Spatiotemporal prediction of dynamic particulate matter using deep learning methods

Abstract

With the unprecedented increase in the world population, there is an increase in the number of vehicles and industries, which results in a worldwide increase in air pollution. Currently, the world population touched the mark of 8 billion. Vehicular emission is the main reason for the increase in air pollution. Thus, measuring and modeling air pollution and taking preventive actions efficiently becomes extremely important. For modeling, traffic characteristics including traffic volume and density near fixed monitoring sites, play an important role. Other factors such as meteorological data like Relative Humidity (RH), Atmospheric Temperature (AT), Wind Speed (WS), and Barometric Pressure (BP) are also used. The pollution decreases during summers as the temperature increases, wind speed increases in summers, and the humidity is less. Due to all these reasons air pollution decreases as winds blow away the pollution. As the winter approaches at the start of November, the pollutants accumulate in the air due to high humidity and wind speeds. There are only 36 monitoring stations located in Delhi. There is a large need of monitoring stations, but it is not feasible to install new monitoring stations as there is a high cost of setting and maintenance of the static stations. So, there is a need to develop a prediction model for the prediction of the pollutants at the locations that are away from the monitoring stations. The study mainly takes four types of features for model development. These features are: Meteorological features, Traffic flow features, Point of Interest (POI) features, and historical data on pollution gathered from the monitoring stations. The research study aims to develop a model to minimize the error between the actual values and the predicted values.