pre-monsoon maximum and minimum temperatures over india in

13
INTERNATIONAL JOURNAL OF CLIMATOLOGY, VOL. 17, 1115–1127 (1997) PRE-MONSOON MAXIMUM AND MINIMUM TEMPERATURES OVER INDIA IN RELATION TO THE SUMMER MONSOON RAINFALL K. KRISHNA KUMAR*, K. RUPA KUMAR AND G. B. PANT Climatology and Hydrometeorology Division, Indian Institute of Tropical Meteorology, Homi Bhabha Road, Pune - 411 008, India email: [email protected] Received 31 May 1996 Revised 6 February 1997 Accepted 4 May 1997 The pre-monsoon thermal field over the Indian landmass has an important bearing on the land–sea heating contrast in the region, consequently influencing the establishment, advance and overall performance of the Indian summer monsoon rainfall. This paper examines the relationship between the pre-monsoon thermal field over India and the following summer monsoon rainfall, in order to identify possible predictors for long-range forecasting of Indian summer monsoon rainfall. Based on the spatial patterns of correlations of monsoon rainfall with maximum and minimum temperatures at 121 stations well distributed over India, during the recent period 1951–80, two predictors showing a significant contribution to the variance in monsoon rainfall have been identified. They are (i) March minimum temperature in east peninsular India and (ii) May minimum temperature in west central India. These two predictors have performed very well in terms of their significant contribution to the multiple regression models during 1951–1987, vis-a `-vis several other known predictors. They have also shown a consistently significant relationship with the monsoon rainfall during the recent period, from the mid-1940s till the end of the data period. A stepwise regression model for long-range forecasting of all-India summer monsoon rainfall, involving three regional predictors, has been developed, and has shown a multiple correlation of 089. # 1997 by the Royal Meteorological Society. Int. J. Climatol., 17: 1115–1127 (1997) (No. of Figures: 8 No. of Tables: 3 No. of References: 21) KEY WORDS: long-range forecasting; Indian summer monsoon rainfall; interannual variability; maximum/minimum temperatures; multiple regression; correlation coefficient patterns 1. INTRODUCTION The problem of long-range forecasting (LRF) of Indian summer monsoon rainfall has been one of the major tasks of Indian meteorologists for more than a century (Blandford, 1884; Walker, 1924; Montgomery, 1940). The interannual variability in the monsoon affects not only the economies of the countries coming under its influence, but also has an important role in the global general circulation (Webster and Song Yang, 1992). A number of regional and global predictor parameters have been identified for the LRF of monsoon rainfall. A comprehensive review of LRF studies on Indian summer monsoon rainfall has been given by Krishna Kumar et al. (1995). A careful examination of the magnitudes of correlation coefficients of various predictors with the Indian summer monsoon rainfall (Parthasarathy et al., 1988) indicates that most of the regional parameters show relatively higher correlations than the global teleconnection parameters, including those related to the El Nin ˜o–Southern Oscillation (ENSO). The correlations between the ENSO parameters and the monsoon rainfall are at the maximum during and after the monsoon season, thus limiting their use in foreshadowing the monsoon performance. However, the winter to spring tendencies of some of the ENSO related parameters provide some predictive utility (Shukla and Paolino, 1983). Studies on LRF in the recent past have brought out several regional parameters based on sea-level pressure, temperature and wind fields over India and sea surface temperature (SST) data from the adjoining Indian seas. Although their performance in seasonal forecasting has been encouraging, there is still a large variance in the monsoon rainfall unaccounted by the predictors identified so far. Therefore, the search for identifying new CCC 0899-8418/97/101115-13 $17.50 # 1997 by the Royal Meteorological Society * Correspondence to: K. K. Kumar, Climatology and Hydrometeorology Division, Indian Institute of Tropical Meteorology, Homi Bhabha Road, Pune-411 008, India

Upload: vodiep

Post on 04-Jan-2017

219 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Pre-monsoon maximum and minimum temperatures over India in

INTERNATIONAL JOURNAL OF CLIMATOLOGY, VOL. 17, 1115±1127 (1997)

PRE-MONSOON MAXIMUM AND MINIMUM TEMPERATURES OVERINDIA IN RELATION TO THE SUMMER MONSOON RAINFALL

K. KRISHNA KUMAR*, K. RUPA KUMAR AND G. B. PANT

Climatology and Hydrometeorology Division, Indian Institute of Tropical Meteorology, Homi Bhabha Road, Pune - 411 008, Indiaemail: [email protected]

