Evaluation of PRECIS simulation
Precipitation
The regional simulations generated from the three QUMPs using PRECIS were studied in detail to evaluate the skill of the model in representing the regional climatological features, especially the summer monsoon (June–September) and winter (December–February) rainfall characteristics. Seasonal rainfall statistics for the three simulations for three time slices and a baseline are given in Table 1. All three simulations resulted in a substantial wet bias over the basin. Of the three simulations, Q1 estimated lowest amount for the monsoon season over the whole basin with 236 mm rainfall and 54.8 mm standard deviation. This value was closest to the APHRODITE estimated value of 191 mm. Q0 and Q14 estimated rainfall of 333 and 286 mm with standard deviations of 75.2 and 53.1 mm, respectively. In the winter season, the Q0 simulation of 90 mm was closest to the APHRODITE estimate of 58 mm. Wet bias over South Asia has been reported by several previous studies (Rupa Kumar et al. 2006; Kumar et al. 2011; Syed et al. 2013). Rupa Kumar et al. (2006) suggested that wet bias may be partly due to the procedures used in determining the spatially averaged observed rainfall quantities. Weakening of the zonal temperature gradient in equatorial Pacific sea surface temperature (SST) in the model leading to increase in monsoon rainfall in response to warming has been attributed to wet bias (Turner and Annamalai 2012). A climate change study over India showed the Q0 and Q14 simulations to have a wet bias and Q1 a dry bias (Kumar et al. 2011), whereas other similar studies over India showed a wet bias (Rupa Kumar et al. 2006; Syed et al. 2013). The PRECIS simulations also overestimated the number of rainy days in a year (a day with area average rainfall ≥1.0 mm). APHRODITE estimated 106 rainy days annual average for the baseline period 1961–1990, whereas PRECIS estimated 159, 154, and 172 days for the Q0, Q1, and Q14 simulations, respectively.
Table 1 Seasonal and annual rainfall (mm) with standard deviation for baseline (1961–1990) and A1B future scenarios
The model has captured the general seasonal accumulation and monthly progression of the annual cycle reasonably well (Fig. 2). Although the model generally overestimated, the monthly values are within one standard deviation of observed values except for the months of July and August in the Q0 simulation. Q1 simulated values are closer to observed values for the monsoon months, and Q0 values are closer to observed values for the other months.
The spatial distribution of seasonal rainfall estimated by APHRODITE and as simulated by PRECIS for the baseline period (1961–1990) is shown in Fig. 3 for summer monsoon (June–September). Rainfall distribution over the whole basin was simulated reasonably well by the model for both the winter (December-February) and summer seasons. The baseline simulations appear to provide an adequate representation of the distribution pattern except over the eastern section of the northern parts of the basin. There were also quantitative biases in the simulated rainfall.
Temperature
Baseline maximum and minimum temperatures were compared with CRU data to evaluate the model. The maximum temperatures were used to evaluate the skill for the summer monsoon season (June–September) and the minimum temperatures for the winter season (December-February). Annual cycle of the mean monthly maximum and minimum temperature for CRU and three QUMP simulations is provided in Fig. 4. Over the Indus basin, the model simulated maximum temperatures were close to observed during the pre-monsoon season, during May and June the model overestimated and from October to February the model underestimates but all values were within one standard deviation. In case of minimum temperature, except during summer monsoon season the model underestimates.
Seasonal statistics for the maximum and minimum temperatures for the three simulations for three time slices and the baseline period are given in Table 2 and Table 3. During the summer season, the model estimated value for maximum temperature over the whole basin of 29.0 °C (Q0 simulation) was very close to the observed value of 28.8 °C. The mean maximum temperature for the summer monsoon season was highest in the Q14 simulation: 30.5 °C with a standard deviation of 0.7 °C. The values for Q0 and Q1 were 29.0 and 29.6 °C, with standard deviation of 1.0–0.8 °C, respectively. However, during the winter season, the model appears to underestimate the minimum temperature with values of −6.2, −5.0, and −4.9 °C for Q0, Q1, and Q14 simulations respectively, compared to the CRU estimate of −0.9 °C.
