4.1 Analysis and Comparison of Available and Bias-Corrected Datasets of Precipitation
The boxplot (Fig. 3.0) shows considerable variations in the dataset's minimum, median, and maximum values. The IMD data is the closest to the Observed data, and NEX-GDDP is better than the SMHI dataset for the sub-catchment at Sarasgaon. The NEX-GDDP and SMHI projected datasets with five common GCMs exhibit underestimated values at quarterly and yearly scales. For the sub-catchment at Sarasgaon, the IMD data shows the best correlations with observed data with coefficients of 0.36,0.95,0.99, and 0.86 for daily, monthly, quarterly, and yearly time scales, respectively (Fig. 4.0). The following best values are given by NEX- GDDP data with correlation values as 0.25,0.82,0.92 and 0.42, respectively. The same pattern follows for NSE values, with IMD as best and the NEX-GDDP as following best. The percentage difference, in general, for the daily time scale goes high, up to 300% for all datasets. Still, it then comes down to the range of (-1 to 1.5) % for IMD, (-2 to 2.5) % for NEX-GDDP and (-10 to 20) % for SMHI at a monthly scale, about -0.5% quarterly for all datasets; (-0.2 to - 0.08) % for IMD, (-0.62 to -0.5) % for NEX-GDDP and (-0.75 to -0.55) % for SMHI at yearly scale. The same analysis for sub-catchment at Sarasgaon but with five different GCMs under NEX-GDDP (the top 5 available but not in SMHI) was also analyzed. There was minor improvement magnitude-wise and the distribution of values about the median.
Finally, the GCM ensemble data of the top five models available in the NEX-GDDP dataset - GFDL-CM3, GFDL-ESM2G, INM-CM4, MIROC5 and MPI-ESM-MR was selected to be used for Historical (1971-1999), RCP 4.5 (2071-2099) and RCP 8.5 (2071-2099) Scenarios for the sub-catchment at Sarasgaon.
It's noteworthy that the ensemble data combining RCP 4.5 and RCP 8.5 for the timeframe between June 1, 2016, and May 31, 2021, did not exhibit significant deviations when compared to observed data on a daily scale. At monthly, quarterly, and yearly timescales, it shows considerable differences. Thus, quantile mapping bias correction was applied to correct the under-estimated data values.
After bias correction, the median precipitation value of corrected RCP 4.5 and 8.5 datasets gets upgraded by 2-3 mm on a daily scale, 60-70 mm on a monthly scale, 150-200 mm on a quarterly scale and 700-800 mm on a yearly scale, for the catchment at Sarasgaon. The RCP 4.5 and 8.5 datasets show variations in 75% quantiles and maximum values with respect to the Historical dataset. These variations sum up to produce a difference of about 100-150 mm in median values of RCP 4.5 (2071-2099) to the Historical (1971-1999) and about 200-250 mm in median values of RCP 8.5 (2071-2099) to the Historical (1971-1999) at yearly scale as shown in (Fig. 5).
4.2 Maximum Temperature
From (Fig. 6), for the catchment at Sarasgaon, the corrected RCP 4.5 and 8.5 datasets show an upgrade of 4℃-5℃ at daily, monthly, quarterly, and yearly scales in the median values. The IMD values are a bit over-estimated and, thus, corrected by reducing them by about 2°C. (Fig. 7) shows that the corrected RCP 4.5 and 8.5 datasets have median maximum temperature values greater by 2°C-4°C with respect to the Historical period at daily, monthly, and quarterly scales, but at yearly scale, the median value of RCP 4.5 is found to be greater by about 2.5℃ and RCP 8.5 by 4.5℃ to the Historical one.
4.3 Minimum Temperature
For the catchment at Sarasgaon, (Fig. 8) shows that the median values of corrected RCP 4.5 and 8.5 datasets are upgraded by about 2.5°C at daily, monthly, and quarterly scales, but at yearly scale, they show a difference correction of 4℃ on the higher side of the median value. The IMD values show a large spread of all four quartiles at the yearly scale, but it is corrected to match the observed data. (Fig. 9) indicates that the median value of RCP 4.5 is about 3℃ higher than that of the Historical one at all timescales, and that of RCP 8.5 is about 6℃ higher than the Historical period median value at all timescales.
