research article an algorithm for retrieving precipitable...

Research ArticleAn Algorithm for Retrieving Precipitable Water Vapor overLand Based on Passive Microwave Satellite Data

Fang-Cheng Zhou12 Xiaoning Song2 Pei Leng3 Hua Wu14 and Bo-Hui Tang14

1State Key Laboratory of Resources and Environment Information System Institute of Geographic Sciences andNatural Resources Research Chinese Academy of Sciences Beijing 100101 China2University of Chinese Academy of Sciences Beijing 100049 China3Key Laboratory of Agri-Informatics Ministry of AgricultureInstitute of Agricultural Resources and Regional PlanningChinese Academy of Agricultural Sciences Beijing 100081 China4Jiangsu Center for Collaborative Innovation in Geographical Information Resource Development and ApplicationNanjing 210023 China

Correspondence should be addressed to Xiaoning Song songxnucasaccn

Received 20 October 2015 Revised 5 January 2016 Accepted 11 January 2016

Academic Editor James Cleverly

Copyright copy 2016 Fang-Cheng Zhou et al This is an open access article distributed under the Creative Commons AttributionLicense which permits unrestricted use distribution and reproduction in any medium provided the original work is properlycited

Precipitable water vapor (PWV) is one of the most variable components of the atmosphere in both space and time In this study apassive microwave-based retrieval algorithm for PWV over land without land surface temperature (LST) data was developed Tobuild the algorithm two assumptions exist (1) land surface emissivities (LSE) at two adjacent frequencies are equal and (2) thereare simple parameterizations that relate transmittance atmospheric effective radiating temperature and PWV Error analyses wereperformed using radiosonde sounding observations fromZhangye China and CE318measurements of Dalanzadgad (43∘3410158403710158401015840N104∘251015840810158401015840E) and Singapore (1∘1710158405210158401015840N 103∘4610158404810158401015840E) sites from Aerosol Robotic Network (AERONET) respectively In Zhangyethe algorithm had a Root Mean Square Error (RMSE) of 439mm and a bias of 036mm on cloud-free days while on cloudy daysthere was an RMSE of 484mm and a bias of 052mm because of the effect of liquid water in cloudsThe validations in Dalanzadgadand Singapore sites showed that the retrieval algorithm had an RMSE of 473mm and a bias of 084mm and the bigger errorsappeared when the water vapor was very dry or very moist

1 Introduction

Precipitable water vapor (PWV) is an important atmosphericcomponent that influencesmany atmospheric processes [1] Itis also crucial for studies of climate change and global warm-ing because water vapor is themost abundant greenhouse gas[2] In the field of quantitative remote sensing knowledgeof PWV can help improve the retrieval accuracy for manyparameters because of the atmospheric correction to remotesensing images [3ndash6]

A number of methods have been developed to estimatePWV either from ground surveys or from remote sens-ing [7] Instrumental measurements of PWV are generallymade with radiosondes ground-based Global PositioningSystems (GPS) microwave radiometers sun photometers

Raman lidar systems Fourier-Transform spectrometers andspace-based satellites with near-infrared or thermal-infraredspectrometers or microwave sounders [8ndash25] In particularradiosonde PWV were the standard PWV for many decades[11] However the use of radiosondes is restricted by theirhigh operational cost and poor spatial resolution Sun pho-tometers operating in the strongwater vapor absorption bandhave been used to determine PWV but they are limitedto be clear of clouds on the path to the Sun [13ndash16] Themain advantage of Raman lidar approach is that the lasersource does not have to turn a specific water vapor absorp-tion line [17ndash21] Dual-frequency ground-based microwaveradiometers have been used for more than 30 years toderive both PWV and cloud liquid water they use frequencybands around the water vapor absorption line at 22235GHz

Hindawi Publishing CorporationAdvances in MeteorologyVolume 2016 Article ID 4126393 11 pageshttpdxdoiorg10115520164126393

2 Advances in Meteorology

Ground-based microwave radiometers can operate in a con-tinuous and unattended manner under almost all weatherconditions with high temporal resolution and low uncer-tainty [22ndash25] Ground-based GPS is an increasingly usefultool for measuring PWV agreeing with radiosonde datawithin 1-2mm [26ndash28] However ground-based microwaveradiometers and GPS methods cannot assess the distributionof PWV on a global scale a problem that also affects groundsurveys because of their poor coverage over oceans and lakesThus in comparisonwith these ground-basedmeasurementsspace-based surveys are an effective way to obtain globalPWVdataMany algorithms have been developed to estimatePWV from satellite data For example MODIS includestwo PWV products derived from observations in the near-infrared and thermal-infrared channels [29] However Li etal [30] found that MODIS algorithms overestimated PWVcompared to data from radiosondes and ground-based GPSMoreover because the MODIS PWV products are based onmeasurements from the near-infrared or thermal-infraredchannels they perform poorly in cloudy conditions JetPropulsion Laboratory (JPL) Atmospheric Infrared Sounder(AIRS) is the first of a new generation of advanced satellite-based atmospheric sounders with the capability of obtaininghigh-vertical resolution profiles of temperature and watervapor [31 32] Although it can producemoisture profile underpartial cloud cover the spatial resolutions of PWV productsare sparse [33]

Comparedwith infraredmeasurements space-based pas-sivemicrowave observations have the advantage of being ableto detect PWV on both cloud-free and cloudy days becausemicrowaves can penetrate clouds However the use of passivemicrowave satellite data for PWVestimation has traditionallybeen applied only over oceans because of their relativelysimple surface conditions compared to the land surface Ingeneral two methods are used to estimate PWV over oceansThe first uses physical algorithms that employ radiativetransfer theory in the model derivations (eg the multiplelinear regression algorithms developed by Schluessel andEmery [34] and the nonlinear iterative algorithms developedby Wentz and Meissner [35]) The second method involvespurely statistical methods with little or no consideration ofthe underlying physics (eg the postlaunch in situ regres-sion algorithms of Basili et al [36]) Compared to oceansdevelopment of PWV retrieval algorithms over land surfaceshas been slow because these surfaces have high pixel-to-pixelvariability in their surface emissivity which makes it diffi-cult to distinguish the atmospheric signal from the surfacebackground [37] Nevertheless several recent approacheshave been proposed that ignore the sensitivity of PWV toland surface emissivity (LSE) Some use a neural networkapproach with a database acquired from the radiative transfermodel [38ndash40] and some take advantage of high frequencywater vapor channels such as 15000 and around 18331 GHzwhich show very low sensitivity to the change in LSE duringPWV estimation [41 42] While these two methods arefeasible both have disadvantages the former has a limitedphysical basis and the latter cannot be used for datasets thatdo not include high frequency water vapor channels such asthose from the Special Sensor Microwave Imager (SSMI)

the ScanningMultichannelMicrowave Radiometer (SMMR)and the Advanced Microwave Scanning Radiometer-EarthObserving System (AMSR-E) Although Deeter [43] pro-posed an algorithm for PWV estimation over both land andocean based on brightness temperatures at 187 and 238GHzthe algorithm did not explain why these two frequencies wereselected [44] Based on these deficiencies the objective ofthis study is to develop an algorithm for retrieving PWVover land using passive microwave satellite data at lowfrequencies

The paper is organized as follows Section 2 describesthe data used in this study Section 3 describes the retrievalalgorithm from a semiempirical point of view Section 4explains the sensitivity analysis and validation with indepen-dent datasets and Section 5 concludes the paper

2 Data

21 The Thermodynamic Initial Guess Retrieval (TIGR)Dataset The TIGR dataset was constructed by the Labora-toire de Meteorologie Dynamique and includes 2311 atmo-spheric profiles selected from 80000 radiosonde datasetsThese profile data are located around the world and representreal atmosphere conditions Each atmospheric profile recordstemperature water vapor and ozone concentrations on agiven pressure grid from the surface to the top of theatmosphere

In this study 378 cloud-free and overland atmosphericprofiles located between 30 and 60 degrees north fromthe TIGR dataset were downloaded from the AtmosphericRadiation Analysis (ARA) website [46] Using these profileswe built a dataset including transmittances atmosphericeffective radiating temperatures and PWV to develop thePWV retrieval model

22 Microwave Radiation Imager (MWRI) Data Launchedon 5 November 2010 FengYun-3B (FY-3B) is the secondsatellite in the Chinese FY satellite series FY-3B crosses theequator in the ascendingmode at 1340 h local solar timeTheMicrowave Radiation Imager (MWRI) on board FY-3B is apassive microwave imager that uses conical scanning at anincidence angle of 53∘ It is a total-power passive radiometerthat measures radiation at five frequencies (1065 187 238365 and 890GHz) with vertical and horizontal polarization[45] A detailed description of the characteristics of the FY-3BMWRI is provided in Table 1

In this study brightness temperatures derived from theFY-3B MWRI with a unified resolution of 10 km wereacquired from the National Satellite Meteorological Centre[47] and used to provide input parameters for the PWVretrieval algorithm

23 Radiosonde Sounding Observations Located in ZhangyeCity Gansu Province China the Zhangye National ClimateObservatory (39∘51015840156810158401015840N 100∘161015840391110158401015840E) was selected asthe validation area for this study The estimated PWV wasverified using PWV data from the observatoryrsquos soundingobservations

