By: César Pinzon-Acosta, Franchesca González-Olivardia, Miguel Hidalgo-Rodríguez, Ana Valdés-Montenegro, Joseph Asprilla-González, Edmanuel Cruz
Abstract
Air pollution from on-road mobile sources is a major concern in urban environments, particularly in regions where high-resolution emission inventories are unavailable. In Panama City, the absence of detailed spatial and temporal information on vehicular emissions limits air quality management. This study presents an integrated methodology to estimate on-road emissions by combining automated vehicle detection based on computer vision with a standardized emission estimation framework. High-definition traffic video data were collected during ten monitoring campaigns at key urban corridors. Vehicles were detected and classified into four categories using a YOLOv5-based model: passenger cars (PC), light commercial vehicles (LCV), heavy-duty vehicles (HDV), and L-category vehicles (L-CAT). Emissions of CO2, CO, NOx, NMVOC, PM and SO2 were estimated using fuel consumption and emission factors from a Tier 1 methodology. The trained YOLOv5 model achieved a precision of 0.99, a recall of 0.98, and a mean average precision of 0.99 on a dataset of 2943 annotated vehicle images. Results show that PCs are the dominant source of CO2 (61.80%), CO (88.16%), and SO2 (93.37%) of the reported emissions, while HDVs are the major contributor to NOx (45.56%) and LCV of PM (69.59%). Spatial analysis identified Villa Lucre, Centenario, and Tocumen as the most emission intensive zones, reflecting the influence of suburban commuting and logistics related traffic. The inventory uncertainty was estimated at ±156.14%. The proposed framework offers a scalable and cost-effective solution for developing urban emission inventories in datascarce environments and supports evidence-based air quality management.