Anthropogenic emissions of greenhouse gases accelerate global warming and contribute to further temperature increases. Global warming increases the likelihood of a shift towards more warm days and seasons and fewer cold days and seasons. Additionally, it causes changes in precipitation patterns. In earlier research, as ambient temperatures increase, cognitive performance decreases and the risk of crashing increases. Earlier, crash-frequency models were developed using various methodologies, but time-series crash-frequency prediction studies considering the effects of climate change are scarce. Therefore, the purpose of this study is to identify the correlation between crashes and climate change using big data and to develop crash-frequency models using an econometric model and a deep-learning model. Econometric models use autoregressive-integrated moving average and autoregressive-integrated moving average with exogenous variable that are traditional time-series methodologies. Deep-learning models use long short-term memory. This study approached crash occurrence by comprehensively considering climate change and traffic factors. Also, it differs from earlier studies in detailing the influence of independent variables on crashes. Through the results, the impact of climate change on accidents can be identified and it can contribute as an engineering basis for improving traffic safety.

Conventional Safety Performance Functions (SPFs) primarily rely on static exposure measures such as Annual Average Daily Traffic (AADT), often failing to capture real-time, individual-level risky driving behaviors. To address this gap, this study proposes a Risky Driving Indicator (RDI) that integrates large-scale smartphone-based hard braking event data with traffic detector occupancy measures. The RDI was evaluated against traditional models across three specific aggregation levels: AADT, Annual Average Weekday Daily Traffic (AAWDT), and AAWDT excluding the overnight period. A case study was conducted using data from 2021 to 2022, a period coinciding with the COVID-19 pandemic, on South Korea’s busiest freeway to evaluate RDI-based SPFs. The results showed that models using the COM-Poisson framework outperformed traditional volume-based versions, showing superior performance across Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Akaike Information Criterion (AIC) values. These findings confirm that integrating crowdsourced behavioral data enhances predictive accuracy, offering transportation agencies a cost-effective, scalable solution for proactive hotspot identification and dynamic safety monitoring. By improving safety management through scalable and cost-effective mobile sensing, this study contributes to the development of more sustainable highway transportation systems.

Extreme weather such as heavy rain and typhoons increasingly disrupts freeway traffic. However, few methods quantify the resulting loss of system performance in a systematic and scalable way. This study proposes the Transportation System Resilience Index (TSRI), which measures how far observed traffic speed deviates from statistically normal conditions for each road segment and time period. Unlike existing resilience metrics, the TSRI considers spatial and temporal heterogeneity. To validate its value for forecasting, the TSRI was added to a Spatial-Temporal Graph Convolutional Network (STGCN). The STGCN outperformed all baseline models across every horizon, improving accuracy by 11.4% on average, and an ablation study showed that removing the TSRI raised the 60-minute forecast error by 7.46%. A case study confirmed that the index tracks disruption and recovery across regions. The TSRI can support real-time monitoring, resilience assessment, and disaster response planning for road authorities.

Effective mitigation of urban traffic congestion requires accurate speed prediction. It also requires a proactive understanding of how traffic states evolve. Existing deep learning models often work as black boxes. They do not capture the underlying dynamics of traffic flow transitions, which limits their use for practical congestion management. To address this gap, this study proposes a theory guided deep learning framework that integrates traffic flow theory with short-term speed prediction. The K-means clustering algorithm was used to classify traffic into three states based on the fundamental speed and flow diagram. These states are free-flow, transition, and congested. The state label was then added as an input feature to nine deep learning models. The method was validated on thirty-eight congested freeway segments in South Korea. On average, the improvement in prediction accuracy was modest. However, the benefit was concentrated in worsening regime transitions, in which the state feature reduced errors approximately four times more than in steady conditions. A statistical test confirmed that this effect was significant in most cases. The framework is therefore most useful when the flow begins to break down, which is when an early warning matters most. This allows operators to act early with measures such as variable speed limits before congestion fully develops, shifting traffic management from reactive monitoring toward proactive control.