Predicting Long-Term Urban Traffic Patterns Under Varying Weather Conditions with AI

Abstract: Adverse weather events significantly impair the performance of transportation systems, causing delays, speed reductions, and traffic flow breakdowns. This study proposes a predictive model that forecasts long-term spatio-temporal traffic conditions at the city level to enhance system resilience against such events. By leveraging a spatio-temporal graph convolutional neural network (ST-GCNN) that integrates meteorological variables such as precipitation, temperature, humidity, soil moisture, and so forth, with traffic data including speed and network topology, the developed artificial intelligence (AI) aims to identify road segments at high risk of weather-related disruptions, with a particular focus on flooding and pluvial events. Unlike short-term predictions that offer limited preparation time for road users and agencies, our long-term forecasting approach enables strategic planning and resource allocation up to 14 days in advance, thus significantly improving emergency preparedness and response efficiency. The research demonstrates the AI's notable predictive accuracy in identifying vulnerable road segments, providing transportation agencies with actionable insights for preemptive measures. This capacity to anticipate and mitigate the impact of extreme weather events on urban traffic not only advances emergency management practices but also contributes to the development of more adaptive and resilient urban transportation networks, enhancing the safety and efficiency of city-level transportation systems in the face of climate unpredictability.