Incorporating spatial autocorrelation in dasymetric mapping: A hierarchical Poisson spatial disaggregation regression model

Abstract

The growing demand for spatially detailed population products in various fields continues to rise, as users shift their focus from aggregated areal totals to high-resolution grid estimates. Aggregating demographic data to areas, such as census tracts or block groups, can mask localized heterogeneities within those areas. This paper presents a new pycnophylactic (density-preserving) geospatial model for disaggregating population to high-resolution grids. We describe a Bayesian Hierarchical Poisson Spatial Disaggregation Regression Model (HPSDRM), which incorporates land cover covariates and two levels of spatial autocorrelation. We evaluated the model's predictive ability first with simulation studies, and then by disaggregating census population data for Davidson County, TN, from the census tract-level to a fine grid and comparing predicted to actual block-level population counts. The interpolated population map successfully identified spatial heterogeneities, such as hot- and cold-spots within census tracts. The HPDSRM model out-performed three other types of disaggregation modeling, which suggests the value of incorporating spatial autocorrelation. Based upon this study, HPSDRM has potential for disaggregating other demographic data, such as socioeconomic indicators.