Projection of hydro-climatological changes over eastern Himalayan catchment by the evaluation of RegCM4 RCM and CMIP5 GCM models

An assessment of climate change impacts and available water resources as per 21st-century demands is essential for long-term sustainable development at national or regional level. The energy balance of our climate system has been changing due to the profusion of greenhouse gases and aerosols in the atmosphere, which has further increased radiative forcing (Taylor et al. 2012; IPCC 2013). Many studies have indicated that the impacts of climate change on natural and human systems have been more pronounced in the last few years (IPCC 2013). As per the Intergovernmental Panel on Climate Change (IPCC) fifth climate assessment report (AR5; IPCC 2013), the realistic future climatic scenarios (under representative concentration pathways such as RCP 2.6, 4.5, 6.0, and 8.5) were generated in consideration of radiative forcing (Taylor et al. 2012; IPCC 2013). The global climate models (GCMs) under Coupled Model Intercomparison Project phase 5 (CMIP5)-based generated scenarios are available to analyze the freshwater-related risks of climate change (Taylor et al. 2012; Field et al. 2014; Das et al. 2016). CMIP5 models have been used for analysis of the various climatological and hydrological applications (Kharin et al. 2013; Akhter et al. 2016; Das et al. 2016). As per the CMIP5 GCMs and their RCP experiments, it was projected that there would be a decrease in water resources in dry subtropical regions and at mid latitudes; however, there will be an increase in water resources in humid mid-latitude and at high latitudes regions (IPCC 2013).

The Coordinated Regional Climate Downscaling Experiment (CORDEX) and the CMIP5 model data have been constructing varied meteorological information at both regional and global scales under multiple RCP experiments (Gao et al. 2012; Park et al. 2013; Lee et al. 2014). To overcome the limitations of GCMs, i.e., coarser resolution (grid size ∼100 km), regional climate model (RCM)-based projections are generally adopted, which have a relatively finer spatial resolution (Park et al. 2013). RCM-based future projections of temperature changes were found to be efficient in highlighting the local scale variability (Park et al. 2013). However, a few studies have also highlighted the bad performance of RCMs in simulating precipitation for different locations (Zou & Zhou 2013). The performance of the RCMs (e.g., RegCM4), as part of the CORDEX project in the simulation of precipitation at various geographical locations has been validated (Gao et al. 2012; Oh et al. 2014). However, the relative performance of RegCM4 RCM and CMIP5 GCMs together at local scale has not been considered, especially over Himalayan catchments.

This study primarily demonstrates an integrated analysis framework for an assessment of the Regional Climate Model 4 (RegCM4) and CMIP5 Coupled Physical Model (CM3), Coupled Climate Model phase 1 (CM2P1), and Earth System Model (ESM-2M), and their applicability in future climate projections of hydro-climatological variables at different locations over Teesta River Himalayan catchments. The RegCM4 has been considered as the finest resolution climate model (Park et al. 2013; Xue-Jie et al. 2013). The CMIP5 GCMs have already shown their usefulness in various climate change studies by various researchers around the world at different spatial scales (Kharin et al. 2013). Understanding water cycle and water balance components (e.g., streamflow, precipitation, snowmelt, and water yield) and their responses to future climate with enhanced temperature was the secondary objective of this study. A physical and meteorological weather generator parameter-based deterministic hydrological Soil and Water Assessment Tool model (SWAT; Arnold et al. 1998) was used in the simulation and projection of water balance components.

In hydrological modeling, several uncertainties (e.g., parameter-based uncertainty) generally arise, which affect the performance of the model in accurately simulating the water balance components (Zhang et al. 2014; Singh et al. 2015). The SWAT Calibration and Uncertainty Program (SWATCUP)-based Sequential Uncertainty Parameter Fitting version 2 (SUFI2; Abbaspour et al. 2007; Abbaspour 2011) was used for model calibration and parameterization sensitivity analysis. Uncertainties make GCM datasets unreliable (Taylor et al. 2012; Kharin et al. 2013). This study does not consider GCM data uncertainties. However, a statistical method of bias correction has been applied under the presence of observed variables to reduce the gaps between large-scale variables and local-scale variables. The bias correction method, initially utilized by Mahmood & Babel (2012), was applied to reduce the uncertainties in the downscaled variables. The Mann–Kendall (MK) test (Kendall 1975; Mann 1975) was applied to detect significant trends in the projected precipitation and temperature scenarios (2006–2060).

