A multiscale masking method for point geographic data

Abstract

This research introduces and tests a new masking technique for sensitive point-based geospatial data. The goal of masking is to allow data to be used for general measurements of spatial distributions, but to not permit the matching of points with their true underlying observations, usually street addresses, in order to protect identity and privacy. The proposed method first converts coordinates to the Military Grid Reference System (MGRS), then uses digit switching to mask coordinates. This permits encryption at five spatial levels of precision, equivalent to 1, 10, 100, 1000, and 10,000 m. Of the 1.3M possible digit switching combinations, one is chosen that minimizes the difference between the aggregate descriptive spatial statistics for the point set as a whole between the masked and the unmasked data. Most heavily weighted in the selection is the nearest neighbor statistic, a measure of clustering in the distribution. The masking method was implemented for point sets within a single MGRS cell square using NGA’s GEOTRANS software and new custom C language code, with both forward and inverse algorithms. The forward masking saves the translation code in a separate file without which the inversion is practically impossible, meaning that the masked data can be offered publicly. Four test point distributions were used to show the method in action. The method appears to offer new possibilities for the protection of sensitive geospatial data.