Evaluating the impact of climate change on extreme temperature and precipitation events over the Kashmir Himalaya

The frequency and severity of climatic extremes is expected to escalate in the future primarily because of the increasing greenhouse gas concentrations in the atmosphere. This study aims to assess the impact of climate change on the extreme temperature and precipitation scenarios using climate indices in the Kashmir Himalaya. The analysis has been carried out for the twenty-first century under different representative concentration pathways (RCPs) through the Statistical Downscaling Model (SDSM) and ClimPACT2. The simulation reveals that the climate in the region will get progressively warmer in the future by increments of 0.36–1.48 °C and 0.65–1.07 °C in mean maximum and minimum temperatures respectively, during 2080s (2071–2100) relative to 1980–2010 under RCP8.5. The annual precipitation is likely to decrease by a maximum of 2.09–6.61% (2080s) under RCP8.5. The seasonal distribution of precipitation is expected to alter significantly with winter, spring, and summer seasons marking reductions of 9%, 5.7%, and 1.7%, respectively during 2080s under RCP8.5. The results of extreme climate evaluation show significant increasing trends for warm temperature-based indices and decreasing trends for cold temperature-based indices. Precipitation indices on the other hand show weaker and spatially incoherent trends with a general tendency towards dry regimes. The projected scenarios of extreme climate indices may result in large-scale adverse impacts on the environment and ecological resource base of the Kashmir Himalaya.


