A new arXiv preprint, EarthAlign, proposes learning language-aligned representations of Earth system data. The abstract states that interpreting Earth system conditions is central to understanding environmental risks and informing decisions across weather- and climate-sensitive sectors.

Existing multimodal weather systems, the abstract notes, primarily target task-specific objectives. EarthAlign appears to take a different approach by aligning Earth system data with language, although the abstract does not provide details on the model architecture or training procedure.

Because this is a preprint, the work has not yet been peer-reviewed. The abstract also gives limited information about evaluation, so the practical benefits of the method remain to be demonstrated.