learning-based predictionSnow removal dispatch decisions during winter nights require not only the accurate prediction of dispatch necessity but also the intuitive presentation of road conditions and decision-relevant information to human operators.In this study, we propose a snow removal dispatch decision support method that integrates fixed-camera images and meteorological information to both predict dispatch necessity and visualize information for operational support.The proposed method employs a machine learning model with feature selection, integrating meteorological observations, short-term forecasts, snow depth measurements, and an imagederived road condition indicator called the snow coverage ratio (SCR).From a sensing perspective, the proposed method integrates heterogeneous environmental sensing modalitiesincluding fixed-camera imaging, snow-depth sensing, and meteorological sensing-to represent road surface conditions relevant to dispatch decision-making.By visualizing road conditions together with prediction results, the system provides operators with objective and consistent situational awareness to support decision-making.Experimental results confirm that SCR effectively reflects temporal changes in road conditions and contributes to dispatch prediction.Furthermore, evaluation using data collected under similar winter meteorological conditions shows that a dispatch prediction model with L1 regularization tends to outperform human operators' decisions.Additional evaluation applying models trained on past-year data to subsequent years demonstrates that training with multi-year datasets improves generalization under varying winter conditions.These results indicate that the proposed method provides practical and reliable support for snow removal dispatch decisions in real operational environments.
Yamashita et al. (Wed,) studied this question.