Air Quality Inference With Spatial Correlations and Heterogeneity of IoT-Based Monitoring Data

Accurate monitoring of air quality at high spatial resolution is essential, but is hampered by the sparse and uneven distribution of monitoring stations. Data inference using monitoring data is a powerful tool. However, the complex spatial relationships of IoT-based multimodal monitoring data are often not fully exploited by existing inference methods. In this paper, we propose a Spatial Correlation-Heterogeneity Network (SCHNet) to explicitly exploit both spatial heterogeneity and correlations of monitoring data to improve air quality inference in unmonitored areas. SCHNet integrates air quality data with population density and elevation to capture the geographic context that influences pollutant distribution through a spatial heterogeneity feature extraction module. A stepwise station selection strategy is developed to identify correlated stations, and a spatial correlation module is proposed to characterise spatial dependencies between data. A spatial fusion module is then developed to combine these features for accurate air quality inference. Experiments on a real-world dataset demonstrate the superior performance of SCHNet over existing methods, highlighting its effectiveness in addressing the challenges of sparse and uneven monitoring data.