Search papers, labs, and topics across Lattice.
This paper investigates shortcut learning in geospatial foundation models (GFMs) for cross-disaster damage assessment by evaluating generalization to unseen disaster types. Holding out entire disasters significantly degrades performance, revealing reliance on non-causal cues like pre-event imagery and building counts, which achieve high accuracy independently. Feature transformations, particularly supervised contrastive learning and PCA, demonstrate efficacy in improving robustness to novel disaster scenarios.
Geospatial foundation models for disaster assessment exhibit severe shortcut learning, with performance plummeting from 0.80+ to 0.63 accuracy when evaluated on unseen disaster types, and simple building counts alone achieve comparable performance.
Geospatial foundation models can support disaster damage assessment, but high accuracy on familiar disasters may hide shortcut learning: prediction from event or scene characteristics rather than evidence that transfers to a new disaster. We study embeddings from TerraMind, Prithvi-EO-2.0, DOFA, and DINOv3 on the multi-disaster datasets xView2 and BRIGHT. Holding out an entire disaster, instead of randomly splitting image tiles, reduces mean balanced accuracy from 0.807 to 0.638 on xView2 and from 0.818 to 0.568 on BRIGHT. Four sets of tests examine what information supports prediction: event and acquisition context, before/after image inputs, building count, and correct image pairing. Pre-event images remain predictive, and a classifier given only the annotated number of buildings reaches balanced accuracy of 0.629 and 0.653. Combining before/after features performs best on average on xView2, but does not improve over pre-only features on BRIGHT. Shuffling image pairs within the same disaster and damage class causes modest losses. Motivated by these findings, we test whether feature transformations learned only from training disasters improve generalization to unseen disasters. These transformations improve performance, with supervised contrastive learning performing best on xView2 and PCA on BRIGHT. Our results highlight the importance of evaluation on unseen domains for assessing model robustness and show that simple feature-level transformations can improve generalization. Code and experiments can be found at https://github.com/alishibli97/gfm-shortcut-learning.
