PAKDDrsquo06) pp 189ndash198 2006

[17] P Xiong Data Mining Algorithm and Clementine PracticeTsinghua University Press Beijing China 2011 (Chinese)

[18] S L Crawford ldquoExtensions to the CART algorithmrdquo Interna-tional Journal ofMan-Machine Studies vol 31 no 2 pp 197ndash2171989

[19] D M Allen ldquoMean square error of prediction as a criterion forselecting variablesrdquo Technometrics vol 13 pp 469ndash475 1971

[20] S Yu and J Shen ldquoForward and backward algorithms forselecting predictors on the basis of the criterion frompredictionsumof squares and their applicationrdquoActaMeteorologica Sinicavol 1 pp 83ndash90 1988

[21] D Yao and S Yu ldquoThe stepwise algorithm of selecting forecastfactors based on PRESS rulerdquo Journal of Atmospheric Sciencesvol 2 pp 129ndash135 1992 (Chinese)

[22] Y LuMathematical Statistics Methods East China University ofScience and Technology Press Shanghai China 2005 (Chi-nese)

[23] K Xie and B Liu ldquoAn ENSO-forecast independent statisticalmodel for the prediction of annual Atlantic tropical cyclonefrequency in Aprilrdquo Advances in Meteorology vol 2014 ArticleID 248148 11 pages 2014

[24] J Yu J Tang Y Dai and B Yu ldquoThe error and cause analysis ofChinarsquos typhoon path predictionrdquo Journal of Weather vol 6 pp695ndash700 2012

[25] SMa A Qu and Z Yu ldquoThe parallelization of typhoon numer-ical prediction model of and track forecast error analysisrdquo Jour-nal of Applied Meteorology vol 3 pp 322ndash328 2004 (Chinese)

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ClimatologyJournal of

EcologyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

EarthquakesJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom

Applied ampEnvironmentalSoil Science

Volume 2014

Mining

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014

International Journal of

Geophysics

OceanographyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of Computational Environmental SciencesHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal ofPetroleum Engineering

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

GeochemistryHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of

Atmospheric SciencesInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

OceanographyHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

MineralogyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

MeteorologyAdvances in

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Paleontology JournalHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ScientificaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Geological ResearchJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Geology Advances in

2 Advances in Meteorology

0

2

4

6

8

9

Year

1949

1954

1959

1964

1969

1974

1979

1984

1989

1994

1999

2004

2009

2012

Num

ber o

f TCs

Number of TCs landing on Hainan Island

Figure 1 The yearly statistical result

1 2 3 4 5 6 7 8 9 10 11 120

5

10

15

20

25

30

35

Month

Num

ber o

f TCs

Number of TCs landing on Hainan Island

Tropical depressionTropical stormSevere tropical storm

TyphoonSevere typhoonSuper typhoon

Figure 2 The monthly statistical result

Agency (JMA) and 120215 km for National MeteorologicalCentre of China (NMCC) [11] Zhang et al comparedthe monitoring data from HY-2 and QuikSCATrsquos satellitescatterometers with the actual typhoon data from groundobservation The result shows that the deviations of typhoonpath and intensity are large and their standard deviationsare also very big [12] Therefore although there are manytyphoon forecast methods in use at present their precisionstill cannot meet the need for real-time typhoon warning

By using data mining technology in combination withstatistical methods a new TC forecast method based on thehistorical data is proposed in this paper Firstly the regionwhere typhoon centers were located 48 (or 72 as a compar-ative experiment) hours before landing on Hainan Island is

divided into five (or a number of 1 3 7 as a comparativeexperiment) areas using119870-means clustering algorithmThenthe TC landing criterion of each area is formed by classifi-cation and regression trees (CART) Further prediction sumof squares (PRESS) algorithm and its progressive optimalalgorithm are applied to optimize forecast factor sets Finallypart of the historical data is used to establish predictionequations by multiple linear regression model (MLRM) andthe accuracy of these equations is examined by the remaininghistorical data All results show that thismethodology ismoreaccurate compared with present existing forecast methods

2 Data and Methodology

21 Data Data used in this research is based on TCsrsquo besttrack datasets of the years 1949sim2012 in the northwesternPacificwaters (including the SouthChina Sea northern of theequator and western of 180∘E) [13] which are derived fromthe TC information center of China Meteorological Admin-istration (CMA) (httptcdatatyphoongovcn) CMA besttrack datasets contain 2172 TCs which in total have 62663observation points Every observation point may provideinformation as follows the observation time strength gradelatitude longitude the lowest center pressure (hereinafterreferred to as air pressure) 2-minute-average-near-center-maximumwind speed (hereinafter referred to aswind speed)and average wind speed in 2 minutes Because the aver-age wind speed in 2 minutes of most observation pointscannot be obtained strength grade (SG) latitude (LAT)longitude (LON) air pressure (AR) wind speed (WS) lati-tude migration velocity (LATMV) and longitude migrationvelocity (LONMV) are selected as seven predictors (here-inafter referred to as observation point information) CurrentLATMV and LONMV can be calculated using the followingmethod

Set the moments of current observation point and pre-vious two observation points as (119864

119905 119873

119905) (119864119905minus1

119873

119905minus1) and

(119864

119905minus2 119873

119905minus2) respectively (where 119864 is on behalf of longitude

and 119873 is on behalf of latitude) Then the LONMV andLATMV of the current observation time are LONMV

119905and

LATMV119905 which are calculated as

LONMV119905=

1

2

119877 sdot cosminus1 [cos2119873119905sdot cos (119864

119905minus 119864

119905minus1) + sin2119873

119905]

sdot sgn (119864119905minus 119864

119905minus1)

+ cosminus1 [cos2119873119905minus1

sdot cos (119864119905minus1

minus 119864

119905minus2)

+ sin2119873119905minus1

]

sdot sgn (119864119905minus1

minus 119864

119905minus2) times 6

minus1

(1)

LATMV119905=

1

2

119877 sdot

(119873

119905minus 119873

119905minus1) + (119873

119905minus1minus 119873

119905minus2)

6

=

119877 sdot (119873

119905minus 119873

119905minus2)

12

(2)

Advances in Meteorology 3

1085 109 1095 110 1105 111 111518

185

19

195

20

205

The fitting curve of Hainan IslandThe actual outer boundary of Hainan Island

Latit

ude (

∘N

)

Longitude (∘E)

Figure 3 The curve fitting of the outer boundary of Hainan Island

where 119877 is the mean radius of the Earth with the value of6370856 km sgn(119909) represents the sign function the unit ofLONMV

119905and LATMV

119905is kmh

22 Methodology As mentioned in the Introduction a land-ing TC is defined as a TC that theminimumdistance betweenthe typhoon center and Hainan Island is no more than thepreset influencing radius Hence in order to distinguish TCsbetween landing and not landing onHainan Island themini-mum distance between each TCrsquos track and the outer bound-ary ofHainan Islandneeds to be calculated according toCMAbest track datasets Due to the variety of TCsrsquo tracks applyinggeneral curve-fitting directly does not provide a good resultTherefore polynomial fitting [14] is applied in this paperwhere an intermediate variable is introduced to conductcurve-fitting with the latitude and longitude respectivelyTaking an arbitrary TC for example the specific fitting effectsare shown in Figures 3 and 4 Using the fitting polynomials ofeach TCrsquos and the outer boundary of Hainan Islandrsquos latitudeand longitudewith respect to the corresponding intermediatevariables the distance between any point of each TCrsquos trackand any point on the outer boundary of Hainan Island canbe calculated from which the minimum distance can beselected The great circle distance (GCD) between any twopoints on the Earth can be calculated using formula (3) TheGCD is the shortest distance between any two points on theEarth Set any two points on the Earth as 119889

1(119864

1 119873

1) and

119889

2(119864

2 119873

2) and the GCD is

1003816

1003816

1003816

1003816

119889

1119889

2

1003816

1003816

1003816

1003816

= 119877 sdot arccos [cos1198731sdot cos119873

2sdot cos (119864

1minus 119864

2)

+ sin1198731sdot sin119873

2]

(3)

The time intervals used to forecast TCs by three mainprediction centers (NHC JMA and NMCC) are 24 48 and72 hours In order to forecast TCs in a timely manner andcompare forecast accuracy among different methods here

100 110 120 130 1404

6

8

10

12

14

16

18

20

22

The actual trackThe fitting trackHainan Island

Longitude (∘E)

Latit

ude (

∘N

)

The actual valueThe fitting value

0 20 40 605

10

15

20

Intermediate variable

0 20 40 60110

120

130

140

Intermediate variable

The actual valueThe fitting value

The l

ongi

tude

of T

C ce

nter

(∘E)

The l

atitu

de o

f TC

cent

er(∘

N)

Figure 4 The curve fitting of an arbitrary TC

110 115 120 125

12

14

16

18

20

22

24

26

Hainan IslandThe landing positions

The positions where TCsrsquo centerswere located 48 hours before theylanded on Hainan Island

Longitude (∘E)

Latit

ude (

∘N

)

Figure 5 The region selected as research object

the region where TCsrsquo centers were located 48 hours beforethey landed on Hainan Island (shown in Figure 5) is selectedas research object In order to narrow the research scope 119870-means clustering algorithm [15 16] is applied to divide theregion where TCsrsquo centers were located 48 hours before theylanded on Hainan Island into five areas In this section thesituation inwhich the regionwhere TCsrsquo centers were located48 hours before they landed is selected as research object andthe research object is divided into five areas is taken as

4 Advances in Meteorology

Screening out the landing TCs and not landing TCs

Filtrating all the observation points of both landing and not landing TCs entering into each area

Forming the landing criterion of each area using CART algorithm

Obtaining the TCsrsquo records aroundHainan Island

Finding out the region where TCsrsquo centers were located48 (or 72) hours before they landed

Dividing the research region into 5 (or a number of 1 3 7) areas

Figure 6 Flow diagram of forming landing criterions

an example for a convenient statement Other situations ofa comparative experiment are also conducted in Section 33

For each of the five areas all the observation points ofboth landing and not landing TCs which entered into thisarea are filtrated With strength grade latitude longitude airpressure wind speed latitude migration velocity and longi-tude migration velocity as classification properties the TClanding criterion of each area is formed by using CART algo-rithm [17 18]The flow diagram of forming landing criterionsis shown in Figure 6

For TCs which are judged to be landing on HainanIsland PRESS and its progressive optimal algorithm andMLRM can be used to forecast TCsrsquo characteristic factors(including latitude longitude the lowest center pressureand wind speed) Forecasts in this paper contain landingprediction pattern and dynamic prediction pattern Landingprediction pattern is defined as employing the observationpoint information of those points which first enter into anyarea to predict the characteristic factors when TC landsDynamic prediction pattern is defined as 24 hoursrsquo and 48hoursrsquo prediction with respect to the observation point whichenters into any area The flow diagrams of landing forecastpattern and dynamic forecast pattern are shown in Figures 7and 8 respectively Here PRESS [19] and its progressiveoptimal algorithm [20 21] are used to select the best forecastfactor set from seven predictors which will be used toforecast corresponding characteristic factor MLRM [22] isused to establish corresponding forecast equations MLRM isexpressed as [23]

119910

119894= 120573

0+ 120573

1119909

1198941+ 120573

2119909

1198942+ sdot sdot sdot + 120573

119898119909

119894119898+ 120576

119894

(119894 = 1 2 sdot sdot sdot 119899)

(4)

where119910119894is the estimated value120573

0sim 120573

119898is the regression coef-

ficients 120576119894is the random error and 119909

1198941sim 119909

119894119898are the forecast

factors of the observation point

Table 1 The geometric centers and scopes of five areas

Area number LongitudeUnit ∘E

LatitudeUnit ∘N

RadiusUnit degree

1 1184511 152845 279022 1131435 149623 317063 1169517 208023 330484 1239859 144951 270985 1226830 185854 28346

Table 2 The number of OPs in five areas

Area numberNumber oflanding(1198851

)

Number of notlanding(1198852

)

The ratio oflanding

119885

1

(119885

1

+ 119885

2

)

1 537 1031 34252 855 1389 38103 788 1514 34234 378 1070 26105 340 1419 1933

3 Results and Discussions

In this section the situation in which the region where TCsrsquocenters were located 48 hours before they landed is selectedas research object and the research object is divided into fiveareas is firstly researched Other situations of a comparativeexperiment in which the research object may be the regionwhere TCsrsquo centers were located 72 hours before they landedand the number of areas of divided research object may beany number of 1 3 5 7 are also discussed at the end of thesection

31 Dividing the Research Region into Five Areas 119870-meansclustering algorithm is used to divide our research region intofive areas as described in Section 22 of which the geometriccenters and scopes are shown in Table 1 With respect to eacharea all the observation points of both landing and not land-ing TCs which entered into this area are filtrated The posi-tions of these observation points are shown in Figure 9 andare used to form TCsrsquo landing criterions The numbers ofthese observation points (OPs) for both landing and notlanding TCs are shown in Table 2 The division of five areasfurther narrows the research scope and makes the selectionof the observation points more pertinent so as to form theeffective landing criterions which will be illustrated furtherin Section 33

32 The Formation of Landing Criterions in Five AreasAccording to the CART algorithm the landing criterionsin five areas are shown in Figure 10 (refer to Section 21for the meaning of seven predictors) The correspondingprobability of accurate judgment (119875AJ) probability of falsealarm (119875FA) and probability of false dismissal (119875FD) are shown

Advances in Meteorology 5

Finding out the area in which the observation point first enters into

Is the TC landing on Hainan Island

Using the corresponding landing criterion

Need not forecast

Screening out all the observation points of historical landing TCs when they firstly

entered into the corresponding area

Finding out the observation information of these observation points when they landed

Using PRESS and CART to establish corresponding forecast equation for each

characteristic factor

Obtaining four characteristic factors when forecasted TC lands

on Hainan Island

Yes

No

Figure 7 Flow diagram of landing prediction pattern

Finding out the area in which the observation point first enters into

Is the TC landing onHainan Island

Using the landing criterion of

corresponding area

Need not forecast

Screening out all the observation points of

historical landing TCs when they firstly entered into the

corresponding area

Finding out the observation information of these

observation points 24 (or 48) hours later

Using PRESS and CART to establish corresponding forecast equation for each

characteristic factor

Finding out the area in which the observation point appears (it is not

necessary that the observation point enters into this area for the first time)

Screening out all the observation points of

historical landing TCs when they appeared in the corresponding area

1 Forecast model 12 Forecast model 2

Finding out the observation information of these

observation points 24 (or 48) hours later

Using PRESS and CART to establish corresponding

forecast equation for each characteristic factor

Obtaining four characteristic factors of the TC 24 (or 48)

hours later

Yes

No

12

Figure 8 Flow diagram of dynamic prediction pattern

in Table 3 Set the numbers ofOPs for landing and not landingTCs in any area to be 119885

1and 119885

2 respectively the number

of OPs which are judged to be landing according to landing

criterions when they landed truly is denoted as 1198721and the

number of OPs which are judged to be not landing accordingto landing criterions when they did not land truly is denoted

6 Advances in Meteorology

110 115 120 12510

12

14

16

18

20

22

24

26

The positions of OPs for landing and not landing TCs in area 1

OPs of not landing TCsOPs of landing TCsHainan Island

The positions when TCs landedThe boundary of area 1

OPs of not landing TCsOPs of landing TCsHainan Island

The positions when TCs landedThe boundary of area 2

OPs of not landing TCsOPs of landing TCsHainan Island

The positions when TCs landedThe boundary of area 3

OPs of not landing TCsOPs of landing TCsHainan Island

The positions when TCs landedThe boundary of area 4

OPs of not landing TCsOPs of landing TCsHainan Island

The positions when TCs landedThe boundary of area 5

110 115 120 12510

12

14

16

18

20

22

24

26

The positions of OPs for landing and not landing TCs in area 5

Longitude (∘E)

Longitude (∘E)

110 115 120 12510

12

14

16

18

20

22

24

26

The positions of OPs for landing and not landing TCs in area 2

Longitude (∘E)

Latit

ude (

∘N

)

Latit

ude (

∘N

)

Latit

ude (

∘N

)

110 115 120 12510

12

14

16

18

20

22

24

26

The positions of OPs for landing and not landing TCs in area 3

Longitude (∘E)

Latit

ude (

∘N

)

110 115 120 12510

12

14

16

18

20

22

24

26

The positions of OPs for landing and not landing TCs in area 4

Longitude (∘E)

Latit

ude (

∘N

)

Figure 9 The positions of OPs for landing and not landing TCs which entered into each area

Advances in Meteorology 7

Calculate the seven predictors

of any OP in area 1

Calculate the seven predictors

of any OP in area 2

LONMVYes No

LAT

LONMV

LONMV

LAT

LON

No

NO

Yes

Yes

No

No

No

Yes

Yes

Yes

Landing

Landing

Landing

Not landing

Not landing Not

landingNot

landing

Not landing

Not landing

Not landing

Not landing

Not landing

Not landing Not

landing

Not landing

Not landing

Not landing

Not landing

Not landing

Not landing

LATYes

LONMVLAT

WS

AP

No

Yes

Yes

Yes

No

Yes

No

No

No

Landing

Landing

Calculate the seven predictors

of any OP in area 3

LONMV

LAT

LON

LONMV

LATMV

Landing

Landing

Landing

Yes

Yes

No

No

Yes

Yes

Yes No

No

No

Calculate the seven predictors

of any OP in area 4

LONMV

LAT

LONMV

SG

AP

YesYes

Yes

Yes

Yes

No

No

No

NoNo

Landing

Landing

Landing

Calculate the seven predictors

of any OP in area 5

LONMV

LONMV

LATMV LAT

LATMV

AP

AP

No

Yes

Yes No

Yes NoYes

NoNo No

No

YesYes

Yes

Landing Landing

le14250

le116850

le minus27918

leminus5358

leminus16795le15950

leminus0892gt14450

le11

le10055

leminus6058

le20650

le116650

le22310

le2317

le minus13357

le12550

le1500

le989000

le minus9666

le minus22048

le10656

le19450le10656

le968500

le1002500

leminus21621

gt14950

Figure 10 The landing criterions in five areas

Table 3 119875AJ 119875FA and 119875FD of each criterion

Area number 119875AJ 119875FA 119875FD

1 7468 1794 39482 7531 1555 39533 7750 1176 43154 6623 3654 25935 7976 1649 3588

as 1198722 Then 119875AJ 119875FA and 119875FD of this area are calculated as

follows

119875AJ =119872

1+119872

2

119885

1+ 119885

2

119875FA =

119885

2minus119872

2

119885

2

119875FD =

119885

1minus119872

1

119885

1

(5)

Table 4 The labels of different situations

Ti Nu Label of corresponding situation48 1 FE148 3 FE348 5 FE548 7 FE772 1 ST172 3 ST372 5 ST572 7 ST7

33 Other Situations as a Comparative Experiment In Sec-tions 31 and 32 the region where TCsrsquo centers were located48 hours before they landed is selected as research object andthe research object is divided into five areas In this sectionother situations as a comparative experiment are researched

8 Advances in Meteorology

Table 5119875AJ119875FA and119875FD for each of the remaining seven situations

(a) 119875AJ 119875FA and 119875FD of each area for FE1

Area number 119875AJ 119875FA 119875FD

1 07884 00696 05279

(b) 119875AJ 119875FA and 119875FD of each area for FE3

Area number 119875AJ 119875FA 119875FD

1 07454 01139 051912 07636 01875 030543 07829 00446 07814

(c) 119875AJ 119875FA and 119875FD of each area for FE7

Area number 119875AJ 119875FA 119875FD

1 08522 00154 084142 07486 01991 039593 07302 01808 041764 07719 01165 043985 07236 00740 067356 07834 01512 029757 07631 0 1

(d) 119875AJ 119875FA and 119875FD of each area for ST1

Area number 119875AJ 119875FA 119875FD

1 08164 00412 06998

(e) 119875AJ 119875FA and 119875FD of each area for ST3

Area number 119875AJ 119875FA 119875FD

1 08320 00677 058772 07396 02062 034623 08270 0 1

(f) 119875AJ 119875FA and 119875FD of each area for ST5

Area number 119875AJ 119875FA 119875FD

1 08284 0 12 07613 0 13 07566 01590 035644 08596 0 15 08281 00678 05813

(g) 119875AJ 119875FA and 119875FD of each area for ST7

Area number 119875AJ 119875FA 119875FD

1 07874 0 12 07020 01664 047233 08799 00118 087774 08358 0 15 08537 0 16 07761 00955 050477 07407 01028 06720

and compared with each other Finally we select the situationwhich produces the best result

In order to distinguish different situations the labels ofthemare denoted inTable 4 where the parameter Ti is used to

Table 6 The Index for each of eight situations

Label of corresponding situation IndexFE1 06750FE3 06659FE5 07026FE7 06625ST1 06297ST3 06572ST5 06391ST7 06231

illustrate the research object is the region where TCsrsquo centersare located Ti hours before they landed onHainan Island Nudenotes the number of areas of divided research object

It can be seen from Table 4 that FE5 is the situation whichhas been researched in Sections 31 and 32 The remainingseven situations are researched as follows using the methodswhich are identical with FE5

The 119875AJ 119875FA and 119875FD of each area for each of the remain-ing seven situations can be calculated according to formula(5) the results of which are shown in Table 5

In order to select the best of these eight situations inTable 4 an evaluation method is introduced with which theIndex of each situation is calculated where Index is definedaccording to formula (6) For any situation 119873 denotes thenumber of areas of divided research object and119875AJ 119894119875FA 119894 and119875FD 119894 denote the 119875AJ 119875FA and 119875FD of the 119894th area respectivelyConsider the following

Index =

1

119873

119873

sum

119894=1

radic[

1

10

119875

2

AJ 119894 +3

10

(1 minus 119875FA 119894)2

+

3

5

(1 minus 119875FD 119894)2

]

(6)

It is obvious that the higher the Index is the better theresult of the landing criterions on the whole is The Indexfor each of eight situations is shown in Table 6 The situationFE5 shows the best result which also illustrates that theresearch scheme in Sections 31 and 32 is better comparedwith other situations Finally situation FE5 is selected to formthe landing criterions

34 Forecast of TCsrsquo Characteristic Factors

341 Landing Prediction Pattern The landing forecast pat-tern is defined as follows obtaining the observation pointinformation (seven predictors) when landing TCsrsquo centersfirst enter into any area which can be used to forecast thecharacteristic factors (LAT LONAP andWS)whenTCs landon Hainan Island The flow diagram is shown in Figure 6Taking area 1 for example the OPs of historical landing TCsrsquocenters when they first entered into area 1 are shown inFigure 11 and the tracks of historical landing TCs whichpassed through area 1 are shown in Figure 12

Dividing the historical landing TCs passing through eacharea into two groups with the same number one group of TCsis used to establish prediction equations and the other group

Advances in Meteorology 9

110 115 120 125

10

12

14

16

18

20

22

24

26

Hainan IslandThe boundary of area 1

The LPs of TCsThe OPs of landing TCs enteringinto area 1 for the first time

Longitude (∘E)

Latit

ude (

∘N

)

Figure 11 The OPs of historical landing TCsrsquo centers when theyfirstly entered into area 1 (LP landing position)

Hainan IslandThe boundary of area 1

The LPs of TCs

110 115 120 1258

10

12

14

16

18

20

22

24

Latit

ude (

∘N

)

Longitude (∘E)

The TCs which passedthrough area 1

Figure 12 The tracks of historical landing TCs which passedthrough area 1

is used to test the accuracy of these equations The results oftesting of these prediction equations for TCs which passedthrough each area are shown in Table 7 Making use of theactual and predicted longitude and latitude of TCsrsquo centersin combination with the formula (3) the calculated meanand standard deviation of GCD in the landing predictionpattern are shown in Figures 13 and 14 respectively Averagingthe results of five areas it can be obtained that the averageof the meanstandard deviation (SD) of GCD is 1446382978740 km In [24] Yu et al analyze the averageGCD error of48 hoursrsquo forecast in the South China Sea which is 2226 kmTherefore the landing prediction pattern proposed in this

1 2 3 4 50

20

40

60

80

100

120

140

160

Area number

The m

ean

of G

CD (k

m)

The mean of GCD for each area in landing prediction pattern

Figure 13 The mean of GCD for each area in landing predictionpattern

1 2 3 4 50

20

40

60

80

100

Area number

The s

tand

ard

devi

atio

n of

GCD

(km

)The standard deviation of GCD for each area in

landing prediction pattern

Figure 14 The standard deviation of GCD for each area in landingprediction pattern

paper shows good prediction accuracy For TCs which arejudged to be landing on Hainan Island as long as the obser-vation point information when their centers first enter intoany area are obtained the corresponding forecast equationscan be used to predict characteristic factors when they land

342 Dynamic Prediction Pattern The dynamic predictionpattern is using the current observation point informationto conduct 24 hoursrsquo and 48 hoursrsquo forecast which is alsoillustrated in Figure 8There are two different forecast modelsin dynamic prediction pattern that are described as follows

Forecast Model 1 It is to obtain the current observation pointinformation (seven predictors) when landing TCsrsquo centersenter into any area for the first time making use of which toconduct 24 hoursrsquo and 48 hoursrsquo forecast

10 Advances in Meteorology

1 2 3 4 50

50

100

150

200

Area number

The m

ean

of G

CD (k

m)

The mean of GCD for 24 hoursrsquo forecast in each areausing forecast models 1 and 2

Forecast model 1Forecast model 2

(a) The mean of GCD for 24 hoursrsquo forecast of two forecast models

Forecast model 1Forecast model 2

1 2 3 4 50

20

40

60

80

100

Area number

The s

tand

ard

devi

atio

n of

GCD

(km

)

The standard deviation of GCD for 24 hoursrsquo forecastin each area using forecast models 1 and 2

(b) The standard deviation of GCD for 24 hoursrsquo forecast of two forecastmodels

Figure 15 The meanstandard deviation of GCD for 24 hoursrsquo forecast of two forecast models

Forecast model 1Forecast model 2

1 2 3 4 50

50

100

150

200

250

300

Area number

The m

ean

of G

CD (k

m)

The mean of GCD for 48 hoursrsquo forecast in each areausing forecast models 1 and 2

(a) The mean of GCD for 48 hoursrsquo forecast of two forecast models

Forecast model 1Forecast model 2

1 2 3 4 50

50

100

150

200

Area number

The s

tand

ard

devi

atio

n of

GCD

(km

)

The standard deviation of GCD for 48 hoursrsquo forecastin each area using forecast models 1 and 2

(b) The standard deviation of GCD for 48 hoursrsquo forecast of two forecastmodels

Figure 16 The meanstandard deviation of GCD for 48 hoursrsquo forecast of two forecast models

Forecast Model 2 It is to obtain the current observation pointinformation (seven predictors) when landing TCsrsquo centers arein any area (not necessarily enter into any area for the firsttime)making use ofwhich to conduct 24 hoursrsquo and 48hoursrsquoforecast

In the process of actual prediction for TCs which arejudged to be landing on Hainan Island the observation pointinformationwhen their centers enter into any area for the first

time and the established equations in forecast model 1 areused to conduct dynamic prediction Furthermore the obser-vation point information when TCsrsquo centers are in area (it isnot necessary that TCsrsquo centers enters into this area for thefirst time) any area and the established equations in forecastmodel 2 can be used to conduct dynamic prediction

Similar to Section 341 the historical observation pointsthat meet the corresponding requirements in corresponding

Advances in Meteorology 11

Table 7 Prediction results in landing prediction pattern

Area numberThe meanSD of LAT

deviationUnit ∘N

The meanSD of LATdeviationUnit ∘E

The meanSD of APdeviationUnit hpa

The meanSD of WSdeviationUnit ms

1 0984908671 0726406404 9288967882 59540438392 0526806009 0867708973 7994665339 53036389973 0971007805 0634007711 10694183758 70915526374 1153708640 0715006162 10346361761 68151474985 1115206941 0672807103 12113885744 8040353759

Table 8 The results of testing forecast model 1 and forecast model 2

Forecast hour Area number Forecast model(1 or 2)

The meanSD ofLAT deviation

Unit ∘N

The meanSD ofLON deviation

Unit ∘E

The meanSD ofAP deviationUnit hpa

The meanSD ofWS deviationUnit ms

24 h

1 1 0812806515 1029507600 4927839733 37787288582 0700405153 0880207011 5411443187 3963729843

2 1 0913105832 0986407794 6341244247 38268278572 0753806538 1005008670 4917549673 3848730261

3 1 0755505945 1052809093 7814969316 47342417822 0772706183 0894907255 7348966512 5171843368

4 1 0658205365 0879706251 6637355333 36128264862 0640404911 0950006746 6228056617 3752125645

5 1 0689904989 0861806124 6221863820 29503258232 0699605634 1001707915 6320361131 4186732785

48 h

1 1 1219808343 1773514304 9778257246 68402402892 1259808949 1694112488 9455772439 6503044361

2 1 1225009604 2196016746 10977584764 79032593072 1285710131 2033015552 8143463813 6529343273

3 1 1240707253 2165517786 9658069204 75541529892 1325009565 1841415666 8102472309 6245253772

4 1 0916208120 1761813047 7900955968 57899453712 0989607848 1502011732 7097848208 5282637708

5 1 0965007833 1696014327 9110774706 87411640062 1131510845 1973115516 9244968041 6805748502

forecast model (1 or 2) are divided into two groups with thesame number of observation points One group of observa-tion points is used to establish prediction equations and theother group of points is used to test the accuracies of theseequations The results of testing these prediction equationsin forecast model 1 and forecast model 2 are shown inTable 8 In combinationwith formula (3) the calculatedmeanand standard deviation of GCD in two forecast models areshown in Figures 15 and 16 respectively Averaging theresults of five areas it can be obtained that the averagesof the meanstandard deviation of GCD under forecastmodel 1 and forecast model 2 for 24 hoursrsquo forecast fare1505192846156 km and 1415464812509 km respectivelyFor 48 hoursrsquo forecast the averages of the meanstandarddeviation of GCD under two different forecast modelsare 26175171456345 km and 25671091452903 km respec-tively Even though the mean of GCD in dynamic prediction

pattern is no less than three main prediction centers (NHCJMA and NMCC) it is much less than the numerical pre-diction model in [25] the means of which are 18633195 kmbased on System T106 and 16182958 km based on T213 bothfor 2448 hoursrsquo forecast Besides forecast models 1 and 2 areall more accurate than the forecast using satellite scatterome-terrsquos monitoring data in the sense of the standard deviationof GCD and the mean of weed speed error which are1496002 km and 119618ms in [12] It can be seen fromTable 8 and Figures 15 and 16 that the accuracies of forecastmodel 1 and forecast model 2 vary from different areas anddifferent characteristic factors The more accurate forecastmodel can be selected from forecast models 1 and 2 accordingto actual conditionsThe results of statistical significance testsfor each equation used to forecast corresponding charac-teristic factor in forecast model 1 and forecast model 2 areshown in Table 9 which show that 119875 value is much less than

12 Advances in Meteorology

Table 9 The results of statistical significance tests for each equation used to forecast corresponding characteristic factor in forecast model 1and forecast model 2

Forecast hour Area numberForecastmodel(1 or 2)

119877-square119875 valuefor LAT

119877-square119875 valuefor LON

119877-square119875 valuefor AP

119877-square119875 valuefor WS

24 h

1 1 05883952 times 10minus11 06011791 times 10minus9 06222446 times 10minus10 07615547 times 10minus15

2 04493226 times 10minus32 05809593 times 10minus50 07030131 times 10minus68 07731101 times 10minus81

2 1 07177186 times 10minus14 04391366 times 10minus5 07556503 times 10minus16 07995357 times 10minus18

2 06808517 times 10minus101 06753177 times 10minus99 07481213 times 10minus122 07277371 times 10minus114

3 1 06471636 times 10minus11 06559348 times 10minus11 06310183 times 10minus10 06495324 times 10minus10

2 06802646 times 10minus94 07506119 times 10minus114 05509140 times 10minus62 05573984 times 10minus65

4 1 07572105 times 10minus12 05348332 times 10minus8 07594115 times 10minus13 08174247 times 10minus15

2 06405483 times 10minus42 06266164 times 10minus40 06903101 times 10minus44 07652925 times 10minus57

5 1 05031240 times 10minus6 05154151 times 10minus6 08732273 times 10minus14 08524503 times 10minus15

2 06239284 times 10minus35 06464167 times 10minus37 07835698 times 10minus52 07550152 times 10minus48

48 h

1 1 0133700240 03943391 times 10minus5 03208177 times 10minus4 04725534 times 10minus6

2 0217700049 0039903398 0253200016 04275792 times 10minus6

2 1 04515905 times 10minus6 0266800218 06720172 times 10minus10 04622599 times 10minus6

2 04022164 times 10minus36 03095651 times 10minus26 02960505 times 10minus23 03332627 times 10minus27

3 1 0225100218 03935457 times 10minus5 0002607270 01362015962 03073618 times 10minus25 04369213 times 10minus39 01768251 times 10minus12 02479891 times 10minus18

4 1 04992960 times 10minus7 03055274 times 10minus4 04303153 times 10minus5 06249184 times 10minus9

2 03323377 times 10minus16 03270102 times 10minus16 02471170 times 10minus9 03580365 times 10minus16

5 1 04546619 times 10minus5 0255400043 03705191 times 10minus4 03535000122 04219109 times 10minus18 03825344 times 10minus17 02120218 times 10minus7 03322246 times 10minus12

005 in almost every case and prove that the correspondingprediction equation is significant

4 Summary

In this paper the CMA best track datasets from 1949 to 2012are used in combination with data mining technology andstatistical methods to put forward a new methodology toforecast TCsrsquo characteristic factors This methodology canaccurately judge whether TCs land on Hainan Island or notand forecast their characteristic factors (including longitudelatitude the lowest center pressure and wind speed) Theaverage of the probabilities of accurate judgment for landingcriterions is 7470 and the highest accuracy can reach7976 For the forecast of landing TCsrsquo characteristic factorslanding prediction pattern and dynamic prediction patternare proposed which not only can accurately forecast thecharacteristic factors when TCs land but also realize dynam-ically 24 hoursrsquo and 48 hoursrsquo forecast The effect of thelanding prediction pattern is better of which the mean ofGCD is 1446382 km compared with the current 48 hoursrsquoforecast in the South China Sea which is 2226 km Eventhough the mean of GCD in dynamic prediction pattern isno less than three main prediction centers (NHC JMA andNMCC) it is much less than the numerical prediction modelin [25] and the method using satellite scatterometerrsquos mon-itoring data in [12] The forecast methodology proposed inthis paper provides a new method for typhoon warning on

