Learning joint probabilistic weather forecasts from station observations alone
CLARA learns joint probabilistic surface-weather forecasts from station observations without numerical weather models.
CLARA (Calibrated Advection-Routing Attention), an approximately 28,000-parameter model, learns joint Gaussian forecasts of five surface weather variables from station observations alone, without numerical weather prediction or reanalysis, and supports CPU training. Across six multi-year folds on 96 stations, its lead-mean energy score is 4.9% lower than a matched learned comparator and 11-65% lower than statistical baselines. Holding marginal variances fixed, removing learned correlations worsens joint negative log-likelihood by 1.0-2.8 nats per station. Retrained in ten regions on six continents, it outperforms persistence in all 60 comparisons and a similarly sized learned model in 57 of 60.
- About 28,000 parameters and trainable on CPU
- Energy score 4.9% better than a matched learned comparator
- Removing correlations worsens likelihood by 1.0-2.8 nats
- Beats persistence in all 60 region-lead comparisons
From the original article
1 min read at arxiv.orgAssessing 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.
By Chaeyeon Yi · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.