Exploring climate change impacts during first half of the 21st century on flow regime of the transboundary Kabul River in the Hindukush region

In transboundary river basins, climate change is being considered as a concern of higher degree than it is in other parts of the world (Milly et al. 2008). As is already being felt, climate change is likely to exacerbate the prevailing timing, quantity and quality of river flows across the globe (IPCC 2013), and transboundary basins are no exception. This may lead to ineffectiveness of the traditional practices of water resources management being reliant on past records of different components of the hydrological cycle (Milly et al. 2008). Hence, the nation states will feel more inter-dependency in terms of their shared water resources (Biermann & Dingwerth 2004), particularly in the highly populated South Asia region. As a matter of fact, the co-riparian in a transboundary river basin setting will not only be facing climate change impacts within their own territory but they will also be affected due to a climate-induced large-scale calamity across the border (Wolf 2009). This may call for a major shift in the focus of climate-related policies to not only address national or sub-national challenges but also to cater for transboundary threats. In the absence of such a shift, there is a high likelihood of disputes among the riparians leading to instability on both sides (Wolf 2009). This may also impact the effectiveness of existing treaties and water management regimes in the long term (Adger et al. 2009; Odom & Wolf 2011). Avoiding such problems and efficiently dealing with the threats induced by climate change will require, as a first essential step, substantial improvement in the understanding of likely climatic changes and its potential impacts on the transboundary waters.

One good example of a south Asian transboundary river basin facing the above-mentioned challenges and hence requiring immediate attention is the Kabul River Basin (KRB). It is a sub-basin of the Indus River system situated in the Hindukush mountain range. The Kabul River is an important right bank tributary of the Indus River and a transboundary river basin in South Asia between Afghanistan and Pakistan. It has a total catchment area of about 91,297 km² in both Afghanistan and Pakistan territories. Interestingly, both these countries are in a unique setting on the Kabul River that both are lower and upper riparian at the same time. The source of the Kabul River water is mainly the melt of seasonal snow which covers most of the basin during the winter season and some glacier-melt from the northern high mountains of the basin (Pervaz & Khan 2014). On average, the Kabul River contributes about 15.5 km³ annually to the Indus River flows (FAO 2011), which makes it very critical to analyse climate-induced variations in the streamflows of this river to better plan downstream uses of the Indus River flows.

Water resources from this basin are shared by Afghanistan and Pakistan and serve as a water supply source for more than 20 million people (Wi et al. 2015), making the shared use of transboundary water between these two countries central in establishing regional water resources development for this area. It is crucial to develop tools that can support engineering plans for existing and potential water infrastructure to take full advantage of the snow and glacier melt dependent water resources in the basin, especially under the existing and likely future threats from rising temperatures and climate change.

Serious implications are posed on snow and glacier coverage by warming and climate change as indicated by a few studies (e.g., Tahir et al. 2011; Khadka et al. 2014). Studies of glacierized basins in different regions have been carried out to improve the understanding of climate change impacts on the basin hydrology (Masood et al. 2018). Similarly, climate change is a threat of a higher degree for the large-sized basins as such basins are sustaining millions of human lives, wildlife and many different ecosystems which makes modelling of such large basins extremely important and challenging at the same time (Malagó et al. 2018; Remesan et al. 2018).

