Taking Advantage of Chaos for Efficient Control of Extreme Weather

Since the weather system is chaotic, small differences generally lead to big differences, particularly for storms associated with dynamical instability. We have been exploring controllability through the Control Simulation Experiment (CSE), the same procedure as the well-known Observing Systems Simulation Experiment (OSSE) but with control inputs to the nature run. Our first paper (Miyoshi and Sun, 2022, NPG) proposed the CSE for the first time with proof-of-concept experiments addressing controllability to stay on a chosen regime of the two-regime Lorenz-63 3-variable model. Our second paper (Miyoshi, Sun, and Richard, 2023, NPG) investigated controllability to avoid extreme events with the Lorenz-96 40-variable model. Moreover, we performed CSEs with our global and regional NWP systems for realistic typhoon and heavy rain cases, respectively. In parallel, we applied reinforcement learning to find effective control inputs to a parameter of the Lorenz-63 model. Here, we explored single-sided control inputs to increase the model’s instability, considering realistic situations such that e.g., adding cloud condensation nuclei is more feasible than removing them. We tested the reinforcement learning approach with a case of tropical cyclone genesis and rapid intensification using the 2-dimensional Cloud Model 1 (CM1). In this presentation, we will summarize RIKEN’s activities toward efficient control of extreme weather events.