PEACE enhances multimodal large language models (MLLMs) with geologic expertise, enabling accurate interpretation of complex, high-resolution maps. By integrating structured extraction, domain knowledge, and reasoning, it supports critical tasks in disaster risk, resource discovery, and infrastructure planning—turning general AI into a specialized tool for geoscience.
Open PEACE repo
PEACE has been released for research purposes.
PEACE (emPowering gEologic mAp holistiC undErstanding) enhances multimodal large language models (MLLMs) for expert-level geologic map understanding. Geologic maps, which provide critical insights into the structure and composition of Earth’s subsurface and surface, are vital tools in disaster detection, resource exploration, and civil engineering. But their complexity—featuring high-resolution visuals, symbolic representations, and domain-specific knowledge—poses significant challenges for current AI models. General-purpose MLLMs often fall short when interpreting such data due to the intricacies of cartographic generalization and geoscientific reasoning.
To bridge this gap, Microsoft researchers and collaborators introduced GeoMap-Bench, the first benchmark specifically designed to evaluate MLLMs across five capabilities essential to geologic map interpretation: extracting, referring, grounding, reasoning, and analyzing. They also developed GeoMap-Agent, an AI system tailored to these challenges.
GeoMap-Agent is composed of three key modules:
1. Hierarchical Information Extraction (HIE) for parsing structured content from complex maps
2. Domain Knowledge Injection (DKI) for embedding geological expertise
3. Prompt-enhanced Question Answering (PEQA) for improved interpretive and reasoning capabilities
Together, these modules enable GeoMap-Agent to outperform existing models with superior accuracy and depth in geologic tasks.
Rather than modifying MLLMs themselves, PEACE builds intelligent, domain-specific layers on top of them, turning general models into specialized agents capable of handling real-world geoscientific problems. This advancement marks a critical step toward applying AI in Earth science, empowering faster, more accurate geological assessments for both researchers and practitioners.
Open PEACE repo
PEACE has been released for research purposes.