Hainan Islandwithout getting toomuch knowledge ofmeteo-rology involved and thus simplifies the implementation of theprediction process andmeanwhile guarantees the accuracy ofprediction

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

Acknowledgments

This work is in part supported by the National Science-Technology Support Plan in the domain of advanced energytechnology Hainan Power Grid Corporation provided sup-port for this research through ldquoIntegrated DemonstrationProject of Regional Smart Gridrdquo no 2013BAA01B03 DrXinhong Huang a Research Engineer in the Department ofElectrical amp Computer Engineering from the University ofWestern Ontario also puts forward valuable revision com-ments on this paper

References

[1] National Climate Center of China Typhoon ldquoDamreyrdquo HasCaused Serious Damages to Hainan Province 2005 httpncccmagovcnWebsiteindexphpNewsID=1454

Advances in Meteorology 13

[2] J S Pedro F Burstein and A Sharp ldquoA case-based fuzzy mul-ticriteria decision support model for tropical cyclone forecast-ingrdquo European Journal of Operational Research vol 160 no 2pp 308ndash324 2005

[3] X Qin and M Mu ldquoA Study on the reduction of forecast errorvariance by three adaptive observation approaches for tropicalcyclone predictionrdquoMonthlyWeather Review vol 139 no 7 pp2218ndash2232 2011

[4] Y Wang W Zhang and W Fu ldquoBack Propogation(BP)-neuralnetwork for tropical cyclone track forecastrdquo in Proceedings of the19th International Conference on Geoinformatics pp 1ndash4 IEEEShanghai China June 2011

[5] B Feng and J N K Liu ldquoAn adaptive neural network classifierfor tropical cyclone prediction using a two-layer feature selec-torrdquo in Advances in Neural NetworksmdashISNN 2005 vol 3497of Lecture Notes in Computer Science pp 399ndash404 SpringerBerlin Germany 2005

[6] H-J Song S-H Huh J-H Kim C-H Ho and S-K ParkldquoTyphoon track prediction by a support vector machine usingdata reduction methodsrdquo in Computational Intelligence andSecurity vol 3801 of Lecture Notes in Computer Science pp 503ndash511 Springer Berlin Germany 2005

[7] M DeMaria ldquoA simplified dynamical system for tropicalcyclone intensity predictionrdquoMonthly Weather Review vol 137no 1 pp 68ndash82 2009

[8] H R Winterbottom E W Uhlhorn and E P Chassignet ldquoAdesign and an application of a regional coupled atmosphere-ocean model for tropical cyclone predictionrdquo Journal ofAdvances in Modeling Earth Systems vol 4 no 10 Article IDM10002 2012

[9] L M Ma and Z M Tan ldquoImproving the behavior of the cumu-lus parameterization for tropical cyclone prediction convectiontriggerrdquo Atmospheric Research vol 92 no 2 pp 190ndash211 2009

[10] J S Gall I Ginis S-J Lin T P Marchok and J-H ChenldquoExperimental tropical cyclone prediction using the GFDL 25-km-resolution global atmospheric modelrdquo Weather and Fore-casting vol 26 no 6 pp 1008ndash1019 2011

[11] G Lv ldquoThe Northwestern Pacific typhoon track forecast in2005rdquo in Proceedings of the Meteorological Science and Technol-ogy Symposium on Taiwan Strait p 7 Chinese MeteorologicalSociety 2006 (Chinese)

[12] D Zhang Y Zhang T Hu B Xie and J Xu ldquoA comparison ofHY-2 and QuikSCAT vector wind products for tropical cyclonetrack and intensity development monitoringrdquo IEEE Geoscienceand Remote Sensing Letters vol 11 no 8 pp 1365ndash1369 2014

[13] MYingWZhangHYu et al ldquoAnoverviewof theChinamete-orological administration tropical cyclone databaserdquo Journal ofAtmospheric and Oceanic Technology vol 31 no 2 pp 287ndash3012014

[14] G E Forsythe ldquoGeneration and use of orthogonal polynomialsfor data-fittingwith a digital computerrdquo Journal of the Society forIndustrial amp Applied Mathematics vol 5 no 2 pp 74ndash88 1957

[15] KWagstaff C Cardie S Rogers and S Schroedl ldquoConstrainedk-means clustering with background knowledgerdquo in Proceed-ings of the 18th International Conference on Machine Learningvol 1 pp 577ndash584 2001

[16] Y Ye ldquoNeighborhood density method for selecting initialcluster centers in K-mean clusteringrdquo in Proceedings of theWorkshop on Data Mining for Biomedical Applications (PAKDDrsquo06) pp 189ndash198 2006

[17] P Xiong Data Mining Algorithm and Clementine PracticeTsinghua University Press Beijing China 2011 (Chinese)

[18] S L Crawford ldquoExtensions to the CART algorithmrdquo Interna-tional Journal ofMan-Machine Studies vol 31 no 2 pp 197ndash2171989

[19] D M Allen ldquoMean square error of prediction as a criterion forselecting variablesrdquo Technometrics vol 13 pp 469ndash475 1971

[20] S Yu and J Shen ldquoForward and backward algorithms forselecting predictors on the basis of the criterion frompredictionsumof squares and their applicationrdquoActaMeteorologica Sinicavol 1 pp 83ndash90 1988

[21] D Yao and S Yu ldquoThe stepwise algorithm of selecting forecastfactors based on PRESS rulerdquo Journal of Atmospheric Sciencesvol 2 pp 129ndash135 1992 (Chinese)

[22] Y LuMathematical Statistics Methods East China University ofScience and Technology Press Shanghai China 2005 (Chi-nese)

[23] K Xie and B Liu ldquoAn ENSO-forecast independent statisticalmodel for the prediction of annual Atlantic tropical cyclonefrequency in Aprilrdquo Advances in Meteorology vol 2014 ArticleID 248148 11 pages 2014

[24] J Yu J Tang Y Dai and B Yu ldquoThe error and cause analysis ofChinarsquos typhoon path predictionrdquo Journal of Weather vol 6 pp695ndash700 2012

[25] SMa A Qu and Z Yu ldquoThe parallelization of typhoon numer-ical prediction model of and track forecast error analysisrdquo Jour-nal of Applied Meteorology vol 3 pp 322ndash328 2004 (Chinese)

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ClimatologyJournal of

EcologyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

EarthquakesJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom

Applied ampEnvironmentalSoil Science

Volume 2014

Mining

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014

International Journal of

Geophysics

OceanographyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of Computational Environmental SciencesHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal ofPetroleum Engineering

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

GeochemistryHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of

Atmospheric SciencesInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

OceanographyHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

MineralogyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

MeteorologyAdvances in

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Paleontology JournalHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ScientificaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Geological ResearchJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Geology Advances in

Advances in Meteorology 3

1085 109 1095 110 1105 111 111518

185

19

195

20

205

The fitting curve of Hainan IslandThe actual outer boundary of Hainan Island

Latit

ude (

∘N

)

Longitude (∘E)

Figure 3 The curve fitting of the outer boundary of Hainan Island

where 119877 is the mean radius of the Earth with the value of6370856 km sgn(119909) represents the sign function the unit ofLONMV

119905and LATMV

119905is kmh

22 Methodology As mentioned in the Introduction a land-ing TC is defined as a TC that theminimumdistance betweenthe typhoon center and Hainan Island is no more than thepreset influencing radius Hence in order to distinguish TCsbetween landing and not landing onHainan Island themini-mum distance between each TCrsquos track and the outer bound-ary ofHainan Islandneeds to be calculated according toCMAbest track datasets Due to the variety of TCsrsquo tracks applyinggeneral curve-fitting directly does not provide a good resultTherefore polynomial fitting [14] is applied in this paperwhere an intermediate variable is introduced to conductcurve-fitting with the latitude and longitude respectivelyTaking an arbitrary TC for example the specific fitting effectsare shown in Figures 3 and 4 Using the fitting polynomials ofeach TCrsquos and the outer boundary of Hainan Islandrsquos latitudeand longitudewith respect to the corresponding intermediatevariables the distance between any point of each TCrsquos trackand any point on the outer boundary of Hainan Island canbe calculated from which the minimum distance can beselected The great circle distance (GCD) between any twopoints on the Earth can be calculated using formula (3) TheGCD is the shortest distance between any two points on theEarth Set any two points on the Earth as 119889

1(119864

1 119873

1) and

119889

2(119864

2 119873

2) and the GCD is

1003816

1003816

1003816

1003816

119889

1119889

2

1003816

1003816

1003816

1003816

= 119877 sdot arccos [cos1198731sdot cos119873

2sdot cos (119864

1minus 119864

2)

+ sin1198731sdot sin119873

2]

(3)

The time intervals used to forecast TCs by three mainprediction centers (NHC JMA and NMCC) are 24 48 and72 hours In order to forecast TCs in a timely manner andcompare forecast accuracy among different methods here

100 110 120 130 1404

6

8

10

12

14

16

18

20

22

The actual trackThe fitting trackHainan Island

Longitude (∘E)

Latit

ude (

∘N

)

The actual valueThe fitting value

0 20 40 605

10

15

20

Intermediate variable

0 20 40 60110

120

130

140

Intermediate variable

The actual valueThe fitting value

The l

ongi

tude

of T

C ce

nter

(∘E)

The l

atitu

de o

f TC

cent

er(∘

N)

Figure 4 The curve fitting of an arbitrary TC

110 115 120 125

12

14

16

18

20

22

24

26

Hainan IslandThe landing positions

The positions where TCsrsquo centerswere located 48 hours before theylanded on Hainan Island

Longitude (∘E)

Latit

ude (

∘N

)

Figure 5 The region selected as research object

the region where TCsrsquo centers were located 48 hours beforethey landed on Hainan Island (shown in Figure 5) is selectedas research object In order to narrow the research scope 119870-means clustering algorithm [15 16] is applied to divide theregion where TCsrsquo centers were located 48 hours before theylanded on Hainan Island into five areas In this section thesituation inwhich the regionwhere TCsrsquo centers were located48 hours before they landed is selected as research object andthe research object is divided into five areas is taken as

4 Advances in Meteorology

Screening out the landing TCs and not landing TCs

Filtrating all the observation points of both landing and not landing TCs entering into each area

Forming the landing criterion of each area using CART algorithm

Obtaining the TCsrsquo records aroundHainan Island

Finding out the region where TCsrsquo centers were located48 (or 72) hours before they landed

Dividing the research region into 5 (or a number of 1 3 7) areas

Figure 6 Flow diagram of forming landing criterions

an example for a convenient statement Other situations ofa comparative experiment are also conducted in Section 33

For each of the five areas all the observation points ofboth landing and not landing TCs which entered into thisarea are filtrated With strength grade latitude longitude airpressure wind speed latitude migration velocity and longi-tude migration velocity as classification properties the TClanding criterion of each area is formed by using CART algo-rithm [17 18]The flow diagram of forming landing criterionsis shown in Figure 6

For TCs which are judged to be landing on HainanIsland PRESS and its progressive optimal algorithm andMLRM can be used to forecast TCsrsquo characteristic factors(including latitude longitude the lowest center pressureand wind speed) Forecasts in this paper contain landingprediction pattern and dynamic prediction pattern Landingprediction pattern is defined as employing the observationpoint information of those points which first enter into anyarea to predict the characteristic factors when TC landsDynamic prediction pattern is defined as 24 hoursrsquo and 48hoursrsquo prediction with respect to the observation point whichenters into any area The flow diagrams of landing forecastpattern and dynamic forecast pattern are shown in Figures 7and 8 respectively Here PRESS [19] and its progressiveoptimal algorithm [20 21] are used to select the best forecastfactor set from seven predictors which will be used toforecast corresponding characteristic factor MLRM [22] isused to establish corresponding forecast equations MLRM isexpressed as [23]

119910

119894= 120573

0+ 120573

1119909

1198941+ 120573

2119909

1198942+ sdot sdot sdot + 120573

119898119909

119894119898+ 120576

119894

(119894 = 1 2 sdot sdot sdot 119899)

(4)

where119910119894is the estimated value120573

0sim 120573

119898is the regression coef-

ficients 120576119894is the random error and 119909

1198941sim 119909

119894119898are the forecast

factors of the observation point

Table 1 The geometric centers and scopes of five areas

Area number LongitudeUnit ∘E

LatitudeUnit ∘N

RadiusUnit degree

1 1184511 152845 279022 1131435 149623 317063 1169517 208023 330484 1239859 144951 270985 1226830 185854 28346

Table 2 The number of OPs in five areas

Area numberNumber oflanding(1198851

)

Number of notlanding(1198852

)

The ratio oflanding

119885

1

(119885

1

+ 119885

2

)

1 537 1031 34252 855 1389 38103 788 1514 34234 378 1070 26105 340 1419 1933

3 Results and Discussions

In this section the situation in which the region where TCsrsquocenters were located 48 hours before they landed is selectedas research object and the research object is divided into fiveareas is firstly researched Other situations of a comparativeexperiment in which the research object may be the regionwhere TCsrsquo centers were located 72 hours before they landedand the number of areas of divided research object may beany number of 1 3 5 7 are also discussed at the end of thesection

31 Dividing the Research Region into Five Areas 119870-meansclustering algorithm is used to divide our research region intofive areas as described in Section 22 of which the geometriccenters and scopes are shown in Table 1 With respect to eacharea all the observation points of both landing and not land-ing TCs which entered into this area are filtrated The posi-tions of these observation points are shown in Figure 9 andare used to form TCsrsquo landing criterions The numbers ofthese observation points (OPs) for both landing and notlanding TCs are shown in Table 2 The division of five areasfurther narrows the research scope and makes the selectionof the observation points more pertinent so as to form theeffective landing criterions which will be illustrated furtherin Section 33

32 The Formation of Landing Criterions in Five AreasAccording to the CART algorithm the landing criterionsin five areas are shown in Figure 10 (refer to Section 21for the meaning of seven predictors) The correspondingprobability of accurate judgment (119875AJ) probability of falsealarm (119875FA) and probability of false dismissal (119875FD) are shown

Advances in Meteorology 5

Finding out the area in which the observation point first enters into

Is the TC landing on Hainan Island

Using the corresponding landing criterion

Need not forecast

Screening out all the observation points of historical landing TCs when they firstly

entered into the corresponding area

Finding out the observation information of these observation points when they landed

Using PRESS and CART to establish corresponding forecast equation for each

characteristic factor

Obtaining four characteristic factors when forecasted TC lands

on Hainan Island

Yes

No

Figure 7 Flow diagram of landing prediction pattern

Finding out the area in which the observation point first enters into

Is the TC landing onHainan Island

Using the landing criterion of

corresponding area

Need not forecast

Screening out all the observation points of

historical landing TCs when they firstly entered into the

corresponding area

Finding out the observation information of these

observation points 24 (or 48) hours later

Using PRESS and CART to establish corresponding forecast equation for each

characteristic factor

Finding out the area in which the observation point appears (it is not

necessary that the observation point enters into this area for the first time)

Screening out all the observation points of

historical landing TCs when they appeared in the corresponding area

1 Forecast model 12 Forecast model 2

Finding out the observation information of these

observation points 24 (or 48) hours later

Using PRESS and CART to establish corresponding

forecast equation for each characteristic factor

Obtaining four characteristic factors of the TC 24 (or 48)

hours later

Yes

No

12

Figure 8 Flow diagram of dynamic prediction pattern

in Table 3 Set the numbers ofOPs for landing and not landingTCs in any area to be 119885

1and 119885

2 respectively the number

of OPs which are judged to be landing according to landing

criterions when they landed truly is denoted as 1198721and the

number of OPs which are judged to be not landing accordingto landing criterions when they did not land truly is denoted

6 Advances in Meteorology

110 115 120 12510

12

14

16

18

20

22

24

26

The positions of OPs for landing and not landing TCs in area 1

OPs of not landing TCsOPs of landing TCsHainan Island

The positions when TCs landedThe boundary of area 1

OPs of not landing TCsOPs of landing TCsHainan Island

The positions when TCs landedThe boundary of area 2

OPs of not landing TCsOPs of landing TCsHainan Island

The positions when TCs landedThe boundary of area 3

OPs of not landing TCsOPs of landing TCsHainan Island

The positions when TCs landedThe boundary of area 4

OPs of not landing TCsOPs of landing TCsHainan Island

The positions when TCs landedThe boundary of area 5

110 115 120 12510

12

14

16

18

20

22

24

26

The positions of OPs for landing and not landing TCs in area 5

Longitude (∘E)

Longitude (∘E)

110 115 120 12510

12

14

16

18

20

22

24

26

The positions of OPs for landing and not landing TCs in area 2

Longitude (∘E)

Latit

ude (

∘N

)

Latit

ude (

∘N

)

Latit

ude (

∘N

)

110 115 120 12510

12

14

16

18

20

22

24

26

The positions of OPs for landing and not landing TCs in area 3

Longitude (∘E)

Latit

ude (

∘N

)

110 115 120 12510

12

14

16

18

20

22

24

26

The positions of OPs for landing and not landing TCs in area 4

Longitude (∘E)

Latit

ude (

∘N

)

Figure 9 The positions of OPs for landing and not landing TCs which entered into each area

Advances in Meteorology 7

Calculate the seven predictors

of any OP in area 1

Calculate the seven predictors

of any OP in area 2

LONMVYes No

LAT

LONMV

LONMV

LAT

LON

No

NO

Yes

Yes

No

No

No

Yes

Yes

Yes

Landing

Landing

Landing

Not landing

Not landing Not

landingNot

landing

Not landing

Not landing

Not landing

Not landing

Not landing

Not landing Not

landing

Not landing

Not landing

Not landing

Not landing

Not landing

Not landing

LATYes

LONMVLAT

WS

AP

No

Yes

Yes

Yes

No

Yes

No

No

No

Landing

Landing

Calculate the seven predictors

of any OP in area 3

LONMV

LAT

LON

LONMV

LATMV

Landing

Landing

Landing

Yes

Yes

No

No

Yes

Yes

Yes No

No

No

Calculate the seven predictors

of any OP in area 4

LONMV

LAT

LONMV

SG

AP

YesYes

Yes

Yes

Yes

No

No

No

NoNo

Landing

Landing

Landing

Calculate the seven predictors

of any OP in area 5

LONMV

LONMV

LATMV LAT

LATMV

AP

AP

No

Yes

Yes No

Yes NoYes

NoNo No

No

YesYes

Yes

Landing Landing

le14250

le116850

le minus27918

leminus5358

leminus16795le15950

leminus0892gt14450

le11

le10055

leminus6058

le20650

le116650

le22310

le2317

le minus13357

le12550

le1500

le989000

le minus9666

le minus22048

le10656

le19450le10656

le968500

le1002500

leminus21621

gt14950

Figure 10 The landing criterions in five areas

Table 3 119875AJ 119875FA and 119875FD of each criterion

Area number 119875AJ 119875FA 119875FD

1 7468 1794 39482 7531 1555 39533 7750 1176 43154 6623 3654 25935 7976 1649 3588

as 1198722 Then 119875AJ 119875FA and 119875FD of this area are calculated as

follows

119875AJ =119872

1+119872

2

119885

1+ 119885

2

119875FA =

119885

2minus119872

2

119885

2

119875FD =

119885

1minus119872

1

119885

1

(5)

Table 4 The labels of different situations

Ti Nu Label of corresponding situation48 1 FE148 3 FE348 5 FE548 7 FE772 1 ST172 3 ST372 5 ST572 7 ST7

33 Other Situations as a Comparative Experiment In Sec-tions 31 and 32 the region where TCsrsquo centers were located48 hours before they landed is selected as research object andthe research object is divided into five areas In this sectionother situations as a comparative experiment are researched

8 Advances in Meteorology

Table 5119875AJ119875FA and119875FD for each of the remaining seven situations

(a) 119875AJ 119875FA and 119875FD of each area for FE1

Area number 119875AJ 119875FA 119875FD

1 07884 00696 05279

(b) 119875AJ 119875FA and 119875FD of each area for FE3

Area number 119875AJ 119875FA 119875FD

1 07454 01139 051912 07636 01875 030543 07829 00446 07814

(c) 119875AJ 119875FA and 119875FD of each area for FE7

Area number 119875AJ 119875FA 119875FD

1 08522 00154 084142 07486 01991 039593 07302 01808 041764 07719 01165 043985 07236 00740 067356 07834 01512 029757 07631 0 1

(d) 119875AJ 119875FA and 119875FD of each area for ST1

Area number 119875AJ 119875FA 119875FD

1 08164 00412 06998

(e) 119875AJ 119875FA and 119875FD of each area for ST3

Area number 119875AJ 119875FA 119875FD

1 08320 00677 058772 07396 02062 034623 08270 0 1

(f) 119875AJ 119875FA and 119875FD of each area for ST5

Area number 119875AJ 119875FA 119875FD

1 08284 0 12 07613 0 13 07566 01590 035644 08596 0 15 08281 00678 05813

(g) 119875AJ 119875FA and 119875FD of each area for ST7

Area number 119875AJ 119875FA 119875FD

1 07874 0 12 07020 01664 047233 08799 00118 087774 08358 0 15 08537 0 16 07761 00955 050477 07407 01028 06720

and compared with each other Finally we select the situationwhich produces the best result

In order to distinguish different situations the labels ofthemare denoted inTable 4 where the parameter Ti is used to

Table 6 The Index for each of eight situations

Label of corresponding situation IndexFE1 06750FE3 06659FE5 07026FE7 06625ST1 06297ST3 06572ST5 06391ST7 06231

illustrate the research object is the region where TCsrsquo centersare located Ti hours before they landed onHainan Island Nudenotes the number of areas of divided research object

It can be seen from Table 4 that FE5 is the situation whichhas been researched in Sections 31 and 32 The remainingseven situations are researched as follows using the methodswhich are identical with FE5

The 119875AJ 119875FA and 119875FD of each area for each of the remain-ing seven situations can be calculated according to formula(5) the results of which are shown in Table 5

In order to select the best of these eight situations inTable 4 an evaluation method is introduced with which theIndex of each situation is calculated where Index is definedaccording to formula (6) For any situation 119873 denotes thenumber of areas of divided research object and119875AJ 119894119875FA 119894 and119875FD 119894 denote the 119875AJ 119875FA and 119875FD of the 119894th area respectivelyConsider the following

Index =

1

119873

119873

sum

119894=1

radic[

1

10

119875

2

AJ 119894 +3

10

(1 minus 119875FA 119894)2

+

3

5

(1 minus 119875FD 119894)2

]

(6)

It is obvious that the higher the Index is the better theresult of the landing criterions on the whole is The Indexfor each of eight situations is shown in Table 6 The situationFE5 shows the best result which also illustrates that theresearch scheme in Sections 31 and 32 is better comparedwith other situations Finally situation FE5 is selected to formthe landing criterions

34 Forecast of TCsrsquo Characteristic Factors

341 Landing Prediction Pattern The landing forecast pat-tern is defined as follows obtaining the observation pointinformation (seven predictors) when landing TCsrsquo centersfirst enter into any area which can be used to forecast thecharacteristic factors (LAT LONAP andWS)whenTCs landon Hainan Island The flow diagram is shown in Figure 6Taking area 1 for example the OPs of historical landing TCsrsquocenters when they first entered into area 1 are shown inFigure 11 and the tracks of historical landing TCs whichpassed through area 1 are shown in Figure 12

Dividing the historical landing TCs passing through eacharea into two groups with the same number one group of TCsis used to establish prediction equations and the other group

Advances in Meteorology 9

110 115 120 125

10

12

14

16

18

20

22

24

26

Hainan IslandThe boundary of area 1

The LPs of TCsThe OPs of landing TCs enteringinto area 1 for the first time

Longitude (∘E)

Latit

ude (

∘N

)

Figure 11 The OPs of historical landing TCsrsquo centers when theyfirstly entered into area 1 (LP landing position)

Hainan IslandThe boundary of area 1

The LPs of TCs

110 115 120 1258

10

12

14

16

18

20

22

24

Latit

ude (

∘N

)

Longitude (∘E)

The TCs which passedthrough area 1

Figure 12 The tracks of historical landing TCs which passedthrough area 1

is used to test the accuracy of these equations The results oftesting of these prediction equations for TCs which passedthrough each area are shown in Table 7 Making use of theactual and predicted longitude and latitude of TCsrsquo centersin combination with the formula (3) the calculated meanand standard deviation of GCD in the landing predictionpattern are shown in Figures 13 and 14 respectively Averagingthe results of five areas it can be obtained that the averageof the meanstandard deviation (SD) of GCD is 1446382978740 km In [24] Yu et al analyze the averageGCD error of48 hoursrsquo forecast in the South China Sea which is 2226 kmTherefore the landing prediction pattern proposed in this

1 2 3 4 50

20

40

60

80

100

120

140

160

Area number

The m

ean

of G

CD (k

m)

The mean of GCD for each area in landing prediction pattern

Figure 13 The mean of GCD for each area in landing predictionpattern

1 2 3 4 50

20

40

60

80

100

Area number

The s

tand

ard

devi

atio

n of

GCD

(km

)The standard deviation of GCD for each area in

landing prediction pattern

Figure 14 The standard deviation of GCD for each area in landingprediction pattern

paper shows good prediction accuracy For TCs which arejudged to be landing on Hainan Island as long as the obser-vation point information when their centers first enter intoany area are obtained the corresponding forecast equationscan be used to predict characteristic factors when they land

342 Dynamic Prediction Pattern The dynamic predictionpattern is using the current observation point informationto conduct 24 hoursrsquo and 48 hoursrsquo forecast which is alsoillustrated in Figure 8There are two different forecast modelsin dynamic prediction pattern that are described as follows

Forecast Model 1 It is to obtain the current observation pointinformation (seven predictors) when landing TCsrsquo centersenter into any area for the first time making use of which toconduct 24 hoursrsquo and 48 hoursrsquo forecast

10 Advances in Meteorology

1 2 3 4 50

50

100

150

200

Area number

The m

ean

of G

CD (k

m)

The mean of GCD for 24 hoursrsquo forecast in each areausing forecast models 1 and 2

Forecast model 1Forecast model 2

(a) The mean of GCD for 24 hoursrsquo forecast of two forecast models

Forecast model 1Forecast model 2

1 2 3 4 50

20

40

60

80

100

Area number

The s

tand

ard

devi

atio

n of

GCD

(km

)

The standard deviation of GCD for 24 hoursrsquo forecastin each area using forecast models 1 and 2

(b) The standard deviation of GCD for 24 hoursrsquo forecast of two forecastmodels

Figure 15 The meanstandard deviation of GCD for 24 hoursrsquo forecast of two forecast models

Forecast model 1Forecast model 2

1 2 3 4 50

50

100

150

200

250

300

Area number

The m

ean

of G

CD (k

m)

The mean of GCD for 48 hoursrsquo forecast in each areausing forecast models 1 and 2

(a) The mean of GCD for 48 hoursrsquo forecast of two forecast models

Forecast model 1Forecast model 2

1 2 3 4 50

50

100

150

200

Area number

The s

tand

ard

devi

atio

n of

GCD

(km

)

The standard deviation of GCD for 48 hoursrsquo forecastin each area using forecast models 1 and 2

(b) The standard deviation of GCD for 48 hoursrsquo forecast of two forecastmodels

Figure 16 The meanstandard deviation of GCD for 48 hoursrsquo forecast of two forecast models

Forecast Model 2 It is to obtain the current observation pointinformation (seven predictors) when landing TCsrsquo centers arein any area (not necessarily enter into any area for the firsttime)making use ofwhich to conduct 24 hoursrsquo and 48hoursrsquoforecast

In the process of actual prediction for TCs which arejudged to be landing on Hainan Island the observation pointinformationwhen their centers enter into any area for the first

time and the established equations in forecast model 1 areused to conduct dynamic prediction Furthermore the obser-vation point information when TCsrsquo centers are in area (it isnot necessary that TCsrsquo centers enters into this area for thefirst time) any area and the established equations in forecastmodel 2 can be used to conduct dynamic prediction

Similar to Section 341 the historical observation pointsthat meet the corresponding requirements in corresponding

Advances in Meteorology 11

Table 7 Prediction results in landing prediction pattern

Area numberThe meanSD of LAT

deviationUnit ∘N

The meanSD of LATdeviationUnit ∘E

The meanSD of APdeviationUnit hpa

The meanSD of WSdeviationUnit ms

1 0984908671 0726406404 9288967882 59540438392 0526806009 0867708973 7994665339 53036389973 0971007805 0634007711 10694183758 70915526374 1153708640 0715006162 10346361761 68151474985 1115206941 0672807103 12113885744 8040353759

Table 8 The results of testing forecast model 1 and forecast model 2

Forecast hour Area number Forecast model(1 or 2)

The meanSD ofLAT deviation

Unit ∘N

The meanSD ofLON deviation

Unit ∘E

The meanSD ofAP deviationUnit hpa

The meanSD ofWS deviationUnit ms

24 h

1 1 0812806515 1029507600 4927839733 37787288582 0700405153 0880207011 5411443187 3963729843

2 1 0913105832 0986407794 6341244247 38268278572 0753806538 1005008670 4917549673 3848730261

3 1 0755505945 1052809093 7814969316 47342417822 0772706183 0894907255 7348966512 5171843368

4 1 0658205365 0879706251 6637355333 36128264862 0640404911 0950006746 6228056617 3752125645

5 1 0689904989 0861806124 6221863820 29503258232 0699605634 1001707915 6320361131 4186732785

48 h

1 1 1219808343 1773514304 9778257246 68402402892 1259808949 1694112488 9455772439 6503044361

2 1 1225009604 2196016746 10977584764 79032593072 1285710131 2033015552 8143463813 6529343273

3 1 1240707253 2165517786 9658069204 75541529892 1325009565 1841415666 8102472309 6245253772

4 1 0916208120 1761813047 7900955968 57899453712 0989607848 1502011732 7097848208 5282637708

5 1 0965007833 1696014327 9110774706 87411640062 1131510845 1973115516 9244968041 6805748502

forecast model (1 or 2) are divided into two groups with thesame number of observation points One group of observa-tion points is used to establish prediction equations and theother group of points is used to test the accuracies of theseequations The results of testing these prediction equationsin forecast model 1 and forecast model 2 are shown inTable 8 In combinationwith formula (3) the calculatedmeanand standard deviation of GCD in two forecast models areshown in Figures 15 and 16 respectively Averaging theresults of five areas it can be obtained that the averagesof the meanstandard deviation of GCD under forecastmodel 1 and forecast model 2 for 24 hoursrsquo forecast fare1505192846156 km and 1415464812509 km respectivelyFor 48 hoursrsquo forecast the averages of the meanstandarddeviation of GCD under two different forecast modelsare 26175171456345 km and 25671091452903 km respec-tively Even though the mean of GCD in dynamic prediction

pattern is no less than three main prediction centers (NHCJMA and NMCC) it is much less than the numerical pre-diction model in [25] the means of which are 18633195 kmbased on System T106 and 16182958 km based on T213 bothfor 2448 hoursrsquo forecast Besides forecast models 1 and 2 areall more accurate than the forecast using satellite scatterome-terrsquos monitoring data in the sense of the standard deviationof GCD and the mean of weed speed error which are1496002 km and 119618ms in [12] It can be seen fromTable 8 and Figures 15 and 16 that the accuracies of forecastmodel 1 and forecast model 2 vary from different areas anddifferent characteristic factors The more accurate forecastmodel can be selected from forecast models 1 and 2 accordingto actual conditionsThe results of statistical significance testsfor each equation used to forecast corresponding charac-teristic factor in forecast model 1 and forecast model 2 areshown in Table 9 which show that 119875 value is much less than

12 Advances in Meteorology

Table 9 The results of statistical significance tests for each equation used to forecast corresponding characteristic factor in forecast model 1and forecast model 2

Forecast hour Area numberForecastmodel(1 or 2)

119877-square119875 valuefor LAT

119877-square119875 valuefor LON

119877-square119875 valuefor AP

119877-square119875 valuefor WS

24 h

1 1 05883952 times 10minus11 06011791 times 10minus9 06222446 times 10minus10 07615547 times 10minus15

2 04493226 times 10minus32 05809593 times 10minus50 07030131 times 10minus68 07731101 times 10minus81

2 1 07177186 times 10minus14 04391366 times 10minus5 07556503 times 10minus16 07995357 times 10minus18

2 06808517 times 10minus101 06753177 times 10minus99 07481213 times 10minus122 07277371 times 10minus114

3 1 06471636 times 10minus11 06559348 times 10minus11 06310183 times 10minus10 06495324 times 10minus10

2 06802646 times 10minus94 07506119 times 10minus114 05509140 times 10minus62 05573984 times 10minus65

4 1 07572105 times 10minus12 05348332 times 10minus8 07594115 times 10minus13 08174247 times 10minus15

2 06405483 times 10minus42 06266164 times 10minus40 06903101 times 10minus44 07652925 times 10minus57

5 1 05031240 times 10minus6 05154151 times 10minus6 08732273 times 10minus14 08524503 times 10minus15

2 06239284 times 10minus35 06464167 times 10minus37 07835698 times 10minus52 07550152 times 10minus48

48 h

1 1 0133700240 03943391 times 10minus5 03208177 times 10minus4 04725534 times 10minus6

2 0217700049 0039903398 0253200016 04275792 times 10minus6

2 1 04515905 times 10minus6 0266800218 06720172 times 10minus10 04622599 times 10minus6

2 04022164 times 10minus36 03095651 times 10minus26 02960505 times 10minus23 03332627 times 10minus27

3 1 0225100218 03935457 times 10minus5 0002607270 01362015962 03073618 times 10minus25 04369213 times 10minus39 01768251 times 10minus12 02479891 times 10minus18

4 1 04992960 times 10minus7 03055274 times 10minus4 04303153 times 10minus5 06249184 times 10minus9

2 03323377 times 10minus16 03270102 times 10minus16 02471170 times 10minus9 03580365 times 10minus16

5 1 04546619 times 10minus5 0255400043 03705191 times 10minus4 03535000122 04219109 times 10minus18 03825344 times 10minus17 02120218 times 10minus7 03322246 times 10minus12

