Modeling crop rotation productivity using an adaptive neuro-fuzzy inference system

The study was conducted with the goal of developing a model and forecasting crop rotation productivity using adaptive neuro-fuzzy inference. The work utilized data from long-term field experiments (data from 9 types of crop rotations focused on grain production) conducted by the Siberian Research Institute of Agriculture and Chemicalization of Agriculture (Siberian Federal Scientific Centre of Agro-BioTechnologies of the Russian Academy of Sciences) from 1999 to 2019. During the study, an artificial neural network (ANN) training algorithm was applied using a hybrid optimization method, combining the least squares method and backpropagation, to set up fuzzy rules with appropriate membership functions based on input and output data. Based on the use of adaptive neuro-fuzzy modeling and the MATLAB development environment, ANFIS crop rotation productivity model has been developed. The ANFIS rules formed during training allow for fairly accurate determination of significant factor combinations that influence the productivity of the given crop rotations. In predictive modeling of three types of crop rotations, the significant role of winter crops and crop rotation elements in enhancing crop rotation resilience to adverse atmospheric moisture conditions and improving the effectiveness of agrochemical application was revealed. A comprehensive analysis using various accuracy metrics (coefficient of determination – 0.78; root mean square error – 5.66; mean absolute error – 4.31; mean absolute percentage error – 20.07 %) indicates the model has good predictive ability. The developed ANFIS model demonstrates a strong ability to account for complex nonlinear relationships between the features influencing crop rotation productivity and can be used in production decision-making for short- and long-term planning.