Snow and ice melt mainly control the runoff regimes in the large Himalayan sub-basins which are mainly governed by air temperature fluctuations (Archer & Fowler 2004). The progress in satellite remote sensing technology of the cryosphere have made it possible to apply those hydrological models that depend on snow coverage data for the basin, such as snowmelt runoff model (SRM) which can now be considered for simulation of larger basins (Tahir et al. 2011). Another example is Kumar et al. (2010) where SRM was employed to study the effect of climate change on streamflow in the Sutlej River basin (a sub-basin of the Indus Basin), which has contributions from snowmelt, rainfall and base flow. More examples can be found showing extensive use of the SRM around the world in snow-dominated mountainous basins (Abudu et al. 2012). Various agencies, institutes and universities have applied the model to over 100 basins, situated in 29 countries (Martinec et al. 2008). The largest basin on which the SRM has been applied is the Ganges River Basin, that covers a catchment area of 917,444 km² (Seidel 2013). The World Meteorological Organization (WMO) also performed successful tests with regard to SRM-based runoff simulations (WMO 1986). Immerzeel et al. (2009) studied the effects of snow cover dynamics on the discharge of the upper Indus and established that the prediction of streamflows can be efficiently made by using a high degree of accuracy of Moderate Resolution Image Spectro-radiometer (MODIS) snow cover data in the SRM. The combination of SRM with MODIS snow cover and Tropical Rainfall Measuring Mission (TRMM) rainfall data has significantly improved regional runoff modelling for the Himalayas (Bookhagen & Burbank 2010). First application of SRM on the KRB was reported in 1989, and initial simulations showed overestimation of streamflows as compared to the observed values (Dey et al. 1989). Rasouli et al. (2015) successfully applied SRM on the upper KRB to assess the impact of climate change on the overall discharge.

The application of SRM to various regions around the world, including the mountainous catchments of the Himalayas, makes the selection of this model for the snow- and glacier-fed KRB to simulate daily streamflows and analysis of projected flows under future climate change scenarios reasonably correct. Also, as the SRM simulations depend on snow-cover data rather than on precipitation data, the limited availability and the quality of precipitation data have a minimal effect on the SRM performance (Munir 2013), which further increases confidence in model selection (Mack et al. 2010). In this study, the WinSRM version 1.12 was applied to the KRB.

The main objectives of this study are as follows:

• 1.

  Simulation of daily streamflow in the snow- and glacier-fed transboundary KRB with limited ground observation stations.

• 2.

  Determine whether rising temperature and associated climate change has significant impact on the streamflow regime of the transboundary KRB under different future climate change scenarios (RCP 4.5 and RCP 8.5).

The study presented in this paper is unique in multiple ways:

• First, it is the only study which considers the larger transboundary KRB (includes the complete area of the basin on both sides of the border) for a climate change analysis.

• Second, it uses for the first time the bias-corrected APHRODITE (the Asian Precipitation Highly Resolved Observational Data Integration Toward Evaluation of Water Resources) data of fine spatial and temporal resolution (0.25° and daily, respectively), produced especially for the Upper Indus Basin (UIB) by Burhan et al. (2015) (will be referred to as ‘AP-Cor’ in the rest of the paper) in a hydrological modelling study of a large snow-fed basin.

The KRB lies between 65 to 75° east and 32.5 to 37.5° north and drains an area of about 91,297 km². It is a shared river basin between Pakistan and Afghanistan. However, as an upstream country, Afghanistan cannot unilaterally undertake water resources development in this basin without establishing cooperation with the downstream country. Much of the discharge of the Kabul River results from the melting snow accumulated during the winter season in the mountains. However, winter rains, which are common in late winter and early spring, falling on a ripe snow pack in the highlands, can greatly augment the flow of the main streams. The dominant climate in the KRB is semi-arid, i.e., higher evaporation rates than the annual total precipitation (Mack et al. 2013) which is characterized by cold winters with maximum precipitation (mostly snow) from November to May and warm to very hot summers with little or no precipitation (Akhtar & Iqbal 2017).

The Kabul River is 435 miles (700 km) long, of which, 350 miles (563 km) are in Afghanistan. It joins the Indus River northwest of Islamabad. Geographic location of the study watershed is shown in Figure 1 (distribution of the basin into the eight elevation bands shown in this figure will be explained in a later relevant section).

Methodology of the study, presented in Figure 2, involved the following components:

• 1.

  Data acquisition: snow cover (MODIS10A2) 8-daily images (processed snow cover images captured at 8-day interval), Shuttle Radar Topography Mission (SRTM) 30-metre digital elevation model (DEM) and hydro-meteorological data (AP-Cor (see the Introduction) precipitation and temperature; and observed river discharge) have been used in this study.

• 2.

  Mosaicking DEM tiles and delineating KRB watershed (originally done for a previous study by the main authors, i.e., Masood et al. 2018).