005 in almost every case and prove that the correspondingprediction equation is significant

4 Summary

In this paper the CMA best track datasets from 1949 to 2012are used in combination with data mining technology andstatistical methods to put forward a new methodology toforecast TCsrsquo characteristic factors This methodology canaccurately judge whether TCs land on Hainan Island or notand forecast their characteristic factors (including longitudelatitude the lowest center pressure and wind speed) Theaverage of the probabilities of accurate judgment for landingcriterions is 7470 and the highest accuracy can reach7976 For the forecast of landing TCsrsquo characteristic factorslanding prediction pattern and dynamic prediction patternare proposed which not only can accurately forecast thecharacteristic factors when TCs land but also realize dynam-ically 24 hoursrsquo and 48 hoursrsquo forecast The effect of thelanding prediction pattern is better of which the mean ofGCD is 1446382 km compared with the current 48 hoursrsquoforecast in the South China Sea which is 2226 km Eventhough the mean of GCD in dynamic prediction pattern isno less than three main prediction centers (NHC JMA andNMCC) it is much less than the numerical prediction modelin [25] and the method using satellite scatterometerrsquos mon-itoring data in [12] The forecast methodology proposed inthis paper provides a new method for typhoon warning on

Hainan Islandwithout getting toomuch knowledge ofmeteo-rology involved and thus simplifies the implementation of theprediction process andmeanwhile guarantees the accuracy ofprediction

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

Acknowledgments

This work is in part supported by the National Science-Technology Support Plan in the domain of advanced energytechnology Hainan Power Grid Corporation provided sup-port for this research through ldquoIntegrated DemonstrationProject of Regional Smart Gridrdquo no 2013BAA01B03 DrXinhong Huang a Research Engineer in the Department ofElectrical amp Computer Engineering from the University ofWestern Ontario also puts forward valuable revision com-ments on this paper

References

[1] National Climate Center of China Typhoon ldquoDamreyrdquo HasCaused Serious Damages to Hainan Province 2005 httpncccmagovcnWebsiteindexphpNewsID=1454

Advances in Meteorology 13

[2] J S Pedro F Burstein and A Sharp ldquoA case-based fuzzy mul-ticriteria decision support model for tropical cyclone forecast-ingrdquo European Journal of Operational Research vol 160 no 2pp 308ndash324 2005

[3] X Qin and M Mu ldquoA Study on the reduction of forecast errorvariance by three adaptive observation approaches for tropicalcyclone predictionrdquoMonthlyWeather Review vol 139 no 7 pp2218ndash2232 2011

[4] Y Wang W Zhang and W Fu ldquoBack Propogation(BP)-neuralnetwork for tropical cyclone track forecastrdquo in Proceedings of the19th International Conference on Geoinformatics pp 1ndash4 IEEEShanghai China June 2011

[5] B Feng and J N K Liu ldquoAn adaptive neural network classifierfor tropical cyclone prediction using a two-layer feature selec-torrdquo in Advances in Neural NetworksmdashISNN 2005 vol 3497of Lecture Notes in Computer Science pp 399ndash404 SpringerBerlin Germany 2005

[6] H-J Song S-H Huh J-H Kim C-H Ho and S-K ParkldquoTyphoon track prediction by a support vector machine usingdata reduction methodsrdquo in Computational Intelligence andSecurity vol 3801 of Lecture Notes in Computer Science pp 503ndash511 Springer Berlin Germany 2005

[7] M DeMaria ldquoA simplified dynamical system for tropicalcyclone intensity predictionrdquoMonthly Weather Review vol 137no 1 pp 68ndash82 2009

[8] H R Winterbottom E W Uhlhorn and E P Chassignet ldquoAdesign and an application of a regional coupled atmosphere-ocean model for tropical cyclone predictionrdquo Journal ofAdvances in Modeling Earth Systems vol 4 no 10 Article IDM10002 2012

[9] L M Ma and Z M Tan ldquoImproving the behavior of the cumu-lus parameterization for tropical cyclone prediction convectiontriggerrdquo Atmospheric Research vol 92 no 2 pp 190ndash211 2009

[10] J S Gall I Ginis S-J Lin T P Marchok and J-H ChenldquoExperimental tropical cyclone prediction using the GFDL 25-km-resolution global atmospheric modelrdquo Weather and Fore-casting vol 26 no 6 pp 1008ndash1019 2011

[11] G Lv ldquoThe Northwestern Pacific typhoon track forecast in2005rdquo in Proceedings of the Meteorological Science and Technol-ogy Symposium on Taiwan Strait p 7 Chinese MeteorologicalSociety 2006 (Chinese)

[12] D Zhang Y Zhang T Hu B Xie and J Xu ldquoA comparison ofHY-2 and QuikSCAT vector wind products for tropical cyclonetrack and intensity development monitoringrdquo IEEE Geoscienceand Remote Sensing Letters vol 11 no 8 pp 1365ndash1369 2014

[13] MYingWZhangHYu et al ldquoAnoverviewof theChinamete-orological administration tropical cyclone databaserdquo Journal ofAtmospheric and Oceanic Technology vol 31 no 2 pp 287ndash3012014

[14] G E Forsythe ldquoGeneration and use of orthogonal polynomialsfor data-fittingwith a digital computerrdquo Journal of the Society forIndustrial amp Applied Mathematics vol 5 no 2 pp 74ndash88 1957

[15] KWagstaff C Cardie S Rogers and S Schroedl ldquoConstrainedk-means clustering with background knowledgerdquo in Proceed-ings of the 18th International Conference on Machine Learningvol 1 pp 577ndash584 2001

[16] Y Ye ldquoNeighborhood density method for selecting initialcluster centers in K-mean clusteringrdquo in Proceedings of theWorkshop on Data Mining for Biomedical Applications (PAKDDrsquo06) pp 189ndash198 2006

[17] P Xiong Data Mining Algorithm and Clementine PracticeTsinghua University Press Beijing China 2011 (Chinese)

[18] S L Crawford ldquoExtensions to the CART algorithmrdquo Interna-tional Journal ofMan-Machine Studies vol 31 no 2 pp 197ndash2171989

[19] D M Allen ldquoMean square error of prediction as a criterion forselecting variablesrdquo Technometrics vol 13 pp 469ndash475 1971

[20] S Yu and J Shen ldquoForward and backward algorithms forselecting predictors on the basis of the criterion frompredictionsumof squares and their applicationrdquoActaMeteorologica Sinicavol 1 pp 83ndash90 1988

[21] D Yao and S Yu ldquoThe stepwise algorithm of selecting forecastfactors based on PRESS rulerdquo Journal of Atmospheric Sciencesvol 2 pp 129ndash135 1992 (Chinese)

[22] Y LuMathematical Statistics Methods East China University ofScience and Technology Press Shanghai China 2005 (Chi-nese)

[23] K Xie and B Liu ldquoAn ENSO-forecast independent statisticalmodel for the prediction of annual Atlantic tropical cyclonefrequency in Aprilrdquo Advances in Meteorology vol 2014 ArticleID 248148 11 pages 2014

[24] J Yu J Tang Y Dai and B Yu ldquoThe error and cause analysis ofChinarsquos typhoon path predictionrdquo Journal of Weather vol 6 pp695ndash700 2012

[25] SMa A Qu and Z Yu ldquoThe parallelization of typhoon numer-ical prediction model of and track forecast error analysisrdquo Jour-nal of Applied Meteorology vol 3 pp 322ndash328 2004 (Chinese)

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ClimatologyJournal of

EcologyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

EarthquakesJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom

Applied ampEnvironmentalSoil Science

Volume 2014

Mining

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014

International Journal of

Geophysics

OceanographyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of Computational Environmental SciencesHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal ofPetroleum Engineering

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

GeochemistryHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of

Atmospheric SciencesInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

OceanographyHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

MineralogyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

MeteorologyAdvances in

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Paleontology JournalHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ScientificaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Geological ResearchJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Geology Advances in

4 Advances in Meteorology

Screening out the landing TCs and not landing TCs

Filtrating all the observation points of both landing and not landing TCs entering into each area

Forming the landing criterion of each area using CART algorithm

Obtaining the TCsrsquo records aroundHainan Island

Finding out the region where TCsrsquo centers were located48 (or 72) hours before they landed

Dividing the research region into 5 (or a number of 1 3 7) areas

Figure 6 Flow diagram of forming landing criterions

an example for a convenient statement Other situations ofa comparative experiment are also conducted in Section 33

For each of the five areas all the observation points ofboth landing and not landing TCs which entered into thisarea are filtrated With strength grade latitude longitude airpressure wind speed latitude migration velocity and longi-tude migration velocity as classification properties the TClanding criterion of each area is formed by using CART algo-rithm [17 18]The flow diagram of forming landing criterionsis shown in Figure 6

For TCs which are judged to be landing on HainanIsland PRESS and its progressive optimal algorithm andMLRM can be used to forecast TCsrsquo characteristic factors(including latitude longitude the lowest center pressureand wind speed) Forecasts in this paper contain landingprediction pattern and dynamic prediction pattern Landingprediction pattern is defined as employing the observationpoint information of those points which first enter into anyarea to predict the characteristic factors when TC landsDynamic prediction pattern is defined as 24 hoursrsquo and 48hoursrsquo prediction with respect to the observation point whichenters into any area The flow diagrams of landing forecastpattern and dynamic forecast pattern are shown in Figures 7and 8 respectively Here PRESS [19] and its progressiveoptimal algorithm [20 21] are used to select the best forecastfactor set from seven predictors which will be used toforecast corresponding characteristic factor MLRM [22] isused to establish corresponding forecast equations MLRM isexpressed as [23]

119910

119894= 120573

0+ 120573

1119909

1198941+ 120573

2119909

1198942+ sdot sdot sdot + 120573

119898119909

119894119898+ 120576

119894

(119894 = 1 2 sdot sdot sdot 119899)

(4)

where119910119894is the estimated value120573

0sim 120573

119898is the regression coef-

ficients 120576119894is the random error and 119909

1198941sim 119909

119894119898are the forecast

factors of the observation point

Table 1 The geometric centers and scopes of five areas

Area number LongitudeUnit ∘E

LatitudeUnit ∘N

RadiusUnit degree

1 1184511 152845 279022 1131435 149623 317063 1169517 208023 330484 1239859 144951 270985 1226830 185854 28346

Table 2 The number of OPs in five areas

Area numberNumber oflanding(1198851

)

Number of notlanding(1198852

)

The ratio oflanding

119885

1

(119885

1

+ 119885

2

)

1 537 1031 34252 855 1389 38103 788 1514 34234 378 1070 26105 340 1419 1933

3 Results and Discussions

In this section the situation in which the region where TCsrsquocenters were located 48 hours before they landed is selectedas research object and the research object is divided into fiveareas is firstly researched Other situations of a comparativeexperiment in which the research object may be the regionwhere TCsrsquo centers were located 72 hours before they landedand the number of areas of divided research object may beany number of 1 3 5 7 are also discussed at the end of thesection

31 Dividing the Research Region into Five Areas 119870-meansclustering algorithm is used to divide our research region intofive areas as described in Section 22 of which the geometriccenters and scopes are shown in Table 1 With respect to eacharea all the observation points of both landing and not land-ing TCs which entered into this area are filtrated The posi-tions of these observation points are shown in Figure 9 andare used to form TCsrsquo landing criterions The numbers ofthese observation points (OPs) for both landing and notlanding TCs are shown in Table 2 The division of five areasfurther narrows the research scope and makes the selectionof the observation points more pertinent so as to form theeffective landing criterions which will be illustrated furtherin Section 33

32 The Formation of Landing Criterions in Five AreasAccording to the CART algorithm the landing criterionsin five areas are shown in Figure 10 (refer to Section 21for the meaning of seven predictors) The correspondingprobability of accurate judgment (119875AJ) probability of falsealarm (119875FA) and probability of false dismissal (119875FD) are shown

Advances in Meteorology 5

Finding out the area in which the observation point first enters into

Is the TC landing on Hainan Island

Using the corresponding landing criterion

Need not forecast

Screening out all the observation points of historical landing TCs when they firstly

entered into the corresponding area

Finding out the observation information of these observation points when they landed

Using PRESS and CART to establish corresponding forecast equation for each

characteristic factor

Obtaining four characteristic factors when forecasted TC lands

on Hainan Island

Yes

No

Figure 7 Flow diagram of landing prediction pattern

Finding out the area in which the observation point first enters into

Is the TC landing onHainan Island

Using the landing criterion of

corresponding area

Need not forecast

Screening out all the observation points of

historical landing TCs when they firstly entered into the

corresponding area

Finding out the observation information of these

observation points 24 (or 48) hours later

Using PRESS and CART to establish corresponding forecast equation for each

characteristic factor

Finding out the area in which the observation point appears (it is not

necessary that the observation point enters into this area for the first time)

Screening out all the observation points of

historical landing TCs when they appeared in the corresponding area

1 Forecast model 12 Forecast model 2

Finding out the observation information of these

observation points 24 (or 48) hours later

Using PRESS and CART to establish corresponding

forecast equation for each characteristic factor

Obtaining four characteristic factors of the TC 24 (or 48)

hours later

Yes

No

12

Figure 8 Flow diagram of dynamic prediction pattern

in Table 3 Set the numbers ofOPs for landing and not landingTCs in any area to be 119885

1and 119885

2 respectively the number

of OPs which are judged to be landing according to landing

criterions when they landed truly is denoted as 1198721and the

number of OPs which are judged to be not landing accordingto landing criterions when they did not land truly is denoted

6 Advances in Meteorology

110 115 120 12510

12

14

16

18

20

22

24

26

The positions of OPs for landing and not landing TCs in area 1

OPs of not landing TCsOPs of landing TCsHainan Island

The positions when TCs landedThe boundary of area 1

OPs of not landing TCsOPs of landing TCsHainan Island

The positions when TCs landedThe boundary of area 2

OPs of not landing TCsOPs of landing TCsHainan Island

The positions when TCs landedThe boundary of area 3

OPs of not landing TCsOPs of landing TCsHainan Island

The positions when TCs landedThe boundary of area 4

OPs of not landing TCsOPs of landing TCsHainan Island

The positions when TCs landedThe boundary of area 5

110 115 120 12510

12

14

16

18

20

22

24

26

The positions of OPs for landing and not landing TCs in area 5

Longitude (∘E)

Longitude (∘E)

110 115 120 12510

12

14

16

18

20

22

24

26

The positions of OPs for landing and not landing TCs in area 2

Longitude (∘E)

Latit

ude (

∘N

)

Latit

ude (

∘N

)

Latit

ude (

∘N

)

110 115 120 12510

12

14

16

18

20

22

24

26

The positions of OPs for landing and not landing TCs in area 3

Longitude (∘E)

Latit

ude (

∘N

)

110 115 120 12510

12

14

16

18

20

22

24

26

The positions of OPs for landing and not landing TCs in area 4

Longitude (∘E)

Latit

ude (

∘N

)

Figure 9 The positions of OPs for landing and not landing TCs which entered into each area

Advances in Meteorology 7

Calculate the seven predictors

of any OP in area 1

Calculate the seven predictors

of any OP in area 2

LONMVYes No

LAT

LONMV

LONMV

LAT

LON

No

NO

Yes

Yes

No

No

No

Yes

Yes

Yes

Landing

Landing

Landing

Not landing

Not landing Not

landingNot

landing

Not landing

Not landing

Not landing

Not landing

Not landing

Not landing Not

landing

Not landing

Not landing

Not landing

Not landing

Not landing

Not landing

LATYes

LONMVLAT

WS

AP

No

Yes

Yes

Yes

No

Yes

No

No

No

Landing

Landing

Calculate the seven predictors

of any OP in area 3

LONMV

LAT

LON

LONMV

LATMV

Landing

Landing

Landing

Yes

Yes

No

No

Yes

Yes

Yes No

No

No

Calculate the seven predictors

of any OP in area 4

LONMV

LAT

LONMV

SG

AP

YesYes

Yes

Yes

Yes

No

No

No

NoNo

Landing

Landing

Landing

Calculate the seven predictors

of any OP in area 5

LONMV

LONMV

LATMV LAT

LATMV

AP

AP

No

Yes

Yes No

Yes NoYes

NoNo No

No

YesYes

Yes

Landing Landing

le14250

le116850

le minus27918

leminus5358

leminus16795le15950

leminus0892gt14450

le11

le10055

leminus6058

le20650

le116650

le22310

le2317

le minus13357

le12550

le1500

le989000

le minus9666

le minus22048

le10656

le19450le10656

le968500

le1002500

leminus21621

gt14950

Figure 10 The landing criterions in five areas

Table 3 119875AJ 119875FA and 119875FD of each criterion

Area number 119875AJ 119875FA 119875FD

1 7468 1794 39482 7531 1555 39533 7750 1176 43154 6623 3654 25935 7976 1649 3588

as 1198722 Then 119875AJ 119875FA and 119875FD of this area are calculated as

follows

119875AJ =119872

1+119872

2

119885

1+ 119885

2

119875FA =

119885

2minus119872

2

119885

2

119875FD =

119885

1minus119872

1

119885

1

(5)

Table 4 The labels of different situations

Ti Nu Label of corresponding situation48 1 FE148 3 FE348 5 FE548 7 FE772 1 ST172 3 ST372 5 ST572 7 ST7

33 Other Situations as a Comparative Experiment In Sec-tions 31 and 32 the region where TCsrsquo centers were located48 hours before they landed is selected as research object andthe research object is divided into five areas In this sectionother situations as a comparative experiment are researched

8 Advances in Meteorology

Table 5119875AJ119875FA and119875FD for each of the remaining seven situations

(a) 119875AJ 119875FA and 119875FD of each area for FE1

Area number 119875AJ 119875FA 119875FD

1 07884 00696 05279

(b) 119875AJ 119875FA and 119875FD of each area for FE3

Area number 119875AJ 119875FA 119875FD

1 07454 01139 051912 07636 01875 030543 07829 00446 07814

(c) 119875AJ 119875FA and 119875FD of each area for FE7

Area number 119875AJ 119875FA 119875FD

1 08522 00154 084142 07486 01991 039593 07302 01808 041764 07719 01165 043985 07236 00740 067356 07834 01512 029757 07631 0 1

(d) 119875AJ 119875FA and 119875FD of each area for ST1

Area number 119875AJ 119875FA 119875FD

1 08164 00412 06998

(e) 119875AJ 119875FA and 119875FD of each area for ST3

Area number 119875AJ 119875FA 119875FD

1 08320 00677 058772 07396 02062 034623 08270 0 1

(f) 119875AJ 119875FA and 119875FD of each area for ST5

Area number 119875AJ 119875FA 119875FD

1 08284 0 12 07613 0 13 07566 01590 035644 08596 0 15 08281 00678 05813

(g) 119875AJ 119875FA and 119875FD of each area for ST7

Area number 119875AJ 119875FA 119875FD

1 07874 0 12 07020 01664 047233 08799 00118 087774 08358 0 15 08537 0 16 07761 00955 050477 07407 01028 06720

and compared with each other Finally we select the situationwhich produces the best result

In order to distinguish different situations the labels ofthemare denoted inTable 4 where the parameter Ti is used to

Table 6 The Index for each of eight situations

Label of corresponding situation IndexFE1 06750FE3 06659FE5 07026FE7 06625ST1 06297ST3 06572ST5 06391ST7 06231

illustrate the research object is the region where TCsrsquo centersare located Ti hours before they landed onHainan Island Nudenotes the number of areas of divided research object

It can be seen from Table 4 that FE5 is the situation whichhas been researched in Sections 31 and 32 The remainingseven situations are researched as follows using the methodswhich are identical with FE5

The 119875AJ 119875FA and 119875FD of each area for each of the remain-ing seven situations can be calculated according to formula(5) the results of which are shown in Table 5

In order to select the best of these eight situations inTable 4 an evaluation method is introduced with which theIndex of each situation is calculated where Index is definedaccording to formula (6) For any situation 119873 denotes thenumber of areas of divided research object and119875AJ 119894119875FA 119894 and119875FD 119894 denote the 119875AJ 119875FA and 119875FD of the 119894th area respectivelyConsider the following

Index =

1

119873

119873

sum

119894=1

radic[

1

10

119875

2

AJ 119894 +3

10

(1 minus 119875FA 119894)2

+

3

5

(1 minus 119875FD 119894)2

]

(6)

It is obvious that the higher the Index is the better theresult of the landing criterions on the whole is The Indexfor each of eight situations is shown in Table 6 The situationFE5 shows the best result which also illustrates that theresearch scheme in Sections 31 and 32 is better comparedwith other situations Finally situation FE5 is selected to formthe landing criterions

34 Forecast of TCsrsquo Characteristic Factors

341 Landing Prediction Pattern The landing forecast pat-tern is defined as follows obtaining the observation pointinformation (seven predictors) when landing TCsrsquo centersfirst enter into any area which can be used to forecast thecharacteristic factors (LAT LONAP andWS)whenTCs landon Hainan Island The flow diagram is shown in Figure 6Taking area 1 for example the OPs of historical landing TCsrsquocenters when they first entered into area 1 are shown inFigure 11 and the tracks of historical landing TCs whichpassed through area 1 are shown in Figure 12

Dividing the historical landing TCs passing through eacharea into two groups with the same number one group of TCsis used to establish prediction equations and the other group

Advances in Meteorology 9

110 115 120 125

10

12

14

16

18

20

22

24

26

Hainan IslandThe boundary of area 1

The LPs of TCsThe OPs of landing TCs enteringinto area 1 for the first time

Longitude (∘E)

Latit

ude (

∘N

)

Figure 11 The OPs of historical landing TCsrsquo centers when theyfirstly entered into area 1 (LP landing position)

Hainan IslandThe boundary of area 1

The LPs of TCs

110 115 120 1258

10

12

14

16

18

20

22

24

Latit

ude (

∘N

)

Longitude (∘E)

The TCs which passedthrough area 1

Figure 12 The tracks of historical landing TCs which passedthrough area 1

is used to test the accuracy of these equations The results oftesting of these prediction equations for TCs which passedthrough each area are shown in Table 7 Making use of theactual and predicted longitude and latitude of TCsrsquo centersin combination with the formula (3) the calculated meanand standard deviation of GCD in the landing predictionpattern are shown in Figures 13 and 14 respectively Averagingthe results of five areas it can be obtained that the averageof the meanstandard deviation (SD) of GCD is 1446382978740 km In [24] Yu et al analyze the averageGCD error of48 hoursrsquo forecast in the South China Sea which is 2226 kmTherefore the landing prediction pattern proposed in this

1 2 3 4 50

20

40

60

80

100

120

140

160

Area number

The m

ean

of G

CD (k

m)

The mean of GCD for each area in landing prediction pattern

Figure 13 The mean of GCD for each area in landing predictionpattern

1 2 3 4 50

20

40

60

80

100

Area number

The s

tand

ard

devi

atio

n of

GCD

(km

)The standard deviation of GCD for each area in

landing prediction pattern

Figure 14 The standard deviation of GCD for each area in landingprediction pattern

paper shows good prediction accuracy For TCs which arejudged to be landing on Hainan Island as long as the obser-vation point information when their centers first enter intoany area are obtained the corresponding forecast equationscan be used to predict characteristic factors when they land

342 Dynamic Prediction Pattern The dynamic predictionpattern is using the current observation point informationto conduct 24 hoursrsquo and 48 hoursrsquo forecast which is alsoillustrated in Figure 8There are two different forecast modelsin dynamic prediction pattern that are described as follows

Forecast Model 1 It is to obtain the current observation pointinformation (seven predictors) when landing TCsrsquo centersenter into any area for the first time making use of which toconduct 24 hoursrsquo and 48 hoursrsquo forecast

10 Advances in Meteorology

1 2 3 4 50

50

100

150

200

Area number

The m

ean

of G

CD (k

m)

The mean of GCD for 24 hoursrsquo forecast in each areausing forecast models 1 and 2

Forecast model 1Forecast model 2

(a) The mean of GCD for 24 hoursrsquo forecast of two forecast models

Forecast model 1Forecast model 2

1 2 3 4 50

20

40

60

80

100

Area number

The s

tand

ard

devi

atio

n of

GCD

(km

)

The standard deviation of GCD for 24 hoursrsquo forecastin each area using forecast models 1 and 2

(b) The standard deviation of GCD for 24 hoursrsquo forecast of two forecastmodels

Figure 15 The meanstandard deviation of GCD for 24 hoursrsquo forecast of two forecast models

Forecast model 1Forecast model 2

1 2 3 4 50

50

100

150

200

250

300

Area number

The m

ean

of G

CD (k

m)

The mean of GCD for 48 hoursrsquo forecast in each areausing forecast models 1 and 2

(a) The mean of GCD for 48 hoursrsquo forecast of two forecast models

Forecast model 1Forecast model 2

1 2 3 4 50

50

100

150

200

Area number

The s

tand

ard

devi

atio

n of

GCD

(km

)

The standard deviation of GCD for 48 hoursrsquo forecastin each area using forecast models 1 and 2

(b) The standard deviation of GCD for 48 hoursrsquo forecast of two forecastmodels

Figure 16 The meanstandard deviation of GCD for 48 hoursrsquo forecast of two forecast models

Forecast Model 2 It is to obtain the current observation pointinformation (seven predictors) when landing TCsrsquo centers arein any area (not necessarily enter into any area for the firsttime)making use ofwhich to conduct 24 hoursrsquo and 48hoursrsquoforecast

In the process of actual prediction for TCs which arejudged to be landing on Hainan Island the observation pointinformationwhen their centers enter into any area for the first

time and the established equations in forecast model 1 areused to conduct dynamic prediction Furthermore the obser-vation point information when TCsrsquo centers are in area (it isnot necessary that TCsrsquo centers enters into this area for thefirst time) any area and the established equations in forecastmodel 2 can be used to conduct dynamic prediction

Similar to Section 341 the historical observation pointsthat meet the corresponding requirements in corresponding

Advances in Meteorology 11

Table 7 Prediction results in landing prediction pattern

Area numberThe meanSD of LAT

deviationUnit ∘N

The meanSD of LATdeviationUnit ∘E

The meanSD of APdeviationUnit hpa

The meanSD of WSdeviationUnit ms

1 0984908671 0726406404 9288967882 59540438392 0526806009 0867708973 7994665339 53036389973 0971007805 0634007711 10694183758 70915526374 1153708640 0715006162 10346361761 68151474985 1115206941 0672807103 12113885744 8040353759

Table 8 The results of testing forecast model 1 and forecast model 2

Forecast hour Area number Forecast model(1 or 2)

The meanSD ofLAT deviation

Unit ∘N

The meanSD ofLON deviation

Unit ∘E

The meanSD ofAP deviationUnit hpa

The meanSD ofWS deviationUnit ms

24 h

1 1 0812806515 1029507600 4927839733 37787288582 0700405153 0880207011 5411443187 3963729843

2 1 0913105832 0986407794 6341244247 38268278572 0753806538 1005008670 4917549673 3848730261

3 1 0755505945 1052809093 7814969316 47342417822 0772706183 0894907255 7348966512 5171843368

4 1 0658205365 0879706251 6637355333 36128264862 0640404911 0950006746 6228056617 3752125645

5 1 0689904989 0861806124 6221863820 29503258232 0699605634 1001707915 6320361131 4186732785

48 h

1 1 1219808343 1773514304 9778257246 68402402892 1259808949 1694112488 9455772439 6503044361

2 1 1225009604 2196016746 10977584764 79032593072 1285710131 2033015552 8143463813 6529343273

3 1 1240707253 2165517786 9658069204 75541529892 1325009565 1841415666 8102472309 6245253772

4 1 0916208120 1761813047 7900955968 57899453712 0989607848 1502011732 7097848208 5282637708

5 1 0965007833 1696014327 9110774706 87411640062 1131510845 1973115516 9244968041 6805748502

forecast model (1 or 2) are divided into two groups with thesame number of observation points One group of observa-tion points is used to establish prediction equations and theother group of points is used to test the accuracies of theseequations The results of testing these prediction equationsin forecast model 1 and forecast model 2 are shown inTable 8 In combinationwith formula (3) the calculatedmeanand standard deviation of GCD in two forecast models areshown in Figures 15 and 16 respectively Averaging theresults of five areas it can be obtained that the averagesof the meanstandard deviation of GCD under forecastmodel 1 and forecast model 2 for 24 hoursrsquo forecast fare1505192846156 km and 1415464812509 km respectivelyFor 48 hoursrsquo forecast the averages of the meanstandarddeviation of GCD under two different forecast modelsare 26175171456345 km and 25671091452903 km respec-tively Even though the mean of GCD in dynamic prediction

pattern is no less than three main prediction centers (NHCJMA and NMCC) it is much less than the numerical pre-diction model in [25] the means of which are 18633195 kmbased on System T106 and 16182958 km based on T213 bothfor 2448 hoursrsquo forecast Besides forecast models 1 and 2 areall more accurate than the forecast using satellite scatterome-terrsquos monitoring data in the sense of the standard deviationof GCD and the mean of weed speed error which are1496002 km and 119618ms in [12] It can be seen fromTable 8 and Figures 15 and 16 that the accuracies of forecastmodel 1 and forecast model 2 vary from different areas anddifferent characteristic factors The more accurate forecastmodel can be selected from forecast models 1 and 2 accordingto actual conditionsThe results of statistical significance testsfor each equation used to forecast corresponding charac-teristic factor in forecast model 1 and forecast model 2 areshown in Table 9 which show that 119875 value is much less than

12 Advances in Meteorology

Table 9 The results of statistical significance tests for each equation used to forecast corresponding characteristic factor in forecast model 1and forecast model 2

Forecast hour Area numberForecastmodel(1 or 2)

119877-square119875 valuefor LAT

119877-square119875 valuefor LON

119877-square119875 valuefor AP

119877-square119875 valuefor WS

24 h

1 1 05883952 times 10minus11 06011791 times 10minus9 06222446 times 10minus10 07615547 times 10minus15

2 04493226 times 10minus32 05809593 times 10minus50 07030131 times 10minus68 07731101 times 10minus81

2 1 07177186 times 10minus14 04391366 times 10minus5 07556503 times 10minus16 07995357 times 10minus18

2 06808517 times 10minus101 06753177 times 10minus99 07481213 times 10minus122 07277371 times 10minus114

3 1 06471636 times 10minus11 06559348 times 10minus11 06310183 times 10minus10 06495324 times 10minus10

2 06802646 times 10minus94 07506119 times 10minus114 05509140 times 10minus62 05573984 times 10minus65

4 1 07572105 times 10minus12 05348332 times 10minus8 07594115 times 10minus13 08174247 times 10minus15

2 06405483 times 10minus42 06266164 times 10minus40 06903101 times 10minus44 07652925 times 10minus57

5 1 05031240 times 10minus6 05154151 times 10minus6 08732273 times 10minus14 08524503 times 10minus15

2 06239284 times 10minus35 06464167 times 10minus37 07835698 times 10minus52 07550152 times 10minus48

48 h

1 1 0133700240 03943391 times 10minus5 03208177 times 10minus4 04725534 times 10minus6

2 0217700049 0039903398 0253200016 04275792 times 10minus6

2 1 04515905 times 10minus6 0266800218 06720172 times 10minus10 04622599 times 10minus6

2 04022164 times 10minus36 03095651 times 10minus26 02960505 times 10minus23 03332627 times 10minus27

3 1 0225100218 03935457 times 10minus5 0002607270 01362015962 03073618 times 10minus25 04369213 times 10minus39 01768251 times 10minus12 02479891 times 10minus18

4 1 04992960 times 10minus7 03055274 times 10minus4 04303153 times 10minus5 06249184 times 10minus9

2 03323377 times 10minus16 03270102 times 10minus16 02471170 times 10minus9 03580365 times 10minus16

5 1 04546619 times 10minus5 0255400043 03705191 times 10minus4 03535000122 04219109 times 10minus18 03825344 times 10minus17 02120218 times 10minus7 03322246 times 10minus12

005 in almost every case and prove that the correspondingprediction equation is significant

4 Summary

In this paper the CMA best track datasets from 1949 to 2012are used in combination with data mining technology andstatistical methods to put forward a new methodology toforecast TCsrsquo characteristic factors This methodology canaccurately judge whether TCs land on Hainan Island or notand forecast their characteristic factors (including longitudelatitude the lowest center pressure and wind speed) Theaverage of the probabilities of accurate judgment for landingcriterions is 7470 and the highest accuracy can reach7976 For the forecast of landing TCsrsquo characteristic factorslanding prediction pattern and dynamic prediction patternare proposed which not only can accurately forecast thecharacteristic factors when TCs land but also realize dynam-ically 24 hoursrsquo and 48 hoursrsquo forecast The effect of thelanding prediction pattern is better of which the mean ofGCD is 1446382 km compared with the current 48 hoursrsquoforecast in the South China Sea which is 2226 km Eventhough the mean of GCD in dynamic prediction pattern isno less than three main prediction centers (NHC JMA andNMCC) it is much less than the numerical prediction modelin [25] and the method using satellite scatterometerrsquos mon-itoring data in [12] The forecast methodology proposed inthis paper provides a new method for typhoon warning on

Hainan Islandwithout getting toomuch knowledge ofmeteo-rology involved and thus simplifies the implementation of theprediction process andmeanwhile guarantees the accuracy ofprediction

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

Acknowledgments

This work is in part supported by the National Science-Technology Support Plan in the domain of advanced energytechnology Hainan Power Grid Corporation provided sup-port for this research through ldquoIntegrated DemonstrationProject of Regional Smart Gridrdquo no 2013BAA01B03 DrXinhong Huang a Research Engineer in the Department ofElectrical amp Computer Engineering from the University ofWestern Ontario also puts forward valuable revision com-ments on this paper

References

[1] National Climate Center of China Typhoon ldquoDamreyrdquo HasCaused Serious Damages to Hainan Province 2005 httpncccmagovcnWebsiteindexphpNewsID=1454

Advances in Meteorology 13

[2] J S Pedro F Burstein and A Sharp ldquoA case-based fuzzy mul-ticriteria decision support model for tropical cyclone forecast-ingrdquo European Journal of Operational Research vol 160 no 2pp 308ndash324 2005

[3] X Qin and M Mu ldquoA Study on the reduction of forecast errorvariance by three adaptive observation approaches for tropicalcyclone predictionrdquoMonthlyWeather Review vol 139 no 7 pp2218ndash2232 2011