For the present study, the Teesta River catchment (up to Chungthang gauge location) was selected, which is a part of north Sikkim eastern Himalayas, India. The catchment area is around 2,552.57 km² and it exhibits diverse topographical characteristics (Figure 1). Teesta River originates from Khangchung Chho, which is a highly elevated glacial lake situated in the eastern Himalayas. The river flows southwards through gorges in the northeastern Sikkim Himalayas over a 309 km stretch. During its course through the Himalayas, the river is fed by many rivulets. Lachung River is the largest tributary of the Teesta River, and meets it at Chungthang. Originating from the Lachung Chhu glacier, Lachung River's catchment corresponds to extreme elevation variations (Figure 1). The upstream parts of the catchment correspond to extreme elevation peaks (around 7,400 m), while the lower parts of the catchment correspond to moderate elevation peaks (around 1,500 m) (Figure 1). The predominant monsoon climate of the Teesta catchment supports the evergreen rainforests, including grasses. In this region, about 80% of the annual rainfall is brought by southwest monsoon during the June to September months (Singh & Goyal 2016). The maximum, minimum, and average annual average precipitation variations across all the sub-catchments (SB) are 3,238.60 mm to 3,354.60 mm, 1,399 mm to 1,569 mm, and 2,346.71 to 2,460.05 mm, respectively. The high elevated upstream parts of the Teesta River catchment are mostly covered by snow and glaciers and the moderate elevated lower parts are supplied with rainfall during the monsoon and summer seasons.

Fine resolution gridded (0.5° × 0.5°) daily precipitation and minimum–maximum temperature datasets (0.5° × 0.5°) were used for this study. These datasets were collected from the Indian Institute of Tropical Meteorology (IITM) and Indian Meteorological Department (IMD) for the period of 1981 to 2005. The datasets were prepared by the utilization of quality-controlled observed precipitation data collected from more than 1,800 gauges. Additionally, measured daily precipitation datasets (from 1981 to 2005), available at the Lachung and Chungthang gauging stations, were utilized for this study. For SWAT model calibration and validation, about 15 years' daily observed discharge datasets (1991–2005), available at the Central Water Commission (CWC), were used. Out of the 15 years' available data, the first 10 years' data were used for calibration and the remaining 5 years were used to validate the model. The other meteorological variables, such as humidity, wind speed, and solar radiation (1981–2005), were obtained from the global weather data of SWAT (https://globalweather.tamu.edu/). The landuse/landcover (LULC) map and soil map of the study area were downloaded from the WaterBase website (http://www.waterbase.org/download_data.html).

To highlight the topographical variations, the whole catchment was divided into seven sub-catchments (SB), which also highlighted the spatial variations in climatological factors (e.g., temperature and precipitation). Due to the extreme to moderate elevation differences, the temperature lapse rate (TLR) and precipitation lapse rate (PLR) varied significantly from the upstream parts to downstream parts of the catchment (Singh & Goyal 2016). Seven climate stations – Chopta valley (SB1), Thangu (SB2), Muguthang (SB3), Lachen (SB4), Yumthang (SB5), Lachung (SB6), and Chungthang (SB7) – were defined based on the elevation variation to project the local scale changes (Figure 1). The gridded and measured minimum–maximum temperature and precipitation were adjusted at each SB by calculating the TLR and PLR (Gardner et al. 2009; Singh & Goyal 2016). In this study, due to large variations in the topography and climate of the Teesta, we observed significant variations from SB1 to SB7 (Das et al. 2012). For the computation of TLR and PLR, the elevation differences within the SBs were calculated from Cartographic Satellite (CARTOSAT) Digital Elevation Model (DEM), which was downloaded from Bhuwan portal, ISRO, India (www.bhuwan.nrsc.gov.in).

For the downscaling of daily temperature and precipitation datasets, the Statistical Downscaling Model (SDSM; Wilby et al. 2002) was used on a daily scale at each SB utilizing RegCM4 (RCP4.5), ESM-2M (RCP4.5), CM3 (RCP4.5), and CM2P1 (RCP4.5) climate models. To adequately cover the various circulation domains, six GCM grids (2.5° × 2.5°) and 12 RegCM4 grids (0.5° × 0.5°) covering the Teesta catchment were chosen as the spatial domain of the 18 most relevant GCM predictors and nine most relevant RCM predictors (Figure 1 and Table 1). These six GCM and 12 RCM grid points were spatially interpolated (point scale) at each climate station using an inverse distance weighted averaging (IDWA) technique. The IDWA interpolates the GCM and RCM grid points (for each predictor) as per the weighted average of each grid point at the known observed point (predictand, such as temperature and precipitation).

