Predictive spatial network analysis for high-resolution transport modeling,applied to cyclist flows,mode choice,and targeting investment

Betweenness is a measure long used in spatial network analysis (SpNA) to predict flows of pedestrians and vehicles, and more recently in public health research. We improve on this approach with a methodology for combining multiple betweenness computations using cross-validated ridge regression to create wide-scale, high-resolution transport models. This enables computationally efficient calibration of distance decay, agglomeration effects, and multiple trip purposes. Together with minimization of the Geoffrey E. Havers (GEH) statistic commonly used to evaluate transport models, this bridges a gap between SpNA and mainstream transport modeling practice. The methodology is demonstrated using models of bicycle transport, where the higher resolution of the SpNA models compared to mainstream (four-step) models is of particular use. Additional models are developed incorporating heterogeneous user preferences (cyclist aversion to motor traffic). Based on network shape and flow data alone the best model gives reasonable correlation against cyclist flows on individual links, weighted to optimize GEH (r² = 0.78, GEH = 1.9). As SpNA models use a single step rather than four, and can be based on flow data alone rather than demographics and surveys, the cost of calibration is lower, ensuring suitability for small-scale infrastructure projects as well as large-scale studies.