Spatial-temporal prediction of air quality by deep learning and kriging interpolation approach

Abstract

Air quality level is closely associated with our day-to-day life due to its serious negative impact on human health. Air pollution monitoring is one of the major steps of air pollution control and prevention. However, limited air pollution monitoring sites make it difficult to measure each corner of a region's pollution level. This research work proposes a methodology framework incorporating a deep learning network, namely CNN-BIGRU-ANN and geostatistical Ordinary Kriging Interpolation model, to address this research gap. The proposed CNN-BIGRU-ANN time series prediction model predicts the $P{M_{10}}$ pollutant level for existing monitoring sites. Each monitoring site's predicted output is transferred as input to the geostatistical Ordinary Kriging interpolation layer to generate the entire region's spatial-temporal interpolation prediction map. The experimental results show the effectiveness of the proposed method in regional control of air pollution.