Introduction
Climate change is central to the global environmental management and sustainability issues of the twenty-first century, given its implications to natural and human systems (Campbell et al. 2011;Gan et al. 2015). The perturbation of global radiation balance owing to increasing emissions of CO 2 and other greenhouse gases is heating the atmosphere and causing global warming (Chu et al. 2010;Huang et al. 2011). From the perspective of climate change impacts, the most noticeable and heavily felt are the extreme weather and climate events because of their immediate and disastrous consequences on the natural and social environment.
Climate change has enhanced the probability of temperature and precipitation extremes (Fischer et al 2013), and several studies (Orlowsky and Seneviratne 2012;Kharin et al. 2013;Sillmann et al. 2013) have projected the changing patterns of extreme climate events along with rising anthropogenic greenhouse gas emissions. The change in the frequency and magnitude of extreme weather events is also considered as an indicator of changing climate in a region (Easterling et al. 2000;Zhang et al. 2011). The study of extreme weather and climate events has attained significant attention, however, the majority of studies on long-term climate changes have focused on changes in mean values (Kostopoulou and Jones 2005;Alexander et al. 2006). This is primarily because of the lack of long-term high-quality observational data, required for the detection and attribution of the extremes (Zhang et al. 2005). Global circulation models (GCMs) provide a promising option for simulating the present and future climates and generation of long-term time-series data, which can be used to analyze the possible changes in future extreme events (Sillman and Roeckner 2008).
GCMs have been the primary tools for understanding the present and future state of the climate of the planet earth. The data provided by GCMs are being used to carry out the impact studies from global to regional scales; however, the coarser grid resolution limits their direct applicability at the regional and local levels. The larger grid size of GCMs does not account for regional, sub-grid features (e.g., topography, cloud, and land use) that play a significant role in determining the local climate (Mearns et al. 2003;Coulibaly et al. 2005;Koukidis and Berg 2009). This scale mismatch is being dealt with using a wide range of dynamical and statistical downscaling techniques (Nguyen et al. 2006(Nguyen et al. , 2007Fowler et al. 2007). The dynamical downscaling involves the nestling of a higher resolution Regional Climate Model (RCM) within a GCM to obtain the resolved climate scenarios. This approach can generate the climate data at much finer scales ∼ 0.5° latitude and longitude scale (Fowler et al. 2007); however, being computationally intensive (Wilby and Wigley 1997), it has relatively restricted application in the local impact assessment studies (Hay and Clark 2003;Fowler et al. 2007). Statistical downscaling on the other hand relies upon the empirical relationships between the observed data and the large-scale predictors and has the advantages of being simple in application and also computationally less demanding (Samadi et al. 2013). Among the different Statistical downscaling approaches, Statistical Downscaling Model (SDSM) is a popular method used globally for climate change assessment and impact studies (Wilby et al. 2002;Gagnon et al. 2005;Chu et al. 2010;Huang et al. 2011;Mahmood and Babel 2014;Zhu et al. 2019). Numerous comparative studies have pointed out the ability of SDSM to perform better than other downscaling techniques in simulating the current and future climate variability with confidence Khan et al. 2006;Chen et al.2011;Teutschbein et al.2011;Samadi et al.2013). In addition to simulation of the mean climate variables, SDSM has a good capability in simulating climatic extremes (Hashmi et al.2010;Huang et al.2011) and thus has been adopted for the present study as well.
Numerous studies across the globe have demonstrated the increase in the magnitude, frequency, and duration of the climatic extremes in response to global warming (Trenberth 2011;Otto et al.2012;Fischer et al.2013;Morak et al.2013). In the central and south Asian region percentage of warm days/nights has increased while as there has been a decrease in cold days/nights from 1961 to 2000 over 70% of the sampled stations (Klein Tank et al.2006). These imprints are also well established over the Indian landmass. Kothawale et al.(2010) examined the trends in pre-monsoon temperature extremes for 121 stations in India and found widespread positive trends in the frequency of hot days and nights, while negative trends were observed for cold days and nights. Similar findings were reported by Dash and Mamgain (2011) for 1969and Revadekar et al.(2012) for 1970. Joshi et al.(2020 assessed the changing patterns of hot extremes over India and reported that the frequency of hot days has augmented by 24.7% in the recent past  compared with the past . This pattern is projected to continue in the future with marked increases and decreases in the frequency of hot and cold extremes respectively towards the end of the twenty-first century (Revadekar et al.2012). In India, the frequency of concurrent hot day and hot night (CNDHN) heat waves has increased substantially post-1984 and is projected to amplify by about 4-fold and 12-fold towards the mid-21st and end-twenty-first century respectively under RCP8.5 (Mukherjee and Mishra 2018). For the western Himalayan region, decreases have been reported in the percentage of cold nights while as there have been increases in the percentage of warm nights during the winter season (Dimri and Dash 2012). Global analysis of the precipitation-based extreme indices, on the other hand, manifests spatially heterogeneous trends with a tendency towards wetter conditions in some regions while drying trends in other regions (Donat et al.2013 also reported the same for the central and south Asian region. From 1910 to 2000 there has been an increase in the extreme precipitation events in contagious areas from NWH in Kashmir to the Deccan plateau in India (Roy and Balling 2004). In the Hindu Kush Himalayan (HKH) region significant positive trends have been reported for light and heavy precipitation events from 1961 to 2012 (Zhan et al.2017). Similar results have been reported in the Pir-Panjal range from 1991 to 2016 (Shekhar et al.2017). In the regional context, few studies have been carried employing SDSM for downscaling the mean temperature and precipitation (Mahmood and Babel 2014;Mahmood et al.2015;Shafiq et al.2019). Mahmood and Babel (2014) have also used SDSM for simulating some of the extreme temperature indices using the HadCM3 model and reported that there will be more warm extremes and fewer cold extremes in the future. Rashid et al.(2015) using the PRE-CIS simulations at a spatial resolution of 0.5° × 0.5° projected an increase of 7.23 °C (± 1.84) °C and 4.89 (± 1.51) °C in average annual maximum and minimum temperatures over Pahalgam station from 2011 to 2098. Furthermore, the average annual maximum and minimum temperatures over Gulmarg station were projected to increase by 7.68 (± 2.01) °C and 5.88(± 1.51) °C. Gujree et al.(2017) explored the historic and future changes in extreme temperature and precipitation events for the twenty-first century using the PRECIS RCM model at a grid resolution of 50 × 50 km and reported climatic extremities over the region are growing owing to changing climate. However, in this study (Gujree et al.2017), no bias correction of model results was done using the observed climate data; essential to assess the inherent systematic bias (Varis et al.2004;Ines and Hansen 2006;Christensen et al.2008;Teutschbein and Seibert 2010;Turco et al.2013). No detailed analysis has been carried out in this region earlier which assesses the spatio-temporal variations in extreme climate indices for the twenty-first century. The previous studies projecting the future climate change in the region have largely focused on changes in the mean climatology. However, the mean climatology smoothens a lot of information that can be pivotal in carrying out an impact assessment on the different climate-dependent natural and bio-physical sectors like agriculture, water resources, human health, etc. The present study is thus attempting to provide an updated and detailed description of future climatic variations and their impact on the extreme climate indices under different climate change scenarios. While evaluating the extreme climate, the study employs the suite of indices defined by the expert team on sector-specific climate indices (ET-SCI). The study utilizes the IPCC recommended representative concentration pathways (RCPs) in projecting the changes in the mean values (RCP2.6, RCP4.5, and RCP8.5) and extreme climate indices (RCP4.5 and RCP8.5). Besides the station scale investigation, the study projects the regional dimensions of the future climate change and will serve as a baseline study for formulating the regional and sector-specific adaptation strategies to mitigate the impacts of climate change in the fragile ecosystems of the Kashmir Himalaya.

