Abstract
The identification of projected changes in temperature caused by global warming at a fine-scale spatial resolution is of great importance for the high-altitude glacier and snow covered Upper Indus Basin. This study used a multimodel ensemble mean bias-correction technique which uses the ensemble empirical mode decomposition method to correct the bias of ensemble mean of seven CMIP6 GCMs outputs with reference to the European Centre for Medium-Range Weather Forecasts Reanalysis 5 (ERA5). The bias-corrected data have a nonlinear trend of seven GCMs but interannual variance and mean climate of ERA5 dataset. The dataset spans from 1985 to 2100 for historical and future climate scenarios (SSP126, SSP245, SSP370, and SSP585) at daily time intervals with a 1 km grid resolution. The result of different scenarios indicates that the increase in maximum (Tmax) and minimum temperature (Tmin) ranging from 1.5 to 5.4 °C and 1.8 to 6.8 °C from 2015 to 2100, respectively. Similarly, elevation-dependent warming is identified in Tmin from 1.7 to 7.0 °C at elevations <2,000 to 6,000 m asl, while the contrary relationship in Tmax is projected under different scenarios from 2015 to 2100. This study provides an insight into how to improve the GCMs projections and can be helpful for further climate change impact studies.
HIGHLIGHTS
Bias correction of temperature maximum and minimum (Tmax and Tmin) of the multimodel ensemble mean of seven CMIP6 GCMs is carried out.
A reduction in the diurnal temperature range (DTR) is anticipated in the future due to high warming in Tmin as compared to Tmax.
Elevation-dependent warming (EDW) is only pronounced in Tmin.
The duration of snow and glacier melt can expand by 1–2 months due to rise in temperature.
INTRODUCTION
Temperature and precipitation changes caused by global climate warming and their impact on water resources and agriculture have received considerable attention from the research community (Mukhopadhyay & Khan 2016; Shrestha et al. 2017; Ghaderpour et al. 2023; Shawky et al. 2023). Climate variability is greatly influenced by changes in precipitation and temperature that have a significant impact on global climate change and energy cycles (Guan et al. 2015, 2017; Su et al. 2016; Giorgi et al. 2019). As a result of climate change, it is projected that temperatures will increase by roughly 1 °C over preindustrial levels. Further, temperatures are expected to continue increasing by 2–5 °C by the end of the 21st century, depending on different emission scenarios (IPCC 2013, 2018; Pomee & Hertig 2022). It is worth noting that the effect of global warming will not be consistent across all geographical regions, and different areas will be impacted in varying ways. Generally, high latitudes and mountain regions of the Northern Hemisphere are experiencing more warming than other parts of the globe (Pepin et al. 2015; Karmalkar & Bradley 2017). In particular, the Hindukush–Karakorum–Himalayan mountain ranges, which contain a significant number of glaciers that are replenished by precipitation, are expected to be adversely affected by global warming (Azam et al. 2016; Kumar et al. 2019).
The Upper Indus Basin (UIB) receives water from the Hindukush, Karakoram, Himalayan (HKH), and Tibetan Plateau (TP) mountain ranges, which are known as the Asian water towers with the highest number of glaciers (Dahri et al. 2016; You et al. 2017). UIB water is very periodic as it relies on runoff from snow and glacial melt during spring (March–May) as well as summer monsoon (June–September) rainfall and provides water to billions of people (Immerzeel et al. 2013; Lutz et al. 2014, 2016; Orr et al. 2022). The UIB's water resources, agriculture, and livelihoods are highly vulnerable to climate change-induced risks (Abbasi et al. 2017; Hussain et al. 2021b; Tuladhar et al. 2023). There is evidence that the effects of climate change are likely to be intensified by growing water demand in agriculture, and industrial sectors as well as the occurrence of frequent seasonal floods and droughts in the UIB (Jamal et al. 2018; Virk et al. 2020; Rizwan et al. 2022).
Temperature is widely recognized as the most influential and sensitive variable that affects the quantity of streamflow resulting from snow and glacier melt, particularly in high-altitude regions (IPCC 2014). According to the Himalayan mountain climate change study, the temperature in the region has increased by 0.7 °C over the past century. This is notably higher than the global average increase (Team 2007; IPCC 2014; Latif et al. 2020). It is projected that global warming of 1.5 °C will experience a temperature increase of 2.1 °C within the high mountain regions of Asia. As the temperature rises at high latitudes, greenhouse gas emissions increase as well, which is a common phenomenon (Kraaijenbrink et al. 2017).
