The article proposes an approach to optimize the division of initial data into training and test subsets for modeling the spatial distribution of a component using artificial neural networks (ANN). This approach takes into account the spatial irregularities and the scatter values of the modeled component. Data were obtained during soil screening of urban areas in the Russian subarctic zone: Tarko-Sale and Noyabrsk cities. For modeling, the chemical element chromium (Cr) was chosen. On Tarko-Sale, areas with an abnormally high value of the modeled element were discovered. On Noyabrsk, no anomalies in Cr values were detected. Using the feedforward neural network multilayer perceptron (MLP) method, Cr spatial distribution in the surface soil layer is constructed. The MLP structure was chosen by computer simulation based on minimizing the root mean square error (RMSE). Comparison of the models that use random division of the initial data into a training and test subset, and models that are based on the proposed approach was performed. For each approach, the mean absolute error (MAE), RMSE, and the mean square relative error (RMSRE) were calculated. For both areas, models using the proposed approach showed more accurate results (up to 50% improvement).

Buevich Alexander Gennadyevich, Engeneer in Laboratory of Physics and Ecology, Institute of Industrial Ecology UB RAS. 620219, Ekaterinburg, S. Kovalevskaya str., 20. E-mail: bagalex3@gmail.com.

Rakhmatova Anna Yuryevna, Junior Researcher in Laboratory of Physics and Ecology, Institute of Industrial Ecology UB RAS. 620219, Ekaterinburg, S. Kovalevskaya str., 20. E-mail: anyarakhmatova@gmail.com.

Sergeev Aleksandr Petrovich, Candidate of Physical and Mathematical Sciences, Acting Head in Laboratory of Physics and Ecology, Institute of Industrial Ecology UB RAS. 620219, Ekaterinburg, S. Kovalevskaya str., 20. E-mail: aleksandrpsergeev@gmail.com.

Baglaeva Elena Mikhailovna, Candidate of Physical and Mathematical Sciences, Senior Researcher in Laboratory of Physics and Ecology, Institute of Industrial Ecology UB RAS. 620219, Ekaterinburg, S. Kovalevskaya str., 20. E-mail: elenbaglaeva@gmail.com.

Shichkin Andrey Vasilevich, Engeneer in Laboratory of Physics and Ecology, Institute of Industrial Ecology UB RAS. 620219, Ekaterinburg, S. Kovalevskaya str., 20. E-mail: and@ecko.uran.ru.

Subbotina Irina Evgenievna, Candidate of Physical and Mathematical Sciences, Researcher in Laboratory of Physics and Ecology, Institute of Industrial Ecology UB RAS. 620219, Ekaterinburg, S. Kovalevskaya str., 20. E-mail: iesub@mail.ru.

Sergeeva Marina Viktorovna, Researcher in Laboratory of Physics and Ecology, Institute of Industrial Ecology UB RAS. 620219, Ekaterinburg, S. Kovalevskaya str., 20. E-mail: marin@ecko.uran.ru.

Markelov Yuri Ivanovich, Candidate of Physical and Mathematical Sciences, Head in the Center of Collective Use, Institute of Industrial Ecology UB RAS. 620219, Ekaterinburg, S. Kovalevskaya str., 20. E-mail: markelov@ecko.uran.ru.

1. Sergeev A.P., Buevich A.G., Baglaeva E.M., Shichkin A.V. Combining spatial autocorrelation with machine learning increases prediction accuracy of soil heavy metals // Catena. 2019. Vol. 174. P. 425-435.
2. Guo G.H., Wu F., Xie F., Zhang R. Spatial distribution and pollution assessment of heavy metals in urban soils from southwest China // Journal of Environmental Sciences. 2012. Vol. 24, Issue 3. P. 410-418.
3. Liu F., He X., Zhou L. Application of generalized regression neural network residual kriging for terrain surface interpolation // Proc. SPIE 7492, International Symposium on Spatial Analysis, Spatial-Temporal Data Modeling, and Data Mining. 2009. 74925F.
4. Land cover and landscape as predictors of groundwater contamination: a neural-network modelling approach applied to Dobrogea, Romania / R.R. Shaker [et al.] // Journal of Environmental Protection and Ecology. 2010. Vol. 11, Issue 1. P. 337-348.
5. Shaker R.R., Ehlinger T.J. Exploring non-linear relationships between landscape and aquatic ecological condition in southern Wisconsin: A GWR and ANN approach // International Journal of Applied Geospatial Research. 2014. Vol. 5, Issue 4. P. 1-20.
6. Use of trans-Gaussian kriging for national soil geochemical mapping in Ireland / C. Zhang [et al.] // Geochemistry: Exploration Environment Analysis. 2008. Vol. 8, Issue 3-4. P. 255-265.
7. Koike K., Matsuda S., Suzuki T., Ohmi M. Neural Network-Based Estimation of Principal Metal Contents in the Hokuroku District, Northern Japan, for Exploring Kuroko-Type Deposits // Natural Resources Research. 2002. Vol. 11, Issue 2. P. 135-156.
8. Samanta B., Ganguli R., Bandopadhyay S. Comparing the Predictive Performance of Neural Networks with Ordinary Kriging in a Bauxite Deposit // Transactions of Institute of Mining and Metallurgy, Section A, Mining Technology. 2005. Vol. 114, Issue 3. P. 129-139.
9. Spatial prediction of soil organic matter content integrating artificialneural network and ordinary kriging in Tibetan Plateau / F. Dai [et al.] // Ecological Indicators. 2014. Vol. 45. P. 184-194.
10. Falamaki A. Artificial neural network application for predicting soil distribution coefficient of nickel // Journal of Environmental Radioactivity. 2014. Vol. 115. P. 6-12.
11. Li Y., Li C., Tao J.-J., Wang L.-D. Study on Spatial Distribution of Soil Heavy Metals in Huizhou City Based on BP-ANN Modeling and GIS // Procedia Environmental Sciences. 2011. Vol. 10. P. 1953-1960.
12. Qualitative and quantitative investigation of chromium-polluted soils by laser-induced breakdown spectroscopy combined with neural networks analysis / J.-B. Sirven [et al.] // Anal Bioanal Chem. 2006. Vol. 385, Issue 2. P. 256-262.
13. Anagu I., Ingwersen J., Utermann J., Streck T. Estimation of heavy metal sorption in German soils using artificial neural networks // Geoderma. 2009. Vol. 152. P. 104-112.
14. Dobrovolsky G.V., Urusevskaya I.S. Soils geography. Moscow : MSU Publishing House ; Kolos-s Publishing House, 2004. 460 p.
15. Environment Geochemistry / Yu.E. Saet [et al.]. Moscow : Nedra, 1990. 335 p.
16. Geochemistry handbook / G.V. Voitkevich [et al.]. Moscow : Nedra, 1990. 479 p.