AUTHOR=Argoubi Majdi , Mili Khaled 

TITLE=Stochastic multi-objective optimization for flood control in multi-reservoir systems: an adaptive Progressive Hedging approach with scenario clustering

JOURNAL=Frontiers in Water

VOLUME=Volume 7 - 2025

YEAR=2025

URL=https://www.frontiersin.org/journals/water/articles/10.3389/frwa.2025.1606096

DOI=10.3389/frwa.2025.1606096

ISSN=2624-9375

ABSTRACT=IntroductionFlood-prone regions face growing challenges due to climate-induced variability, rapid urbanization, and competing demands on water infrastructure. Multi-reservoir systems play a critical role in mitigating flood damages, but climate change has intensified management complexity through increased hydroclimatic volatility and reduced reaction times for dam operators.MethodsThis study presents a stochastic multi-objective optimization framework that integrates ensemble-based inflow scenarios, high-resolution hydraulic simulations, and an adaptive version of the Progressive Hedging Algorithm enhanced by K-means scenario clustering. The approach was applied to Tunisia's Medjerda River basin, focusing on five interconnected reservoirs. The model balances downstream flood risk reduction with long-term water storage security by optimizing reservoir release policies across 1,000 synthetic inflow scenarios, reduced to 10 representative scenarios through clustering.ResultsThe proposed method achieved robust performance with normalized objective values of 0.087 for storage security and 0.094 for flood control. More than 93% of simulated scenarios satisfied both storage and flood-related constraints, demonstrating superior reliability compared to traditional rule-based methods (60–70%). The framework converged in 42 iterations with a computational time of 3.2 hours, representing a 6.7-fold reduction compared to full-scenario optimization while maintaining only 6–7% performance degradation. Peak discharge reductions of 25–30% were achieved through coordinated reservoir operations.DiscussionThe framework provides operationally feasible release policies that perform consistently across diverse flood conditions while significantly reducing computational costs. By combining hydrological realism with optimization scalability, this work supports the design of resilient and anticipatory flood management strategies in semi-arid regions, directly contributing to global efforts toward sustainable water governance (SDG 6), climate resilience (SDG 13), and disaster risk reduction in human settlements (SDG 11).