A city-based PM2.5 forecasting framework using Spatially Attentive Cluster-based Graph Neural Network model

Urban environments globally are under threat due to recent climate changes caused by a variety of factors such as growing industrialization, rapid migration, increasing traffic flow, etc. An effective data-driven air pollution modeling system helps in increasing regular awareness regarding the severity of the air quality at the local level, play a preventive role in addressing the root causes and hence can be extremely useful for the urban administration. Graph Neural Networks have recently emerged for various classification and estimation tasks on graph-structured data. A Spatially Attentive Cluster-based Graph Neural Network-enabled PM2.5 concentration forecasting model (SA-GNN) is proposed to predict short-term PM2.5 concentrations by considering monitoring stations as nodes of a graph structure and exploring their spatial relationships. This modeling procedure takes into account relevant meteorological variables like wind speed, wind direction, relative humidity etc. An efficient clustering-based spatiotemporal feature extraction method is proposed within a graph neural network setting. This technique makes use of cluster-wise separated graph-structured spatiotemporal features by utilizing disjoint intermediate spatiotemporal GRU networks in order to handle spatial heterogeneity. Additionally, the use of graph attentional network (GATs) makes the modeling framework efficient. The proposed short-term PM2.5 concentrations forecasting framework is applied to the highly polluted Indian capital city, Delhi. The proposed SA-GNN model achieves R2 value 0.75, RMSE and MAE, 25.13 and 21.28 μg/m3 respectively on test data, achieving significant improvement with respect to the baseline models. In fact, even the high pollution episodes can be predicted by the SA-GNN model with better accuracy. Evidently, the proposed GNN-based air pollution modeling framework can be a potential option for forecasting of pollutants in other similar cities globally with high pollution records.