Individual And Pairwise Representativeness Of Sampling Points In Interpolation Tasks Of Heavy Metals Distribution In The Topsoil

The optimization of environmental soil monitoring based on representative selection of a training subset for an artificial neural network is an unresolved problem in the tasks of interpolation of the distribution of metals in the topsoil. The soil survey data, often used as input for artificial neural network modeling, are datasets at irregular points. Usually, the division of the input data into training and test subsets is carried out randomly in a ratio of 70% to 30% points, respectively. The question of the individual and collective representativeness of local sampling points on the element content in the soil in a given area for a training subset remains beyond the scope of interpolation problems. In this work, the representativeness of the sampling points plays a crucial role in reducing the ANN error and enhancing the correlation between the results of model calculations on the test subset and natural measurements when the points are part of the training subset. When evaluating the pairwise representativeness, we found two types of effects: synergy and anti-synergy. The synergy was achieved with an increase in model accuracy when the pair entered the training subset. The anti-synergy manifested in a decrease informativeness of the point pair for modeling. The various sampling locations have different information and unequal meaning for feature interpolation. The scale-free network structures were found to have pairwise representativeness by RMSE .