Assessment of Impact of air pollution using deep learning-based Air Quality data Model

Air pollution poses a major concern to people’s lives. Over two million Indians are said to lose their life to causes attributed to air pollution (Balakrishnan et al., 2019). Most of this pollution comes from industries closely followed by vehicular pollution, unrestrained emission sources, periodic agricultural pollutants, and household pollution. The air quality data obtained from orbital sensors like Sentinel 5 and 5P, MODIS and Landsat and ground sensors (Central pollution control board sensors, CPCB) provide a large amount of information about the particle pollutants present in the atmosphere. This study explored the fusion of data obtained from ground (CPCB) and orbital sensors to better estimate the air quality parameters like PM2.5, NO2, CO and SO2. The combination of these improved parameters will subsequently enhance the air quality index (AQI) approximation. The region of study is the Indo-Gangetic plain, North India because this region hosts eight out of the top ten most polluted cities worldwide.

In this study, we propose a deep neural network-based air quality model by combining fine-grained properties of in-situ (CPCB) data with coarse-grained satellite images. Deep Neural Networks like Convolutional Neural Network (CNN) and Long short-term memory (LSTM) have shown major advantage in solving nonlinear spatio-temporal problems. They can extract valuable contextual features to combine temporal attributes, and model temporal and spatial dependencies accurately. Therefore, we propose the combination of CNN and LSTM models for precise air quality prediction in the Indo Gangetic Plain for the upcoming twenty-four hours based on data acquired on the preceding twenty-four hours.