To investigate past, present, and future trends of climate change, many general circulation models (GCMs) have been utilized since the 1980s. Recently, scholars have employed the Coupled Model Intercomparison Project Phase 5 & 6 (CMIP5 and CMIP6) GCMs to evaluate the climate change impact on temperature and precipitation in several studies (Richter & Tokinaga 2020; Lun et al. 2021; Ngoma et al. 2021; Xu et al. 2021; Ghazi & Jeihouni 2022; Jose & Dwarakish 2022; Noël et al. 2022). Almazroui et al. (2020) carried out a projected temperature study over South Asian countries. The study found that the mean temperature would increase by 1.2, 2.1, and 4.3 °C under the SSP126, SSP245, and SSP585 scenarios, respectively, by the end of the century. Zamani et al. (2020) have compared the CMIP5 and CMIP6 in the precipitation projection in Northeastern Iran. The results showed that CMIP6 GCMs' ensemble showed better performance in most stations and seasons than CMIP5. Ghazi & Jeihouni (2022) carried out a projected temperature study using CMIP5 and CMIP6 GCMs and found that the temperature in all months would increase but variation would be dramatic under CMIP6. Nazeer et al. (2022) utilized two CMIP6 GCMs, EC-Earth3, and MPI-ESM to assess the future water availability in the sub-basins of the UIB. The study concluded a significant increase in temperature (1.1–8.6 °C). Shafeeque & Luo (2021) employed an envelope-based approach and past performance analysis to select CMIP6 GCMs for the UIB. Their future projections under different SSP scenarios (SSP126, SSP245, and SSP585) suggested temperature increases of 1.7 ± 0.6 °C, 3.3 ± 1.3 °C, and 6.3 ± 2.6 °C, respectively, for the period 2071–2100 compared to 1985–2014. Ougahi et al. (2022) focused on downscaled CMIP5 GCMs in the sub-basins of the UIB and examined two emission scenarios (RCP4.5 and RCP8.5) for the periods 2020–2050 and 2070–2100. Their findings revealed a warming of over 6 °C relative to the historical period from 1974 to 2004. Baig et al. (2021) carried out a study using a fine-resolution MRI-AGCM to quantify the impacts of climate change on streamflow in the UIB. Their results indicated that temperatures would increase in all seasons, with the highest temperature increase of 8 °C projected from September to October. Additionally, the annual temperature was projected to increase by 5 °C. Glaciers were found to contribute two-thirds of the annual river flows in the UIB, while rainfall and snow contributed 19 and 11%, respectively.
Moreover, temperature and precipitation changes are extremely influential on crop production. Jabal et al. (2022) conducted a correlation analysis of crop production with temperature and precipitation and found that 59–63% of winter crops (wheat and barley) production is associated with precipitation and about 20–40% of summer crops are associated with temperature (rice, maize, and sunflower). Modi et al. (2021) have found the reverse effects of climate change on crop production until the year 2090. The study showed that the increase in temperature would significantly decrease (p < 0.05) the corn crop production (expected to fall by 24%) due to the decreased bioavailability of water in plants.
Further, some studies have demonstrated that high-altitude mountains are more susceptible to the impacts of climate change than low-altitude mountains due to the presence of snow and ice in these regions (Miller et al. 2012; Pepin et al. 2015). This phenomenon is called ‘elevation-dependent warming’ (EDW), which is accepted globally (Miller et al. 2012; Pepin et al. 2015). The relevance of this theory in the context of the UIB (the altitude varies from 361 to 8,569 m asl) has been argued in some recent studies (for instance, Latif et al. 2020; Yaseen et al. 2020; Pomee & Hertig 2022). The significance of temperature as a fundamental component in the formation of global climate cannot be overstated. Even small temperature change can alter the weather patterns across different regions of the globe. Therefore, climatology studies place significant emphasis on investigating long-term temperature trends across various locations and periods (Zahraei et al. 2020). The diurnal temperature range (DTR) is also an important climate change indicator that has attracted increasing researcher's attention in recent years. In meteorology, DTR is defined as the difference between the maximum and minimum temperatures over the course of a day and is an important parameter of urbanization and climate change. It reveals whether the weather conditions are stable or not for the day. The change in DTR can have negative effects on human health and comfort as well as cardiovascular, nervous, and immune systems. These effects can result in a reduced ability to cope with the significant fluctuations that occur over a single day (Braganza et al. 2004; Kan et al. 2007; Cao et al. 2009; Tam et al. 2009; Lim et al. 2012a, 2012b; Yang et al. 2013; Vutcovici et al. 2014).
Future climate projections at a fine spatial resolution are of great importance in climate-related studies, such as climate extremes, water resources, agriculture, biodiversity, air quality, and wind power (Hoogenboom et al. 2004; Adam et al. 2011; Xu et al. 2021). Climate projections, especially in high-altitude mountainous regions, are challenging due to large spatial variations in natural landscapes, where orography, climate (temperature and precipitation), water resources, crops, and soil management vary across small distances (Hijmans et al. 2005). To overcome this gap, scientists have developed various downscaling techniques to study the effect of climate change at a local scale (Challinor et al. 2005; Baigorria et al. 2008; Wilby et al. 2009; Adam et al. 2011). Therefore, the projection of climate change and its impacts at a local scale rather than at a global scale is necessary to downscale and bias-correct the GCM data (Zhang 2005; Jarvis et al. 2008; Seguí et al. 2010; Tabor & Williams 2010; Hawkins et al. 2013; Jones & Thornton 2013; Musau et al. 2013).
