A hybrid model using artificial neural network, kriging, and residual kriging for forecasting the abnormal pollutant distribution

Forecasting of soil pollutants distribution is a significant field of research in view of the current concerns regarding environment all over the world. Increasing of emissions, particularly at urban locations, is an essential... [ view full abstract ]

Forecasting of soil pollutants distribution is a significant field of research in view of the current concerns regarding environment all over the world. Increasing of emissions, particularly at urban locations, is an essential threat to habitat quality. In view of the health and environment hazards posed by increasing of soil pollution, it will be helpful to have a model, which is able to predict the distribution of contaminants and to forecast the pollutants over the analyzed domain. This paper presents a hybrid model using artificial neural networks, kriging and residual kriging to predict the abnormal distribution of a soil pollutant (Cr) comparing with the normally distributed element (Cu) at the particular location in subarctic Tarko-Sale, Russian Federation using data of the chemical analysis. The proposed model is developed, implemented and tested using ArcGIS; combining with the MLP developed in MATLAB. The network structure is selected individually for each element ((2-10-1) for Cr and (2-9-1) for Cu) during the computer simulation based on the minimization of the RMSE. The model consists of several steps: 1) comparing of kriging estimates and network ones; 2) building residues; 3) residual kriging and combination with ANN estimates. The work confirms that trained ANN is suitable for modeling both the normal and abnormal spatial distribution of pollutants. The prognostic accuracy of the ANN is higher than in geostatistical (kriging) and deterministic (IDW) methods. The hybrid model (ANN-kriging) allows improving the predictive accuracy of the model for the both simulated elements.