Robust Regional Predictability of North Atlantic Tropical Cyclones on Seasonal Timescale

Dan Fu , Xue Liu, Fred Castruccio, Stephen Gerald Yeager, Gokhan Danabasoglu, Ping Chang

Tropical cyclones (TCs) are intermittent extreme weather events, yet their seasonal occurrence can be constrained by slowly varying climate variability and regional predictability remains poorly understood. Using 44 years of May-initialized ensemble hindcasts, we show robust seasonal predictability of North Atlantic TC track density across two prediction systems with contrasting resolutions: nominal 1° versus 0.25° atmosphere/0.1° ocean. Both show similar skill over the Gulf region and Caribbean Sea, with correlations approaching 0.6, despite underestimated TC climatology in the low-resolution system. This predictability is linked to a large-scale sea surface temperature pattern predicted months in advance, with correlations exceeding 0.8, and associated atmospheric circulation changes. Yet similar observed skill arises from sharply contrasting model-internal signal-to-noise characteristics: weak ensemble-coherent TC variability in the low-resolution system but a stronger predictable signal in the high-resolution system. These results suggest that weak model signal-to-noise does not necessarily imply weak physical predictability for intermittent extreme weather.