Investigating the projected changes in water balance components under climate change considering the effect of storage structures

Hydrological modelling is a tool that is commonly used in water balance studies to understand the movement and distribution of water in a particular area (Devia et al. 2015; Sood & Smakhtin 2015; Abdulkareem et al. 2018). It involves the use of mathematical models to simulate the hydrological processes that occur in a given area, such as precipitation, evapotranspiration, infiltration, and runoff. The results of these models can provide valuable information about the water resources availability in an area and help in decision making for water management (Nandi & Reddy 2020a; Chandu et al. 2022). One such example for the use of hydrological modelling is performing water balance studies considering the climate change impacts. By simulating the hydrological processes in an area under different climate scenarios, researchers have predicted how changes in precipitation and temperature will affect the availability of water resources. This information can then be used to develop strategies for adaptation, and mitigating the impacts of climate change on water resources. Hydrological modelling has been also adopted for agricultural water management, groundwater assessment, flood and drought monitoring (Singh et al. 1999; Saraf et al. 2004; Corzo Perez et al. 2011; Nandi & Reddy 2022b). Overall, the hydrological modelling plays a pivotal role in assessing water availability, especially in river basins where climate change and anthropogenic activities are drastically impacting water resources. While numerous models exist, there is an increasing emphasis on incorporating reservoir storage structures to better predict streamflow and manage water resources.

Physics-based spatially distributed hydrological models offer several advantages over lumped or black box models (Liu & Gupta 2007; Fatichi et al. 2016; Clark et al. 2017). One of the main advantages of physics-based models is their ability to represent the spatial and temporal variability of hydrological processes, such as precipitation, evaporation, and infiltration. This allows for a more accurate representation of the movement of water in a given area and can provide valuable insights into the dynamics of the hydrological system (Butts et al. 2004; Antonetti et al. 2017; Pilz et al. 2019). Another advantage of physics-based models is their ability to incorporate detailed data on the physical characteristics of the study area, such as topography, soil type, and vegetation. This allows for a more accurate representation of the interactions between water and the environment and can provide valuable information about the factors that influence the availability of water in a given area. One of the most popular physics-based hydrological models is the variable infiltration capacity (VIC) model. The VIC model, developed at the University of Washington, is a physically-based, distributed hydrological model that has been widely used in water balance and other hydrological studies (Liang et al. 1994). The VIC model has several key benefits, including its ability to represent the effects of vegetation, snow, and frozen soils on the hydrological processes in a given area. It has been used in a wide range of applications, including the study of the impacts of climate change on water resources, prediction of water availability for irrigation and other purposes, and assessing the impacts of land use changes on water resources (Hengade & Eldho 2016; Das et al. 2018; Nandi & Reddy 2020a; Chandu et al. 2022). The routing application for parallel computation of discharge (RAPID) river routing model in combination with the VIC model can be used effectively to simulate the hydrological response of a river basin. However, few have incorporated storage structures into this modelling framework. Recent research by Dang et al. (2020) and Wang et al. (2021) highlighted the potential improvements in model accuracy upon including reservoir dynamics.

While hydrological models provide vital ground-level information, understanding the broader impacts of climate change necessitates a more global perspective. Many recent studies collectively shed light on the complex facets of climate change and its repercussions on water resources and ecosystems (Jabal et al. 2022; Nama et al. 2022). This is where general circulation models (GCMs) come into play. GCMs provide future climate scenarios by simulating global atmospheric, oceanic, and land surface processes. Combining the fine-scale resolution of spatially distributed hydrological models with the global climate outlook offered by GCMs can yield comprehensive and reliable projections, thereby enriching our understanding of future water resource availability (Venkataraman et al. 2016; Singh et al. 2019; Vemula et al. 2019; Hamed et al. 2022). One challenge in using GCMs in hydrological studies is that the resolution of the models is typically too coarse to accurately represent the processes that occur at the local level (Giorgi & Mearns 1991; Xu 1999; Lebel et al. 2000). To address this limitation, dynamic and statistical downscaling techniques are often used to refine the GCM output and provide more detailed information about the local climate (Prudhomme et al. 2002; Chokkavarapu & Mandla 2019; Daniel & Abate 2022; Daniel 2023b). Statistical downscaling involves the use of statistical techniques to relate the output of a GCM to observed climate data at the local level. This approach is less computationally intensive and can be applied to any region for which observed climate data are available (Kannan et al. 2014; Shashikanth et al. 2014; Okkan & Kirdemir 2016; Daniel 2023a). It is therefore a useful tool for providing detailed information about the local climate and has been widely used in hydrological studies. One example of the use of GCMs in water balance studies is studying the impacts of climate change on water resources. By simulating the climate of a given region under different climate scenarios, some studies have projected how changes in temperature and precipitation can affect the availability of water in the future (Kristvik et al. 2019; Saranya & Nair Vinish 2021; Rajabi et al. 2022; Silva et al. 2022). This information can be used to develop suitable strategies for adaptation and mitigating the impacts of climate change on water resources. Using multiple GCMs in hydrological studies can provide a more robust assessment of the likely changes in temperature and precipitation that will occur in a given region (Johnson & Sharma 2011; Deb et al. 2018; Pandey et al. 2019; Nandi & Reddy 2020a). By comparing the results of multiple GCMs, researchers can better assess the uncertainty associated with their projections and can develop more robust adaptation and mitigation strategies.