The SDSM downscaling method involves the empirical relationships between the large-scale predictor(s) and local-scale predictands (Wilby et al. 2002; Mahmood & Babel 2012). SDSM utilizes linear regression for estimation of the parameters of daily precipitation and minimum–maximum temperatures, separately. However, SDSM also incorporates conditional and stochastic weather generators. Therefore, SDSM is known to be sophisticated compared to a simple regression model (Wilby et al. 2002). In SDSM, regression coefficients for linear regression are computed using the ordinary least-square method (Mahmood & Babel 2012). Selection of predictors plays an important role in statistical downscaling. The predictor selection was carried out using correlation and partial correlation analysis. First, the highly correlated predictors were selected from the 18 GCM predictors and nine RCM predictors with reference to predictand (as listed in Table 1); 12 GCM and nine RCM most highly correlated predictors were finally used (as shown in Tables 2 and 3, respectively, for GCM and RCM). Further, a quantitative method was adopted for final selection of the most suitable predictors, as discussed by Mahmood & Babel (2012). In this method, correlation matrices between 12 predictors for GCMs and nine predictors for RCM (as shown in Tables 2 and 3) and the predictand (observed data) were prepared. The predictors with high correlation coefficients were taken and arranged in descending order. The highest correlated or first ranked predictor was taken as the first suitable super predictor (FSSP). The absolute correlation (R) between predictand and predictor variables and the coefficient of correlation between individual predictor variables were also calculated. We should not chose highly correlated predictors to avoid multi-co-linearity in the predictors. Thus, the other highly correlated predictors were not considered to reduce multi-co-linearity among the predictor variables. For precipitation and temperature, the thresholds were taken as 0.5 and 0.7 between the two predictors, respectively. The selection of the other predictors was carried out using percentage reduction in an absolute partial correlation (PRP) with respect to absolute correlation, as discussed by Mahmood & Babel (2012). The smallest value (more negative values) shows minimal co-linearity among the predictors. As per the correlation matrix among the predictors, the second most suitable predictor was identified in the same way such that the predictor corresponded to minimum correlation (or the minimum PRP in partial correlation) (Tables 2 and 3). The minimum PRP predictor variable had nearly no or a very insignificant multi-co-linearity with the FSSP. The third, fourth, and following predictors were selected by reiterating the same procedure.

For future scenario generation, the combined calibration/validation of SDSM was performed using observed daily minimum and maximum temperature, and precipitation datasets. For the model calibration and parameterization, SDSM was developed on monthly sub-models (Wilby et al. 2002). The model performance was evaluated using multiple evaluation methods such as coefficient of determination (R²), root mean square error (RMSE), mean bias and standard deviation.

A non-parametric significance test such as MK (Kendall 1975; Mann 1975) was used for the trend analysis of the meteorological variables. The MK test quantifies the magnitude of change in the time series datasets (Subash & Sikka 2014). The significance level to determine significant increase or decrease was defined (α) = 0.05. As per significance level (α), the Z-statistic values larger than +1.96 or lower than −1.96 designate the significantly increasing (positive) or decreasing (negative) trends, respectively. In this study, trend analysis was performed on monthly time series of precipitation and temperature.

For the SWAT setup, weather generator parameters daily precipitation, daily minimum and maximum temperature, daily humidity, daily wind speed, and daily net solar radiation were used. The physical and topographical parameters LULC, soil parameters, and DEM were processed to generate the drainage and physiographical characteristics of the catchment.