Received 31 May 1996Revised 6 February 1997

Accepted 4 May 1997

The pre-monsoon thermal ®eld over the Indian landmass has an important bearing on the land±sea heating contrast in theregion, consequently in¯uencing the establishment, advance and overall performance of the Indian summer monsoon rainfall.This paper examines the relationship between the pre-monsoon thermal ®eld over India and the following summer monsoonrainfall, in order to identify possible predictors for long-range forecasting of Indian summer monsoon rainfall. Based on thespatial patterns of correlations of monsoon rainfall with maximum and minimum temperatures at 121 stations well distributedover India, during the recent period 1951±80, two predictors showing a signi®cant contribution to the variance in monsoonrainfall have been identi®ed. They are (i) March minimum temperature in east peninsular India and (ii) May minimumtemperature in west central India. These two predictors have performed very well in terms of their signi®cant contribution tothe multiple regression models during 1951±1987, vis-aÁ-vis several other known predictors. They have also shown aconsistently signi®cant relationship with the monsoon rainfall during the recent period, from the mid-1940s till the end of thedata period. A stepwise regression model for long-range forecasting of all-India summer monsoon rainfall, involving threeregional predictors, has been developed, and has shown a multiple correlation of 0�89. # 1997 by the Royal MeteorologicalSociety. Int. J. Climatol., 17: 1115±1127 (1997)

(No. of Figures: 8 No. of Tables: 3 No. of References: 21)

KEY WORDS: long-range forecasting; Indian summer monsoon rainfall; interannual variability; maximum/minimum temperatures; multipleregression; correlation coef®cient patterns

1. INTRODUCTION

The problem of long-range forecasting (LRF) of Indian summer monsoon rainfall has been one of the major tasks

of Indian meteorologists for more than a century (Blandford, 1884; Walker, 1924; Montgomery, 1940). The

interannual variability in the monsoon affects not only the economies of the countries coming under its in¯uence,

but also has an important role in the global general circulation (Webster and Song Yang, 1992). A number of

regional and global predictor parameters have been identi®ed for the LRF of monsoon rainfall. A comprehensive

review of LRF studies on Indian summer monsoon rainfall has been given by Krishna Kumar et al. (1995). A

careful examination of the magnitudes of correlation coef®cients of various predictors with the Indian summer

monsoon rainfall (Parthasarathy et al., 1988) indicates that most of the regional parameters show relatively higher

correlations than the global teleconnection parameters, including those related to the El NinÄo±Southern

Oscillation (ENSO). The correlations between the ENSO parameters and the monsoon rainfall are at the

maximum during and after the monsoon season, thus limiting their use in foreshadowing the monsoon

performance. However, the winter to spring tendencies of some of the ENSO related parameters provide some

predictive utility (Shukla and Paolino, 1983).

Studies on LRF in the recent past have brought out several regional parameters based on sea-level pressure,

temperature and wind ®elds over India and sea surface temperature (SST) data from the adjoining Indian seas.

Although their performance in seasonal forecasting has been encouraging, there is still a large variance in the

monsoon rainfall unaccounted by the predictors identi®ed so far. Therefore, the search for identifying new

CCC 0899-8418/97/101115-13 $17.50

# 1997 by the Royal Meteorological Society

* Correspondence to: K. K. Kumar, Climatology and Hydrometeorology Division, Indian Institute of Tropical Meteorology, Homi BhabhaRoad, Pune-411 008, India

Page 2: Pre-monsoon maximum and minimum temperatures over India in

parameters and to diagnose the existing parameters for the improvement of LRF schemes continues

(Parthasarathy et al., 1991a; Krishna Kumar et al., 1992). The present study is one such attempt in that direction.

Parthasarathy et al. (1990), based on the analysis of the mean monthly surface temperature data at 73 stations

in India, have identi®ed the regional mean temperature over west central India during March±April±May as a

good predictor representing the pre-monsoon heating over India. Mooley and Paolino (1988), using maximum

and minimum temperature data for the period 1901±1975, have shown that a parameter based on minimum

temperatures at some stations in the western Indian region during the month of May has a good potential for the

forecasting of Indian summer monsoon rainfall. The India Meteorological Department, in their operational LRF

model (Gowariker et al., 1991), uses two temperature parameters; one from north India and the other from the

east coast of India. From the above studies and from the fact that the monsoons are a consequence of the land±sea

thermal contrast, it is evident that the thermal conditions during the pre-monsoon season over India play a

signi®cant role in the performance of the ensuing monsoon.

Mooley and Paolino (1988) have used maximum and minimum temperature data up to 1975 only in their study,

and considered the entire data period (1901±1975) in the correlation analysis to identify predictor parameters.

Many recent studies have indicated secular variations in the relationships between the summer monsoon rainfall

and its predictors (Krishna Kumar et al., 1995). In view of this and the fact that a period of 30 years is a necessary

and suf®cient requirement to establish a reliable climatological relationship, it is desirable to identify predictors

based on the recent three decades or so. Therefore, the present study attempts to re-examine the relationship

between the thermal ®eld during the winter and pre-monsoon months and the all-India summer monsoon rainfall

(AISMR) in more detail by making use of the maximum and minimum temperature data from a well distributed

observatory network for the period 1901±1987. The main objective of the present study is to identify spatially

coherent regions over which the maximum and minimum temperatures show a signi®cant relationship with

AISMR during the recent period, 1951±1980, and to develop suitable predictor parameters using the temperature

data from those regions. The parameters that are identi®ed in the study have been examined for their temporal

consistency. Their relationship with the rainfall of different subdivisions of India has also been examined to see

the spatial preferences, if any. Suitable multiple regression schemes using these parameters, together with other

LRF parameters, are also presented.

