Spectral Gaps and Spatial Priors: Studying Hyperspectral Downstream Adaptation Using TerraMind

Julia Anna Leonardi, Johannes Jakubik, Paolo Fraccaro, Maria Antonia Brovelli

TL;DR: In this paper, we explore the transferability of a multimodal Geospatial Foundation Model, pretrained without hyperspectral data, to hyperspectral applications via channel adaptation strategies.

Abstract: Geospatial Foundation Models (GFMs) typically lack native support for Hyperspectral Imaging (HSI) due to the complexity and sheer size of high-dimensional spectral data. This study investigates the adaptability of TerraMind, a multimodal GFM, to HSI downstream tasks \emph{without} HSI-specific pretraining by comparing two channel adaptation strategies: Naive Band Selection and physics-aware Spectral Response Function (SRF) grouping. Our results confirm the general superiority of HSI-native architectures, though TerraMind demonstrates robust adaptability to spectral tasks through simple band selection. By establishing this baseline, we underscore the necessity of developing native spectral tokenization for future multimodal GFMs.

Submission Number: 46