Transient upper-ocean warming (from the ocean surface layer to depths of several hundred of meters) is a main contributor to the process of hurricane development. An unusually warmer ocean can potentially “fuel” the atmosphere with more energy, favoring storm development with higher intensification rates. Yet, surprisingly, little progress has been made to improve representation of oceanic conditions in hurricane intensity prediction by considering what happens in the ocean. Traditionally, in addition to the atmospheric variables, hurricane tracking and intensity predictions primarily rely on near-surface ocean-related variables (heat fluxes or surface temperature). However, surface variables do not fully capture the large amounts of heat trapped at depths far from surface interaction that can emerge to the surface and fuel storm systems. We believe that improved understanding of pre-hurricane conditions and forecasts can be achieved using a hybrid model that includes knowledge of the underlying temperature distribution in the ocean and machine learning (ML) models. Subsurface in-situ temperature data (from Copernicus) will be used, as well as satellite-based observations (AVISOS’s MSLA) and hurricane data (e.g., ESA’s Storm portal, NASA’s TCI HIRAD v2.1) to provide additional support. Integration of data-driven ML techniques, such as non-linear kernel-based regression and feature extraction methods, will help better rank the level of importance of each specific subsurface layer and discriminate between the drivers that yield extreme storms. Hence, the main goals are to: 1) identify relationships (from subsurface temperature) in the region where the system first originates, and along the path of the storm; 2) propose a new metric as a “severity score” in anticipation of the development of a storm or hurricane. A better understanding of the ocean changes dominating storm formation will help improve hurricane prediction methods and future space weather missions.