Study area
This study is carried out for the Kashmir Himalaya that forms a substantial part of the northwest Himalaya (Fig. 1). The broad geomorphic architecture of the Kashmir Himalaya has evolved as result of collision between India and Eurasia (Alam et al. 2015(Alam et al. , 2017. The elevation varies between 1450 and 5500 a.m.s.l and results in microclimatic variations within the area (Fig. 1). Based on stratigraphy and elevation, the region has three distinct physiographic divisions i.e., mountains, Karewas (lacustrine deposits), and the valley floor with a network of tributaries draining into trunk river Jhelum (Bhat et al. 2019). Although located in the subtropical latitudes the climate of the area falls under the Sub-Mediterranean type because of the orographic controls and altitudinal variations. The average mean maximum and minimum temperature of the region is 19.3 °C and 7.3 °C respectively with an average annual precipitation of about 84 cm (Husain 1987). The climate is marked by sharp seasonality characterized by four distinct seasons based on the mean temperature and precipitation viz., spring, summer, autumn, and winter. The precipitation mainly occurs in the form of snowfall from the westerly disturbances in the winter season from November to April. With some monsoon incursions into the region, the enclosing Pir-Panjal mountain

Observed data
The observed time-series data for the maximum temperature (T max ), the mean minimum temperature (T min ), and the precipitation at daily and monthly scales were obtained from the Indian Meteorological Department (see Supplementary Material, S1). A quality check was done to identify the errors in the data before using it for the analysis. Negative daily precipitations were removed and both daily maximum and minimum temperatures were set to zero if daily T max was less than daily T min (e.g., Alexander et al.2006). Outliers in daily T max and T min , defined by values outside the range of four standard deviations (std) of the climatological mean of the value for the day i.e., mean ± 4 × std (Zhang et al. 2005;Alexander et al. 2006;Athar 2014) were identified and manually adjusted on a case to case basis to the data of neighboring stations. The next quality criterion was the homogeneity check of data series to make sure that the variations in the data are caused only by the variations in climate rather than other external factors (Aguilar et al. 2003;Campozano et al. 2014) using the double-mass curve analysis (Tabari et al. 2011). Various stations had missing daily data gaps ranging from months to years; however, monthly data was available which was later utilized for the bias correction. Individual scattered missing entries were filled with multiple imputation techniques using Statistical Package for the Social Sciences (SPSS) software. The data gaps of the range from months to years were substituted with the bias-corrected Climate Forecast System Reanalysis (CFSR) climate data.

Climate Forecast System Reanalysis (CFSR) data
CFSR climate data is valuable global data set provided by the National Centers for Environmental Prediction (NCEP) (Saha et al. 2010;Fuka et al. 2014;Dile and Srinivasan 2014). The data was downloaded for the study area using a bounding box of latitude 33.53°-34.54° N and longitude 74.25°-75.41 °E which generated the data at daily scale for 14 stations. The nearest CFSR data station ( Fig. 1) was chosen to infill observed station data. Raw CFSR data suitability was checked statistically by plotting it against the available data of the respective stations and a systematic bias was found in the raw CFSR climate data, therefore it was imperative to correct the bias utilizing the available observed monthly data. Bias in the CFSR data was corrected using the Linear Scaling (LS) method which aims to perfectly match the monthly mean of corrected values with that of observed ones (Lenderink et al. 2007). It operates with monthly correction values based on the differences between observed and raw data using Eqs. (1) and (2) where T bc, m,d and P bc,m,d are bias-corrected temperature and precipitation on dth day of m-th month and T raw,m,d and P raw,m,d are raw temperature and precipitation on the dth day of mth month. µ(.) is the mean value of the variable for mth month (Fang et al. 2015).