Advances in Meteorology 3

Table 1 Performance requirements for the FY-3B MWRI [45]

Frequencies (GHz) 1065 187 238 365 89

Polarization VH VH VH VH VHBandwidth (MHz) 180 200 400 400 3000Sensitivity (K) 05 05 05 05 08Calibration error (K) 15 15 15 15 15Dynamic range (K) 3ndash340Samplesscan 254Main beam efficiency gt90Ground resolution(km)

51 times 85 30 times 50 27 times 45 18 times 30 9 times 15

Scan mode Conical scanningOrbit width (km) 1400Viewing angle (deg) 45Scan period (s) 18

At Zhangye National Climate Observatory L bandVaisala RS92 radiosonde balloons were launched three timesper day at 700ndash800 h 1300ndash1400 h and 1900ndash2000 hlocal solar time from June 1 2012 to August 31 2012 Inthis study we used the observations from 1300 to 1400 hbecause they corresponded to the transit time of FY-3BThe Lband radiosonde system [48] records pressure temperaturerelative humidity wind speed and wind direction profilesThe radiosonde balloons ascend to approximately 30ndash35 kmin the sky namely they record atmospheric information fromthe land surface to 30 km above the ground

Vaisala radiosondes use thin-film capacitance rela-tive humidity sensors The capacitance measured by theradiosonde is proportional to the number of water moleculescaptured at binding sites in the polymer structure whichin turn is proportional to the ambient water vapor concen-tration Miloshevich et al [49] found that RS92 mean biasfor daytime soundings and high solar altitude angles was adry bias that increases from about 5 at 700mbar to 45in the upper troposphere and varies somewhat with relativehumidity The radiosonde PWV would be used as the truevalues to validate the retrieval algorithm consequently themean bias error of Vaisala RS92 measurements was firstlyremoved by an empirical correction [49] in this study

24 Aerosol Robotic Network (AERONET) Sun Photome-ters Aerosol Robotic Network (AERONET) [50] makesmeasurements of the Sun direct irradiance and performsPWV retrievals based on CE318 sun photometers measure-ments in the water vapor absorption band around 940 nmAERONET was created basically for studying columnaraerosol properties and the quality of those retrievalswas well established [50ndash52] Two sites located at Dalan-zadgad (43∘3410158403710158401015840N 104∘251015840810158401015840E in dry area) and Singa-pore (1∘1710158405210158401015840N 103∘4610158404810158401015840E in tropical region) operatedby AERONET were chosen as other validation areas Thevalidation time was still from June 1 2012 to August 31 2012at around 0540 h Greenwich Mean Time (GMT)

In general even Level 20 (best data quality offered byAERONET) PWV products were lower than those obtainedby ground-based microwave radiometers and GPS by sim60ndash90 and sim60ndash80 respectively The AERONET valueswere also lower by approximately 5 than those obtainedfrom the balloon radiosondes [12] Consequently the PWVfrom AERONET need to be corrected based on [12] firstlyand then to be used to validate the retrieval PWV

3 Methodology

31 Model Development The brightness temperature 119879119861

measured using space-basedmicrowave radiometers consistsof three radiative components (1) the upwelling radiationemitted by the atmosphere119879uarr

119886

(2) the radiation emitted fromthe surface that is attenuated by the atmosphere 120591 sdot 120576 sdot 119879

119904

and (3) the downwelling radiation from the atmosphere andthe cosmic background that is reflected by the surface andattenuated by the atmosphere (1 minus 120576)120591(119879darr

119886

+ 119879sky) Assumingthat the land surface is Lambertian and that the upwelling anddownwelling atmospheric radiances are identical [53] theradiative transfer equations at two frequencies can be writtenas follows

119879119861119894= 119879119886119894+ 120591119894sdot 120576119894sdot 119879119904+ (1 minus 120576

119894) 120591119894(119879119886119894+ 119879sky) (1a)

119879119861119895= 119879119886119895+ 120591119895sdot 120576119895sdot 119879119904+ (1 minus 120576

119895) 120591119895(119879119886119895+ 119879sky) (1b)

where 119879119861119894

and 119879119861119895

are the brightness temperatures at 119894and 119895 GHz (Kelvins hereafter referred as K) 120591

119894and 120591

119895

are transmittances 119879119886119894

and 119879119886119895

are the effective radiatingtemperatures of the atmosphere (K) 120576

119894and 120576119895are LSEs 119879

119904

is the land surface temperature (LST K) and 119879sky is thecosmic background radiation temperature (approximately27 K) These equations implicitly include the dependence ofthe radiance on the sensor angle

To reduce the number of parameters we make twoassumptions in this studyThe first assumption is that 120576

119894= 120576119895

which means that 120576 sdot 119879119904can be cancelled out when (1a) and

(1b) are combined This leads to a new combined equationwithout 119879

119904

119879119861119894sdot 120591119895minus 119879119861119895sdot 120591119894+ 119879119886119895sdot 120591119894minus 119879119886119894sdot 120591119895

= (1 minus 120576) 120591119894sdot 120591119895(119879119886119894minus 119879119886119895)

(2)

Several studies [54ndash56] have noted that a significant linearrelationship exists between 120591 and PWV and between 119879

119886and

PWVAs a result the terms119879119886119895sdot120591119894minus119879119886119894sdot120591119895and 120591119894sdot120591119895(119879119886119894minus119879119886119895)

in (2) can be expressed as quadratic and cubic equations aboutPWV However because it is more difficult to solve higherorder equations than first order ones the best forms about thetwo terms and PWV should be linear Therefore we make asecond assumption that there are linear relationships between119879119886119895sdot 120591119894minus119879119886119894sdot 120591119895and PWV and between 120591

119894sdot 120591119895(119879119886119894minus119879119886119895) and

PWV Under this assumption (2) changes form as follows

119879119861119894sdot (1205721sdot PWV + 120572

2) minus 119879119861119895sdot (1205723sdot PWV + 120572

4)

+ (1205725sdot PWV + 120572

6) = (1 minus 120576) (120572

7sdot PWV + 120572

8)

(3)


The retrieval algorithm of PWV over land from passivemicrowave satellite data can be developed by simplifying (3)to the following form

PWV =1205731+ 1205732sdot 120576 + 120573

3sdot 119879119861119894+ 1205734sdot 119879119861119895

1205735+ 1205736sdot 120576 + 120573

7sdot 119879119861119894+ 1205738sdot 119879119861119895

(4)

where 120573119894(119894 = 1 8) are coefficients 120576 is the LSE and

119879119861119894

and 119879119861119895

are brightness temperatures at 119894 and 119895 GHz (K)The retrieval algorithm is based on the two assumptions that(1) LSE at two adjacent frequencies are equal and (2) thereare simple parameterizations that can relate transmittancesatmospheric effective radiating temperatures and PWV Thetwo assumptions depend on the frequency and polarizationof the brightness temperature data Consequently it is impor-tant to verify the accuracy of both assumptions by selectingthe suitable combinations of frequency and polarization

32 Verification of Assumptions To retrieve PWV over landbased on two types of brightness temperature data bothfrequencies should possess two important qualities signifi-cantly different atmospheric absorptions and approximatelyequal LSEs The best way to ensure different atmosphericabsorptions is to use one frequency in the vapor channelwhile the other is not Thus for the FY-3B MWRI 238GHzwas selected because it is the only frequency in the watervapor channel In addition to the 238GHz band two otherbands at 187 and 365GHz were considered as candidatesbecause their LSE are approximately equal to the LSE for238GHz To find the best solution we further analyzed theLSE for 187 238 and 365GHz Because the LSE is influencedby the rough surface and soil moisture content the AdvancedIntegral Equation Model (AIEM) [57] was used to simulateLSE for different surfaces and soil moisture contents AIEMcan predict the backscattering and emissivity of Gaussiancorrelated surfaces as demonstrated by comparisons ofAIEMrsquos results with experimental measurements [58 59] Inthis study LSE were simulated at 187 238 and 365GHzwith an incidence angle of 53∘ a correlation length of 15 cman RMS height of 10 cm (correlation length and RMS heightare two fundamental physical quantities for describing thestatistical characteristic of random rough surfaces) a soiltemperature of 300K and a relative soil moisture contentvarying from 2 to 50 with a step of 2 Using thesimulated data we calculated the variation in LSE at adjacentfrequencies with increasing relative soil moisture contentThe results are shown in Figure 1 and it is apparent that themaximumΔLSE

238-187 is approximately 002 in contrast theΔLSE365-238 is approximately 005 which indicates that the

LSE at 187 GHz is much closer to the LSE at 238GHz than tothat at 365 GHz As a consequence 187 GHz was selected asthe second frequency for the PWV retrieval model

To further verify the first assumption that LSE187

isequal to LSE

238 the polarization combination of the two

frequencies was determined because different polarizationspossess different LSE even at the same frequency The LSEpolarization differences at 187 and 238GHz were comparedto select the combination with a result closest to 0The differ-ent LSE polarization combinations at 187 and 238GHz are

006

005

004

003

002

001

000

ΔLS

E

0 10 20 30 40 50 60

Relative soil moisture content ()

