An integrated multi-stage analytical method tailored for the spatiotemporal evolution of microscopic road segment traffic risk

Previous research has identified accident black spots at the urban road network level, but spatiotemporal characteristics at the microscopic segment scale remain understudied. To address this gap, this study develops an integrated multi-stage analytical framework and examines parameter selection issues for crash risk warning and traffic supervision in spatiotemporal contexts. The framework first applies the Peaks over Threshold (POT) extreme value model to determine the traffic conflict threshold. It then employs the emerging space–time cube–based hotspot analysis to partition road segments into spatiotemporal units and to identify their evolving traffic risk patterns. A key methodological contribution lies in integrating Fast Fourier Transform (FFT) periodograms, autocorrelation analysis, and a spatial heterogeneity-modified global Moran’s I index to objectively determine optimal spatiotemporal step lengths and neighborhood scales. Empirical analysis uses traffic conflicts from 24-h traffic data on two-lane roads in Jinan City. Results show that the derived spatiotemporal scale parameters effectively integrate road structure heterogeneity with risk periodicity, aligning well with road geometries and lane dimensions. Risk hotspots display pronounced spatial heterogeneity and temporal clustering, forming high-risk clusters particularly near ramps and specific lane sections. These findings provide a scientific basis for lane-level traffic risk management, with substantial theoretical and practical implications.