2. DATA

The majority of the predictors currently being used for the LRF of Indian summer monsoon rainfall have been

identi®ed using the data period 1951±1980. Considering this, in the present study the maximum/minimum

temperature data during the recent 30-year period, 1951±1980, have been considered for the correlation analysis

so that the predictors that may arise out of the analysis can be used in conjunction with other predictors to develop

a suitable LRF scheme. The temporal consistency of the parameters identi®ed, however, has been examined using

data on a longer period, namely 1901±1994.

All-India summer monsoon rainfall

The summer monsoon (June through September) rainfall for all-India and 29 meteorological sub-divisions for

the period 1901±1994, prepared by area-weighting 306 well-distributed rain-gauge stations in the plain areas of

the country (Parthasarathy et al., 1987, 1992, 1994) have been used in the present study.

Maximum and minimum temperatures

Monthly mean maximum/minimum temperature data from 121 well-distributed observatories (Figure 1) during

the period 1951±1980 have been used in the present study. The missing data at a few stations have been estimated

from the best related neighbouring stations by the linear regression method. In all, the total number of missing

values estimated is less than 10 per cent of the total data. More details about this data set are given by Rupa

Kumar et al. (1994). For examining the long-term consistency of the correlations shown by selected predictors,

data for the period 1901±1994 for the corresponding stations are used.

1116 K. K. KUMAR, K. R. KUMAR AND G. B. PANT

INT. J. CLIMATOL, VOL. 17: 1115±1127 (1997) # 1997 Royal Meteorological Society

Page 3: Pre-monsoon maximum and minimum temperatures over India in

3. RELATIONSHIPS BETWEEN AISMR AND MAXIMUM AND MINIMUM TEMPERATURES

Spatial correlation patterns

The correlation coef®cients between AISMR and maximum and minimum temperatures at all the stations have

been worked out using the data period 1951±1980, for all the winter (December, January and February) and pre-

monsoon (March, April and May) months, in order to identify spatially coherent regions with high predictive

potential for their use in preparing suitable LRF parameters. The stationwise correlation coef®cients (CCs) for all

the six months from the previous December to the current May are plotted and isocorrelation maps are prepared

in order to study their spatial patterns. The CCs during the winter months are found to be statistically insigni®cant

over almost all the country, and therefore they are not presented here. The spatial patterns of CCs signi®cant at 5

per cent and 1 per cent levels for the pre-monsoon months of March, April and May, for both maximum and

minimum temperatures are presented in Figures 2 and 3. The salient features are described below.

Maximum temperature

In the month of March, the maximum temperature of any part of the country does not show spatially coherent

and statistically signi®cant (5 per cent or above) CCs (Figure 2(a)). For the month of April (Figure 2(b)), the CCs

are signi®cant over central and west central parts of the country. The CCs are also signi®cant over a small region

in the north. The CCs for the month of May (Figure 2(c)) are again not spatially very coherent. There are a couple

of isolated patches in the eastern parts of north central India and central India where the CCs are signi®cant. On

Figure 1. Location of stations at which maximum/minimum temperature data are utilized in the study. Circles enclosing the station locationsindicate those considered for WCTNMAY and squares enclosing the stations indicate those considered for EPTNMAR

PREDICTORS OF INDIAN SUMMER RAIN 1117

# 1997 Royal Meteorological Society INT. J. CLIMATOL, VOL. 17: 1115±1127 (1997)

Page 4: Pre-monsoon maximum and minimum temperatures over India in

the whole, the spatial patterns of the CCs suggest that the maximum temperature pre-monsoon months over India

may not provide a strong and spatially coherent signal for its use as a predictor parameter, excepting for the

month of April in which the CCs are signi®cant over central and west central Indian regions.

Minimum temperature

In contrast to the relationships shown by the maximum temperatures (Figure 2), the minimum temperature in

all three pre-monsoon months shows statistically signi®cant and spatially coherent CCs (Figure 3) with AISMR.

In March (Figure 3(a)), the CCs along the east coast, right from the southern tip of the peninsula to West Bengal

in the north and even extending into the inland areas in the central parts of the coastal belt, are very high and

spatially coherent, indicating good potential for developing a predictive parameter for use in LRF of monsoon

rainfall. Signi®cant CCs are also found elsewhere, but are restricted to a few isolated stations. It is noteworthy

that such a large area along the east coast showing a strong predictive signal is conspicuously missing in the

earlier study of Mooley and Paolino (1988). This could be partly because of the differences in the data period

considered in the analysis. From Figure 3(b) for April, it can be seen that the statistically signi®cant CCs are

present mainly over two regions, one over the northern parts of west coast and adjoining areas and the other over

