Utilizing a Geospatial Foundation Model for Coastline Delineation in Small Sandy Islands (Papers Track)
Tishya Chhabra (MIT); Walter Zesk (Self Assembly Lab at MIT); Skylar Tibbits (Self Assembly Lab at MIT)
Abstract
We present an initial evaluation of NASA and IBM’s Prithvi-EO-2.0 geospatial foundation model for the task of shoreline delineation on small sandy islands using satellite imagery. We curated and labeled a dataset of 225 multispectral Sentinel-2 images of two Maldivian islands, which we publicly release, and fine-tuned both the 300M and 600M parameter versions of Prithvi-EO-2.0 on training subsets ranging from 5 to 181 images. Our experiments show that even with as few as 5 training images, the models achieve high segmentation performance (F1 > 0.94, IoU > 0.79). We observed minimal differences between the 300M and 600M models, suggesting that the larger model’s additional computational cost may not be justified for this task. The results demonstrate the strong transfer learning capability of Prithvi, underscoring the potential of such models to support coastal monitoring in data-poor regions.