ΔLSE238-187ΔLSE365-238

Figure 1 LSE differences at adjacent frequencies (187 and 238GHzand 238 and 365GHz) versus increasing relative soil moisturecontent

expressed as ΔLSE187v-238v ΔLSE187h-238h ΔLSE187v-238h

and ΔLSE187h-238v (v and h represent vertical and horizontal

polarizations resp) Except for the previous simulationthis database also covers a wide range of surface dielectricconstants which are the averages calculated by the modelsof Dobson et al [60] Mironov et al [61] and Wang andSchmugge [62] The corresponding volumetric soil moisturecontent varied from 2 to 50 with a step of 2 theRMS height varied from 025 cm to 300 cm with a step of025 cm and the correlation length varied from 25 cm to30 cm with a step of 25 cm Figure 2 depicts the box chartsof the four different LSE polarization combinations The boxchart of ΔLSE

187v-238v is closest to the 119910 = 0 line fol-lowed by ΔLSE

187h-238h ΔLSE187v-238h and ΔLSE187h-238vHowever the simulated results showed no significant differ-ence between ΔLSE

187h-238h and ΔLSE187v-238v Moreover

ΔLSE187h-238h showed a minimum standard mean value of

00065 while the values for ΔLSE187v-238v ΔLSE187v-238h

andΔLSE187h-238v were 00069 0064 and 0075 respectively

These results indicate that the differences in LSE for thesame polarization at 187 and 238GHz are less than forthose for different polarizations and the LSE differencesfor horizontal-horizontal polarizations are close to those forvertical-vertical polarizations In addition the box chart forΔLSE187h-238h is the narrowest indicating that the LSE for

horizontal polarization is less sensitive to changes in soildielectric constants than the LSE for vertical polarizationThis insensitivity suggests that the horizontally polarized LSEis relatively stable even with large changes in land surfaceparameters Given that it is important to maintain modelrobustness the horizontal-horizontal polarization combina-tion was selected for the PWV retrieval model

The above analysis allowed us to select the best combi-nation of frequencies and polarization (187 and 238GHzwith horizontal-horizontal polarization) for determination of

Advances in Meteorology 5Δ

LSE

02

04

00

minus02

minus04

ΔLSE187v-238v ΔLSE187v-238hΔLSE187h-238h ΔLSE187h-238v

stands for 99 and 1 valuesstands for max and min valuesstands for the mean value

Figure 2 Box charts of four LSE polarization differencesΔLSE

187v-238v ΔLSE187h-238h ΔLSE187v-238h and ΔLSE187h-238v

PWV Using these frequencies and polarization we furtherinvestigate our second assumption

To verify the relationships between the terms119879119886238sdot120591187minus

119879119886187sdot 120591238

and PWV as well as between 120591187sdot 120591238(119879119886187minus

119879119886238) and PWV a database including 119879

119886187 119879119886238

120591187

120591238

and PWV is necessary In this study MonoRTM wasused to simulate these parameters along with TIGR profilesMonoRTM is a radiative transfer model designed to processa number of monochromatic wavenumber values [63] It isparticularly useful in the microwave spectrum The relation-ships between PWV and 119879

119886238sdot 120591187minus 119879119886187sdot 120591238

andbetween PWV and 120591

187sdot 120591238(119879119886187minus 119879119886238) are shown

in Figure 3 As we expected there is a significant linearrelationship between 119879

119886238sdot 120591187minus 119879119886187sdot 120591238

and PWVwith a coefficient of determination (1198772) of 099 and a RootMean Square Error (RMSE) of 064K A linear relationshipalso exists between 120591

187sdot 120591238(119879119886187minus119879119886238) and PWV with

1198772 of 098 and an RMSE of 074K We note that Figure 3(b)

displays more of a second order trend (with an 1198772 of 099not shown in the figure) than a linear trend However it ismore difficult to solve a second order equation about PWVthan a first order one in addition the trend of the secondorder equation is not much more significant than the trendof the first order one (1198772 of 099 versus 1198772 of 098) Thus weconsider the first order equation for 120591

187sdot120591238(119879119886187minus119879119886238)

and PWV to be acceptable especially when PWV rangesfrom 5 to 40mm These results demonstrate that the secondassumption is satisfied

As stated previously verification of the two assumptionsprovides a statistical basis for the PWV retrieval algorithmAssuming that the lowest air temperature of each TIGRprofile is the LST [64] and using randomly generatednumbers between 07 and 099 as the LSE the brightnesstemperatures at 187 and 238GHz were calculated with theforward models in (1a) and (1b) respectively Afterwardsthe coefficients (120573

119894 see Table 2) were calculated using the

Table 2 Regression coefficients of the PWV retrieval algorithmapplied only when using 187 and 238GHz with horizontal polar-ization

Coefficients 1205731

1205732

1205733

1205734

1205735

1205736

1205737

1205738

Values minus77404 minus21141 minus1157 1572 249 minus1915 02 minus011

PWV LSE and brightness temperatures in (4) based on thenonlinear least squares optimization

4 Results and Discussions

41 Sensitivity Analysis

411 Sensitivity of the PWV Retrieval Model to LSE ErrorOur retrieval algorithm made the assumption that LSEwas known through measurement or retrieval Howevereither method can generate errors To assess the sensitivityof the PWV retrieval model to LSE error two series ofnormally distributed errors errorLSE sim 119873(0 001

2

) anderrorLSE sim 119873(0 002

2

) (the white noises whose mean is 0 andvariances are 001 and 002 resp) were added to the originalLSE data Using these new LSE data as input for the retrievalalgorithm and leaving the other parameters unchanged twonew sets of estimated PWV were obtained with (4) Figure 4shows the histograms of errors for the newly estimated PWVand the original PWV It is apparent that there is an RMSE of060mm and a bias of 0019mmbetween the newly estimatedPWV and the originally calculated values and that no LSEerror existsMoreover over 99of the errors are under 2mmwhich indicates that the PWV model is not very sensitive toerrors in the LSE Similarly for LSE errors with a standarddeviation of 002 there is an RMSE of 125mm and a biasof 0056mm between the newly estimated and the originalPWVvaluesThis result further confirms that the error in LSEdoes not significantly affect the accuracy of PWV estimatedusing the proposed model

412 Sensitivity of the PWV Retrieval Model to Bright-ness Temperature Error In the retrieval model 119879

119861187and

119879119861238

can be directly obtained from the brightness tem-peratures at 187 and 238GHz with horizontal polariza-tion Consequently errors are mainly introduced becauseof the noise-equivalent differential temperature (NEΔ119879)of the microwave radiometers which are system errorsFor FY-3B the NEΔ119879 values for brightness temperaturesat 187 and 238GHz are respectively 05 and 08 KTo assess the sensitivity of the PWV retrieval model toerrors in 119879

119861 two series of normally distributed errors

error119879119861187sim 119873(0 05

2

) and error119879119861238sim 119873(0 08

2

) (the whitenoises whose mean is 0 and variances are 05 and 08 resp)were added to the original 119879

119861187and 119879

119861238 respectively

Thesemodified valueswere then used to obtain new estimatesof PWV while leaving the other parameters unchangedFigure 5 shows the histograms of errors between the newlyestimated PWV and the original PWV An RMSE of 123mmand a bias of minus0014mm result when the standard deviationof 119879119861187

is 05 K Moreover over 90 of errors are less than


50

40

30

20

10

0

Ta238120591187minusTa187120591238

(K)

0 10 20 30 40 50

PWV (mm)

Intercept 186KSlope 089KmmR2= 099

RMSE = 064K

(a)

0

minus5

minus10

minus15

minus20

minus25

minus30

120591187120591238(T

a187minusTa238)

(K)

0 10 20 30 40 50

PWV (mm)

Intercept minus35KSlope minus064KmmR2= 098

RMSE = 074K

(b)

Figure 3 Linear relationships between (a) 119879119886238

sdot 120591187

minus 119879119886187

sdot 120591238

and PWV and (b) 120591187

sdot 120591238

(119879119886187

minus 119879119886238

) and PWV

ΔPWV (mm)minus3 minus2 minus1 0 1 2 3 4

Freq

uenc

y (

)

30

20

10

0

Variance 001Bias 0019mmRMSE 060mm

(a)


Freq

uenc

y (

)30

20

25

10

15

0

5


(b)

Figure 4 Histograms of errors between the newly estimated PWV (using the new LSE which includes added errors) and the original PWVwhen (a) the standard deviation of the LSE errors is 001 and (b) the standard deviation of the LSE errors is 002

2mm and less than 03 of errors are over 4mm which indi-cates that the PWV model is not sensitive to error in 119879

119861187

On the contrary a 119879119861238

error with a standard deviation of08 K results in an RMSE of 205mm which is greater thanthe RMSE for LSE and119879

119861187 In addition approximately 67

of errors are less than 2mm and approximately 5 of errorsare greater than 4mm Thus the PWV algorithm is moresensitive to errors in 119879

119861238than to errors in LSE and 119879

119861187

In realistic case the errors normally occur simultaneously inboth channels thus we discuss the sensitivity of the PWVretrieval model to both channels errors (05 K for 119879