Figure 2. Spatial patterns of signi®cant correlations between AISMR and maximum temperatures during 1951±1980; (a) March, (b) April and(c) May. Light shading indicates signi®cance at 5 per cent level and dark shading at 1 per cent level

Figure 3. Same as Figure 2 but for minimum temperatures

1118 K. K. KUMAR, K. R. KUMAR AND G. B. PANT

INT. J. CLIMATOL, VOL. 17: 1115±1127 (1997) # 1997 Royal Meteorological Society

Page 5: Pre-monsoon maximum and minimum temperatures over India in

the south-eastern parts of peninsular India. These two regions are similar to the ones identi®ed by Mooley and

Paolino (1988) for the minimum temperature in the month of April. The CCs for the month of May (Figure 3(c))

show highly signi®cant values over a very large area covering a major part of west central India and the adjoining

central Indian and north-west Indian regions. The CCs at most of the stations are signi®cant at 1 per cent and

some are even signi®cant at the 0�1 per cent level.

Development of predictors

From the above analysis, two broad regions emerge over which the CCs are highly signi®cant and spatially

coherent. The ®rst region is along the east coast of India and interior parts of peninsular India, and the second

over west central India. The minimum temperatures during the month of March for the ®rst region, and during the

month of May for the second region, indicate good predictive potential for LRF of Indian summer monsoon

rainfall. Keeping this in view, two predictive parameters have been developed using the minimum temperature

data of the stations that are located in the core regions of the highest correlations for the respective months. The

stations in proximity to the stations chosen but showing non-signi®cant CCs with AISMR have been excluded

from the regional indices, in order to maximize the signal. Minimum temperature data from 12 stations located

along the east coast of India and over the adjoining interior parts of peninsular India have been considered to

prepare a parameter for the month of March, which hereafter is classi®ed as the east-peninsular-India March

minimum temperature (EPTNMAR). Similarly, minimum temperature data at 10 stations from the west central

Indian region have been used for developing the other parameter, hereafter called west-central-India May

minimum temperature (WCTNMAY). These two sets of stations are shown in Figure 1. In Figure 4 are given the

time series of the anomalies of EPTNMAR, WCTNMAY, and AISMR for the period 1951±1994. Although the

signi®cant CCs shown by the maximum temperatures are spatially less extensive, some stations over west central

India show signi®cant CCs in the month of April. Therefore, a parameter representing the maximum temperature

(WCTXAPR) has also been worked out using April maximum temperature at nine stations. The details of stations

used to develop the above three predictors, along with the corresponding CCs, are given in Table I.

The monthly evolution of the CCs of the predictors EPTNMAR, WCTNMAY and WCTXAPR, from the

previous year's July to the next year's June, with AISMR, are shown in Figure 5. For east peninsular India, the

CCs are statistically signi®cant for the months July and August of the year preceding and March of the current

Figure 4. Time series of WCTNMAY, EPTNMAR and AISMR during 1951±1994

PREDICTORS OF INDIAN SUMMER RAIN 1119

# 1997 Royal Meteorological Society INT. J. CLIMATOL, VOL. 17: 1115±1127 (1997)

Page 6: Pre-monsoon maximum and minimum temperatures over India in

year. The CC (0�69) for the month of March is the highest. In the case of west central India, the CCs of minimum

temperatures are signi®cant for last year's July and current year's May. Again the CC is highest (0�66) for the

month (May) from which the temperature index for this region was de®ned. Maximum temperatures over west

central India show almost the same pattern of CCs, except that the peak CC (0�52) is observed in the month of

April. The correlation patterns indicate a gradual increase of CCs from winter months to their peak values in the

pre-monsoon season, and thereafter suddenly drop to the negative CCs obviously caused by the monsoon activity.

The negative CCs during the monsoon season are higher in the case of maximum temperature, presumably due to

its higher sensitivity to cloudiness and rainfall. The high positive CCs observed during the previous July and

August could be regarded as part of the links between the preceding year's monsoon and the current monsoon.

These CCs apparently ®t into the quasi-biennial oscillation (QBO) observed in the monsoon rainfall variability

(Mooley and Parthasarathy, 1984). These results also support the hypothesis that the convective heating

Table I. List of stations considered for the preparation of predictor parameters

Predictor parameter Name of station Latitude (�N) Longitude (�E) Elevation (m) CC withAISMR

East peninsular India(March minimumtemperature)

Calcutta 22�32 88�20 6 0�44Balasore 21�30 86�56 20 0�59Gopalpur 19�16 84�53 17 0�52Puri 19�48 85�49 6 0�48Kalingapatnam 18�20 84�08 6 0�51Hanamkonda 18�01 79�34 269 0�43Gulbarga 17�21 76�51 458 0�54Masulipatnam 16�11 81�08 3 0�61Kakinada 16�57 82�14 8 0�58Bellary 15�09 76�51 449 0�44Kurnool 15�50 78�04 350 0�55Nellore 14�27 79�59 20 0�46

West central India(May minimumtemperature)