GCM/NCEP data
Historical and future simulated data on the predictors required for the SDSM were derived from the second-generation Canadian Earth System Model (CanESM2) developed by the Canadian center for climate modeling and analysis (CCCma) (von Salzen et al. 2013;Khadka and Pathak 2016;Huang et al. 2016;Zhu et al. 2019). The model is a part of the fifth Coupled Model Inter Comparison Project (CMIP5) and is currently the only CMIP5 model for which readyto-use SDSM predictors are available. The future modeled predictors are available for RCP2.6, RCP4.5, and RCP8.5. In addition to CanESM2 predictors, the National Centers for Environmental Prediction (NCEP) reanalysis predictors for the period 1961-2005 were used to calibrate and validate the SDSM model.

Description of SDSM model
The most popularly used downscaling model SDSM version 4.2 (Wilby and Dawson 2007) was used in this study. SDSM is a hybrid of multiple linear regression (MLR) and stochastic weather generator (SWG) (Meenu et al. 2013;Mahmood and Babel 2014). The SWG can generate a maximum of 100 ensembles while 20 ensembles are considered sufficient (Wilby et al. 2002;Gagnon et al. 2005;Chu et al. 2010). SDSM contains two kinds of optimization algorithms: (1) ordinary least squares (OLS) and (2) dual simplex (DS). The OLS was used because it is faster than DS and produces comparable results with DS (Huang et al. 2011). Moreover, annual and monthly sub-models can be developed during the calibration and generation of future scenarios. The annual model uses the same transfer function for 12 months while as in the monthly model 12 regression equations are developed for 12 months (Huang et al. 2011). The model can be set unconditional or conditional model depending upon the nature of predictant; for temperature unconditional model is used whereas precipitation is downscaled using a conditional model (Wilby et al. 2002;Chu et al. 2010). Unconditional models assume that regional predictors and the local predictand are directly linked e.g., the temperature is dependent directly on the large-scale predictors. In contrast to this conditional model have an intermediate variable e.g., daily precipitation amounts or evaporation is conditioned by the probability of wet-day occurrence (Wilby et al. 2002). Since the precipitation data generally does not possess a normal distribution, SDSM provides several transformations before using it in developing regression equations (Khan et al. 2006;Huang et al. 2011). The SDSM performs a sequence of steps i.e., screening of predictors, calibration and validation, and generation of future scenarios.

Screening of predictors
The selection of appropriate predictors is the most important step of any downscaling method (Wilby et al. 2002). This study employed a combination of the correlation matrix, partial correlation, and P-value (Gagnon et al. 2005;Huang et al. 2011;Mahmood and Babel 2014). The selection of the first predictor also called a super predictor is based on the highest correlation coefficient with the local predictant. The selection of subsequent predictors was done using the concept of percentage reduction in partial correlation (Mahmood and Babel 2014) and is given by the Eq. (3).
where PRP is the percentage reduction in partial correlation with respect to the correlation coefficient, P.r is the partial correlation coefficient of a predictor in the presence of super predictor, and R1 is the correlation coefficient between the predictor and predictants. The predictors having P-value > 0.05 are ruled out of the selection process and the predictor with the lowest PRP is selected as the second predictor. This procedure is repeated up to the desired number of predictors. Multi-collinearity between the predictors was also taken into consideration during the selection of predictors.

Calibration and validation
After the selection of appropriate predictors, the model was calibrated for each of the predictants using NCEP predictors (Fig. 2). Both the annual and monthly sub-models were developed for evaluation and a later monthly model was selected (Mahmood and Babel 2014). The model was set unconditional for temperature while for precipitation conditional model was used (Wilby et al. 2002;Chu et al. 2010). Moreover, the precipitation data were transformed to the 4th root to render it normal before using it for the calibration (Khan et al. 2006). Explained variance and standard error were chosen as indicators of the performance of the calibration process (e.g., Huang et al. 2011Huang et al. , 2012Mahmood et al. 2015). The model was validated by using independent data of 10 years for each of the meteorological stations (S2).
With the calibrated models 20 ensembles were generated for T max , T min , and Prcp. for both the calibration and validation periods. Moreover (R 2 ) was chosen to compare the simulated data (mean of 20 ensembles) with that of observed data (Huang et al. 2011;Mahmood et al. 2015).

