Modeling of surface dust concentrations using neural networks and kriging

• DOI: 10.1063/1.4968425
• Publication Date: 2016-12-20T13:01:45Z
• Authors: Alexander G. Buevich; Alexander N. Medv...; Alexander P. Sergeev; Dmitry A. Tarasov; Andrey V. Shichkin; Marina V. Sergeeva; T. B. Atanasova
• Keywords: Multilayer perceptron; Variogram

Creating models which are able to accurately predict the distribution of pollutants based on a limited set input data is an important task in environmental studies. In paper two neural approaches: (multilayer perceptron (MLP)) and generalized regression network (GRNN)), geostatistical (kriging cokriging), using for modeling forecasting dust concentrations snow cover. The area study under influence emissions from copper quarry several industrial companies. comparison mentioned approaches conducted. Three indices used as indicators accuracy: mean absolute error (MAE), root square (RMSE) relative (RRMSE). Models artificial networks (ANN) have shown better accuracy. When considering all indices, most precision model was GRNN, uses parameters coordinates sampling points distance probable source. results work confirm that trained ANN may be more suitable tool