Nowcasts are high resolution, short-term (e.g. up to 2 hours) weather forecasts of precipitation, cloud coverage or other meteorological quantities widely used in public safety, air traffic control, and many other areas that require high fidelity and rapidly updating forecasts. Numerical weather prediction (NWP) models are insufficient for predicting such near term events. However, having accurate predictions of near-term severe weather events such as thunderstorms is critically important. Meterological data is in no short supply yet the curation of this data can be time consuming and computationally intensive. We present the Storm EVent ImagRy dataset (SEVIR) along with a baseline model that will help you get started.

Please see the notebook in the following repo for further details on the datasets and challenge: