Probabilistic & ensemble forecasting
Skillful joint probabilistic weather forecasting from marginals
In one sentence. FGN generates ensembles through learned model perturbations trained on per-location CRPS, achieving state-of-the-art probabilistic skill and skillful ensemble tropical cyclone tracks.
Abstract
Machine learning (ML)-based weather models have rapidly risen to prominence due to their greater accuracy and speed than traditional forecasts based on numerical weather prediction (NWP), recently outperforming traditional ensembles in global probabilistic weather forecasting. This paper presents FGN, a simple, scalable and flexible modeling approach which significantly outperforms the current state-of-the-art models. FGN generates ensembles via learned model-perturbations with an ensemble of appropriately constrained models. It is trained directly to minimize the continuous rank probability score (CRPS) of per-location forecasts. It produces state-of-the-art ensemble forecasts as measured by a range of deterministic and probabilistic metrics, makes skillful ensemble tropical cyclone track predictions, and captures joint spatial structure despite being trained only on marginals.
Labels
- Topic
- Probabilistic & ensemble forecasting, Global medium-range NWP
- Type
- New model or method
- Method
- Other
- Time scale
- 3–15 days
- Domain
- Global
- Data
- ERA5 / reanalysis
- Status
- Research
- Identifiers
- arXiv:2506.10772 OpenAlex:W4417355657
- Added
- 2026-10-22
Labels and summary are generated by Claude and reviewed weekly. Wrong? Edit this entry.