Over the past few years, bias correction has gained wide acceptance in the research community, resulting in the adoption of various techniques for GCMs bias correction. Several studies have applied different bias-correction techniques, including mean bias correction, mean and variance, linear scaling (LS), delta change (DC) method, empirical quantile mapping (EQM), adjusted quantile mapping (AQM), quantile delta mapping (QDM), trend-preserving quantile, and nested and multimodel ensemble (MME) bias correction (Holland et al. 2010; Colette et al. 2012; Xu & Yang 2012, 2015; Bruyère et al. 2014; Done et al. 2015; Hoffmann et al. 2016; Rocheta et al. 2017; Lange 2019; Dai et al. 2020; Tan et al. 2020). The use of these bias-correction techniques significantly enhances the quality of downscaling outputs; for example, the correction of mean bias over the North Atlantic Ocean improved the dynamical downscaling outputs of tropical cyclones (Bruyère et al. 2014; Done et al. 2015). A variance bias-correction technique is used to enhance the extreme events and interannual variability of the GCM outputs (Xu & Yang 2012). A statistical bias correction is a scale-based method that uses multiplicative or additive scaling factors (e.g., delta correction and LS) to adjust the GCM simulations by applying delta, change factors, or nudging factors calculated from historical observations of GCMs to the model output (Seguí et al. 2010; Jakob Themeßl et al. 2011; Hawkins et al. 2013; Gebrechorkos et al. 2019). EQM and AQM are distribution-based methods that empirically transform some features of the probability distribution function (Tan et al. 2020). MME mean-based bias correction would probably produce more reliable projections of the future climate of the globe (Dai et al. 2020).
However, the raw GCM projections are not accurate representations of the actual climate, and bias correction is therefore required at a fine-scale resolution to accurately identify those aspects of GCM projections that are considered significant for impact assessment at a local scale (Baron et al. 2005; Ines et al. 2011; Hawkins et al. 2013). Despite this, some previous bias-correction methods have limitations as well; for example, quantile‒quantile bias correction cannot retain the intervariable dependencies, introduce biases in the spatial gradient of variables, and may disrupt the temporal sequence (Colette et al. 2012; White & Toumi 2013; Xu et al. 2019; Tan et al. 2020). In some cases, variance bias correction and dynamic downscaling may modify the trend of GCM outputs (Xu & Yang 2012; Hoffmann et al. 2016). Since GCM projections without proper bias correction are highly uncertain, it is difficult to utilize these projected climate data, especially in the fields of water resources and agriculture (IPCC 2013; Latif 2011; Xu et al. 2021).
To the best of the authors' knowledge, previously, no such study was conducted to quantify the projected temperature change and their impacts at altitude zones of UIB on fine-scale spatiotemporal resolution using MME of CMIP6 GCMs. This study is conducted to fill the gap and limitations present in the previous studies and methods to bias-correct the temperature data that have large impacts especially on glacier and snow cover UIB. The study used the MME mean bias-correction technique to bias-correct the ensemble mean of seven CMIP6 GCMs data with reference to the European Centre for Medium-Range Weather Forecasts Reanalysis 5 (ERA5) dataset. This technique utilizes the nonlinear trend of ensemble mean of seven GCMs to provide long-term reliable future temperature trends. The ensemble empirical mode decomposition (EEMD) method is therefore applied to calculate the nonlinear trend of ERA5 and GCMs data. Moreover, the biases in the GCM's climatological mean and interannual variance are also corrected. The 30-year period is selected as a reference for historical (1985–2014), and three future periods, 2030s (2015–2044), 2060s (2045–2074), and 2090s (2075–2100), are chosen to bias-correct the maximum and minimum temperatures on daily time steps at a 1 km grid resolution for four future climate scenarios (SSP 126, SSP245, SSP370, and SSP585) over the UIB.
MATERIALS AND METHODS
Study area
The HKH has diverse meteoclimatic regimes due to the interaction of local and large-scale atmospheric circulation systems. Additionally, the altitudinal topography highly affects the climatic variables. For example, the northern valley plains of the UIB are arid, where annual precipitation varies from 100 to 200 mm, which increases at an altitude of 4,400–600 mm (Hussain et al. 2021a). Some previous studies, specifically glaciology studies, note that annual accumulation varies from 1,500 to 2,000 mm at and above an altitude of 5,500 m asl (Wake 1989; Khan & Koch 2018). The average annual snow cover area varies from 10 to 70% during the snowmelt (June–September) and snow accumulation seasons (December–February) (Khan et al. 2020).