While the integration of general circulation models (GCMs) provides a global perspective on climate change impacts, it is essential to apply this knowledge to regions where these changes have significant implications. In this context, we turn our attention to the Upper Bhima River basin (UBRB) in India, a region of paramount importance for agriculture, hydropower generation, and domestic water supply. The UBRB faces growing challenges stemming from a burgeoning population, escalating water demand, and the looming spectre of climate change (Shukla et al. 2020; Mann & Gupta 2022). The UBRB is characterized by a semi-arid climate and is dependent on monsoon rains for its water supply. In this study, we conduct a spatio-temporal analysis of the water balance components and their projected changes in the near-future under climate change in the UBRB. This study is particularly important for the UBRB, which is vulnerable to the effects of climate changes. In addition, the presence of storage structures in the basin, such as reservoirs, can significantly affect the water balance and must be accounted for in order to accurately predict future water availability.

The primary objective of this study is to evaluate the performance of the VIC–RAPID model in the UBRB, emphasizing the significance of including storage structures within the modelling framework. Furthermore, we aim to assess the potential impacts of climate change on various water balance components in this region. Along with impacts on water balance components, assessing potential changes in flood hazard is also crucial for infrastructure planning and disaster management in the UBRB. Therefore, an analysis of flood frequency and risk under climate change projections will be undertaken in this study which aims to fill the knowledge gap concerning the effects of climate variability and change on flood frequencies and intensities, which holds practical implications for policy formulation and adaptive strategies. Notably, this study is the first to apply this comprehensive approach to the UBRB, offering valuable insights into understanding the effects of climate change on water resources. In the context of modern hydrological models, which encompass a range of processes such as rainfall–runoff relations and evapotranspiration estimates, the role of storage structures like reservoirs is often underestimated. However, these structures can substantially modulate streamflow and influence downstream water availability. Our approach bridges these aspects, creating a holistic modelling framework for a more comprehensive assessment.

The present study uses a distributed hydrological modelling approach based on the VIC model to simulate water balance components in river basins. The RAPID river routing model is used to route the VIC-generated runoff. The inclusion of storage structures within hydrological models is crucial for accurately representing anthropogenic interventions in river basins. These structures have a significant impact on catchment hydrology and the corresponding water budget of the basin. Additionally, dams and reservoirs can also reduce flooding in a basin. Therefore, a storage structure module was developed and integrated with VIC–RAPID modelling framework. To perform the hydrological simulations, the necessary gridded input data for VIC modelling over the UBRB were collected, processed, and prepared in the appropriate format. The configuration of the VIC and RAPID models (e.g., resolution of model grids and number of soil layers) was selected based on the recommendations from previous studies (Nandi & Reddy 2022b). The VIC–RAPID model was automatically calibrated using a self-adaptive differential evolution (SaDE) algorithm (Nandi & Reddy 2020b) and validated using various performance measures. The calibrated VIC–RAPID model was used to simulate water balance components for both the baseline (1999–2010) and near-future (2019–2040) periods with climatic forcing. A spatio-temporal analysis of the water balance components was performed at monthly, seasonal, and annual timescales. The study also investigated the impacts of climate change on water resources in the basin for the near-future. Next, the proposed integrated model was used to estimate future flood events using future hydrological simulations derived from multiple bias-corrected Coupled Model Intercomparison Project Phase 5 (CMIP5) model outputs. Lastly, to evaluate changes in future flood risk, flood frequency analysis was conducted using the simulated streamflow time series. Annual maximum daily flows were extracted to generate peak flow series for the baseline and future periods. Suitable probability distributions were fitted to the respective peak flow data and return levels estimated for different return intervals. The Log-normal, Gumbel, and Pearson type III distributions were tested and selected based on goodness-of-fit metrics.

