Neural network forecasting of productive moisture reserves in soil before sowing grain crops

Abstract

Soil moisture reserves are one of the main factors limiting the growth and development of plants during the vegetation period of agricultural crops, their preliminary assessment plays a major role in the planning of agrotechnical measures for the spring summer period, which in turn affects the yield and efficiency of agricultural production. The article proposes a method of predicting moisture content in a meter layer of soil before sowing grain crops, based on the construction and training of an artificial neural network. To build an artificial neural network we used the data of multifactorial field experience of the Siberian Research Institute of Crop Production of SFNCA RAS (central forest-steppe). The data include the results of studies of agrophysical and agrochemical factors in a four-field grain-fallow crop rotation from 1996 to 2018. T The constructed artificial neural network has the architecture of a multilayer perseptron consisting of an input, hidden and output layer. The input layer accepts data in the form of predictors, namely: predecessor, tillage method, weather conditions, autumn productive moisture reserve of the previous year, i. e. factors affecting the predicted variable. The hidden layer transforms and processes input data, while the output layer generates model predictions. The developed artificial neural network demonstrated a fairly high accuracy of forecasting. The total percentage of reliably predicted observations was 80.6 %. The ROC analysis performed to evaluate the predictive ability of the neural network showed that the area under the ROC curve for each category was close to 1. This indicates that the neural network has high predictive power and is able to accurately identify the different categories of the target indicator.