• 3.

  Snow cover tiles were added to the delineated watershed and the snow cover data were masked out.

• 4.

  A linear interpolation formula was applied to generate a daily time series from the 8-daily snow cover data (assuming that the snow cover changes linearly under linear variation in temperature).

• 5.

  Calibration and validation of the hydrological model (SRM) was carried out using daily snow cover data and daily hydro-meteorological data (AP-Cor precipitation, temperature data and observed river discharge data).

• 6.

  Running the validated SRM model with different sets of future climate data time series representing different scenarios (RCP 4.5 and RCP 8.5) and different future time slices (near future and far future) to simulate the projected flows of the Kabul River.

• 7.

  Analysis of the future river flow time series, obtained for different climate change scenarios, to assess changes in mean and extreme flows.

Groundwater analysis is beyond the scope of this study, since the scope of the project was mainly focused on understanding the impact of climate change on surface water flows. Also, this research work used the same snow cover area data in all model simulations for different climate change scenarios.

The dominant climate in the KRB is semi-arid, i.e., higher evaporation rates than the annual total precipitation (Mack et al. 2013), which is characterized by cold winters with maximum precipitation (mostly snow) from November to May and warm to very hot summers with little or no precipitation (Akhtar & Iqbal 2017).

In terms of hydro-meteorological data, ideally the study needed long-term (at least 25- to 30-year data as is a requirement in a climate analysis) daily data in different elevation zones of the basin and spatially well spread in the basin. In reality, only two climate observation stations with long-term daily data were available in Pakistan, namely, Peshawar (elevation: 500 m above sea level (m.a.s.l)) and Chitral (elevation: 1,500 m.a.s.l), and daily data from Afghanistan for the early 21st century period of the study analysis was not available. Hence, gridded AP-Cor daily precipitation and temperature data were used (Burhan et al. 2015), available from Pakistan Meteorological Department (PMD).

The APHRODITE (Yatagai et al. 2012) has long-term daily gridded precipitation and temperature data sets for Asia including the Upper Indus Basin. Burhan et al. (2015) noted that APHRODITE precipitation was considered to be underestimated at high altitudes, likely due to the valley-bottom location of precipitation observations and due to limitation of data availability from India. Hence, they performed bias corrections for temperature and precipitation of the APHRODITE data for the historical period using well-accepted temperature lapse rates worked out by other researchers for the Himalayan region and generated 1 km horizontal resolution fields. Full details of the bias correction process are available in Burhan et al. (2015).

AP-Cor gridded data were taken and selected a single grid cell in each of the eight elevations bands of the study watershed that was closest to the mean elevation of that particular band to act as a climate station of that band when providing input to the SRM hydrological model (shown in Figure 3).

The observed daily discharge of Kabul River at Nowshera was obtained from Water and Power Development Authority (WAPDA) for the period 1961–2009. As this is a climate change study, accordingly the 30-year period of 1981–2010 was considered as the ‘baseline’ period for analysing the changes in future river flows. Hence, Figures 4 and 5 present the baseline scenario of the Kabul River flows at Nowshera in terms of a linear trend in the daily discharge and annual maximum flow values. Examination of Figures 4 and 5 reveals that under the baseline scenario, there is a statistically steady trend in the long-term daily and annual maximum series.

For area-elevation analysis of the watershed, Shuttle Radar Topography Mission (SRTM) 30-metre digital elevation model (DEM) was downloaded from the US Geological Survey (USGS) website (https://earthexplorer.usgs.gov/). In an earlier study by some of the authors of the current study (i.e., Masood et al. 2018), the DEM tiles were mosaicked and the KRB watershed was delineated at coordinates of Nowshera and the same is also used here for this study. The delineated watershed of KRB is shown in Figure 1. The geographical coordinate system is associated with the ellipsoid WGS 84 (World Geodetic System 1984) and the cartographical coordinates are issued from a UTM projection (Universal Transversal Mercator) in Zone 42N of the northern hemisphere (Niard 2003). The mosaicked DEM was then clipped to the study area using the delineated watershed boundary shape file. Based on the elevation range of the watershed obtained from the watershed DEM, the whole watershed was divided into eight elevation bands, such that each elevation band contains almost the same area except the two highest elevation bands.

