Impact of travel time uncertainties on modeling of spatial accessibility: a comparison of street data sources

Abstract

ABSTRACT GIS-based spatial access measures have been used extensively to monitor social equity and to help develop policy. However, inherent uncertainties in the road datasets used in spatial access estimates remain largely underreported. These uncertainties might result in unrecognized biases within visualization products and decision-making outcomes that strive to improve social equity based on seemingly egalitarian accessibility metrics. To better understand and address these uncertainties, we evaluated variations in travel impedance for car and bus transportation using proprietary, volunteer-information-based, and free (non-volunteer-information-based) street networks. We then interpreted the measured variations through the lens of street data uncertainty and its propagation in a common E2SFCA model of spatial accessibility. Results indicated that travel impedance disagreement propagates through the modeling process to effect agreement of spatial access index (SPAI) estimates among different street sources, with larger uncertainties observed for bus travel than car travel. Higher impedance coefficients (β), a model parameter, reduced the impact of street-source variations on estimates. Less urbanized regions were found to experience higher street-source variations when compared with the core-metropolitan area. We also demonstrated that a relative spatial access measure – the spatial access ratio (SPAR) – reduced uncertainties introduced by the choice of street datasets. Careful selection of reliable street sources and model parameters (e.g. higher β), as well as consideration of the potential for bias, particularly for less urbanized areas and areas reliant on public transportation, is warranted when leveraging SPAI to inform policy. When reliable/accurate road network data are not accessible or data quality information is not available, the SPAR is a suitable alternative or supplement to SPAI for visualization and analyses.