Weather4cast 2024 Competition #2 - Event Prediction

We are proud to present the Weather4cast 2024 Competitions.

In this Competition Challenge, you are asked to pre-train a generic model, which is then finetuned to individual Downstream Tasks.

This page here is for Downstream Task #2 – the now-casting prediction of high-precipitation events that start within 4 hours anywhere in the region.

Leaderboard and Submission links will shortly be provided in the menu on the right.

High-precipitation events are to be characterized by their predicted intensity, duration, start time, location, and size:

• maxPrec: maximum precipitation, in mm/h/m²
• duration: min. 45 mins, max 48h, given in time slots, 1 slot = 15 mins
• start.offset: start time offset in slots, 0=first slot
• mid.diag: a measure of the size (not area!) of the event at mid time, in hi-res OPERA rain radar pixels; mid time is min.time + duration/2.
• mid.x, mid.y: coordinates of the center of the event at mid time, in hi-res OPERA rain radar pixel resolution

For finetuning your generic model, we provide the W4C24-Task-2 training dataset as zip archive of CSV text files containing lists of events characterized by the values above. The detailed description of the training data, along with examples, can be found in our forum post.

Input EUMETSAT satellite band data for the context region for which the prediction is made including satellite data for 1h of input will be provided. You are then asked to predict 5 events starting in the subsequent 4 hours.

As in Task 1, predictions are evaluated using a Continuous Ranked Probability Score (CRPS) that support (but do not require) probabilistic forecasts.

Predictions for intensity, duration, start time, and size will be assessed by the log-error log(pred/truth), for an unbiased test of relative accuracies. The event start time will be offset by the start of the 4h prediction window time0 for this purpose, testing log[ (pred.time–time0) / (true.time–time0) ], where this offset adjusted event start time can range from a minimum of 15 minutes up to a maximum of 4h. The location will be assessed by the generalized log of the absolute error scaled by the true event size. The event score is then the root mean squared error of all these, with the final score averaging over events.