119861187and

08 K for 119879119861238

not shown) An RMSE of 243mm and a biasof 00064mm are obtained with 61 of errors less than 2mmand approximately 9 of errors greater than 4mm It is foundthat the RMSE caused by both channels errors is not the sumof two separate RMSE caused by both119879

119861238and119879119861238

errors

This is because the different error sources are counteracted inpractice Fortunately however because the source of 119879

119861238

errors is NEΔ119879 these errors can be dealt with by promotingbetter design and manufacture of microwave radiometersThese improvements could further reduce the sensitivity ofthe PWV retrieval model to errors in 119879

119861238in the future

42 Validation To further assess the accuracy of the PWVretrieval model validations were performed using L bandradiosonde sounding observations and AERONET CE318-PWV products respectively FY-3B MWRI brightness tem-peratures at 187 and 238GHz with horizontal polarizationwere extracted and used as input data of the algorithmBecause a radiosondersquos investigation radius can be up to200 km (depending on the wind speed and direction atvarious altitudes) the radiosonde sounding observations can


Freq

uenc

y (

)

20

25

10

15

0

5

ΔPWV (mm)minus6minus8 minus4 minus2 0 2 4 6

Variance 05KBias minus0014mmRMSE 123mm

(a)

Freq

uenc

y (

)

20

25

10

15

0

5

ΔPWV (mm)minus10 minus5 0 5 10 15

Variance 08KBias 000097mmRMSE 205mm

(b)

Figure 5Histograms of errors between the newly estimated PWV (using the new brightness temperatures with added errors) and the originalPWV when (a) the standard deviation of the 119879

119861187

errors is 05 K and (b) the standard deviation of the 119879119861238

errors is 08 K

represent the atmospheric conditions of an entire MWRIpixel Considering the orbital gaps in the FY-3B MWRI dataonly 58 available data were selected including 34 cloudy daysand 24 cloud-free days Because no available daily or hourlyLSE products exist the instantaneous LSE values used inthe validation were calculated using the following equation(adapted from (1a) and (1b))

LSE =119879119861187minus 119879119886187minus 120591187(119879119886187+ 119879sky)

120591187sdot 1198791015840

119904

minus 120591187(119879119886187+ 119879sky)

(5)

where 119879119861187

is the brightness temperature at 187 GHz (K)120591187

is the transmittance 119879119886187

is the effective radiatingtemperature of the atmosphere (K) 1198791015840

119904

is the LST (becauseno infrared LST products exist for cloudy days measuredLST values from Zhangye National Climate Observatorywere used) (K) and 119879sky is the cosmic background radiationtemperature (27 K)

Figure 6 shows comparisons of the corrected radiosondePWV and the retrieved PWV for both cloudy and cloud-free days On cloud-free days the retrieval algorithm over-estimates the PWV with an RMSE of 439mm and a biasof 036mm compared to the radiosonde PWV These errorsmainly derive from three sources First the model itself errorstill exists Second the detection zone of radiosonde balloonsranges from the land surface to an altitude of approximately30ndash35 km in contrast the FY-3B satellite orbits the earth at analtitude of 836 km This difference in detection zones causesthe retrieved PWV from FY-3B to theoretically be more thanthe radiosonde PWV This explains the positive bias in thecomparisonThird there are errors from themeasured119879

119861187

and 119879119861238

and the retrieved LSE and analyzed hereinbeforethe effects of these errors are small except 119879

119861238 On cloudy

days the retrieval algorithm overestimates the PWV withan RMSE of 484mm and a bias of 052mm The additionalerrors on cloudy days are mainly caused by liquid waterand water ice in clouds which affect the transmittance

Retr

ieve

d PW

V (m

m)

50

40

30

20

10

0

50403020100

Corrected radiosonde PWV (mm)

Cloudy days

Cloud-free days

1 1 line

RMSE = 484mm bias = 052mm R2 = 057

RMSE = 439mm bias = 036mm R2 = 046

Figure 6 Comparison of the corrected radiosonde PWV andretrieved PWV for cloud-free days (red circles) and cloudy days(blue squares) at Zhangye National Climate Observatory

atmospheric effective radiating temperature and brightnesstemperature because of volume scattering and absorption ofwater Figure 7 compares the corrected AERONET PWV atDalanzadgad (in dry area) and Singapore (in tropical region)and the retrieved PWV on cloud-free conditions On thewhole the retrieval algorithm shows a satisfactory accuracyfor the dry area and tropical region with a total RMSE of473mm and a bias of 084mm But it is noted that theretrieval algorithm reveals an overestimation when PWV islower than 10mm and an underestimation when PWV ishigher than 40mmThismay be because we consider the first

8 Advances in MeteorologyRe

trie

ved

PWV

(mm

)

50

60

40

30

20

10

0

50 60403020100

Corrected AERONET PWV (mm)

1 1 line

DalanzadgadSingapore

RMSE = 473mm bias = 084mm R2 = 093

Figure 7 Comparison of the corrected AERONET PWV and theretrieved PWV for cloud-free conditions at Dalanzadgad (in dryarea) and Singapore (in tropical region) sites

order equation for 120591187sdot 120591238(119879119886187minus119879119886238) and PWV rather

than second order leading to the errors when the water vaporis very dry or very moist

43 Discussions As the input parameters of themodel LSE isvery important because it often has greater error compared tothe brightness temperatures as the high-level remotely sensedproduct But sensitivity analysis reveals that this algorithmis insensitive to the LSE error Besides this algorithm doesnot need LST as the input data which further reducesthe potential error from LST product Both suggest thatthis algorithm has better robustness compared to previousresearches [43 44] On cloudy days although the accuracyis inferior to that of cloud-free days this retrieval algorithmshows the ability to provide the PWV distribution at whichtime no infrared PWV products are available [65]The effectsof water scattering and absorption in clouds are ignored inour retrieval algorithm because of their complexity whileone thing which fascinates us is that from Figure 6 we canfind that RMSE and bias on cloudy days are only a littlehigher than those on cloud-free days the performancesof RMSE and bias are far from the expectation that theaccuracy degrades significantly on cloudy days This maybe caused by two reasons first the cloud liquid water islittle on these cloudy days and the PWV retrieval algorithmis insensitive to it second the validation data are sparseand under representation To further improve the algorithmrsquosaccuracy wewill consider thewater scattering and absorptionfor microwave radiation in future studies

5 Conclusions

An algorithm for retrieving PWV over land was devel-oped based on passive microwave brightness temperaturedata at 187 and 238GHz with horizontal polarization The

algorithm is based on two assumptions first that LSE isequal at two adjacent frequencies and second that simpleparameterizations exist which relate transmittance atmo-spheric effective radiating temperature and PWV Usingsimulated data from two widely used and physically basedmodels AIEM and MonoRTM we confirmed these twoassumptions To assess the precision of the retrieval modelL band radiosonde sounding observations and AERONETsun photometers were collected and were used to verify theretrieved PWVOur results revealed a significant relationshipbetween the retrieved PWV and the corrected radiosondeand AERONET PWV products both on cloud-free and oncloudy days and both in dry area and in tropical region

It is difficult to completely distinguish the atmosphericsignal from the surface background The advance in thisretrieval model lies in the reduction in input parametersWithout LST observations this PWV retrievalmodel ismucheasier and reduces the potential errors from the LSTproductsMeanwhile this model possesses some physical foundationbecause it is developed from the radiative transfer equationValidation suggests that the algorithm performs well oncloud-free days but worse on cloudy days due to the volumescattering and absorption of water which is ignored in thismodel To generalize the model the effects of cloud liquidwater on the PWV retrieval algorithm should be consideredin future work The accuracy of the model could also befurther assessed using additional PWV data

Conflict of Interests

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

Acknowledgments

This work was supported by the National Natural ScienceFoundation of China under Grant no 41231170 and anindependent innovation project of State Key Laboratory ofResource and Environmental Information System CAS (no08R8B6B0YA)The authors thank the ARA for providing theTIGR dataset the AER for providing the MonoRTM codeand WestDC (httpwestdcwestgisaccn) for providing thedataset of radiosonde sounding observations from ZhangyeNational Climate Observatory

References

[1] M Bevis S Businger T A Herring C Rocken R AAnthes and R H Ware ldquoGPS meteorology remote sensing ofatmospheric water vapor using the global positioning systemrdquoJournal of Geophysical Research vol 97 no 14 pp 15787ndash158011992

[2] I M Held and B J Soden ldquoWater vapor feedback and globalwarmingrdquo inAnnual Review of Energy and the Environment pp441ndash475 Annual Reviews Inc Palo Alto Calif USA 2000

[3] C O Justice T F Eck D Tanre and B N Holben ldquoThe effectof water-vapor on the normalized difference vegetation indexderived for the sahelian region from NOAA AVHRR datardquo


International Journal of Remote Sensing vol 12 no 6 pp 1165ndash1187 1991

[4] J A Sobrino Z-L Li and M P Stoll ldquoImpact of the atmo-spheric transmittance and total water vapor content in thealgorithms for estimating satellite sea surface temperaturesrdquoIEEE Transactions on Geoscience and Remote Sensing vol 31no 5 pp 946ndash952 1993

