Reinforcement Learning for Pre-Landfall Typhoon Intensity Control

This project is very ambitious — and more directly connected to the question of whether science can protect lives. Both China and Japan are actively exploring whether small, targeted perturbations can reduce typhoon intensity before landfall. I have been investigating one specific mechanism: if we could pre-cool a patch of the ocean ahead of an approaching typhoon — by mixing warmer surface water with cooler water below — could we meaningfully reduce the storm’s intensity? And if so, how should we optimize the location, timing, and strength of that intervention? To answer this, I turned to reinforcement learning, the same class of algorithms that underlies AlphaGo and ChatGPT. The result was striking: the AI was able to identify intervention strategies that do work — strategies that reduce typhoon intensity in ways that would not be obvious from physical intuition alone. The idea that AI might one day help us manage one of nature’s most destructive phenomena feels extraordinary to me, and it is a possibility I want to pursue further.