The VIC model is a widely used hydrological model that simulates the water balance at the land surface. It is based on the assumption that the land surface can be divided into a grid of cells, with each cell representing a uniform area of land with uniform properties. The VIC model uses a set of physical equations to calculate the water balance for each cell, considering processes such as evapotranspiration, infiltration, snow accumulation and melt, and runoff. In this study, the three-layer VIC-5 model is used which simulates the water balance using three soil layers, with each layer representing a different depth in the soil. The top two soil layers in the VIC model control processes such as evapotranspiration, infiltration, and runoff. Plant roots are typically found here and most of the water that is taken up by plants is stored in these zones. The bottom soil layer in the VIC model is typically deeper and control processes such as baseflow. Additionally, the VIC model assumes that the effects of vegetation on the water cycle can be represented by a set of empirical parameters that describe the transpiration and interception of water by plants. These flexible parametrization schemes enable the VIC model to simulate the complex processes involved in the water cycle at high spatial and temporal resolution, making it a valuable tool for studying the effects of climate change and land use on water resources. One of the key advantages of the VIC model is its flexibility and adaptability. The model can be easily configured to simulate the water balance for a wide range of different climates, land use types, and vegetation cover. Additionally, the model source code is freely available and well-documented, making it easy for researchers to use and modify the model for their specific needs.

In this work, a separate routing model called the RAPID model is used to transfer the produced runoff from the VIC model grid to the outflow of the basin. The RAPID model determines the flow through the river network to the basin outlet by resolving a matrix-based Muskingum equation (David et al. 2011). The VIC model does not have a routing module, so the RAPID model is necessary for this purpose. The surface and subsurface runoff time series from the VIC model is used as the input to the RAPID model for obtaining the streamflow at the basin outlet. One of the key advantages of the RAPID model is its ability to handle large-scale, parallel computations, making it well-suited for studying complex river systems. This contrasts with many older routing models, that are limited in their ability to handle multiple input variables and computational demands.

To account for the storage structures, the streamflow routing in the RAPID model is modified by implementing the following steps. First, the location of the storage structures is determined, and they are linked to the associated river reaches in the RAPID model. Next, the river network connection for the whole study area is broken at the river reaches that have storage structures. This creates a number of smaller river networks with no storage structures within them. Routing in these smaller river networks is carried out similarly to the original RAPID model routing scheme. However, the transfer of water from an upstream river network to a downstream network is made using a simple reservoir operation policy in the case of two river networks separated by a storage structure. In the study region, reservoirs are mainly created to meet irrigation demands during non-monsoon and manage floodwater during monsoon. To incorporate this information in reservoir operation, all the reservoirs are designed to store the incoming water during the monsoon and release the stored water uniformly during the non-monsoon season. Also, a storage space about 20% of live storage capacity is kept empty for attenuating the peak flows of hydrograph to account for flood control. The violation of minimum and maximum storage capacity is checked for all time periods. This simple reservoir operation policy is especially helpful where there is not enough information regarding the rule curves or the provisions for the various water uses of the storage structures that need to be modelled. However, a provision can also be made in the current reservoir operation module to incorporate the operational rule curve for better release scheduling in case of related reservoir operation information is available.

The SaDE algorithm has been found to be a reliable and efficient choice for the automatic calibration of the hydrological model. This algorithm offers several improvements over traditional DE algorithms, making it a desirable option for calibration of complex hydrological models. In particular, SaDE has been shown to be significantly faster in terms of convergence to optimal solutions compared to other popular hydrological model calibration techniques such as simple DE, genetic algorithm, and shuffled complex evolution algorithms. In this study, a total of 15 parameters from the VIC and RAPID models were considered for the SaDE-based automatic calibration. These parameters are essential for accurately simulating the hydrological processes within the model, and their selection and calibration can significantly affect the overall performance and reliability of the model. The selected parameters and their descriptions are presented in Table 1, and further implementation details and their sensitivity analysis can be found in Nandi and Reddy (2022a, 2022b).

