Integration of crop growth model and random forest for winter wheat yield estimation from UAV hyperspectral imagery

Siqi Yang, Ling Hu, Haobo Wu, Huazhong Ren, Hongbo Qiao, Peijun Li, Wenjie Fan
2021 IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing  
Accurate and timely crop yield estimation is critical for food security and sustainable development. The rapid development of unmanned aerial vehicles (UAVs) offers a new approach to acquire high spatio-temporal resolution imagery of farmland at a low cost. In order to realize the full potential of UAV platform and sensor, machine learning has been introduced to estimate crop yield, but the shortages of field measurements have troubled researchers. In this study, the CW-RF model, a new wheat
more » ... ld estimation model suitable for the North China Plain, was established using random forest, and the crop growth model (the CERES-Wheat model) was chosen to simulate abundant training samples for random forest at field plot scale. According to CERES-Wheat model simulation, the leaf area index (LAI) and leaf nitrogen content (LNC) at the wheat jointing and heading stages were selected as the most sensitive parameters, and were retrieved from UAV hyperspectral imagery using the directional second derivative (DSD) and angular insensitivity vegetation index (AIVI) methods respectively. Then the retrieved LAI and LNC results were input into the CW-RF model to estimate winter wheat yield. The field validation in Luohe, Henan showed that the root mean squared error (RMSE) of the retrieved LAI and LNC were 6.27% and 12.17% at jointing stages, 9.21% and 13.64% at heading stages, respectively. The RMSE of estimated yield was 1,008.08 kg/ha, and the mean absolute percent error (MAPE) of estimated yield was 9.36%, demonstrating the available of the CW-RF model in wheat yield estimation at field plot scale. Apart from Luohe, validations in some other fields (e.g. Xiaotangshan, Beijing), prove the wide applicability of the CW-RF model. In addition, the UAV hyperspectral data were found to significantly improve the retrieval accuracy, and further improve CW-RF model estimation accuracy. In conclusion, this study showed that the CERES-Wheat model simulation can be important data source for machine learning-based wheat yield estimation model at field plot scale, and the hyperspectral sensor mounted on a UAV is a feasible remote sensing data acquisition mode for winter wheat growth monitoring and yield estimation. Index Terms-wheat yield estimation, the CERES-Wheat model, random forest, unmanned aerial vehicle (UAV), hyperspectral remote sensing
doi:10.1109/jstars.2021.3089203 fatcat:eog3tfykv5g6nodr7bchbqseuu