Hypsometric mean elevation is defined as the elevation with equal areas above and below that elevation in the zone. The study watershed's hypsometric curve was calculated using the area-elevation information derived from the delineated DEM. Figure 6 presents a plot between hypsometric mean elevations and cumulative areas of zones. The study watershed was divided into eight elevation zones/bands. Based on the authors' modelling experience in the HKH (Hindukush-Karakoram-Himalaya) region in terms of size of the basin under consideration and data availability issues and review of other relevant studies (e.g., Tahir et al. 2011; Rasouli et al. 2015), it was decided to select eight bands as the optimum number to divide this basin into. More than eight bands would increase the complexity of the model, especially considering data scarcity in the basin and a number less than eight may miss some important details of the basin. The elevation zones and hypsometric mean elevation values are listed in Table 1.

Based on the elevation range (difference of maximum and minimum elevation values) of the watershed obtained from the watershed DEM, the watershed was divided into eight elevation bands in such a way that the total watershed area is properly divided into these eight bands. The highest zone is very steep, hence change in area with respect to elevation is much less. This lead us to set the elevation range for zone 8 in a way that the area contained by this zone is not drastically different to other zones. The hypsometric mean elevation of each zone given in Table 1 refers to an elevation value which has almost equal zone area above and below it.

The snowmelt runoff model has been widely used in the simulation and forecast of streamflow in snow-dominated mountainous basins around the world (Khadka et al. 2014). The model is specifically designed to simulate and forecast daily streamflow in mountainous basins where snowmelt is a major runoff factor. Most of the precipitation in the KRB is in the form of snow. Therefore, for a reliable watershed simulation using the SRM, one needs good quality and reliable daily data of snow cover and non-availability of daily rainfall data does not affect the reliability and quality of simulation.

The model has three inputs, namely, temperature, precipitation and snow-covered area (SCA). Observed discharge of the Kabul River was used to compare the simulated hydrograph with that of the observed during the calibration and validation process.

• •

  Temperature input: The program accepts either temperature data from a single station, i.e., option 0 for basin-wide or from several stations, i.e., option 1, zone wise. Daily temperature data derived from AP-Cor (as explained earlier) were used by opting for option 1, zone-wise approach.

• •

  Precipitation input: The SRM accepts either a single, basin-wide precipitation input (option 0) or different precipitation inputs zone by zone (option 1). For this study, option 1 was opted for in the manner explained earlier.

• •

  Snow-covered area (SCA): The SRM takes snow-covered area as input on daily time intervals in the form of conventional depletion curve (CDC). CDC is the ratio of SCA in a zone to the total area of the zone. MODIS (Onboard TERRA Satellite) 8-daily images of 500 m spatial resolution have been processed for snow cover extraction. The snow cover has been calculated for the whole KRB. The following explicit assumptions were used in the calculation of SCA:

• • i.

    Cloud cover over snow covered area is taken as SCA.

  • ii.

    Cloud cover over land surface is ignored.

  • iii.

    Lake ice area is always added in the SCA.

• •

  Model parameters: Default model parameters were used which were determined based on hydrological judgement but not exceeding the physically acceptable values. The range of values for the parameters was deduced by the different research institutes in 42 basins of various size and elevation in 11 countries through the application of SRM (Martinec & Rango 1986). The ranges of parameter values used for this study are presented in Table 2. During the calibration, which is basically an iterative process, model tuning was done by changing the parameter values in each of the elevation bands within the preselected range, until a satisfactory level of agreement was reached between the observed and model simulated flow time series. Model evaluation (explained in the forthcoming text) was used to gauge the level of agreement between the two flow time series.

• •

  Model evaluation: It is based on the values of two statistics, i.e., (i) coefficient of determination and (ii) volume difference calculated by the model for each year of simulation.

