Modeling North Atlantic Tropical Cyclone Counts Directly From Sea Surface Temperature Maps

Abstract Annual North Atlantic tropical cyclone (TC) counts are frequently modeled as a Poisson process with a state‐dependent rate. Current models based on Poisson regression can explain roughly 50% of the annual variance using three climate indices: El Niño/Southern Oscillation, average sea surface temperature (SST) in the main development region of the North Atlantic, and the North Atlantic oscillation atmospheric circulation index. We introduce a new method, based on the Elastic Net (EN) that predicts TC counts directly from global SST maps. We show it achieves performance on par with current models, without requiring manually constructed indices. To understand the performance of the EN we argue that, when TC counts are generated by independent Poisson draws, statistical models are subject to a lower limit on prediction error. We estimate this limit and show that it is saturated by both current models and our new method.