Research On Wheat Yield Estimation Method Under Abnormal Climate Based On Random Forest

In order to explore the impact of climate change on crop yield,obtain a yield estimation model satisfying the prediction accuracy under abnormal climate conditions,this paper used different methods to fit the trend yield,screened out four major types of meteorological disasters,then established the corresponding meteorological disaster index respectively,combining with remote sensing vegetation index,established wheat yield estimation models in the five wheat districts using the random forest algorithm,accuracy verification was carried out in meteorological disaster years,finally analyzed the characteristics of the importance of different wheat input variables.Through the test results and analysis,the following conclusions are drawn.(1)The 3A,5A linear moving average method and HP filtering method were used to perform trend yield fitting.The actual yield was subtracted from trend yield to obtain meteorological yield.The random forest algorithm was used to perform regression for various meteorological factors and meteorological yield,and the fitting accuracy of trend yield was indirectly verified by analyzing regression accuracy.Through comprehensive comparison of R~2,root mean square error and mean absolute error of each wheat area model,it is concluded that 3A linear moving average method has a better effect in fitting trend yield.(2)Using four types of meteorological disaster indexes,remote sensing index and trend yield per unit area as input variables and actual yield per unit area as output variables,a random forest yield estimation model was established in the five major wheat regions of China and its accuracy was verified.The results showed that the regression model of each wheat area had high fitting accuracy,and the determination coefficient R~2 reached above 0.95.The mean relative error,root mean square error and mean absolute error of the verified samples were all lower than 0.073,26.778 jin per unit area and 17.554 jin per unit area respectively.The average relative errors of actual yield per unit area and predicted yield per unit area in disaster years in all counties were lower than 0.060,and the relative errors of actual yield per unit area and predicted yield per unit area in all counties were lower than 0.049.(3)The trend yield of each wheat area had the highest importance in all parameters,reaching above 0.9,which was the most important feature affecting the final wheat yield.The SPEI drought index has a high ranking of importance in different wheat regions,indicating that drought is still an important factor affecting wheat production in five wheat regions.The importance of dry hot wind index is higher in spring wheat area of northwest and north,which is in line with the local characteristics of dry hot wind mainly occurring in Hexi and Huang-Huai-Hai region.The importance of monthly mean precipitation anomaly is related to the growth stage of wheat in each wheat area.Seedling stage,jointing stage and heading and filling stage of wheat are the key periods for water demand,and the importance of the corresponding monthly mean precipitation anomaly is relatively high.The importance of monthly mean temperature anomaly in wheat fields was not significantly different.The importance of NDVI at jointing and heading stage of wheat in each wheat area is relatively higher,and the growth condition at the early stage of wheat planting can not be ignored.