Generation of future scenarios
The future scenarios for T max , T min , and Prcp. were generated by forcing the calibrated models with the CanESM2 predictors for the RCP2.6, RCP4.5, and RCP8.5. The future period (2006-2100) was divided into 3-time slices of near future, mid future, and far future i.e., 2020s (2011-2040), 2050s (2041-2070), and 2080s (2071-2100) respectively (e.g., Mahmood and Babel 2013). A total of 20 ensembles (Gagnon et al. 2005;Chu et al. 2010) were generated for each variable. The simulated values (mean of 20 ensembles) were compared with the observed means from the baseline data  to calculate the magnitude of changes in T max , T min , and Prcp. during the various spans of the twenty-first century with respect to the baseline climatology (1980-2010).

Calculation of extreme climatic indices
The future daily data series of T max , T min , and Prcp. simulated by SDSM was used to calculate the extreme climatic indices using the ClimPACT2 model for RCP4.5 (stabilization scenario) and RCP8.5 (business as usual scenario). From the suite of indices, 15 temperature indices (S3) and 14 precipitation indices (S4) were selected for the present study.

Trend analysis of extreme indices
Linear trends were computed for each index to explore the variability in the magnitude of indices under a changing climate during the various spans of the twenty-first century. The magnitude of the trend was estimated using Sen's slope estimator (Sen 1968) and the statistical significance of trends was established using Man Kendall's tau test at confidence levels of 90% and 95%. The trends were calculated for the future period of 2010-2100 and also for the near future (2010-2040), mid future (2041-2070), and far future (2071-2100) to explore the variability of extremes within each period. Moreover, the regional averages were computed for each index and were also subjected to trend analysis.

Screening of predictors
The selected predictors for the downscaling temperature (T max , T min ) and precipitation (Prcp.) are given in (S5 and S6) respectively. The temperature at 2 m height (temp) was the super predictor for both T max and T min ; precipitation being heterogeneous, the partial correlation coefficient between the predictors and predictants was generally low (e.g., Dibike and Coulibaly 2005;Hashmi et al. 2010;Huang et al. 2011) and as such the screening of predictors for downscaling the precipitation was complex. Among the predictors, 500 hPa geopotential height (p500), the meridional

Calibration and validation
The model was calibrated using the station-wise observed and NCEP predictors and validated for independent data of 10 years for each predictant. The percentage of explained variance E (%) and standard error (SE) were chosen as a measure of the performance of the calibrated model. E (%) for T max and T min varied between 48 and 70% and 54-66% respectively. The standard error (SE) ranged between 2.34 and 3.14 °C and 1.67-2.69 °C for T max and T min respectively. The E (%) for precipitation was generally low as compared to temperature and ranged between 12.6 and 27% with SE ranging between 0.43 and 0.51 mm. The E (%) is generally low for the heterogeneous variable like daily precipitation amounts (Wilby et al. 2002). The R 2 computed on the monthly series, varied between 94 and 97%, 93-96%, and 78-85% for T max , T min, and precipitation, respectively during the calibration period. For the validation period, R 2 varied between 91 and 95% and 89-93% for T max and T min respectively whereas for precipitation it ranged between 58 and 75%. The results are well in agreement with the previous studies (Liu et al. 2008;Huang et al. 2011;Souvignet and Heinrich 2011), shown graphically for the Srinagar station (illustrative purposes) in Fig. 2 to visualize the inter-annual variations during the calibration and validation process.  (Fig. 3a, b).

