Using geospatial networking tools to optimize source locations as applied to the study of food availability: a study in Guilford County, North Carolina

In the study of local-level food security, terms such as food variety, availability, accessibility and utilization represent quantitative metrics to describe one's relationship to the tangible and intangible food environment. Food availability entails how close one is located to the nearest food location. These locations could be healthy and fresh food as applied explicitly to the study food deserts, generally considered to be low-income areas that are far from healthy and fresh food. In the Geographic Information Systems (GIS) network model where travel times and distances are either calculated along a line network such as a series of roads or via more traditional techniques such as Manhattan or Euclidean distance, healthy and fresh food locations are defined as destinations. The places people are traveling from are referred to as sources. However, modeling source locations can be increasingly complex. In just measuring food availability between all residential parcels to the closest healthy food destination in Guilford County, North Carolina, it requires more than 177,000 route calculations, one for each of the residential parcels in Guilford County, North Carolina. Research (Zandbergen and Hart 2009; Fischer 2004; Sahar et al. 2019; Winn 2014) has highlighted the challenges in efficiently locating many addresses and calculating so many routes. In order to simplify the number of network calculations, this research explores ways to model, agglomerate or simplify source locations to decrease the sheer number of calculations while not degrading results when compared to calculations using all original 177,000 source locations. Studies in the field of food security have modeled source locations as census tract centroids, block group centroids, as well as random points and even fishnets or grids. In this paper, we explore the use of different techniques to simulate source locations in the study of food availability in Guilford County, North Carolina. These results are compared to calculations using all residential source locations in the county as a baseline. While all eleven techniques, which include random, stratified and systematic, as well as combinations of them, showed some level of agreement with baseline measurements, sources simulated as block centroids, population-weighted block group centroids and even a randomized-strata technique were strongest using t-tests of two means and equivalence tests for dissimilarity for both drive-distance and drive-time.