Road Safety Assessment and Diagnosis of Local Road Networks through a Spatial-Temporal Analysis

Road safety is a critical issue that requires a comprehensive understanding of accident risk factors and the spatial distribution of hazardous road locations (HRLs). This thesis focuses on identifying HRLs in the Piedmont Region (Italy) using spatial-temporal analysis, particularly with Kernel Density Estimations (KDEs) and employing GIS-based modeling. Despite significant advancements in road safety across Italy, urban road networks with complex traffic patterns continue to face challenges, making it essential to identify crash-prone locations, so the road infrastructure administration can implement effective countermeasures by prioritizing safety interventions based on realistic analyses’ results. All the work is grounded in the understanding of Density-Based Methods functionality, more specifically with KDEs, and general insights into data spatial distribution. Even though all the obtained results were reached by implementing KDEs, considering the theoretical principles from spatial distribution of points, it was possible to identify accident clusters. In consequence, these crash clusters were crucial to determine the HRL and hotspots along the study area. Complementary, the implemented GIS software, ArcGIS Pro, lead to the study of critical HRL and within them where the corresponding hotspots were located. Moreover, the Piedmont Region was selected for this case study due to its complex urban landscape, diverse traffic conditions, and availability of high-quality accident data, provided by the Regione Piemonte administration, covering over a 10-year period (2012–2022) 105,322 crashes. Additionally, a key methodological insight was the influence of KDEs parameters on crash density estimations. The results confirm that smaller search radiation improves accuracy in urban areas, while larger search radii help prevent data fragmentation in rural areas. However, KDEs also produced false positive hotspots, leading to cross-validation with raw crash data for increased accuracy. Whereas even though implementing building constraints enhances spatial precision within the KDEs analysis performed, by restricting the density results along the road network; its implementation reduces computational efficiency and increases analyses’ elapsed time, which eventually crashes the software. Overall, the findings indicate that HRLs are primarily concentrated in built-up areas, principally at the province capitals from the region, and particularly at roundabouts and intersections. Then, KDE analysis shows persistent crash clustering, revealing that crash occurrences are not random but instead exhibit spatial dependency. In contrast, even rural areas show lower crash densities, it doesn’t mean that they should receive less attention to their crash-prone locations.