Modified Projection of Synthesized Households on Buildings for Real-Scale Social Simulation

This study refines a geospatial attribution method for synthetic microdata used in real-scale social simulation by improving how synthesized households are projected onto buildings. A conventional approach first allocates municipality-level synthetic households to chome/aza (small-area units) to match published tabulations and then uniformly assigns households to buildings within each area. This can produce unrealistic outcomes in mixed neighborhoods, such as assigning many households to detached houses. We introduce the census “basic unit district” (the smallest census area) as an intermediate spatial unit and extend the workflow to a three-stage allocation: (1) allocation to chome/aza, (2) re-allocation to basic unit districts within each chome/aza using simulated annealing, and (3) assignment to buildings within each basic unit district. Because polygon boundaries of basic unit districts are not publicly available, we estimate basic-unit polygons by extracting block polygons from road-edge data in Japan’s Fundamental Geospatial Data and assigning them using basic-unit centroid points and proximity rules; buildings are then labeled with basic-unit identifiers. Experiments in Takatsuki City, Osaka (≈145k households; ≈330k people) show that the proposed method mitigates excessive household concentration on detached-house blocks, allocates more households to apartment-dominant blocks, and avoids assigning households to uninhabited blocks. Future work includes validating and improving the estimated basic-unit boundaries and allocation accuracy.