The SUFI2 algorithm is incorporated in the SWATCUP tool (Abbaspour et al. 2007). The parameter-based sensitivity analysis actually helps to recognize the significance of a particular parameter in calibration, and whether the process is influenced by the parameter values if it changes. A global sensitivity analysis approach, using two sensitivity evaluation functions, P value test and statistical t-test, was used to rank the highly sensitive and non-sensitive parameters. In SUFI2, an iterative process updates the old coefficient parameters into new coefficients during the calibration process to achieve the final estimates using the stratified Latin hypercube sampling method (Abbaspour 2011). The SUFI2 algorithm basically assumes a large parameter uncertainty range (or physically meaningful range that can occur due to the used data inputs such as LULC, precipitation, temperature, etc.) to make sure that the observed data lie in the 95% prediction uncertainty (95PPU) band for the first iteration and then reduces this uncertainty band in a step-wise manner while monitoring the p-factor and r-factor. The 95PPU (e.g., upper and lower uncertainty ranges) can be calculated between 2.5% and 97.5% levels of the cumulative distribution of an output variable obtained through Latin hypercube sampling. It disallows 5% of the very bad simulations (Abbaspour 2011). Theoretically, the value of p-factor ranges between 0 and 100% and r-factor ranges between 0 and infinity. The value of p-factor = 1 and r-factor = 0 corresponds to measured data equivalent to observed data.

The predictor selection was carried out separately according to the methodology discussed. Tables 2–4 present a comparison between the predictors selected for GCMs and RCM. For example, for precipitation and SB1 in GCM, tas has the highest value of R (=0.62), hence, tas was chosen as the FSSP (Table 2). Similarly, the FSSP were chosen for other variables and SBs. The selection of other predictors was carried out based on the values of partial r and PRP. For example, for precipitation and SB1 in GCM, rhs, tas, and vas were selected as the second, third and fourth predictors, respectively, based on the values of PRP and partial r (Table 2). A similar procedure, as discussed in the Methodology section, was used for selection of predictors for different variables, SBs, and models (i.e., GCM and RCM). Tables 2–4 also highlight the differences in the suitable predictors for both model types. For example, in the case of GCM, tas, tasmin, vas, rhs, tasmax, and pr were the most widely selected predictors for precipitation, whereas for RCM, huss, psl, and rsds were the most widely selected (Su et al. 2016). Singh & Goyal (2016) highlighted the scope of similar GCM predictors over the eastern Himalayan catchment and Mahmood & Babel (2012) also revealed the same over the western Himalayan-based Jhelum River catchment. Similarly, GCMs and RCM also showed little dissimilarity in suitable variables for maximum and minimum temperatures. For maximum temperature, tas was the FSSP for most of the stations in the case of RCM and for all stations in the case of GCM. For minimum temperature, again, tas was the first suitable predictor for all stations except SB5 in the case of GCM. On the other hand, like precipitation, for minimum temperature also, huss was the most used FSSP for RCM. Shivam et al. (2017) worked on the Subansiri River basin and also applied the same approach to find out the best suitable super predictors.

The performance of GCMs and RCM for calibration and validation periods (within historical time series duration) was evaluated and compared using three statistics, R, R², and RMSE, for all three downscaled variables, as shown in Table 5. For precipitation, RCM showed higher R and R² values compared to GCM for all the stations, which indicated a better correlation between downscaled and observed precipitation records. However, RMSE for RCM was higher compared to GCM, which indicated the presence of larger error in the case of RCM (Table 5). Based on RMSE, it followed that GCM downscaled precipitation was more accurate compared to RCM downscaled precipitation. For maximum temperature, RCM showed higher R and R² compared to GCM. Considering RMSE, it can be seen that RCM performed better at four out of seven stations, which showed that the RCM was more accurate than GCM. Similar to the previous two variables, RCM was able to follow the trend for minimum temperature and showed better R and R². RCM showed less RMSE than GCM for all stations except SB7, which showed that RCM was better than GCMs for minimum temperature also.

