Data-efficient flood depth prediction through domain-aware coreset selection and tabular foundation models

In three linesDomain-aware coreset selection pipeline for near-real-time flood depth prediction. Uses tabular foundation model conditioned at inference time. Achieves R²=0.663 across nine Houston-area watersheds with 0.7% of training data, without per-watershed retraining.

Summary generated by Claude — human-verified