Towards Scalable and Generalizable Earth Observation Data Mining via Foundation Model Composition

TL;DR

This paper explores combining pretrained foundation models from remote sensing and general vision to enhance Earth Observation data mining, demonstrating that ensembling smaller models can match larger models' performance with less resource use.

Contribution

It investigates the effectiveness of reusing and combining existing pretrained models for Earth Observation tasks, highlighting feature ensembling and knowledge distillation as practical strategies.

Findings

Feature ensembling of smaller models matches larger models' performance.

Ensembling reduces training time and computational resources.

Knowledge distillation creates compact models retaining ensemble strengths.

Abstract

Foundation models are rapidly transforming Earth Observation data mining by enabling generalizable and scalable solutions for key tasks such as scene classification and semantic segmentation. While most efforts in the geospatial domain have focused on developing large models trained from scratch using massive Earth Observation datasets, an alternative strategy that remains underexplored is the reuse and combination of existing pretrained models. In this study, we investigate whether foundation models pretrained on remote sensing and general vision datasets can be effectively combined to improve performance across a diverse set of key Earth Observation tasks. Using the GEO-Bench benchmark, we evaluate several prominent models, including Prithvi, Hiera, and DOFA, on eleven datasets covering a range of spatial resolutions, sensor modalities, and task types. The results show that feature-level ensembling of smaller pretrained models can match or exceed the performance of much larger models, while requiring less training time and computational resources. Moreover, the study highlights the potential of applying knowledge distillation to transfer the strengths of ensembles into more compact models, offering a practical path for deploying foundation models in real-world Earth Observation applications.