Elevation zone . | Elevation range (m asl) . | Zone area . | Glaciers area . | ||
---|---|---|---|---|---|
(km2) . | (%) . | (km2) . | (%) . | ||
Zone-1 | <2,000 | 3,221 | 1.60 | 0.0 | 0.0 |
Zone-2 | 2,000 − 3,000 | 9,183 | 4.56 | 39.1 | 0.2 |
Zone-3 | 3,000 − 4,000 | 25,894 | 12.85 | 813.6 | 4.2 |
Zone-4 | 4,000 − 5,000 | 85,953 | 42.67 | 5,164.6 | 26.5 |
Zone-5 | 5,000 − 6,000 | 73,675 | 36.57 | 11,259.3 | 57.8 |
Zone-6 | >6,000 | 3,521 | 1.75 | 2,212.3 | 11.4 |
Total | 361 − 8,569 | 201,448 | 100 | 19,488.8 | 100 |
Elevation zone . | Elevation range (m asl) . | Zone area . | Glaciers area . | ||
---|---|---|---|---|---|
(km2) . | (%) . | (km2) . | (%) . | ||
Zone-1 | <2,000 | 3,221 | 1.60 | 0.0 | 0.0 |
Zone-2 | 2,000 − 3,000 | 9,183 | 4.56 | 39.1 | 0.2 |
Zone-3 | 3,000 − 4,000 | 25,894 | 12.85 | 813.6 | 4.2 |
Zone-4 | 4,000 − 5,000 | 85,953 | 42.67 | 5,164.6 | 26.5 |
Zone-5 | 5,000 − 6,000 | 73,675 | 36.57 | 11,259.3 | 57.8 |
Zone-6 | >6,000 | 3,521 | 1.75 | 2,212.3 | 11.4 |
Total | 361 − 8,569 | 201,448 | 100 | 19,488.8 | 100 |
Data description
The study used seven CMIP6, GCMs historical (1985–2014), and four future (2015–2100) climate scenarios, named SSP126, SSP245, SSP370, and SSP585 daily temperature maximum (Tmax) and temperature minimum (Tmin). CMIP6 climate scenarios are a ‘high priority’ for IPCC assessment report six (AR6). For example, SSP126 is a 2 °C sustainable scenario with a 2.6 W/m2 radiative forcing level by 2100. Second, SSP245 is a middle-of-the-road scenario that belongs to a socioeconomic family with a radiative forcing level of 4.5 W/m2 by 2100. Third, SSP370 is a medium-high scenario that belongs to a socioeconomic family with a radiative forcing level of 4.5 W/m2 by 2100. SSP585 is a high reference scenario with high fossil fuel development with a radiative forcing level of 8.5 W/m2 throughout the 21st century (Meinshausen et al. 2020).
The climate model's name, institution, and grid resolutions are given in Table 2. The data were collected and downloaded from https://esgf-node.llnl.gov/search/cmip6/. Additionally, we used the European Centre for Medium-Range Weather Forecasts ERA5 reanalysis hourly dataset which is produced using 4D-var data assimilation (Hersbach et al. 2020), from the 1985 to 2014 historical period. The ERA5 dataset has a 30 km horizontal grid resolution. The dataset was downloaded from https://www.ecmwf.int/en/forecasts/datasets/reanalysis-datasets/era5.
No. . | Model . | Institution . | Country . | Resolution . |
---|---|---|---|---|
1 | BCC-CSM2-MR | Beijing Climate Center | China | 1.125° × 1.125° |
2 | CanESM5 | Canadian Centre for Climate Modeling and Analysis | Canada | 2.81° × 2.81° |
3 | INM-CM4-8 | Institute for Numerical Mathematics, Russian Academy of Science | Russia | 2.0° × 1.5° |
4 | INM-CM5-0 | Institute for Numerical Mathematics, Russian Academy of Science | Russia | 2.0° × 1.5° |
5 | IPSL-CM6A-LR | Institute Pierre Simon Laplace | France | 2.5° × 1.26° |
6 | MPI-ESM1-2-HR | Max Planck Institute for Meteorology | Germany | 0.94° × 0.94° |
7 | MRI-ESM2-0 | Meteorological Research Institute, Japan Meteorological Agency | Japan | 1.125° × 1.125° |
No. . | Model . | Institution . | Country . | Resolution . |
---|---|---|---|---|
1 | BCC-CSM2-MR | Beijing Climate Center | China | 1.125° × 1.125° |
2 | CanESM5 | Canadian Centre for Climate Modeling and Analysis | Canada | 2.81° × 2.81° |
3 | INM-CM4-8 | Institute for Numerical Mathematics, Russian Academy of Science | Russia | 2.0° × 1.5° |
4 | INM-CM5-0 | Institute for Numerical Mathematics, Russian Academy of Science | Russia | 2.0° × 1.5° |
5 | IPSL-CM6A-LR | Institute Pierre Simon Laplace | France | 2.5° × 1.26° |
6 | MPI-ESM1-2-HR | Max Planck Institute for Meteorology | Germany | 0.94° × 0.94° |
7 | MRI-ESM2-0 | Meteorological Research Institute, Japan Meteorological Agency | Japan | 1.125° × 1.125° |
Climate change scenarios are mainly published by the Intergovernmental Panel of Climate Change (IPCC) and include SRES (Special Report on Emission Scenario) scenarios (IPCC Third Report), RCP (Representative Concentration Pathways) scenarios (IPCC 5th Report), and SSP (Shared Socioeconomic Pathways) scenarios. Currently, widely used RCP scenarios are based on how future greenhouse gas concentrations will change. In contrast, SSP scenarios predict how climate change will change in response to socioeconomic indicators such as population, economy, land use, and energy change (Lee et al. 2017).
