Multi-scale facility context and micro-spatial motor-vehicle-cyclist conflict risk at signalized intersections: a characterized spatial squares framework

Connected and automated intersection management requires spatially explicit safety methods that translate real-time traffic data into micro-scale risk information. This study introduced a CSS framework to explore how facility location, visibility-based spatial configuration, and local traffic exposure jointly relate to motor-vehicle-cyclist conflicts at signalized intersections. The framework divides the internal intersection into fine grid cells, assigns PET-based conflict outcomes and local traffic exposure to each cell, and uses visibility indicators and multi-scale facility variables to characterize them. Using UAV data from signalized intersections in Chengdu, China, the study built 4481 core grid cells and identified 3669 conflicts. A Boruta-RF model screened relevant predictors, including traffic exposure, spatial configuration, positional controls, and facility variables, at five spatial scales. Results show that conflict risk is highly micro-spatial and not fully explained by site-level conflict totals. Facility variables were most informative at the heavyfar and ultrafar scales; local and near facility encodings were not Confirmed, but all four facility-traffic interaction terms were. These findings suggest that intersection facilities relate to conflict risk through neighborhood-scale spatial contexts rather than as isolated design features. Although the evidence comes from eight intersections in a single city, the CSS framework identifies where facility-related conflict risk concentrates and can be used to guide targeted facility redesign. Applications such as real-time risk mapping and adaptive intersection management still require further validation.