New AI framework BaguanCyclone improves tropical cyclone forecasting by correcting systematic biases in track and intensity predictions

BaguanCyclone addresses limitations of current AI weather forecasting systems that rely on coarse-resolution reanalysis data like ERA5 at 0.25 degree resolution

The framework uses a probabilistic center refinement module to model continuous spatial distribution, overcoming discretization errors that constrain predictions to fixed grids

Tackles intensity forecasting challenges for strong tropical cyclones by addressing the smoothing effect of coarse meteorological fields and regression losses that bias predictions toward conditional means

Unified framework combines two key innovations to enhance both track accuracy and intensity predictions for improved tropical cyclone forecasting in tropical and subtropical regions