Spatiotemporal prediction of unsafe road conditions using weather and infrastructure information

Abstract

This paper presents the spatiotemporal prediction results of unsafe road conditions (URC) across the Dutch highway network, based on sparse, historical multi-source data. Climate change increases the risk of extreme weather-related URCs on road networks, which can cause disruptions and road user safety risks. Hence, the objective of this paper is to develop a data-driven model for the spatiotemporal prediction of URCs for highway networks. We propose a Graph Convolutional Network – Bidirectional Long Short Term Memory Network (GCN-BLSTM) for this goal, and benchmark this model against state-of-the-art deep learning algorithms. The GCN-BLSTM shows overall, higher predictive performance, with road ponding and roadside fires being predicted most accurately among the URCs. This study highlights the predictive potential of a hybrid spatiotemporal model for multi-URC predictions, compared to state-of-the-art data-driven methods, and proposes implications for future research on multi-URC prediction.