[4] Y Wang W Zhang and W Fu ldquoBack Propogation(BP)-neuralnetwork for tropical cyclone track forecastrdquo in Proceedings of the19th International Conference on Geoinformatics pp 1ndash4 IEEEShanghai China June 2011

[5] B Feng and J N K Liu ldquoAn adaptive neural network classifierfor tropical cyclone prediction using a two-layer feature selec-torrdquo in Advances in Neural NetworksmdashISNN 2005 vol 3497of Lecture Notes in Computer Science pp 399ndash404 SpringerBerlin Germany 2005

[6] H-J Song S-H Huh J-H Kim C-H Ho and S-K ParkldquoTyphoon track prediction by a support vector machine usingdata reduction methodsrdquo in Computational Intelligence andSecurity vol 3801 of Lecture Notes in Computer Science pp 503ndash511 Springer Berlin Germany 2005

[7] M DeMaria ldquoA simplified dynamical system for tropicalcyclone intensity predictionrdquoMonthly Weather Review vol 137no 1 pp 68ndash82 2009

[8] H R Winterbottom E W Uhlhorn and E P Chassignet ldquoAdesign and an application of a regional coupled atmosphere-ocean model for tropical cyclone predictionrdquo Journal ofAdvances in Modeling Earth Systems vol 4 no 10 Article IDM10002 2012

[9] L M Ma and Z M Tan ldquoImproving the behavior of the cumu-lus parameterization for tropical cyclone prediction convectiontriggerrdquo Atmospheric Research vol 92 no 2 pp 190ndash211 2009

[10] J S Gall I Ginis S-J Lin T P Marchok and J-H ChenldquoExperimental tropical cyclone prediction using the GFDL 25-km-resolution global atmospheric modelrdquo Weather and Fore-casting vol 26 no 6 pp 1008ndash1019 2011

[11] G Lv ldquoThe Northwestern Pacific typhoon track forecast in2005rdquo in Proceedings of the Meteorological Science and Technol-ogy Symposium on Taiwan Strait p 7 Chinese MeteorologicalSociety 2006 (Chinese)

[12] D Zhang Y Zhang T Hu B Xie and J Xu ldquoA comparison ofHY-2 and QuikSCAT vector wind products for tropical cyclonetrack and intensity development monitoringrdquo IEEE Geoscienceand Remote Sensing Letters vol 11 no 8 pp 1365ndash1369 2014

[13] MYingWZhangHYu et al ldquoAnoverviewof theChinamete-orological administration tropical cyclone databaserdquo Journal ofAtmospheric and Oceanic Technology vol 31 no 2 pp 287ndash3012014

[14] G E Forsythe ldquoGeneration and use of orthogonal polynomialsfor data-fittingwith a digital computerrdquo Journal of the Society forIndustrial amp Applied Mathematics vol 5 no 2 pp 74ndash88 1957

[15] KWagstaff C Cardie S Rogers and S Schroedl ldquoConstrainedk-means clustering with background knowledgerdquo in Proceed-ings of the 18th International Conference on Machine Learningvol 1 pp 577ndash584 2001

[16] Y Ye ldquoNeighborhood density method for selecting initialcluster centers in K-mean clusteringrdquo in Proceedings of theWorkshop on Data Mining for Biomedical Applications (PAKDDrsquo06) pp 189ndash198 2006

[17] P Xiong Data Mining Algorithm and Clementine PracticeTsinghua University Press Beijing China 2011 (Chinese)

[18] S L Crawford ldquoExtensions to the CART algorithmrdquo Interna-tional Journal ofMan-Machine Studies vol 31 no 2 pp 197ndash2171989

[19] D M Allen ldquoMean square error of prediction as a criterion forselecting variablesrdquo Technometrics vol 13 pp 469ndash475 1971

[20] S Yu and J Shen ldquoForward and backward algorithms forselecting predictors on the basis of the criterion frompredictionsumof squares and their applicationrdquoActaMeteorologica Sinicavol 1 pp 83ndash90 1988

[21] D Yao and S Yu ldquoThe stepwise algorithm of selecting forecastfactors based on PRESS rulerdquo Journal of Atmospheric Sciencesvol 2 pp 129ndash135 1992 (Chinese)

[22] Y LuMathematical Statistics Methods East China University ofScience and Technology Press Shanghai China 2005 (Chi-nese)

[23] K Xie and B Liu ldquoAn ENSO-forecast independent statisticalmodel for the prediction of annual Atlantic tropical cyclonefrequency in Aprilrdquo Advances in Meteorology vol 2014 ArticleID 248148 11 pages 2014

[24] J Yu J Tang Y Dai and B Yu ldquoThe error and cause analysis ofChinarsquos typhoon path predictionrdquo Journal of Weather vol 6 pp695ndash700 2012

[25] SMa A Qu and Z Yu ldquoThe parallelization of typhoon numer-ical prediction model of and track forecast error analysisrdquo Jour-nal of Applied Meteorology vol 3 pp 322ndash328 2004 (Chinese)

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ClimatologyJournal of

EcologyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

EarthquakesJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom

Applied ampEnvironmentalSoil Science

Volume 2014

Mining

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014

International Journal of

Geophysics

OceanographyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of Computational Environmental SciencesHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal ofPetroleum Engineering

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

GeochemistryHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of

Atmospheric SciencesInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

OceanographyHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

MineralogyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

MeteorologyAdvances in

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Paleontology JournalHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ScientificaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Geological ResearchJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Geology Advances in

Advances in Meteorology 5

Finding out the area in which the observation point first enters into

Is the TC landing on Hainan Island

Using the corresponding landing criterion

Need not forecast

Screening out all the observation points of historical landing TCs when they firstly

entered into the corresponding area

Finding out the observation information of these observation points when they landed

Using PRESS and CART to establish corresponding forecast equation for each

characteristic factor

Obtaining four characteristic factors when forecasted TC lands

on Hainan Island

Yes

No

Figure 7 Flow diagram of landing prediction pattern

Finding out the area in which the observation point first enters into

Is the TC landing onHainan Island

Using the landing criterion of

corresponding area

Need not forecast

Screening out all the observation points of

historical landing TCs when they firstly entered into the

corresponding area

Finding out the observation information of these

observation points 24 (or 48) hours later

Using PRESS and CART to establish corresponding forecast equation for each

characteristic factor

Finding out the area in which the observation point appears (it is not

necessary that the observation point enters into this area for the first time)

Screening out all the observation points of

historical landing TCs when they appeared in the corresponding area

1 Forecast model 12 Forecast model 2

Finding out the observation information of these

observation points 24 (or 48) hours later

Using PRESS and CART to establish corresponding

forecast equation for each characteristic factor

Obtaining four characteristic factors of the TC 24 (or 48)

hours later

Yes

No

12

Figure 8 Flow diagram of dynamic prediction pattern

in Table 3 Set the numbers ofOPs for landing and not landingTCs in any area to be 119885

1and 119885

2 respectively the number

of OPs which are judged to be landing according to landing

criterions when they landed truly is denoted as 1198721and the

number of OPs which are judged to be not landing accordingto landing criterions when they did not land truly is denoted

6 Advances in Meteorology

110 115 120 12510

12

14

16

18

20

22

24

26

The positions of OPs for landing and not landing TCs in area 1

OPs of not landing TCsOPs of landing TCsHainan Island

The positions when TCs landedThe boundary of area 1

OPs of not landing TCsOPs of landing TCsHainan Island

The positions when TCs landedThe boundary of area 2

OPs of not landing TCsOPs of landing TCsHainan Island

The positions when TCs landedThe boundary of area 3

OPs of not landing TCsOPs of landing TCsHainan Island

The positions when TCs landedThe boundary of area 4

OPs of not landing TCsOPs of landing TCsHainan Island

The positions when TCs landedThe boundary of area 5

110 115 120 12510

12

14

16

18

20

22

24

26

The positions of OPs for landing and not landing TCs in area 5

Longitude (∘E)

Longitude (∘E)

110 115 120 12510

12

14

16

18

20

22

24

26

The positions of OPs for landing and not landing TCs in area 2

Longitude (∘E)

Latit

ude (

∘N

)

Latit

ude (

∘N

)

Latit

ude (

∘N

)

110 115 120 12510

12

14

16

18

20

22

24

26

The positions of OPs for landing and not landing TCs in area 3

Longitude (∘E)

Latit

ude (

∘N

)

110 115 120 12510

12

14

16

18

20

22

24

26

The positions of OPs for landing and not landing TCs in area 4

Longitude (∘E)

Latit

ude (

∘N

)

Figure 9 The positions of OPs for landing and not landing TCs which entered into each area

Advances in Meteorology 7

Calculate the seven predictors

of any OP in area 1

Calculate the seven predictors

of any OP in area 2

LONMVYes No

LAT

LONMV

LONMV

LAT

LON

No

NO

Yes

Yes

No

No

No

Yes

Yes

Yes

Landing

Landing

Landing

Not landing

Not landing Not

landingNot

landing

Not landing

Not landing

Not landing

Not landing

Not landing

Not landing Not

landing

Not landing

Not landing

Not landing

Not landing

Not landing

Not landing

LATYes

LONMVLAT

WS

AP

No

Yes

Yes

Yes

No

Yes

No

No

No

Landing

Landing

Calculate the seven predictors

of any OP in area 3

LONMV

LAT

LON

LONMV

LATMV

Landing

Landing

Landing

Yes

Yes

No

No

Yes

Yes

Yes No

No

No

Calculate the seven predictors

of any OP in area 4

LONMV

LAT

LONMV

SG

AP

YesYes

Yes

Yes

Yes

No

No

No

NoNo

Landing

Landing

Landing

Calculate the seven predictors

of any OP in area 5

LONMV

LONMV

LATMV LAT

LATMV

AP

AP

No

Yes

Yes No

Yes NoYes

NoNo No

No

YesYes

Yes

Landing Landing

le14250

le116850

le minus27918

leminus5358

leminus16795le15950

leminus0892gt14450

le11

le10055

leminus6058

le20650

le116650

le22310

le2317

le minus13357

le12550

le1500

le989000

le minus9666

le minus22048

le10656

le19450le10656

le968500

le1002500

leminus21621

gt14950

Figure 10 The landing criterions in five areas

Table 3 119875AJ 119875FA and 119875FD of each criterion

Area number 119875AJ 119875FA 119875FD

1 7468 1794 39482 7531 1555 39533 7750 1176 43154 6623 3654 25935 7976 1649 3588

as 1198722 Then 119875AJ 119875FA and 119875FD of this area are calculated as

follows

119875AJ =119872

1+119872

2

119885

1+ 119885

2

119875FA =

119885

2minus119872

2

119885

2

119875FD =

119885

1minus119872

1

119885

1

(5)

Table 4 The labels of different situations

Ti Nu Label of corresponding situation48 1 FE148 3 FE348 5 FE548 7 FE772 1 ST172 3 ST372 5 ST572 7 ST7

33 Other Situations as a Comparative Experiment In Sec-tions 31 and 32 the region where TCsrsquo centers were located48 hours before they landed is selected as research object andthe research object is divided into five areas In this sectionother situations as a comparative experiment are researched

8 Advances in Meteorology

Table 5119875AJ119875FA and119875FD for each of the remaining seven situations

(a) 119875AJ 119875FA and 119875FD of each area for FE1

Area number 119875AJ 119875FA 119875FD

1 07884 00696 05279

(b) 119875AJ 119875FA and 119875FD of each area for FE3

Area number 119875AJ 119875FA 119875FD

1 07454 01139 051912 07636 01875 030543 07829 00446 07814

(c) 119875AJ 119875FA and 119875FD of each area for FE7

Area number 119875AJ 119875FA 119875FD

1 08522 00154 084142 07486 01991 039593 07302 01808 041764 07719 01165 043985 07236 00740 067356 07834 01512 029757 07631 0 1

(d) 119875AJ 119875FA and 119875FD of each area for ST1

Area number 119875AJ 119875FA 119875FD

1 08164 00412 06998

(e) 119875AJ 119875FA and 119875FD of each area for ST3

Area number 119875AJ 119875FA 119875FD

1 08320 00677 058772 07396 02062 034623 08270 0 1

(f) 119875AJ 119875FA and 119875FD of each area for ST5

Area number 119875AJ 119875FA 119875FD

1 08284 0 12 07613 0 13 07566 01590 035644 08596 0 15 08281 00678 05813

(g) 119875AJ 119875FA and 119875FD of each area for ST7

Area number 119875AJ 119875FA 119875FD

1 07874 0 12 07020 01664 047233 08799 00118 087774 08358 0 15 08537 0 16 07761 00955 050477 07407 01028 06720

and compared with each other Finally we select the situationwhich produces the best result

In order to distinguish different situations the labels ofthemare denoted inTable 4 where the parameter Ti is used to

Table 6 The Index for each of eight situations

Label of corresponding situation IndexFE1 06750FE3 06659FE5 07026FE7 06625ST1 06297ST3 06572ST5 06391ST7 06231

illustrate the research object is the region where TCsrsquo centersare located Ti hours before they landed onHainan Island Nudenotes the number of areas of divided research object

It can be seen from Table 4 that FE5 is the situation whichhas been researched in Sections 31 and 32 The remainingseven situations are researched as follows using the methodswhich are identical with FE5

The 119875AJ 119875FA and 119875FD of each area for each of the remain-ing seven situations can be calculated according to formula(5) the results of which are shown in Table 5

In order to select the best of these eight situations inTable 4 an evaluation method is introduced with which theIndex of each situation is calculated where Index is definedaccording to formula (6) For any situation 119873 denotes thenumber of areas of divided research object and119875AJ 119894119875FA 119894 and119875FD 119894 denote the 119875AJ 119875FA and 119875FD of the 119894th area respectivelyConsider the following

Index =

1

119873

119873

sum

119894=1

radic[

1

10

119875

2

AJ 119894 +3

10

(1 minus 119875FA 119894)2

+

3

5

(1 minus 119875FD 119894)2

]

(6)

It is obvious that the higher the Index is the better theresult of the landing criterions on the whole is The Indexfor each of eight situations is shown in Table 6 The situationFE5 shows the best result which also illustrates that theresearch scheme in Sections 31 and 32 is better comparedwith other situations Finally situation FE5 is selected to formthe landing criterions

34 Forecast of TCsrsquo Characteristic Factors

341 Landing Prediction Pattern The landing forecast pat-tern is defined as follows obtaining the observation pointinformation (seven predictors) when landing TCsrsquo centersfirst enter into any area which can be used to forecast thecharacteristic factors (LAT LONAP andWS)whenTCs landon Hainan Island The flow diagram is shown in Figure 6Taking area 1 for example the OPs of historical landing TCsrsquocenters when they first entered into area 1 are shown inFigure 11 and the tracks of historical landing TCs whichpassed through area 1 are shown in Figure 12

Dividing the historical landing TCs passing through eacharea into two groups with the same number one group of TCsis used to establish prediction equations and the other group

Advances in Meteorology 9

110 115 120 125

10

12

14

16

18

20

22

24

26

Hainan IslandThe boundary of area 1

The LPs of TCsThe OPs of landing TCs enteringinto area 1 for the first time

Longitude (∘E)

Latit

ude (

∘N

)

Figure 11 The OPs of historical landing TCsrsquo centers when theyfirstly entered into area 1 (LP landing position)

Hainan IslandThe boundary of area 1

The LPs of TCs

110 115 120 1258

10

12

14

16

18

20

22

24

Latit

ude (

∘N

)

Longitude (∘E)

The TCs which passedthrough area 1

Figure 12 The tracks of historical landing TCs which passedthrough area 1

is used to test the accuracy of these equations The results oftesting of these prediction equations for TCs which passedthrough each area are shown in Table 7 Making use of theactual and predicted longitude and latitude of TCsrsquo centersin combination with the formula (3) the calculated meanand standard deviation of GCD in the landing predictionpattern are shown in Figures 13 and 14 respectively Averagingthe results of five areas it can be obtained that the averageof the meanstandard deviation (SD) of GCD is 1446382978740 km In [24] Yu et al analyze the averageGCD error of48 hoursrsquo forecast in the South China Sea which is 2226 kmTherefore the landing prediction pattern proposed in this

1 2 3 4 50

20

40

60

80

100

120

140

160

Area number

The m

ean

of G

CD (k

m)

The mean of GCD for each area in landing prediction pattern

Figure 13 The mean of GCD for each area in landing predictionpattern

1 2 3 4 50

20

40

60

80

100

Area number

The s

tand

ard

devi

atio

n of

GCD

(km

)The standard deviation of GCD for each area in

landing prediction pattern

Figure 14 The standard deviation of GCD for each area in landingprediction pattern

paper shows good prediction accuracy For TCs which arejudged to be landing on Hainan Island as long as the obser-vation point information when their centers first enter intoany area are obtained the corresponding forecast equationscan be used to predict characteristic factors when they land

342 Dynamic Prediction Pattern The dynamic predictionpattern is using the current observation point informationto conduct 24 hoursrsquo and 48 hoursrsquo forecast which is alsoillustrated in Figure 8There are two different forecast modelsin dynamic prediction pattern that are described as follows

Forecast Model 1 It is to obtain the current observation pointinformation (seven predictors) when landing TCsrsquo centersenter into any area for the first time making use of which toconduct 24 hoursrsquo and 48 hoursrsquo forecast

10 Advances in Meteorology

1 2 3 4 50

50

100

150

200

Area number

The m

ean

of G

CD (k

m)

The mean of GCD for 24 hoursrsquo forecast in each areausing forecast models 1 and 2

Forecast model 1Forecast model 2

(a) The mean of GCD for 24 hoursrsquo forecast of two forecast models

Forecast model 1Forecast model 2

1 2 3 4 50

20

40

60

80

100

Area number

The s

tand

ard

devi

atio

n of

GCD

(km

)

The standard deviation of GCD for 24 hoursrsquo forecastin each area using forecast models 1 and 2

(b) The standard deviation of GCD for 24 hoursrsquo forecast of two forecastmodels

Figure 15 The meanstandard deviation of GCD for 24 hoursrsquo forecast of two forecast models

Forecast model 1Forecast model 2

1 2 3 4 50

50

100

150

200

250

300

Area number

The m

ean

of G

CD (k

m)

The mean of GCD for 48 hoursrsquo forecast in each areausing forecast models 1 and 2

(a) The mean of GCD for 48 hoursrsquo forecast of two forecast models

Forecast model 1Forecast model 2

1 2 3 4 50

50

100

150

200

Area number

The s

tand

ard

devi

atio

n of

GCD

(km

)

The standard deviation of GCD for 48 hoursrsquo forecastin each area using forecast models 1 and 2

(b) The standard deviation of GCD for 48 hoursrsquo forecast of two forecastmodels

Figure 16 The meanstandard deviation of GCD for 48 hoursrsquo forecast of two forecast models

Forecast Model 2 It is to obtain the current observation pointinformation (seven predictors) when landing TCsrsquo centers arein any area (not necessarily enter into any area for the firsttime)making use ofwhich to conduct 24 hoursrsquo and 48hoursrsquoforecast

In the process of actual prediction for TCs which arejudged to be landing on Hainan Island the observation pointinformationwhen their centers enter into any area for the first

time and the established equations in forecast model 1 areused to conduct dynamic prediction Furthermore the obser-vation point information when TCsrsquo centers are in area (it isnot necessary that TCsrsquo centers enters into this area for thefirst time) any area and the established equations in forecastmodel 2 can be used to conduct dynamic prediction

Similar to Section 341 the historical observation pointsthat meet the corresponding requirements in corresponding

Advances in Meteorology 11

Table 7 Prediction results in landing prediction pattern

Area numberThe meanSD of LAT

deviationUnit ∘N

The meanSD of LATdeviationUnit ∘E

The meanSD of APdeviationUnit hpa

The meanSD of WSdeviationUnit ms

1 0984908671 0726406404 9288967882 59540438392 0526806009 0867708973 7994665339 53036389973 0971007805 0634007711 10694183758 70915526374 1153708640 0715006162 10346361761 68151474985 1115206941 0672807103 12113885744 8040353759

Table 8 The results of testing forecast model 1 and forecast model 2

Forecast hour Area number Forecast model(1 or 2)

The meanSD ofLAT deviation

Unit ∘N

The meanSD ofLON deviation

Unit ∘E

The meanSD ofAP deviationUnit hpa

The meanSD ofWS deviationUnit ms

24 h

1 1 0812806515 1029507600 4927839733 37787288582 0700405153 0880207011 5411443187 3963729843

2 1 0913105832 0986407794 6341244247 38268278572 0753806538 1005008670 4917549673 3848730261

3 1 0755505945 1052809093 7814969316 47342417822 0772706183 0894907255 7348966512 5171843368

4 1 0658205365 0879706251 6637355333 36128264862 0640404911 0950006746 6228056617 3752125645

5 1 0689904989 0861806124 6221863820 29503258232 0699605634 1001707915 6320361131 4186732785

48 h

1 1 1219808343 1773514304 9778257246 68402402892 1259808949 1694112488 9455772439 6503044361

2 1 1225009604 2196016746 10977584764 79032593072 1285710131 2033015552 8143463813 6529343273

3 1 1240707253 2165517786 9658069204 75541529892 1325009565 1841415666 8102472309 6245253772

4 1 0916208120 1761813047 7900955968 57899453712 0989607848 1502011732 7097848208 5282637708

5 1 0965007833 1696014327 9110774706 87411640062 1131510845 1973115516 9244968041 6805748502

forecast model (1 or 2) are divided into two groups with thesame number of observation points One group of observa-tion points is used to establish prediction equations and theother group of points is used to test the accuracies of theseequations The results of testing these prediction equationsin forecast model 1 and forecast model 2 are shown inTable 8 In combinationwith formula (3) the calculatedmeanand standard deviation of GCD in two forecast models areshown in Figures 15 and 16 respectively Averaging theresults of five areas it can be obtained that the averagesof the meanstandard deviation of GCD under forecastmodel 1 and forecast model 2 for 24 hoursrsquo forecast fare1505192846156 km and 1415464812509 km respectivelyFor 48 hoursrsquo forecast the averages of the meanstandarddeviation of GCD under two different forecast modelsare 26175171456345 km and 25671091452903 km respec-tively Even though the mean of GCD in dynamic prediction

pattern is no less than three main prediction centers (NHCJMA and NMCC) it is much less than the numerical pre-diction model in [25] the means of which are 18633195 kmbased on System T106 and 16182958 km based on T213 bothfor 2448 hoursrsquo forecast Besides forecast models 1 and 2 areall more accurate than the forecast using satellite scatterome-terrsquos monitoring data in the sense of the standard deviationof GCD and the mean of weed speed error which are1496002 km and 119618ms in [12] It can be seen fromTable 8 and Figures 15 and 16 that the accuracies of forecastmodel 1 and forecast model 2 vary from different areas anddifferent characteristic factors The more accurate forecastmodel can be selected from forecast models 1 and 2 accordingto actual conditionsThe results of statistical significance testsfor each equation used to forecast corresponding charac-teristic factor in forecast model 1 and forecast model 2 areshown in Table 9 which show that 119875 value is much less than

12 Advances in Meteorology

Table 9 The results of statistical significance tests for each equation used to forecast corresponding characteristic factor in forecast model 1and forecast model 2

Forecast hour Area numberForecastmodel(1 or 2)

119877-square119875 valuefor LAT

119877-square119875 valuefor LON

119877-square119875 valuefor AP

119877-square119875 valuefor WS

24 h

1 1 05883952 times 10minus11 06011791 times 10minus9 06222446 times 10minus10 07615547 times 10minus15

2 04493226 times 10minus32 05809593 times 10minus50 07030131 times 10minus68 07731101 times 10minus81

2 1 07177186 times 10minus14 04391366 times 10minus5 07556503 times 10minus16 07995357 times 10minus18

2 06808517 times 10minus101 06753177 times 10minus99 07481213 times 10minus122 07277371 times 10minus114

3 1 06471636 times 10minus11 06559348 times 10minus11 06310183 times 10minus10 06495324 times 10minus10

2 06802646 times 10minus94 07506119 times 10minus114 05509140 times 10minus62 05573984 times 10minus65

4 1 07572105 times 10minus12 05348332 times 10minus8 07594115 times 10minus13 08174247 times 10minus15

2 06405483 times 10minus42 06266164 times 10minus40 06903101 times 10minus44 07652925 times 10minus57

5 1 05031240 times 10minus6 05154151 times 10minus6 08732273 times 10minus14 08524503 times 10minus15

2 06239284 times 10minus35 06464167 times 10minus37 07835698 times 10minus52 07550152 times 10minus48

48 h

1 1 0133700240 03943391 times 10minus5 03208177 times 10minus4 04725534 times 10minus6

2 0217700049 0039903398 0253200016 04275792 times 10minus6

2 1 04515905 times 10minus6 0266800218 06720172 times 10minus10 04622599 times 10minus6

2 04022164 times 10minus36 03095651 times 10minus26 02960505 times 10minus23 03332627 times 10minus27

3 1 0225100218 03935457 times 10minus5 0002607270 01362015962 03073618 times 10minus25 04369213 times 10minus39 01768251 times 10minus12 02479891 times 10minus18

4 1 04992960 times 10minus7 03055274 times 10minus4 04303153 times 10minus5 06249184 times 10minus9

2 03323377 times 10minus16 03270102 times 10minus16 02471170 times 10minus9 03580365 times 10minus16

5 1 04546619 times 10minus5 0255400043 03705191 times 10minus4 03535000122 04219109 times 10minus18 03825344 times 10minus17 02120218 times 10minus7 03322246 times 10minus12

005 in almost every case and prove that the correspondingprediction equation is significant

4 Summary

In this paper the CMA best track datasets from 1949 to 2012are used in combination with data mining technology andstatistical methods to put forward a new methodology toforecast TCsrsquo characteristic factors This methodology canaccurately judge whether TCs land on Hainan Island or notand forecast their characteristic factors (including longitudelatitude the lowest center pressure and wind speed) Theaverage of the probabilities of accurate judgment for landingcriterions is 7470 and the highest accuracy can reach7976 For the forecast of landing TCsrsquo characteristic factorslanding prediction pattern and dynamic prediction patternare proposed which not only can accurately forecast thecharacteristic factors when TCs land but also realize dynam-ically 24 hoursrsquo and 48 hoursrsquo forecast The effect of thelanding prediction pattern is better of which the mean ofGCD is 1446382 km compared with the current 48 hoursrsquoforecast in the South China Sea which is 2226 km Eventhough the mean of GCD in dynamic prediction pattern isno less than three main prediction centers (NHC JMA andNMCC) it is much less than the numerical prediction modelin [25] and the method using satellite scatterometerrsquos mon-itoring data in [12] The forecast methodology proposed inthis paper provides a new method for typhoon warning on

Hainan Islandwithout getting toomuch knowledge ofmeteo-rology involved and thus simplifies the implementation of theprediction process andmeanwhile guarantees the accuracy ofprediction

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

Acknowledgments

This work is in part supported by the National Science-Technology Support Plan in the domain of advanced energytechnology Hainan Power Grid Corporation provided sup-port for this research through ldquoIntegrated DemonstrationProject of Regional Smart Gridrdquo no 2013BAA01B03 DrXinhong Huang a Research Engineer in the Department ofElectrical amp Computer Engineering from the University ofWestern Ontario also puts forward valuable revision com-ments on this paper

References

[1] National Climate Center of China Typhoon ldquoDamreyrdquo HasCaused Serious Damages to Hainan Province 2005 httpncccmagovcnWebsiteindexphpNewsID=1454

Advances in Meteorology 13

[2] J S Pedro F Burstein and A Sharp ldquoA case-based fuzzy mul-ticriteria decision support model for tropical cyclone forecast-ingrdquo European Journal of Operational Research vol 160 no 2pp 308ndash324 2005

[3] X Qin and M Mu ldquoA Study on the reduction of forecast errorvariance by three adaptive observation approaches for tropicalcyclone predictionrdquoMonthlyWeather Review vol 139 no 7 pp2218ndash2232 2011

[4] Y Wang W Zhang and W Fu ldquoBack Propogation(BP)-neuralnetwork for tropical cyclone track forecastrdquo in Proceedings of the19th International Conference on Geoinformatics pp 1ndash4 IEEEShanghai China June 2011

[5] B Feng and J N K Liu ldquoAn adaptive neural network classifierfor tropical cyclone prediction using a two-layer feature selec-torrdquo in Advances in Neural NetworksmdashISNN 2005 vol 3497of Lecture Notes in Computer Science pp 399ndash404 SpringerBerlin Germany 2005

[6] H-J Song S-H Huh J-H Kim C-H Ho and S-K ParkldquoTyphoon track prediction by a support vector machine usingdata reduction methodsrdquo in Computational Intelligence andSecurity vol 3801 of Lecture Notes in Computer Science pp 503ndash511 Springer Berlin Germany 2005

[7] M DeMaria ldquoA simplified dynamical system for tropicalcyclone intensity predictionrdquoMonthly Weather Review vol 137no 1 pp 68ndash82 2009

[8] H R Winterbottom E W Uhlhorn and E P Chassignet ldquoAdesign and an application of a regional coupled atmosphere-ocean model for tropical cyclone predictionrdquo Journal ofAdvances in Modeling Earth Systems vol 4 no 10 Article IDM10002 2012

[9] L M Ma and Z M Tan ldquoImproving the behavior of the cumu-lus parameterization for tropical cyclone prediction convectiontriggerrdquo Atmospheric Research vol 92 no 2 pp 190ndash211 2009

[10] J S Gall I Ginis S-J Lin T P Marchok and J-H ChenldquoExperimental tropical cyclone prediction using the GFDL 25-km-resolution global atmospheric modelrdquo Weather and Fore-casting vol 26 no 6 pp 1008ndash1019 2011

[11] G Lv ldquoThe Northwestern Pacific typhoon track forecast in2005rdquo in Proceedings of the Meteorological Science and Technol-ogy Symposium on Taiwan Strait p 7 Chinese MeteorologicalSociety 2006 (Chinese)

[12] D Zhang Y Zhang T Hu B Xie and J Xu ldquoA comparison ofHY-2 and QuikSCAT vector wind products for tropical cyclonetrack and intensity development monitoringrdquo IEEE Geoscienceand Remote Sensing Letters vol 11 no 8 pp 1365ndash1369 2014

[13] MYingWZhangHYu et al ldquoAnoverviewof theChinamete-orological administration tropical cyclone databaserdquo Journal ofAtmospheric and Oceanic Technology vol 31 no 2 pp 287ndash3012014

[14] G E Forsythe ldquoGeneration and use of orthogonal polynomialsfor data-fittingwith a digital computerrdquo Journal of the Society forIndustrial amp Applied Mathematics vol 5 no 2 pp 74ndash88 1957

[15] KWagstaff C Cardie S Rogers and S Schroedl ldquoConstrainedk-means clustering with background knowledgerdquo in Proceed-ings of the 18th International Conference on Machine Learningvol 1 pp 577ndash584 2001

[16] Y Ye ldquoNeighborhood density method for selecting initialcluster centers in K-mean clusteringrdquo in Proceedings of theWorkshop on Data Mining for Biomedical Applications (PAKDDrsquo06) pp 189ndash198 2006

[17] P Xiong Data Mining Algorithm and Clementine PracticeTsinghua University Press Beijing China 2011 (Chinese)

[18] S L Crawford ldquoExtensions to the CART algorithmrdquo Interna-tional Journal ofMan-Machine Studies vol 31 no 2 pp 197ndash2171989

[19] D M Allen ldquoMean square error of prediction as a criterion forselecting variablesrdquo Technometrics vol 13 pp 469ndash475 1971

[20] S Yu and J Shen ldquoForward and backward algorithms forselecting predictors on the basis of the criterion frompredictionsumof squares and their applicationrdquoActaMeteorologica Sinicavol 1 pp 83ndash90 1988

[21] D Yao and S Yu ldquoThe stepwise algorithm of selecting forecastfactors based on PRESS rulerdquo Journal of Atmospheric Sciencesvol 2 pp 129ndash135 1992 (Chinese)

[22] Y LuMathematical Statistics Methods East China University ofScience and Technology Press Shanghai China 2005 (Chi-nese)

[23] K Xie and B Liu ldquoAn ENSO-forecast independent statisticalmodel for the prediction of annual Atlantic tropical cyclonefrequency in Aprilrdquo Advances in Meteorology vol 2014 ArticleID 248148 11 pages 2014

[24] J Yu J Tang Y Dai and B Yu ldquoThe error and cause analysis ofChinarsquos typhoon path predictionrdquo Journal of Weather vol 6 pp695ndash700 2012

[25] SMa A Qu and Z Yu ldquoThe parallelization of typhoon numer-ical prediction model of and track forecast error analysisrdquo Jour-nal of Applied Meteorology vol 3 pp 322ndash328 2004 (Chinese)

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ClimatologyJournal of

EcologyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

EarthquakesJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom

Applied ampEnvironmentalSoil Science

Volume 2014

Mining

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014

International Journal of

Geophysics

OceanographyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of Computational Environmental SciencesHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal ofPetroleum Engineering

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

GeochemistryHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of

Atmospheric SciencesInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

OceanographyHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

MineralogyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

MeteorologyAdvances in

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Paleontology JournalHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ScientificaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Geological ResearchJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Geology Advances in

6 Advances in Meteorology

110 115 120 12510

12

14

16

18

20

22

24

26

The positions of OPs for landing and not landing TCs in area 1

OPs of not landing TCsOPs of landing TCsHainan Island

The positions when TCs landedThe boundary of area 1

OPs of not landing TCsOPs of landing TCsHainan Island

The positions when TCs landedThe boundary of area 2

OPs of not landing TCsOPs of landing TCsHainan Island

The positions when TCs landedThe boundary of area 3

OPs of not landing TCsOPs of landing TCsHainan Island

The positions when TCs landedThe boundary of area 4

OPs of not landing TCsOPs of landing TCsHainan Island

The positions when TCs landedThe boundary of area 5

110 115 120 12510

12

14

16

18

20

22

24

26

The positions of OPs for landing and not landing TCs in area 5

Longitude (∘E)

Longitude (∘E)

110 115 120 12510

12

14

16

18

20

22

24

26

The positions of OPs for landing and not landing TCs in area 2

Longitude (∘E)

