Cadastral-based expert dasymetric system (CEDS) using census and parcel data

Population censuses are commonly aggregated and mapped as uniformly distributed areal units. This leads to generalization of geographical patterns where occupied land is indistinguishable from unoccupied land. Many applications, including demographic profiling, calculation of vulnerable populations requiring access to healthcare, and electoral districting can be measured more precisely if cartographic representations are more disaggregate. An example of disaggregate mapping is the dasymetric principle, which has been widely used to estimate the population of census tracts which are classified as occupied. These methods include simple areal weighting, centroid-based moving windows, land use interpolation from remote sensing and other more complex mathematical models. Each attempts to combine two or more data sets, redistribute population across known occupied land use, and some even abide to the smoothing and masspreserving principles known as pycnophylactic interpolation. However, some of these methods are rigid and at best only replicate the output from basic choropleth maps by assuming uniformity across occupied space or by overly-reapportioning population, neglecting the underlying population distribution, and operating at single scales of census collection. In contrast, the cadastral-based expert dasymetric system (CEDS) is a non areal weighting algorithm that interpolates census data using cadastral parcel data at multiple scales. In demonstrating, this paper will explore the United States population census at three scales–tract, block group and block level–as well as cadastral parcel data (sometimes referred to as taxlot data) containing information on land use type, number of residential units (RU), and number of square feet of living area (RA). The objective of our CEDS example is to estimate the population of a parcel, where: POPl = POPc * Ul / Uc POP l = dasymetrically-derived cadastral parcel level estimated population POPc = census population (at the tract, block group, or block level) Ul = the number of proxy units at the cadastral parcel level (RU or RA) Uc = the number of proxy units at the census level (RU or RA per tract, block group, or block level). The CEDS method calculates error rates for the number of residential units and square feet of living area. It then chooses the model that best fits each individual census polygon (across all three scales). The CEDS method has been tested on cadastral parcel data for the county of Hillsborough in the US city of Tampa, Florida. It has shown to perform well against many common methods of population estimation, and offers flexibility in terms of choice of predictive factor and scale of population information. Our research found no significant differences between final population estimates using three predictive factors but found significant differences between two scales of census population data disaggregation.