Table 1

Details of the VIC–RAPID parameters with their associated ranges selected for sensitivity analysis

Model .	Parameters .	Description .	Unit .	Lower .	Upper .
VIC model	binflt	Infiltration parameter	–	0	0.4
	d1	Depth for soil layer 1	m	0	0.5
	d2	Depth for soil layer 2	m	0	3
	d3	Depth for soil layer 3	m	0	3
	Dm	Maximum baseflow velocity	mm/day	0	30
	DS	Fraction of Dm for nonlinear baseflow	–	0	1
	WS	Fraction of max soil moisture for nonlinear baseflow	–	0	1
	KS1	Saturated hydraulic conductivity for soil layer 1	mm/day	1	10,000
	KS2	Saturated hydraulic conductivity for soil layer 2	mm/day	1	10,000
	KS3	Saturated hydraulic conductivity for soil layer 3	mm/day	1	10,000
	Ex1	Brooks–Corey exponent for soil layer 1	–	3	30
	Ex2	Brooks–Corey exponent for soil layer 2	–	3	30
	Ex3	Brooks–Corey exponent for soil layer 3	–	3	30
RAPID model		Multiplier for Muskingum storage constant	–	0	0.6
	X	Muskingum weighting factor	–	0	0.5

Model .	Parameters .	Description .	Unit .	Lower .	Upper .
VIC model	binflt	Infiltration parameter	–	0	0.4
	d1	Depth for soil layer 1	m	0	0.5
	d2	Depth for soil layer 2	m	0	3
	d3	Depth for soil layer 3	m	0	3
	Dm	Maximum baseflow velocity	mm/day	0	30
	DS	Fraction of Dm for nonlinear baseflow	–	0	1
	WS	Fraction of max soil moisture for nonlinear baseflow	–	0	1
	KS1	Saturated hydraulic conductivity for soil layer 1	mm/day	1	10,000
	KS2	Saturated hydraulic conductivity for soil layer 2	mm/day	1	10,000
	KS3	Saturated hydraulic conductivity for soil layer 3	mm/day	1	10,000
	Ex1	Brooks–Corey exponent for soil layer 1	–	3	30
	Ex2	Brooks–Corey exponent for soil layer 2	–	3	30
	Ex3	Brooks–Corey exponent for soil layer 3	–	3	30
RAPID model		Multiplier for Muskingum storage constant	–	0	0.6
	X	Muskingum weighting factor	–	0	0.5

View Large

In this study, the Nash–Sutcliffe efficiency (NSE) is used as the objective function for the SaDE-based automatic calibration of the VIC–RAPID model. The performance of the calibrated model is evaluated using two additional statistical measures: the correlation of determination (R²) and percentage bias (PBIAS). The formulation and optimal ranges for these three performance measures are listed in Table 2.

Table 2

Statistical performance measures used for calibration and validation of the VIC–RAPID model

Name .	Formulation .	Optimal Value .
NSE		1
Correlation of determination (R2)		1
PBIAS		0

Name .	Formulation .	Optimal Value .
NSE		1
Correlation of determination (R2)		1
PBIAS		0

*Sim_i and Obs_i are the simulated and observed streamflow at ith time step; and are the mean of simulated and observed streamflow for all time periods.

View Large

In this study, multiple datasets are used to simulate and predict the current and future streamflow values for the UBRB. The baseline period for this study is 1999–2010, and daily meteorological data for this period, including rainfall, minimum and maximum temperature, and wind speed, are obtained from a range of sources. This includes the Indian Meteorological Department (IMD), the National Centre for Environmental Prediction (NCEP) reanalysis data, and the Indian Monsoon Data Assimilation and Analysis (IMDAA) project. In addition to meteorological data, this study also uses land cover, soil, and elevation data for the hydrological simulation. The land cover data are obtained from the MODIS land cover database, while the soil data are obtained from the Food and Agricultural Organization's (FAO) harmonized world soil database. Elevation data are obtained from the Shuttle Radar Topography Mission 90 m digital elevation model. To simulate streamflows in the near-future (2019–2040), this study uses bias-corrected and downscaled output from 19 GCMs (Representative Concentration Pathways (RCP) 4.5 and 8.5) obtained from NASA. These GCMs are described in detail in Table 3. Detailed information about the river network, which is required for streamflow routing, is taken from the HydroRIVERS project (https://hydrosheds.org). In total, 10 major and medium-sized reservoirs over the UBRB are considered for the present work. The locations of the reservoirs in the study area are shown in Figure 1. The details of the location and the capacity of reservoirs are presented in Table 4. This reservoir information is collected from the India Water Resources Information System. The Takli River monitoring station is located downstream of all the reservoirs and chosen as the streamflow observation point for calibration and validation of the hydrological model.