Latit

ude (

∘N

)

Latit

ude (

∘N

)

Latit

ude (

∘N

)

110 115 120 12510

12

14

16

18

20

22

24

26

The positions of OPs for landing and not landing TCs in area 3

Longitude (∘E)

Latit

ude (

∘N

)

110 115 120 12510

12

14

16

18

20

22

24

26

The positions of OPs for landing and not landing TCs in area 4

Longitude (∘E)

Latit

ude (

∘N

)

Figure 9 The positions of OPs for landing and not landing TCs which entered into each area

Advances in Meteorology 7

Calculate the seven predictors

of any OP in area 1

Calculate the seven predictors

of any OP in area 2

LONMVYes No

LAT

LONMV

LONMV

LAT

LON

No

NO

Yes

Yes

No

No

No

Yes

Yes

Yes

Landing

Landing

Landing

Not landing

Not landing Not

landingNot

landing

Not landing

Not landing

Not landing

Not landing

Not landing

Not landing Not

landing

Not landing

Not landing

Not landing

Not landing

Not landing

Not landing

LATYes

LONMVLAT

WS

AP

No

Yes

Yes

Yes

No

Yes

No

No

No

Landing

Landing

Calculate the seven predictors

of any OP in area 3

LONMV

LAT

LON

LONMV

LATMV

Landing

Landing

Landing

Yes

Yes

No

No

Yes

Yes

Yes No

No

No

Calculate the seven predictors

of any OP in area 4

LONMV

LAT

LONMV

SG

AP

YesYes

Yes

Yes

Yes

No

No

No

NoNo

Landing

Landing

Landing

Calculate the seven predictors

of any OP in area 5

LONMV

LONMV

LATMV LAT

LATMV

AP

AP

No

Yes

Yes No

Yes NoYes

NoNo No

No

YesYes

Yes

Landing Landing

le14250

le116850

le minus27918

leminus5358

leminus16795le15950

leminus0892gt14450

le11

le10055

leminus6058

le20650

le116650

le22310

le2317

le minus13357

le12550

le1500

le989000

le minus9666

le minus22048

le10656

le19450le10656

le968500

le1002500

leminus21621

gt14950

Figure 10 The landing criterions in five areas

Table 3 119875AJ 119875FA and 119875FD of each criterion

Area number 119875AJ 119875FA 119875FD

1 7468 1794 39482 7531 1555 39533 7750 1176 43154 6623 3654 25935 7976 1649 3588

as 1198722 Then 119875AJ 119875FA and 119875FD of this area are calculated as

follows

119875AJ =119872

1+119872

2

119885

1+ 119885

2

119875FA =

119885

2minus119872

2

119885

2

119875FD =

119885

1minus119872

1

119885

1

(5)

Table 4 The labels of different situations

Ti Nu Label of corresponding situation48 1 FE148 3 FE348 5 FE548 7 FE772 1 ST172 3 ST372 5 ST572 7 ST7

33 Other Situations as a Comparative Experiment In Sec-tions 31 and 32 the region where TCsrsquo centers were located48 hours before they landed is selected as research object andthe research object is divided into five areas In this sectionother situations as a comparative experiment are researched

8 Advances in Meteorology

Table 5119875AJ119875FA and119875FD for each of the remaining seven situations

(a) 119875AJ 119875FA and 119875FD of each area for FE1

Area number 119875AJ 119875FA 119875FD

1 07884 00696 05279

(b) 119875AJ 119875FA and 119875FD of each area for FE3

Area number 119875AJ 119875FA 119875FD

1 07454 01139 051912 07636 01875 030543 07829 00446 07814

(c) 119875AJ 119875FA and 119875FD of each area for FE7

Area number 119875AJ 119875FA 119875FD

1 08522 00154 084142 07486 01991 039593 07302 01808 041764 07719 01165 043985 07236 00740 067356 07834 01512 029757 07631 0 1

(d) 119875AJ 119875FA and 119875FD of each area for ST1

Area number 119875AJ 119875FA 119875FD

1 08164 00412 06998

(e) 119875AJ 119875FA and 119875FD of each area for ST3

Area number 119875AJ 119875FA 119875FD

1 08320 00677 058772 07396 02062 034623 08270 0 1

(f) 119875AJ 119875FA and 119875FD of each area for ST5

Area number 119875AJ 119875FA 119875FD

1 08284 0 12 07613 0 13 07566 01590 035644 08596 0 15 08281 00678 05813

(g) 119875AJ 119875FA and 119875FD of each area for ST7

Area number 119875AJ 119875FA 119875FD

1 07874 0 12 07020 01664 047233 08799 00118 087774 08358 0 15 08537 0 16 07761 00955 050477 07407 01028 06720

and compared with each other Finally we select the situationwhich produces the best result

In order to distinguish different situations the labels ofthemare denoted inTable 4 where the parameter Ti is used to

Table 6 The Index for each of eight situations

Label of corresponding situation IndexFE1 06750FE3 06659FE5 07026FE7 06625ST1 06297ST3 06572ST5 06391ST7 06231

illustrate the research object is the region where TCsrsquo centersare located Ti hours before they landed onHainan Island Nudenotes the number of areas of divided research object

It can be seen from Table 4 that FE5 is the situation whichhas been researched in Sections 31 and 32 The remainingseven situations are researched as follows using the methodswhich are identical with FE5

The 119875AJ 119875FA and 119875FD of each area for each of the remain-ing seven situations can be calculated according to formula(5) the results of which are shown in Table 5

In order to select the best of these eight situations inTable 4 an evaluation method is introduced with which theIndex of each situation is calculated where Index is definedaccording to formula (6) For any situation 119873 denotes thenumber of areas of divided research object and119875AJ 119894119875FA 119894 and119875FD 119894 denote the 119875AJ 119875FA and 119875FD of the 119894th area respectivelyConsider the following

Index =

1

119873

119873

sum

119894=1

radic[

1

10

119875

2

AJ 119894 +3

10

(1 minus 119875FA 119894)2

+

3

5

(1 minus 119875FD 119894)2

]

(6)

It is obvious that the higher the Index is the better theresult of the landing criterions on the whole is The Indexfor each of eight situations is shown in Table 6 The situationFE5 shows the best result which also illustrates that theresearch scheme in Sections 31 and 32 is better comparedwith other situations Finally situation FE5 is selected to formthe landing criterions

34 Forecast of TCsrsquo Characteristic Factors

341 Landing Prediction Pattern The landing forecast pat-tern is defined as follows obtaining the observation pointinformation (seven predictors) when landing TCsrsquo centersfirst enter into any area which can be used to forecast thecharacteristic factors (LAT LONAP andWS)whenTCs landon Hainan Island The flow diagram is shown in Figure 6Taking area 1 for example the OPs of historical landing TCsrsquocenters when they first entered into area 1 are shown inFigure 11 and the tracks of historical landing TCs whichpassed through area 1 are shown in Figure 12

Dividing the historical landing TCs passing through eacharea into two groups with the same number one group of TCsis used to establish prediction equations and the other group

Advances in Meteorology 9

110 115 120 125

10

12

14

16

18

20

22

24

26

Hainan IslandThe boundary of area 1

The LPs of TCsThe OPs of landing TCs enteringinto area 1 for the first time

Longitude (∘E)

Latit

ude (

∘N

)

Figure 11 The OPs of historical landing TCsrsquo centers when theyfirstly entered into area 1 (LP landing position)

Hainan IslandThe boundary of area 1

The LPs of TCs

110 115 120 1258

10

12

14

16

18

20

22

24

Latit

ude (

∘N

)

Longitude (∘E)

The TCs which passedthrough area 1

Figure 12 The tracks of historical landing TCs which passedthrough area 1

is used to test the accuracy of these equations The results oftesting of these prediction equations for TCs which passedthrough each area are shown in Table 7 Making use of theactual and predicted longitude and latitude of TCsrsquo centersin combination with the formula (3) the calculated meanand standard deviation of GCD in the landing predictionpattern are shown in Figures 13 and 14 respectively Averagingthe results of five areas it can be obtained that the averageof the meanstandard deviation (SD) of GCD is 1446382978740 km In [24] Yu et al analyze the averageGCD error of48 hoursrsquo forecast in the South China Sea which is 2226 kmTherefore the landing prediction pattern proposed in this

1 2 3 4 50

20

40

60

80

100

120

140

160

Area number

The m

ean

of G

CD (k

m)

The mean of GCD for each area in landing prediction pattern

Figure 13 The mean of GCD for each area in landing predictionpattern

1 2 3 4 50

20

40

60

80

100

Area number

The s

tand

ard

devi

atio

n of

GCD

(km

)The standard deviation of GCD for each area in

landing prediction pattern

Figure 14 The standard deviation of GCD for each area in landingprediction pattern

paper shows good prediction accuracy For TCs which arejudged to be landing on Hainan Island as long as the obser-vation point information when their centers first enter intoany area are obtained the corresponding forecast equationscan be used to predict characteristic factors when they land

342 Dynamic Prediction Pattern The dynamic predictionpattern is using the current observation point informationto conduct 24 hoursrsquo and 48 hoursrsquo forecast which is alsoillustrated in Figure 8There are two different forecast modelsin dynamic prediction pattern that are described as follows

Forecast Model 1 It is to obtain the current observation pointinformation (seven predictors) when landing TCsrsquo centersenter into any area for the first time making use of which toconduct 24 hoursrsquo and 48 hoursrsquo forecast

10 Advances in Meteorology

1 2 3 4 50

50

100

150

200

Area number

The m

ean

of G

CD (k

m)

The mean of GCD for 24 hoursrsquo forecast in each areausing forecast models 1 and 2

Forecast model 1Forecast model 2

(a) The mean of GCD for 24 hoursrsquo forecast of two forecast models

Forecast model 1Forecast model 2

1 2 3 4 50

20

40

60

80

100

Area number

The s

tand

ard

devi

atio

n of

GCD

(km

)

The standard deviation of GCD for 24 hoursrsquo forecastin each area using forecast models 1 and 2

(b) The standard deviation of GCD for 24 hoursrsquo forecast of two forecastmodels

Figure 15 The meanstandard deviation of GCD for 24 hoursrsquo forecast of two forecast models

Forecast model 1Forecast model 2

1 2 3 4 50

50

100

150

200

250

300

Area number

The m

ean

of G

CD (k

m)

The mean of GCD for 48 hoursrsquo forecast in each areausing forecast models 1 and 2

(a) The mean of GCD for 48 hoursrsquo forecast of two forecast models

Forecast model 1Forecast model 2

1 2 3 4 50

50

100

150

200

Area number

The s

tand

ard

devi

atio

n of

GCD

(km

)

The standard deviation of GCD for 48 hoursrsquo forecastin each area using forecast models 1 and 2

(b) The standard deviation of GCD for 48 hoursrsquo forecast of two forecastmodels

Figure 16 The meanstandard deviation of GCD for 48 hoursrsquo forecast of two forecast models

Forecast Model 2 It is to obtain the current observation pointinformation (seven predictors) when landing TCsrsquo centers arein any area (not necessarily enter into any area for the firsttime)making use ofwhich to conduct 24 hoursrsquo and 48hoursrsquoforecast

In the process of actual prediction for TCs which arejudged to be landing on Hainan Island the observation pointinformationwhen their centers enter into any area for the first

time and the established equations in forecast model 1 areused to conduct dynamic prediction Furthermore the obser-vation point information when TCsrsquo centers are in area (it isnot necessary that TCsrsquo centers enters into this area for thefirst time) any area and the established equations in forecastmodel 2 can be used to conduct dynamic prediction

Similar to Section 341 the historical observation pointsthat meet the corresponding requirements in corresponding

Advances in Meteorology 11

Table 7 Prediction results in landing prediction pattern

Area numberThe meanSD of LAT

deviationUnit ∘N

The meanSD of LATdeviationUnit ∘E

The meanSD of APdeviationUnit hpa

The meanSD of WSdeviationUnit ms

1 0984908671 0726406404 9288967882 59540438392 0526806009 0867708973 7994665339 53036389973 0971007805 0634007711 10694183758 70915526374 1153708640 0715006162 10346361761 68151474985 1115206941 0672807103 12113885744 8040353759

Table 8 The results of testing forecast model 1 and forecast model 2

Forecast hour Area number Forecast model(1 or 2)

The meanSD ofLAT deviation

Unit ∘N

The meanSD ofLON deviation

Unit ∘E

The meanSD ofAP deviationUnit hpa

The meanSD ofWS deviationUnit ms

24 h

1 1 0812806515 1029507600 4927839733 37787288582 0700405153 0880207011 5411443187 3963729843

2 1 0913105832 0986407794 6341244247 38268278572 0753806538 1005008670 4917549673 3848730261

3 1 0755505945 1052809093 7814969316 47342417822 0772706183 0894907255 7348966512 5171843368

4 1 0658205365 0879706251 6637355333 36128264862 0640404911 0950006746 6228056617 3752125645

5 1 0689904989 0861806124 6221863820 29503258232 0699605634 1001707915 6320361131 4186732785

48 h

1 1 1219808343 1773514304 9778257246 68402402892 1259808949 1694112488 9455772439 6503044361

2 1 1225009604 2196016746 10977584764 79032593072 1285710131 2033015552 8143463813 6529343273

3 1 1240707253 2165517786 9658069204 75541529892 1325009565 1841415666 8102472309 6245253772

4 1 0916208120 1761813047 7900955968 57899453712 0989607848 1502011732 7097848208 5282637708

5 1 0965007833 1696014327 9110774706 87411640062 1131510845 1973115516 9244968041 6805748502

forecast model (1 or 2) are divided into two groups with thesame number of observation points One group of observa-tion points is used to establish prediction equations and theother group of points is used to test the accuracies of theseequations The results of testing these prediction equationsin forecast model 1 and forecast model 2 are shown inTable 8 In combinationwith formula (3) the calculatedmeanand standard deviation of GCD in two forecast models areshown in Figures 15 and 16 respectively Averaging theresults of five areas it can be obtained that the averagesof the meanstandard deviation of GCD under forecastmodel 1 and forecast model 2 for 24 hoursrsquo forecast fare1505192846156 km and 1415464812509 km respectivelyFor 48 hoursrsquo forecast the averages of the meanstandarddeviation of GCD under two different forecast modelsare 26175171456345 km and 25671091452903 km respec-tively Even though the mean of GCD in dynamic prediction

pattern is no less than three main prediction centers (NHCJMA and NMCC) it is much less than the numerical pre-diction model in [25] the means of which are 18633195 kmbased on System T106 and 16182958 km based on T213 bothfor 2448 hoursrsquo forecast Besides forecast models 1 and 2 areall more accurate than the forecast using satellite scatterome-terrsquos monitoring data in the sense of the standard deviationof GCD and the mean of weed speed error which are1496002 km and 119618ms in [12] It can be seen fromTable 8 and Figures 15 and 16 that the accuracies of forecastmodel 1 and forecast model 2 vary from different areas anddifferent characteristic factors The more accurate forecastmodel can be selected from forecast models 1 and 2 accordingto actual conditionsThe results of statistical significance testsfor each equation used to forecast corresponding charac-teristic factor in forecast model 1 and forecast model 2 areshown in Table 9 which show that 119875 value is much less than

12 Advances in Meteorology

Table 9 The results of statistical significance tests for each equation used to forecast corresponding characteristic factor in forecast model 1and forecast model 2

Forecast hour Area numberForecastmodel(1 or 2)

119877-square119875 valuefor LAT

119877-square119875 valuefor LON

119877-square119875 valuefor AP

119877-square119875 valuefor WS

24 h

1 1 05883952 times 10minus11 06011791 times 10minus9 06222446 times 10minus10 07615547 times 10minus15

2 04493226 times 10minus32 05809593 times 10minus50 07030131 times 10minus68 07731101 times 10minus81

2 1 07177186 times 10minus14 04391366 times 10minus5 07556503 times 10minus16 07995357 times 10minus18

2 06808517 times 10minus101 06753177 times 10minus99 07481213 times 10minus122 07277371 times 10minus114

3 1 06471636 times 10minus11 06559348 times 10minus11 06310183 times 10minus10 06495324 times 10minus10

2 06802646 times 10minus94 07506119 times 10minus114 05509140 times 10minus62 05573984 times 10minus65

4 1 07572105 times 10minus12 05348332 times 10minus8 07594115 times 10minus13 08174247 times 10minus15

2 06405483 times 10minus42 06266164 times 10minus40 06903101 times 10minus44 07652925 times 10minus57

5 1 05031240 times 10minus6 05154151 times 10minus6 08732273 times 10minus14 08524503 times 10minus15

2 06239284 times 10minus35 06464167 times 10minus37 07835698 times 10minus52 07550152 times 10minus48

48 h

1 1 0133700240 03943391 times 10minus5 03208177 times 10minus4 04725534 times 10minus6

2 0217700049 0039903398 0253200016 04275792 times 10minus6

2 1 04515905 times 10minus6 0266800218 06720172 times 10minus10 04622599 times 10minus6

2 04022164 times 10minus36 03095651 times 10minus26 02960505 times 10minus23 03332627 times 10minus27

3 1 0225100218 03935457 times 10minus5 0002607270 01362015962 03073618 times 10minus25 04369213 times 10minus39 01768251 times 10minus12 02479891 times 10minus18

4 1 04992960 times 10minus7 03055274 times 10minus4 04303153 times 10minus5 06249184 times 10minus9

2 03323377 times 10minus16 03270102 times 10minus16 02471170 times 10minus9 03580365 times 10minus16

5 1 04546619 times 10minus5 0255400043 03705191 times 10minus4 03535000122 04219109 times 10minus18 03825344 times 10minus17 02120218 times 10minus7 03322246 times 10minus12

005 in almost every case and prove that the correspondingprediction equation is significant

4 Summary

In this paper the CMA best track datasets from 1949 to 2012are used in combination with data mining technology andstatistical methods to put forward a new methodology toforecast TCsrsquo characteristic factors This methodology canaccurately judge whether TCs land on Hainan Island or notand forecast their characteristic factors (including longitudelatitude the lowest center pressure and wind speed) Theaverage of the probabilities of accurate judgment for landingcriterions is 7470 and the highest accuracy can reach7976 For the forecast of landing TCsrsquo characteristic factorslanding prediction pattern and dynamic prediction patternare proposed which not only can accurately forecast thecharacteristic factors when TCs land but also realize dynam-ically 24 hoursrsquo and 48 hoursrsquo forecast The effect of thelanding prediction pattern is better of which the mean ofGCD is 1446382 km compared with the current 48 hoursrsquoforecast in the South China Sea which is 2226 km Eventhough the mean of GCD in dynamic prediction pattern isno less than three main prediction centers (NHC JMA andNMCC) it is much less than the numerical prediction modelin [25] and the method using satellite scatterometerrsquos mon-itoring data in [12] The forecast methodology proposed inthis paper provides a new method for typhoon warning on

Hainan Islandwithout getting toomuch knowledge ofmeteo-rology involved and thus simplifies the implementation of theprediction process andmeanwhile guarantees the accuracy ofprediction

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

Acknowledgments

This work is in part supported by the National Science-Technology Support Plan in the domain of advanced energytechnology Hainan Power Grid Corporation provided sup-port for this research through ldquoIntegrated DemonstrationProject of Regional Smart Gridrdquo no 2013BAA01B03 DrXinhong Huang a Research Engineer in the Department ofElectrical amp Computer Engineering from the University ofWestern Ontario also puts forward valuable revision com-ments on this paper

References

[1] National Climate Center of China Typhoon ldquoDamreyrdquo HasCaused Serious Damages to Hainan Province 2005 httpncccmagovcnWebsiteindexphpNewsID=1454

Advances in Meteorology 13

[2] J S Pedro F Burstein and A Sharp ldquoA case-based fuzzy mul-ticriteria decision support model for tropical cyclone forecast-ingrdquo European Journal of Operational Research vol 160 no 2pp 308ndash324 2005

[3] X Qin and M Mu ldquoA Study on the reduction of forecast errorvariance by three adaptive observation approaches for tropicalcyclone predictionrdquoMonthlyWeather Review vol 139 no 7 pp2218ndash2232 2011

[4] Y Wang W Zhang and W Fu ldquoBack Propogation(BP)-neuralnetwork for tropical cyclone track forecastrdquo in Proceedings of the19th International Conference on Geoinformatics pp 1ndash4 IEEEShanghai China June 2011

[5] B Feng and J N K Liu ldquoAn adaptive neural network classifierfor tropical cyclone prediction using a two-layer feature selec-torrdquo in Advances in Neural NetworksmdashISNN 2005 vol 3497of Lecture Notes in Computer Science pp 399ndash404 SpringerBerlin Germany 2005

[6] H-J Song S-H Huh J-H Kim C-H Ho and S-K ParkldquoTyphoon track prediction by a support vector machine usingdata reduction methodsrdquo in Computational Intelligence andSecurity vol 3801 of Lecture Notes in Computer Science pp 503ndash511 Springer Berlin Germany 2005

[7] M DeMaria ldquoA simplified dynamical system for tropicalcyclone intensity predictionrdquoMonthly Weather Review vol 137no 1 pp 68ndash82 2009

[8] H R Winterbottom E W Uhlhorn and E P Chassignet ldquoAdesign and an application of a regional coupled atmosphere-ocean model for tropical cyclone predictionrdquo Journal ofAdvances in Modeling Earth Systems vol 4 no 10 Article IDM10002 2012

[9] L M Ma and Z M Tan ldquoImproving the behavior of the cumu-lus parameterization for tropical cyclone prediction convectiontriggerrdquo Atmospheric Research vol 92 no 2 pp 190ndash211 2009

[10] J S Gall I Ginis S-J Lin T P Marchok and J-H ChenldquoExperimental tropical cyclone prediction using the GFDL 25-km-resolution global atmospheric modelrdquo Weather and Fore-casting vol 26 no 6 pp 1008ndash1019 2011

[11] G Lv ldquoThe Northwestern Pacific typhoon track forecast in2005rdquo in Proceedings of the Meteorological Science and Technol-ogy Symposium on Taiwan Strait p 7 Chinese MeteorologicalSociety 2006 (Chinese)

[12] D Zhang Y Zhang T Hu B Xie and J Xu ldquoA comparison ofHY-2 and QuikSCAT vector wind products for tropical cyclonetrack and intensity development monitoringrdquo IEEE Geoscienceand Remote Sensing Letters vol 11 no 8 pp 1365ndash1369 2014

[13] MYingWZhangHYu et al ldquoAnoverviewof theChinamete-orological administration tropical cyclone databaserdquo Journal ofAtmospheric and Oceanic Technology vol 31 no 2 pp 287ndash3012014

[14] G E Forsythe ldquoGeneration and use of orthogonal polynomialsfor data-fittingwith a digital computerrdquo Journal of the Society forIndustrial amp Applied Mathematics vol 5 no 2 pp 74ndash88 1957

[15] KWagstaff C Cardie S Rogers and S Schroedl ldquoConstrainedk-means clustering with background knowledgerdquo in Proceed-ings of the 18th International Conference on Machine Learningvol 1 pp 577ndash584 2001

[16] Y Ye ldquoNeighborhood density method for selecting initialcluster centers in K-mean clusteringrdquo in Proceedings of theWorkshop on Data Mining for Biomedical Applications (PAKDDrsquo06) pp 189ndash198 2006

[17] P Xiong Data Mining Algorithm and Clementine PracticeTsinghua University Press Beijing China 2011 (Chinese)

[18] S L Crawford ldquoExtensions to the CART algorithmrdquo Interna-tional Journal ofMan-Machine Studies vol 31 no 2 pp 197ndash2171989

[19] D M Allen ldquoMean square error of prediction as a criterion forselecting variablesrdquo Technometrics vol 13 pp 469ndash475 1971

[20] S Yu and J Shen ldquoForward and backward algorithms forselecting predictors on the basis of the criterion frompredictionsumof squares and their applicationrdquoActaMeteorologica Sinicavol 1 pp 83ndash90 1988

[21] D Yao and S Yu ldquoThe stepwise algorithm of selecting forecastfactors based on PRESS rulerdquo Journal of Atmospheric Sciencesvol 2 pp 129ndash135 1992 (Chinese)

[22] Y LuMathematical Statistics Methods East China University ofScience and Technology Press Shanghai China 2005 (Chi-nese)

[23] K Xie and B Liu ldquoAn ENSO-forecast independent statisticalmodel for the prediction of annual Atlantic tropical cyclonefrequency in Aprilrdquo Advances in Meteorology vol 2014 ArticleID 248148 11 pages 2014

[24] J Yu J Tang Y Dai and B Yu ldquoThe error and cause analysis ofChinarsquos typhoon path predictionrdquo Journal of Weather vol 6 pp695ndash700 2012

[25] SMa A Qu and Z Yu ldquoThe parallelization of typhoon numer-ical prediction model of and track forecast error analysisrdquo Jour-nal of Applied Meteorology vol 3 pp 322ndash328 2004 (Chinese)

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ClimatologyJournal of

EcologyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

EarthquakesJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom

Applied ampEnvironmentalSoil Science

Volume 2014

Mining

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014

International Journal of

Geophysics

OceanographyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of Computational Environmental SciencesHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal ofPetroleum Engineering

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

GeochemistryHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of

Atmospheric SciencesInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

OceanographyHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

MineralogyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

MeteorologyAdvances in

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Paleontology JournalHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ScientificaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Geological ResearchJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Geology Advances in

Advances in Meteorology 7

Calculate the seven predictors

of any OP in area 1

Calculate the seven predictors

of any OP in area 2

LONMVYes No

LAT

LONMV

LONMV

LAT

LON

No

NO

Yes

Yes

No

No

No

Yes

Yes

Yes

Landing

Landing

Landing

Not landing

Not landing Not

landingNot

landing

Not landing

Not landing

Not landing

Not landing

Not landing

Not landing Not

landing

Not landing

Not landing

Not landing

Not landing

Not landing

Not landing

LATYes

LONMVLAT

WS

AP

No

Yes

Yes

Yes

No

Yes

No

No

No

Landing

Landing

Calculate the seven predictors

of any OP in area 3

LONMV

LAT

LON

LONMV

LATMV

Landing

Landing

Landing

Yes

Yes

No

No

Yes

Yes

Yes No

No

No

Calculate the seven predictors

of any OP in area 4

LONMV

LAT

LONMV

SG

AP

YesYes

Yes

Yes

Yes

No

No

No

NoNo

Landing

Landing

Landing

Calculate the seven predictors

of any OP in area 5

LONMV

LONMV

LATMV LAT

LATMV

AP

AP

No

Yes

Yes No

Yes NoYes

NoNo No

No

YesYes

Yes

Landing Landing

le14250

le116850

le minus27918

leminus5358

leminus16795le15950

leminus0892gt14450

le11

le10055

leminus6058

le20650

le116650

le22310

le2317

le minus13357

le12550

le1500

le989000

le minus9666

le minus22048

le10656

le19450le10656

le968500

le1002500

leminus21621

gt14950

Figure 10 The landing criterions in five areas

Table 3 119875AJ 119875FA and 119875FD of each criterion

Area number 119875AJ 119875FA 119875FD

1 7468 1794 39482 7531 1555 39533 7750 1176 43154 6623 3654 25935 7976 1649 3588

as 1198722 Then 119875AJ 119875FA and 119875FD of this area are calculated as

follows

119875AJ =119872

1+119872

2

119885

1+ 119885

2

119875FA =

119885

2minus119872

2

119885

2

119875FD =

119885

1minus119872

1

119885

1

(5)

Table 4 The labels of different situations

Ti Nu Label of corresponding situation48 1 FE148 3 FE348 5 FE548 7 FE772 1 ST172 3 ST372 5 ST572 7 ST7

33 Other Situations as a Comparative Experiment In Sec-tions 31 and 32 the region where TCsrsquo centers were located48 hours before they landed is selected as research object andthe research object is divided into five areas In this sectionother situations as a comparative experiment are researched

8 Advances in Meteorology

Table 5119875AJ119875FA and119875FD for each of the remaining seven situations

(a) 119875AJ 119875FA and 119875FD of each area for FE1

Area number 119875AJ 119875FA 119875FD

1 07884 00696 05279

(b) 119875AJ 119875FA and 119875FD of each area for FE3

Area number 119875AJ 119875FA 119875FD

1 07454 01139 051912 07636 01875 030543 07829 00446 07814

(c) 119875AJ 119875FA and 119875FD of each area for FE7

Area number 119875AJ 119875FA 119875FD

1 08522 00154 084142 07486 01991 039593 07302 01808 041764 07719 01165 043985 07236 00740 067356 07834 01512 029757 07631 0 1

(d) 119875AJ 119875FA and 119875FD of each area for ST1

Area number 119875AJ 119875FA 119875FD

1 08164 00412 06998

(e) 119875AJ 119875FA and 119875FD of each area for ST3

Area number 119875AJ 119875FA 119875FD

1 08320 00677 058772 07396 02062 034623 08270 0 1

(f) 119875AJ 119875FA and 119875FD of each area for ST5

Area number 119875AJ 119875FA 119875FD

1 08284 0 12 07613 0 13 07566 01590 035644 08596 0 15 08281 00678 05813

(g) 119875AJ 119875FA and 119875FD of each area for ST7

Area number 119875AJ 119875FA 119875FD

1 07874 0 12 07020 01664 047233 08799 00118 087774 08358 0 15 08537 0 16 07761 00955 050477 07407 01028 06720

and compared with each other Finally we select the situationwhich produces the best result

In order to distinguish different situations the labels ofthemare denoted inTable 4 where the parameter Ti is used to

Table 6 The Index for each of eight situations

Label of corresponding situation IndexFE1 06750FE3 06659FE5 07026FE7 06625ST1 06297ST3 06572ST5 06391ST7 06231

illustrate the research object is the region where TCsrsquo centersare located Ti hours before they landed onHainan Island Nudenotes the number of areas of divided research object

It can be seen from Table 4 that FE5 is the situation whichhas been researched in Sections 31 and 32 The remainingseven situations are researched as follows using the methodswhich are identical with FE5

The 119875AJ 119875FA and 119875FD of each area for each of the remain-ing seven situations can be calculated according to formula(5) the results of which are shown in Table 5

In order to select the best of these eight situations inTable 4 an evaluation method is introduced with which theIndex of each situation is calculated where Index is definedaccording to formula (6) For any situation 119873 denotes thenumber of areas of divided research object and119875AJ 119894119875FA 119894 and119875FD 119894 denote the 119875AJ 119875FA and 119875FD of the 119894th area respectivelyConsider the following

Index =

1

119873

119873

sum

119894=1

radic[

1

10

119875

2

AJ 119894 +3

10

(1 minus 119875FA 119894)2

+

3

5

(1 minus 119875FD 119894)2

]

(6)

It is obvious that the higher the Index is the better theresult of the landing criterions on the whole is The Indexfor each of eight situations is shown in Table 6 The situationFE5 shows the best result which also illustrates that theresearch scheme in Sections 31 and 32 is better comparedwith other situations Finally situation FE5 is selected to formthe landing criterions

34 Forecast of TCsrsquo Characteristic Factors

341 Landing Prediction Pattern The landing forecast pat-tern is defined as follows obtaining the observation pointinformation (seven predictors) when landing TCsrsquo centersfirst enter into any area which can be used to forecast thecharacteristic factors (LAT LONAP andWS)whenTCs landon Hainan Island The flow diagram is shown in Figure 6Taking area 1 for example the OPs of historical landing TCsrsquocenters when they first entered into area 1 are shown inFigure 11 and the tracks of historical landing TCs whichpassed through area 1 are shown in Figure 12

Dividing the historical landing TCs passing through eacharea into two groups with the same number one group of TCsis used to establish prediction equations and the other group

Advances in Meteorology 9

110 115 120 125

10

12

14

16

18

20

22

24

26

Hainan IslandThe boundary of area 1

The LPs of TCsThe OPs of landing TCs enteringinto area 1 for the first time

Longitude (∘E)

Latit

ude (

∘N

)

Figure 11 The OPs of historical landing TCsrsquo centers when theyfirstly entered into area 1 (LP landing position)

Hainan IslandThe boundary of area 1

The LPs of TCs

110 115 120 1258

10

12

14

16

18

20

22

24

Latit

ude (

∘N

)

Longitude (∘E)

The TCs which passedthrough area 1

Figure 12 The tracks of historical landing TCs which passedthrough area 1

is used to test the accuracy of these equations The results oftesting of these prediction equations for TCs which passedthrough each area are shown in Table 7 Making use of theactual and predicted longitude and latitude of TCsrsquo centersin combination with the formula (3) the calculated meanand standard deviation of GCD in the landing predictionpattern are shown in Figures 13 and 14 respectively Averagingthe results of five areas it can be obtained that the averageof the meanstandard deviation (SD) of GCD is 1446382978740 km In [24] Yu et al analyze the averageGCD error of48 hoursrsquo forecast in the South China Sea which is 2226 kmTherefore the landing prediction pattern proposed in this

1 2 3 4 50

20

40

60

80

100

120

140

160

Area number

The m

ean

of G

CD (k

m)

The mean of GCD for each area in landing prediction pattern

Figure 13 The mean of GCD for each area in landing predictionpattern

1 2 3 4 50

20

40

60

80

100

Area number

The s

tand

ard

devi

atio

n of

GCD

(km

)The standard deviation of GCD for each area in

landing prediction pattern

Figure 14 The standard deviation of GCD for each area in landingprediction pattern

paper shows good prediction accuracy For TCs which arejudged to be landing on Hainan Island as long as the obser-vation point information when their centers first enter intoany area are obtained the corresponding forecast equationscan be used to predict characteristic factors when they land

342 Dynamic Prediction Pattern The dynamic predictionpattern is using the current observation point informationto conduct 24 hoursrsquo and 48 hoursrsquo forecast which is alsoillustrated in Figure 8There are two different forecast modelsin dynamic prediction pattern that are described as follows

Forecast Model 1 It is to obtain the current observation pointinformation (seven predictors) when landing TCsrsquo centersenter into any area for the first time making use of which toconduct 24 hoursrsquo and 48 hoursrsquo forecast

10 Advances in Meteorology

1 2 3 4 50

50

100

150

200

Area number

The m

ean

of G

CD (k

m)

The mean of GCD for 24 hoursrsquo forecast in each areausing forecast models 1 and 2

