LEPA: Learning Geometric Equivariance in Satellite Remote Sensing Data with a Predictive Architecture
This work provides a method to accurately adjust precomputed geospatial embeddings for geometric mismatches, which is crucial for Earth observation applications using large-scale satellite data, improving data utility without re-encoding.
The paper addresses the issue of geometric mismatches in precomputed embeddings for satellite remote sensing data, where standard latent-space interpolation performs poorly (MRR below 0.2). They propose LEPA, a Learned Equivariance-Predicting Architecture, which achieves an MRR over 0.8 by directly predicting transformed embeddings conditioned on geometric augmentations.
Geospatial foundation models provide precomputed embeddings that serve as compact feature vectors for large-scale satellite remote sensing data. While these embeddings can reduce data-transfer bottlenecks and computational costs, Earth observation (EO) applications can still face geometric mismatches between user-defined areas of interest and the fixed precomputed embedding grid. Standard latent-space interpolation is unreliable in this setting because the embedding manifold is highly non-convex, yielding representations that do not correspond to realistic inputs. We verify this using Prithvi-EO-2.0 to understand the shortcomings of interpolation applied to patch embeddings. As a substitute, we propose a Learned Equivariance-Predicting Architecture (LEPA). Instead of averaging vectors, LEPA conditions a predictor on geometric augmentations to directly predict the transformed embedding. We evaluate LEPA on NASA/USGS Harmonized Landsat-Sentinel (HLS) imagery and ImageNet-1k. Experiments show that standard interpolation achieves a mean reciprocal rank (MRR) below 0.2, whereas LEPA increases MRR to over 0.8, enabling accurate geometric adjustment without re-encoding.