GeoForge: Non-Parametric Self-Evolving Agents for Earth-Observation Reasoning

2026-08-11

AI Summary

This work addresses the challenges faced by Earth observation agents in constructing scientific tool workflows—namely, sensor semantics, product dependencies, spatiotemporal compatibility, and parameter constraints—which hinder existing methods from efficiently distilling heterogeneous task trajectories into reusable knowledge. To overcome this, the authors propose GeoForge, a training-free, self-evolving framework that models execution states in a non-parametric manner to transform historical trajectories into structured knowledge. GeoForge introduces a novel tripartite memory mechanism comprising workflow graph memory, action-level experience, and adaptive skill standard operating procedures, integrated with context-aware operational constraints and task-conditioned retrieval to enable safe and efficient knowledge distillation and reuse. Experiments demonstrate that GeoForge significantly improves task accuracy and tool invocation quality across multiple geospatial benchmarks, effectively reducing planning and reasoning errors in large language models while continuously enhancing agent capabilities without updating the underlying model.

Abstract

Earth observation (EO) agents construct scientifically valid tool workflows and ground their conclusions in current geospatial evidence. This is challenging because EO workflows are constrained by sensing semantics, product dependencies, spatial and temporal compatibility, and parameter requirements. Existing agents often search a broad operation space for each query, while recent self-evolving systems do not fully organize heterogeneous EO trajectories into reusable knowledge across different decision levels. To solve this problem, we present GeoForge, a training-free, self-evolving framework that transforms completed trajectories into a structured nonparametric execution state. GeoForge constrains the operation space according to the sensing context, then retrieves a task-conditioned prior from three complementary memories. Workflow Graph Memory captures global operation order, Action-Level Experiences provide local corrections, and the Adapted Skill Standard Operating Procedure preserves procedural and data constraints. The retrieved prior guides tool execution, while current observations remain the basis of the final answer. After each task, a safety-gated distillation process converts grounded trajectories into reusable execution knowledge for future retrieval. This execution, distillation, and reuse loop improves planning without updating the backbone LLM. Experiments on multiple geospatial benchmarks demonstrate that GeoForge consistently improves both task accuracy and tool-use trajectory quality across diverse LLM backbones, while substantially reducing tool-planning and reasoning errors for most LLMs.

Problem

• Earth observation
• tool workflow
• self-evolving agents
• operation space
• knowledge reuse

Innovation

• non-parametric
• self-evolving agents
• execution state distillation
• tool-use reasoning
• earth observation