Efficiency of crop yield forecasting depending on the moment of prediction based on large remote sensing data set

Agricultural yields can be predicted from detailed multi-year remote sensing image sequences using measured features of vegetation conditions. In this paper, the dependency between the moment of prediction and the accuracy of the forecast is studied. The linear model is selected as a basic approach of yield forecasting. Then, the model is extended with non-linear components (factors) in order to improve the accuracy of the forecasts. The extensions take into consideration long-term technological advances in agricultural productivity as well as regional variations in yields (fertility of the lands). The accuracy of the model has been estimated based on the time period between the moment of the forecast formation and the harvest time.