"Unified Preprocessed Weather-Radar Nowcasting Datasets: SEVIR-VIL + HKO-7 + MeteoNet"

"This dataset release provides a unified, preprocessed benchmark pack for precipitation nowcasting across three widely used weather-radar datasets: SEVIR (VIL), HKO-7 radar echo, and MeteoNet radar. All three sources are converted into a consistent spatiotemporal tensor format with harmonized data organization, enabling drop-in training and evaluation of deep learning nowcasting models under a single pipeline. The release includes standardized train\/validation\/test splits, per-sample sequences of past observations and future targets (e.g., [N_in] input frames \u2192 [N_out] forecast frames at [\u0394t] intervals), and basic metadata needed for reproducible experiments. By consolidating multiple radar benchmarks with consistent preprocessing, this pack supports fair cross-dataset comparisons, robustness studies, and uncertainty-aware nowcasting research. Original dataset licenses and attributions apply; this package distributes processed derivatives for research use."