Figure 2 shows that daily streamflow simulations from the two configurations (i.e., with and without storage structures) are well matching with the observation dataset, especially for the low and medium flows. It is clearly seen from Figure 2 that the VIC–RAPID model by incorporating storage structures produced better streamflow simulation as compared to the model without storage structures. The VIC–RAPID model, without incorporating storage structures, overestimated the streamflow. To quantitively evaluate the model performance, the NSE, R², and PBIAS values of the two model configurations are computed and are presented in Table 6. For the calibration period, a significant 30% improvement is found in the daily NSE score in the case of model simulations with storage structures as compared to the model simulations without incorporating the effect of storage structures. The daily streamflows simulated without considering storage structures produced an R² value of 0.59, which was increased to 0.68 when storages structures are incorporated into the VIC–RAPID model. The PBIAS score also improved to a great extent (i.e., deceased by five times) when the effect of storage structures is incorporated in the modelling. Similar improvements can also be seen during the validation period for the VIC–RAPID modelling incorporating the storage structures. Although improvement in streamflow simulation is observed for the model with storage structure configuration, it is not clear from Table 6 how much of this improvement is happening solely due to the incorporation of the storage structure in the modelling. To get this estimate, the calibrated parameters of the model without storage structure are used for streamflow simulation by the model incorporating storage structure configuration. The resulting NSE, R², and PBIAS score are found to be 0.54, 0.77, −137, and 0.47, 0.73, −91 for the calibration and validation periods, respectively. These performance measures show a noticeable improvement in streamflow simulation during the calibration period attributed to incorporation of the storage structures in the model.

The present study offers significant contributions to the understanding of the hydrological dynamics of the UBRB in the context of climate change. Utilizing the VIC–RAPID model and incorporating a storage structures module, this work provides an extensive analysis of the water balance components under both historical and future climate scenarios. This research is particularly timely and impactful, considering the growing water scarcity issues and the anticipated shifts in climatic patterns due to global warming. The findings of this study underscore the critical role of storage structures in enhancing the reliability of hydrological simulations, thereby aiding in effective water resource management. Moreover, the projections of future water balance components and streamflows present valuable insights for policy formulations in flood management and sustainable agriculture, especially in monsoon-dependent regions like the UBRB.

One of the salient findings of this study is the impact of climate change on the seasonal and annual water balance components, including precipitation, evapotranspiration, surface flow, and baseflow. The results reveal a notable decrease in annual precipitation and a subsequent increase in evapotranspiration, which can exacerbate existing drought conditions in the region. Similar patterns have been observed in other studies that focused on climate change impacts on water resources in the UBRB or Krishna River basin (Raje et al. 2014; Chanapathi et al. 2018; Nandi & Reddy 2020a). The study also found that the decline in monsoon precipitation could have a cascading effect on agriculture, aligning with the findings of Mehrotra et al. (2013). However, Rudraswamy et al. (2023) indicated a significant increase in precipitation and streamflow in the future (2015–2100) over the Tungabhadra River basin, which contrasts with our observations of a decrease in annual precipitation and streamflows in the UBRB. Such disparate trends could be due to differences in geographic location, climate models used (CMIP5 vs CMIP6), or other factors (difference in hydrological models and evaluation period), underscoring the need for region-specific assessments and adaptation strategies, particularly in light of the high uncertainties associated with different hydrologic and climate models.

Another key aspect of this work is the consideration of storage structures in hydrological modelling. The inclusion of these structures led to a significant improvement in the model's performance, with around a 30% enhancement in terms of NSE. This aligns well with existing research that advocates for the inclusion of human influences such as dams and reservoirs in hydrological models for better water availability assessment (Zhao et al. 2016). Our study advances this line of research by providing evidence of how storage structures can also modulate the impacts of climate change on water balance components, an area less explored in existing literature.

