Summer Maize Yield Estimation Based On Assimilation Of UAV Remote Sensing Data Into WOFOST Model

Yield estimation at field scale can provide important assistance for agricultural operation and management.Crop models can simulate crop growth.Unmanned aerial vehicle(UAV)remote sensing can quickly and accurately acquire crop growth status in medium-sized areas.Date assimilation method combines with remotely sensed data and crop growth model has become a significant hotspot in crop yield forecasting.In this paper,we took summer maize as research object,obtained multi-spectral image of research area by remote sensing of UAV,completed LAI inversion of summer maize by regression method,and WOFOST model was used to simulate the growth period of summer maize,then we assimilated UAV remote sensing data with WOFOST model by ensemble Kalman filter.Meanwhile,we use data assimilation combined historical meteorological data to forecast the yield.The main results are as follows:(1)Under each treatment,the correlation between NDVI and LAI is the highest,while the correlation between SAVI and LAI was lowest.MSAVI2 was negatively correlated with LAI.Modeling results show that LAI-NDVI models perform better under all treatments.Under CK treatment,LAI-NDVI model R~2 is 0.84,RMSE and RE of model predicted value and measured value are 0.21 and 11.04%respectively.LAI-NDVI model R~2 was 0.70 under N1 treatment,RMSE and RE of model predicted value and measured value were 0.29 and13.75%respectively,and LAI-NDVI model R~2 was 0.63 under N2 treatment,while RMSE and RE of model predicted value and measured value were 0.28 and 8.98%respectively.Under N3 treatment,LAI-NDVI model R~2 is 0.79,RMSE and RE of model predicted value and measured value are 0.25 and 9.95%respectively.(2)Development parameters and soil parameters of WOFOST model were adjusted by field measured data.We used PEST to adjust more than ten crop parameters with high sensitivity.Among them,the quota consistency index d of each validation index is more than0.9,and n RMSE is less than 30%.The simulation accuracy of yield after assimilation were improved,and the relative error RE were less than 3%.Remote sensing data assimilation can well correct the simulation results of the model.We also analyzed the effects of different assimilation schemes on simulated yield after assimilation.It was clearly that the optimum assimilation times should be more than or equal to three times,from anthesis to maturity is the best assimilation stage.(3)Combined with historical meteorological data and En KF algorithm can effectively predict summer maize yield.Under different treatments,the coefficient of variation of yield forecast set on September 21 was the smallest.The results showed that the relative error of predicted and measured yield on September 6 was the smallest.Among them,RE of predicted yield and measured yield under CK treatment is 2.44%,RE of predicted yield and measured yield under N1 treatment is 0.94%,RE of predicted yield and measured yield under N2 treatment is 0.84%,RE of predicted yield and measured yield under N3 treatment is 1.93%.The forecast yield after 4 assimilations is more accurate than that of the other four assimilations.In summary,this study used UAV remote sensing and crop model assimilation to estimate summer maize yield,and the yield prediction method using historical weather data-driven model combined with data assimilation achieved good accuracy.Combining remote sensing information and using ensemble Kalman filtering algorithm can realize the estimation of summer maize yield of WOFOST model.