Projected changes in the temperature
Spatial variability is being observed in climatic variables as the rates of increase vary considerably between the different altitudinal ranges (S7). A relatively higher rate of warming is associated with the high-altitude stations (Pahalgam and Gulmarg) as compared to the low altitude stations. In the lower troposphere, the rate of warming varies directly with the altitude thus implying that temperatures will increase more in high mountains than at low altitudes (e.g., Bradley et al. 2006). On the seasonal scale, autumn is projected to have the maximum increases both in T max and T min (averaged over six stations) under all RCPs followed by the summer season (S8). Furthermore, the station scale analysis reveals that high-altitude areas will mark substantial increments in winter T min of order 0.23-0.37 °C (RCP2.6), 0.32-0.43 °C (RCP4.5), and 0.38-0.68 °C (RCP8.5) for the Gulmarg station and 0.43-0.86 °C (RCP2.6), 0.80-0.94 °C (RCP4.5) and 0.79-1.50 °C (RCP8.5) for the Pahalgam station. These are the only available stations representative of major glaciated and snow-covered areas suggesting that climate change will have a significant impact on the glacial health and snow accumulation (e.g., Akhtar et al. 2008;Immerzeel et al. 2010). These effects will be further amplified by the increases in average spring temperature that will result in early snowmelt (e.g., Barnett et al. 2005), thereby altering the surface runoff and water availability (Sharma et al. 2000;Beniston and Stoffel 2014). The inferences of the present study concur with other relevant studies (e.g., Babel 2013, 2014;Mahmood et al. 2015).

Projected changes in the precipitation
Precipitation shows a general decreasing trend in response to rising greenhouse gas concentrations when compared with the baseline precipitation . All the stations except the Gulmarg (showing marginal increases) are projected to experience reductions in annual precipitation of varying magnitudes under different RCPs during the twentyfirst century (S7). Under RCP2.6 annual precipitation shows a decrease of 1.05% (2020s), 0.75% (2050s) and 1.43% (2080s) in the region. RCP4.5 shows an average decrease of about 0.81% (2020s), 1.83% (2050s) and 2.30% (2080s). The precipitation is projected to decrease by 0.86% (2020s), 2.32% (2050s) and 3.36% (2080s) under RCP8.5 (Fig. 3c). On the station scale, Kupwara shows the maximum decrease of 6.61% (2080s) followed by Qazigund (5.69%) (2050s) and the Srinagar Station (5.5%) (2080s) under RCP8.5. The seasonal distribution of precipitation is projected to get altered during the twenty-first century with substantial decreases in the winter precipitation under all RCPs (S8). Changes in winter precipitation are directly related to glacial health and water resource availability in the region (Akhtar et al. 2008). spring precipitation shows an increase under RCP2.6 but decreases during the 2050s (2.07%) and 2080s (4.7%) under RCP8.5. On the other hand, summer precipitation shows an increase of about 0.68-2.82%, and autumn precipitation is projected to increase by 2.24-6.9% under different RCPs.
The changes in the seasonal distribution of precipitation will alter the runoff regimes and the water availability (Immerzeel et al. 2010;Beniston and Stoffel 2014). The spatial patterns of changes in mean annual temperature (Fig. 4) and annual precipitation (Fig. 5) under RCP2.6 and RCP8.5 were obtained by using the Inverse Distance Weighted (IDW) interpolation method. The average changes in seasonal T max , T min , and Prcp. under RCP2.6 and RCP8.5 are shown graphically (S9) for the comparison.  Fig. 4 Spatial pattern of the changes in the annual mean temperature (°C) under RCP2.6 (a-2020s, b-2050s, c-2080s) and RCP8.5 (d-2020s, e-2050s, f-2080s) with respect to baseline mean temperature     a-2020s, b-2050s, c-2080s) and RCP8.5 (d-2020s, e-2050s, f-2080s) with respect to baseline precipitation  indices and a decrease in the cold temperature-based indices have been reported (e.g., Zhang et al. 2005;Klein Tank et al. 2006;Tebaldi et al. 2006;Sillmann and Roeckner 2008;Athar 2014).

Trends in extreme precipitation indices
A total of 14 precipitation indices were selected for the analysis (Zhang et al. 2005;Tebaldi et al. 2006;Sillmann and Roeckner 2008;Athar 2014). The trends in precipitation indices are weak with only a few indices showing statistically significant trends over the various spans of the twentyfirst century (S12 and S13). Moreover, the trends, unlike the temperature-based indices, are spatially and temporally incoherent partly in the response to the poor performance of GCM models in resolving the local precipitation (Huang et al. 2012) and also due to greater spatio-temporal variation of the precipitation in the study area. This problem has been reported in numerous other studies of this type (e.g., Kostopoulou (2080s, 2011-2100) and Kupwara (2080s). Under RCP8.5 Kupwara shows a decrease (statistically significant) of about 6.6 mm/year during 2050s. The annual number of rainy days (rainfall > 2.5 mm) (R2.5 mm) exhibits increasing trends (statistically insignificant) for Srinagar, Kupwara (2050s), and Kokernag (2020s, 2080s, 2011-2100) while decreasing trends (statistically significant) were found at Kupwara (2080s, 2011-2100) under RCP4.5. Under RCP8.5 the decreases in R2.5 mm are statistically significant for Kupwara (2050s, 2011Kupwara (2050s, -2100 and Kokernag (2080s). The trends of other extreme precipitation indices of absolute thresholds (R10mm, R20mm), percentile-based thresholds (R95P, R99P, R95PTOT, R99PTOT), or duration based (RX1DAY, RX5DAY, RX7DAY) show a mix of increasing and decreasing trends without spatial coherence.

