Exploiting temporal information in parcel data to refine small area population estimates

Abstract

Temporal analysis of small-area demographic data commonly relies on areal interpolation methods to create temporally consistent and compatible areal units. In this study, cadastral (parcel) data are used to identify residential land and to dasymetrically refine census tracts, with the goal of achieving more accurate small-area estimates. The built date recorded for residential parcel units is used to create residential land layers for two different time points used in the areal interpolation. Three different areal interpolation methods are employed with and without dasymetric refinement, including areal weighting (AW), target density weighting (TDW) and pycnophylactic modeling (PM). The methods interpolate tract-level population counts in Hennepin County, Minnesota, in 2000 into census tract boundaries from the year 2010. The mean absolute error, median absolute error, root mean square error and the 90th percentile of absolute error are calculated for each of the methods, and spatial variation in the interpolations are displayed in maps. Parcel-based refinements are also compared with refinements using the National Land Cover Dataset (NLCD). Results show that spatial refinement using residential parcels has the potential to improve the accuracy of areal interpolation for temporal analysis. Parcel-refined TDW out-performs the other tested methods, as well as the NLCD-refined TDW in this example. Parcel data identify residential land more reliably in rural areas. However, parcel units can have very large extents potentially biasing residential area delineation and population counts. Parcel-based refinement has the potential to further advance demographic change analysis over long time periods and large areas where the built date attribute is included in the dataset.

Estimating changes in urban land and urban population using refined areal interpolation techniques

“…All methods described are run for two time periods, 1990 to 2010 and 2000 to 2010, respectively. The methods are briefly described below, but the reader can refer to previous works (e.g., Zoraghein et al 2016) for more detailed explanations and mathematical formulae. Importantly, in this study the spatially refined temporal interpolation framework is applied to urban population using ancillary variables that are known to be associated with urban lands and thus delineate areas where urban population is expected to reside.…”

“…Areal interpolation coupled with spatial refinement has been demonstrated as an effective approach to reduce estimation errors in temporally interpolating population enumerated in a set of source zones (source census year) to target zones defined by the boundaries of the target census year (e.g. Ruther et al 2015;Zoraghein et al 2016). In this study, this approach is tested for estimating urban population of census tracts in 1990 and 2000 (i.e., the source zones) within census tract boundaries in 2010 (i.e., the target zones) using different ancillary variables for spatial refinement to create a temporally consistent time series of urban population distributions at the tract level.…”

Hybrid areal interpolation of census counts from 2000 blocks to 2010 geographies

“…For example, the Longitudinal Tract Database (LTDB), which supplies 1970–2010 census data for 2010 tracts (Logan, Xu, & Stults, 2014), estimates 2000 population totals by interpolating from blocks using areal weighting (AW), which allocates source zone counts in proportion to the area of intersection with each target zone (Goodchild & Lam, 1980). Other research has applied AW to block data as a benchmark against which to assess the interpolation of tract data (Buttenfield, Ruther, & Leyk, 2015; Ruther, Leyk, & Buttenfield, 2015; Zoraghein et al, 2016). AW’s basic assumption—that characteristics are uniformly distributed within each source zone—may often be inaccurate, and numerous studies have shown that, in settings with larger source zones, more sophisticated models are more effective (e.g., Goodchild, Anselin, & Deichmann, 1993; Fisher & Langford, 1995; Mrozinski & Cromley, 1999; Gregory, 2002; Reibel & Bufalino, 2005; Langford, 2006; Reibel & Agrawal, 2007; Schroeder, 2007; Zandbergen & Ignizio, 2010; etc.).…”

“…In an early example, Wright (1936) identifies zones through interpretation of topographic maps. More recently, the most common ancillary data type has been land cover and land use data (e.g., Eicher & Brewer, 2001; Holt, Lo, & Hodler, 2004; Langford, 2006; Mennis & Hultgren, 2006; Lin, Cromley, & Zhang, 2011; Cromley, Hanink, & Bentley, 2012; Lin et al, 2013; Schroeder & Van Riper, 2013; Buttenfield, Ruther, & Leyk, 2015; Lin & Cromley, 2015; Ruther, Leyk, & Buttenfield, 2015; Zoraghein et al, 2016). Under a strict definition of dasymetric mapping, the ancillary data must represent zones, which excludes models based on road lengths (as discussed in Section 2.1) or counts of address points (e.g., Tapp, 2010), but line and point data can still be used to define BD models through the use of buffering.…”

Spatiotemporal enabled Content-based Image Retrieval

“…The first strategy applies TDW to only refined sub-areas of source and target zones delineated by residential parcels as the ancillary variable (Zoraghein et al 2016). The built-year attribute that records when the main structure of a parcel was built is used to match parcels with the census year.…”