A MACHINE LEARNING COUPLED DISTRIBUTED HYDROLOGICAL MODEL FOR EL NIÑO IMPACT ASSESSMENT ON THE WATER-ENERGY NEXUS

Abstract

The inter-annual variability of Indian Summer Monsoon Rainfall (ISMR) significantly impacts India's water and energy security, as it is the primary source of rainfall for the Indian subcontinent. Climate teleconnections, such as El Niño events, are crucial in influencing ISMR, leading to substantial rainfall deficits and drought occurrences across India. These reductions in rainfall have profound implications in altering the streamflow patterns and consequent hydroelectric power generation in affected river basins. Assessing the impacts of El Niño events on ISMR, water security, and energy sustainability is imperative for the nation's sustainable development. This study evaluated the impacts of El Niño events on the water-energy nexus within the Godavari River Basin (GRB), India. The study employed a comprehensive framework that integrates hydrological modelling, energy planning, and climate analysis to understand the intricate interdependencies between the water and energy sectors under the influence of El Niño events. To assess the impacts of El Niño events on the water-energy nexus, the development/application of hydrological models plays a vital role. Hydrological models, such as the Variable Infiltration Capacity (VIC) model, are powerful computer-based tools for better process understanding and detailed representation of complex hydrological processes. However, accurate estimation of parameters in VIC by developing effective calibration procedures can improve the characterization of real-world phenomena of hydrological fluxes. This is vital for effective water resources management and decision-making, particularly during El Niño episodes. This study proposed a novel framework for automatically calibrating complex macroscale models that integrates sensitivity analysis and an adaptive pseudo-modelling-based optimization algorithm. The framework was applied to calibrate catchment-specific sensitive parameters of the VIC model at a spatial resolution of 0.15° for streamflow simulations over the GRB. Both single- (scenario 1) and multi-site (scenario 2) calibration approaches were employed, with scenario 2 demonstrating superior performance in terms of objective calibration, equifinality reduction, and reduced parameter uncertainty. Besides improving model accuracy, the Lola-Voronoi adaptive sampling technique was utilized to generate the ensemble pseudo model (PM), significantly reducing the total simulation time compared to the classical 'one shot' approach of constructing PMs. Specifically, the total simulation time was reduced by about 0.19, 0.48, and 0.76 times at Polavaram (downstream), Tekra (middle) and Konta (downstream) gauging stations. Overall, the developed framework enhanced the simulation capacity of the VIC model and facilitated a better understanding of hydrological responses in GRB, particularly during El Niño events. After calibrating the VIC model using this framework, the study analyzed grid-wise flux outputs from the model across a composite of El Niño and normal years to assess El Niño's impacts on the basin's hydrology. The findings revealed an inverse relationship of precipitation, evapotranspiration and soil moisture with the increasing magnitude of El Niño events. This relationship highlighted the basin's increased vulnerability to more frequent droughts during El Niño episodes. To mitigate the vulnerability of GRB during El Niño events and enhance the understanding of their impacts on the water-energy nexus, the study employed an integrated framework combining the VIC model with geospatial tools. This framework explicitly simulated spatio-temporal variations in streamflow within the GRB and identified sites suitable for run-of-river small hydropower plants (RoRSHP) to meet the region's energy needs during El Niño events. The study utilized long-term (1951-2020) daily streamflow data, simulated with the VIC model for design discharge computation at 30%, 75%, and 90% flow dependability for identifying the most suitable locations which are ungauged. The analysis revealed significant potential for RoR development within the GRB, identifying 226 initial sites based on the head along the river, with a combined power and annual energy generation estimate of 92 MW and 0.4 TWh/yr, respectively, at 90% flow dependability. Further analysis of these potential sites during El Niño years demonstrated a decline of approximately 46%, 38%, and 18% in total annual energy at 30%, 75%, and 90% flow dependability, respectively, compared to normal years. This finding underscored the increased risk of power shortages in the GRB during El Niño years, emphasizing the imperative need for implementing water-energy nexus strategies to cope with the risks associated with El Niño events. Consequently, to bolster long-term energy security and resilience against future El Niño events, the study meticulously screened initial sites and identified nine potential locations for RoR-SHP development. These sites were selected based on criteria such as head, power potential, and viability under El Niño conditions, ensuring their capability to maintain firm power even during El Niño years. Future energy planning can leverage these insights to develop appropriate strategies that effectively manage the risks associated with El Niño and ensure the resilience of the region's energy infrastructure. Besides assessing the impacts of El Niño events in the present climate, it is critical to comprehend their implications on the water-energy nexus in the future climate to meet the rising needs for energy and water. The study employed an extreme gradient boosting (XGBoost) algorithm to generate locally unbiased climate input variables for the VIC model, thereby reducing uncertainty in future El Niño impact assessment. The methodology leveraged diverse training datasets, including five general circulation models and topographic data (elevation, slope, aspect) for robust training of XGBoost. Compared to the conventional Quantile Delta Mapping (QDM) method, the XGBoost framework significantly reduced biases in downscaled climate variables across the GRB. For instance, the proposed model exhibited significant improvement in achieving the maximum NSE values of 0.44, 0.96 and 0.97, corresponding to precipitation, maximum and minimum temperature, while the conventional QDM method yielded the NSE of -0.34, 0.56 and 0.75, respectively over the GRB. By leveraging bias-corrected and downscaled climate variables from the XGBoost framework, the study projected future changes in the water-energy nexus under the SSP245 scenario for initially proposed small hydropower sites. The findings revealed an absolute increase of 7 TWh/yr (i.e. 201%), 2 TWh/yr (i.e. 362%) and 0.6 TWh/yr (i.e. 181%) in the total annual energy at 30%, 75%, and 90% flow dependability, respectively. The projected increase suggests reduced vulnerability of the basin during future El Niño events, offering valuable insights for sustainable planning and management strategies. In conclusion, the present study provides valuable insights into advancing sustainable development in India's water and energy sectors, particularly in light of potential El Niño episodes in the future. Implementation of enhanced management strategies and operational policies can minimize vulnerabilities and facilitate adaptation to upcoming El Niño occurrences nationwide.