Global climate model (GCM)-based future climate change scenarios have been generated by using the methodology of linear interpolation and bias correction (LIBC) statistical downscaling technique (shown in Figure 7; full details available in Burhan et al. 2015). Before applying the LIBC downscaling process, four GCMs were selected based on three-point criteria that ensures the data produced by the historical runs of these GCMs: (a) have a good correlation (greater than or equal to 0.88) with the baseline AP-Cor time series; (b) have a normalized root mean square error with the AP-Cor less than or equal to 0.15; and (c) have normalized standard deviations that lie within ±0.4 to that of normalized standard deviation of the AP-Cor data set. For full details of the GCM selection process and results, readers are referred to Burhan et al. (2015). Based on the pre-set three-parameter criteria (given above), four GCMs were found to compare very well (i.e., they are able to give the best reproduction of the AP-Cor data as compared to other competing GCMs). The selected GCMs along with their default spatial resolutions and standardized output statistics are shown in Table 3. In the next step, three data sets were obtained for each of the selected GCM, i.e., historical, RCP 4.5 and RCP 8.5. Then a statistical relationship between the AP-Cor and the GCM data (obtained in the previous step) was developed which was used to convert the GCM data into the AP-Cor spatial resolution. In the next step, the spatially downscaled GCM data were temporally disaggregated (from monthly to daily time scale) to make them usable for most of the popular hydrological models. Hence, the final product was downscaled GCM data at 1 km spatial resolution and at daily time scale (Figure 7).

Future climate projections for each of the eight elevation bands were generated from four climate models for near future (2011–2030) and far future (2031–2050) time spans under two RCPs scenarios: RCP 4.5 and RCP 8.5, and in the end a simple arithmetic average was used to assemble the future projections of the four GCMs. Readers are suggested to refer to Bokhari et al. (2018) for a detailed analysis of future climate projections for the Kabul Basin.

In this study, RCP 4.5 and RCP 8.5 scenarios were used to investigate the impact of radiative forcing on climate of the region and, in turn, its impact on the future water availability of transboundary KRB. RCP 4.5 is a stabilized scenario in which total radiative forcing is stabilized shortly after 2100, without overshooting the long-run radiative forcing target level (Burhan et al. 2015). Whereas in RCP 8.5 scenario, the 8.5 Wm⁻² case, carbon dioxide levels rise above a 1,300 parts per million by the end of the century (Burhan et al. 2015). As is obvious from the definition of these two scenarios, the selection of these scenarios was based on a strategy to analyse climate impacts on Kabul River flows under a scenario in which global efforts have resulted in a stability of greenhouse gase (GHG) emissions (i.e., RCP 4.5). Also, to see the impacts of a worst-case scenario wherein the world did not do much to mitigate GHG emissions, resulting in a very high level of GHG concentrations (i.e., RCP 8.5).

The downscaled future climate projections (under RCPs 4.5 and 8.5 and two-time slices, i.e., near future (2011–2030) and far future 2031–2050)) for each of the eight elevational bands were used as input to the validated hydrological model for generating the future river flow projections. The purpose of selecting near-term period and long-term period as future time slices for the analysis of anticipated changes in river flows is to provide information that helps to get better prepared for near and far future climate challenges to the Kabul River flows. Such information will be helpful in developing well-informed short- to long-term strategies and development plans related to KRB.

Variation of snow-covered area derived from the MODIS data (explained earlier) for KRB for the period of 2001 to 2016 is shown in Figure 8. The snow-covered area starts increasing from the month of August and keeps on increasing untill the month of February and then it decreases from February to the month of August. Minimum snow cover is in the months of July, August and September and maximum in December, January and February. A detailed snow-cover data analysis for assessing variations over a 16-year period has been presented by another recent study by the authors (Masood et al. 2018) and readers are encouraged to look at it. Masood et al. (2018) suggests that the cryosphere area in the KRB has a slight increasing trend on a seasonal and annual basis.