[5] J A Sobrino Z-L Li M P Stoll and F Becker ldquoImprovementsin the split-window technique for land surface temperaturedeterminationrdquo IEEE Transactions on Geoscience and RemoteSensing vol 32 no 2 pp 243ndash253 1994

[6] B-H Tang Y Y Bi Z-A Li and J Xia ldquoGeneralized split-window algorithm for estimate of land surface temperaturefrom Chinese geostationary FengYun meteorological satellite(FY-2C) datardquo Sensors vol 8 no 2 pp 933ndash951 2008

[7] J-P Chaboureau A Chedin and N A Scott ldquoRemote sensingof the vertical distribution of atmospheric water vapor fromthe TOVS observations method and validationrdquo Journal ofGeophysical Research Atmospheres vol 103 no 8 pp 8743ndash8752 1998

[8] H Wu L Ni N Wang Y Qian B-H Tang and Z-LLi ldquoEstimation of atmospheric profiles from hyperspectralinfrared IASI sensorrdquo IEEE Journal of Selected Topics in AppliedEarth Observations and Remote Sensing vol 6 no 3 pp 1485ndash1494 2013

[9] Z-L Li L Jia Z B Su Z M Wan and R H Zhang ldquoA newapproach for retrieving precipitable water from ATSR2 split-window channel data over land areardquo International Journal ofRemote Sensing vol 24 no 24 pp 5095ndash5117 2003

[10] S Peng B-H TangHWu R Tang and Z-L Li ldquoEstimating ofthe total atmospheric precipitable water vapor amount from theChinese new generation polar orbit FengYun meteorologicalsatellite (FY-3) datardquo in Proceedings of the IEEE InternationalGeoscience and Remote Sensing Symposium (IGARSS rsquo14) pp3029ndash3032 IEEE Quebec Canada July 2014

[11] A E Niell A J Coster F S Solheim et al ldquoComparison ofmeasurements of atmospheric wet delay by radiosonde watervapor radiometer GPS and VLBIrdquo Journal of Atmospheric andOceanic Technology vol 18 no 6 pp 830ndash850 2001

[12] D Perez-Ramırez D N Whiteman A Smirnov et al ldquoEvalua-tion of AERONET precipitable water vapor versus microwaveradiometry GPS and radiosondes at ARM sitesrdquo Journal ofGeophysical Research Atmospheres vol 119 no 15 pp 9596ndash9613 2014

[13] R N Halthore T F Eck B N Holben and B L MarkhamldquoSun photometric measurements of atmospheric water vaporcolumn abundance in the 940-nm bandrdquo Journal of GeophysicalResearch Atmospheres vol 102 no 4 pp 4343ndash4352 1997

[14] B N Holben and T F Eck ldquoPrecipitable water in the Sahelmeasured using sun photometryrdquo Agricultural and Forest Mete-orology vol 52 no 1-2 pp 95ndash107 1990

[15] J JMichalsky J C Liljegren and L CHarrison ldquoA comparisonof Sun photometer derivations of total column water vaporand ozone to standard measures of same at the Southern GreatPlains Atmospheric Radiation Measurement siterdquo Journal ofGeophysical Research Atmospheres vol 100 no 12 pp 25995ndash26003 1995

[16] F E Volz ldquoEconomical multispectral sun photometer formeasurements of aerosol extinction from044 120583mto 16 120583mandprecipitable waterrdquo Applied Optics vol 13 no 8 pp 1732ndash17331974

[17] D A Leonard ldquoObservation of raman scattering from theatmosphere using a pulsed nitrogen ultraviolet laserrdquo Naturevol 216 no 5111 pp 142ndash143 1967

[18] J A Cooney ldquoMeasurements on the raman component of laseratmospheric backscatterrdquo Applied Physics Letters vol 12 no 2pp 40ndash42 1968

[19] W B Grant ldquoDifferential absorption and Raman lidar for watervapor profile measurements a reviewrdquoOptical Engineering vol30 no 1 pp 40ndash48 1991

[20] M Froidevaux C W Higgins V Simeonov et al ldquoA Ramanlidar to measure water vapor in the atmospheric boundarylayerrdquo Advances in Water Resources vol 51 pp 345ndash356 2013

[21] D N Whiteman K Rush S Rabenhorst et al ldquoAirborne andground-based measurements using a high-performance ramanlidarrdquo Journal of Atmospheric and Oceanic Technology vol 27no 11 pp 1781ndash1801 2010

[22] D C Hogg F O Guiraud J B Snider M T Decker and ER Westwater ldquoA steerable dual-channel microwave radiometerfor measurement of water vapor and liquid in the troposphererdquoJournal of Applied Meteorology vol 22 no 5 pp 789ndash806 1983

[23] F O Guiraud J Howard and D C Hogg ldquoA dual-channelmicrowave radiometer for measurement of precipitable watervapor and liquidrdquo IEEE Transactions on Geoscience Electronicsvol 17 no 4 pp 129ndash136 1979

[24] D Cimini T J Hewison and LMartin ldquoComparison of bright-ness temperatures observed from ground-based microwaveradiometers during TUCrdquo Meteorologische Zeitschrift vol 15no 1 pp 19ndash25 2006

[25] M P Cadeddu J C Liljegren and D D Turner ldquoThe atmo-spheric radiation measurement (ARM) program network ofmicrowave radiometers instrumentation data and retrievalsrdquoAtmospheric Measurement Techniques vol 6 no 9 pp 2359ndash2372 2013

[26] T R Emardson J Johansson and G Elgered ldquoThe systematicbehavior of water vapor estimates using four years of GPSobservationsrdquo IEEE Transactions on Geoscience and RemoteSensing vol 38 no 1 pp 324ndash329 2000

[27] Y A Liou Y T Teng T V Hove and J C Liljegren ldquoCom-parison of precipitable water observations in the near tropicsby GPS microwave radiometer and radiosondesrdquo Journal ofApplied Meteorology vol 40 no 1 pp 5ndash15 2001

[28] J Chen and G Li ldquoDiurnal variations of ground-based GPS-PWV under different solar radiation intensity in the ChengduPlainrdquo Journal of Geodynamics vol 72 pp 81ndash85 2013

[29] B-C Gao and Y J Kaufman ldquoWater vapor retrievals usingModerate Resolution Imaging Spectroradiometer (MODIS)near-infrared channelsrdquo Journal of Geophysical Research DAtmospheres vol 108 no 13 pp 1ndash10 2003

[30] Z Li J-P Muller and P Cross ldquoComparison of precipitablewater vapor derived from radiosonde GPS and moderate-resolution imaging spectroradiometer measurementsrdquo Journalof Geophysical Research Atmospheres vol 108 no 20 pp 10-1ndash10-12 2003

[31] D C Tobin H E Revercomb R O Knuteson et al ldquoAtmo-spheric radiation measurement site atmospheric state best esti-mates for atmospheric infrared sounder temperature and watervapor retrieval validationrdquo Journal of Geophysical Research DAtmospheres vol 111 no 9 pp 831ndash846 2006

[32] H H Aumann M T Chahine C Gautier et alldquoAIRSAMSUHSB on the aqua mission design scienceobjectives data products and processing systemsrdquo IEEE


Transactions on Geoscience and Remote Sensing vol 41 no 2pp 253ndash264 2003

[33] J Susskind C D Barnet and J M Blaisdell ldquoRetrieval ofatmospheric and surface parameters from AIRSAMSUHSBdata in the presence of cloudsrdquo IEEE Transactions on Geoscienceand Remote Sensing vol 41 no 2 pp 390ndash409 2003

[34] P Schluessel and W J Emery ldquoAtmospheric water vapour overoceans from SSMI measurementsrdquo International Journal ofRemote Sensing vol 11 no 5 pp 753ndash766 1990

[35] F Wentz and T Meissner ldquoAMSR ocean algorithm version 2algorithm theoretical basis documentrdquo Tech Rep 121599A-1Remote Sensing System Santa Rosa Calif USA 2000

[36] P Basili S Bonafoni V Mattioli P Ciotti and N PierdiccaldquoMapping the atmospheric water vapor by integratingmicrowave radiometer and GPS measurementsrdquo IEEETransactions on Geoscience and Remote Sensing vol 42 no 8pp 1657ndash1665 2004

[37] S-A Boukabara K Garrett and CWanchun ldquoGlobal coverageof total precipitable water using a microwave variational algo-rithmrdquo IEEE Transactions on Geoscience and Remote Sensingvol 48 no 10 pp 3608ndash3621 2010

[38] F Aires C Prigent W B Rossow and M Rothstein ldquoA newneural network approach including first guess for retrieval ofatmospheric water vapor cloud liquid water path surface tem-perature and emissivities over land from satellite microwaveobservationsrdquo Journal of Geophysical Research Atmospheresvol 106 no 14 pp 14887ndash14907 2001

[39] P Basili S Bonafoni V Mattioli et al ldquoNeural-networkretrieval of integrated precipitable water vapor over land fromsatellite microwave radiometerrdquo in Proceedings of the 11th Spe-cialist Meeting onMicrowave Radiometry and Remote Sensing ofthe Environment (MicroRad rsquo10) pp 161ndash166 IEEEWashingtonDC USA March 2010

