GeoConformal Prediction: A Model-Agnostic Framework for Measuring the Uncertainty of Spatial Prediction

Spatial prediction is a fundamental task in geography, providing essential data support for various scenarios. Recent advancements, empowered by the development of geospatial artificial intelligence (GeoAI), have primarily focused on improving prediction accuracy while overlooking reliable measurements of prediction uncertainty. Such measures are crucial for enhancing model trustworthiness and supporting responsible decision-making. To address this issue, we propose a model-agnostic uncertainty assessment method called GeoConformal Prediction (GeoCP). First, a simulation study is conducted to validate the usefulness of GeoCP. Then, we applied GeoCP to two classic spatial prediction cases, spatial regression and spatial interpolation, to evaluate its reliability. For the case of spatial regression, we used XGBoost to predict housing prices, followed by GeoCP to calculate uncertainty. Our results show that GeoCP achieved a coverage rate of 93.67 percent, whereas bootstrapping methods reached a maximum coverage of 81.00 percent after 2,000 runs. We then applied GeoCP for the case of spatial interpolation models. By comparing a GeoAI-based geostatistical model with a traditional geostatistical model (Kriging), we found that the uncertainty obtained from GeoCP aligned closely with the variance in Kriging. Finally, using GeoCP, we analyzed the sources of uncertainty in spatial prediction. We found that explicitly including local features in AI models can significantly reduce prediction uncertainty, especially in areas with strong local dependence. Our findings suggest that GeoCP holds substantial potential not only for geographic knowledge discovery but also for guiding the design of future GeoAI models, paving the way for more reliable and interpretable spatial prediction frameworks. The method is implemented in an open-source Python package named geoconformal.