When remote sensing data meet ubiquitous urban data: Fine-grained air quality inference

With the growth of the economy, the air quality is becoming a serious issue, especially for those developing countries, such as China. Therefore, it is very important for the public and the government to access real-time air quality information. Unfortunately, the limited number of air quality monitoring stations is unable to provide fine-grained air quality information in a huge city, such as Beijing. One cost-effective approach for obtaining fine-grained air quality information is to infer air quality with those measured data at the monitoring stations. However, existing inference techniques have poor performance because of the extreme data sparsity problem (e.g., only 0.2% data are known). We observe that remote sensing has been a high-quality data source about urban dynamics. In this paper, we propose to integrate remote sensing data and ubiquitous urban data for air quality inference. There are two main challenges, i.e., data heterogeneity and incomplete remote sensing data. In response to the challenges, we propose a two-stage inference approach. In the first stage, we use the AOT remote sensing data and the meteorological data to infer the air quality values with an Artificial Neural Network (ANN). After this stage, we significantly reduce the percentage of empty cells in the tensor representing the spatio-temporal air quality values. In the second stage, we propose a tensor decomposition method to infer the complete set of air quality values. We use the spatial features (i.e., road features and POI features) and the temporal features (i.e., meteorological features) as the constraints in the tensor decomposition process. Experiments with real data sets show that our approach has profound performance advantage over the state-of-the-art methods, such as U-Air.