Climate change and hydropower resilience in Nepal: an integrated modeling approach in the Madi River Basin

Climate change represents one of the most pressing global challenges of our time, a fact affirmed by key international bodies such as the United Nations, the Intergovernmental Panel on Climate Change (IPCC), and the Paris Agreement. The United Nations' 2030 Agenda for Sustainable Development specifically emphasizes the urgent need for substantial action against climate change through Sustainable Development Goal 13, which calls for measures to combat climate change and its impacts (UNFCCC 2015). The IPCC verifies that climate change is real and primarily caused by human activities. Failure to significantly reduce greenhouse gas emissions could result in severe and irreversible changes (IPCC 2014).

In response to these concerns, the Paris Agreement, adopted in 2015, establishes a global framework to mitigate climate change by limiting global warming to well below 2 °C above pre-industrial levels, with an ideal target of 1.5 °C. It also encourages nations to establish and pursue their own nationally determined contributions (NDCs) to achieve these goals. As part of its NDC, Nepal has set an ambitious goal to increase its clean energy generation from a baseline of 1,400 to 15,000 MW by 2030. Nepal aims to meet 15% of its total energy demand through clean energy sources by the same year. This strategy demonstrates Nepal's firm commitment to aligning its national objectives with global climate change mitigation efforts (NDC 2020).

Hydropower plays a crucial role as a renewable energy source, boasting a global installed capacity of 1,360 GW in 2021 (IHA 2021). It provides environmental advantages, contributes to grid stability, and offers storage capacity (IPCC 2014; IRENA 2020; Baniya et al. 2023; Lamsal et al. 2023). However, the climate change is likely to affect hydropower, impacting energy security and grid reliability (Madani & Lund 2010; Hamududu & Killingtveit 2012; van Vliet et al. 2013). Hydropower in Nepal is highly vulnerable to climate-induced variations, including alterations in precipitation patterns and glacier melt dynamics, necessitating focused research (Xu et al. 2009; Bajracharya & Shrestha 2011; Hamududu & Killingtveit 2012; Shrestha et al. 2012; IPCC 2018). Despite the increasing integration of climate considerations into hydropower planning, there is still an emphasis on engineering resilience and risk management (Regmi et al. 2016; Dhakal et al. 2017; Poudel & Duex 2017). Within this broader context, the Madi River Basin emerges as a significant case study due to its hydropower potential and its status as a representative glacier-fed system in Nepal. Also, research and development is crucial for the sector's long-term sustainability and climate resilience (Koirala et al. 2016).

Nepal's hydrological landscape is distinctive, shaped by Himalayan snow and glacier melt (Xu et al. 2009; Bajracharya & Shrestha 2011). The Koshi, Narayani, and Karnali river basins, Nepal's three major basins, hold a gross hydropower potential of 72,544 MW, constituting 94% of the country's total hydropower potential (WECS 2019). Presently, Nepal's hydroelectric production surpasses its peak demand, generating 1,964 MW against a demand of 1,748 MW (NEA 2022). However, to meet its NDC targets, a more diversified approach to renewable energy beyond hydropower is essential (WECS 2022).

Climate change is fundamentally altering global water cycles, impacting precipitation and evaporation patterns, as well as influencing streamflow and extreme weather events, especially in glacier-dependent areas (Hegerl et al. 2015; Van Lanen et al. 2016; Greve et al. 2018; IPCC 2019; Barzkar et al. 2022). These shifts have serious implications for water management, affecting water availability, quality, and timing (Haddeland et al. 2014). In Nepal, these effects are exacerbated by its unique Himalayan geography, leading to accelerated glacier retreat and altered hydrological characteristics of river basins (Bajracharya & Shrestha 2011; Huss & Hock 2018). Hydrological shifts in pattern, magnitude, and water availability in Nepal impact multiple sectors like hydropower, irrigation, water supply, and traditional livelihoods (Khadka et al. 2014; Ragettli et al. 2015; Dahal et al. 2016a, 2016b; Dhakal et al. 2019; Ojha et al. 2020). Recent studies reveal dual trends in precipitation, with a decline in extreme annual indices and an increase in season-specific dry days (Chhetri et al. 2021). In the Narayani basin, Bhattarai et al. (2018) show that rising temperatures influence both annual and seasonal streamflows, as well as the snowmelt contribution to river flow. This observation aligns with hydrological findings from the Madi River Basin, a sub-basin of Narayani, where researchers elucidated the marked seasonality of river flows, observing that the mean monthly flow during the summer monsoon period was up to 14 times higher than in the dry season, thereby shedding light on the significant role played by topographical and climatic conditions in shaping low-flow regimes (Khanal & Watanabe 2017). Furthermore, a study by Burgan (2022) illustrates the utility of artificial neural networks (ANNs), such as feed-forward back propagation (FFBP), generalized regression neural networks (GRNN), and radial basis function (RBF) neural networks for enhancing daily streamflow predictions in the Kocasu River, Turkey. This advancement underscores the potential of similar predictive technologies to improve the accuracy of streamflow forecasts in Nepal's complex hydrological settings.

