Designing Resilient Multipurpose Reservoir Operation Policies in Presence of Internal Climate Variability

Adaptation planning for water resource systems is fraught with significant challenges, arising from uncertainties associated with diverse climate change scenarios, varying model structures, and Internal Climate Variability (ICV), often captured through multiple initial condition runs. ICV, typically considered irreducible, has received significant attention for state and derived hydrological variables. However, its implications and role in regional decision-making remain elusive. Here, we develop an integrated framework to incorporate uncertainties through hydrological modeling combined with a suite of multi-objective stochastic optimization techniques. This approach is applied to design optimal operating policies for the Sardar Sarovar Dam in Gujarat, India, a multipurpose infrastructure of national importance to meet flood control, hydroelectric generation and domestic, industrial, and irrigation water demands while accounting for two future climate change scenarios, SSP245 and SSP585, with 49 different initializations of each scenario to represent the ICV. We employ Sampling Stochastic Dynamic Programming to incorporate ICV by considering multiple initializations simultaneously, in contrast to Stochastic Dynamic Programming, which evaluates realizations individually. We show that despite the wide range of uncertainties, optimal operating policies can be designed to meet the various demands with reliability of 100%, 59%, and 27% for domestic, irrigation, and industrial water demand, respectively, when all scenarios are considered simultaneously. Our study advocates for the systematic inclusion of a wide array of climate model outputs, emphasizing that such integration is essential not only for crafting robust operating policies, but also for the reliability assessment of current operating policies in light of changing climate and demand scenarios.