Predicting road-specific emissions of the active vehicular fleet over the tier-II city in India: Integrating deep learning and speed information

Emissions from the on-road transport sector are widely recognised as one of the major sources of air pollution in urban areas, especially in developing countries like India. The congested traffic flow pattern and plying of obsolete vehicle technology may result in the heterogeneous spatial distribution of vehicular exhaust emissions, with a high concentration gradient near traffic junctions. Comprehensive information on the spatial and temporal pattern of vehicular fleet emissions is essential to formulating road transport emission reduction policies for effective air quality management. However, computing such detailed emissions is a very complex task as it requires detailed and accurate traffic activity data such as vehicle volume, type, speed, age, fuel and technology share, etc. 
The present study adopted the globally recognised vehicle emission COPERT model to quantify the road transport-related spatial and diurnal emission patterns at 0.01° gridded resolution. The study is novel in its application of deep learning techniques, i.e., the Yolo model for vehicle detection,  achieving greater precision and reducing uncertainty in the activity data for heterogeneous traffic flow patterns. The emission estimate of overall PM is highest at the peak morning, i.e., 8-10 AM time, showcasing approximately 48.5 kg/hr, whereas NOx emissions resulted in being highest in 6-8 PM duration, emitting maximum load from buses, i.e., 106 kg/hr. The hourly emission variations exhibit a distinct bimodal pattern, characterised by prominent peaks in the morning and dominant peaks in the evening, largely associated with traffic congestion and peak travel times. The emissions estimates are observed to be highest for two-wheelers (scooters and motorbikes) and cars at the main traffic junctions inside the city area. Emissions from heavy commercial vehicles are observed to be concentrated on the highways during the nighttime. The developed methodology offers a framework for future real-time emission models in Indian urban regions, using real-time traffic activity data in tier II cities.

Keywords: Road transport emissions, deep learning, urban air quality, YOLO