Large-Spatial Air Pollution Pattern Identification by Combined Approach of Source-Receptor Matrix and GMDH

In this paper, a mathematical model for identifying the static large-spatial pattern of air pollution concentration is developed. Firstly, the source-receptor matrix, which represents a linear relationship between the air pollution sources and the pollutant densities at the monitoring stations (receptors), is identified by a regression analysis. This rough model is used as a first-order approximation. Then, the difference between the output of this rough model and the real system is identified by GMDH (Group Method of Data Handling), that is, the completely unknown nonlinear part of the system is identified by GMDH. The combined approach of source-receptor matrix and GMDH can reduce the average relative error of predicting the concentration pattern by half compared with the use of the rough model of source-receptor matrix alone.