SHRUG-FM: Reliability-Aware Foundation Models for Earth Observation

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Gonzalez-Calabuig, Maria, Cohrs, Kai-Hendrik, Nedungadi, Vishal, Osika, Zuzanna, Cartuyvels, Ruben, Knoblauch, Steffen, Massant, Joppe, Nath, Shruti, Ebel, Patrick, Sitokonstantinou, Vasileios
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908979223855104
author Gonzalez-Calabuig, Maria
Cohrs, Kai-Hendrik
Nedungadi, Vishal
Osika, Zuzanna
Cartuyvels, Ruben
Knoblauch, Steffen
Massant, Joppe
Nath, Shruti
Ebel, Patrick
Sitokonstantinou, Vasileios
author_facet Gonzalez-Calabuig, Maria
Cohrs, Kai-Hendrik
Nedungadi, Vishal
Osika, Zuzanna
Cartuyvels, Ruben
Knoblauch, Steffen
Massant, Joppe
Nath, Shruti
Ebel, Patrick
Sitokonstantinou, Vasileios
contents Geospatial foundation models (GFMs) for Earth observation often fail to perform reliably in environments underrepresented during pretraining. We introduce SHRUG-FM, a framework for reliability-aware prediction that enables GFMs to identify and abstain from likely failures. Our approach integrates three complementary signals: geophysical out-of-distribution (OOD) detection in the input space, OOD detection in the embedding space, and task-specific predictive uncertainty. We evaluate SHRUG-FM across three high-stakes rapid-mapping tasks: burn scar segmentation, flood mapping, and landslide detection. Our results show that SHRUG-FM consistently reduces prediction risk on retained samples, outperforming established single-signal baselines like predictive entropy. Crucially, by utilizing a shallow "glass-box" decision tree for signal fusion, SHRUG-FM provides interpretable abstention thresholds. It builds a pathway toward safer and more interpretable deployment of GFMs in climate-sensitive applications, bridging the gap between benchmark performance and real-world reliability.
format Preprint
id arxiv_https___arxiv_org_abs_2511_10370
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle SHRUG-FM: Reliability-Aware Foundation Models for Earth Observation
Gonzalez-Calabuig, Maria
Cohrs, Kai-Hendrik
Nedungadi, Vishal
Osika, Zuzanna
Cartuyvels, Ruben
Knoblauch, Steffen
Massant, Joppe
Nath, Shruti
Ebel, Patrick
Sitokonstantinou, Vasileios
Computer Vision and Pattern Recognition
Artificial Intelligence
Machine Learning
Geospatial foundation models (GFMs) for Earth observation often fail to perform reliably in environments underrepresented during pretraining. We introduce SHRUG-FM, a framework for reliability-aware prediction that enables GFMs to identify and abstain from likely failures. Our approach integrates three complementary signals: geophysical out-of-distribution (OOD) detection in the input space, OOD detection in the embedding space, and task-specific predictive uncertainty. We evaluate SHRUG-FM across three high-stakes rapid-mapping tasks: burn scar segmentation, flood mapping, and landslide detection. Our results show that SHRUG-FM consistently reduces prediction risk on retained samples, outperforming established single-signal baselines like predictive entropy. Crucially, by utilizing a shallow "glass-box" decision tree for signal fusion, SHRUG-FM provides interpretable abstention thresholds. It builds a pathway toward safer and more interpretable deployment of GFMs in climate-sensitive applications, bridging the gap between benchmark performance and real-world reliability.
title SHRUG-FM: Reliability-Aware Foundation Models for Earth Observation
topic Computer Vision and Pattern Recognition
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2511.10370