The paper is present a comparison of modern approaches for predicting the spatial distribution in the upper soil layer of a chemical element chromium (Cr), which had spots of anomalously high concentration in the investigated region. The distribution of a normally distributed element copper (Cu) was also predicted. The data were obtained as a result of soil screening in the city of Tarko-Sale, Russia. Models based on artificial neural networks (multilayer perceptron MLP), random forests (RF), and also a model based on a random forest in which MLP used as a tree - a random perceptron forest (RMLPF) - were considered. The models were implemented in MATLAB. Approaches using artificial neural networks (MLP and RMLPF) were significantly more accurate for anomalously distributed Cr. Models based on RF algorithms proved to be more accurate for normally distributed copper. In general, the proposed model RMLPF was the most universal and accurate.

Time series forecasting is relevant in many fields of human activity. In particular, when studying the processes associated with global warming, such forecasts are very important. The present study used data of the concentration of the greenhouse gases (methane) in the surface layer of atmospheric air on the Arctic island Belyi, Russia. For the work, the time interval of 170 hours (about a week) was chosen during the summer period, characterized by significant daily fluctuations of methane concentration. Models based on artificial neural networks (ANN) such as Nonlinear Autoregressive Neural Network with an External Input (NARX), Elman Neural Network (ENN), and Multi-Layer Perceptron (MLP) were used for modelling. Methane concentrations corresponding to the first 150 hours of the interval used for ANN training, then the concentrations were predicted for the next 20 hours. The model based on the ANN type NARX showed the best accuracy.

The work deals with the application of neural networks residual kriging (NNRK) to the spatial prediction of the abnormally distributed soil pollutant (Cr). It is known that combination of geostatistical interpolation approaches (kriging) and neural networks leads to significantly better prediction accuracy and productivity. Generalized regression neural networks and multilayer perceptrons are classes of neural networks widely used for the continuous function mapping. Each network has its own pros and cons; however both demonstrated fast training and good mapping possibilities. In the work, we examined and compared two combined techniques: generalized regression neural network residual kriging (GRNNRK) and multilayer perceptron residual kriging (MLPRK). The case study is based on the real data sets on surface contamination by chromium at a particular location of the subarctic Novy Urengoy, Russia, obtained during the previously conducted screening. The proposed models have been built, implemented and validated using ArcGIS and MATLAB environments. The networks structures have been chosen during a computer simulation based on the minimization of the RMSE. MLRPK showed the best predictive accuracy comparing to the geostatistical approach (kriging) and even to GRNNRK.