Table 2 Seasonal and annual maximum temperature (°C) with standard deviation for baseline (1961–1990) and A1B future scenarios
Table 3 Seasonal and annual minimum temperature (°C) with standard deviation for baseline (1961–1990) and A1B future scenarios
The spatial temperature distribution estimated from CRU data and as simulated by PRECIS for the baseline period (1961–1990) is shown for mean maximum temperature during the summer monsoon season in Fig. 5. The spatial pattern of mean minimum temperature also captured well by the model simulations, but with a distinct cold bias over the upper Indus. Of the three simulations, Q1 has a relatively cold bias, as previously reported by (Rupa Kumar et al. 2006) over India. The cold bias has also been reported by other studies and has been attributed to the prescribed land-use distribution in Biosphere Atmosphere Transfer Scheme (BATS) (Xue et al. 1996; Suh and Lee 2004; Dimri 2012).
Projections of future climate
Projected changes in annual cycles, precipitation
The baseline and projected monthly precipitation over Indus is shown in Fig. 6 (bars). The Q0 projections suggest a gradual increase in precipitation from the 2020s (2011–2040), to the 2050s (2041–2070), and 2080s (2071–2098) for all seasons except the months May and June. The increase is more marked in the monsoon season. The Q1 and Q14 simulations do not provide a clear indication of change; several periods show an increase in precipitation in the 2020s followed by a decrease in the 2050s and 2080s.
The Q0 and Q14 projections suggest changes in the intra-annual pattern in precipitation in the upper Indus with a significant overall increase and that there could be an increase in precipitation from November to March. The changes are clear for all three simulations in the upper Indus, but not as clear in the lower Indus (not shown).
Projected changes in annual cycles, temperature
The baseline and projected monthly maximum and minimum temperatures for three simulations are shown in Fig. 6 (lines). The monthly mean minimum and maximum temperatures show a consistent rise in the 2020s, 2050s, and 2080s in all three simulations over both the upper and lower parts of the Indus basin, but the overall shape of the annual cycle is unchanged. The temperature rise may be greater in the non-monsoon months.
Projected changes in spatial distribution, precipitation
The projected spatial distribution of seasonal rainfall changes over the whole Indus basin in the three future periods as simulated by Q0, Q1, and Q14 is given for the summer monsoon in Fig. 7. All three simulations suggest generally increasing monsoon precipitation in the 2020s. Q0 simulations suggest a much lower rise over the whole Indus basin in the 2050s and 2080s than in the 2020s. Q1 simulations indicate a decrease in monsoon precipitation in the southern and north-western parts of the basin in the 2050s and 2080s. Q14 simulations suggest little change in seasonal precipitation in the 2050s, but an increase in precipitation in the north eastern part of the basin compared to the baseline in the 2080s.
All simulations suggest a decrease in precipitation during the winter season over the southern part of the basin, and an increase over the upper part of the basin, in the 2020s. In the 2050s, Q0 suggests a marked increase in rainfall over most parts of the basin whereas Q1 and Q14 suggest a marked decrease over most parts of the lower Indus.
Overall the projected regional rainfall changes are broadly consistent with the general observation “the wet gets wetter and the dry become drier” (IPCC 2007).
Projected changes in spatial distribution, temperature
The spatial distribution of mean maximum summer temperatures is shown in Fig. 8. The annual values for the maximum and minimum temperatures are given in Tables 2 and 3 respectively. All three scenarios indicate a warming trend during the summer monsoon season, with a small rise in maximum mean temperature (about 1.0–1.5 °C) in the 2020s. Q14 suggests the highest warming, with the greatest change in the northernmost part of the basin and significant warming across a large area in the west of the lower Indus.