While this study adds considerable value to hydrological modelling and water resource management under climate change scenarios, it is not without limitations. For example, although the NSE values of 0.61–0.65 achieved in this study are considered satisfactory performance, it is clear from the results that further improvements can be made to enhance the reliability and robustness of the hydrological simulations, especially given the higher PBIAS values from the streamflow simulation. The NSE metric provides a general indicator of model fit but does not fully capture the ability to simulate complex spatial and temporal dynamics. Several potential sources of uncertainty may be limiting model performance. Firstly, incomplete or incorrect parameterization likely contributes to the modelling uncertainties. The calibration process aims to estimate suitable parameters, but identifying the globally optimal set is challenging. Using multi-objective calibration and sampling from posterior parameter distributions could better quantify parametric uncertainties (Adeyeri et al. 2020). Secondly, the model structure itself has limitations in fully capturing the catchment hydrology. The VIC hydrologic model, while robust, simplifies subsurface flows and can offer limited understanding of groundwater dynamics without consideration of lateral movement of water (Sridhar et al. 2018). Therefore, introducing better surface–groundwater linkages and lateral movement of water may improve flow simulation in the model. Thirdly, uncertainties in input data propagate into the hydrologic simulations. For example, land use changes over time are not considered in the current study. Integrating remote sensing data to regularly update land cover parameters could improve simulations (Teklay et al. 2019). Fourthly, the choice of using the CMIP5 dataset over the CMIP6 dataset represents another limitation. Although CMIP5 has been widely employed, we acknowledge that the CMIP6 models provide advantages such as higher resolution and updated model physics. Fifthly, assumptions made in the study also contribute to uncertainties. One such assumption was the use of a simple reservoir operation policy for simulating water release. In the absence of detailed rule curves, this approach serves as a first approximation but may deviate from actual reservoir operations. Another assumption was the use of an ensemble of 19 GCMs from the CMIP5 dataset without evaluating the individual performance of these models for the study area. While an ensemble approach provides a broad view of possible climatic impacts, more targeted use of high-performing models could refine our simulations. Lastly, lack of human factors like small-scale irrigation abstractions may contribute to streamflow biases, especially in the non-monsoon low flow periods. Introducing such processes into the model could further enhance simulations. Future studies could aim to address these identified limitations in the current modelling framework.

This study aimed to fill research gaps in hydrological modelling that incorporates storage structures, as well as the localized impacts of climate change on the UBRB. The findings demonstrate that considering storage structures like reservoirs lead to more accurate hydrological models, which is of significant value for water resources management in the region. Particularly, the study focuses on evaluating the performance of the VIC–RAPID model in simulating the water balance of the UBRB, and to assess the effects of climate change on water balance components in the region by introducing a storage structures module into the VIC–RAPID modelling framework. The results affirm that control structures like reservoir storages play an important role in hydrological modelling applications and their reliable use in water resources management. These findings are crucial considering future growth in hydropower development and flood management activities, especially in developing countries like India. The main findings of this study are as follows:

• (1) A significant portion of the rainfall in the UBRB is getting lost in the form of evapotranspiration. The analysis of the future water budget indicated a negative climatic effect, i.e., a significant decrease in the precipitation during monsoon seasons, which would further worsen the drought conditions in the basin.

• (2) Consideration of storage structures in the VIC–RAPID model led to a significant improvement (∼30% in terms of NSE) in the estimation of streamflows. The results from the study also highlighted that apart from different parametrization schemes, the incorporation of the storage structure in the modelling could also significantly improve the streamflow simulations and water availability assessment in the basin.

• (3) There may be a slight decrease in the annual streamflows of the UBRB, but the monsoon streamflow might decrease by a significant margin which would bring large impact on the monsoon-dependent agriculture productivity in the basin. This identifies the risk to water security during the critical kharif cropping.

• (4) The baseflow in the UBRB was found to be most sensitive to climate change as compared to other water balance components, and its significant decrease in near-future may bring decreased biodiversity and an overall decline in the health of the river ecosystem.

• (5) Flood magnitude is projected to reduce compared to current levels. This provides information relevant for future hydropower expansion and flood mitigation planning.

Finally, this study encourages the development of adaptive water management strategies focusing on the efficient use of storage structures to mitigate the impacts of climate change in water availability. Future work should explore the integration of more sophisticated climate model output and detailed reservoir operation policies to refine predictions and management rules.

Data cannot be made publicly available; readers should contact the corresponding author for details.

The authors declare there is no conflict.