Understanding the vulnerabilities of regions like the Madi River Basin necessitates the utilization of advanced climate and hydrological models. Climate models, particularly those from the climate model intercomparison project (CMIP), serve as foundational tools for understanding hydrological responses under variant climatic conditions (Milly et al. 2008). The most recent phase, CMIP6, enriches these evaluations by accounting for anthropogenic influences, notably greenhouse gas emissions (Eyring et al. 2016). By integrating shared socioeconomic pathways (SSPs), CMIP6 facilitates intricate hydrological analyses that are especially vital for localized contexts, such as Nepal (O'Neill et al. 2016; Pandey et al. 2018; Reidmiller et al. 2018). Specifically, this study integrates two representative SSPs – SSP245 and SSP585 for comprehensive insights into moderate and high-emission scenarios, respectively. Downscaling techniques are employed to enhance the granularity of these models, thereby making them aptly suited for localized studies in areas such as Nepal (Teutschbein & Seibert 2012; Mishra et al. 2020).

Hydrological modeling, essential for managing water resources in mountainous terrains, has benefited from methodological advancements tailored to complex environments like the Himalayas (Bajracharya & Shrestha 2011; Immerzeel et al. 2020). Among various methods, the soil and water assessment tool (SWAT) stands out for its flexibility in representing intricate landscapes and hydrological conditions, including its applicability for climate change impact assessments (Arnold et al. 1998; Gassman et al. 2007; Zhang et al. 2009). Various case studies in Nepalese Himalayan Basins substantiate the capability of capturing the complexities of these regions (Shrestha et al. 2013; Babel et al. 2014; Ragettli et al. 2015; Bhattarai et al. 2016; Dahal et al. 2016a, 2016b; Bista et al. 2021).

Nepal's hydropower infrastructure, predominantly reliant on glacier- and snow-fed rivers, is highly vulnerable to climate change (Xu et al. 2009; Shrestha et al. 2012). Despite hydropower's centrality to Nepal's energy portfolio, studies at the river basin level addressing climate impact are limited, creating a knowledge gap (Dhakal et al. 2019; Shrestha et al. 2019). While studies have begun to investigate the vulnerabilities of Nepal's hydropower sector to climate change, they often focus on larger basins or issues like glacier melt and monsoon patterns (Shrestha et al. 2013; Bhattarai et al. 2016). The urgency of addressing climate impacts on hydropower generation at the sub-basin level remains underemphasized (Shrestha et al. 2019). The Madi River Basin, a tributary of the larger Narayani basin, exemplifies this research gap. Although prior work has identified 72 potential hydropower sites along the Madi's tributaries (Pokharel et al. 2020), comprehensive studies assessing the climate change impacts on hydropower generation within this specific basin are markedly absent.

In response to identified research gaps and the pressing need for focused inquiry, this study is primarily aimed at investigating the effects of climate change on hydrological characteristics and hydropower projects in the Madi River Basin. The study endeavors to achieve three main objectives: (1) assess and analyze the baseline hydroclimatic trends in the region; (2) quantify the impacts of climate change on streamflow patterns; and (3) examine how these climatic changes influence hydroelectric energy production concerning established baseline energy figures. By pursuing these objectives, the study seeks to enhance understanding of the Madi Basin's resilience to evolving climatic conditions, thereby contributing to Nepal's strategies for climate-resilient hydropower development.

