Project Coordinator :  Prof. Dr. S. Levent Kuzu   (Istanbul Technical University)
Investigators:  Prof. Dr. Şeref Naci Engin   (Yıldız Technical University )

Air quality modeling studies are conventionally carried out using dispersion models, aiming to predict pollutant concentrations at ground level or at a specified height. In urban areas, the exclusion of buildings from the modeling domain can lead to prediction errors, particularly in terrains such as streets or avenues where vehicles pass. Additionally, emissions are often estimated using assumptions that may result in unrealistic predictions. The innovative aspects of the proposed project are as follows:  i) Incorporating real terrain conditions, including buildings, into the modeling study, ii) Collecting real-time data on vehicle types, numbers, and speeds for the modeled region and time frame.  Subsequently, the collected data will be used as input for a high-resolution computational fluid dynamics (CFD) model, enabling pollutant predictions at any point within the geometry.  The study will focus on the modeling and calculation of NO, NO₂, and CO emissions to represent traffic emissions. Model results will be validated against data obtained from air quality monitoring stations. A more innovative approach, deep learning, will be employed to estimate vehicle emissions. Images captured from fixed traffic cameras will be processed using the YOLO object detection algorithm to create a comprehensive dataset, including real-time information on vehicle types, speeds, and counts. This approach will enable the dynamic determination of each vehicle's contribution to the pollution level in the region. Total vehicle emissions will be calculated using COPERT, following the most comprehensive methodology, Tier-3. The data obtained will serve as input for the CFD model.

                      
(a) Image labelling and vehicle detection           (b) Air quality model