Data-driven control reveals distributed flood adaptation priorities across large river networks

Jeil Oh , Matthew Bartos

In the face of growing flood risks, decentralized adaptation measures like detention storage, enhanced infiltration, and floodplain reconnection have the potential to mitigate flood impacts at the river basin scale. However, optimal spatial allocation of flood control measures is complicated by the high dimensionality of hydrologic systems and the sensitivity of proposed strategies to climate uncertainty. To overcome these challenges, we propose a diagnostic framework that combines reduced-order data-driven modeling with optimal control to directly estimate reach-level attenuation targets without the need for iterative simulation and optimization. First, Proper Orthogonal Decomposition (POD) and Dynamic Mode Decomposition with control (DMDc) are used to create a low-rank linear surrogate model of basin rainfall-runoff dynamics. A Linear Quadratic Regulator (LQR) is then applied to compute optimal reach-scale attenuation targets that mitigate flood impacts. Applied to a large river basin under a multi-model climate ensemble, the framework successfully determines distributed attenuation strategies that reduce bankfull discharge exceedances under varying adaptation budgets. Across scenarios, marginal increases to flow attenuation are found to yield diminishing returns to flood mitigation, while higher-emission scenarios retain substantially greater residual flood volume for the same effort level. Although attenuation allocation generally scales with mean flow, we identify tributary and transitional reaches where attenuation demand is disproportionate to local hydrologic size, showing that network-level flood dynamics produce spatial priorities that cannot be recovered from reach attributes alone. Taken together, the proposed framework provides a scalable approach for flood adaptation planning that is effective for basin- to continental-scale applications under climate uncertainty.