Trends in the regional average indices
To capture an overall picture of extremes in the region, regional averages of the indices were computed by averaging the values of indices over all the stations for the 2020s, 2050s, 2080s, and 2011-2100 (Tables 1, 2)  Among the precipitation indices, CDD shows insignificant decreasing trends under RCP4.5 while as, the trend shows an increase (statistically significant) under RCP8.5 over 2011-2100. The annual precipitation (PRCPTOT) decreases at the rate of 6.4 mm/decade (RCP4.5) and 4.2 mm/decade (RCP8.5) during 2011-2100. R2.5 mm is projected to decrease under both RCPs statistically significant during the 2080s (RCP4.5) and 2011-2100 (RCP8.5). The number of heavy rainfall days (R10mm) shows statistically significant decreasing trends under both the RCPs. In contrast to this, the number of very heavy rainfall days (R20mm) shows insignificant increases under RCP8.5. The total annual rainfall from very wet and extremely wet rain days (R95P, R99P) as well as the contribution of these to total rainfall (R95PTOT, R99PTOT) mark an increase although statistically insignificant under the RCP8.5. Similarly, Rx1day and Rx5day also show decreasing trends under RCP4.5 but increasing trends under RCP8.5. The Simple Daily Intensity Index (SDII) shows a decrease of 0.004 mm/ day under RCP4.5, while an increasing trend (statistically insignificant) at the rate of 0.001 mm/day is found under RCP8.5 for the period 2011-2100. The regional averages of the indices under RCP4.5 and RCP8.5 are shown in Figs. 6 and 7 to visualize the inter-annual variations for the period 2011-2100.