The simulations also suggest an increasing trend in temperature during the winter season (Fig. 9). Generally the rise in minimum temperature was greater over the upper Indus basin, especially in the northern part, with only a small rise over the central part of the lower Indus basin. The rise in minimum temperature over the whole basin was up to 2 °C in the 2020s, 2.5–4 °C in the 2050s, and more than 4 °C in the 2080s.
The three simulations Q0, Q1, and Q14 gave average rises across the whole basin in the 2080s from the baseline of 4.3, 3.9, and 5.1 °C in the annual minimum temperature, and 4.0, 3.4, and 4.6 °C in the summer maximum temperature, respectively. In other words, the projected rise in minimum temperature is more than the rise in maximum temperature. This indicates that the Indus basin may experience warmer winters in the future.
Analysis of extreme events
Precipitation
The Indus basin is particularly prone to floods and flash floods. Prediction of changes in extreme precipitation events can provide insights into likely changes in the magnitude and frequency of floods and flash floods, which in turn can help in planning appropriate mitigation measures. In order to study the impact of extreme events, we analysed the frequency of rainy days and the rainfall intensity. Figure 10 shows the number of rainy days for the baseline period 1961–1990 as calculated from gridded daily rainfall data and estimated by Q0, Q1, and Q14 simulations. The spatial pattern of the model results is very similar to the observed pattern but the model results show a higher number of rainy days especially in the north eastern part of the basin.
The projected changes in rainy days are shown in Fig. 11. All three simulations show an increase in the number of rainy days in the 2020s over most parts of the basin, with a decrease in the border area between the upper and lower Indus basins, the area with the highest amount of rainfall. All three simulations show an increase in rainy days over the northeastern part of the upper Indus basin in all three time periods. The Q0 simulation showed an increase in the number of rainy days in the 2050s, followed by a decrease over the central part of the basin in the 2080s. The Q1 and Q14 simulations indicate a decrease in rainy days over most part of the basin in the 2050s and 2080s with a slight increase over the eastern portion of the central Indus area.
Figure 12 shows the simulated rainfall intensity (mm/day) for the baseline period 1961-1990 as calculated from gridded daily rainfall data and estimated by Q0, Q1, and Q14 simulations. The spatial pattern of the model results is very similar to the observed pattern. The projected changes in rainfall intensity are shown in Fig. 13. Q0 simulation show slight decrease in the intensity towards northwest part of the basin in 2020s and progressively increases towards 2080s. All three simulations indicate increase of rainfall intensity over the northeast part of the border area between lower Indus and upper Indus towards 2080s.
Temperature
Temperature extremes were investigated by analysing highest maximum (summer daytime) and lowest minimum (winter night time) temperatures. The 1 °C bias corrected maximum and minimum temperature datasets generated by the Princeton University Hydroclimatology group (Sheffield et al. 2006) available for 1948–2007 were used as observation estimates for extreme temperature events. Figure 14 shows the lowest minimum temperatures as estimated by the Princeton University dataset, Q0, Q1, and Q14 simulations for the baseline period 1961–1990. The spatial patterns of the PRECIS simulations are similar to those of the observed values, but model values are higher for extreme maximum values and lower for extreme minimum values.
All three simulations show an increase in highest maximum temperatures throughout the basin in the 2050s and 2080s. In the 2020s, the Q1 and Q14 simulations show a decrease in extreme temperature events in small pocket areas. The Q0 and Q14 simulations show a rise in maximum temperatures of 4–8 °C, and Q1 of 4–6 °C, over most parts of the basin in the 2080s.
The projected increases in lowest minimum temperature are greater than the projected increases in highest maximum temperature, but the projections do not show any systematic pattern (Fig. 15). All simulations show a rise of more than 4 °C throughout the basin in the 2080s, with an increase of more than 8 °C over the border area between the upper and lower Indus basins. Significant warming, particularly in the upper basin could mean enhanced melting of the snow cover and glaciers leading to changes in the hydrological regime of the basin.