On the seasonal predictability of the 2020 North Atlantic tropical cyclone season

Emma Lilly Levin, Mu-Ting Chien , Elizabeth Barnes, Haozhe He, Gabriel Vecchi, Wenchang Yang 

The 2020 Atlantic tropical cyclone (TC) season was exceptionally active, producing over twenty named storms, yet several seasonal forecasts failed to predict such extreme activity across their ensemble spread. Even when forced with the observed 2020 sea surface temperatures (SSTs), physics-based models simulated only a moderately active season across their ensemble members. Using observations and statistical, dynamical, and deep learning (DL) models, we evaluate several hypothesis regarding why the observed hyperactive outcome fell outside the ensemble range of the physics-based models forced with observed SSTs. Analysis of observed large-scale conditions indicates that 2020 did not exhibit favorable predictors of hyperactivity, indicating that the moderate activity in the models should not be unexpected. We also find support for a role in subseasonal atmospheric variability in enhancing the 2020 activity relative to predictions based on monthly and seasonal characteristics. To comprehensively characterize the range of outcomes for the 2020 season, we construct a 1,000-member ensemble using an DL emulator forced with observed SSTs. The observed hyperactivity corresponds to a 0.5 percent event within this ensemble. Although highly unlikely in any single year, such an outcome has roughly a 20 percent chance of occurring at least once in a 45-year period. Taken together, our findings support the interpretation that 2020 represents an unlikely but possible outcome, potentially enhanced by subseasonal atmospheric variability, given current understanding, models, and observations. These results serve to remind us that rare events will occur in a chaotic climate system, and large ensembles are one approach to sample them.