arXiv:2511.15658v1 Announce Type: new 
Abstract: Geospatial Foundation Models (GeoFMs) are transforming Earth Observation (EO), but evaluation lacks standardized protocols. GEO-Bench-2 addresses this with a comprehensive framework spanning classification, segmentation, regression, object detection, and instance segmentation across 19 permissively-licensed datasets. We introduce ''capability'' groups to rank models on datasets that share common characteristics (e.g., resolution, bands, temporality). This enables users to identify which models excel in each capability and determine which areas need improvement in future work. To support both fair comparison and methodological innovation, we define a prescriptive yet flexible evaluation protocol. This not only ensures consistency in benchmarking but also facilitates research into model adaptation strategies, a key and open challenge in advancing GeoFMs for downstream tasks.
  Our experiments show that no single model dominates across all tasks, confirming the specificity of the choices made during architecture design and pretraining. While models pretrained on natural images (ConvNext ImageNet, DINO V3) excel on high-resolution tasks, EO-specific models (TerraMind, Prithvi, and Clay) outperform them on multispectral applications such as agriculture and disaster response. These findings demonstrate that optimal model choice depends on task requirements, data modalities, and constraints. This shows that the goal of a single GeoFM model that performs well across all tasks remains open for future research. GEO-Bench-2 enables informed, reproducible GeoFM evaluation tailored to specific use cases. Code, data, and leaderboard for GEO-Bench-2 are publicly released under a permissive license.

يقدم GEO-Bench-2 إطار تقييم شامل لنماذج الأساس الجغرافية (GeoFMs) في مراقبة الأرض (EO). يعالج نقص البروتوكولات القياسية من خلال تغطيته لمهام متنوعة مثل التصنيف، والتجزئة، واكتشاف الكائنات عبر 19 مجموعة بيانات. يقدم الإطار مجموعات 'قدرة' لتصنيف النماذج بناءً على الخصائص المشتركة، مما يسهل المقارنات العادلة والابتكار المنهجي في استراتيجيات تكيف النماذج، وهو أمر حاسم لتقدم GeoFMs.

GEO-Bench-2 presenta un marco de evaluación integral para los Modelos de Fundación Geoespacial (GeoFMs) en la Observación de la Tierra (EO). Aborda la falta de protocolos estandarizados abarcando diversas tareas como clasificación, segmentación y detección de objetos en 19 conjuntos de datos. El marco introduce grupos de 'capacidad' para clasificar modelos según características compartidas, facilitando comparaciones justas e innovación metodológica en estrategias de adaptación de modelos, cruciales para el avance de los GeoFMs.

GEO-Bench-2 présente un cadre d'évaluation complet pour les modèles de fond géospatial (GeoFMs) dans l'observation de la Terre (EO). Il répond au manque de protocoles standardisés en couvrant diverses tâches telles que la classification, la segmentation et la détection d'objets sur 19 ensembles de données. Le cadre introduit des groupes de 'capacité' pour classer les modèles en fonction de caractéristiques communes, facilitant des comparaisons équitables et l'innovation méthodologique dans les stratégies d'adaptation des modèles, cruciales pour faire avancer les GeoFMs.

GEO-Bench-2 introduces a comprehensive evaluation framework for Geospatial Foundation Models (GeoFMs) in Earth Observation (EO). It addresses the lack of standardized protocols by spanning various tasks such as classification, segmentation, and object detection across 19 datasets. The framework introduces 'capability' groups to rank models based on shared characteristics, facilitating fair comparisons and methodological innovation in model adaptation strategies, crucial for advancing GeoFMs.