As mentioned in the Introduction, due to the fact that most of the precipitation in the KRB is in the form of snowfall, a reliable simulation of the Kabul River watershed using the SRM mostly depends on reliable daily data of snow cover of the watershed rather than daily precipitation data. Since the MODIS snow cover data, which were used here to fulfil the SRM input requirement, are not available prior to the year 2001 and the daily river discharge time series is available untill the year 2009, it was decided that the hydrological model will be run for some selected years from the period of 2001 to 2009, being the most recent decade of the observed data.

As has been mentioned in more than one place in this paper, that simulation of the Kabul Basin is highly challenging due to severe data scarcity as well as quality issues, therefore, after a number of initial trials, the period of 2002–2005 was selected for use as the hydrological model's calibration and validation period (presented in Figure 4). There are several studies in the published literature which have used a similar approach or even shorter time periods (especially for the Kabul Basin). For example, Tahir et al. (2011) used a four-year period for the model calibration/validation in their Hunza Basin study. Similarly, there are other good examples that used a similar approach (e.g., Ma & Cheng 2003; Rasouli et al. 2015 (Kabul Basin study); Firouzi & Sadeghian 2016). Certainly, if the model is performing very well for our selected calibration/validation period, it can perform equally well for other years provided climate data of a reasonable quality are available. This is the basic modelling assumption of this study, based on which the validated model was trusted for generating future flow projections.

Moreover, the SRM model treats each modelling year separately, which makes performing a multi-year continuous simulation very challenging, but still found a better choice to model data-scarce basins (Martinec & Rango 1986; Martinec et al. 1998, 2008). This makes it extremely difficult to come up with a single consistent set of SRM parameters that are valid for each simulation year and are able to produce daily river discharge values similar to the observed river discharge values for each year of simulation (Rasouli et al. 2015).

The SRM was calibrated for two consecutive years, 2002 and 2003, by providing three input variables and seven parameters into the model (given in Table 2). The runoff simulated by the model was calibrated against the observed discharge by tuning the model parameter values within the selected range (see Table 2). The parameter values for each year calibrated individually were adjusted further to obtain a universal set of parameters that gives maximum correlation factor and minimum average percentage volume difference for all the years of the selected calibration period. This universal set of parameters obtained in the calibration process was then used without any change for simulating the flow of the years 2004 and 2005, set as the validation years of the model.

The results of the calibration and the validation periods are shown in Figure 9 in the form of observed vs simulated hydrographs and values of the model performance evaluation statistics (which is a standard practice in hydrological modelling studies, e.g., Tahir et al. 2011). Examination of Figure 9 reveals the correlation coefficient (coefficient of determination in the SRM) for the calibration period was 0.86 whereas the correlation coefficient for the validation period was 0.71. The volume difference for calibration and validation was 4.13% and 10%, respectively. These results establish the reliability of the model and provide reasonable confidence to use this model as a tool for generating future flow projections to assess water availability under climate change scenarios.

Figures 10 and 11 present a summary of the projected changes in precipitation and temperature estimated for the watershed under two climate change scenarios for near and far future time slices (more details are available in Bokhari et al. (2018)). Overall, precipitation is projected to increase under all scenarios but the highest increase is for the RCP 8.5 scenario in the far future time slice (Figure 10 shows relative change values).

Figure 11 presents the projections of average temperature change in the watershed under two climate change scenarios and for two future time slices. As generally expected, overall, average temperature is projected to increase for both climate change scenarios in the near future. This increase is significantly higher for the far future under both RCP 4.5 and RCP 8.5, as compared to the near future time period under the same scenarios.

Daily time scale future projections' data of precipitation and temperature for the two selected climate change scenarios (RCP 4.5 and RCP 8.5) and two future time slices were used as input in the validated hydrological model (SRM) of the transboundary KRB to determine the projected future flows. Future daily time series of river flows thus generated was analysed for the two 20-year time slices, i.e., near future (2011–2030; 2020s) and far future (2031–2050; 2040s). The baseline period for this analysis was selected as 1981–2010 (a 30-year period) which was used to analyse changes in each future time slice.