Forecast model 1Forecast model 2

(a) The mean of GCD for 24 hoursrsquo forecast of two forecast models

Forecast model 1Forecast model 2

1 2 3 4 50

20

40

60

80

100

Area number

The s

tand

ard

devi

atio

n of

GCD

(km

)

The standard deviation of GCD for 24 hoursrsquo forecastin each area using forecast models 1 and 2

(b) The standard deviation of GCD for 24 hoursrsquo forecast of two forecastmodels

Figure 15 The meanstandard deviation of GCD for 24 hoursrsquo forecast of two forecast models

Forecast model 1Forecast model 2

1 2 3 4 50

50

100

150

200

250

300

Area number

The m

ean

of G

CD (k

m)

The mean of GCD for 48 hoursrsquo forecast in each areausing forecast models 1 and 2

(a) The mean of GCD for 48 hoursrsquo forecast of two forecast models

Forecast model 1Forecast model 2

1 2 3 4 50

50

100

150

200

Area number

The s

tand

ard

devi

atio

n of

GCD

(km

)

The standard deviation of GCD for 48 hoursrsquo forecastin each area using forecast models 1 and 2

(b) The standard deviation of GCD for 48 hoursrsquo forecast of two forecastmodels

Figure 16 The meanstandard deviation of GCD for 48 hoursrsquo forecast of two forecast models

Forecast Model 2 It is to obtain the current observation pointinformation (seven predictors) when landing TCsrsquo centers arein any area (not necessarily enter into any area for the firsttime)making use ofwhich to conduct 24 hoursrsquo and 48hoursrsquoforecast

In the process of actual prediction for TCs which arejudged to be landing on Hainan Island the observation pointinformationwhen their centers enter into any area for the first

time and the established equations in forecast model 1 areused to conduct dynamic prediction Furthermore the obser-vation point information when TCsrsquo centers are in area (it isnot necessary that TCsrsquo centers enters into this area for thefirst time) any area and the established equations in forecastmodel 2 can be used to conduct dynamic prediction

Similar to Section 341 the historical observation pointsthat meet the corresponding requirements in corresponding

Advances in Meteorology 11

Table 7 Prediction results in landing prediction pattern

Area numberThe meanSD of LAT

deviationUnit ∘N

The meanSD of LATdeviationUnit ∘E

The meanSD of APdeviationUnit hpa

The meanSD of WSdeviationUnit ms

1 0984908671 0726406404 9288967882 59540438392 0526806009 0867708973 7994665339 53036389973 0971007805 0634007711 10694183758 70915526374 1153708640 0715006162 10346361761 68151474985 1115206941 0672807103 12113885744 8040353759

Table 8 The results of testing forecast model 1 and forecast model 2

Forecast hour Area number Forecast model(1 or 2)

The meanSD ofLAT deviation

Unit ∘N

The meanSD ofLON deviation

Unit ∘E

The meanSD ofAP deviationUnit hpa

The meanSD ofWS deviationUnit ms

24 h

1 1 0812806515 1029507600 4927839733 37787288582 0700405153 0880207011 5411443187 3963729843

2 1 0913105832 0986407794 6341244247 38268278572 0753806538 1005008670 4917549673 3848730261

3 1 0755505945 1052809093 7814969316 47342417822 0772706183 0894907255 7348966512 5171843368

4 1 0658205365 0879706251 6637355333 36128264862 0640404911 0950006746 6228056617 3752125645

5 1 0689904989 0861806124 6221863820 29503258232 0699605634 1001707915 6320361131 4186732785

48 h

1 1 1219808343 1773514304 9778257246 68402402892 1259808949 1694112488 9455772439 6503044361

2 1 1225009604 2196016746 10977584764 79032593072 1285710131 2033015552 8143463813 6529343273

3 1 1240707253 2165517786 9658069204 75541529892 1325009565 1841415666 8102472309 6245253772

4 1 0916208120 1761813047 7900955968 57899453712 0989607848 1502011732 7097848208 5282637708

5 1 0965007833 1696014327 9110774706 87411640062 1131510845 1973115516 9244968041 6805748502

forecast model (1 or 2) are divided into two groups with thesame number of observation points One group of observa-tion points is used to establish prediction equations and theother group of points is used to test the accuracies of theseequations The results of testing these prediction equationsin forecast model 1 and forecast model 2 are shown inTable 8 In combinationwith formula (3) the calculatedmeanand standard deviation of GCD in two forecast models areshown in Figures 15 and 16 respectively Averaging theresults of five areas it can be obtained that the averagesof the meanstandard deviation of GCD under forecastmodel 1 and forecast model 2 for 24 hoursrsquo forecast fare1505192846156 km and 1415464812509 km respectivelyFor 48 hoursrsquo forecast the averages of the meanstandarddeviation of GCD under two different forecast modelsare 26175171456345 km and 25671091452903 km respec-tively Even though the mean of GCD in dynamic prediction

pattern is no less than three main prediction centers (NHCJMA and NMCC) it is much less than the numerical pre-diction model in [25] the means of which are 18633195 kmbased on System T106 and 16182958 km based on T213 bothfor 2448 hoursrsquo forecast Besides forecast models 1 and 2 areall more accurate than the forecast using satellite scatterome-terrsquos monitoring data in the sense of the standard deviationof GCD and the mean of weed speed error which are1496002 km and 119618ms in [12] It can be seen fromTable 8 and Figures 15 and 16 that the accuracies of forecastmodel 1 and forecast model 2 vary from different areas anddifferent characteristic factors The more accurate forecastmodel can be selected from forecast models 1 and 2 accordingto actual conditionsThe results of statistical significance testsfor each equation used to forecast corresponding charac-teristic factor in forecast model 1 and forecast model 2 areshown in Table 9 which show that 119875 value is much less than

12 Advances in Meteorology

Table 9 The results of statistical significance tests for each equation used to forecast corresponding characteristic factor in forecast model 1and forecast model 2

Forecast hour Area numberForecastmodel(1 or 2)

119877-square119875 valuefor LAT

119877-square119875 valuefor LON

119877-square119875 valuefor AP

119877-square119875 valuefor WS

24 h

1 1 05883952 times 10minus11 06011791 times 10minus9 06222446 times 10minus10 07615547 times 10minus15

2 04493226 times 10minus32 05809593 times 10minus50 07030131 times 10minus68 07731101 times 10minus81

2 1 07177186 times 10minus14 04391366 times 10minus5 07556503 times 10minus16 07995357 times 10minus18

2 06808517 times 10minus101 06753177 times 10minus99 07481213 times 10minus122 07277371 times 10minus114

3 1 06471636 times 10minus11 06559348 times 10minus11 06310183 times 10minus10 06495324 times 10minus10

2 06802646 times 10minus94 07506119 times 10minus114 05509140 times 10minus62 05573984 times 10minus65

4 1 07572105 times 10minus12 05348332 times 10minus8 07594115 times 10minus13 08174247 times 10minus15

2 06405483 times 10minus42 06266164 times 10minus40 06903101 times 10minus44 07652925 times 10minus57

5 1 05031240 times 10minus6 05154151 times 10minus6 08732273 times 10minus14 08524503 times 10minus15

2 06239284 times 10minus35 06464167 times 10minus37 07835698 times 10minus52 07550152 times 10minus48

48 h

1 1 0133700240 03943391 times 10minus5 03208177 times 10minus4 04725534 times 10minus6

2 0217700049 0039903398 0253200016 04275792 times 10minus6

2 1 04515905 times 10minus6 0266800218 06720172 times 10minus10 04622599 times 10minus6

2 04022164 times 10minus36 03095651 times 10minus26 02960505 times 10minus23 03332627 times 10minus27

3 1 0225100218 03935457 times 10minus5 0002607270 01362015962 03073618 times 10minus25 04369213 times 10minus39 01768251 times 10minus12 02479891 times 10minus18

4 1 04992960 times 10minus7 03055274 times 10minus4 04303153 times 10minus5 06249184 times 10minus9

2 03323377 times 10minus16 03270102 times 10minus16 02471170 times 10minus9 03580365 times 10minus16

5 1 04546619 times 10minus5 0255400043 03705191 times 10minus4 03535000122 04219109 times 10minus18 03825344 times 10minus17 02120218 times 10minus7 03322246 times 10minus12

005 in almost every case and prove that the correspondingprediction equation is significant

4 Summary

In this paper the CMA best track datasets from 1949 to 2012are used in combination with data mining technology andstatistical methods to put forward a new methodology toforecast TCsrsquo characteristic factors This methodology canaccurately judge whether TCs land on Hainan Island or notand forecast their characteristic factors (including longitudelatitude the lowest center pressure and wind speed) Theaverage of the probabilities of accurate judgment for landingcriterions is 7470 and the highest accuracy can reach7976 For the forecast of landing TCsrsquo characteristic factorslanding prediction pattern and dynamic prediction patternare proposed which not only can accurately forecast thecharacteristic factors when TCs land but also realize dynam-ically 24 hoursrsquo and 48 hoursrsquo forecast The effect of thelanding prediction pattern is better of which the mean ofGCD is 1446382 km compared with the current 48 hoursrsquoforecast in the South China Sea which is 2226 km Eventhough the mean of GCD in dynamic prediction pattern isno less than three main prediction centers (NHC JMA andNMCC) it is much less than the numerical prediction modelin [25] and the method using satellite scatterometerrsquos mon-itoring data in [12] The forecast methodology proposed inthis paper provides a new method for typhoon warning on

Hainan Islandwithout getting toomuch knowledge ofmeteo-rology involved and thus simplifies the implementation of theprediction process andmeanwhile guarantees the accuracy ofprediction

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

Acknowledgments

This work is in part supported by the National Science-Technology Support Plan in the domain of advanced energytechnology Hainan Power Grid Corporation provided sup-port for this research through ldquoIntegrated DemonstrationProject of Regional Smart Gridrdquo no 2013BAA01B03 DrXinhong Huang a Research Engineer in the Department ofElectrical amp Computer Engineering from the University ofWestern Ontario also puts forward valuable revision com-ments on this paper

References

[1] National Climate Center of China Typhoon ldquoDamreyrdquo HasCaused Serious Damages to Hainan Province 2005 httpncccmagovcnWebsiteindexphpNewsID=1454

Advances in Meteorology 13

[2] J S Pedro F Burstein and A Sharp ldquoA case-based fuzzy mul-ticriteria decision support model for tropical cyclone forecast-ingrdquo European Journal of Operational Research vol 160 no 2pp 308ndash324 2005

[3] X Qin and M Mu ldquoA Study on the reduction of forecast errorvariance by three adaptive observation approaches for tropicalcyclone predictionrdquoMonthlyWeather Review vol 139 no 7 pp2218ndash2232 2011

[4] Y Wang W Zhang and W Fu ldquoBack Propogation(BP)-neuralnetwork for tropical cyclone track forecastrdquo in Proceedings of the19th International Conference on Geoinformatics pp 1ndash4 IEEEShanghai China June 2011

[5] B Feng and J N K Liu ldquoAn adaptive neural network classifierfor tropical cyclone prediction using a two-layer feature selec-torrdquo in Advances in Neural NetworksmdashISNN 2005 vol 3497of Lecture Notes in Computer Science pp 399ndash404 SpringerBerlin Germany 2005

[6] H-J Song S-H Huh J-H Kim C-H Ho and S-K ParkldquoTyphoon track prediction by a support vector machine usingdata reduction methodsrdquo in Computational Intelligence andSecurity vol 3801 of Lecture Notes in Computer Science pp 503ndash511 Springer Berlin Germany 2005

[7] M DeMaria ldquoA simplified dynamical system for tropicalcyclone intensity predictionrdquoMonthly Weather Review vol 137no 1 pp 68ndash82 2009

[8] H R Winterbottom E W Uhlhorn and E P Chassignet ldquoAdesign and an application of a regional coupled atmosphere-ocean model for tropical cyclone predictionrdquo Journal ofAdvances in Modeling Earth Systems vol 4 no 10 Article IDM10002 2012

[9] L M Ma and Z M Tan ldquoImproving the behavior of the cumu-lus parameterization for tropical cyclone prediction convectiontriggerrdquo Atmospheric Research vol 92 no 2 pp 190ndash211 2009

[10] J S Gall I Ginis S-J Lin T P Marchok and J-H ChenldquoExperimental tropical cyclone prediction using the GFDL 25-km-resolution global atmospheric modelrdquo Weather and Fore-casting vol 26 no 6 pp 1008ndash1019 2011

[11] G Lv ldquoThe Northwestern Pacific typhoon track forecast in2005rdquo in Proceedings of the Meteorological Science and Technol-ogy Symposium on Taiwan Strait p 7 Chinese MeteorologicalSociety 2006 (Chinese)

[12] D Zhang Y Zhang T Hu B Xie and J Xu ldquoA comparison ofHY-2 and QuikSCAT vector wind products for tropical cyclonetrack and intensity development monitoringrdquo IEEE Geoscienceand Remote Sensing Letters vol 11 no 8 pp 1365ndash1369 2014

[13] MYingWZhangHYu et al ldquoAnoverviewof theChinamete-orological administration tropical cyclone databaserdquo Journal ofAtmospheric and Oceanic Technology vol 31 no 2 pp 287ndash3012014

[14] G E Forsythe ldquoGeneration and use of orthogonal polynomialsfor data-fittingwith a digital computerrdquo Journal of the Society forIndustrial amp Applied Mathematics vol 5 no 2 pp 74ndash88 1957

[15] KWagstaff C Cardie S Rogers and S Schroedl ldquoConstrainedk-means clustering with background knowledgerdquo in Proceed-ings of the 18th International Conference on Machine Learningvol 1 pp 577ndash584 2001

[16] Y Ye ldquoNeighborhood density method for selecting initialcluster centers in K-mean clusteringrdquo in Proceedings of theWorkshop on Data Mining for Biomedical Applications (PAKDDrsquo06) pp 189ndash198 2006

[17] P Xiong Data Mining Algorithm and Clementine PracticeTsinghua University Press Beijing China 2011 (Chinese)

[18] S L Crawford ldquoExtensions to the CART algorithmrdquo Interna-tional Journal ofMan-Machine Studies vol 31 no 2 pp 197ndash2171989

[19] D M Allen ldquoMean square error of prediction as a criterion forselecting variablesrdquo Technometrics vol 13 pp 469ndash475 1971

[20] S Yu and J Shen ldquoForward and backward algorithms forselecting predictors on the basis of the criterion frompredictionsumof squares and their applicationrdquoActaMeteorologica Sinicavol 1 pp 83ndash90 1988

[21] D Yao and S Yu ldquoThe stepwise algorithm of selecting forecastfactors based on PRESS rulerdquo Journal of Atmospheric Sciencesvol 2 pp 129ndash135 1992 (Chinese)

[22] Y LuMathematical Statistics Methods East China University ofScience and Technology Press Shanghai China 2005 (Chi-nese)

[23] K Xie and B Liu ldquoAn ENSO-forecast independent statisticalmodel for the prediction of annual Atlantic tropical cyclonefrequency in Aprilrdquo Advances in Meteorology vol 2014 ArticleID 248148 11 pages 2014

[24] J Yu J Tang Y Dai and B Yu ldquoThe error and cause analysis ofChinarsquos typhoon path predictionrdquo Journal of Weather vol 6 pp695ndash700 2012

[25] SMa A Qu and Z Yu ldquoThe parallelization of typhoon numer-ical prediction model of and track forecast error analysisrdquo Jour-nal of Applied Meteorology vol 3 pp 322ndash328 2004 (Chinese)

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ClimatologyJournal of

EcologyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

EarthquakesJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom

Applied ampEnvironmentalSoil Science

Volume 2014

Mining

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014

International Journal of

Geophysics

OceanographyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of Computational Environmental SciencesHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal ofPetroleum Engineering

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

GeochemistryHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of

Atmospheric SciencesInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

OceanographyHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

MineralogyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

MeteorologyAdvances in

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Paleontology JournalHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ScientificaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Geological ResearchJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Geology Advances in

8 Advances in Meteorology

Table 5119875AJ119875FA and119875FD for each of the remaining seven situations

(a) 119875AJ 119875FA and 119875FD of each area for FE1

Area number 119875AJ 119875FA 119875FD

1 07884 00696 05279

(b) 119875AJ 119875FA and 119875FD of each area for FE3

Area number 119875AJ 119875FA 119875FD

1 07454 01139 051912 07636 01875 030543 07829 00446 07814

(c) 119875AJ 119875FA and 119875FD of each area for FE7

Area number 119875AJ 119875FA 119875FD

1 08522 00154 084142 07486 01991 039593 07302 01808 041764 07719 01165 043985 07236 00740 067356 07834 01512 029757 07631 0 1

(d) 119875AJ 119875FA and 119875FD of each area for ST1

Area number 119875AJ 119875FA 119875FD

1 08164 00412 06998

(e) 119875AJ 119875FA and 119875FD of each area for ST3

Area number 119875AJ 119875FA 119875FD

1 08320 00677 058772 07396 02062 034623 08270 0 1

(f) 119875AJ 119875FA and 119875FD of each area for ST5

Area number 119875AJ 119875FA 119875FD

1 08284 0 12 07613 0 13 07566 01590 035644 08596 0 15 08281 00678 05813

(g) 119875AJ 119875FA and 119875FD of each area for ST7

Area number 119875AJ 119875FA 119875FD

1 07874 0 12 07020 01664 047233 08799 00118 087774 08358 0 15 08537 0 16 07761 00955 050477 07407 01028 06720

and compared with each other Finally we select the situationwhich produces the best result

In order to distinguish different situations the labels ofthemare denoted inTable 4 where the parameter Ti is used to

Table 6 The Index for each of eight situations

Label of corresponding situation IndexFE1 06750FE3 06659FE5 07026FE7 06625ST1 06297ST3 06572ST5 06391ST7 06231

illustrate the research object is the region where TCsrsquo centersare located Ti hours before they landed onHainan Island Nudenotes the number of areas of divided research object

It can be seen from Table 4 that FE5 is the situation whichhas been researched in Sections 31 and 32 The remainingseven situations are researched as follows using the methodswhich are identical with FE5

The 119875AJ 119875FA and 119875FD of each area for each of the remain-ing seven situations can be calculated according to formula(5) the results of which are shown in Table 5

In order to select the best of these eight situations inTable 4 an evaluation method is introduced with which theIndex of each situation is calculated where Index is definedaccording to formula (6) For any situation 119873 denotes thenumber of areas of divided research object and119875AJ 119894119875FA 119894 and119875FD 119894 denote the 119875AJ 119875FA and 119875FD of the 119894th area respectivelyConsider the following

Index =

1

119873

119873

sum

119894=1

radic[

1

10

119875

2

AJ 119894 +3

10

(1 minus 119875FA 119894)2

+

3

5

(1 minus 119875FD 119894)2

]

(6)

It is obvious that the higher the Index is the better theresult of the landing criterions on the whole is The Indexfor each of eight situations is shown in Table 6 The situationFE5 shows the best result which also illustrates that theresearch scheme in Sections 31 and 32 is better comparedwith other situations Finally situation FE5 is selected to formthe landing criterions

34 Forecast of TCsrsquo Characteristic Factors

341 Landing Prediction Pattern The landing forecast pat-tern is defined as follows obtaining the observation pointinformation (seven predictors) when landing TCsrsquo centersfirst enter into any area which can be used to forecast thecharacteristic factors (LAT LONAP andWS)whenTCs landon Hainan Island The flow diagram is shown in Figure 6Taking area 1 for example the OPs of historical landing TCsrsquocenters when they first entered into area 1 are shown inFigure 11 and the tracks of historical landing TCs whichpassed through area 1 are shown in Figure 12

Dividing the historical landing TCs passing through eacharea into two groups with the same number one group of TCsis used to establish prediction equations and the other group

Advances in Meteorology 9

110 115 120 125

10

12

14

16

18

20

22

24

26

Hainan IslandThe boundary of area 1

The LPs of TCsThe OPs of landing TCs enteringinto area 1 for the first time

Longitude (∘E)

Latit

ude (

∘N

)

Figure 11 The OPs of historical landing TCsrsquo centers when theyfirstly entered into area 1 (LP landing position)

Hainan IslandThe boundary of area 1

The LPs of TCs

110 115 120 1258

10

12

14

16

18

20

22

24

Latit

ude (

∘N

)

Longitude (∘E)

The TCs which passedthrough area 1

Figure 12 The tracks of historical landing TCs which passedthrough area 1

is used to test the accuracy of these equations The results oftesting of these prediction equations for TCs which passedthrough each area are shown in Table 7 Making use of theactual and predicted longitude and latitude of TCsrsquo centersin combination with the formula (3) the calculated meanand standard deviation of GCD in the landing predictionpattern are shown in Figures 13 and 14 respectively Averagingthe results of five areas it can be obtained that the averageof the meanstandard deviation (SD) of GCD is 1446382978740 km In [24] Yu et al analyze the averageGCD error of48 hoursrsquo forecast in the South China Sea which is 2226 kmTherefore the landing prediction pattern proposed in this

1 2 3 4 50

20

40

60

80

100

120

140

160

Area number

The m

ean

of G

CD (k

m)

The mean of GCD for each area in landing prediction pattern

Figure 13 The mean of GCD for each area in landing predictionpattern

1 2 3 4 50

20

40

60

80

100

Area number

The s

tand

ard

devi

atio

n of

GCD

(km

)The standard deviation of GCD for each area in

landing prediction pattern

Figure 14 The standard deviation of GCD for each area in landingprediction pattern

paper shows good prediction accuracy For TCs which arejudged to be landing on Hainan Island as long as the obser-vation point information when their centers first enter intoany area are obtained the corresponding forecast equationscan be used to predict characteristic factors when they land

342 Dynamic Prediction Pattern The dynamic predictionpattern is using the current observation point informationto conduct 24 hoursrsquo and 48 hoursrsquo forecast which is alsoillustrated in Figure 8There are two different forecast modelsin dynamic prediction pattern that are described as follows

Forecast Model 1 It is to obtain the current observation pointinformation (seven predictors) when landing TCsrsquo centersenter into any area for the first time making use of which toconduct 24 hoursrsquo and 48 hoursrsquo forecast

10 Advances in Meteorology

1 2 3 4 50

50

100

150

200

Area number

The m

ean

of G

CD (k

m)

The mean of GCD for 24 hoursrsquo forecast in each areausing forecast models 1 and 2

Forecast model 1Forecast model 2

(a) The mean of GCD for 24 hoursrsquo forecast of two forecast models

Forecast model 1Forecast model 2

1 2 3 4 50

20

40

60

80

100

Area number

The s

tand

ard

devi

atio

n of

GCD

(km

)

The standard deviation of GCD for 24 hoursrsquo forecastin each area using forecast models 1 and 2

(b) The standard deviation of GCD for 24 hoursrsquo forecast of two forecastmodels

Figure 15 The meanstandard deviation of GCD for 24 hoursrsquo forecast of two forecast models

Forecast model 1Forecast model 2

1 2 3 4 50

50

100

150

200

250

300

Area number

The m

ean

of G

CD (k

m)

The mean of GCD for 48 hoursrsquo forecast in each areausing forecast models 1 and 2

(a) The mean of GCD for 48 hoursrsquo forecast of two forecast models

Forecast model 1Forecast model 2

1 2 3 4 50

50

100

150

200

Area number

The s

tand

ard

devi

atio

n of

GCD

(km

)

The standard deviation of GCD for 48 hoursrsquo forecastin each area using forecast models 1 and 2

(b) The standard deviation of GCD for 48 hoursrsquo forecast of two forecastmodels

Figure 16 The meanstandard deviation of GCD for 48 hoursrsquo forecast of two forecast models

Forecast Model 2 It is to obtain the current observation pointinformation (seven predictors) when landing TCsrsquo centers arein any area (not necessarily enter into any area for the firsttime)making use ofwhich to conduct 24 hoursrsquo and 48hoursrsquoforecast

In the process of actual prediction for TCs which arejudged to be landing on Hainan Island the observation pointinformationwhen their centers enter into any area for the first

time and the established equations in forecast model 1 areused to conduct dynamic prediction Furthermore the obser-vation point information when TCsrsquo centers are in area (it isnot necessary that TCsrsquo centers enters into this area for thefirst time) any area and the established equations in forecastmodel 2 can be used to conduct dynamic prediction

Similar to Section 341 the historical observation pointsthat meet the corresponding requirements in corresponding

Advances in Meteorology 11

Table 7 Prediction results in landing prediction pattern

Area numberThe meanSD of LAT

deviationUnit ∘N

The meanSD of LATdeviationUnit ∘E

The meanSD of APdeviationUnit hpa

The meanSD of WSdeviationUnit ms

1 0984908671 0726406404 9288967882 59540438392 0526806009 0867708973 7994665339 53036389973 0971007805 0634007711 10694183758 70915526374 1153708640 0715006162 10346361761 68151474985 1115206941 0672807103 12113885744 8040353759

Table 8 The results of testing forecast model 1 and forecast model 2

Forecast hour Area number Forecast model(1 or 2)

The meanSD ofLAT deviation

Unit ∘N

The meanSD ofLON deviation

Unit ∘E

The meanSD ofAP deviationUnit hpa

The meanSD ofWS deviationUnit ms

24 h

1 1 0812806515 1029507600 4927839733 37787288582 0700405153 0880207011 5411443187 3963729843

2 1 0913105832 0986407794 6341244247 38268278572 0753806538 1005008670 4917549673 3848730261

3 1 0755505945 1052809093 7814969316 47342417822 0772706183 0894907255 7348966512 5171843368

4 1 0658205365 0879706251 6637355333 36128264862 0640404911 0950006746 6228056617 3752125645

5 1 0689904989 0861806124 6221863820 29503258232 0699605634 1001707915 6320361131 4186732785

48 h

1 1 1219808343 1773514304 9778257246 68402402892 1259808949 1694112488 9455772439 6503044361

2 1 1225009604 2196016746 10977584764 79032593072 1285710131 2033015552 8143463813 6529343273

3 1 1240707253 2165517786 9658069204 75541529892 1325009565 1841415666 8102472309 6245253772

4 1 0916208120 1761813047 7900955968 57899453712 0989607848 1502011732 7097848208 5282637708

5 1 0965007833 1696014327 9110774706 87411640062 1131510845 1973115516 9244968041 6805748502

forecast model (1 or 2) are divided into two groups with thesame number of observation points One group of observa-tion points is used to establish prediction equations and theother group of points is used to test the accuracies of theseequations The results of testing these prediction equationsin forecast model 1 and forecast model 2 are shown inTable 8 In combinationwith formula (3) the calculatedmeanand standard deviation of GCD in two forecast models areshown in Figures 15 and 16 respectively Averaging theresults of five areas it can be obtained that the averagesof the meanstandard deviation of GCD under forecastmodel 1 and forecast model 2 for 24 hoursrsquo forecast fare1505192846156 km and 1415464812509 km respectivelyFor 48 hoursrsquo forecast the averages of the meanstandarddeviation of GCD under two different forecast modelsare 26175171456345 km and 25671091452903 km respec-tively Even though the mean of GCD in dynamic prediction

pattern is no less than three main prediction centers (NHCJMA and NMCC) it is much less than the numerical pre-diction model in [25] the means of which are 18633195 kmbased on System T106 and 16182958 km based on T213 bothfor 2448 hoursrsquo forecast Besides forecast models 1 and 2 areall more accurate than the forecast using satellite scatterome-terrsquos monitoring data in the sense of the standard deviationof GCD and the mean of weed speed error which are1496002 km and 119618ms in [12] It can be seen fromTable 8 and Figures 15 and 16 that the accuracies of forecastmodel 1 and forecast model 2 vary from different areas anddifferent characteristic factors The more accurate forecastmodel can be selected from forecast models 1 and 2 accordingto actual conditionsThe results of statistical significance testsfor each equation used to forecast corresponding charac-teristic factor in forecast model 1 and forecast model 2 areshown in Table 9 which show that 119875 value is much less than

12 Advances in Meteorology

Table 9 The results of statistical significance tests for each equation used to forecast corresponding characteristic factor in forecast model 1and forecast model 2

Forecast hour Area numberForecastmodel(1 or 2)

119877-square119875 valuefor LAT

119877-square119875 valuefor LON

119877-square119875 valuefor AP

119877-square119875 valuefor WS

24 h

1 1 05883952 times 10minus11 06011791 times 10minus9 06222446 times 10minus10 07615547 times 10minus15

2 04493226 times 10minus32 05809593 times 10minus50 07030131 times 10minus68 07731101 times 10minus81

2 1 07177186 times 10minus14 04391366 times 10minus5 07556503 times 10minus16 07995357 times 10minus18

2 06808517 times 10minus101 06753177 times 10minus99 07481213 times 10minus122 07277371 times 10minus114

3 1 06471636 times 10minus11 06559348 times 10minus11 06310183 times 10minus10 06495324 times 10minus10

2 06802646 times 10minus94 07506119 times 10minus114 05509140 times 10minus62 05573984 times 10minus65

4 1 07572105 times 10minus12 05348332 times 10minus8 07594115 times 10minus13 08174247 times 10minus15

2 06405483 times 10minus42 06266164 times 10minus40 06903101 times 10minus44 07652925 times 10minus57

5 1 05031240 times 10minus6 05154151 times 10minus6 08732273 times 10minus14 08524503 times 10minus15

2 06239284 times 10minus35 06464167 times 10minus37 07835698 times 10minus52 07550152 times 10minus48

48 h

1 1 0133700240 03943391 times 10minus5 03208177 times 10minus4 04725534 times 10minus6

2 0217700049 0039903398 0253200016 04275792 times 10minus6

2 1 04515905 times 10minus6 0266800218 06720172 times 10minus10 04622599 times 10minus6

2 04022164 times 10minus36 03095651 times 10minus26 02960505 times 10minus23 03332627 times 10minus27

3 1 0225100218 03935457 times 10minus5 0002607270 01362015962 03073618 times 10minus25 04369213 times 10minus39 01768251 times 10minus12 02479891 times 10minus18

4 1 04992960 times 10minus7 03055274 times 10minus4 04303153 times 10minus5 06249184 times 10minus9

2 03323377 times 10minus16 03270102 times 10minus16 02471170 times 10minus9 03580365 times 10minus16

5 1 04546619 times 10minus5 0255400043 03705191 times 10minus4 03535000122 04219109 times 10minus18 03825344 times 10minus17 02120218 times 10minus7 03322246 times 10minus12

005 in almost every case and prove that the correspondingprediction equation is significant

4 Summary

In this paper the CMA best track datasets from 1949 to 2012are used in combination with data mining technology andstatistical methods to put forward a new methodology toforecast TCsrsquo characteristic factors This methodology canaccurately judge whether TCs land on Hainan Island or notand forecast their characteristic factors (including longitudelatitude the lowest center pressure and wind speed) Theaverage of the probabilities of accurate judgment for landingcriterions is 7470 and the highest accuracy can reach7976 For the forecast of landing TCsrsquo characteristic factorslanding prediction pattern and dynamic prediction patternare proposed which not only can accurately forecast thecharacteristic factors when TCs land but also realize dynam-ically 24 hoursrsquo and 48 hoursrsquo forecast The effect of thelanding prediction pattern is better of which the mean ofGCD is 1446382 km compared with the current 48 hoursrsquoforecast in the South China Sea which is 2226 km Eventhough the mean of GCD in dynamic prediction pattern isno less than three main prediction centers (NHC JMA andNMCC) it is much less than the numerical prediction modelin [25] and the method using satellite scatterometerrsquos mon-itoring data in [12] The forecast methodology proposed inthis paper provides a new method for typhoon warning on

Hainan Islandwithout getting toomuch knowledge ofmeteo-rology involved and thus simplifies the implementation of theprediction process andmeanwhile guarantees the accuracy ofprediction

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

Acknowledgments

This work is in part supported by the National Science-Technology Support Plan in the domain of advanced energytechnology Hainan Power Grid Corporation provided sup-port for this research through ldquoIntegrated DemonstrationProject of Regional Smart Gridrdquo no 2013BAA01B03 DrXinhong Huang a Research Engineer in the Department ofElectrical amp Computer Engineering from the University ofWestern Ontario also puts forward valuable revision com-ments on this paper

References

[1] National Climate Center of China Typhoon ldquoDamreyrdquo HasCaused Serious Damages to Hainan Province 2005 httpncccmagovcnWebsiteindexphpNewsID=1454

Advances in Meteorology 13

[2] J S Pedro F Burstein and A Sharp ldquoA case-based fuzzy mul-ticriteria decision support model for tropical cyclone forecast-ingrdquo European Journal of Operational Research vol 160 no 2pp 308ndash324 2005

[3] X Qin and M Mu ldquoA Study on the reduction of forecast errorvariance by three adaptive observation approaches for tropicalcyclone predictionrdquoMonthlyWeather Review vol 139 no 7 pp2218ndash2232 2011

[4] Y Wang W Zhang and W Fu ldquoBack Propogation(BP)-neuralnetwork for tropical cyclone track forecastrdquo in Proceedings of the19th International Conference on Geoinformatics pp 1ndash4 IEEEShanghai China June 2011

[5] B Feng and J N K Liu ldquoAn adaptive neural network classifierfor tropical cyclone prediction using a two-layer feature selec-torrdquo in Advances in Neural NetworksmdashISNN 2005 vol 3497of Lecture Notes in Computer Science pp 399ndash404 SpringerBerlin Germany 2005

[6] H-J Song S-H Huh J-H Kim C-H Ho and S-K ParkldquoTyphoon track prediction by a support vector machine usingdata reduction methodsrdquo in Computational Intelligence andSecurity vol 3801 of Lecture Notes in Computer Science pp 503ndash511 Springer Berlin Germany 2005

[7] M DeMaria ldquoA simplified dynamical system for tropicalcyclone intensity predictionrdquoMonthly Weather Review vol 137no 1 pp 68ndash82 2009

[8] H R Winterbottom E W Uhlhorn and E P Chassignet ldquoAdesign and an application of a regional coupled atmosphere-ocean model for tropical cyclone predictionrdquo Journal ofAdvances in Modeling Earth Systems vol 4 no 10 Article IDM10002 2012

[9] L M Ma and Z M Tan ldquoImproving the behavior of the cumu-lus parameterization for tropical cyclone prediction convectiontriggerrdquo Atmospheric Research vol 92 no 2 pp 190ndash211 2009