Figure 2(a) shows the annual average maximum and minimum temperature trends for the observed period (1981–2005). A significant increase in observed minimum temperature can be seen in Figure 2(a). Figure 2(b) and 2(c) show the projected scenarios (2006–2060) for maximum and minimum temperatures, respectively, for three GCMs and RCM. An increasing trend in both minimum and maximum temperatures can be observed for all models. The RCP-based projected scenarios of minimum and maximum temperatures show a consistent increase in temperature (2006–2060). Comparing both the models, the RegCM4 projection showed higher increase than GCMs. Shivam et al. (2017) and Su et al. (2016) also used the CMIP5 GCMs and observed a consistent increase in temperature over the Himalaya. To find and quantify the trends present in the sub-catchment (SB)-wise projected scenarios of precipitation, maximum and minimum temperatures, the MK test was performed. Figure 3 shows the MK-based significant and non-significant trend analyses and their comparisons for projected maximum and minimum temperatures, and precipitation utilizing RegCM4 (RCP 4.5), CM3 (RCP 4.5), ESM-2M (RCP 4.5), and CM2P1 (RCP 4.5) models. The magnitude of (e.g., increasing or decreasing) the trends for significant trends was assessed by MK test statistic Z (as explained in the Methodology section). For maximum temperature (Figure 3(a)), an increasing trend was found for all the sub-catchments and for most of the months with a few exceptions like January, June, and July where a decreasing trend was also observed. Similar trends were observed by Shivam et al. (2017) over the eastern Himalayan catchment. In January, a decreasing trend was found for all sub-catchments except the SB7 where a significant increasing trend was found. A significant decreasing trend was found for the months of June and July using RegCM4 RCM, whereas different GCMs showed mixed trends (increasing as well as decreasing). For the rest of the months, a clear significant increasing trend was found. However, their magnitude of change varied for different SBs and months. In the case of maximum temperature, SB1 and SB7 showed maximum changes compared to other SBs, which illustrated that these two portions of the catchment would be highly affected by climate change. Similar to maximum temperature, in the case of minimum temperature (Figure 3(b)), an increasing trend was obtained for all months and for most of the SBs. For post-monsoon months (July–November), the CM3 model showed a very significant increase in the minimum temperature. The Z-statistics for other models, for monsoon months, were less compared to the CM3 model. For SB1 (most elevated station), in the month of July, a very significant decreasing trend was observed using CM2P1 and RegCM4 models, with Z-statistics less than −5.00, whereas for the same case the other two models showed a significant increasing trend. Similar to maximum temperature, the minimum temperature also showed maximum change for SB1 and SB7 compared to other SBs. The overall trends showed significant warming signatures over the hilly terrains of the Teesta River catchment as projected by GCM and RCM datasets. Figure 3(c) shows the MK trend analysis for precipitation scenarios (2006–2060). The Indian subcontinent gets precipitation mainly in monsoon, which generally happens during the months of June to September (Singh & Goyal 2016). All four models showed an increasing trend for precipitation in pre-monsoon months (March–May), whereas a decreasing trend was observed for monsoon months (June, July, and August). Similar findings were also observed by Shivam et al. (2017) over the Brahmaputra region. Projections from all models were found to be highly variable from SB1 to SB7. The uppermost sub-catchment (SB7) corresponded with very high elevation zones and was thus always covered with snow and glaciers, while SB1 belongs to moderate elevation ranges and does not receive snow during summer. Therefore, the climatology of the SB7 significantly differs from that of other catchments. Several previous studies also highlighted similar observations over the high hilly terrains of Himalayas (Collins et al. 2013).

As discussed in the Methodology section, SWATCUP was used for calibration and sensitivity analysis of SWAT model parameters. The sensitivity of the streamflow and snowmelt-related parameters was computed using 15 parameters, curve number (CN2), baseflow (ALPHA_BF), groundwater delayed flow (GW_DELAY), minimum flow for groundwater (GWQMN), soil evaporation compensation factor (ESCO), snowfall temperature (SFTMP), soil water content (SOL_AWC), soil hydraulic conductivity (SOL_K), Manning roughness for main channel (V__CH_N2), effective hydraulic conductivity in main channel (V__CH_K2), bank erosion (V__ALPHA_BNK), snow cover maximum (SNOCOVMX), snow cover minimum (SNOCOVMN), snow cover at 50% snow coverage (SNO50COV), and precipitation lapse rate (PLR) in this case, to overcome the model uncertainty issues in the model calibration. The parameterization and uncertainty analysis was performed using R² and NSE objective functions. The best optimized coefficients and parameter values were opted for after running a number of iterations. The most sensitive parameters were found to be curve number (CN2), groundwater delay (GW_DELAY), maximum snow water content (SNOCOVMX), and fraction of snow volume (SNO50COV) (Singh et al. 2015). These parameters showed a significant response in optimizing the best coefficient values to project the future scenarios of streamflow, snowfall, and snowmelt. A total of three iterations were performed and in each iteration, 500 simulations were carried out to get the optimal coefficient and parameter values.