Jodhpur 26�18 73�02 217 0�49Deesa 24�12 72�12 136 0�59Nimach 24�28 74�54 496 0�52Rajkot 22�18 70�47 138 0�58Indore 22�43 75�48 567 0�52Surat 21�12 72�50 12 0�43Khandwa 21�50 76�22 318 0�53Veraval 20�54 70�22 8 0�62Akola 20�42 77�02 282 0�49Bombay 18�54 72�49 11 0�55

West central India(April maximumtemperature)

Jodhpur 26�18 73�02 217 0�36Deesa 24�12 72�12 136 0�45Nimach 24�28 74�54 496 0�35Dwaraka 22�22 69�05 11 0�44Rajkot 22�18 70�47 138 0�52Bhavnagar 21�45 72�12 12 0�53Surat 21�12 72�50 12 0�37Malegaon 20�33 74�32 437 0�45Bombay 18�54 72�49 11 0�52

1120 K. K. KUMAR, K. R. KUMAR AND G. B. PANT

INT. J. CLIMATOL, VOL. 17: 1115±1127 (1997) # 1997 Royal Meteorological Society

Page 7: Pre-monsoon maximum and minimum temperatures over India in

anomalies over Africa and the western Paci®c Ocean associated with the tropical biennial oscillation (TBO) alter

the mid-latitude circulation over Asia and maintain the temperature anomalies over South Asia through winter to

set up a land±sea temperature contrast for subsequent monsoon development (Meehl, 1994). Although the west

central India maximum temperatures during April may be representing the pre-monsoon heating, the minimum

temperatures in May may be indicating the moisture conditions just before the monsoon onset. However, the

Figure 5. Monthly evolution of CCs during 1951±1980 between AISMR and minimum temperatures over (i) east peninsular India and (ii)west central India, and maximum temperatures over (iii) west central India from previous July until the following June, relative to the

monsoon year

PREDICTORS OF INDIAN SUMMER RAIN 1121

# 1997 Royal Meteorological Society INT. J. CLIMATOL, VOL. 17: 1115±1127 (1997)

Page 8: Pre-monsoon maximum and minimum temperatures over India in

physical link of east peninsular minimum temperatures as early as in March is not clear. Further, the anomalous

surface temperatures over India may be part of anomalies in atmospheric circulation over a much larger area,

which might have persisted to affect the Indian summer monsoon.

4. TEMPORAL CONSISTENCY OF CORRELATIONS

Most of the LRF parameters are known to have shown signi®cant secular variation in their strength of

relationship with the monsoon rainfall (Parthasarathy et al., 1988, 1991b; Krishna Kumar et al., 1995). Therefore,

it is essential to examine the temporal consistency of any new predictor parameter for assessing its use in a LRF

model. For this purpose, 31-year sliding correlations have been worked out for EPTNMAR and WCTNMAY

during the period 1901±1994 and for WCTXAPR during 1901±1990 (Figure 6). This analysis shows that the CCs

in general have been strong since the 1940s in all three cases. During the period before 1940, the CCs were even

of the opposite sign, although not statistically signi®cant. Further, the CCs in the case of WCTXAPR were

statistically signi®cant for only a very short period of time, whereas the CCs in the case of EPTNMAR and

WCTNMAY were very high and statistically signi®cant right from the early 1940s until the end of the data

period. Similar changes in the strength and sign of the relationship have been observed previously in the case of

several other predictor parameters (Parthasarathy et al., 1993; Krishna Kumar et al., 1995). The turning point

around 1940 in the sign of the relationship is in general agreement with those noticed in the case of Bombay sea-

level pressure (Parthasarathy et al., 1991b). Such low-frequency changes in the CCs may be related to the slowly

varying large-scale changes in the monsoon regime, possibly caused by changes in deep oceanic circulations

related to the monsoon circulation. This analysis has indicated that the parameter WCTXAPR has not shown

promising performance in its temporal consistency for the current period and therefore has not been considered

for the development of the LRF scheme in the present study.

5. RELATIONSHIPS BETWEEN SUBDIVISIONAL MONSOON RAINFALL AND EPTNMAR AND

WCTNMAY

To understand the spatial representativeness of the relationship of the two predictors identi®ed in the present

study with AISMR, the CCs of the two parameters EPTNMAR and WCTNMAY with the monsoon rainfall of 29

meteorological subdivisions over India have been worked out (Figure 7). The CCs are statistically signi®cant

over the north-western and central parts of India in the case of EPTNMAR and over the northern parts of India,

excepting the north-east Indian region for WCTNMAY. It is interesting to note here that the March minimum

temperature over the east coast of India has no signi®cant relation with the rainfall over the same region.

However, this kind of spatial distribution of CCs has been observed with most of the LRF parameters known so

far, with some slight variations at the peripheral areas (Krishna Kumar et al., 1995). This could be due primarily

to the spatial coherency of the monsoon rainfall over this region (Parthasarathy et al., 1993).