[10] J S Gall I Ginis S-J Lin T P Marchok and J-H ChenldquoExperimental tropical cyclone prediction using the GFDL 25-km-resolution global atmospheric modelrdquo Weather and Fore-casting vol 26 no 6 pp 1008ndash1019 2011

[11] G Lv ldquoThe Northwestern Pacific typhoon track forecast in2005rdquo in Proceedings of the Meteorological Science and Technol-ogy Symposium on Taiwan Strait p 7 Chinese MeteorologicalSociety 2006 (Chinese)

[12] D Zhang Y Zhang T Hu B Xie and J Xu ldquoA comparison ofHY-2 and QuikSCAT vector wind products for tropical cyclonetrack and intensity development monitoringrdquo IEEE Geoscienceand Remote Sensing Letters vol 11 no 8 pp 1365ndash1369 2014

[13] MYingWZhangHYu et al ldquoAnoverviewof theChinamete-orological administration tropical cyclone databaserdquo Journal ofAtmospheric and Oceanic Technology vol 31 no 2 pp 287ndash3012014

[14] G E Forsythe ldquoGeneration and use of orthogonal polynomialsfor data-fittingwith a digital computerrdquo Journal of the Society forIndustrial amp Applied Mathematics vol 5 no 2 pp 74ndash88 1957

[15] KWagstaff C Cardie S Rogers and S Schroedl ldquoConstrainedk-means clustering with background knowledgerdquo in Proceed-ings of the 18th International Conference on Machine Learningvol 1 pp 577ndash584 2001

[16] Y Ye ldquoNeighborhood density method for selecting initialcluster centers in K-mean clusteringrdquo in Proceedings of theWorkshop on Data Mining for Biomedical Applications (PAKDDrsquo06) pp 189ndash198 2006

[17] P Xiong Data Mining Algorithm and Clementine PracticeTsinghua University Press Beijing China 2011 (Chinese)

[18] S L Crawford ldquoExtensions to the CART algorithmrdquo Interna-tional Journal ofMan-Machine Studies vol 31 no 2 pp 197ndash2171989

[19] D M Allen ldquoMean square error of prediction as a criterion forselecting variablesrdquo Technometrics vol 13 pp 469ndash475 1971

[20] S Yu and J Shen ldquoForward and backward algorithms forselecting predictors on the basis of the criterion frompredictionsumof squares and their applicationrdquoActaMeteorologica Sinicavol 1 pp 83ndash90 1988

[21] D Yao and S Yu ldquoThe stepwise algorithm of selecting forecastfactors based on PRESS rulerdquo Journal of Atmospheric Sciencesvol 2 pp 129ndash135 1992 (Chinese)

[22] Y LuMathematical Statistics Methods East China University ofScience and Technology Press Shanghai China 2005 (Chi-nese)

[23] K Xie and B Liu ldquoAn ENSO-forecast independent statisticalmodel for the prediction of annual Atlantic tropical cyclonefrequency in Aprilrdquo Advances in Meteorology vol 2014 ArticleID 248148 11 pages 2014

[24] J Yu J Tang Y Dai and B Yu ldquoThe error and cause analysis ofChinarsquos typhoon path predictionrdquo Journal of Weather vol 6 pp695ndash700 2012

[25] SMa A Qu and Z Yu ldquoThe parallelization of typhoon numer-ical prediction model of and track forecast error analysisrdquo Jour-nal of Applied Meteorology vol 3 pp 322ndash328 2004 (Chinese)

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ClimatologyJournal of

EcologyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

EarthquakesJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom

Applied ampEnvironmentalSoil Science

Volume 2014

Mining

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014

International Journal of

Geophysics

OceanographyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of Computational Environmental SciencesHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal ofPetroleum Engineering

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

GeochemistryHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of

Atmospheric SciencesInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

OceanographyHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

MineralogyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

MeteorologyAdvances in

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Paleontology JournalHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ScientificaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Geological ResearchJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Geology Advances in

Advances in Meteorology 9

110 115 120 125

10

12

14

16

18

20

22

24

26

Hainan IslandThe boundary of area 1

The LPs of TCsThe OPs of landing TCs enteringinto area 1 for the first time

Longitude (∘E)

Latit

ude (

∘N

)

Figure 11 The OPs of historical landing TCsrsquo centers when theyfirstly entered into area 1 (LP landing position)

Hainan IslandThe boundary of area 1

The LPs of TCs

110 115 120 1258

10

12

14

16

18

20

22

24

Latit

ude (

∘N

)

Longitude (∘E)

The TCs which passedthrough area 1

Figure 12 The tracks of historical landing TCs which passedthrough area 1

is used to test the accuracy of these equations The results oftesting of these prediction equations for TCs which passedthrough each area are shown in Table 7 Making use of theactual and predicted longitude and latitude of TCsrsquo centersin combination with the formula (3) the calculated meanand standard deviation of GCD in the landing predictionpattern are shown in Figures 13 and 14 respectively Averagingthe results of five areas it can be obtained that the averageof the meanstandard deviation (SD) of GCD is 1446382978740 km In [24] Yu et al analyze the averageGCD error of48 hoursrsquo forecast in the South China Sea which is 2226 kmTherefore the landing prediction pattern proposed in this

1 2 3 4 50

20

40

60

80

100

120

140

160

Area number

The m

ean

of G

CD (k

m)

The mean of GCD for each area in landing prediction pattern

Figure 13 The mean of GCD for each area in landing predictionpattern

1 2 3 4 50

20

40

60

80

100

Area number

The s

tand

ard

devi

atio

n of

GCD

(km

)The standard deviation of GCD for each area in

landing prediction pattern

Figure 14 The standard deviation of GCD for each area in landingprediction pattern

paper shows good prediction accuracy For TCs which arejudged to be landing on Hainan Island as long as the obser-vation point information when their centers first enter intoany area are obtained the corresponding forecast equationscan be used to predict characteristic factors when they land

342 Dynamic Prediction Pattern The dynamic predictionpattern is using the current observation point informationto conduct 24 hoursrsquo and 48 hoursrsquo forecast which is alsoillustrated in Figure 8There are two different forecast modelsin dynamic prediction pattern that are described as follows

Forecast Model 1 It is to obtain the current observation pointinformation (seven predictors) when landing TCsrsquo centersenter into any area for the first time making use of which toconduct 24 hoursrsquo and 48 hoursrsquo forecast

10 Advances in Meteorology

1 2 3 4 50

50

100

150

200

Area number

The m

ean

of G

CD (k

m)

The mean of GCD for 24 hoursrsquo forecast in each areausing forecast models 1 and 2

Forecast model 1Forecast model 2

(a) The mean of GCD for 24 hoursrsquo forecast of two forecast models

Forecast model 1Forecast model 2

1 2 3 4 50

20

40

60

80

100

Area number

The s

tand

ard

devi

atio

n of

GCD

(km

)

The standard deviation of GCD for 24 hoursrsquo forecastin each area using forecast models 1 and 2

(b) The standard deviation of GCD for 24 hoursrsquo forecast of two forecastmodels

Figure 15 The meanstandard deviation of GCD for 24 hoursrsquo forecast of two forecast models

Forecast model 1Forecast model 2

1 2 3 4 50

50

100

150

200

250

300

Area number

The m

ean

of G

CD (k

m)

The mean of GCD for 48 hoursrsquo forecast in each areausing forecast models 1 and 2

(a) The mean of GCD for 48 hoursrsquo forecast of two forecast models

Forecast model 1Forecast model 2

1 2 3 4 50

50

100

150

200

Area number

The s

tand

ard

devi

atio

n of

GCD

(km

)

The standard deviation of GCD for 48 hoursrsquo forecastin each area using forecast models 1 and 2

(b) The standard deviation of GCD for 48 hoursrsquo forecast of two forecastmodels

Figure 16 The meanstandard deviation of GCD for 48 hoursrsquo forecast of two forecast models

Forecast Model 2 It is to obtain the current observation pointinformation (seven predictors) when landing TCsrsquo centers arein any area (not necessarily enter into any area for the firsttime)making use ofwhich to conduct 24 hoursrsquo and 48hoursrsquoforecast

In the process of actual prediction for TCs which arejudged to be landing on Hainan Island the observation pointinformationwhen their centers enter into any area for the first

time and the established equations in forecast model 1 areused to conduct dynamic prediction Furthermore the obser-vation point information when TCsrsquo centers are in area (it isnot necessary that TCsrsquo centers enters into this area for thefirst time) any area and the established equations in forecastmodel 2 can be used to conduct dynamic prediction

Similar to Section 341 the historical observation pointsthat meet the corresponding requirements in corresponding

Advances in Meteorology 11

Table 7 Prediction results in landing prediction pattern

Area numberThe meanSD of LAT

deviationUnit ∘N

The meanSD of LATdeviationUnit ∘E

The meanSD of APdeviationUnit hpa

The meanSD of WSdeviationUnit ms

1 0984908671 0726406404 9288967882 59540438392 0526806009 0867708973 7994665339 53036389973 0971007805 0634007711 10694183758 70915526374 1153708640 0715006162 10346361761 68151474985 1115206941 0672807103 12113885744 8040353759

Table 8 The results of testing forecast model 1 and forecast model 2

Forecast hour Area number Forecast model(1 or 2)

The meanSD ofLAT deviation

Unit ∘N

The meanSD ofLON deviation

Unit ∘E

The meanSD ofAP deviationUnit hpa

The meanSD ofWS deviationUnit ms

24 h

1 1 0812806515 1029507600 4927839733 37787288582 0700405153 0880207011 5411443187 3963729843

2 1 0913105832 0986407794 6341244247 38268278572 0753806538 1005008670 4917549673 3848730261

3 1 0755505945 1052809093 7814969316 47342417822 0772706183 0894907255 7348966512 5171843368

4 1 0658205365 0879706251 6637355333 36128264862 0640404911 0950006746 6228056617 3752125645

5 1 0689904989 0861806124 6221863820 29503258232 0699605634 1001707915 6320361131 4186732785

48 h

1 1 1219808343 1773514304 9778257246 68402402892 1259808949 1694112488 9455772439 6503044361

2 1 1225009604 2196016746 10977584764 79032593072 1285710131 2033015552 8143463813 6529343273

3 1 1240707253 2165517786 9658069204 75541529892 1325009565 1841415666 8102472309 6245253772

4 1 0916208120 1761813047 7900955968 57899453712 0989607848 1502011732 7097848208 5282637708

5 1 0965007833 1696014327 9110774706 87411640062 1131510845 1973115516 9244968041 6805748502

forecast model (1 or 2) are divided into two groups with thesame number of observation points One group of observa-tion points is used to establish prediction equations and theother group of points is used to test the accuracies of theseequations The results of testing these prediction equationsin forecast model 1 and forecast model 2 are shown inTable 8 In combinationwith formula (3) the calculatedmeanand standard deviation of GCD in two forecast models areshown in Figures 15 and 16 respectively Averaging theresults of five areas it can be obtained that the averagesof the meanstandard deviation of GCD under forecastmodel 1 and forecast model 2 for 24 hoursrsquo forecast fare1505192846156 km and 1415464812509 km respectivelyFor 48 hoursrsquo forecast the averages of the meanstandarddeviation of GCD under two different forecast modelsare 26175171456345 km and 25671091452903 km respec-tively Even though the mean of GCD in dynamic prediction

pattern is no less than three main prediction centers (NHCJMA and NMCC) it is much less than the numerical pre-diction model in [25] the means of which are 18633195 kmbased on System T106 and 16182958 km based on T213 bothfor 2448 hoursrsquo forecast Besides forecast models 1 and 2 areall more accurate than the forecast using satellite scatterome-terrsquos monitoring data in the sense of the standard deviationof GCD and the mean of weed speed error which are1496002 km and 119618ms in [12] It can be seen fromTable 8 and Figures 15 and 16 that the accuracies of forecastmodel 1 and forecast model 2 vary from different areas anddifferent characteristic factors The more accurate forecastmodel can be selected from forecast models 1 and 2 accordingto actual conditionsThe results of statistical significance testsfor each equation used to forecast corresponding charac-teristic factor in forecast model 1 and forecast model 2 areshown in Table 9 which show that 119875 value is much less than

12 Advances in Meteorology

Table 9 The results of statistical significance tests for each equation used to forecast corresponding characteristic factor in forecast model 1and forecast model 2

Forecast hour Area numberForecastmodel(1 or 2)

119877-square119875 valuefor LAT

119877-square119875 valuefor LON

119877-square119875 valuefor AP

119877-square119875 valuefor WS

24 h

1 1 05883952 times 10minus11 06011791 times 10minus9 06222446 times 10minus10 07615547 times 10minus15

2 04493226 times 10minus32 05809593 times 10minus50 07030131 times 10minus68 07731101 times 10minus81

2 1 07177186 times 10minus14 04391366 times 10minus5 07556503 times 10minus16 07995357 times 10minus18

2 06808517 times 10minus101 06753177 times 10minus99 07481213 times 10minus122 07277371 times 10minus114

3 1 06471636 times 10minus11 06559348 times 10minus11 06310183 times 10minus10 06495324 times 10minus10

2 06802646 times 10minus94 07506119 times 10minus114 05509140 times 10minus62 05573984 times 10minus65

4 1 07572105 times 10minus12 05348332 times 10minus8 07594115 times 10minus13 08174247 times 10minus15

2 06405483 times 10minus42 06266164 times 10minus40 06903101 times 10minus44 07652925 times 10minus57

5 1 05031240 times 10minus6 05154151 times 10minus6 08732273 times 10minus14 08524503 times 10minus15

2 06239284 times 10minus35 06464167 times 10minus37 07835698 times 10minus52 07550152 times 10minus48

48 h

1 1 0133700240 03943391 times 10minus5 03208177 times 10minus4 04725534 times 10minus6

2 0217700049 0039903398 0253200016 04275792 times 10minus6

2 1 04515905 times 10minus6 0266800218 06720172 times 10minus10 04622599 times 10minus6

2 04022164 times 10minus36 03095651 times 10minus26 02960505 times 10minus23 03332627 times 10minus27

3 1 0225100218 03935457 times 10minus5 0002607270 01362015962 03073618 times 10minus25 04369213 times 10minus39 01768251 times 10minus12 02479891 times 10minus18

4 1 04992960 times 10minus7 03055274 times 10minus4 04303153 times 10minus5 06249184 times 10minus9

2 03323377 times 10minus16 03270102 times 10minus16 02471170 times 10minus9 03580365 times 10minus16

5 1 04546619 times 10minus5 0255400043 03705191 times 10minus4 03535000122 04219109 times 10minus18 03825344 times 10minus17 02120218 times 10minus7 03322246 times 10minus12

005 in almost every case and prove that the correspondingprediction equation is significant

4 Summary

In this paper the CMA best track datasets from 1949 to 2012are used in combination with data mining technology andstatistical methods to put forward a new methodology toforecast TCsrsquo characteristic factors This methodology canaccurately judge whether TCs land on Hainan Island or notand forecast their characteristic factors (including longitudelatitude the lowest center pressure and wind speed) Theaverage of the probabilities of accurate judgment for landingcriterions is 7470 and the highest accuracy can reach7976 For the forecast of landing TCsrsquo characteristic factorslanding prediction pattern and dynamic prediction patternare proposed which not only can accurately forecast thecharacteristic factors when TCs land but also realize dynam-ically 24 hoursrsquo and 48 hoursrsquo forecast The effect of thelanding prediction pattern is better of which the mean ofGCD is 1446382 km compared with the current 48 hoursrsquoforecast in the South China Sea which is 2226 km Eventhough the mean of GCD in dynamic prediction pattern isno less than three main prediction centers (NHC JMA andNMCC) it is much less than the numerical prediction modelin [25] and the method using satellite scatterometerrsquos mon-itoring data in [12] The forecast methodology proposed inthis paper provides a new method for typhoon warning on

Hainan Islandwithout getting toomuch knowledge ofmeteo-rology involved and thus simplifies the implementation of theprediction process andmeanwhile guarantees the accuracy ofprediction

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

Acknowledgments

This work is in part supported by the National Science-Technology Support Plan in the domain of advanced energytechnology Hainan Power Grid Corporation provided sup-port for this research through ldquoIntegrated DemonstrationProject of Regional Smart Gridrdquo no 2013BAA01B03 DrXinhong Huang a Research Engineer in the Department ofElectrical amp Computer Engineering from the University ofWestern Ontario also puts forward valuable revision com-ments on this paper

References

[1] National Climate Center of China Typhoon ldquoDamreyrdquo HasCaused Serious Damages to Hainan Province 2005 httpncccmagovcnWebsiteindexphpNewsID=1454

Advances in Meteorology 13

[2] J S Pedro F Burstein and A Sharp ldquoA case-based fuzzy mul-ticriteria decision support model for tropical cyclone forecast-ingrdquo European Journal of Operational Research vol 160 no 2pp 308ndash324 2005

[3] X Qin and M Mu ldquoA Study on the reduction of forecast errorvariance by three adaptive observation approaches for tropicalcyclone predictionrdquoMonthlyWeather Review vol 139 no 7 pp2218ndash2232 2011

[4] Y Wang W Zhang and W Fu ldquoBack Propogation(BP)-neuralnetwork for tropical cyclone track forecastrdquo in Proceedings of the19th International Conference on Geoinformatics pp 1ndash4 IEEEShanghai China June 2011

[5] B Feng and J N K Liu ldquoAn adaptive neural network classifierfor tropical cyclone prediction using a two-layer feature selec-torrdquo in Advances in Neural NetworksmdashISNN 2005 vol 3497of Lecture Notes in Computer Science pp 399ndash404 SpringerBerlin Germany 2005

[6] H-J Song S-H Huh J-H Kim C-H Ho and S-K ParkldquoTyphoon track prediction by a support vector machine usingdata reduction methodsrdquo in Computational Intelligence andSecurity vol 3801 of Lecture Notes in Computer Science pp 503ndash511 Springer Berlin Germany 2005

[7] M DeMaria ldquoA simplified dynamical system for tropicalcyclone intensity predictionrdquoMonthly Weather Review vol 137no 1 pp 68ndash82 2009

[8] H R Winterbottom E W Uhlhorn and E P Chassignet ldquoAdesign and an application of a regional coupled atmosphere-ocean model for tropical cyclone predictionrdquo Journal ofAdvances in Modeling Earth Systems vol 4 no 10 Article IDM10002 2012

[9] L M Ma and Z M Tan ldquoImproving the behavior of the cumu-lus parameterization for tropical cyclone prediction convectiontriggerrdquo Atmospheric Research vol 92 no 2 pp 190ndash211 2009

[10] J S Gall I Ginis S-J Lin T P Marchok and J-H ChenldquoExperimental tropical cyclone prediction using the GFDL 25-km-resolution global atmospheric modelrdquo Weather and Fore-casting vol 26 no 6 pp 1008ndash1019 2011

[11] G Lv ldquoThe Northwestern Pacific typhoon track forecast in2005rdquo in Proceedings of the Meteorological Science and Technol-ogy Symposium on Taiwan Strait p 7 Chinese MeteorologicalSociety 2006 (Chinese)

[12] D Zhang Y Zhang T Hu B Xie and J Xu ldquoA comparison ofHY-2 and QuikSCAT vector wind products for tropical cyclonetrack and intensity development monitoringrdquo IEEE Geoscienceand Remote Sensing Letters vol 11 no 8 pp 1365ndash1369 2014

[13] MYingWZhangHYu et al ldquoAnoverviewof theChinamete-orological administration tropical cyclone databaserdquo Journal ofAtmospheric and Oceanic Technology vol 31 no 2 pp 287ndash3012014

[14] G E Forsythe ldquoGeneration and use of orthogonal polynomialsfor data-fittingwith a digital computerrdquo Journal of the Society forIndustrial amp Applied Mathematics vol 5 no 2 pp 74ndash88 1957

[15] KWagstaff C Cardie S Rogers and S Schroedl ldquoConstrainedk-means clustering with background knowledgerdquo in Proceed-ings of the 18th International Conference on Machine Learningvol 1 pp 577ndash584 2001

[16] Y Ye ldquoNeighborhood density method for selecting initialcluster centers in K-mean clusteringrdquo in Proceedings of theWorkshop on Data Mining for Biomedical Applications (PAKDDrsquo06) pp 189ndash198 2006

[17] P Xiong Data Mining Algorithm and Clementine PracticeTsinghua University Press Beijing China 2011 (Chinese)

[18] S L Crawford ldquoExtensions to the CART algorithmrdquo Interna-tional Journal ofMan-Machine Studies vol 31 no 2 pp 197ndash2171989

[19] D M Allen ldquoMean square error of prediction as a criterion forselecting variablesrdquo Technometrics vol 13 pp 469ndash475 1971

[20] S Yu and J Shen ldquoForward and backward algorithms forselecting predictors on the basis of the criterion frompredictionsumof squares and their applicationrdquoActaMeteorologica Sinicavol 1 pp 83ndash90 1988

[21] D Yao and S Yu ldquoThe stepwise algorithm of selecting forecastfactors based on PRESS rulerdquo Journal of Atmospheric Sciencesvol 2 pp 129ndash135 1992 (Chinese)

[22] Y LuMathematical Statistics Methods East China University ofScience and Technology Press Shanghai China 2005 (Chi-nese)

[23] K Xie and B Liu ldquoAn ENSO-forecast independent statisticalmodel for the prediction of annual Atlantic tropical cyclonefrequency in Aprilrdquo Advances in Meteorology vol 2014 ArticleID 248148 11 pages 2014

[24] J Yu J Tang Y Dai and B Yu ldquoThe error and cause analysis ofChinarsquos typhoon path predictionrdquo Journal of Weather vol 6 pp695ndash700 2012

[25] SMa A Qu and Z Yu ldquoThe parallelization of typhoon numer-ical prediction model of and track forecast error analysisrdquo Jour-nal of Applied Meteorology vol 3 pp 322ndash328 2004 (Chinese)

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ClimatologyJournal of

EcologyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

EarthquakesJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom

Applied ampEnvironmentalSoil Science

Volume 2014

Mining

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014

International Journal of

Geophysics

OceanographyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of Computational Environmental SciencesHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal ofPetroleum Engineering

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

GeochemistryHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of

Atmospheric SciencesInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

OceanographyHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

MineralogyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

MeteorologyAdvances in

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Paleontology JournalHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ScientificaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Geological ResearchJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Geology Advances in

10 Advances in Meteorology

1 2 3 4 50

50

100

150

200

Area number

The m

ean

of G

CD (k

m)

The mean of GCD for 24 hoursrsquo forecast in each areausing forecast models 1 and 2

Forecast model 1Forecast model 2

(a) The mean of GCD for 24 hoursrsquo forecast of two forecast models

Forecast model 1Forecast model 2

1 2 3 4 50

20

40

60

80

100

Area number

The s

tand

ard

devi

atio

n of

GCD

(km

)

The standard deviation of GCD for 24 hoursrsquo forecastin each area using forecast models 1 and 2

(b) The standard deviation of GCD for 24 hoursrsquo forecast of two forecastmodels

Figure 15 The meanstandard deviation of GCD for 24 hoursrsquo forecast of two forecast models

Forecast model 1Forecast model 2

1 2 3 4 50

50

100

150

200

250

300

Area number

The m

ean

of G

CD (k

m)

The mean of GCD for 48 hoursrsquo forecast in each areausing forecast models 1 and 2

(a) The mean of GCD for 48 hoursrsquo forecast of two forecast models

Forecast model 1Forecast model 2

1 2 3 4 50

50

100

150

200

Area number

The s

tand

ard

devi

atio

n of

GCD

(km

)

The standard deviation of GCD for 48 hoursrsquo forecastin each area using forecast models 1 and 2

(b) The standard deviation of GCD for 48 hoursrsquo forecast of two forecastmodels

Figure 16 The meanstandard deviation of GCD for 48 hoursrsquo forecast of two forecast models

Forecast Model 2 It is to obtain the current observation pointinformation (seven predictors) when landing TCsrsquo centers arein any area (not necessarily enter into any area for the firsttime)making use ofwhich to conduct 24 hoursrsquo and 48hoursrsquoforecast

In the process of actual prediction for TCs which arejudged to be landing on Hainan Island the observation pointinformationwhen their centers enter into any area for the first

time and the established equations in forecast model 1 areused to conduct dynamic prediction Furthermore the obser-vation point information when TCsrsquo centers are in area (it isnot necessary that TCsrsquo centers enters into this area for thefirst time) any area and the established equations in forecastmodel 2 can be used to conduct dynamic prediction

Similar to Section 341 the historical observation pointsthat meet the corresponding requirements in corresponding

Advances in Meteorology 11

Table 7 Prediction results in landing prediction pattern

Area numberThe meanSD of LAT

deviationUnit ∘N

The meanSD of LATdeviationUnit ∘E

The meanSD of APdeviationUnit hpa

The meanSD of WSdeviationUnit ms

1 0984908671 0726406404 9288967882 59540438392 0526806009 0867708973 7994665339 53036389973 0971007805 0634007711 10694183758 70915526374 1153708640 0715006162 10346361761 68151474985 1115206941 0672807103 12113885744 8040353759

Table 8 The results of testing forecast model 1 and forecast model 2

Forecast hour Area number Forecast model(1 or 2)

The meanSD ofLAT deviation

Unit ∘N

The meanSD ofLON deviation

Unit ∘E

The meanSD ofAP deviationUnit hpa

The meanSD ofWS deviationUnit ms

24 h

1 1 0812806515 1029507600 4927839733 37787288582 0700405153 0880207011 5411443187 3963729843

2 1 0913105832 0986407794 6341244247 38268278572 0753806538 1005008670 4917549673 3848730261

3 1 0755505945 1052809093 7814969316 47342417822 0772706183 0894907255 7348966512 5171843368

4 1 0658205365 0879706251 6637355333 36128264862 0640404911 0950006746 6228056617 3752125645

5 1 0689904989 0861806124 6221863820 29503258232 0699605634 1001707915 6320361131 4186732785

48 h

1 1 1219808343 1773514304 9778257246 68402402892 1259808949 1694112488 9455772439 6503044361

2 1 1225009604 2196016746 10977584764 79032593072 1285710131 2033015552 8143463813 6529343273

3 1 1240707253 2165517786 9658069204 75541529892 1325009565 1841415666 8102472309 6245253772

4 1 0916208120 1761813047 7900955968 57899453712 0989607848 1502011732 7097848208 5282637708

5 1 0965007833 1696014327 9110774706 87411640062 1131510845 1973115516 9244968041 6805748502

forecast model (1 or 2) are divided into two groups with thesame number of observation points One group of observa-tion points is used to establish prediction equations and theother group of points is used to test the accuracies of theseequations The results of testing these prediction equationsin forecast model 1 and forecast model 2 are shown inTable 8 In combinationwith formula (3) the calculatedmeanand standard deviation of GCD in two forecast models areshown in Figures 15 and 16 respectively Averaging theresults of five areas it can be obtained that the averagesof the meanstandard deviation of GCD under forecastmodel 1 and forecast model 2 for 24 hoursrsquo forecast fare1505192846156 km and 1415464812509 km respectivelyFor 48 hoursrsquo forecast the averages of the meanstandarddeviation of GCD under two different forecast modelsare 26175171456345 km and 25671091452903 km respec-tively Even though the mean of GCD in dynamic prediction

pattern is no less than three main prediction centers (NHCJMA and NMCC) it is much less than the numerical pre-diction model in [25] the means of which are 18633195 kmbased on System T106 and 16182958 km based on T213 bothfor 2448 hoursrsquo forecast Besides forecast models 1 and 2 areall more accurate than the forecast using satellite scatterome-terrsquos monitoring data in the sense of the standard deviationof GCD and the mean of weed speed error which are1496002 km and 119618ms in [12] It can be seen fromTable 8 and Figures 15 and 16 that the accuracies of forecastmodel 1 and forecast model 2 vary from different areas anddifferent characteristic factors The more accurate forecastmodel can be selected from forecast models 1 and 2 accordingto actual conditionsThe results of statistical significance testsfor each equation used to forecast corresponding charac-teristic factor in forecast model 1 and forecast model 2 areshown in Table 9 which show that 119875 value is much less than

12 Advances in Meteorology

Table 9 The results of statistical significance tests for each equation used to forecast corresponding characteristic factor in forecast model 1and forecast model 2

Forecast hour Area numberForecastmodel(1 or 2)

119877-square119875 valuefor LAT

119877-square119875 valuefor LON

119877-square119875 valuefor AP

119877-square119875 valuefor WS

24 h

1 1 05883952 times 10minus11 06011791 times 10minus9 06222446 times 10minus10 07615547 times 10minus15

2 04493226 times 10minus32 05809593 times 10minus50 07030131 times 10minus68 07731101 times 10minus81

2 1 07177186 times 10minus14 04391366 times 10minus5 07556503 times 10minus16 07995357 times 10minus18

2 06808517 times 10minus101 06753177 times 10minus99 07481213 times 10minus122 07277371 times 10minus114

3 1 06471636 times 10minus11 06559348 times 10minus11 06310183 times 10minus10 06495324 times 10minus10

2 06802646 times 10minus94 07506119 times 10minus114 05509140 times 10minus62 05573984 times 10minus65

4 1 07572105 times 10minus12 05348332 times 10minus8 07594115 times 10minus13 08174247 times 10minus15

2 06405483 times 10minus42 06266164 times 10minus40 06903101 times 10minus44 07652925 times 10minus57

5 1 05031240 times 10minus6 05154151 times 10minus6 08732273 times 10minus14 08524503 times 10minus15

2 06239284 times 10minus35 06464167 times 10minus37 07835698 times 10minus52 07550152 times 10minus48

48 h

1 1 0133700240 03943391 times 10minus5 03208177 times 10minus4 04725534 times 10minus6

2 0217700049 0039903398 0253200016 04275792 times 10minus6

2 1 04515905 times 10minus6 0266800218 06720172 times 10minus10 04622599 times 10minus6

2 04022164 times 10minus36 03095651 times 10minus26 02960505 times 10minus23 03332627 times 10minus27

3 1 0225100218 03935457 times 10minus5 0002607270 01362015962 03073618 times 10minus25 04369213 times 10minus39 01768251 times 10minus12 02479891 times 10minus18

4 1 04992960 times 10minus7 03055274 times 10minus4 04303153 times 10minus5 06249184 times 10minus9

2 03323377 times 10minus16 03270102 times 10minus16 02471170 times 10minus9 03580365 times 10minus16

5 1 04546619 times 10minus5 0255400043 03705191 times 10minus4 03535000122 04219109 times 10minus18 03825344 times 10minus17 02120218 times 10minus7 03322246 times 10minus12

005 in almost every case and prove that the correspondingprediction equation is significant

4 Summary

In this paper the CMA best track datasets from 1949 to 2012are used in combination with data mining technology andstatistical methods to put forward a new methodology toforecast TCsrsquo characteristic factors This methodology canaccurately judge whether TCs land on Hainan Island or notand forecast their characteristic factors (including longitudelatitude the lowest center pressure and wind speed) Theaverage of the probabilities of accurate judgment for landingcriterions is 7470 and the highest accuracy can reach7976 For the forecast of landing TCsrsquo characteristic factorslanding prediction pattern and dynamic prediction patternare proposed which not only can accurately forecast thecharacteristic factors when TCs land but also realize dynam-ically 24 hoursrsquo and 48 hoursrsquo forecast The effect of thelanding prediction pattern is better of which the mean ofGCD is 1446382 km compared with the current 48 hoursrsquoforecast in the South China Sea which is 2226 km Eventhough the mean of GCD in dynamic prediction pattern isno less than three main prediction centers (NHC JMA andNMCC) it is much less than the numerical prediction modelin [25] and the method using satellite scatterometerrsquos mon-itoring data in [12] The forecast methodology proposed inthis paper provides a new method for typhoon warning on

Hainan Islandwithout getting toomuch knowledge ofmeteo-rology involved and thus simplifies the implementation of theprediction process andmeanwhile guarantees the accuracy ofprediction

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

Acknowledgments

This work is in part supported by the National Science-Technology Support Plan in the domain of advanced energytechnology Hainan Power Grid Corporation provided sup-port for this research through ldquoIntegrated DemonstrationProject of Regional Smart Gridrdquo no 2013BAA01B03 DrXinhong Huang a Research Engineer in the Department ofElectrical amp Computer Engineering from the University ofWestern Ontario also puts forward valuable revision com-ments on this paper

References

[1] National Climate Center of China Typhoon ldquoDamreyrdquo HasCaused Serious Damages to Hainan Province 2005 httpncccmagovcnWebsiteindexphpNewsID=1454

Advances in Meteorology 13

[2] J S Pedro F Burstein and A Sharp ldquoA case-based fuzzy mul-ticriteria decision support model for tropical cyclone forecast-ingrdquo European Journal of Operational Research vol 160 no 2pp 308ndash324 2005

[3] X Qin and M Mu ldquoA Study on the reduction of forecast errorvariance by three adaptive observation approaches for tropicalcyclone predictionrdquoMonthlyWeather Review vol 139 no 7 pp2218ndash2232 2011

[4] Y Wang W Zhang and W Fu ldquoBack Propogation(BP)-neuralnetwork for tropical cyclone track forecastrdquo in Proceedings of the19th International Conference on Geoinformatics pp 1ndash4 IEEEShanghai China June 2011

[5] B Feng and J N K Liu ldquoAn adaptive neural network classifierfor tropical cyclone prediction using a two-layer feature selec-torrdquo in Advances in Neural NetworksmdashISNN 2005 vol 3497of Lecture Notes in Computer Science pp 399ndash404 SpringerBerlin Germany 2005

[6] H-J Song S-H Huh J-H Kim C-H Ho and S-K ParkldquoTyphoon track prediction by a support vector machine usingdata reduction methodsrdquo in Computational Intelligence andSecurity vol 3801 of Lecture Notes in Computer Science pp 503ndash511 Springer Berlin Germany 2005

[7] M DeMaria ldquoA simplified dynamical system for tropicalcyclone intensity predictionrdquoMonthly Weather Review vol 137no 1 pp 68ndash82 2009

[8] H R Winterbottom E W Uhlhorn and E P Chassignet ldquoAdesign and an application of a regional coupled atmosphere-ocean model for tropical cyclone predictionrdquo Journal ofAdvances in Modeling Earth Systems vol 4 no 10 Article IDM10002 2012

