Comparing Modelled and Recalled Routes of Children's School Travel : Winter Maintenance-Weighted Routing in Oulu, Finland

Abstract

Background: Active school travel supports children's physical activity and independent mobility, but subarctic winter conditions introduce a spatially differentiated barrier that most GIS-based routing models fail to capture. Existing shortest path models assume travellers minimise geomet ric distance, an assumption increasingly challenged by empirical route choice research, yet few studies have tested whether incorporating winter maintenance quality as a routing impedance improves correspondence with children's actual recalled routes, and none have done so in a subarctic winter city.

Methods: This thesis compared two GIS-based routing models against 168 child-reported school travel routes (101 cycling, 67 walking) collected via PPGIS survey across four primary schools in Oulu, Finland, in February 2026. Model 1 generated shortest path routes on the Digiroad national road network. Model 2 applied a winter maintenance-weighted impedance function, tested under two multiplier configurations: a theoretically derived conservative configuration and an expert-informed configuration developed through practitioner consultation. Spatial correspondence between modelled and reported routes was assessed using corridor buffer similarity (CBS), discrete Fréchet distance, detour ratio, and a compound agreement classification. Binary logistic regression tested whether individual perceived winter surface concern predicted spatial deviation from the winter-weighted model, controlling for travel mode and school.

Results: The shortest path model achieved moderate but spatially variable correspondence with reported routes (median CBS = 0.542) across schools. Winter maintenance weighting did not improve spatial correspondence with children’s reported routes at the aggregate level: median spatial overlap declined and route divergence increased under both weight configurations tested, contrary to the study's central hypothesis. This pattern varied substantially by school catchment, with one school (C) showing improved correspondence under the weighted model while others showed pronounced decline, suggesting the effect of maintenance weighting depends on local network structure rather than acting uniformly. Individual perceived winter surface concern did not significantly predict route deviation from the weighted model, while school catchment membership did, indicating that neighbourhood-level structural factors influence model correspondence more strongly than individual perception at this sample size. These results suggest that administrative maintenance classification, without empirically calibrated behavioural weights, is an insufficient basis for modelling children's winter route choices. These findings are drawn from a single city and winter season and should be interpreted with this scope in mind.

Conclusion: Detecting the spatial signal of winter maintenance in children's route behaviour requires finer-grained, revealed-preference impedance data than administrative classification alone can provide. For winter cities pursuing catchment-targeted maintenance interventions, these findings suggest that behavioural change is most achievable where the local network already offers genuine route alternatives.