The Madi River, also known as Madikhola, originates from the Annapurna II glacier and is supplemented by other features like the Kahphuche Lake. The lake was formed as a result of the melting of ice sheets that pooled in a lower area (Koirala et al. 2017). Uniquely, the lake is formed at a relatively low altitude for glacial lakes in the Nepal Himalayas. By 2020, the lake's area was estimated to be 0.12 ± 0.015 km², with a volume of 1.42 × 10⁶ m³ according to bathymetric data (Kandel et al. 2023). The Madi River has a total of 14 tributaries, the largest is the Midim, and the others include Rudi, Madkyu, Kyaduso, Nyache, Idi, and Madame supporting various capacities of hydropower projects. The remaining tributaries – Paste, Birdi, Khalte, Pisti, Risti, Sange, and Kalesti – are located at the lower elevations of the watershed.

At present, there are 23 hydropower projects either at the operation, generation, or survey license phase in the Madi basin. Taken together, the potential power generation for the Madi River Basin stands at 377.37 MW, a notable output for a basin of its size. To address the impacts of climate change and ensure the representation of diverse geographical areas, projects have been selected to optimally represent the flow conditions of the tributaries. The three selected hydro projects are presented in Table 1.

The historical hydroclimatic data required for this study include daily precipitation, minimum and maximum temperature, and observed streamflow data, which were sourced from the Department of Hydrology and Meteorology (DHM), Government of Nepal. In the vicinity of the Madi River Basin, six climatic stations and one hydrologic station were strategically selected to provide relevant and overall coverage (Figure 2). The statistical measures for annual hydroclimatic variables at hydrometeorological stations (Table 3), encompassing descriptive statistics, measures of variability, shape characteristics, confidence intervals, and distribution properties, are detailed in Supplementary Table S1.

Future climate projection refers to the use of scientific methods and computational models that enable the prediction of future climatic conditions over various time scales, such as decades or centuries (IPCC 2013). Bias in precipitation and temperature data from global climate models (GCM) often arises from the models' coarse resolution or the way they are parameterized (Mishra et al. 2020). To address this, Mishra et al. (2020) utilized 13 CMIP6-GCM models with spatial resolutions ranging from 0.7° to 0.2° under r1i1p1f1 initial conditions for both historical (1850–2014) and future (2015–2100) climate under four SSPs (SSP126, SSP245, SSP370, and SSP585). Using both dynamic and statistical approaches, climate data were produced for six South Asian countries, including Nepal. Thus, downscaled and bias-corrected daily climate data to a spatial resolution of approximately 0.25° (around 28 km at the equator) were used in this study (available at https://doi.org/10.5281/zenodo.3987736).

Furthermore, two scenarios, SSP245 and SSP585, were selected for this study, so even with less computation, we could explore a wide range of potential future climate conditions, i.e., from a moderate pathway where some mitigation efforts are made, representing a ‘balanced case’ scenario (SSP245) to a more extreme pathway with high emissions and little mitigation are made, typically representing ‘worst case’ scenario (SSP585) (Riahi et al. 2017). The GCM models used in this study for performance evaluation are listed in Table 2.

Trend analysis on hydroclimatic variables is an indispensable technique in understanding the dynamic nature of climatic processes over time. In this research, the trend analysis is performed at chosen hydrometeorological stations listed in Table 3, focusing on available time series data at multiple temporal scales, i.e., seasonal and annual. The analysis is performed using two methods: Mann–Kendall test and Sen's slope test. The Mann–Kendall test is a non-parametric statistical method used to identify trends in time series data (Mann 1945; Kendall 1975). Following the trend detection by the Mann–Kendall test, Sen's slope test is employed to quantify the slope of the trend (Sen 1968).

