Multi-sensor Weather Forecast Competition
- Study multi-channel weather movies.
- Predict weather products in various Earth regions.
- Apply transfer learning to new earth regions.
The goal of the competition is a short-term prediction of selected weather products based on meteorological satellites data obtained in collaboration with AEMET/ NWC SAF. Following recent success of our Traffic4cast competitions at NeurIPS in 2019 and 2020, this challenge presents weather forecast as a video frame prediction task. The weather movies consist of multi-channel images encoding the cloud properties, temperature, turbulence, and rainfall. The images are recorded at 15 minute intervals through the entire year. Each pixel in the images represents the area of about 4 km x 4 km, and each region contains 256 x 256 pixels. The regions span varying landscapes including mountains, deserts, islands and seas, and others. The training and validation data contain observations in various earth regions and the test data include additional new regions. The challenge is to predict the encoded weather products in all regions. The competition outcomes are expected to go beyond methodological advances in weather forecasting and video frame prediction. The challenge offers real-world benchmark for few shot and transfer learning and allows testing multi-sensor data fusion. Learn more…
The competition goal is to predict the next 32 images (8 hours in 15 minute intervals) of the weather movies. The images contain four channels encoding the following weather products: temperature (on accessible surface: top cloud or earth), convective rainfall rate, probability of occurrence of tropopause folding, and cloud mask.
The competition data are split into training, validation, and test sets. The core challenge data contain the training, validation, and test sets in three regions: Nile region, Eastern Europe, and South West Europe. The resulting training set contains about 25 billion data points. The transfer learning challenge data contain only the test set in three additional regions: Central Maghreb, South Mediterranean, and Central Europe.
Stage 1 of the competition offers two challenges:
- Core challenge: Predict the encoded weather products in the three training regions. Get the data.
- Transfer learning: Apply transfer learning to predict weather products in the three additional regions. Get the data.
For each day, the images are grouped into a separate folder, within which there are several folders containing the weather products. Learn more…
Regions in blue: core challenge; regions in orange: transfer learning.