UC Santa Barbara

Natural hazards such as wildfires and floods are becoming increasingly frequent and severe, increasing the need for effective evacuation and emergency planning. A key aspect of this planning is ensuring that road infrastructure can support emergency response and safe evacuations. In this study, we introduce a novel geospatial uncertainty-aware road detection (GUARD) method to identify road segments of substandard width from remote sensing data. GUARD leverages airborne optical imagery and light detection and ranging (LiDAR) data for road region extraction and applies a Bayesian classification model with moment-based features to account for measurement uncertainty. By identifying substandard road segments for follow-up inspection, GUARD can facilitate more efficient road infrastructure monitoring and emergency planning. Evaluation using ground-truth data from a case study in Santa Barbara, California, United States, demonstrates the method’s effectiveness. Results indicate that a significant portion of the road segments in the study area have passages narrower than required standards. This finding highlights the potential vulnerability of many older wildland–urban interface neighborhoods and how the GUARD method can contribute to road infrastructure monitoring and adaptation in emergency planning.