A New Method for Detecting and Explaining Local Irregularities in Spatial Synthetic Data

Spatial synthetic populations are increasingly used in agent‐based models, spatial microsimulation, and other forms of place‐based analysis. Their usefulness, however, depends not only on reproducing aggregate patterns but also on maintaining plausible local distributions. This creates a difficult validation problem when independent benchmarks are limited and validation cannot proceed beyond checking the technical implementation of model constraints. We develop a moment‐based validation framework for detecting and interpreting local irregularities in spatial synthetic data. Local distributions are characterized through their first four moments and we combine hierarchical clustering with SHapley Additive exPlanations to identify and explain areas whose modeled distributions depart from expected local patterns and contextual variables. The framework is demonstrated using modeled salary distributions from the Synthetic Population Catalyst in West Yorkshire, a setting in which fine‐scale synthetic populations are assembled from sources observed at coarser spatial and substantive resolutions. The application identifies locally unusual areas, shows how these can be interpreted against contextual characteristics and local knowledge, and illustrates how the framework can support judgments about whether synthetic data are fit for intended use. More broadly, the article argues that local coherence should be treated as a central criterion in the validation of spatial synthetic populations.