South China Sea marine heatwaves split into two types, with ocean dynamics playing a surprising role

Researchers extend classical CNOP method for deep learning forecasting models with multi-time-slice-input structure.  It reveals when—not just where—input errors matter most in targeted observations. This improves forecasts for ocean-atmospheric variables, especially high-impact environmental events.

In an era of intensifying extreme weather, this review offers a clear message: to better project the future of tropical cyclones in a warmer climate, we must first understand the patterns of the warming seas.