Methodology
This study used the EEMD method to calculate the nonlinear trend of ERA5 and GCM data (Wu & Huang 2009). This method excludes the long-term nonlinear trend in the perturbation term to avoid modification of the long-term trend during variance bias correction. Some previous studies (Holland et al. 2010; Xu & Yang 2012; Hoffmann et al. 2016) decompose the GCM into a linear trend or climatological mean plus perturbation term.
The bias-corrected daily GCM data over the future period, therefore, have a base climate provided by the ERA dataset over the historical period, with the change in future climate relative to the historical climatology generated by the MME and the future bias-corrected weather and climate variability derived from a single GCM. Equation (6) is the final equation that corrects the GCM mean and variance biases after replacing the single GCM nonlinear trend with the MME nonlinear trend (hereafter MVT bias-correction method).
The calculation of nonlinear trends using the EEMD method is computationally costly if we used it to process the daily datasets at 1 km resolution over multiple decades. It also helps to average out the uncertainties present in daily data within a month. Therefore, we considered the long-term nonlinear trends for the daily value of a variable (Tmax and Tmin) to remain the same within the same month. The detailed methodology is described by Xu et al. (2021).
Further, the code used in this study to bias-correct the GCM data is publicly available. The code consists of NCL (NCAR Command Language) and CDO (Climate Data Operators) scripts. NCL is a data analysis and visualization language, developed by the National Center for Atmospheric Research (NCAR). CDO is a command-line tool for manipulating and analyzing climate and weather data. It was developed by the Max Planck Institute for Meteorology in Hamburg, Germany, and is now maintained and distributed by the Climate Data Center at the Deutscher Wetterdienst (German Meteorological Service). NCL script is used to calculate nonlinear trends using the EEMD method, while CDO script re-grid and bias-correct the GCM data. The step-by-step usage of code is also provided in the readme.txt file. The code is simple to use and can be downloaded from https://doi.org/10.11922/sciencedb.00487.
Performance parameter
The performance of bias-corrected GCM data and seven CMIP6 models were computed by using statistical parameters such as the correlation coefficient, mean error, the centered root-mean-square difference (cRMSD), and standard deviation. To evaluate the overall performance of the bias-corrected GCM dataset, a multivariable integrated skill score (MISS) was utilized. The MISS summarizes the power of how well the model simulates multiple variables. A score of 1 indicates that the bias-corrected GCM dataset for multiple variables is exactly the observed values.
These statistical parameters are calculated by the multivariable integrated evaluation method developed by Xu et al. (2017) and modified by Zhang et al. (2021) using the MVIETool v1.0 tool. This tool is written in NCL, which can be easily used in Windows, Linux, and Mac operating systems. The detailed application of these statistical parameters is described in previous studies (Xu et al. 2016, 2017, 2021; Xu & Han 2020; Zhang et al. 2021).
RESULTS AND DISCUSSION
Statistical analysis
Different types of statistics are separated from each other by a thick black line. To facilitate the comparison of the metrics from different variables, in the centered mode, the SD (cRMSL), cRMSD (cRMSVD), and ME (VME) of the models are normalized by dividing by the corresponding SD (cRMSL) of the reference. In the uncentered mode, rms (RMSL) and RMSD (RMSVD) are normalized using rms (RMSL). The metrics table of the centered statistics decomposes the original field into mean and anomaly fields for evaluation. The anomaly fields are further evaluated from the perspective of pattern similarity, variance, and overall difference between the model and observation. The metrics table can clearly explain how much of the overall error comes from the mean error (ME, VME), the amplitude error of the anomaly field (SD, cRMSL), or the error in pattern similarity of the anomaly field (CORR, cVSC).
For example, the ME and cRMSD of MPI-ESM1-2-HR in simulating Tmin are 0.285 and 0.661, respectively, indicating that the overall error is caused mainly by the error in the anomaly field (Figure 5). The cRMSD can be further attributed to the amplitude (0.795) and pattern similarity (0.752). Similarly, one can also decompose model errors into mean error (VME) and overall error of the anomaly field (cRMSVD) in terms of the simulation of multiple variables.
In order to ensure a good comparison, all metrics were normalized by dividing them with the observed standard deviation (ERA5 data) of the corresponding variable. This helps to standardize the values and account for any differences in the scales or units of measurement between the variables being compared.
Five out of seven CMIP6 models showed a relatively poor performance in modeling the spatial pattern of Tmax and Tmin, which is associated with the value of the correlation coefficient (value less than 0.5). The spatial pattern was relatively far better for MPI-ESM1-2-HR (MRI-ESM2-0), as it has a correlation value of 0.698 (0.652) for Tmax and 0.790 (0.594) for Tmin. The results show that 4 (3) out of 7 CMIP6 climate models overestimate (underestimate) Tmax values, as indicated by positive (negative) mean errors. In contrast, 4 (3) out of 7 CMIP6 climate models underestimate (overestimate) Tmin values as directed by negative (positive) mean errors. To summarize and rank the overall performance of a model in simulating multiple fields, the MISSs in both centered and uncentered models are provided in the metrics table and are expected to provide a more accurate evaluation compared with MIEI. Figure 5 shows that MPI-ESM1-2-HR ranks first out of seven models when referring to the values of the centered MIEI (cMIEI), while it also ranks first based on the centered MISS (cMISS) values. It also ranks second based on the uncentered MISS (uMISS) values.
