Scholarlink

<p>This study evaluates the role of Large Language Models (LLMs) in supporting spatial interpolation workflows within Geographic Information Systems (GIS), particularly under conditions of incomplete geospatial data. Using a comparative experimental framework, LLM-assisted implementations of Inverse Distance Weighting (IDW), Kriging, and Spline interpolation were assessed against conventional GIS approaches using two spatial datasets: precipitation observations and urban mobility indicators derived from taxi GPS trajectories. Model performance was evaluated using standard accuracy metrics (RMSE, MAE, and R²), alongside qualitative criteria including interpretability, workflow completeness, and computational efficiency. Results indicate that Kriging consistently achieves the highest predictive accuracy and produces smooth, spatially coherent surfaces, while IDW captures local variability but remains sensitive to parameter selection and uneven data density. Spline interpolation generates visually smooth surfaces but may obscure finer spatial patterns. Among LLM platforms, ChatGPT produced the most consistent and interpretable R-based workflows, Gemini AI demonstrated robust handling of dense datasets, and Microsoft Copilot was efficient for deterministic methods but required additional refinement for Kriging and Spline. Overall, LLMs reduce technical barriers, enhance reproducibility, and support exploratory spatial analysis, suggesting their potential to complement traditional geostatistical methods in Spatial Information Science.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Integrating large language models into spatial analysis: experimental insights from GIS interpolation

  • 期刊论文

  • 发表日期：

    2026年1月26日

查看全文
  • Rashima Kachari

摘要

This study evaluates the role of Large Language Models (LLMs) in supporting spatial interpolation workflows within Geographic Information Systems (GIS), particularly under conditions of incomplete geospatial data. Using a comparative experimental framework, LLM-assisted implementations of Inverse Distance Weighting (IDW), Kriging, and Spline interpolation were assessed against conventional GIS approaches using two spatial datasets: precipitation observations and urban mobility indicators derived from taxi GPS trajectories. Model performance was evaluated using standard accuracy metrics (RMSE, MAE, and R²), alongside qualitative criteria including interpretability, workflow completeness, and computational efficiency. Results indicate that Kriging consistently achieves the highest predictive accuracy and produces smooth, spatially coherent surfaces, while IDW captures local variability but remains sensitive to parameter selection and uneven data density. Spline interpolation generates visually smooth surfaces but may obscure finer spatial patterns. Among LLM platforms, ChatGPT produced the most consistent and interpretable R-based workflows, Gemini AI demonstrated robust handling of dense datasets, and Microsoft Copilot was efficient for deterministic methods but required additional refinement for Kriging and Spline. Overall, LLMs reduce technical barriers, enhance reproducibility, and support exploratory spatial analysis, suggesting their potential to complement traditional geostatistical methods in Spatial Information Science.