6. DEVELOPMENT OF A REGRESSION MODEL FOR ALL-INDIA

MONSOON RAINFALL PREDICTION

The importance of the newly developed minimum temperature parameters relative to the other known LRF

parameters has been assessed using a stepwise regression model. For this purpose, the two newly identi®ed

parameters have been combined with 17 other LRF parameters, in order to prepare a set of predictors for a

stepwise regression analysis with the AISMR as the predictand. The details of the parameters considered for this

analysis are given in Table II. These parameters can be classi®ed broadly into four groups depending on their role

in the monsoon circulation, as (i) regional, (ii) ENSO, (iii) cross-equatorial ¯ow and (iv) global/hemispheric

conditions. These parameters cover most of the widely used predictors in many LRF schemes for the monsoon

rainfall. However, some other parameters, such as Eurasian snow cover and stratospheric winds, are not

considered because the period of data available for those parameters is very short. The CCs based on the data for

the period 1951±1980 between these parameters and AISMR are also given in Table II.

1122 K. K. KUMAR, K. R. KUMAR AND G. B. PANT

INT. J. CLIMATOL, VOL. 17: 1115±1127 (1997) # 1997 Royal Meteorological Society

Page 9: Pre-monsoon maximum and minimum temperatures over India in

Stepwise multiple regression

A multiple linear regression equation with AISMR has been developed using a stepwise procedure based on

the algorithm of Jennrich (1977). In this procedure, starting with the simplest model containing a single predictor

that has shown the highest correlation with the predictand (AISMR), further predictors are added in a stepwise

manner based on the F-value criterion, in order to obtain a statistically optimized multivariate regression model.

The stepping is terminated when the F-value of the most prospective predictor candidate drops below the 5 per

cent signi®cance level.

The ®nal regression equation obtained in the present study has shown a multiple correlation coef®cient of 0�89

based on 30-year data (1951±1980) and is shown below:

AISMR � ÿ350�8ÿ 54�976002� �WCPMAMÿ 1000� � 61�93805� EPTNMAR � 19�60� APR500

Figure 6. Sliding CCs over 31-year window between AISMR and (a) EPTNMAR, (b) WCTNMAY and (c) WCTXAPR

PREDICTORS OF INDIAN SUMMER RAIN 1123

# 1997 Royal Meteorological Society INT. J. CLIMATOL, VOL. 17: 1115±1127 (1997)

Page 10: Pre-monsoon maximum and minimum temperatures over India in

where AISMR is all-India summer monsoon rainfall in mm, WCPMAM is the west central Indian pressure

(March±April±May) in hPa, EPTNMAR is the east-peninsular-India minimum temperature during March in �C,

and APR500 is the mean latitudinal location (�N) of the 500 hPa ridge along 75�E longitude in April.

The EPTNMAR parameter identi®ed in this study has entered the regression equation at step two, signifying its

importance compared with the other LRF parameters. The observed and estimated values of monsoon rainfall are

given in Figure 8. In the ®gure, an independent veri®cation of the regression equation during the period 1981±

1994 is also given. In the independent veri®cation, eight years (1981, 1982, 1984, 1987±1990 and 1992) have

shown good agreement (within � 5 per cent) between the observed and estimated values, whereas six years

(1983, 1985, 1986, 1991, 1993 and 1994) have shown large differences. This indicates that the regression model

has limited success over extended independent periods. However, it may be noted in this context that, in a

majority of the above years in which the predictions were off the mark, many other models, including those

developed from the latest data have also failed in the predictions. It appears, therefore, that there are still

additional forcings on the monsoon to be identi®ed in order to achieve a better performance in long-range

prediction.

30-year sliding regression analysis

In order to assess the performance of these two new predictors regarding the dependency of their entry into the

regression on the data window considered, 30-year sliding regression analysis, following the above stepwise

procedure, has been performed for the period 1951±1987. The list of parameters that entered the ®nal equation

and the corresponding multiple CCs for each sliding period are given in Table III. From the table it can be seen

that the parameter EPTNMAR has entered the regression model ®ve times (1951±1980, 1952±1981, 1954±1983,

1955±1984 and 1956±1985), and the other parameter, WCTNMAY, has dominated the regression schemes

continuously for six times from 1953±1982 to 1958±1987. For three periods both the parameters have entered,

Figure 7. Correlation patterns between subdivisional monsoon rainfall and EPTNMAR and WCTNMAY during 1951±1980. Light shadingindicates signi®cance at 5 per cent level and dark shading at 1 per cent level

1124 K. K. KUMAR, K. R. KUMAR AND G. B. PANT

INT. J. CLIMATOL, VOL. 17: 1115±1127 (1997) # 1997 Royal Meteorological Society

Page 11: Pre-monsoon maximum and minimum temperatures over India in

indicating their importance relative to the other LRF parameters that are being widely used in several prediction

schemes.

