Searching for Chaos in Tropical Cyclone Intensity: A Machine Learning Approach

Do tropical cyclones (TC) possess chaotic dynamics at any stage of their development? This is an open yet important question in current TC research, as it sets a limit on how much one can further improve intensity forecast in the future. This study presents a novel use of machine learning (ML) to quantify TC intensity chaos. By treating TC scales as input features for different ML models, we show that TC dynamics displays a limited predictability range of similar to 3 hours at the maximum intensity (PI) state under a fixed environment, which confirms the existence of a chaotic regime in TC development. Using the minimum central pressure as a metric for TC intensity could extend the predictability range up to 9 hours, yet the low-dimensional chaos of TC intensity is still captured in all ML models. Additional sensitivity experiments with different ML model configurations, the number of input features, or sampling frequency all confirm the robustness of such limited predictability for TC intensity, thus supporting the existence of low-dimensional chaos at the PI limit. The existence of such intensity chaos has a profound implication that TCs must possess an intrinsic intensity variability even under an idealized condition. This internal variability dictates a lower bound for the absolute intensity error in TC models regardless of how perfect the TC models or initial condition will be.