A multi-scale loss formulation for learning a probabilistic model with proper score optimisation
By: Simon Lang, Martin Leutbecher, Pedro Maciel
Potential Business Impact:
Improves weather forecasts by seeing small details.
We assess the impact of a multi-scale loss formulation for training probabilistic machine-learned weather forecasting models. The multi-scale loss is tested in AIFS-CRPS, a machine-learned weather forecasting model developed at the European Centre for Medium-Range Weather Forecasts (ECMWF). AIFS-CRPS is trained by directly optimising the almost fair continuous ranked probability score (afCRPS). The multi-scale loss better constrains small scale variability without negatively impacting forecast skill. This opens up promising directions for future work in scale-aware model training.
Similar Papers
A Probabilistic U-Net Approach to Downscaling Climate Simulations
Machine Learning (CS)
Makes climate predictions more detailed and accurate.
Trajectory learning for ensemble forecasts via the continuous ranked probability score: a Lorenz '96 case study
Numerical Analysis
Improves weather forecasts by learning from past predictions.
PIMRL: Physics-Informed Multi-Scale Recurrent Learning for Spatiotemporal Prediction
Machine Learning (CS)
Predicts weather and science events more accurately.