Toward Trustworthy Short-Range Forecasts with AFNO: From Skill Metrics to Conservation Checks

TL;DR

This paper explores the use of AFNO to improve short-range weather forecasts, focusing on skill metrics and conservation checks to ensure dynamical consistency in data-driven models.

Contribution

It introduces a novel approach combining AFNO with conservation checks to enhance the reliability of data-driven weather forecasts.

Findings

AFNO improves forecast skill metrics.

Conservation checks ensure dynamical consistency.

Enhanced models outperform traditional methods in short-range predictions.

Abstract

Data driven weather models now approach traditional numerical weather prediction (NWP) skill at short to medium lead times, but their dynamical consistency during autoregressive rollout remains uncertain.