The uncentered statistical parameters are used to evaluate the model performance in terms of the original field. These metrics are calculated by comparing the model output directly to the observed values, without any adjustments or normalization. In contrast, the centered statistics are used to evaluate model performance in simulating anomaly fields, which are deviations from the long-term average or climatology of a particular variable (Zhang et al. 2021). Overall, the bias-corrected GCM (MPI-ESM1-2-HR_bc) performance in simulating individual or multiple variables is extraordinary based on the ERA5 dataset for the historical period (1985–2014). The bias-corrected GCM ranked first out of all climate models.
Projected temperature over the UIB
The GCM data overestimate the temperature maximum relative to ERA5 data by approximately 2.8 °C during the historical period 1985–2014 (Figure 7(a)). The bias-correction method comprehensively removed the mean bias in the GCM data. The raw GCM data underestimate the amplitude of the interannual variability when compared to ERA5 data. The amplitude is described by the standard deviation of GCM (1.91) and ERA5 (2.14) data. The amplitude is relatively enhanced in the bias-corrected data (SD = 2.12), which is closer to the ERA data. Moreover, both GCM bias-corrected and original GCM datasets have different long-term trends because the bias-corrected dataset has a nonlinear trend of the ensemble mean of the seven GCMs. From Figure 7(a), it can be seen that the bias-corrected data have an interannual variation of the MPI-ESM1-2-HR data. The cumulative frequency distribution provides information about the probability of occurrence of a specific event. We considered different temperature bins (ranges) for the probability distribution of bias-corrected, GCM, and ER5 data. For example, at a temperature range from 10.5 to 11 °C, the value of cumulative frequency is approximately 0.3 for ERA5 and GCM_bc, indicating that there is a 30% chance of observing the same temperature range (10.5–11 °C) or below for the historical period (1985–2014). At this temperature range, the probability for GCM data was just 0.1. Similarly, at probability 0.3, the value of temperature for GCM is approximately 12.5–13 °C. Overall, the Tmax shows that the GCM considerably overvalues high-temperature events relative to the ERA5 data (Figure 7(b)). In contrast, the bias-corrected data show a similar cumulative frequency distribution relative to the ERA5 dataset. This indicates that the bias-correction method significantly improves the characteristics of GCM data for the historical and future periods (Figure 7(b) and 7(c)).
The impact of atmospheric oscillation on temperature is beyond the scope of our study. This study is limited to the bias correction of the temperature data using novel techniques and did not discuss any atmospheric oscillation to get valuable insights from our findings. Some previous studies reported that atmospheric oscillation is an influential factor behind temperature changes in south Asian countries (Iqbal et al. 2016; Mallick et al. 2022; Shawky et al. 2023). Iqbal et al. (2016) conducted a correlation analysis study in Pakistan and found a relationship of maximum and minimum temperatures with oscillation indices. The highest correlations were noticed in the pre-monsoon (January to March) with North Atlantic Oscillation (NAO) and May with ENSO (El Nino-Southern Oscillation), and the late pre-monsoon month of May with the North Sea Caspian (NCP) index.
In a previous study, Nazeer et al. (2022) used CMIP6 GCMs in the UIB and found a significant increase in mean temperature ranging from 1.1 to 8.6 °C. The positive change in projected mean temperature is similar to our study but the magnitude of temperature increase is different from our temperature ranges of 1.65–6.1 °C. The different range of projected increases in mean temperature may be associated with the study which utilized only two GCMs. Another study by Ougahi et al. (2022) used CMIP5 GCMs in the UIB and their findings revealed a warming of over 6 °C in mean temperature under a high emission scenario (RCP8.5). The results are similar to our study's mean temperature increase of about 6.1 °C.
The DTR exhibits a declining pattern because the increase in minimum temperature is at a higher rate compared to the temperature maximum (Figure 9). The DTR is narrowing or damping by 0.2–0.4 and 0.3–1.4 °C for SSP126 and SSP585, respectively, from 2030s to 2090s. Overall, SSP585 gives the highest warming signal at both temperatures, while in SSP126, the warming signals are at a low rate compared to all other scenarios. The reason is associated with the development of the scenario as the radiative forcing level of 2.6 W/m2 by 2100.
