Comparative Evaluation of Prompting and Fine-Tuning for Applying Large Language Models to Grid-Structured Geospatial Data

2025-05-21

AI Summary

This study addresses the limited spatiotemporal reasoning capability of large language models (LLMs) in understanding gridded geospatial data. We systematically compare structured prompt engineering against supervised fine-tuning (SFT). Our method introduces a geospatial-semantic–aware encoding scheme for gridded spatiotemporal data, constructs user-assistant interactive training data, and performs domain-adaptive fine-tuning. Through quantitative experiments—first of their kind—we reveal the fundamental limitations of zero-shot structured prompting on complex spatiotemporal reasoning tasks. We further demonstrate that domain-specific fine-tuning substantially enhances LLMs’ joint modeling capacity for geographic entities, temporal relations, and spatial topologies. On geographic question answering and spatiotemporal inference tasks, the fine-tuned model achieves an average accuracy improvement of 32.7% over zero-shot prompting, with markedly improved generalization and robustness.

Technology Category

Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal ReasoningMachine Learning: Large Multimodal Models (LMMs)Data Mining & Knowledge Management: Mining of Spatial, Temporal or Spatio-Temporal Data

Application Category

Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMsGraph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphsSearch and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information

Abstract

This paper presents a comparative study of large language models (LLMs) in interpreting grid-structured geospatial data. We evaluate the performance of a base model through structured prompting and contrast it with a fine-tuned variant trained on a dataset of user-assistant interactions. Our results highlight the strengths and limitations of zero-shot prompting and demonstrate the benefits of fine-tuning for structured geospatial and temporal reasoning.

Problem

• Compare prompting vs fine-tuning for LLMs on geospatial data
• Evaluate zero-shot performance on grid-structured data interpretation
• Assess fine-tuning benefits for spatiotemporal reasoning tasks

Innovation

• Comparative study of LLMs for geospatial data
• Evaluates structured prompting versus fine-tuning
• Demonstrates benefits of fine-tuning for reasoning