[40] S Bonafoni V Mattioli P Basili P Ciotti and N PierdiccaldquoSatellite-based retrieval of precipitable water vapor over landby using a neural network approachrdquo IEEE Transactions onGeoscience and Remote Sensing vol 49 no 9 pp 3236ndash32482011

[41] J Miao K Kunzi G Heygster T A Lachlan-Cope and JTurner ldquoAtmospheric water vapor over Antarctica derivedfrom Special Sensor MicrowaveTemperature 2 datardquo Journal ofGeophysical Research Atmospheres vol 106 no 10 pp 10187ndash10203 2001

[42] K-P Johnsen J Miao and S Q Kidder ldquoComparison of atmo-sphericwater vapor overAntarctica derived fromCHAMPGPSand AMSU-B datardquo Physics and Chemistry of the Earth PartsABC vol 29 no 2-3 pp 251ndash255 2004

[43] M N Deeter ldquoA new satellite retrieval method for precipitablewater vapor over land and oceanrdquo Geophysical Research Lettersvol 34 no 2 Article ID L02815 2007

[44] D Ji and J Shi ldquoWater vapor retrieval over cloud cover areaon land using AMSR-E and MODISrdquo IEEE Journal of SelectedTopics in Applied Earth Observations and Remote Sensing vol 7no 7 pp 3105ndash3116 2014

[45] H Yang X Zou X Li and R You ldquoEnvironmental datarecords from FengYun-3B microwave radiation imagerrdquo IEEETransactions on Geoscience and Remote Sensing vol 50 no 12pp 4986ndash4993 2012

[46] Thermodynamic Initial Guess Retrieval (TIGR) August 2015httparaabctlmdpolytechniquefrindexphppage=tigr

[47] FENGYUN Satellite Data Center August 2015 httpsatellitecmagovcnPortalSiteDefaultaspx

[48] J C Bian H B Chen H Vomel Y J Duan Y J Xuan andD R Lu ldquoIntercomparison of humidity and temperature sen-sors GTS1 Vaisala RS80 and CFHrdquo Advances in AtmosphericSciences vol 28 no 1 pp 139ndash146 2011

[49] L M Miloshevich H Vomel D N Whiteman and TLeblanc ldquoAccuracy assessment and correction of Vaisala RS92radiosonde water vapor measurementsrdquo Journal of GeophysicalResearch Atmospheres vol 114 no 11 pp 1013ndash1033 2009

[50] B N Holben T F Eck I Slutsker et al ldquoAERONETmdashafederated instrument network and data archive for aerosolcharacterizationrdquo Remote Sensing of Environment vol 66 no1 pp 1ndash16 1998

[51] A Smirnov B N Holben T F Eck O Dubovik and ISlutsker ldquoCloud-screening and quality control algorithms forthe AERONET databaserdquo Remote Sensing of Environment vol73 no 3 pp 337ndash349 2000

[52] O Dubovik A Smirnov B N Holben et al ldquoAccuracyassessments of aerosol optical properties retrieved fromAerosolRobotic Network (AERONET) sun and sky radiance measure-mentsrdquo Journal of Geophysical Research Atmospheres vol 105no 8 pp 9791ndash9806 2000

[53] Z-L Li B-H Tang HWu et al ldquoSatellite-derived land surfacetemperature current status and perspectivesrdquo Remote Sensingof Environment vol 131 pp 14ndash37 2013

[54] Z-L Liu HWu B-H Tang S Qiu and Z-L Li ldquoAtmosphericcorrections of passive microwave data for estimating landsurface temperaturerdquo Optics Express vol 21 no 13 pp 15654ndash15663 2013

[55] Z-L Liu HWu B-H Tang S Qiu and Z-L Li ldquoAn empiricalrelationship of bare soil microwave emissions between verticaland horizontal polarization at 1065GHzrdquo IEEE Geoscience andRemote Sensing Letters vol 11 no 9 pp 1479ndash1483 2014

[56] E R Westwater J B Snider and M J Falls ldquoGround-based radiometric observations of atmospheric emission andattenuation at 206 3165 and 900 GHz a comparison ofmeasurements and theoryrdquo IEEE Transactions on Antennas andPropagation vol 38 no 10 pp 1569ndash1580 1990

[57] K S Chen T-D Wu L Tsang Q Li J Shi and A K FungldquoEmission of rough surfaces calculated by the integral equa-tion method with comparison to three-dimensional momentmethod simulationsrdquo IEEE Transactions on Geoscience andRemote Sensing vol 41 no 1 pp 90ndash101 2003

[58] J C Shi L M Jiang L X Zhang K-S Chen J-P Wigneronand A Chanzy ldquoA parameterized multifrequency-polarizationsurface emission modelrdquo IEEE Transactions on Geoscience andRemote Sensing vol 43 no 12 pp 2831ndash2841 2005

[59] L Chen J Shi J-P Wigneron and K-S Chen ldquoA param-eterized surface emission model at L-band for soil moistureretrievalrdquo IEEE Geoscience and Remote Sensing Letters vol 7no 1 pp 127ndash130 2010

[60] M C Dobson F T Ulaby M T Hallikainen and M AEl-Rayes ldquoMicrowave dielectric behavior of wet soilmdashpart IIdielectric mixingmodelsrdquo IEEE Transactions on Geoscience andRemote Sensing vol 23 no 1 pp 35ndash46 1985

[61] V L Mironov M C Dobson V H Kaupp S A Komarov andV N Kleshchenko ldquoGeneralized refractive mixing dielectricmodel for moist soilsrdquo IEEE Transactions on Geoscience andRemote Sensing vol 42 no 4 pp 773ndash785 2004

[62] J R Wang and T J Schmugge ldquoAn empirical model for thecomplex dielectric permittivity of soils as a function of watercontentrdquo IEEE Transactions on Geoscience and Remote Sensingvol 18 no 4 pp 288ndash295 1980


[63] MonoRTMAugust 2015 httprtwebaercommonortm framehtml

[64] M J McFarland R L Miller and C M U Neale ldquoLandsurface temperature derived from the SSMI passive microwavebrightness temperaturesrdquo IEEE Transactions on Geoscience andRemote Sensing vol 28 no 5 pp 839ndash845 1990

[65] Y J Kaufman and B-C Gao ldquoRemote sensing of water vaporin the near IR from EOSMODISrdquo IEEE Transactions onGeoscience and Remote Sensing vol 30 no 5 pp 871ndash884 1992

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ClimatologyJournal of

EcologyInternational Journal of


EarthquakesJournal of


Hindawi Publishing Corporationhttpwwwhindawicom

Applied ampEnvironmentalSoil Science

Volume 2014

Mining


Journal of

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014

International Journal of

Geophysics

OceanographyInternational Journal of


Journal of Computational Environmental SciencesHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal ofPetroleum Engineering


GeochemistryHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of

Atmospheric SciencesInternational Journal of


OceanographyHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in


MineralogyInternational Journal of


MeteorologyAdvances in

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Paleontology JournalHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ScientificaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Geological ResearchJournal of


Geology Advances in


Ground-based microwave radiometers can operate in a con-tinuous and unattended manner under almost all weatherconditions with high temporal resolution and low uncer-tainty [22ndash25] Ground-based GPS is an increasingly usefultool for measuring PWV agreeing with radiosonde datawithin 1-2mm [26ndash28] However ground-based microwaveradiometers and GPS methods cannot assess the distributionof PWV on a global scale a problem that also affects groundsurveys because of their poor coverage over oceans and lakesThus in comparisonwith these ground-basedmeasurementsspace-based surveys are an effective way to obtain globalPWVdataMany algorithms have been developed to estimatePWV from satellite data For example MODIS includestwo PWV products derived from observations in the near-infrared and thermal-infrared channels [29] However Li etal [30] found that MODIS algorithms overestimated PWVcompared to data from radiosondes and ground-based GPSMoreover because the MODIS PWV products are based onmeasurements from the near-infrared or thermal-infraredchannels they perform poorly in cloudy conditions JetPropulsion Laboratory (JPL) Atmospheric Infrared Sounder(AIRS) is the first of a new generation of advanced satellite-based atmospheric sounders with the capability of obtaininghigh-vertical resolution profiles of temperature and watervapor [31 32] Although it can producemoisture profile underpartial cloud cover the spatial resolutions of PWV productsare sparse [33]

