Application of Resonon Pika L on winter wheat yield estimation
Crop yield is one of the most critical issues affecting national economic development and food security. Accurate and timely crop yield estimation is essential for precision agriculture and sustainable development, and can provide strong support for agricultural decision-making and management. Traditional crop yield estimation heavily depends on ground field surveys, which are costly, time consuming and prone to large errors. 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.
Based on this, in the attached article, a group of Chinese scientists from Peking University and Henan Agricultural University introduced crop growth model simulation to increase the number of samples, then using the random forest algorithm to build a crop yield estimation model suitable for the UAV imagery. The primary objectives are: (1) developing a new winter wheat yield estimation model (the CW-RF model) using the random forest regression algorithm in combination with the CERES-Wheat model suitable for UAV hyperspectral imagery; (2) testing the performance of the LAI and LNC retrieval methods for the hyperspectral sensor mounted on the UAV; and (3) assessing the potential for UAV remote sensing in yield estimation and analyze the possible error sources.
The study was conducted at the experimental station of National Agriculture Production base for high quality wheat, which is located in Luohe, Henan Province, China, (113°52′54″ E, 33°41′59″ N) at an altitude of 63 m (Fig. 1). A hyperspectral imaging instrument, Pika L (Resonon Inc., Bozeman, MT, USA) was mounted on the UAV to acquire hyperspectral images (Fig. 2(c)). Field measurements and UAV flights were conducted simultaneously. The collected data mainly included the LAI, LNC, and the spectral reflectance of the canopy during the growth stage of winter wheat.
Fig. 1. Overview of the study area. (a) Geographical location of the experimental site. (b) Illustration of field data collection. (c)RGB orthomosaic imagery collected by a UAV on March 22, 2019 showing the spatial location of the 40 plots.
Fig. 2. UAV systems and hyperspectral sensor. (a) DJI M600 Pro Hexacopter platform. (b) Autonomous flight control system and telemetry and telecontrol system. (c) Pika L hyperspectral sensor.
Fig. 3. Concept map for winter wheat yield estimation based on the integration of the CERES-Wheat model and random forest algorithm.
[Results]
The field validation in Luohe, Henan showed that the root-mean-squared error 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 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.
Fig. 4. LAI spatial distribution map retrieved from UAV-based hyperspectral data. (a) Jointing stage. (b) Heading stage
Fig. 5. Comparison between the retrieved and measured LAI at the jointing and heading stages. (a) Jointing stage. (b) Heading stage.
Fig. 6. LNC spatial distribution map retrieved from UAV-based hyperspectral data. (a) Jointing stage. (b) Heading stage.
Fig. 7. Comparison between the retrieved and measured LNC at the jointing and heading stages. (a) Jointing stage. (b) Heading stage.
Fig. 8. Relationship between measured and estimated winter wheat yield.
[Conclusion]
Estimating winter wheat yield early and accurately at field plot scale is of great significance for field management and agricultural operation. In this article, a winter wheat yield estimation model, the CW-RF model, was established based on the CERES-Wheat model simulation data using the random forest regression algorithm. A total of 4200 groups of samples were simulated using the CERES-Wheat model, involving four different weather parameters, five different soil parameters, ten different wheat cultivars, seven different nitrogen fertilizer parameters and three irrigation parameters that were based on the current situation in the North China Plain. The jointing and heading stages were identified as the two key growth periods, and the LAI and LNC were chosen as the main growth parameters for winter wheat yield estimation. Therefore, the LAI and LNC of winter wheat during the jointing and heading stages retrieved from UAV hyperspectral images were input into the CW-RF model to estimate winter wheat yield.
Field validation shows that the CW-RF model has a high accuracy and could provide an accurate yield estimation at the field plot scale. The CERES-Wheat model simulation could solve the problem of few samples in the application of random forest algorithm for crop yield estimation, and could also ensure estimation accuracy. The model performed well in two typical areas of the North China Plain, Luohe (Henan) and Xiaotangshan (Beijing). Compared with the traditional winter wheat yield estimation model, the CW-RF model owns a more general applicability. More UAV flights and ground measurement experiments will be conducted in other locations to confirm the applicability of the model to the North China Plain.
The prior knowledge and empirical methods for the LAI and LNC retrieval may dilute the transferability of the model and reduce the physical interpretability of the CW-RF model. The quantitatively remote sensed models for LAI and LNC deserve future study.
As the simulation errors and uncertainty of the CERES-Wheat model could be transferred to the CW-RF model, and this effect was positively correlated, more field measurements should be introduced into the CW-RF model to reduce the errors and uncertainties. In addition, simulation and analysis results showed that the UAV hyperspectral data could significantly improve the winter wheat LAI and LNC inversion accuracy, and further improve the accuracy of winter wheat yield estimation.
6377827554221702492828547.pdf