Methodological aspects of forecasting winter rye yield on lands withdrawn from agricultural cultivation

The study was conducted to compare the suitability of two mathematical methods for the procedure of forecasting the winter rye yield within abandoned lands. The training set for adjusting the forecast models (regression and neural network analyses) was the results of long-term monitoring of the winter rye yield on an agroecological transect laid within a terminal moraine hill near the city of Tver. The regression equations and perceptrons obtained for different climatic conditions, describing the dependence of rye yield on landscape conditions, were used to forecast its productivity in abandoned lands located in the same region. The developed forecasts based on regression and neural network models differ only in details. Maps of the forecast winter rye yield, created on the basis of various mathematical approaches, indicate the suitability of the studied territory for growing this crop – its forecast average weighted yield within its limits fluctuates from 1.52 to 2.61 t/ha. The optimal location for rye crops based on the map created based on the forecast results will be the landscape of the moraine-glacial plain, and areas unfavorable for growing this crop are located within the outwash plain and the Volga River valley. For the purposes of adaptive landscape farming in the Non-Black Earth Region, it is necessary to recommend the combined use of regression and neural network analysis to obtain the most reliable and informative forecasts. The creation of maps of the predicted yield of winter rye within the boundaries of an abandoned site based on the available archival data has not only a methodological but also an applied meaning, since this procedure allows us to assess its suitability for growing this crop and, thus, answer the question of the advisability of conducting expensive expeditionary research on it.

adaptive landscape farming systems, monitoring, forecast, regression analysis, neural networks