Physics-Informed Sampling-Based Model Predictive Control for Event-Scale Precipitation Regulation

Abstract. Extreme rainfall events pose serious risks to human life, infrastructure, and economic systems. To mitigate such events, this paper proposes an effective event-scale precipitation control framework that deploys distributed offshore drag-inducing actuators to perturb the local wind and humidity fields and thereby influence precipitation formation. Existing rainfall-intervention methods are generally limited either to long-term, open-loop interventions at the climate scale or to simplified models that are difficult to extend to realistic weather systems. In contrast, the proposed framework directly integrates high-fidelity numerical weather prediction (NWP) models to preserve realistic atmospheric conditions while enabling closed-loop regulation of individual rainfall events through model predictive control (MPC). Sampling-based optimization makes this integration feasible by avoiding the need for gradient information and supporting efficient parallel computation for real-time implementation. Crucially, a physics-informed spatial prior is introduced to support the sampling process by guiding candidate layouts toward physically plausible regions with high control potential, thereby improving search efficiency. This is key to enhancing the robustness and practical viability of sampling-based optimization for such a complex rainfall control problem. Ensemble-based numerical experiments demonstrate that the proposed framework achieves effective and robust event-scale rainfall regulation under realistic weather conditions.