Statistical versus machine learning-based spatial interpolation of post-processed ensemble weather forecasts
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.
Related
- The influence of ensemble size and composition on the performance of combined real-time COVID-19 forecasts
- Learning joint probabilistic weather forecasts from station observations alone
- Flowz: When Does Learning Forecast Errors Add Value? Conditional Flow Matching for Financial Price Forecasts
- Generative Refinement Learning for Temporal Interpolation in Scientific Machine Learning
- Strictly Replicable Realizable Learning: Unknown Marginals, Finite-Class Interpolation, and the Cost of Shared Randomness
BibTeX
@misc{lakatos2026statistical,
title = {{Statistical versus machine learning-based spatial interpolation of post-processed ensemble weather forecasts}},
author = {Mária Lakatos},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.07512},
eprint = {2609.07512},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.07512},
}
RIS
TY - UNPB
TI - Statistical versus machine learning-based spatial interpolation of post-processed ensemble weather forecasts
AU - Mária Lakatos
PY - 2026
DA - 2026/09/07
PB - arXiv
DO - 10.48550/arxiv.2609.07512
UR - https://arxiv.org/abs/2609.07512
AB - 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.
ER -