Comparedwith infraredmeasurements space-based pas-sivemicrowave observations have the advantage of being ableto detect PWV on both cloud-free and cloudy days becausemicrowaves can penetrate clouds However the use of passivemicrowave satellite data for PWVestimation has traditionallybeen applied only over oceans because of their relativelysimple surface conditions compared to the land surface Ingeneral two methods are used to estimate PWV over oceansThe first uses physical algorithms that employ radiativetransfer theory in the model derivations (eg the multiplelinear regression algorithms developed by Schluessel andEmery [34] and the nonlinear iterative algorithms developedby Wentz and Meissner [35]) The second method involvespurely statistical methods with little or no consideration ofthe underlying physics (eg the postlaunch in situ regres-sion algorithms of Basili et al [36]) Compared to oceansdevelopment of PWV retrieval algorithms over land surfaceshas been slow because these surfaces have high pixel-to-pixelvariability in their surface emissivity which makes it diffi-cult to distinguish the atmospheric signal from the surfacebackground [37] Nevertheless several recent approacheshave been proposed that ignore the sensitivity of PWV toland surface emissivity (LSE) Some use a neural networkapproach with a database acquired from the radiative transfermodel [38ndash40] and some take advantage of high frequencywater vapor channels such as 15000 and around 18331 GHzwhich show very low sensitivity to the change in LSE duringPWV estimation [41 42] While these two methods arefeasible both have disadvantages the former has a limitedphysical basis and the latter cannot be used for datasets thatdo not include high frequency water vapor channels such asthose from the Special Sensor Microwave Imager (SSMI)

the ScanningMultichannelMicrowave Radiometer (SMMR)and the Advanced Microwave Scanning Radiometer-EarthObserving System (AMSR-E) Although Deeter [43] pro-posed an algorithm for PWV estimation over both land andocean based on brightness temperatures at 187 and 238GHzthe algorithm did not explain why these two frequencies wereselected [44] Based on these deficiencies the objective ofthis study is to develop an algorithm for retrieving PWVover land using passive microwave satellite data at lowfrequencies

The paper is organized as follows Section 2 describesthe data used in this study Section 3 describes the retrievalalgorithm from a semiempirical point of view Section 4explains the sensitivity analysis and validation with indepen-dent datasets and Section 5 concludes the paper

2 Data

21 The Thermodynamic Initial Guess Retrieval (TIGR)Dataset The TIGR dataset was constructed by the Labora-toire de Meteorologie Dynamique and includes 2311 atmo-spheric profiles selected from 80000 radiosonde datasetsThese profile data are located around the world and representreal atmosphere conditions Each atmospheric profile recordstemperature water vapor and ozone concentrations on agiven pressure grid from the surface to the top of theatmosphere

In this study 378 cloud-free and overland atmosphericprofiles located between 30 and 60 degrees north fromthe TIGR dataset were downloaded from the AtmosphericRadiation Analysis (ARA) website [46] Using these profileswe built a dataset including transmittances atmosphericeffective radiating temperatures and PWV to develop thePWV retrieval model

22 Microwave Radiation Imager (MWRI) Data Launchedon 5 November 2010 FengYun-3B (FY-3B) is the secondsatellite in the Chinese FY satellite series FY-3B crosses theequator in the ascendingmode at 1340 h local solar timeTheMicrowave Radiation Imager (MWRI) on board FY-3B is apassive microwave imager that uses conical scanning at anincidence angle of 53∘ It is a total-power passive radiometerthat measures radiation at five frequencies (1065 187 238365 and 890GHz) with vertical and horizontal polarization[45] A detailed description of the characteristics of the FY-3BMWRI is provided in Table 1

In this study brightness temperatures derived from theFY-3B MWRI with a unified resolution of 10 km wereacquired from the National Satellite Meteorological Centre[47] and used to provide input parameters for the PWVretrieval algorithm

23 Radiosonde Sounding Observations Located in ZhangyeCity Gansu Province China the Zhangye National ClimateObservatory (39∘51015840156810158401015840N 100∘161015840391110158401015840E) was selected asthe validation area for this study The estimated PWV wasverified using PWV data from the observatoryrsquos soundingobservations



Frequencies (GHz) 1065 187 238 365 89








3 Methodology



119886


119904


119886


119879119861119894= 119879119886119894+ 120591119894sdot 120576119894sdot 119879119904+ (1 minus 120576

119894) 120591119894(119879119886119894+ 119879sky) (1a)

119879119861119895= 119879119886119895+ 120591119895sdot 120576119895sdot 119879119904+ (1 minus 120576

119895) 120591119895(119879119886119895+ 119879sky) (1b)

where 119879119861119894

and 119879119861119895


119894and 120591

119895


and 119879119886119895


119894and 120576119895are LSEs 119879

119904



119894= 120576119895



119904


= (1 minus 120576) 120591119894sdot 120591119895(119879119886119894minus 119879119886119895)

(2)


119886and



119894sdot 120591119895(119879119886119894minus119879119886119895) and


119879119861119894sdot (1205721sdot PWV + 120572


4)

+ (1205725sdot PWV + 120572

6) = (1 minus 120576) (120572

7sdot PWV + 120572

8)

(3)



PWV =1205731+ 1205732sdot 120576 + 120573

3sdot 119879119861119894+ 1205734sdot 119879119861119895

1205735+ 1205736sdot 120576 + 120573

7sdot 119879119861119894+ 1205738sdot 119879119861119895

(4)


119879119861119894

and 119879119861119895






isequal to LSE



006

005

004

003

002

001

000

ΔLS

E

0 10 20 30 40 50 60


ΔLSE238-187ΔLSE365-238















LSE

02

04

00

minus02

minus04







119879119886187sdot 120591238



119886187 119879119886238

120591187

120591238


119886238sdot 120591187minus 119879119886187sdot 120591238




119886238sdot 120591187minus 119879119886187sdot 120591238





187sdot120591238(119879119886187minus119879119886238)






1205732

1205733

1205734

1205735

1205736

1205737

1205738






2


2



119861187and

119879119861238



error119879119861187sim 119873(0 05

2

) and error119879119861238sim 119873(0 08

2


119861187and 119879





50

40

30

20

10

0

Ta238120591187minusTa187120591238

(K)

0 10 20 30 40 50

PWV (mm)


RMSE = 064K

(a)

0

minus5

minus10

minus15

minus20

minus25

minus30

120591187120591238(T

a187minusTa238)

(K)

0 10 20 30 40 50

PWV (mm)


RMSE = 074K

(b)


sdot 120591187

minus 119879119886187

sdot 120591238


sdot 120591238

(119879119886187

minus 119879119886238

) and PWV


Freq

uenc

y (

)

30

20

10

0


(a)


Freq

uenc

y (

)30

20

25

10

15

0

5


(b)



119861187






119861187


119861187and

08 K for 119879119861238


119861238and119879119861238

errors


119861238





Freq

uenc

y (

)

20

25

10

15

0

5



(a)

Freq

uenc

y (

)

20

25

10

15

0

5



(b)


119861187


errors is 08 K



120591187sdot 1198791015840

119904

minus 120591187(119879119886187+ 119879sky)

(5)

where 119879119861187




119904



119861187

and 119879119861238


119861238 On cloudy


Retr

ieve

d PW

V (m

m)

50

40

30

20

10

0

50403020100


Cloudy days

Cloud-free days

1 1 line

RMSE = 484mm bias = 052mm R2 = 057

RMSE = 439mm bias = 036mm R2 = 046




trie

ved

PWV

(mm

)

50

60

40

30

20

10

0

50 60403020100


1 1 line


RMSE = 473mm bias = 084mm R2 = 093





5 Conclusions






Acknowledgments


References
















































































Volume 2014

Mining


Journal of



Geophysics







Journal of




Advances in











Geology Advances in



Frequencies (GHz) 1065 187 238 365 89








3 Methodology



119886


119904


119886


119879119861119894= 119879119886119894+ 120591119894sdot 120576119894sdot 119879119904+ (1 minus 120576

119894) 120591119894(119879119886119894+ 119879sky) (1a)

119879119861119895= 119879119886119895+ 120591119895sdot 120576119895sdot 119879119904+ (1 minus 120576

119895) 120591119895(119879119886119895+ 119879sky) (1b)

where 119879119861119894

and 119879119861119895


119894and 120591

119895


and 119879119886119895


119894and 120576119895are LSEs 119879

119904



119894= 120576119895



119904


= (1 minus 120576) 120591119894sdot 120591119895(119879119886119894minus 119879119886119895)

(2)


119886and



119894sdot 120591119895(119879119886119894minus119879119886119895) and


119879119861119894sdot (1205721sdot PWV + 120572


4)

+ (1205725sdot PWV + 120572

6) = (1 minus 120576) (120572

7sdot PWV + 120572

8)

(3)



PWV =1205731+ 1205732sdot 120576 + 120573

3sdot 119879119861119894+ 1205734sdot 119879119861119895

1205735+ 1205736sdot 120576 + 120573

7sdot 119879119861119894+ 1205738sdot 119879119861119895

(4)


119879119861119894

and 119879119861119895






isequal to LSE



006

005

004

003

002

001

000

ΔLS

E

0 10 20 30 40 50 60


ΔLSE238-187ΔLSE365-238















LSE

02

04

00

minus02

minus04







119879119886187sdot 120591238



119886187 119879119886238

120591187

120591238


119886238sdot 120591187minus 119879119886187sdot 120591238




119886238sdot 120591187minus 119879119886187sdot 120591238





187sdot120591238(119879119886187minus119879119886238)






1205732

1205733

1205734

1205735

1205736

1205737

1205738






2


2



119861187and

119879119861238



error119879119861187sim 119873(0 05

2

) and error119879119861238sim 119873(0 08

2


119861187and 119879





50

40

30

20

10

0

Ta238120591187minusTa187120591238

(K)

0 10 20 30 40 50

PWV (mm)


RMSE = 064K

(a)

0

minus5

minus10

minus15

minus20

minus25

minus30

120591187120591238(T

a187minusTa238)

(K)

0 10 20 30 40 50

PWV (mm)


