Deep learning for Temporal Analysis of Urban Air Quality Assessment

Abstract

Air pollution is a significant environmental problem in many countries, partic ularly in developing nations such as Pakistan. Exposure to high levels of air pollutants has been linked to serious health issues, including respiratory diseases, cardiovascular disorders, and cancer. Therefore, it is critical to accurately fore cast air pollutant levels and take appropriate actions to mitigate emissions.In this study, we utilized data gathered by the Sentinel-5P satellite to predict future air pollution levels in Islamabad, Pakistan. We employed both deep learning and machine learning models to forecast the levels of NO2, SO2, and CO in the at mosphere. The deep learning models included long short-term memory (LSTM) and bi-directional LSTM, while the machine learning models included decision tree and random forest regression. Our findings indicate that the bi-directional LSTM model outperformed the traditional LSTM model in predicting air pollu tant concentrations for NO2, SO2 and CO, achieving MSE values of 0.41, 0.38 and 0.34 respectively, compared to LSTM’s MSE values of 0.50, 0.44 and 0.47. In comparison, the decision tree and random forest regression models did not perform as well as the deep learning models, with higher MSE and MAE values for all pollutants. The best-performing machine learning model was the random forest regression model for predicting NO2 concentrations, with an MSE value of 0.68. Overall, the bi-directional LSTM model demonstrated superior perfor mance in predicting NO2 and SO2 concentrations, while the LSTM model was better for predicting CO concentrations. These results provide valuable insights into forecasting air pollution levels and can help inform policy decisions aimed at mitigating the impacts of air pollution on public health.