Simulating winter maintenance efforts: A multiscale geographically weighted regression model

Despite its cruciality for road mobility and safety, Winter Road Maintenance (WRM) is highly expensive and environmentally impactful. This suggests that it needs to be optimized. Simulation of WRM operations might help in optimizing these services. This study focuses on upgrading “Effort Model,” a regression-based model used to estimate WRM operations, including salting, plowing, and combined plowing-salting efforts, across Norway's state road network. This model would be the computational core for a WRM-simulation tool. The earlier version, a Generalized Linear Regression (GLR) model, showed limitations in capturing the spatial variability of operations due to Norway's diverse climate and topography. To address this, the authors adopted the Multiscale Geographically Weighted Regression (MGWR) method to upgrade three sub-models for salting, plowing, and plowing-salting efforts. MGWR allows for different spatial scales of explanatory variables. The current proposed models are calibrated using three winter seasons (2020−2023) and include both weather and non-weather variables, such as cycle time, average annual daily traffic (AADT), snow days, and cold days. Findings showed that the MGWR approach significantly improved estimation accuracy compared to the GLR, with higher adjustedR 2 and lower Akaike Information Criterion (AIC) scores. Based on the results, the spatial variation of coefficients is not the same; while some variables like cycle time behave more globally, others such as cold days show localized impacts. Despite the improvements, the model still needs additional refinements in terms of predicting an unseen winter (2023–2024). • WRM in Norway is costly and environmentally impactful. Optimizing it can lead to major savings and sustainability gains. • Introduced an upgraded, MGWR-based Effort Model to simulate seasonal WRM operations across Norway's state roads. • MGWR significantly improved model performance over GLR by capturing spatial heterogeneity in variables. • The model enables better-informed decision-making for WRM policy, contract tendering, and regional planning. • The nature of WRM—shaped by topography, climate, and regional practices—fits especially well with the MGWR approach.