Application of machine learning methods for crop rotation selection in organic farming system

Abstract

This study demonstrates the possibility of crop rotation selection based on the assessment of productivity and sustainability of crop production under different atmospheric moisture conditions. The study considers 8 crop rotations oriented to grain production. The data obtained in long-term field experiments in the forest-steppe of the Novosibirsk region were used. As a result of the implementation of the decision tree (CART) and the use of ensemble algorithm (Random Forest) the construction of a model characterized by a fairly high predictive ability was performed. Standardized Precipitation Index was chosen as the main predictor characterizing atmospheric moistening in different periods of vegetation. The most stable from the point of view of stability of crop yield – grain-fallow with winter rye, grain-fallow with legumes (vetch-oats), in conditions of manifestation of atmospheric drought of different severity were selected. The possibility of using machine learning methods (CART, Random Forest) as effective tools in the selection of crop rotation for sustainable grain production without the use of chemicalization in soil and climatic conditions of Siberia, as well as the assessment of possible risks in the transition of crop production to organic technologies were scientifically substantiated.