Geographically weighted regression for spatial network data: an application to traffic volumes estimation

Estimating traffic volumes on road networks represents a critical issue in various areas of research such as transport studies and road safety analyses. In these cases, the traffic figures are usually recorded via sparse manual counts or expensive automatic tools (e.g. cameras or inductive loops). However, given the increasing availability of mobile sensors (e.g. smartphones and GPS sat-nav), in the last years several methods were developed to extract traffic information from geo-referenced mobile devices. This paper proposes a geographically weighted regression (GWR) approach to combine fixed counts and GPS data to estimate traffic flows, re-adapting the appropriate statistical methods to the spatial network context. The suggested methodology is exemplified using data collected in the City of Leeds (UK).