Discussion
Changing form of the climatic extremes is important evidence and impact of anthropogenic induced climate change. In the current study, future projections of temperature and precipitation showed significant changes for the twenty-first century with the temperature manifesting increasing trends and precipitation showing a decrease over the basin. The present study established the imprints of global warming on the climatic extremes in the region. The model projections show increasing trends in the warm temperature-based indices while decreasing trends were found in the cold temperature-based indices for the twentyfirst century. In the present study the warm extremes viz., amount of hot days/nights, fraction of days with aboveaverage temperature, warm spell duration, summer days, and tropical nights witnessed increasing trends under RCP4.5&8.5. Consistent patterns of the trends have been reported among the different regions of the world Tebaldi et al. 2006;Sillmann and Roeckner 2008). A similar pattern of trends resonates over the Indian landmass and the results of the present study conform with the previous studies like Kothawale et al. (2010), Dash and Mamgain (2011), and Revadekar et al. (2012). Revadekar et al. (2012) and Mukherjee and Mishra (2018) also projected the enhancement of the hot extremes in India over the twenty-first century under different climate change scenarios. While the cold extremes like the amount of cool days/nights, cold spell duration, and frost days are projected to decrease over the twenty-first century in the study area. Similar results have been reported by Gujree et al. (2017) who used the PRECIS RCM for projecting the extreme climate events in the Kashmir valley. The increase in the frequency and occurrence of the warm temperature-based extreme indices is consistent with global warming but the magnitude of change varies spatially due to the large-scale atmospheric circulation patterns associated with it (Meehl and Tebaldi 2004). Changes in the geopotential height anomalies at 500 hPa have been linked to the enhanced prevalence of the hot extremes (Ding et al. 2010;Loikith and Broccoli 2012;Lee and Lee 2016). Joshi et al. (2020) explored the physical driving mechanisms responsible for the observed increases in the frequency and intensity of hot extremes over the Indian landmass. The study cited that the augmentation of the geopotential height anomalies at 500 hPa over the northern parts of India is leading to increases in hot extremes. The persistent atmospheric subsidence associated with it results in adiabatic warming and increases atmospheric stability (Lee and Lee 2016). This phenomenon produces clear skies and decreases in the cloud cover, thus leading to increases in hot extremes due to pronounced surface warming. The decrease in the amount of the precipitable water is yet another mechanism yielding the enhancements in hot extremes over India (Joshi et al. 2020). The augmentation of temperature extremes over the twenty-first century could also be related to the inverse trends in the precipitation in the Kashmir Himalaya. The model projections manifested a decrease in the precipitation with similar trends for the extreme precipitation-based indices. The length of the dry spell and the consequent soil moisture deficit is projected to increase in the future. The positive feedback loop induced by the reduced precipitation is causing the reduction in the soil moisture content which in turn alters the atmospheric heat fluxes. The reduced soil moisture conditions favor the decreases in latent heat flux and increases in sensible heat flux which also leads to the enhanced surface warming (Fischer et al. 2007).
Some indicators of intensification of the precipitation regimes in the region are also there e.g., the number of very heavy rainfall days, total annual precipitation from heavy and very heavy rain days, maximum 1-day precipitation, and maximum 5-day precipitation show increasing trends. Although the trends are weak but such signals cannot be ignored knowing the flood susceptibility of the region. These indices could have a direct bearing on the flood risks in the basin as the region is very prone to flood disasters. Floods have been a recurrent feature in the basin but the flood risk has increased in the recent past due to encroachment of the wetlands/waterways, urban expansion, landfilling and dissection of the floodplain by railway/roads Bhat et al. 2019). Climate change has the potential to further exacerbate the flood risks and the losses. On the other hand, the length of dry spells showing increasing trends may cause droughts in different parts of the Kashmir Himalaya. Long dry spells are an emerging concern in the study area that lead to acute water scarcity and an increase in the drought episodes (Himayoun and Roshni 2019).

Conclusions
Climate change will have serious implications for the mountain ecosystems like the Kashmir Himalaya. In this study, a combination of the Statistical Downscaling Model (SDSM), and ClimPACT2 were used to assess the projected changes in the extreme climate (frequency and magnitude) over the twenty-first century. The average annual temperatures (T max and T min ) have shown an increase at all the stations although with different magnitudes and patterns under the different RCPs. The projected increase varied directly with the levels of radiative forcing. The patterns of increase also revealed the orographic controls with the high altitude areas registering relatively higher increases as compared to the low altitude areas. Seasonally, autumn exhibited the highest warming followed by summer season. On the other hand, precipitation has shown a decrease in most of the area with some increases at the Gulmarg station with respect to the baseline precipitation . Winter will be the most affected season showing significant decreases in the precipitation under all the RCPs. The projected variations in the extreme climate revealed that in response to the rising greenhouse gas concentrations in the atmosphere, the extreme climate is likely to intensify further in future. Conforming to the global trends, Kashmir Himalaya showed increasing trends in the warm temperature-based indices and a reduction in the cold extremes over the twenty-first century. The trends were more pronounced under the RCP8.5 due to higher levels of forcing and resultant warming associated with it. Precipitation indices comparatively showed divergent and spatially incoherent trends with an overall inclination towards drier regimes. The changes in the climate will have serious implications for sectors like agriculture, water resources, and human health. A decrease in the number of frost days, extensions in growing season length may be favorable for the agriculture sector but the increasing temperatures and decreasing precipitation will induce greater demands of water for irrigation purposes. Water resources and water dependent systems will be affected directly under the changing temperature and precipitation scenarios. Dry spells are a new trend of the climate in Kashmir Himalaya and recent years have marked abnormally long dry periods, creating acute water shortages. Prolonged dry spells indicated by consecutive dry days impact the surface water availability during the summer and autumn seasons. Increases in the number of hot days and nights will affect human health by increasing the thermal stresses. The results of the study can be valuable for formulating the sector-specific mitigation of climate change; however, more studies are required for the analysis of climate extremes using multi-model ensembles to further refine the quantification of the changes and uncertainty associated with GCM outputs. Furthermore, increasing the density of meteorological observatories would further help in understanding the microclimatic variation and climate change signals more lucidly.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.