Comparing methods to improve weather forecasts at missing stations
Statistical versus machine learning-based spatial interpolation of post-processed ensemble weather forecasts
Machine Learning
Summary
Weather forecasts can be better when using several predictions together and adjusting them statistically. But it's hard to make good predictions for places without weather stations. The authors compared different mathematical and machine learning methods to improve temperature and wind forecasts in Germany, including new combinations of forecasts that consider altitude. They found that no single method was best for every situation, but their new altitude-aware approach gave a small improvement for places without measurements.
ensemble weather forecastsstatistical post-processingmachine learningEMOSdistributional regressiontransformersgraph neural networksspatial interpolationaltitude-aware linear pool
Authors
Mária Lakatos
Abstract
Statistical post-processing improves ensemble weather forecasts, but generating calibrated predictions at locations without observations remains challenging. This study compares statistical and machine-learning-based methods for post-processing ECMWF 2-m temperature and 10-m wind speed forecasts at observed and unobserved stations in Germany. We consider EMOS-based approaches, distributional regression networks, Transformers, and graph neural networks under both limited and extended predictor settings. For temperature, we also investigate linear forecast combinations and propose an altitude-aware linear pool (ALP). The results show that post-processing improves upon the raw ensemble in most settings, but no single method performs best across all variables, station groups, and evaluation metrics. The proposed ALP provides a small but significant improvement over the standard linear pool at unobserved locations.