The r-factor and p-factor are computed as 0.48 and 0.30 for calibration, and 0.54 and 0.45 for validation, respectively, at Lachung station, whereas the values of p-factor and r-factor were recorded as 0.58 and 0.14 for calibration, and 0.55 and 0.45 for calibration and validation, respectively, at Chungthang station. The uncertainty analysis outcomes show that the simulated data were mostly falling within the uncertainty band (upper 95PPU). As per Zhang et al. (2014) and Singh & Goyal (2017), the above results are acceptable as compared to their studies. The r-factor shows reasonable uncertainty level in simulated discharge. The coefficient of determination (R²) was 0.52 and 0.60 for calibration and validation, respectively, at Lachung station, whereas the values of R² were 0.62 and 0.61 for calibration and validation, respectively, at Chungthang station. NSE was found to be 0.38 and 0.53 for calibration and validation, respectively, at Lachung station, whereas NSE was 0.46 and 0.77 for calibration and validation, respectively, at Chungthang station. The model performance was found to be better at Chungthang, as indicated by the values of R² and NSE. The observed discharge and modeled discharge were recorded as 800 m³/sec during monsoon time and 300–400 m³/sec during the summer season.

The projected scenarios of streamflow and precipitation at basin level were computed and compared as per the GCM and RCM scenarios (Figure 4(a) and 4(b)). A calibrated SWAT model was used to simulate the projected water balance components based on the downscaled precipitation and temperature (i.e., GCM and RCM) as input to the model. Here, it is assumed that up to 2060 there will be no significant changes in LULC and soil except for climatic factors, as the upper portion of the Teesta catchment has a very small population and is mostly covered with snow and glaciers. Figure 4(a) and 4(b) show the basin-level variations in precipitation and water yield scenarios, respectively, from the year 2008 to 2060 (the first two years 2006–2007 were taken as the warm-up period for the SWAT model). Figure 4(a) shows a significant increase in precipitation from the year 2008 to 2060. ESM2M predicted less precipitation compared to other models. Overall, an increasing trend was found for all models. Figure 4(b) shows a basin-level comparison of projected water yield scale, simulated using the SWAT model, under multiple GCMs and RegCM4 projected scenarios from the year 2008 to 2060. All scenarios showed a consistent significant increase in water yield from the year 2008 to 2060. Among all the GCMs and RCM, RegCM4-based projected scenarios showed a maximum increase in water yield.

Further, the projected scenarios of streamflow at both the gauges (i.e., Lachung and Chungthang) were plotted from the year 2008 to 2060 to highlight the increasing trend in streamflow (Figure 5(a) and 5(b)). All the models showed a significant increase in streamflow at both the gauge locations. The average annual streamflow varied from 20 m³/s to 45 m³/s and from 100 m³/s to 200 m³/s for Lachung and Chungthang, respectively. The CM3 and CM2P1 models showed a higher increase compared to ESM-2M and RegCM4. However, a sudden increasing trend in streamflow was observed for all the models after 2040. The RegCM4 and ESM-2M models showed an almost similar increase in streamflow amount in the future time domain.

This study demonstrated a comprehensive research work related to climate change and water balance components over a snow glacier-induced Himalayan catchment. In this study, the three latest CMIP5 GCMs and one RCM were used to highlight and compare the projected scenarios through statistical evaluation methods. This study showed the comparative responses of the high resolution RCM and moderate resolution GCM datasets in the assessment of the main water balance components. The RegGCM4 model performed well for temperature downscaling; whereas, in the case of precipitation, there was not much difference between GCMs and RCM. The overall performance of the RegCM4 model was found to be comparatively superior to GCMs as it showed higher accuracy for the statistical evaluation. The SUFI2-based model parameterization and uncertainty analysis provided great scope for improving modeling performances, especially in the case of future forecasting. The projected scenarios showed a significant increase in precipitation, water yield, and streamflow at each sub-catchment and both the gauge locations. The RegCM4-based projections of the minimum and maximum temperature clearly highlighted the extremity of temperature increments over the snow glacier-induced Himalayan catchment.

This present research work has been carried out under the Department of Science and Technology (DST), Government of India, Research Project No. SB/DGH-66/2013 and is awarded to Dr Manish Kumar Goyal. The financial support from DST is gratefully acknowledged. We also thank Geophysical Fluid Dynamics Laboratory (GFDL), New Jersey, USA, Earth Explorer NASA, Central Water Commission (CWC), New Delhi, India, the Indian Meteorological Department (IMD) Pune, India, and Indian Institute of Tropical Meteorology (IITM), Pune, India for providing the required datasets for the completion of this research work.