Precipitation Forecasting
Overview
Precipitation forecasting spans radar-based nowcasting over minutes to hours and numerical or AI weather-model forecasts over days to weeks. Methods include convolutional and recurrent networks, transformers, multimodal data fusion, physical advection, and probabilistic generators such as diffusion and rectified-flow models; post-processing and downscaling correct biases and add local detail. Evaluation uses measures such as CSI and threat scores for event detection, alongside CRPS, Brier scores, and spatial or spectral measures for uncertainty and structure. A central challenge is producing calibrated, realistic rainfall amounts and extreme-event forecasts at longer lead times while retaining useful spatial detail across different regions and climates.
Latest papers
56 papers4 papers in the last four weeks, down 64% on the four weeks before. 0.0% of all new papers.