The process of setting up the SWAT model in this study begins with the preprocessing and loading of the DEM into the model (Neitsch et al. 2011). Watershed delineation was then performed, with a drainage threshold area (DTA) set at 500 ha, a critical parameter that determines the stream network within the watershed, defining the size and configuration of the sub-basins (Arnold et al. 2012). Monitoring points were manually added, incorporating a hydrological station as an outlet and carefully selecting the final outlet at Byas Cave, and then the watershed was discretized into 12 sub-basins. Subsequently, five slope classes were defined as follows: 0–5%, 5–15%, 15–30%, 30–50%, and above 50%. The threshold area for land, soil, and slope class was set at 10%, ensuring that only major features were represented in the model (Arnold et al. 2012). The discretization continued with the creation of 130 HRUs, each representing a unique combination of land use, soil type, and slope class within a sub-basin. Daily weather inputs, such as precipitation, solar radiation, relative humidity, wind, and maximum and minimum temperature data, were imported into SWAT through the weather input station, and any missing data were generated using the model's inbuilt stochastic weather generator. For surface runoff estimation, the SCS curve number method was applied, and no point sources were defined in this model setup. Lastly, the Penman–Monteith method was employed for potential evapotranspiration (PET) calculations, favored for its comprehensiveness and ability to account for various atmospheric and surface conditions (Allen et al. 1998).

The sequential uncertainty fitting (SUFI-2) algorithm in SWAT calibration and uncertainty programs (SWAT-CUP) was employed to perform sensitivity analysis, calibration, and validation. Specifically, the model was calibrated for one hydrological station, Sisha Ghat, over 12 years from 2004 to 2015, with the first 4 years serving as a warm-up phase. The model was subsequently calibrated using daily discharge data for the next 5 years, 2008–2012, and validated for the following 3 years, 2013–2015. Global sensitivity analysis was adopted in this study. The parameters were methodically selected through a two-stage process: first, by analyzing the behavior of the initial model through a single simulation and performing parameter regionalization as suggested by Abbaspour et al. (2015), and second, by drawing from previous studies conducted on Gandaki Basin and in other Himalayan watersheds in Nepal (Devkota & Gyawali 2015; Pandey et al. 2020; Pokharel et al. 2020). The goodness of fit was quantified by analyzing the 95% prediction uncertainty (95PPU) band. Key metrics in this assessment were the P-factor and R-factor, with optimal values being a P-factor approaching 1 and an R-factor approaching 0, with criteria set for good fit for flow data at P>0.7 and R<1.2 (Abbaspour et al. 2017). This robust methodology ensures the reliability and precision of the hydrological modeling before the model performance indicator.

Three primary error metrics were used for this evaluation: Nash–Sutcliffe Efficiency (NSE), coefficient of determination (R²), and percentage bias (PBIAS). NSE is a significant metric, representing how well the variance in observed data is represented by the model, with a value of 1 indicating perfect estimation and thus signifying an accurate model (Krause et al. 2005). On the other hand, the coefficient of determination, R², informs the correlation between observed and simulated values. The PBIAS is another vital parameter; a positive value represents an overestimation while a negative value indicates underestimation. The low absolute value of PBIAS further signifies the superior performance of the model. Criteria for judging the efficiency of the model include R² values greater than 0.70, NSE values greater than 0.75, and the PBIAS value less than ±10%, solidifying the model's validity and reliability. According to Moriasi et al. (2007), these parameters are indispensable for understanding the accuracy of hydrological models.

For this study, precipitation and temperature projections were analyzed up until the end of the century (2026–2100), focusing on the potential impacts on streamflow. The baseline historical data for comparison purposes were taken from the years 1995 to 2019. Bias correction is a statistical technique that can be used independently of downscaling and is essential to make the model outputs more accurate and useful for impact assessment (Maraun et al. 2010). Due to the elongated characteristics of the watershed, despite its small area, it contained one gridded station near the DHM station Siklesh (824) and three in the vicinity of Khairenitar (815), Kunchha (807), and Damauli (817). Thus, data retrieved from Mishra et al. (2020) were bias-corrected using the nearest station's climate data. The bias correction was carried out using the linear scaling method, a simple yet effective method. This method perfectly adjusts the monthly mean of raw GCM output to the observed value without excessively altering the underlying distribution of values, ensuring the natural variability in the climate (Maraun et al. 2010; Teutschbein & Seibert 2012). Compared to other complex bias correction techniques, the linear scaling method is both simple and effective (Shrestha et al. 2017).

