The neural network ranking of agro-technologies by indices of soil microbiological activity and soil fertility: new possibilities of statistical analysis

Neural networks make it possible to extract previously inaccessible information from physiological and molecular genetic data and visualize implicit relationships. In the presented work we applied the neural network approach to analyze the data of Yu.M. Vozniakovskaya et al. (Sel'skokhozyaistvennaya Biologiya/Agricultural Biology, 1994) obtained in a long-term experiment (1962-1991, North-West region of the Russian Federation) on the influence of crop rotations and mineral fertilizers on microbiological and biochemical characteristics of sod-podzolic soils under potato and barley. The experiment, the aim of which was to identify microbiological indicators that most clearly characterize the level of soil fertility, determined the composition of soil microorganisms, measured the intensity of CO2 release, cellulose decomposition, activity of soil polyphenol oxidases and peroxidases as indicators of the intensity of plant residue humification, as well as invertases and ureases to assess the accumulation of nitrates and ammonium in the soil. On this basis, the authors attributed the most informative microbiological indicators of the level of effective soil fertility to the general soil biogenicity, species diversity of soil microorganisms, the ratio of trophic groups of microorganisms, and the graph of trophic relationships. In the present work, the analysis of fractal profiles of microbial physiological groups isolated on selective media demonstrated unique possibilities of statistical analysis using computational neural networks. Their application for processing empirical microbiological and physico-biochemical soil data reported by Yu.M. Vozniakovskaya et al., 1994 allowed determination of the specificity of the impact of the used agrotechnology elements on soil fertility under different crops. Soil microbiological activity and soil fertility are data characterizing heterogeneous biological objects? Soil microbial community (CSImicro index) and plants (CSIyield index), where CSI means Cognitive Salience Index. Therefore, to visualize the function CSIyield = f (CSImicro) we used neural network with the construction of Scale matrix the cells of which with coordinates CSIyield;CSImicro were filled with corresponding numbers of treatment variants. For this purpose, fractal profiles of soil microbial physiological groups of microorganisms obtained in the experiment were subjected to neural network analysis. As a result, form the Scale matrix, it was possible to estimate the function CSIyield = f (CSImicro) and the influence of fertilizer and crop rotation agrotechnologies on the intensity of humus accumulation in soil and soil fertility. It was found that the maximum soil fertility index CSIyield did not coincide with the maximum value of soil microbiological activity index CSImicro. The original indices proposed by us allow optimization of agrotechnologies for organ mineral fertilizers and bacterial preparations to obtain stably high yields of agricultural crops.