7. SUMMARY AND CONCLUSIONS

In this study an attempt has been made to develop predictor parameters for seasonal forecasting of the Indian

summer monsoon rainfall, using the maximum and minimum temperature data over India. Some studies in the

past have demonstrated the importance of minimum temperature compared with the maximum in foreshadowing

the performance of monsoon rainfall based on the data up to the mid-1970s. Of late, it has been observed in many

studies that the strength of the relationship between several LRF parameters and the monsoon rainfall has been

undergoing secular variations. Most of the predictors are strongly coupled with the monsoon rainfall variability

only since the late 1940s and particularly during the last three or four decades. In view of this, it is felt necessary

to re-examine the relationship between the maximum and minimum temperature and the monsoon rainfall with

an extended data set.

The following are the main conclusions from the above study.

(i) The minimum temperatures exhibit spatially coherent and statistically signi®cant relationships with the Indian

summer monsoon rainfall over two broad regions, one along the east coast and adjoining regions during

March and the other over the west central parts of India during May.

Table II. Parameters considered for the regression analysis and their correlations with AISMR during 1951±1980

Parameters Season/Month Acronym CC with AISMR

Regional conditions Sea-level pressure parameters:Bombay MAM-DJF BBPM-D 7 0�70**West central India MAM WCPMAM 7 0�63**Trivandrum MAM-DJF TRPM-D 7 0�64**Minicoy MAM-DJF MNPM-D 7 0�49**

Temperature parameters:West central India mean MAM WCTMMAM � 0�60**East Peninsular India minimum March EPTNMAR � 0�69**West central India minimum May WCTNMAY � 0�66**

Circulation parameter:500 hPa Ridge location along 75�E April APR500 � 0�70**

ENSO indicators Sea-level pressure parameters:Darwin MAM-DJF DWPM-D 7 0�63**Tahiti-Darwin MAM-DJF T-DPM-D � 0�38*Plaissene MAM-DJF PLPM-D 7 0�40*Santiago-Darwin MAM-DJF S-DPM-D � 0�52**Adelaide MAM-DJF ADPM-D 7 0�35*Cardoba MAM-DJF CDPM-D 7 0�37*Buenos Aires MAM-DJF BUPM-D 7 0�40*

Sea-surface temperature parameter:NINO4-SST MAM-DJF NO4TM-D 7 0�55**

Cross-equatorial ¯ow Agalega SLP MAM-DJF AGPM-D 7 0�44*Nouvelle±Agalega SLP MAM-DJF N-APM-D � 0�50*

Global/hemisphericcondition

Northern Hemisphere surfaceair temperature

JF NHTJF � 0�50**

**Signi®cant at 1 per cent level; *signi®cant at 5 per cent level.

PREDICTORS OF INDIAN SUMMER RAIN 1125

# 1997 Royal Meteorological Society INT. J. CLIMATOL, VOL. 17: 1115±1127 (1997)

Page 12: Pre-monsoon maximum and minimum temperatures over India in

(ii) Two new predictor parameters, EPTNMAR and WCTNMAY, have been developed using the minimum

temperature data of 12 and 10 stations, respectively, from each of the two regions forming the core of the

highest CCs.

(iii) The two newly identi®ed predictors have shown consistently high signi®cant CCs since the mid 1940s to the

end of the data period, i.e. 1994.

(iv) A multiple regression equation developed by a stepwise selection procedure from 19 regional and global

parameters including the two identi®ed above, has shown a multiple CC of 0�89 and contained three

parameters, all of which are regional: EPTNMAR is one of them.

(v) In a 30-year sliding regression analysis with the 19 predictors during the period 1951±1987 the predictors

EPTNMAR and WCTNMAY have entered the regression equation for ®ve and six times, respectively, out

of the total of eight data windows considered.

Figure 8. Observed and estimated AISMR during 1951±1994, using the prediction model developed for the period 1951±1980

Table III. 30-year sliding stepwise regression analysis during 1951±1987

Period Parameters that entered in the regressiona Multiple CC

1951±1980 APR500; EPTNMAR; WCPMAM 0�89321952±1981 APR500; EPTNMAR; WCPMAM; N-APM-D; NHTJF 0�91541953±1982 WCTNMAY; N-APM-D; WCPMAM; NHTJF 0�92591954±1983 WCTNMAY; NO4TM-D; N-APM-D; NHTJF; WCPMAM;

EPTNMAR; APR500; WCTMMAM0�9324

1955±1984 WCTNMAY; ADPM-D; APR500; EPTNMAR 0�88941956±1985 WCTNMAY; TRPM-D; APR500; EPTNMAR; PLPM-D;

WCTMMAM0�9128

1957±1986 WCTNMAY; AGPM-D; ADPM-D; APR500 0�86001958±1987 WCTNMAY; AGPM-D; ADPM-D; APR500 0�8667

aUnderlined predictors are those identi®ed in the present study.

1126 K. K. KUMAR, K. R. KUMAR AND G. B. PANT

INT. J. CLIMATOL, VOL. 17: 1115±1127 (1997) # 1997 Royal Meteorological Society

Page 13: Pre-monsoon maximum and minimum temperatures over India in

(vi) In view of the high predictive potential of these newly identi®ed parameters relative to the other known LRF

parameters, their inclusion in the LRF models could enhance the skill of all-India summer monsoon rainfall

prediction.

