Dear Competitors,
We are happy to announce that all Weather4cast leaderboards are now open and active. We invite you to test your models in different scenarios: long-time prediction, nowcasting, and spatio-temporal transfer learning challenges!
This year, we have prepared for you the following leaderboards:
– Core: The task is to predict 8 hours into the future for so-called known regions: boxi_0015, boxi_0034, boxi_0076, roxi_0004, roxi_0005, roxi_0006, roxi_0007, covering the years 2019 and 2020. For this task, you can train your models using the dataset from the same regions and the same years.
– Nowcasting: The task is the same as for the Core leaderboard, but we ask you to predict 4 hours into the future. As the 8-hour prediction is a challenging task, we are providing another leaderboard with shorter prediction times. We are interested to see if you can design and train a model that will give better results for a shorter prediction time.
– Transfer: The task is to predict 4hours into the future or additional 3 regions (roxi_0008, roxi_0009 and roxi_0010) for years 2019 and 2020 and 10 all regions in 2021. New regions provide a spatial transfer learning challenge in years 2019 and 2020 and a temporal transfer learning challenge in 2021.
To asses the ranking for all the leaderboards, we are using mean CSI metric for defined thresholds: 0.2, 1, 5, 10 and 15. Please refer to this forum post to learn more about the competition metric: https://weather4cast.net/neurips2023/forums/topic/announcing-final-competition-metric/
