TerraBench: Can Agents Reason Over Heterogeneous Earth-System Data?

2026-06-11

AI Summary

This work addresses the challenge of integrating heterogeneous Earth science data—including gridded datasets, remote sensing imagery, geospatial information, and simulation outputs—into coherent decision-making workflows, a task current large language models struggle to support due to their inability to directly process high-dimensional Earth system data or enable interactive reasoning. To bridge this gap, the study introduces TerraBench, a comprehensive benchmark, and TerraAgent, a novel framework that unifies Earth observation, gridded data analysis, GIS reasoning, and simulation through executable interfaces. Leveraging a ReAct-style agent architecture, TerraAgent orchestrates large language models with a scientific toolchain encompassing environmental retrieval, geospatial processing, simulators, and traceable computation. The benchmark comprises 403 complex tasks across three tracks and eight application domains, totaling 24,500 validation steps, and introduces process-level tool-use metrics and tolerance-aware numerical scoring, demonstrating that robust Earth science agents must precisely coordinate heterogeneous workflows, parameterize tools, and maintain provenance of derived products.

Abstract

Climate and environmental decision-making increasingly requires reasoning across heterogeneous inputs, including gridded physical data, satellite imagery, geospatial context, and simulator outputs. Weather and climate foundation models can forecast well, but do not reason interactively in language, while large language models (LLMs) reason in language but cannot operate directly on high-dimensional Earth-system data. As a result, real scientific workflows in Earth-science remain underserved. We introduce TerraBench, a benchmark for grounded Earth-science reasoning, built on TerraAgent, a ReAct-style executable framework that interleaves reasoning, tool calls, and observations to couple LLM planning with scientific tools for environmental retrieval, geospatial processing, simulation, and artifact-backed computation. TerraBench unifies analysis of Earth observation imagery, gridded data, GIS reasoning and simulation in a single executable interface, whereas prior benchmarks isolate these capabilities into narrow individual tasks. It is also the first in this space to pair process-level tool-use metrics with tolerance-aware numeric scoring. The benchmark comprises 403 extensive agentic tasks across three tracks (Fundamentals, Simulator-Grounded, and Document-Grounded Verification) and eight application domains with 24,500 verified execution steps. These results indicate that reliable Earth-science agents must go beyond tool access to coordinate heterogeneous workflows, parameterize tools precisely, and preserve artifact provenance.

Problem

• Earth-system data
• heterogeneous data
• scientific reasoning
• agent benchmark
• geospatial context

Innovation

• Earth-system reasoning
• heterogeneous data integration
• agentic benchmarking
• tool-augmented LLMs
• executable scientific workflows