High resolution, multi-year compatible dasymetric models of US population

Abstract

GIScience 2016 Short Paper Proceedings High Resolution, Multi-Year Compatible Dasymetric Models of US Population A. Dmowska 1,2 , T. F. Stepinski 1 , P. Netzel 1 University of Cincinnati, Cincinnati, OH, 45221, USA Email: {dmowskaa;stepintz;netzelpl}@ucmail.uc.edu Adam Mickiewicz University, Poznan,61-680, Poland Email: dmowska@amu.edu.pl Abstract We developed 30 m resolution grids of the US population in 1990, 2000, and 2010 using a multi-year compatible dasymetric model. These grids are designed to assess population change across the conterminous US at street-level spatial resolution. The model and its novel, computationally efficient implementation in R are described. The grids are available online for interactive exploration and data download using especially developed GeoWeb application SocScape (http://sil.uc.edu/webapps/socscape_usa/). 1. Introduction The goal of this research is to provide an open and convenient access to high resolution, multi-year compatible population data. Such resource is sought after by academic, government, and industry stakeholders as it can provide access to information needed in numerous applications including social and health services, economic development, and planning. Unfortunately, the US Census data, in its original format of aggregated units, is not multi-year compatible, and, except in the densely populated urban areas, is not given at sufficiently high resolution. Boundaries of small Census aggregates, blocks, block groups, and tracts, change from one Census to another making a direct assessment of change impossible without interpolation. In the rural areas large portions of Census blocks are uninhabited with population restricted to small fragments of the blocks resulting in unrealistic estimates of population density. Numerous other shortcomings of aggregated data have also been identified (Sperling, 2012). The solution is to calculate a grid-based model of the US population density. Surprisingly, until now no adequate grid-based model had been available. The Socioeconomic Data and Application Center (SEDAC) provides 1 km resolution (250 m for selected metropolitan areas) demographic grids, but they are constructed using an oversimplified model (areal weighting), have insufficient resolution, and are available only for 1990 and 2000. The 90 m dasymetric model (LandScan–USA) has been developed by the Oak Ridge National Laboratory (Bhaduri et al., 2007) but it is not available, nor is it expected to be in the public domain when and if it becomes available. Since 2014 we have developed (Dmowska and Stepinski, 2014, 2016) a series of dasymetric models covering the entire conterminous US (CONUS). The first generation of our models was the results of sharpening of SEDAC grids into smaller, 90 m cells; these grids were available only for 1990 and 2000. The second generation grid was the result of direct disaggregation of 2010 Census blocks into 30 m grid using land cover (NLCD2011) and land use (NLUD2010, Theobald 2014) as ancillary data. Here we report on direct disaggregation of 1990, 2000, and 2010 Census blocks in a manner that makes the three resultant grids comparable to each other thus enabling change assessment. To achieve this we needed to use ancillary data that are compatible between