ACKNOWLEDGEMENTS

The authors are thankful to the Director, Indian Institute of Tropical Meteorology, Pune, for the necessary

facilities and encouragement to carry out this work. Thanks also are due to Dr B. Parthasarathy and Dr M. K.

Soman for critically going through the manuscript and for many useful discussions during the course of this work.

Data required for this study have been kindly made available by the India Meteorological Department.

REFERENCES

Blandford, H. F. 1884. `On the connection between Himalayan snowfall with dry winds and seasons of drought in India', Proc. Roy. Soc.,London, 37, 3±22.

Gowariker, V., Thapliyal, V., Kulshrestha, S. M., Mandal, G. S., Sen Roy, N. and Sikka, D. R. 1991. `A power regression model for long-range forecasting of southwest monsoon rainfall over India', Mausam, 42(2), 125±130.

Jennrich, R. 1977. `Stepwise regression: statistical methods for digital computers' in Ensleing, K., Relston, A. and Wilf, H. S. (eds), JohnWiley, New York, pp. 58±75.

Krishna Kumar, K., Soman, M. K. and Rupa Kumar, K. 1995. `Seasonal forecasting of Indian summer monsoon rainfall: a review', Weather50(12), 449±467.

Krishna Kumar, K., Rupa Kumar, K. and Pant, G. B. 1992. `Premonsoon ridge location over India and its relation to monsoon rainfall', J.Climate, 5(9), 979±986.

Meehl, G. A. 1994. `Coupled land±ocean±atmosphere processes and South Asian monsoon variability', Science, 266, 263±267.Montgomery, R. B. 1940. `Report on the work of G. T. Walker', Mon. Wea. Rev. (Supplement), 39, 26.Mooley, D. A. and Paolino, D. A. 1988. `A predictive monsoon signal in the surface level thermal ®eld over India', Mon. Wea. Rev., 116(1),

256±264.Mooley, D. A. and Parthasarathy, B. 1984. `Fluctuations in all-India summer monsoon rainfall during 1871±1978', Climate Change, 6, 287±

301.Parthasarathy, B., Sontakke, N. A., Munot, A. A. and Kothawale, D. R. 1987. `Droughts/¯oods in the summer monsoon season over different

meteorological subdivisions of India for the period 1871±1984', J. Climatol., 7, 57±70.Parthasarathy, B., Diaz, H. F. and Eisheid, J. K. 1988. `Prediction of all-India summer monsoon rainfall with regional and large scale

parameters', J. Geophys. Res., 93,(D5), 5341±5350.Parthasarathy, B., Rupa Kumar, K. and Sontakke, N. A. 1990. `Surface and upper air temperatures over India in relation to monsoon rainfall',

Theor. Appl. Climatol., 42, 93±110.Parthasarathy, B., Rupa Kumar, K. and Deshpande, V. R. 1991a. `Indian summer monsoon rainfall and 200-mb meridional wind index:

application for long-range prediction', Int. J. Climatol., 11, 165±176.Parthasarathy, B., Rupa Kumar, K. and Munot, A. A. 1991b. `Evidence of secular variations in Indian monsoon rainfall±circulation

relationships', J. Climate, 4(9), 927±938.Parthasarathy, B., Rupa Kumar, K. and Kothawale, D. R. 1992. `Indian summer monsoon rainfall indices: 1871±1990', Meteorol. Mag., 121,

174±186.Parthasarathy, B., Rupa Kumar, K. and Munot, A. A. 1993. `Homogeneous Indian monsoon rainfall: variability and prediction', Proc. Indian

Acad. Sci (Earth Planet. Sci.), 102(1), 121±155.Parthasarathy, B., Munot, A. A. and Kothawale, D. R. 1994. `All-India monthly and seasonal rainfall series: 1871±1993', Theor. Appl.

Climatol., 49, 217±224.Rupa Kumar, K., Krishna Kumar, K. and Pant, G. B. 1994. `Diurnal asymmetry of surface temperature trends over India', Geophys. Res. Lett.,

21(8), 677±680.Shukla, J. and Paolino, D. A. 1983. `The southern oscillation and long-range forecasting of the summer monsoon rainfall over India', Mon.

Wea. Rev., 111(9), 1830±1837.Walker, G. T. 1924. `Correlation in seasonal variation of weather, X. Application to seasonal forecasting in India', Mem. Indian Meteorol.

Dept., 24, 333±345.Webster, P. and Song Yang. 1992. `Monsoon and ENSO: selectively interactive systems', Q. J. R. Meteorol. Soc., Part B, 118(507), 877±926.

PREDICTORS OF INDIAN SUMMER RAIN 1127

# 1997 Royal Meteorological Society INT. J. CLIMATOL, VOL. 17: 1115±1127 (1997)