In addition to the annual temperature increases, Table 3 also includes information on the seasonal Tmax and Tmin and DTR (which is the difference between the maximum and minimum temperatures in a day). The seasonal analysis shows that warming will occur at both temperatures in all seasons for all climate scenarios. The highest rate of warming is found in the autumn season (September to November) in all scenarios except SSP126, which also shows the highest warming rate in the spring season (March to May) by the 2090s. The finding demonstrates that the temperature increment is more prominent in the later decades of the 2090s, except for SSP126, which exhibits a substantial temperature rise during the summer and autumn seasons in the 2060s. For the autumn season in the 2090s, the temperature increase for Tmin (7.6 °C) is higher than that for Tmax (6.0 °C) in SSP585. Among the different scenarios, SSP126, SSP245, and SSP370 are projected to experience a minimum temperature rise in Tmin by 1.5, 3.3, and 5.0 °C, respectively, in the winter season, while the minimum temperature rise for SSP585 is expected in the summer season in the 2090s. Likewise, the minimum temperature rise in Tmax for SSP245, SSP370, and SSP585 is projected to be 2.5, 3.7, and 4.6 °C, respectively, in the spring season by the 2090s. Overall, our results contradict an earlier study (Jones et al. 1999; Pomee & Hertig 2022), which found maximum warming in the winter season in Tmin, but our results showed maximum warming in the spring (autumn) season under SSP126 (SSP585) by the 21st century. Our findings are in line with some previous downscaling studies (Taylor et al. 2012; Ali et al. 2015) that projected positive Tmin changes. Another study conducted by Baig et al. (2021) in the sub-basin of UIB found that the highest mean temperature increase (8.1 °C) is projected from September to October which is similar to our findings where the maximum increase (6.8 °C) is from September to November (autumn season).
The DTR and temperature changes have an inverse relationship that could have potential implications for agriculture, which is significantly reliant on predictable temperature patterns and any fluctuations in the DTR could potentially impact crop yields. The DTR exhibits declining trends, indicating a reduction in the amplitude or variability of the DTR. During the 2090s, the spring season (March to May) is expected to experience a higher rate of DTR reduction of 2.2 °C under the SSP585 scenario, whereas no changes are anticipated in DTR during the winter season for SSP126. These findings highlight the anticipated rise in temperature and the decrease in DTR in the forthcoming decades under various climate scenarios. Such insights could aid in policymaking and initiatives to counteract the consequences of climate change.
Elevation zone-wise temperature projection
The altitudinal zone-wise projected changes in temperatures for the SSP126, SSP245, SSP370, and SSP585 climate scenarios for 2030s, 2060s, and 2090s are presented in Table 4. EDW is detected in annual and seasonal Tmin. The annual analysis showed warming at a rate of 1.7 (6.3) at elevations <2,000 m asl (Zone-1) to 1.9 (7.0) °C at elevations 5,000–6,000 m asl (Zone-5) for SSP126 (SSP585) from 2030s to 2090s, respectively. Our results are in line with a previous study (Qin et al. 2009) that noted that the warming rate increases with an elevation below 5,000 m asl, and such an increasing trend disappears for elevations higher than 5,000 m asl. The inverse relationship is noted in Tmax, which increases as altitude decreases to 2,000 m asl at a rate of 1.5 (5.4) to 1.6 (5.6) °C for SSP126 (SSP585) from 2030 to 2090s. Our finding contradicts previous findings (Pomee & Hertig 2022), which suggested an EDW for Tmax. Here, EDW is found in Tmin, but both studies found a significant reduction in DTR over the UIB. Some previous studies explained the Tmin warming within the UIB. The effect of increased cloud radiative forcing (Forsythe et al. 2015), surface albedo (Pepin et al. 2015), and soil moisture responses can explain substantial Tmin increases in the UIB. The smaller increase in Tmax may be associated with winter precipitation as snow, which will increase albedo from fresh snow. Similarly, enriched convection that starts during March may also reduce insolation over the UIB (Cannon et al. 2016). During cloudy conditions, increased soil moisture, due to increased precipitation and melting in the UIB, may also exert a positive response to increasing Tmin (Pomee & Hertig 2022).
The EDW variation may also be imitated in lapse rates. The increase in temperature at higher rates with respect to elevation may lead to a decrease in lapse rates. The results also show that in all elevation zones, the DTR decreases from a very minimal value to a higher value. The highest decrease in DTR is detected in Zone-5 for SSP126 (SSP585) at 0.4 (1.6) °C during the 2090s. EDW is also analyzed on a seasonal basis (Table 4). The results showed that there was no or a minimal increasing or decreasing trend with elevation in the winter season for all projected scenarios. The highest Tmin increase of 2.4 (7.9) °C is detected in the summer (autumn) season during the 2060s (2090s) for the SSP126 (SSP585) scenario at elevation Zone-5. Similarly, the highest Tmax increase of 2 (6.1) °C is detected in the autumn season in all zones (Zone-1, Zone-5, and Zone-6) during the 2060s (2090s) for the SSP126 (SSP585) scenario.
The study investigated the seasonal variation in temperature and its correlation with elevation. The results, presented in Table 4, indicate that there is no significant trend in the increase or decrease in temperatures with elevation during the winter season for all projected scenarios. However, during the summer and autumn seasons, the study found that the highest increase in Tmin is detected at an elevation of Zone-5, with an increase of 2.4 (7.9) °C for the SSP126 (SSP585) scenario during the 2060s (2090s). Additionally, the highest increase in maximum temperature (Tmax) was detected during the autumn season for all zones (Zone-1, Zone-5, and Zone-6), with an increase of 2 (6.1) °C for the SSP126 (SSP585) scenario during the 2060s (2090s). These findings suggest a seasonal variation in temperatures with elevation, indicating the importance of considering seasonal changes when studying the effects of climate change on EDW.
