Abstract:

In recent 10 years, the dynamic and refined yield forecast have been promoted accompanied with the development of the agrometeorological observation technology, the remote sensing monitoring technology, crop model simulation technology, and the application of intelligent grid meteorological. All these have improved the accuracy of yield forecast and played an important role to ensure national food security. In this paper, from the perspective of the technical progress of crop yield forecasting and the test of forecast results in National Meteorological Center over the past decade, the statistical models based on key meteorological factors, meteorological influence index, climatic suitability index, multi model integrated forecasting, as well as the crop dynamic yield forecasting technology based on crop model simulation and multi-source data fusion, are systematically introduced. The forecast results of early rice in the main producing provinces in 2020 and in different periods in Fujian province showed that the accuracy of different mathematical statistical forecasting models was generally quite close to each other, ranging between 90.8% and 99.8%, and the climatic suitability index  outperformed the other two methods. The results of the forecast of the main single rice-producing counties in Jiangsu province indicate that the county scale yield forecasting accuracy based on the climate suitability index method was generally high. Specifically, the July 20 forecasts exhibited accuracy rates between 73.9% and 98.1%, while the August 20 forecasts showed rates between 80.4% and 98.3%. The impact index based on daily meteorological data, to a certain extent, can quantitatively assess the effect of meteorological conditions on crop yields at different time scales. Crop Growth Simulating and Monitoring System in China constructed by using different crop models could carry out county-level and provincial-level yield forecasting of different crops, and the forecast accuracy was relatively stable. The accuracy rates for different initial forecast dates were consistently maintained between 88.4% and 97.4%, while Shandong and Hebei province exhibited higher rates than those in other provinces. It is feasible to carry out yield forecast at national level based on the observed yield series and the new yield series could provide new data support for yield forecast in National Meteorological Centre. The county-level yield forecast based on remote sensing data and machine learning has good prediction accuracy, which can improve the technical of yield forecasting. The adoption of suitable yield prediction methodologies can significantly enhance forecast accuracy for diverse crops in various provincial regions.

Key words: National yield forecast, Statistical models, Spatio-temporal refined forecast model, Multi-source data forecast model