The performance analysis of GCM and bias-corrected models (BCM) is an imperative step in enhancing the reliability and accuracy of climate projections. It ensures that the corrected models retain the essential physical processes and spatial patterns captured by the raw GCM (Teutschbein & Seibert 2012). In this study, the performance of climate data was evaluated for both raw GCM and BCM against the baseline monthly observed data at four selected DHM stations (824, 807, 815, and 817). Utilizing three key performance metrics – Nash Sutcliffe efficiency (NSE), ratio of the root mean square error to the standard deviation of observed data (RSR), and PBIAS – the climate data were analyzed, and individual performance metrics were computed for all 13 raw GCMs and BCMs. The ranking done based on the average of combined performance ratings as shown in Table 4 (Timilsina et al. 2021) enabled the subsequent selection of the top six BCMs listed in Table 2. This study adopts a similar approach to Baniya (2022), utilizing heat map representations to effectively convey both individual metric values and their collective ratings for the selected stations.

The model performance of the SWAT model was evaluated for calibration and validation purposes considering streamflow observations from 2004 to 2015. Subsequently, this model was employed for the extended period from 1990 to 2019 with the first 5 years as a warmup, to simulate baseline streamflow. This well-calibrated model was then forced with future climate data spanning from 2026 to 2100, to generate continuous streamflow projections. By comparing these projections with the established baseline, the possible climate change implications on the hydrological characteristics are quantified at annual and seasonal scales. Focusing on three selected hydropower projects, the climate change impact on energy generation was calculated. This analysis was performed on annual and monthly scales at three distinct 25-year intervals: Near Future (2026–2050), Mid Future (2051–2075), and Far Future (2076–2100) scenarios, termed as NF, MF, and FF, respectively.

The annual trend of streamflow at Sisha Ghat shows a significant decrease at a rate of −0.4487 cumec/year (p = 0.01574), indicative of a substantial reduction in flow. This decline is most prominent during the monsoon season with a rate of −1.3066 cumec/year (p= 0.0086), while the pre-monsoon trend is negligible. Post-monsoon and winter seasons present slightly positive trends, though they are not statistically significant. The temperature at Khairenitar demonstrates a consistent and significantly increasing trend across all seasons, with the annual rate at 0.0288 °C/year (p = 0.0001). The precipitation trend of the upper watershed reveals a negligible annual trend, suggesting a stable precipitation pattern in this area. The annual trend in lower watershed indicates a significantly decreasing rate of −16.84 mm/year (p = 0.00412), particularly notable in the monsoon season at a rate of −10.49 mm/year (p = 0.021). The precipitation trend of the upper watershed reveals a negligible annual trend, suggesting a stable precipitation pattern in this area. The annual trend in lower watershed indicates a significantly decreasing rate of −16.84 mm/year (p = 0.00412), particularly notable in the monsoon season at a rate of −10.49 mm/year (p = 0.021).

The significant decrease in annual and monsoon streamflow may be due to altered precipitation patterns, land use, increased water consumption, or groundwater recharge. The significant temperature rise could be indicative of broader climatic shifts or localized anthropogenic influences such as urbanization. The contrasting trends in precipitation between upper and lower watersheds possibly reflect spatial weather variations or topographical influences on rain distribution.

The performance evaluation of the SWAT model is crucial for assessing its ability to simulate real field scenarios. Out of 21 chosen parameters, 15 are identified as the most sensitive. Since the trend analysis reveals topographical effects, the calibration of the SWAT model at the Sisha Ghat station is conducted first, with specific attention to elevation parameters such as temperature lapse rate (TLAPS) and precipitation lapse rate (PLAPS). Among these, the PLAPS parameter emerges as the most sensitive, with a positive value that indicates an increase in precipitation with increasing elevation. Previous studies (Pandey et al. 2020; Baniya 2022) in Himalayan catchment have found similar sensitivity of these parameters due to the orographic effect. Then among snow parameters, snowfall temperature (SFTMP) is identified as the most sensitive.

Further sensitivity analysis highlights that surface release and groundwater movement exert substantial influence on the basin. Geomorphological factors, OV_N (Manning's n value for overland flow) and SLSUBBSN (average slope length) are found to be highly sensitive. The groundwater-sensitive parameter GW_DELAY, controlling groundwater behavior, and the saturated hydraulic conductivity SOL_K (), affecting the basin's surface runoff, have a dominant effect as well. The remaining parameters are ranked and tabulated according to their sensitivity from 1 to 15, with 1 being the most sensitive (Table 5).