Figure 12 shows the mean annual daily discharge of Kabul River for RCP 4.5 and RCP 8.5 compared to observed mean annual daily discharge during the baseline period. Examination of this figure reveals that there is a clear increase in river discharge in the near future and far future. In the near future time slice, flow variability is found to be increased, and in terms of standard deviation (SD) values it is 218, 289 and 284 for the baseline period, RCP 4.5 and RCP 8, respectively. Under the far future scenario, the changes in flow were similar to the near future scenario but to a more enhanced level, causing greater concerns. For example, see Figure 12, where, as compared to the near future scenario, a greater increase in annual average discharge is evident with much more variability of the Kabul River flows as SD values are 218, 240 and 359, for baseline, RCP 4.5 and RCP 8.5, respectively. The increase in the future flows in terms of magnitude and variability can be attributed to the projected increase in both precipitation and temperature. As most of the flow in the Kabul River is contributed by snowmelt, hence more snow means more contribution to river flow causing increased flows. Similarly, temperature rise would cause the snow to melt at faster rates which will, in turn, give rise to extremity of flows, higher annual maximum flows and more flow variability.

The increasing variability of flow is also depicted in Figure 13, where annual maximum (AM) flow time series for the baseline period is plotted along with the same for the near future time period under RCP 4.5 and 8.5 scenarios. The (rectangular) area shows the high to very high flood region for the Kabul River at Nowshera (∼4,000–5,600 m³/sec) set by Pakistan Meteorological Department as per international standards. Examination of Figure 13 reveals some interesting facts, such as the AM values of magnitude ≥5,000 m³/s are expected to occur more often than these have occurred during the baseline period.

To further augment this important finding, as in the case of the near future scenario, a comparison of AM discharge values for baseline and far future (under RCP 4.5 and 8.5) scenarios is presented in Figure 14. Again, the shaded (rectangular) area shows the high to very high flood region for the Kabul River at Nowshera set by Pakistan Meteorological Department as per international standards. A keen look at this figure shows that in the far future under both the RCP scenarios, the AM flow values in the order of ≥6,000 m³/s are becoming very frequent, which is of great concern in terms of risks associated with flooding events in the settled areas of this watershed.

To further analyse changes in intra-annual (monthly) flow patterns under near and far future scenarios, monthly flow hydrographs' comparison is presented in Figures 15 and 16. Findings of this analysis are consistent with the previously presented inter-annual analysis, but with some additional interesting as well as important information. For example, examination of Figure 15 shows that the monthly average discharges for the near future period (for both the RCPs) are increasing with a marked increase during the summer season where rising temperatures are causing an earlier start of snowmelt leading to a higher peak flow occurring one month earlier as compared to the baseline period.

Similar to the near future case, under the far future scenario (Figure 16), more warming is causing a huge increase in the average monthly discharge, especially in the four summer months, i.e., May–August and an earlier flow peak.

In order to assess changes in frequency of flood of different selected return periods under the analysed climate change scenarios, a flood frequency analysis has been performed. The Gumbel distribution (Kotz & Nadarajah 2000) has been fitted to each annual maximum (AM) time series obtained from the observed data and future flow time series data generated by using the validated hydrological model for different climate change scenarios.

The results of this analysis are presented in Figure 17, which is a comparative bar graph of percentage change in the magnitude of flood events of 2-, 5-, 10-. 100-, 200- and 500-year return period. Examination of Figure 17 shows that, overall, an increase in the range of 18–40% is projected for near and far future time slices under both RCP 4.5 and RCP 8.5 climate change scenarios. However, a closer look gives interesting yet alarming information in terms of flooding hazard. Under all climate change scenarios, except RCP 8.5 scenario for far future time slice, there is a general tendency of higher increase in magnitude for lower return period events, i.e., the more the return period, the lower is the projected increase in flood magnitude. We know that the higher frequency (low return period) events are the flash floods that are already a characteristic of this basin. However, under future climate change scenarios, these are projected to occur more frequently and with a higher magnitude. These findings are quite consistent with those presented in the previous section (Future water availability) of this paper and further affirm the threat being posed by the future flow projections for this basin generated in this study.