Projected climate impact on glaciers melt
Furthermore, only 4.4% of the glacier area is located below an elevation of 4,000 m asl, while 84.3% of the glacier area is found between elevations of 4,000 to 6,000 m asl (which corresponds to Zone-4 and Zone-5), as given in Table 1. This result reflects the importance of this elevation range for understanding the glacier response.
Furthermore, the analysis found that during the 2030s, the phase change is noted in Zone-2 and Zone-5 by one (October) and two (July and September) months, respectively, for all climate scenarios. Notably, Zone-5 contains approximately 57.8% of the glacier area and will therefore generate more melt water regardless of other zones. During the 2060s, it is anticipated that the phase change in Zone-6 will occur under both the SSP585 and SSP370 scenarios, which cover approximately 11.4% of the glacier area. By the 2090s, it is noted that the melt season will increase by 2 months across all altitude zones, which will result in more short-term streamflow but may also increase evaporation and evapotranspiration over the whole basin. Additionally, over the long term, glacier melt will reduce the availability of streamflow due to depletion in glacier volume caused by climate change. It is important to note that the reliability of our findings is associated with the ERA5 dataset we used to bias-correct the GCM data and the uncertainties present in the GCMs.
CONCLUSIONS
Climate change is currently one of the great challenges. To cope with this challenge, GCMs are available. The GCM datasets have a coarse spatial resolution and the accuracy of these datasets varies from region to region and from model to model. Also, the datasets are unsuitable especially in any mountainous region for direct application to climate change impact assessment studies. Therefore, for climate change impact assessment on water resources and agriculture, fine-resolution future projections are necessarily required for optimal accuracy. Hence, this study is conducted to utilize seven GCM datasets and adapted a novel MME bias-correction technique. The ERA5 dataset was utilized as a reference dataset for bias correction of daily maximum and minimum temperatures at a 1 km grid resolution.
The investigation of climate change impact on the UIB for a long-term historical period (1985–2014) and future scenarios (SSP126, SSP245, SSP370, and SSP585) from 2015 to 2100 is carried out. The outcomes of the climate change assessment are summarized as follows:
The whole UIB will be nonuniformly warm during the 21st century under all climate scenarios. The projected warming is strong (weaker) under the SSP585 (SSP126) scenario during the 2090s.
Analysis showed that maximum warming is dominated by the temperature minimum (Tmin) compared to the temperature maximum (Tmax).
The reduction in the DTR is noticed as a result of the high rate of increase in projected Tmin.
The periodogram showed a consistent and dominant pattern of increase in temperature occurring every year. Some minor periodicities of 2 and 3 years were found that could be associated with longer-term climate patterns or atmospheric oscillations.
EDW is found in Tmin only, but a projected increase is in inverse relation to the elevation in the UIB.
The high rate of warming is pronounced in Tmin (1.8–1.9 for SSP126 and 6.9–7.0 for SSP585) at an elevation range of 4,000–6,000 m asl, which accounts for approximately 84% of the total glacier area of the UIB. This increase in temperature is substantial and will deplete glacier ice rapidly especially at an elevation range of 4,000–6,000 in the future, which will result in high streamflow during the melt season.
The impact of warming on glaciered melt suggests that melting will expand by 1 or 2 months over the basin under all climate scenarios by the 21st century.
Projected warming will significantly increase the water demand downstream. The temperature rise will cause liquid precipitation during the winter season, which will significantly increase the river flows. The temperature increase may drastically influence glacial stability under SSP585, as many regions remain well above the freezing point temperature (0 °C). Our analysis has some limitations, including a relatively small GCM ensemble, and the use of the ERA5 dataset as a referenced dataset for bias correction and interannual variability is dependent on a single GCM (MPI-ESM1-2-HR). These findings have significant implications for the management of water resources and adaptation strategies in the UIB region in the face of climate change. A detailed study is still required to discuss the impact of atmospheric oscillation as well as the impact of land use and land cover on temperature projections.
ACKNOWLEDGEMENTS
This work is financially supported by The Strategic Priority Research Program of the Chinese Academy of Sciences (XDA20100104) and The National Natural Science Foundation of China (41871280). The authors want to acknowledge the organizations for providing ERA5 and GCM (BCC-CSM2-MR, CanESM5, INM-CM4-8, INM-CM5-0, IPSL-CM6A-LR, MPI-ESM1-2-HR, and MRI-ESM2-0) climate data.
AUTHORS CONTRIBUTION
All authors are involved in the intellectual part of this paper. X.L., K.J., and Y.C. designed the research work. K.J. conducted the research and wrote the draft manuscript and P.M. helped in data management, while M.A., M.R., and Z.S. helped in data analysis. X.L., Y.C., and M.A. revised the article and provided many suggestions. All authors have read and approved the final manuscript.
DATA AVAILABILITY STATEMENT
Data cannot be made publicly available; readers should contact the corresponding author for details.
CONFLICT OF INTEREST
The authors declare there is no conflict.