Learning joint probabilistic weather forecasts from station observations alone
Abstract
Assessing compound weather risks requires forecasts representing dependence between variables. CLARA (Calibrated Advection-Routing Attention) learns joint Gaussian predictive distributions of five surface variables from station observations alone, without numerical weather prediction or reanalysis; the approximately 28,000-parameter model supports CPU training and prediction. Across six multi-year folds on 96 stations, its lead-mean energy score is 4.9% lower than that of a learned comparator with matched temporal inputs (4.7% with a similar parameter count) and 11-65% lower than those of statistical baselines. Holding marginal variances fixed, removing learned correlations worsens joint negative log-likelihood by 1.0-2.8 nats per station. A covariance-scale estimator, proved consistent under stated assumptions, improves short-lead calibration but over-corrects at long leads. Synthetic interventions show an attention-bias coefficient alone does not measure forecast influence. Retrained in ten regions on six continents, CLARA outperforms persistence in all 60 multi-year region-lead comparisons and a similarly sized learned model in 57 of 60.
Related
- Improving global precipitation forecasts with an AI weather model trained on satellite observations
- A Station-Based Evaluation of Machine Learning-based Weather Forecasting Models in Northern Norway
- Butterfly Effect and the Kinetic Energy Cascade in Probabilistic Machine Learning Weather Prediction Models
- Station-trained air-quality forecasts without regional emission inventories
- Can we create a `race to the top' for weather forecasts to inform smallholder farmer decisions?
BibTeX
@misc{yi2026learning,
title = {{Learning joint probabilistic weather forecasts from station observations alone}},
author = {Chaeyeon Yi and Yun Am Seo},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.09898},
eprint = {2610.09898},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.09898},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Learning joint probabilistic weather forecasts from station observations alone
AU - Chaeyeon Yi
AU - Yun Am Seo
PY - 2026
DA - 2026/10/07
PB - arXiv
DO - 10.48550/arxiv.2610.09898
UR - https://arxiv.org/abs/2610.09898
AB - Assessing compound weather risks requires forecasts representing dependence between variables. CLARA (Calibrated Advection-Routing Attention) learns joint Gaussian predictive distributions of five surface variables from station observations alone, without numerical weather prediction or reanalysis; the approximately 28,000-parameter model supports CPU training and prediction. Across six multi-year folds on 96 stations, its lead-mean energy score is 4.9% lower than that of a learned comparator with matched temporal inputs (4.7% with a similar parameter count) and 11-65% lower than those of statistical baselines. Holding marginal variances fixed, removing learned correlations worsens joint negative log-likelihood by 1.0-2.8 nats per station. A covariance-scale estimator, proved consistent under stated assumptions, improves short-lead calibration but over-corrects at long leads. Synthetic interventions show an attention-bias coefficient alone does not measure forecast influence. Retrained in ten regions on six continents, CLARA outperforms persistence in all 60 multi-year region-lead comparisons and a similarly sized learned model in 57 of 60.
ER -