SoftSEEPS improves ML-based precipitation forecasting
Authors developed SoftSEEPS, a differentiable approximation of the SEEPS score, allowing machine learning models to forecast precipitation directly by using it as a training loss. This enables training a decoder for precipitation on latent space data from pre-trained low-resolution models. Results show that combining SoftSEEPS with standard error metrics yields marginal trade-offs while improving forecasting accuracy.