Hurricane Otis: A Case for a Rapid Migration Toward Probabilistic Tropical Cyclone Forecasting

Hurricane Otis was a tragedy compounded by an astounding failure of forecast models. A mere 30 hours ahead of landfall, not a single operational numerical model predicted that Otis would attain even marginal hurricane intensity, a far cry from the Category 5 storm that slammed into Acapulco. But while the deterministic models failed catastrophically, the SHIPS probabilistic forecast 30 hours ahead of landfall indicated a 25% chance of rapid intensification, and several non-operational but publicly available high-resolution ensemble simulations did indicate a non-trivial probability of hurricane force winds near Acapulco.

Lack of observations of Otis and its environment, including the upper ocean, undoubtedly played a role in the deterministic model forecast failures, and NHC forecasters were able to identify some of these problems in real time and issue forecasts of intensity above those of any of the models, though still well short of what transpired. While we can and should take steps to improve observational capabilities, we ought also to be moving more rapidly toward probabilistic forecasting that accounts for uncertainties in observations and models. We will show that some non-operational but real-time ensembles provided ample evidence that high-intensity landfall was possible, information that could have been used in decision-making. The advent of robotic in-situ observations, faster computers, improved models and downscaling techniques, and advanced machine learning heralds an era of improved probabilistic forecasts, but this will require improved means of communicating uncertainty.