4.4 Rainfall-Runoff Modelling
Calibration and Validation of Model Setup
The calibration period was from June 1, 2016, to May 31, 2018, and the validation period was from June 1, 2018, to May 31, 2021 (Fig. 10). The calibration and validation NSE of the model setup of the catchment at Sarasgaon are 0.79 and 0.73, respectively. No snowpack is seen in the calibration output, whereas in the validation output, a small snowpack of more than 20 mm is spread over five layers for a short duration. There are differences in high-peak flow simulations with the observed one during the calibration period, whereas during the validation period, there are differences in the low-flow simulations.
Model Simulations of Past and Future Input Meteorological Data
The streamflow is simulated for the reference period (or Historical period from 1971 to 1999) by inputting past rainfall and temperature data (from June 1, 1971, to May 31, 1999), keeping the calibrated parameters constant. Small snowpacks are seen in the output, and the maximum flow goes up to 25 mm day−1 (Fig. 11). For the future period (for RCP 4.5 scenario from 2071-2099), by the input of future rainfall and temperature data (from June 1, 2071 to May 31, 2099), the streamflow is simulated keeping the calibrated parameters constant. No snowpack is seen in the output, and the maximum flow shows more than 30 mm day−1 (Fig. 12). Similarly, the streamflow is simulated for the future period (for RCP 8.5 from 2071-2099) by the input of future rainfall and temperature data (from June 1, 2071, to May 31, 2099), keeping the calibrated parameters constant. No snowpack is seen in the output, and the maximum flow goes up to 35 mm day−1 (Fig. 13).
To analyze the differences in the discharge time series of all three periods, (Fig 14) is plotted to show that the magnitudes of streamflows in the RCP 8.5 scenario are greater than those in the RCP 4.5 scenario, which are also more significant than that in the Historical period during the monsoon season. Though different for these periods, the time base has the same number of days. The maximum flows for RCP 8.5, RCP 4.5, and the Historical period go up to 200 m3sec−1, 175 m3sec−1, and 140 m3sec−1, respectively.
Changes in Simulated Streamflow and Runoff Depth
The average annual runoff depths of Historical, RCP 4.5, and RCP 8.5 scenarios were 885.15, 912.13, and 1053.60 (in mm), respectively. Also, the plot (Fig. 15) of the average total quarterly runoff depths of all three scenarios gave interesting results for the JJA (June, July, August), SON (September, October, November), DJF (December, January, February) and MAM (March, April, May) quarters. The runoff depth increases in JJA and SON quarters but decreases in DJF and MAM quarters from Historical to RCP 4.5 to RCP 8.5 scenarios.
The boxplot in (Fig. 16) shows that the median values of annual maximum flood peaks of simulated Historical (1971-1999) period, RCP 4.5 (2071-2099) and RCP 8.5 (2071-2099) scenarios lie at about 85 m3sec−1, 95 m3sec−1 and 120 m3sec−1 respectively. The minimum, 25%, and 75% quantiles of RCP 4.5 and 8.5 fall higher than those of the historical period, but the maximum quantiles of RCP 4.5 and RCP 8.5 lie at approximately the same level. The boxplot in (Fig. 17) shows that the median values of the occurrence of the annual maximum flood peaks of simulated Historical (1971-1999) period, RCP 4.5 (2071-2099) and RCP 8.5 (2071-2099) scenarios lie at about 75 days, 65 days and 70 days respectively from the start of the water year (i.e. June 1); which means at August 14, August 4 and August 9 respectively. Interestingly, the 25% quantiles of RCP 4.5 and RCP 8.5 fall earlier than those of the Historical period, i.e. towards the end of July month, whereas the 75% quantiles of RCP 4.5 and RCP 8.5 fall later than those of the Historical period, i.e. towards the end of August. The maximum quantiles of RCP 4.5 and RCP 8.5 fall towards the end of September, whereas the minimum quantiles of RCP 4.5 and RCP 8.5 fall towards mid-July and end-July, respectively.