Evaluating GCMs against observed data is vital in identifying inherent biases and guiding subsequent refinements to enhance model accuracy. The performance evaluation of GCM and BCM is carried out on a monthly scale at four climate stations: Siklesh (824), Kunchha (807), Khairenitar (815), and Damauli (817). Individual statistical parameter performance NSE, RSR, and PBIAS along with their collective performance are computed based on Table 4.

Particularly notable is the PBIAS metrics' correction of precipitation, resulting in a very good performance, also rated as 5. The NSE values for most models have been uplifted from poorer to satisfactory levels, while only a few models display satisfactory results in RSR.

A collective improvement in overall ratings has been observed, with BCC-CSM2-MR at the top of the rank and CanESM5 at the bottom. Out of 13 BCMs, the top six models, namely BCC-CSM2-MR, ACCESS-ESM1-5, MPI-ESM1-2-HR, INM-CNM4-8, MPI-ESM1-2-LR, and INM-CM5-0, have been selected for future climate projection, as they collectively achieve above-satisfactory results for precipitation. Ultimately, the application of linear scaling has further enhanced these models' accuracy and reliability.

Climate change projection utilizes previously selected six BCMs at two emission levels, SSP245 and SSP585 at four stations, Siklesh (824), Kunchha (807), Khairenitar (815), and Damauli (817). The projected timeframe of 2026–2100 has been divided into three intervals, the near future (2026–2050), the mid future (2051–2075), and the far future (2076–2100). Within these intervals, the projections have been interpreted graphically.

For both scenarios, the winter season undergoes relatively smaller increments compared to the other seasons. Given the projected substantial increases in precipitation, particularly under the SSP585 scenario, there is a heightened risk of extreme hydrological events that could exacerbate risks associated with exceptionally high rains, such as dam overflow, increased sediment load, and potential structural failures. Addressing these challenges requires the incorporation of adaptive management strategies in the design and operation of hydropower systems, including enhanced spillway capacity, sediment management practices, integration of Integrated Water Resources Management (IWRM) approaches, and implementation of advanced forecasting and early warning systems to anticipate extreme weather events with reservoir operation protocols.

These collective changes in hydroclimatic variables present both challenges and opportunities for the Madi River Basin. The anticipated increases in precipitation and streamflow may enhance water resources and energy generation but higher flow velocities can result in scouring, erosion, landslides, and increased pressure on dam structures, necessitating the implementation of robust design and maintenance standards, particularly during monsoon season. Such conditions not only pose a risk to the structural integrity of hydropower dams but also to the efficiency of energy production. Moreover, the integration of flexible operation strategies, such as adjustable turbine technology and modular dam designs, could enhance the resilience and adaptability of hydropower systems to fluctuating hydrological conditions. Additionally, diversifying the energy mix with renewable resources and strengthening policy and regulatory frameworks will ensure the longevity and reliability of hydropower projects. In conclusion, the streamflow data at the outlet of Madi River Basin shed light on the complex interplay of climatic and hydrological factors that govern the watershed's flow regime.

Building upon the comprehensive analysis of streamflow and climate scenarios, this section explores three selected hydropower projects. The baseline energy is obtained from SWAT-simulated flow under the design specification of the respective hydropower projects. The projected fluctuations in energy generation from 2026 to 2100 are compared and analyzed with baseline energy, thereby interpreting their implications for the regional hydropower sector.

A collective study of the annual and seasonal scale changes is depicted in Supplementary Table S6. In both scenarios, annual energy generation fluctuates modestly. Seasonally, the pre-monsoon and winter periods manifest the most considerable increment in SSP245 with pre-monsoon rising to 13.90% in MF and winter to 8.13% in the FF. In SSP585, pre-monsoon witnessed the most substantial rise, reaching up to 15.06% in FF, while winter and post-monsoon seasons demonstrated more tempered adjustments.

