Securing Sensitive Data: A Synthetic Data Framework for Protecting Geographic Coordinates

Abstract

High-resolution spatial data are critical for environmental and epidemiological research yet present significant privacy risks when released at the individual level. Traditional protection methods, such as aggregation or masking, often compromise data utility or introduce unrealistic spatial patterns. We propose a robust framework for generating fully synthetic geographic coordinates that balances privacy and analytical validity by modeling the spatial intensity through observed covariates. We extend existing methodology by incorporating continuous attributes into the model and strictly enforcing restricted areas (e.g., uninhabitable zones). Furthermore, we address the limitations of standard risk metrics in multimodal spatial distributions by proposing extended distance-based risk measures that assess identification risk within probabilistic neighborhoods. We evaluate the framework through simulation studies and demonstrate its practical utility by generating synthetic georeferenced COVID-19 notifications from Montes Claros, Brazil. Results demonstrate that our method preserves the spatial intensity structure of the data while providing a secure mechanism for sharing sensitive microdata.