GeoMMBench and GeoMMAgent: Toward Expert-Level Multimodal Intelligence in Geoscience and Remote Sensing

arXiv cs.CV / 4/13/2026

💬 OpinionSignals & Early TrendsIdeas & Deep AnalysisModels & Research

共有:

Key Points

The paper introduces GeoMMBench, a comprehensive multimodal QA benchmark designed to better evaluate geoscience and remote sensing (RS) capabilities across disciplines, sensors, and task types.
Using GeoMMBench, the authors test 36 open-source and proprietary multimodal large language models and identify recurring weaknesses in domain knowledge, perceptual grounding, and reasoning.
To address these limitations, they propose GeoMMAgent, a multi-agent framework that combines retrieval, perception, and reasoning while leveraging domain-specific RS models and tools.
Experiments show GeoMMAgent performs significantly better than standalone LLMs, highlighting the value of tool-augmented, agentic approaches for complex geospatial interpretation.
The work positions the benchmark and agent framework as a pathway toward more rigorous and expert-level multimodal intelligence in geoscience and RS workflows.

Abstract

Recent advances in multimodal large language models (MLLMs) have accelerated progress in domain-oriented AI, yet their development in geoscience and remote sensing (RS) remains constrained by distinctive challenges: wide-ranging disciplinary knowledge, heterogeneous sensor modalities, and a fragmented spectrum of tasks. To bridge these gaps, we introduce GeoMMBench, a comprehensive multimodal question-answering benchmark covering diverse RS disciplines, sensors, and tasks, enabling broader and more rigorous evaluation than prior benchmarks. Using GeoMMBench, we assess 36 open-source and proprietary large language models, uncovering systematic deficiencies in domain knowledge, perceptual grounding, and reasoning--capabilities essential for expert-level geospatial interpretation. Beyond evaluation, we propose GeoMMAgent, a multi-agent framework that strategically integrates retrieval, perception, and reasoning through domain-specific RS models and tools. Extensive experimental results demonstrate that GeoMMAgent significantly outperforms standalone LLMs, underscoring the importance of tool-augmented agents for dynamically tackling complex geoscience and RS challenges.