Furthermore, an examination of the annual and seasonal fluctuations in energy generation is portrayed in Supplementary Table S7. Under the SSP245 scenario, modest annual changes are accompanied by noteworthy seasonal increments; the pre-monsoon and winter seasons exhibit the most significant rises, peaking at 12.05 and 11.72% in the NF, respectively. Similarly, the SSP585 scenario manifests modest annual shifts but presents pronounced seasonal variations. The pre-monsoon season experiences the most substantial increment, reaching 7.30% in the FF, while the winter season demonstrates a divergent trend, declining to −5.38% in the FF. These variations under both scenarios underscore the differentiated impacts of climatic alterations on energy generation across annual and seasonal timescales.

A summative analysis of the annual and seasonal variations in energy generation is presented in Supplementary Table S8. Under the SSP245 scenario, there is a remarkable surge in annual energy generation, peaking at 29.27% in the FF. The most significant seasonal increment is observed in the pre-monsoon period, reaching 231.06% in FF. In contrast, the SSP585 scenario shows its maximum annual change at 40.28% in the MF, with the pre-monsoon season also witnessing the most significant rise at 356.46%. In both scenarios, the winter season demonstrates relatively modest increases, reaching up to 194.02% in SSP585 during the MF.

To bridge the existing knowledge gap regarding the critical implications of climate change on Nepal's hydropower sector, we designed a study focused on the Madi River Basin. The methodologies and insights derived from this study can be applied to other basins with similar characteristics, thus extending the relevance of the findings to a broader hydroclimatic context. In summary, this study affirms the robustness of the SWAT model for hydrological assessments in the Madi River Basin, achieving ‘Very Good’ performance (both R² and NSE greater than 0.8). The necessity for bias correction in CMIP6 GCMs was highlighted, particularly for precipitation variables. Application of the linear scaling method markedly enhanced the predictive reliability, with BCC-CSM2-MR emerging as the better-performing model. Six models were chosen for future climate projection, underscoring the importance of correction in reliable climate impact assessments. Projected analysis reveals an increase in annual precipitation by 76.64% under SSP245 and 108.24% under SSP585 in the far future when compared with the historical baseline level of 2,240.13 mm. Concurrently, the temperature is predicted to follow a consistent upward trajectory, leading to increments of 1.8 and 3.5 °C in the far future under differing scenarios. Increments in annual mean flow fluctuations, ranging from 72.32 to 192.78 m³/s under differing scenarios, from a historical baseline flow of 69.58 m³/s. To enhance the resilience of hydropower systems against the challenges posed by climate change-induced extreme hydrological events, it is crucial to adopt adaptive management strategies, including the implementation of enhanced spillway capacities and advanced forecasting and early warning systems, alongside the integration of robust design standards to mitigate risks from increased sediment load, dam overflow, and structural pressures due to high rains and increased flow velocities.

This study also emphasizes the heterogeneous nature of hydropower potential across different projects within the basin. For instance, Super Madi HPP represents hydropower projects within the main river. Here, annual energy generation is projected to increase by 4.88% under SSP245 and 4.95% under SSP585 in the far future timeframe. Similarly, Madame Khola HPP illustrates the unique challenges faced by projects in the upper tributaries and upper elevations. For the Madame project, annual energy generation in the far future is projected to increase by 3.83 and 2.06% under SSP245 and SSP585, respectively. Both projects are characterized by gradual increases in energy generation, and seasonal patterns of full-capacity operation during monsoon months emphasizing the project's adaptability to varying climatic conditions. Furthermore, Midim Khola HPP presents an encouraging picture for small-scale projects in lower tributaries and middle-lower elevations with annual increment in the far future by 29.27% under SSP245 and 30.6% under SSP585 demonstrating consistent and incremental growth in energy generation. Such outcomes underscore the necessity for location-specific planning and strategic management in optimizing renewable energy sources. The Madi River Basin's projected increase in streamflow and precipitation does not uniformly translate into a corresponding trend in hydropower generation across different projects, revealing the intricate interplay between hydrological variables, hydraulic constraints, and energy production. These results can offer valuable insights for facilitating the planning and development of sustainable energy management and climate adaptation strategies, in evolving global climate challenges.

The authors would like to express sincere thanks to Er. Rupesh Baniya from Nepal Development Research Institute, for his support and invaluable advice during the study, which were instrumental to the success of this study.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.