RMSE = 074K

(b)


sdot 120591187

minus 119879119886187

sdot 120591238


sdot 120591238

(119879119886187

minus 119879119886238

) and PWV


Freq

uenc

y (

)

30

20

10

0


(a)


Freq

uenc

y (

)30

20

25

10

15

0

5


(b)



119861187






119861187


119861187and

08 K for 119879119861238


119861238and119879119861238

errors


119861238





Freq

uenc

y (

)

20

25

10

15

0

5



(a)

Freq

uenc

y (

)

20

25

10

15

0

5



(b)


119861187


errors is 08 K



120591187sdot 1198791015840

119904

minus 120591187(119879119886187+ 119879sky)

(5)

where 119879119861187




119904



119861187

and 119879119861238


119861238 On cloudy


Retr

ieve

d PW

V (m

m)

50

40

30

20

10

0

50403020100


Cloudy days

Cloud-free days

1 1 line

RMSE = 484mm bias = 052mm R2 = 057

RMSE = 439mm bias = 036mm R2 = 046




trie

ved

PWV

(mm

)

50

60

40

30

20

10

0

50 60403020100


1 1 line


RMSE = 473mm bias = 084mm R2 = 093





5 Conclusions






Acknowledgments


References
















































































Volume 2014

Mining


Journal of



Geophysics







Journal of




Advances in











Geology Advances in



PWV =1205731+ 1205732sdot 120576 + 120573

3sdot 119879119861119894+ 1205734sdot 119879119861119895

1205735+ 1205736sdot 120576 + 120573

7sdot 119879119861119894+ 1205738sdot 119879119861119895

(4)


119879119861119894

and 119879119861119895






isequal to LSE



006

005

004

003

002

001

000

ΔLS

E

0 10 20 30 40 50 60


ΔLSE238-187ΔLSE365-238















LSE

02

04

00

minus02

minus04







119879119886187sdot 120591238



119886187 119879119886238

120591187

120591238


119886238sdot 120591187minus 119879119886187sdot 120591238




119886238sdot 120591187minus 119879119886187sdot 120591238





187sdot120591238(119879119886187minus119879119886238)






1205732

1205733

1205734

1205735

1205736

1205737

1205738






2


2



119861187and

119879119861238



error119879119861187sim 119873(0 05

2

) and error119879119861238sim 119873(0 08

2


119861187and 119879





50

40

30

20

10

0

Ta238120591187minusTa187120591238

(K)

0 10 20 30 40 50

PWV (mm)


RMSE = 064K

(a)

0

minus5

minus10

minus15

minus20

minus25

minus30

120591187120591238(T

a187minusTa238)

(K)

0 10 20 30 40 50

PWV (mm)


RMSE = 074K

(b)


sdot 120591187

minus 119879119886187

sdot 120591238


sdot 120591238

(119879119886187

minus 119879119886238

) and PWV


Freq

uenc

y (

)

30

20

10

0


(a)


Freq

uenc

y (

)30

20

25

10

15

0

5


(b)



119861187






119861187


119861187and

08 K for 119879119861238


119861238and119879119861238

errors


119861238





Freq

uenc

y (

)

20

25

10

15

0

5



(a)

Freq

uenc

y (

)

20

25

10

15

0

5



(b)


119861187


errors is 08 K



120591187sdot 1198791015840

119904

minus 120591187(119879119886187+ 119879sky)

(5)

where 119879119861187




119904



119861187

and 119879119861238


119861238 On cloudy


Retr

ieve

d PW

V (m

m)

50

40

30

20

10

0

50403020100


Cloudy days

Cloud-free days

1 1 line

RMSE = 484mm bias = 052mm R2 = 057

RMSE = 439mm bias = 036mm R2 = 046




trie

ved

PWV

(mm

)

50

60

40

30

20

10

0

50 60403020100


1 1 line


RMSE = 473mm bias = 084mm R2 = 093





5 Conclusions






Acknowledgments


References
















































































Volume 2014

Mining


Journal of



Geophysics







Journal of




Advances in











Geology Advances in


LSE

02

04

00

minus02

minus04







119879119886187sdot 120591238



119886187 119879119886238

120591187

120591238


119886238sdot 120591187minus 119879119886187sdot 120591238




119886238sdot 120591187minus 119879119886187sdot 120591238





187sdot120591238(119879119886187minus119879119886238)






1205732

1205733

1205734

1205735

1205736

1205737

1205738






2


2



119861187and

119879119861238



error119879119861187sim 119873(0 05

2

) and error119879119861238sim 119873(0 08

2


119861187and 119879





50

40

30

20

10

0

Ta238120591187minusTa187120591238

(K)

0 10 20 30 40 50

PWV (mm)


RMSE = 064K

(a)

0

minus5

minus10

minus15

minus20

minus25

minus30

120591187120591238(T

a187minusTa238)

(K)

0 10 20 30 40 50

PWV (mm)


RMSE = 074K

(b)


sdot 120591187

minus 119879119886187

sdot 120591238


sdot 120591238

(119879119886187

minus 119879119886238

) and PWV


Freq

uenc

y (

)

30

20

10

0


(a)


Freq

uenc

y (

)30

20

25

10

15

0

5


(b)



119861187






119861187


119861187and

08 K for 119879119861238


119861238and119879119861238

errors


119861238





Freq

uenc

y (

)

20

25

10

15

0

5



(a)

Freq

uenc

y (

)

20

25

10

15

0

5



(b)


119861187


errors is 08 K



120591187sdot 1198791015840

119904

minus 120591187(119879119886187+ 119879sky)

(5)

where 119879119861187




119904



119861187

and 119879119861238


119861238 On cloudy


Retr

ieve

d PW

V (m

m)

50

40

30

20

10

0

50403020100


Cloudy days

Cloud-free days

1 1 line

RMSE = 484mm bias = 052mm R2 = 057

RMSE = 439mm bias = 036mm R2 = 046




trie

ved

PWV

(mm

)

50

60

40

30

20

10

0

50 60403020100


1 1 line


RMSE = 473mm bias = 084mm R2 = 093





5 Conclusions






Acknowledgments


References
















































































Volume 2014

Mining


Journal of



Geophysics







Journal of




Advances in











Geology Advances in


50

40

30

20

10

0

Ta238120591187minusTa187120591238

(K)

0 10 20 30 40 50

PWV (mm)


RMSE = 064K

(a)

0

minus5

minus10

minus15

minus20

minus25

minus30

120591187120591238(T

a187minusTa238)

(K)

0 10 20 30 40 50

PWV (mm)


RMSE = 074K

(b)


sdot 120591187

minus 119879119886187

sdot 120591238


sdot 120591238

(119879119886187

minus 119879119886238

) and PWV


Freq

uenc

y (

)

30

20

10

0


(a)


Freq

uenc

y (

)30

20

25

10

15

0

5


(b)



119861187






119861187


119861187and

08 K for 119879119861238


119861238and119879119861238

errors


119861238





Freq

uenc

y (

)

20

25

10

15

0

5



(a)

Freq

uenc

y (

)

20

25

10

15

0

5



(b)


119861187


errors is 08 K



120591187sdot 1198791015840

119904

minus 120591187(119879119886187+ 119879sky)

(5)

where 119879119861187




119904



119861187

and 119879119861238


119861238 On cloudy


Retr

ieve

d PW

V (m

m)

50

40

30

20

10

0

50403020100


Cloudy days

Cloud-free days

1 1 line

RMSE = 484mm bias = 052mm R2 = 057

RMSE = 439mm bias = 036mm R2 = 046




trie

ved

PWV

(mm

)

50

60

40

30

20

10

0

50 60403020100


1 1 line


RMSE = 473mm bias = 084mm R2 = 093





5 Conclusions






Acknowledgments


References
















































































Volume 2014

Mining


Journal of



Geophysics







Journal of




Advances in











Geology Advances in


Freq

uenc

y (

)

20

25

10

15

0

5



(a)

Freq

uenc

y (

)

20

25

10

15

0

5



(b)


119861187


errors is 08 K



120591187sdot 1198791015840

119904

minus 120591187(119879119886187+ 119879sky)

(5)

where 119879119861187




119904



119861187

and 119879119861238


119861238 On cloudy


Retr

ieve

d PW

V (m

m)

50

40

30

20

10

0

50403020100


Cloudy days

Cloud-free days

1 1 line

RMSE = 484mm bias = 052mm R2 = 057

RMSE = 439mm bias = 036mm R2 = 046




trie

ved

PWV

(mm

)

50

60

40

30

20

10

0

50 60403020100


1 1 line


RMSE = 473mm bias = 084mm R2 = 093





5 Conclusions






Acknowledgments


References
















































































Volume 2014

Mining


Journal of



Geophysics







Journal of




Advances in











Geology Advances in


trie

ved

PWV

(mm

)

50

60

40

30

20

10

0

50 60403020100


1 1 line


RMSE = 473mm bias = 084mm R2 = 093





5 Conclusions






Acknowledgments


References
















































































Volume 2014

Mining


Journal of



Geophysics







Journal of




Advances in











Geology Advances in













































































Volume 2014

Mining


Journal of



Geophysics







Journal of




Advances in











Geology Advances in

research article an algorithm for retrieving precipitable...

Documents