Improving maize yield estimation by assimilating UAV-based LAI into WOFOST model

Abstract

ContextCrop growth models were widely applied for simulating the dynamic growth of crops at multi-scales. The data assimilation by integrating the remote sensing data retrieved crop parameters and crop models have showed great potentials for describing the crop growth and assessing the agricultural yields. ObjectiveThe purpose of this study was to integrates sequential observations of crop phenotyping traits from Unmanned Aerial Vehicles (UAV) remote sensing into World Food Studies (WOFOST) model to improve the simulation of crop growth processes. MethodsTwo years of Leaf Area Index (LAI) of summer maize retrieved from Unmanned Aerial Vehicles (UAV)-RGB images was assimilated into the WOFOST using the Ensemble Kalman Filter (EnKF). The sensitive crop parameters of WOFOST were firstly identified using the Extended Fourier Amplitude Sensitivity Test (EFAST) global sensitivity analysis approach, and then the parameters were adjusted and confirmed using the SUBPLEX optimization algorithm. The LAI data was assimilated into WOFOST model using EnKF by minimizing the differences between the UAV-retrieved LAI and crop-simulated LAI. Results indicated assimilating LAI into WOFOST model significantly improved the accuracy of maize yield prediction. ResultsCompared with non-assimilation, data assimilation reduced the Root Mean Square Error (RMSE) from 413 to 132 kg/ha for 2020, and from 392 to 215 kg/ha for 2021, respectively. Through the effects of different ensemble size and different time-point for data assimilation, it was obtained that the accuracy of yield prediction achieved the highest when ensemble size was 100 at reproductive growth stage. ConclusionsIntegrating UAV-based crop traits into WOFOST model using data assimilation (EnKF) could effectively improve the maize yield accuracy.