Regression Forecasts of Irrigated Winter Crop Yields Us-ing Satellite Vegetation Indexеs: Models, Predictors and Experiments

Abstract

The technologies of agrometeorological crop forecasts (ACF), based on data on the normalized vegetation index (NDVI) are an important element of the modern agricultural industry. Aim. To establish the heuristic yield model and approaches to the development of models of regression forecasts, including the ACF predictors selection procedure by exploring satellite Moderate Imaging Spectroradiometer (MODIS) NDVI data and conduct experimental forecasting. Materials and methods. The official yield statistics of irrigated winter wheat and barley in Diwaniyah province of Iraq and the NDVI MODIS observation for 2001–2019 are used. It is proposed to choose a two-component heuristic yield model containing a yield trend, due to a relatively slow change in crop cultivation technology and a climatic component associated with fluctuations in biological productivity due to the effects of weather conditions. Results. Using of heuristic model as background, an object-oriented approach to the choice of ACF regression model and predictor selection is developed. Firstly, we use NDVI semi-quantitative connection with crop coverage and crop leaf indexes to determine NDVI evolution according with the wheat and barley growing stages. Then, it is shown that in the province level ACF, as the original predictors should choose the NDVI time-series derived on the first and second half of February for three distinct grain-producing regions of the pro-vince. Experiments have shown that the satisfactory quality of the regressive ACF of both cultures can be achieved with 2–3 different original non-collinear predictors by their combination with the last year's yield or by inclusion of linear or quadratic dependencies. Conclusion. Wheat forecast with a relative error of 10% is obtained only by special selecting of time interval to train model and by control the parameters of the auto-regressive predictor. The high quality of the barley forecasting models is due to the fact that the variability of barley yields is dominated by the climatic component. The developed object-oriented approach can be adapted to the conditions of rainfed agriculture and to forecast of yield of other crops.