The study presented in this paper mainly aimed to carry out a modelling based analysis of future scenarios of the flows in the transboundary KRB to assess future water availability in this basin. Major findings and conclusions of the study are as follows.

The study has attempted to quantify the climate change impacts on water availability in the KRB using gridded climate data as there are only two climate observation stations in the Pakistani part of this basin having sufficiently long daily time series and non-availability of daily data on the Afghanistan side of the basin for the early 21st century for which the study analysis was designed. The analysis was performed for two selected future time periods, i.e., near future (2011–2030) and far future (2031–2050). An ensemble of four GCMs' RCP 4.5 and RCP 8.5 scenarios were used in this study. Analysis of the future climate time series shows an overall increase in temperature for the whole watershed within a range of 3–5 °C under RCP 4.5 and RCP 8.5 scenarios in the near and far future time periods. Also, an overall relative change in precipitation for the whole watershed is within the range of 1.3–1.7 (under RCP 4.5 and RCP 8.5 scenarios in the near and far future time periods). The combination of temperature and precipitation increase causes an increase in amount of snow and also in snowmelt rate, producing an increased amount of melt water that becomes available earlier during the year as compared to the baseline period. In summary, we found a larger contribution from snow- and icemelt and shift in the intra-annual flow patterns. Climate change impact assessment in terms of changes in river flows indicated that the basin's hydrologic regime will alter considerably under different climate change scenarios and future time slices. Some key findings related to the alteration in the hydrologic regime are: (i) higher annual average discharge with more variability (SD values: 218, 289 and 284, for baseline, RCP 4.5 and RCP 8.5, respectively, under near future and 218, 240 and 359, for baseline, RCP 4.5 and RCP 8.5, respectively under far future scenario); (ii) more frequent extreme events of higher magnitude, e.g. annual maximum values of magnitude ≥5,000 m³/s and ≥6,000 m³/s are expected to occur more often in the near and far future, respectively, as compared to the baseline period. This was further confirmed by performing a flood frequency analysis of 2-, 5-, 10-. 100-, 200- and 500-year return period flood events. The flood frequency analysis revealed that for most of the scenarios and future time slices, the higher the frequency (the lower the return period value), the smaller is the increase in flood magnitude, with an overall 18–40% increase in flood magnitude for all the climate change scenarios and future time slices. It poses a serious threat of more frequent and higher magnitude extreme flow events in the form of localized flash floods in the mountainous part of the basin and larger floods in the flatter terrain (mostly in KP province of Pakistan, e.g., Nowshera and adjacent areas) of the basin. Hence, it creates a huge safety concern to a major proportion of the population living in the KRB on both sides of the border.

However, the future flooding threat can be taken as an opportunity for a sustained cooperation between the two riparian countries. Following are some policy recommendations that could be adopted by both the countries:

• A joint campaign by the meteorological agencies of the two riparian countries for installation of climate and river flow gauging sites in the basin, especially on the Afghanistan side which has virtually no long-term data available required for a reliable climate analysis. It will help improve hydrological monitoring and climate change impact assessment research studies in this basin. In this regard, Pakistan has better capacity and can offer full technical support to Afghanistan.

• Joint watershed management practices including tree plantation to reduce soil erosion and flood risk, and flood plain management (restoring and creating wetlands) to make beneficial use of flood flows.

• A formal mechanism of cooperation/coordination among the national weather and flood forecasting agencies in both the countries through enhanced knowledge and data sharing and mutual capacity building.

• Developing jointly-run flood early warning systems for saving, as much as possible, precious lives and property in both the countries.

• In future, joint modelling studies for this basin should be carried out by the relevant researchers in both the riparian countries and a multiple-model hydrological modelling approach should also be used to quantify uncertainties in the future projection of flows.

The authors acknowledge with thanks the LEAD Pakistan for collaboration in this research work. The authors are also grateful to WAPDA, MODIS and SRTM DEM teams for providing discharge flow, snow covered and DEM data, respectively, free of cost. The authors are also grateful for the free access to SRM model by USDA. The authors declare there is no conflict of interest.