GEO-Bench-2: From Performance to Capability, Rethinking Evaluation in Geospatial AI

arXiv:2512.02055v1 Announce Type: new 
Abstract: Floods are among the most damaging weather-related hazards, and in 2024, the warmest year on record, extreme flood events affected communities across five continents. Earth observation (EO) satellites provide critical, frequent coverage for mapping inundation, yet operational accuracy depends heavily on labeled datasets and model generalization. Recent Geospatial Foundation Models (GFMs), such as ESA-IBM's TerraMind, offer improved generalizability through large-scale self-supervised pretraining, but their performance on diverse global flood events remains poorly understood.
  We fine-tune TerraMind for flood extent mapping using FloodsNet, a harmonized multimodal dataset containing co-located Sentinel-1 (Synthetic Aperture Radar, SAR data) and Sentinel-2 (optical) imagery for 85 flood events worldwide. We tested four configurations (base vs. large models; frozen vs. unfrozen backbones) and compared against the TerraMind Sen1Floods11 example and a U-Net trained on both FloodsNet and Sen1Floods11. The base-unfrozen configuration provided the best balance of accuracy, precision, and recall at substantially lower computational cost than the large model. The large unfrozen model achieved the highest recall. Models trained on FloodsNet outperformed the Sen1Floods11-trained example in recall with similar overall accuracy. U-Net achieved higher recall than all GFM configurations, though with slightly lower accuracy and precision.
  Our results demonstrate that integrating multimodal optical and SAR data and fine-tuning a GFM can enhance near-real-time flood mapping. This study provides one of the first global-scale evaluations of a GFM for flood segmentation, highlighting both its potential and current limitations for climate adaptation and disaster resilience.

تم استغلال التقدمات الأخيرة في نماذج الأساس الجغرافية متعددة الوسائط المعتمدة على الذكاء الاصطناعي، وخاصة نموذج TerraMind من ESA-IBM، لتحسين رسم خرائط الفيضانات في الوقت شبه الحقيقي على مستوى العالم. يأتي هذا التطور في ضوء الأحداث الشديدة للفيضانات التي أثرت على المجتمعات عبر خمس قارات في عام 2024، وهو العام الأكثر حرارة على الإطلاق. تقوم الدراسة بضبط TerraMind باستخدام FloodsNet، وهو مجموعة بيانات متعددة الوسائط تجمع بين صور Sentinel-1 وSentinel-2 لـ 85 حدث فيضان حول العالم.

Los avances recientes en modelos de fundación geoespacial multimodal basados en IA, particularmente TerraMind de ESA-IBM, se han aprovechado para mejorar la cartografía de inundaciones en tiempo casi real a nivel global. Este desarrollo surge a raíz de eventos de inundación extremos que afectaron a comunidades en cinco continentes en 2024, el año más cálido registrado. El estudio ajusta TerraMind utilizando FloodsNet, un conjunto de datos multimodal que combina imágenes de Sentinel-1 y Sentinel-2 para 85 eventos de inundación en todo el mundo.

Des avancées récentes dans les modèles de fondation géospatiale multimodale basés sur l'IA, en particulier TerraMind d'ESA-IBM, ont été exploitées pour améliorer la cartographie des inondations en temps quasi réel à l'échelle mondiale. Ce développement intervient à la suite d'événements d'inondation extrêmes touchant des communautés sur cinq continents en 2024, l'année la plus chaude jamais enregistrée. L'étude ajuste TerraMind en utilisant FloodsNet, un ensemble de données multimodal combinant des images Sentinel-1 et Sentinel-2 pour 85 événements d'inondation dans le monde.

Recent advancements in AI multimodal geospatial foundation models, particularly ESA-IBM's TerraMind, have been leveraged for enhanced near-real-time flood mapping globally. This development comes in light of extreme flood events impacting communities across five continents in 2024, the warmest year on record. The study fine-tunes TerraMind using FloodsNet, a multimodal dataset combining Sentinel-1 and Sentinel-2 imagery for 85 flood events worldwide.

GEO-Bench-2: From Performance to Capability, Rethinking Evaluation in Geospatial AI

Was this article worth reading? Share it

FETCH HIVE

Cometapi-e0d0fd

Guidejar-4eb95b