[9] L M Ma and Z M Tan ldquoImproving the behavior of the cumu-lus parameterization for tropical cyclone prediction convectiontriggerrdquo Atmospheric Research vol 92 no 2 pp 190ndash211 2009

[10] J S Gall I Ginis S-J Lin T P Marchok and J-H ChenldquoExperimental tropical cyclone prediction using the GFDL 25-km-resolution global atmospheric modelrdquo Weather and Fore-casting vol 26 no 6 pp 1008ndash1019 2011

[11] G Lv ldquoThe Northwestern Pacific typhoon track forecast in2005rdquo in Proceedings of the Meteorological Science and Technol-ogy Symposium on Taiwan Strait p 7 Chinese MeteorologicalSociety 2006 (Chinese)

[12] D Zhang Y Zhang T Hu B Xie and J Xu ldquoA comparison ofHY-2 and QuikSCAT vector wind products for tropical cyclonetrack and intensity development monitoringrdquo IEEE Geoscienceand Remote Sensing Letters vol 11 no 8 pp 1365ndash1369 2014

[13] MYingWZhangHYu et al ldquoAnoverviewof theChinamete-orological administration tropical cyclone databaserdquo Journal ofAtmospheric and Oceanic Technology vol 31 no 2 pp 287ndash3012014

[14] G E Forsythe ldquoGeneration and use of orthogonal polynomialsfor data-fittingwith a digital computerrdquo Journal of the Society forIndustrial amp Applied Mathematics vol 5 no 2 pp 74ndash88 1957

[15] KWagstaff C Cardie S Rogers and S Schroedl ldquoConstrainedk-means clustering with background knowledgerdquo in Proceed-ings of the 18th International Conference on Machine Learningvol 1 pp 577ndash584 2001

[16] Y Ye ldquoNeighborhood density method for selecting initialcluster centers in K-mean clusteringrdquo in Proceedings of theWorkshop on Data Mining for Biomedical Applications (PAKDDrsquo06) pp 189ndash198 2006

[17] P Xiong Data Mining Algorithm and Clementine PracticeTsinghua University Press Beijing China 2011 (Chinese)

[18] S L Crawford ldquoExtensions to the CART algorithmrdquo Interna-tional Journal ofMan-Machine Studies vol 31 no 2 pp 197ndash2171989

[19] D M Allen ldquoMean square error of prediction as a criterion forselecting variablesrdquo Technometrics vol 13 pp 469ndash475 1971

[20] S Yu and J Shen ldquoForward and backward algorithms forselecting predictors on the basis of the criterion frompredictionsumof squares and their applicationrdquoActaMeteorologica Sinicavol 1 pp 83ndash90 1988

[21] D Yao and S Yu ldquoThe stepwise algorithm of selecting forecastfactors based on PRESS rulerdquo Journal of Atmospheric Sciencesvol 2 pp 129ndash135 1992 (Chinese)

[22] Y LuMathematical Statistics Methods East China University ofScience and Technology Press Shanghai China 2005 (Chi-nese)

[23] K Xie and B Liu ldquoAn ENSO-forecast independent statisticalmodel for the prediction of annual Atlantic tropical cyclonefrequency in Aprilrdquo Advances in Meteorology vol 2014 ArticleID 248148 11 pages 2014

[24] J Yu J Tang Y Dai and B Yu ldquoThe error and cause analysis ofChinarsquos typhoon path predictionrdquo Journal of Weather vol 6 pp695ndash700 2012

[25] SMa A Qu and Z Yu ldquoThe parallelization of typhoon numer-ical prediction model of and track forecast error analysisrdquo Jour-nal of Applied Meteorology vol 3 pp 322ndash328 2004 (Chinese)

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ClimatologyJournal of

EcologyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

EarthquakesJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom

Applied ampEnvironmentalSoil Science

Volume 2014

Mining

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014

International Journal of

Geophysics

OceanographyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of Computational Environmental SciencesHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal ofPetroleum Engineering

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

GeochemistryHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of

Atmospheric SciencesInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

OceanographyHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

MineralogyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

MeteorologyAdvances in

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Paleontology JournalHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ScientificaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Geological ResearchJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Geology Advances in

Advances in Meteorology 11

Table 7 Prediction results in landing prediction pattern

Area numberThe meanSD of LAT

deviationUnit ∘N

The meanSD of LATdeviationUnit ∘E

The meanSD of APdeviationUnit hpa

The meanSD of WSdeviationUnit ms

1 0984908671 0726406404 9288967882 59540438392 0526806009 0867708973 7994665339 53036389973 0971007805 0634007711 10694183758 70915526374 1153708640 0715006162 10346361761 68151474985 1115206941 0672807103 12113885744 8040353759

Table 8 The results of testing forecast model 1 and forecast model 2

Forecast hour Area number Forecast model(1 or 2)

The meanSD ofLAT deviation

Unit ∘N

The meanSD ofLON deviation

Unit ∘E

The meanSD ofAP deviationUnit hpa

The meanSD ofWS deviationUnit ms

24 h

1 1 0812806515 1029507600 4927839733 37787288582 0700405153 0880207011 5411443187 3963729843

2 1 0913105832 0986407794 6341244247 38268278572 0753806538 1005008670 4917549673 3848730261

3 1 0755505945 1052809093 7814969316 47342417822 0772706183 0894907255 7348966512 5171843368

4 1 0658205365 0879706251 6637355333 36128264862 0640404911 0950006746 6228056617 3752125645

5 1 0689904989 0861806124 6221863820 29503258232 0699605634 1001707915 6320361131 4186732785

48 h

1 1 1219808343 1773514304 9778257246 68402402892 1259808949 1694112488 9455772439 6503044361

2 1 1225009604 2196016746 10977584764 79032593072 1285710131 2033015552 8143463813 6529343273

3 1 1240707253 2165517786 9658069204 75541529892 1325009565 1841415666 8102472309 6245253772

4 1 0916208120 1761813047 7900955968 57899453712 0989607848 1502011732 7097848208 5282637708

5 1 0965007833 1696014327 9110774706 87411640062 1131510845 1973115516 9244968041 6805748502

forecast model (1 or 2) are divided into two groups with thesame number of observation points One group of observa-tion points is used to establish prediction equations and theother group of points is used to test the accuracies of theseequations The results of testing these prediction equationsin forecast model 1 and forecast model 2 are shown inTable 8 In combinationwith formula (3) the calculatedmeanand standard deviation of GCD in two forecast models areshown in Figures 15 and 16 respectively Averaging theresults of five areas it can be obtained that the averagesof the meanstandard deviation of GCD under forecastmodel 1 and forecast model 2 for 24 hoursrsquo forecast fare1505192846156 km and 1415464812509 km respectivelyFor 48 hoursrsquo forecast the averages of the meanstandarddeviation of GCD under two different forecast modelsare 26175171456345 km and 25671091452903 km respec-tively Even though the mean of GCD in dynamic prediction

pattern is no less than three main prediction centers (NHCJMA and NMCC) it is much less than the numerical pre-diction model in [25] the means of which are 18633195 kmbased on System T106 and 16182958 km based on T213 bothfor 2448 hoursrsquo forecast Besides forecast models 1 and 2 areall more accurate than the forecast using satellite scatterome-terrsquos monitoring data in the sense of the standard deviationof GCD and the mean of weed speed error which are1496002 km and 119618ms in [12] It can be seen fromTable 8 and Figures 15 and 16 that the accuracies of forecastmodel 1 and forecast model 2 vary from different areas anddifferent characteristic factors The more accurate forecastmodel can be selected from forecast models 1 and 2 accordingto actual conditionsThe results of statistical significance testsfor each equation used to forecast corresponding charac-teristic factor in forecast model 1 and forecast model 2 areshown in Table 9 which show that 119875 value is much less than

12 Advances in Meteorology

Table 9 The results of statistical significance tests for each equation used to forecast corresponding characteristic factor in forecast model 1and forecast model 2

Forecast hour Area numberForecastmodel(1 or 2)

119877-square119875 valuefor LAT

119877-square119875 valuefor LON

119877-square119875 valuefor AP

119877-square119875 valuefor WS

24 h

1 1 05883952 times 10minus11 06011791 times 10minus9 06222446 times 10minus10 07615547 times 10minus15

2 04493226 times 10minus32 05809593 times 10minus50 07030131 times 10minus68 07731101 times 10minus81

2 1 07177186 times 10minus14 04391366 times 10minus5 07556503 times 10minus16 07995357 times 10minus18

2 06808517 times 10minus101 06753177 times 10minus99 07481213 times 10minus122 07277371 times 10minus114

3 1 06471636 times 10minus11 06559348 times 10minus11 06310183 times 10minus10 06495324 times 10minus10

2 06802646 times 10minus94 07506119 times 10minus114 05509140 times 10minus62 05573984 times 10minus65

4 1 07572105 times 10minus12 05348332 times 10minus8 07594115 times 10minus13 08174247 times 10minus15

2 06405483 times 10minus42 06266164 times 10minus40 06903101 times 10minus44 07652925 times 10minus57

5 1 05031240 times 10minus6 05154151 times 10minus6 08732273 times 10minus14 08524503 times 10minus15

2 06239284 times 10minus35 06464167 times 10minus37 07835698 times 10minus52 07550152 times 10minus48

48 h

1 1 0133700240 03943391 times 10minus5 03208177 times 10minus4 04725534 times 10minus6

2 0217700049 0039903398 0253200016 04275792 times 10minus6

2 1 04515905 times 10minus6 0266800218 06720172 times 10minus10 04622599 times 10minus6

2 04022164 times 10minus36 03095651 times 10minus26 02960505 times 10minus23 03332627 times 10minus27

3 1 0225100218 03935457 times 10minus5 0002607270 01362015962 03073618 times 10minus25 04369213 times 10minus39 01768251 times 10minus12 02479891 times 10minus18

4 1 04992960 times 10minus7 03055274 times 10minus4 04303153 times 10minus5 06249184 times 10minus9

2 03323377 times 10minus16 03270102 times 10minus16 02471170 times 10minus9 03580365 times 10minus16

5 1 04546619 times 10minus5 0255400043 03705191 times 10minus4 03535000122 04219109 times 10minus18 03825344 times 10minus17 02120218 times 10minus7 03322246 times 10minus12

005 in almost every case and prove that the correspondingprediction equation is significant

4 Summary

In this paper the CMA best track datasets from 1949 to 2012are used in combination with data mining technology andstatistical methods to put forward a new methodology toforecast TCsrsquo characteristic factors This methodology canaccurately judge whether TCs land on Hainan Island or notand forecast their characteristic factors (including longitudelatitude the lowest center pressure and wind speed) Theaverage of the probabilities of accurate judgment for landingcriterions is 7470 and the highest accuracy can reach7976 For the forecast of landing TCsrsquo characteristic factorslanding prediction pattern and dynamic prediction patternare proposed which not only can accurately forecast thecharacteristic factors when TCs land but also realize dynam-ically 24 hoursrsquo and 48 hoursrsquo forecast The effect of thelanding prediction pattern is better of which the mean ofGCD is 1446382 km compared with the current 48 hoursrsquoforecast in the South China Sea which is 2226 km Eventhough the mean of GCD in dynamic prediction pattern isno less than three main prediction centers (NHC JMA andNMCC) it is much less than the numerical prediction modelin [25] and the method using satellite scatterometerrsquos mon-itoring data in [12] The forecast methodology proposed inthis paper provides a new method for typhoon warning on

Hainan Islandwithout getting toomuch knowledge ofmeteo-rology involved and thus simplifies the implementation of theprediction process andmeanwhile guarantees the accuracy ofprediction

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

Acknowledgments

This work is in part supported by the National Science-Technology Support Plan in the domain of advanced energytechnology Hainan Power Grid Corporation provided sup-port for this research through ldquoIntegrated DemonstrationProject of Regional Smart Gridrdquo no 2013BAA01B03 DrXinhong Huang a Research Engineer in the Department ofElectrical amp Computer Engineering from the University ofWestern Ontario also puts forward valuable revision com-ments on this paper

References

[1] National Climate Center of China Typhoon ldquoDamreyrdquo HasCaused Serious Damages to Hainan Province 2005 httpncccmagovcnWebsiteindexphpNewsID=1454

Advances in Meteorology 13

[2] J S Pedro F Burstein and A Sharp ldquoA case-based fuzzy mul-ticriteria decision support model for tropical cyclone forecast-ingrdquo European Journal of Operational Research vol 160 no 2pp 308ndash324 2005

[3] X Qin and M Mu ldquoA Study on the reduction of forecast errorvariance by three adaptive observation approaches for tropicalcyclone predictionrdquoMonthlyWeather Review vol 139 no 7 pp2218ndash2232 2011

[4] Y Wang W Zhang and W Fu ldquoBack Propogation(BP)-neuralnetwork for tropical cyclone track forecastrdquo in Proceedings of the19th International Conference on Geoinformatics pp 1ndash4 IEEEShanghai China June 2011

[5] B Feng and J N K Liu ldquoAn adaptive neural network classifierfor tropical cyclone prediction using a two-layer feature selec-torrdquo in Advances in Neural NetworksmdashISNN 2005 vol 3497of Lecture Notes in Computer Science pp 399ndash404 SpringerBerlin Germany 2005

[6] H-J Song S-H Huh J-H Kim C-H Ho and S-K ParkldquoTyphoon track prediction by a support vector machine usingdata reduction methodsrdquo in Computational Intelligence andSecurity vol 3801 of Lecture Notes in Computer Science pp 503ndash511 Springer Berlin Germany 2005

[7] M DeMaria ldquoA simplified dynamical system for tropicalcyclone intensity predictionrdquoMonthly Weather Review vol 137no 1 pp 68ndash82 2009

[8] H R Winterbottom E W Uhlhorn and E P Chassignet ldquoAdesign and an application of a regional coupled atmosphere-ocean model for tropical cyclone predictionrdquo Journal ofAdvances in Modeling Earth Systems vol 4 no 10 Article IDM10002 2012

[9] L M Ma and Z M Tan ldquoImproving the behavior of the cumu-lus parameterization for tropical cyclone prediction convectiontriggerrdquo Atmospheric Research vol 92 no 2 pp 190ndash211 2009

[10] J S Gall I Ginis S-J Lin T P Marchok and J-H ChenldquoExperimental tropical cyclone prediction using the GFDL 25-km-resolution global atmospheric modelrdquo Weather and Fore-casting vol 26 no 6 pp 1008ndash1019 2011

[11] G Lv ldquoThe Northwestern Pacific typhoon track forecast in2005rdquo in Proceedings of the Meteorological Science and Technol-ogy Symposium on Taiwan Strait p 7 Chinese MeteorologicalSociety 2006 (Chinese)

[12] D Zhang Y Zhang T Hu B Xie and J Xu ldquoA comparison ofHY-2 and QuikSCAT vector wind products for tropical cyclonetrack and intensity development monitoringrdquo IEEE Geoscienceand Remote Sensing Letters vol 11 no 8 pp 1365ndash1369 2014

[13] MYingWZhangHYu et al ldquoAnoverviewof theChinamete-orological administration tropical cyclone databaserdquo Journal ofAtmospheric and Oceanic Technology vol 31 no 2 pp 287ndash3012014

[14] G E Forsythe ldquoGeneration and use of orthogonal polynomialsfor data-fittingwith a digital computerrdquo Journal of the Society forIndustrial amp Applied Mathematics vol 5 no 2 pp 74ndash88 1957

[15] KWagstaff C Cardie S Rogers and S Schroedl ldquoConstrainedk-means clustering with background knowledgerdquo in Proceed-ings of the 18th International Conference on Machine Learningvol 1 pp 577ndash584 2001

[16] Y Ye ldquoNeighborhood density method for selecting initialcluster centers in K-mean clusteringrdquo in Proceedings of theWorkshop on Data Mining for Biomedical Applications (PAKDDrsquo06) pp 189ndash198 2006

[17] P Xiong Data Mining Algorithm and Clementine PracticeTsinghua University Press Beijing China 2011 (Chinese)

[18] S L Crawford ldquoExtensions to the CART algorithmrdquo Interna-tional Journal ofMan-Machine Studies vol 31 no 2 pp 197ndash2171989

[19] D M Allen ldquoMean square error of prediction as a criterion forselecting variablesrdquo Technometrics vol 13 pp 469ndash475 1971

[20] S Yu and J Shen ldquoForward and backward algorithms forselecting predictors on the basis of the criterion frompredictionsumof squares and their applicationrdquoActaMeteorologica Sinicavol 1 pp 83ndash90 1988

[21] D Yao and S Yu ldquoThe stepwise algorithm of selecting forecastfactors based on PRESS rulerdquo Journal of Atmospheric Sciencesvol 2 pp 129ndash135 1992 (Chinese)

[22] Y LuMathematical Statistics Methods East China University ofScience and Technology Press Shanghai China 2005 (Chi-nese)

[23] K Xie and B Liu ldquoAn ENSO-forecast independent statisticalmodel for the prediction of annual Atlantic tropical cyclonefrequency in Aprilrdquo Advances in Meteorology vol 2014 ArticleID 248148 11 pages 2014

[24] J Yu J Tang Y Dai and B Yu ldquoThe error and cause analysis ofChinarsquos typhoon path predictionrdquo Journal of Weather vol 6 pp695ndash700 2012

[25] SMa A Qu and Z Yu ldquoThe parallelization of typhoon numer-ical prediction model of and track forecast error analysisrdquo Jour-nal of Applied Meteorology vol 3 pp 322ndash328 2004 (Chinese)

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ClimatologyJournal of

EcologyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

EarthquakesJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom

Applied ampEnvironmentalSoil Science

Volume 2014

Mining

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014

International Journal of

Geophysics

OceanographyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of Computational Environmental SciencesHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal ofPetroleum Engineering

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

GeochemistryHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of

Atmospheric SciencesInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

OceanographyHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

MineralogyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

MeteorologyAdvances in

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Paleontology JournalHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ScientificaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Geological ResearchJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Geology Advances in

12 Advances in Meteorology

Table 9 The results of statistical significance tests for each equation used to forecast corresponding characteristic factor in forecast model 1and forecast model 2

Forecast hour Area numberForecastmodel(1 or 2)

119877-square119875 valuefor LAT

119877-square119875 valuefor LON

119877-square119875 valuefor AP

119877-square119875 valuefor WS

24 h

1 1 05883952 times 10minus11 06011791 times 10minus9 06222446 times 10minus10 07615547 times 10minus15

2 04493226 times 10minus32 05809593 times 10minus50 07030131 times 10minus68 07731101 times 10minus81

2 1 07177186 times 10minus14 04391366 times 10minus5 07556503 times 10minus16 07995357 times 10minus18

2 06808517 times 10minus101 06753177 times 10minus99 07481213 times 10minus122 07277371 times 10minus114

3 1 06471636 times 10minus11 06559348 times 10minus11 06310183 times 10minus10 06495324 times 10minus10

2 06802646 times 10minus94 07506119 times 10minus114 05509140 times 10minus62 05573984 times 10minus65

4 1 07572105 times 10minus12 05348332 times 10minus8 07594115 times 10minus13 08174247 times 10minus15

2 06405483 times 10minus42 06266164 times 10minus40 06903101 times 10minus44 07652925 times 10minus57

5 1 05031240 times 10minus6 05154151 times 10minus6 08732273 times 10minus14 08524503 times 10minus15

2 06239284 times 10minus35 06464167 times 10minus37 07835698 times 10minus52 07550152 times 10minus48

48 h

1 1 0133700240 03943391 times 10minus5 03208177 times 10minus4 04725534 times 10minus6

2 0217700049 0039903398 0253200016 04275792 times 10minus6

2 1 04515905 times 10minus6 0266800218 06720172 times 10minus10 04622599 times 10minus6

2 04022164 times 10minus36 03095651 times 10minus26 02960505 times 10minus23 03332627 times 10minus27

3 1 0225100218 03935457 times 10minus5 0002607270 01362015962 03073618 times 10minus25 04369213 times 10minus39 01768251 times 10minus12 02479891 times 10minus18

4 1 04992960 times 10minus7 03055274 times 10minus4 04303153 times 10minus5 06249184 times 10minus9

2 03323377 times 10minus16 03270102 times 10minus16 02471170 times 10minus9 03580365 times 10minus16

5 1 04546619 times 10minus5 0255400043 03705191 times 10minus4 03535000122 04219109 times 10minus18 03825344 times 10minus17 02120218 times 10minus7 03322246 times 10minus12

005 in almost every case and prove that the correspondingprediction equation is significant

4 Summary

In this paper the CMA best track datasets from 1949 to 2012are used in combination with data mining technology andstatistical methods to put forward a new methodology toforecast TCsrsquo characteristic factors This methodology canaccurately judge whether TCs land on Hainan Island or notand forecast their characteristic factors (including longitudelatitude the lowest center pressure and wind speed) Theaverage of the probabilities of accurate judgment for landingcriterions is 7470 and the highest accuracy can reach7976 For the forecast of landing TCsrsquo characteristic factorslanding prediction pattern and dynamic prediction patternare proposed which not only can accurately forecast thecharacteristic factors when TCs land but also realize dynam-ically 24 hoursrsquo and 48 hoursrsquo forecast The effect of thelanding prediction pattern is better of which the mean ofGCD is 1446382 km compared with the current 48 hoursrsquoforecast in the South China Sea which is 2226 km Eventhough the mean of GCD in dynamic prediction pattern isno less than three main prediction centers (NHC JMA andNMCC) it is much less than the numerical prediction modelin [25] and the method using satellite scatterometerrsquos mon-itoring data in [12] The forecast methodology proposed inthis paper provides a new method for typhoon warning on

Hainan Islandwithout getting toomuch knowledge ofmeteo-rology involved and thus simplifies the implementation of theprediction process andmeanwhile guarantees the accuracy ofprediction

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

Acknowledgments

This work is in part supported by the National Science-Technology Support Plan in the domain of advanced energytechnology Hainan Power Grid Corporation provided sup-port for this research through ldquoIntegrated DemonstrationProject of Regional Smart Gridrdquo no 2013BAA01B03 DrXinhong Huang a Research Engineer in the Department ofElectrical amp Computer Engineering from the University ofWestern Ontario also puts forward valuable revision com-ments on this paper

References

[1] National Climate Center of China Typhoon ldquoDamreyrdquo HasCaused Serious Damages to Hainan Province 2005 httpncccmagovcnWebsiteindexphpNewsID=1454

Advances in Meteorology 13

[2] J S Pedro F Burstein and A Sharp ldquoA case-based fuzzy mul-ticriteria decision support model for tropical cyclone forecast-ingrdquo European Journal of Operational Research vol 160 no 2pp 308ndash324 2005

[3] X Qin and M Mu ldquoA Study on the reduction of forecast errorvariance by three adaptive observation approaches for tropicalcyclone predictionrdquoMonthlyWeather Review vol 139 no 7 pp2218ndash2232 2011

[4] Y Wang W Zhang and W Fu ldquoBack Propogation(BP)-neuralnetwork for tropical cyclone track forecastrdquo in Proceedings of the19th International Conference on Geoinformatics pp 1ndash4 IEEEShanghai China June 2011

[5] B Feng and J N K Liu ldquoAn adaptive neural network classifierfor tropical cyclone prediction using a two-layer feature selec-torrdquo in Advances in Neural NetworksmdashISNN 2005 vol 3497of Lecture Notes in Computer Science pp 399ndash404 SpringerBerlin Germany 2005

[6] H-J Song S-H Huh J-H Kim C-H Ho and S-K ParkldquoTyphoon track prediction by a support vector machine usingdata reduction methodsrdquo in Computational Intelligence andSecurity vol 3801 of Lecture Notes in Computer Science pp 503ndash511 Springer Berlin Germany 2005

[7] M DeMaria ldquoA simplified dynamical system for tropicalcyclone intensity predictionrdquoMonthly Weather Review vol 137no 1 pp 68ndash82 2009

[8] H R Winterbottom E W Uhlhorn and E P Chassignet ldquoAdesign and an application of a regional coupled atmosphere-ocean model for tropical cyclone predictionrdquo Journal ofAdvances in Modeling Earth Systems vol 4 no 10 Article IDM10002 2012

[9] L M Ma and Z M Tan ldquoImproving the behavior of the cumu-lus parameterization for tropical cyclone prediction convectiontriggerrdquo Atmospheric Research vol 92 no 2 pp 190ndash211 2009

[10] J S Gall I Ginis S-J Lin T P Marchok and J-H ChenldquoExperimental tropical cyclone prediction using the GFDL 25-km-resolution global atmospheric modelrdquo Weather and Fore-casting vol 26 no 6 pp 1008ndash1019 2011

[11] G Lv ldquoThe Northwestern Pacific typhoon track forecast in2005rdquo in Proceedings of the Meteorological Science and Technol-ogy Symposium on Taiwan Strait p 7 Chinese MeteorologicalSociety 2006 (Chinese)

[12] D Zhang Y Zhang T Hu B Xie and J Xu ldquoA comparison ofHY-2 and QuikSCAT vector wind products for tropical cyclonetrack and intensity development monitoringrdquo IEEE Geoscienceand Remote Sensing Letters vol 11 no 8 pp 1365ndash1369 2014

[13] MYingWZhangHYu et al ldquoAnoverviewof theChinamete-orological administration tropical cyclone databaserdquo Journal ofAtmospheric and Oceanic Technology vol 31 no 2 pp 287ndash3012014

[14] G E Forsythe ldquoGeneration and use of orthogonal polynomialsfor data-fittingwith a digital computerrdquo Journal of the Society forIndustrial amp Applied Mathematics vol 5 no 2 pp 74ndash88 1957

[15] KWagstaff C Cardie S Rogers and S Schroedl ldquoConstrainedk-means clustering with background knowledgerdquo in Proceed-ings of the 18th International Conference on Machine Learningvol 1 pp 577ndash584 2001

[16] Y Ye ldquoNeighborhood density method for selecting initialcluster centers in K-mean clusteringrdquo in Proceedings of theWorkshop on Data Mining for Biomedical Applications (PAKDDrsquo06) pp 189ndash198 2006

[17] P Xiong Data Mining Algorithm and Clementine PracticeTsinghua University Press Beijing China 2011 (Chinese)

[18] S L Crawford ldquoExtensions to the CART algorithmrdquo Interna-tional Journal ofMan-Machine Studies vol 31 no 2 pp 197ndash2171989

[19] D M Allen ldquoMean square error of prediction as a criterion forselecting variablesrdquo Technometrics vol 13 pp 469ndash475 1971

[20] S Yu and J Shen ldquoForward and backward algorithms forselecting predictors on the basis of the criterion frompredictionsumof squares and their applicationrdquoActaMeteorologica Sinicavol 1 pp 83ndash90 1988

[21] D Yao and S Yu ldquoThe stepwise algorithm of selecting forecastfactors based on PRESS rulerdquo Journal of Atmospheric Sciencesvol 2 pp 129ndash135 1992 (Chinese)

[22] Y LuMathematical Statistics Methods East China University ofScience and Technology Press Shanghai China 2005 (Chi-nese)

[23] K Xie and B Liu ldquoAn ENSO-forecast independent statisticalmodel for the prediction of annual Atlantic tropical cyclonefrequency in Aprilrdquo Advances in Meteorology vol 2014 ArticleID 248148 11 pages 2014

[24] J Yu J Tang Y Dai and B Yu ldquoThe error and cause analysis ofChinarsquos typhoon path predictionrdquo Journal of Weather vol 6 pp695ndash700 2012

[25] SMa A Qu and Z Yu ldquoThe parallelization of typhoon numer-ical prediction model of and track forecast error analysisrdquo Jour-nal of Applied Meteorology vol 3 pp 322ndash328 2004 (Chinese)

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ClimatologyJournal of

EcologyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

EarthquakesJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom

Applied ampEnvironmentalSoil Science

Volume 2014

Mining

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014

International Journal of

Geophysics

OceanographyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of Computational Environmental SciencesHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal ofPetroleum Engineering

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

GeochemistryHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of

Atmospheric SciencesInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

OceanographyHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

MineralogyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

MeteorologyAdvances in

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Paleontology JournalHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ScientificaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Geological ResearchJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Geology Advances in

Advances in Meteorology 13

[2] J S Pedro F Burstein and A Sharp ldquoA case-based fuzzy mul-ticriteria decision support model for tropical cyclone forecast-ingrdquo European Journal of Operational Research vol 160 no 2pp 308ndash324 2005

[3] X Qin and M Mu ldquoA Study on the reduction of forecast errorvariance by three adaptive observation approaches for tropicalcyclone predictionrdquoMonthlyWeather Review vol 139 no 7 pp2218ndash2232 2011

[4] Y Wang W Zhang and W Fu ldquoBack Propogation(BP)-neuralnetwork for tropical cyclone track forecastrdquo in Proceedings of the19th International Conference on Geoinformatics pp 1ndash4 IEEEShanghai China June 2011

[5] B Feng and J N K Liu ldquoAn adaptive neural network classifierfor tropical cyclone prediction using a two-layer feature selec-torrdquo in Advances in Neural NetworksmdashISNN 2005 vol 3497of Lecture Notes in Computer Science pp 399ndash404 SpringerBerlin Germany 2005

[6] H-J Song S-H Huh J-H Kim C-H Ho and S-K ParkldquoTyphoon track prediction by a support vector machine usingdata reduction methodsrdquo in Computational Intelligence andSecurity vol 3801 of Lecture Notes in Computer Science pp 503ndash511 Springer Berlin Germany 2005

[7] M DeMaria ldquoA simplified dynamical system for tropicalcyclone intensity predictionrdquoMonthly Weather Review vol 137no 1 pp 68ndash82 2009

[8] H R Winterbottom E W Uhlhorn and E P Chassignet ldquoAdesign and an application of a regional coupled atmosphere-ocean model for tropical cyclone predictionrdquo Journal ofAdvances in Modeling Earth Systems vol 4 no 10 Article IDM10002 2012

[9] L M Ma and Z M Tan ldquoImproving the behavior of the cumu-lus parameterization for tropical cyclone prediction convectiontriggerrdquo Atmospheric Research vol 92 no 2 pp 190ndash211 2009

[10] J S Gall I Ginis S-J Lin T P Marchok and J-H ChenldquoExperimental tropical cyclone prediction using the GFDL 25-km-resolution global atmospheric modelrdquo Weather and Fore-casting vol 26 no 6 pp 1008ndash1019 2011

[11] G Lv ldquoThe Northwestern Pacific typhoon track forecast in2005rdquo in Proceedings of the Meteorological Science and Technol-ogy Symposium on Taiwan Strait p 7 Chinese MeteorologicalSociety 2006 (Chinese)

[12] D Zhang Y Zhang T Hu B Xie and J Xu ldquoA comparison ofHY-2 and QuikSCAT vector wind products for tropical cyclonetrack and intensity development monitoringrdquo IEEE Geoscienceand Remote Sensing Letters vol 11 no 8 pp 1365ndash1369 2014

[13] MYingWZhangHYu et al ldquoAnoverviewof theChinamete-orological administration tropical cyclone databaserdquo Journal ofAtmospheric and Oceanic Technology vol 31 no 2 pp 287ndash3012014

[14] G E Forsythe ldquoGeneration and use of orthogonal polynomialsfor data-fittingwith a digital computerrdquo Journal of the Society forIndustrial amp Applied Mathematics vol 5 no 2 pp 74ndash88 1957

[15] KWagstaff C Cardie S Rogers and S Schroedl ldquoConstrainedk-means clustering with background knowledgerdquo in Proceed-ings of the 18th International Conference on Machine Learningvol 1 pp 577ndash584 2001

[16] Y Ye ldquoNeighborhood density method for selecting initialcluster centers in K-mean clusteringrdquo in Proceedings of theWorkshop on Data Mining for Biomedical Applications (PAKDDrsquo06) pp 189ndash198 2006

[17] P Xiong Data Mining Algorithm and Clementine PracticeTsinghua University Press Beijing China 2011 (Chinese)

[18] S L Crawford ldquoExtensions to the CART algorithmrdquo Interna-tional Journal ofMan-Machine Studies vol 31 no 2 pp 197ndash2171989

[19] D M Allen ldquoMean square error of prediction as a criterion forselecting variablesrdquo Technometrics vol 13 pp 469ndash475 1971

[20] S Yu and J Shen ldquoForward and backward algorithms forselecting predictors on the basis of the criterion frompredictionsumof squares and their applicationrdquoActaMeteorologica Sinicavol 1 pp 83ndash90 1988

[21] D Yao and S Yu ldquoThe stepwise algorithm of selecting forecastfactors based on PRESS rulerdquo Journal of Atmospheric Sciencesvol 2 pp 129ndash135 1992 (Chinese)

[22] Y LuMathematical Statistics Methods East China University ofScience and Technology Press Shanghai China 2005 (Chi-nese)

[23] K Xie and B Liu ldquoAn ENSO-forecast independent statisticalmodel for the prediction of annual Atlantic tropical cyclonefrequency in Aprilrdquo Advances in Meteorology vol 2014 ArticleID 248148 11 pages 2014

[24] J Yu J Tang Y Dai and B Yu ldquoThe error and cause analysis ofChinarsquos typhoon path predictionrdquo Journal of Weather vol 6 pp695ndash700 2012

[25] SMa A Qu and Z Yu ldquoThe parallelization of typhoon numer-ical prediction model of and track forecast error analysisrdquo Jour-nal of Applied Meteorology vol 3 pp 322ndash328 2004 (Chinese)

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ClimatologyJournal of

EcologyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

EarthquakesJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom

Applied ampEnvironmentalSoil Science

Volume 2014

Mining

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014

International Journal of

Geophysics

OceanographyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of Computational Environmental SciencesHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal ofPetroleum Engineering

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

GeochemistryHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of

Atmospheric SciencesInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

OceanographyHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

MineralogyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

MeteorologyAdvances in

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Paleontology JournalHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ScientificaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Geological ResearchJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Geology Advances in

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ClimatologyJournal of

EcologyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

EarthquakesJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom

Applied ampEnvironmentalSoil Science

Volume 2014

Mining

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014

International Journal of

Geophysics

OceanographyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of Computational Environmental SciencesHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal ofPetroleum Engineering

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

GeochemistryHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of

Atmospheric SciencesInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

OceanographyHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

MineralogyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

MeteorologyAdvances in

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Paleontology JournalHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ScientificaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Geological ResearchJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Geology Advances in