Understanding the underexplored role of deeper ocean layers in hurricane dynamics is pivotal amidst escalating storm intensities and growing threats to coastal communities. While early forecasting approaches focused almost exclusively on near-surface variables (e.g. sea surface temperature) or upper-ocean layers (0–100 m), some follow-up efforts integrated subsurface information using ocean heat content analyses relative to the 26 °C isotherm, yielding modest enhancement for the most intense storms. By contrast, our convolutional neural network–random forest (RF) framework explicitly exploits the three‐dimensional structure of ocean anomalies from the surface down to 500 m—indirectly capturing heat redistribution, salinity‐driven stratification, and mixed‐layer dynamics across depths. We demonstrate that the layered information within 40–500 m accounts for 44.91 ± 0.24% of total variable importance in the RF model and supports 72-h lead‐time severe‐hurricane forecasts with a precision of 73.04 ± 7.95%. This substantially extends ocean-heat-content‐only approaches by: 1) avoiding a hard threshold, 2) providing the first quantification of layer-specific subsurface contributions to severity forecasting through a multi-depth anomaly analysis across multiple variables using an RF-based framework, and 3) leveraging fine‐scale spatial patterns via convolutional-neural-network‐derived features. Subsurface variables down to 500 m are not merely ancillary—they are indispensable for capturing the ocean precursors of the strongest hurricanes, particularly when the 26 °C isotherm (typically shallower than 100 m) fails to represent the full heat reservoir. This deeper thermal structure may represent a key energy source that can support—and potentially accelerate—severe hurricane intensification. These findings highlight the untapped predictive value of deep-layer ocean information and its potential to enhance early-warning systems for high-impact storm events.

Read original article

Read original article