Optimal Reduction of the Local-scale PM_2.5 Monitoring Network (I): Predicting the Spatial Distribution of PM_2.5 by Regression Kriging

Abstract

In this study, we predicted the spatial distribution of PM_2.5 in the Kyushu area by a regression kriging (RK) model to evaluate how well a subset of the present networks can represent the full network. We developed a RK model with the particulate sulfate concentration in the PM_2.5 obtained by the chemical transport model as an indicator for regional pollution. The predictions were made for the daily average, annual average, and high concentration day average. The prediction performance was assessed by indicators calculated from the leave-one-out cross validation results. The R²-values were 0.90, 0.58, 0.70 and the RMSE-values were 2.47, 1.45, 2.40 µg/m³,respectively. The performance of the RK model in this study was comparable to that found in previous studies. The daily average prediction of the model showed a significantly better prediction accuracy than that of a RK model without the regional pollution indicator. These results suggested that the model in this study showed a good performance and was suitable for further evaluation. These results also suggested that the performance of the RK model might be improved by introducing explanatory variables according to features of the area, especially in areas with sparse observation networks.