Spatial uncertainty refers to the inherent unpredictability in geospatial predictions — the gap between what a model predicts and the true values at unobserved locations. Unlike traditional ML settings, spatial data exhibits spatial autocorrelation: nearby locations tend to have similar values, so prediction quality varies across geographic space.
Conformal prediction (CP) provides a distribution-free framework for quantifying this uncertainty. It wraps any prediction model to produce prediction intervals with guaranteed coverage probability, without assuming a specific data distribution. The spatial CP variants in this tool extend CP to account for geographic proximity, ensuring locally-calibrated prediction intervals.