Kolotii A. Regression models for winter wheat yield prediction with use of satellite data

Dissertation on Acquiring a Technical PhD Degree in Specialization 05.07.12 - Remote aerospace research. - Space Research Institute NAS Ukraine and SSA Ukraine, Kyiv, 2014. This research solved the problem of yield predicting for winter wheat in Ukraine with use of satellite data of different nature. Regression models for yield prediction based on time series of satellite products NDVI, FAPAR and VHI are constructed, the choice of informative features for predictive models is proved and it is shown that there is a temporal consistency of selected features for different types of predictors. It is shown that the accuracy of prediction based on time series of FAPAR and VHI is higher than for NDVI. RMS prediction error is about 0.7 t / ha, and the average prediction error across all regions is about 0.3 t / ha. A regional model and automated information technology for automated winter wheat yield prediction for the territory of Ukraine with use of satellite data at the regional level are developed. The technology is implemented as a Web service that provides within convenient user interface geospatial information to support decision making.