Zhonglin Wang , Xianming Tan c, Yangming Ma, Tao Liu,Limei He,Feng Yang, Chuanhai Shu,Leilei Li,Hao Fu , Biao Li,Yongjian Sun,Zhiyuan Yang
, Zongkui Chen ,Jun Ma
Computers and Electronics in Agriculture,Volume 221, June 2024,29 April 2024(IF=8.3)
https://www.sciencedirect.com/science/article/pii/S0168169924003661
Abstract
Predicting rice grain yield using multi-source remotely sensed data is crucial for improving prediction accuracy, optimizing nitrogen management, and advancing precision agricultural development. However, the feasibility and reliability of using multi-source remotely sensed data to predict the grain yield remain unclear. Therefore, this study aimed to explore the possibility of providing rice leaf chlorophyll content (LCC) estimations and predictions of the grain yield using multi-source remotely sensed data. Two rice field experiments were conducted with various rice cultivars and nitrogen rates, and a field spectrometer and an unmanned aerial vehicle (UAV) equipped with a digital camera were employed to acquire the spectral reflectance and red–green–blue (RGB) images at the tillering, jointing, and full-heading stages. Destructive sampling was then conducted to measure the LCC and grain yield. The LCC was used as a bridge to develop remotely sensed prediction models for the grain yield. First, the linear relationship between grain yield and the LCC was determined at the tillering, jointing, and full-heading stages. A multiple linear regression (MLR) model was then developed to predict grain yield using multi-temporal LCCs at three growth stages. Second, multiple stepwise regression, support vector regression, and back propagation neural network were used to evaluate the estimation performance of spectral reflectance, RGB image data, and their combination for LCC. Third, the most accurate LCC estimation model was selected and coupled with the linear and MLR models of grain yield. The results showed that grain yield was significantly and positively related to the LCC at the tillering, jointing, and full-heading stages, and that the MLR model of grain yield achieved the best estimation accuracy using multi-temporal LCCs. The fusion models established by combining spectral reflectance and RGB image data improved LCC estimation accuracies. Using multi-growth stages, the most accurate predictions of grain yield were obtained from LCC estimation models (R2 = 0.698, RMSE = 0.742 t ha−1, rRMSE = 9.004 %) compared to those using single growth stages. Our study concluded that multi-source remotely sensed data fused from ground-based spectral reference and UAV-based RGB images can better predict and explain the grain yield for both single and multi-growth stages. This study provides a novel method of estimating the crop chlorophyll content and grain yield.
Keywords: Rice Grain yield;Leaf chlorophyll content;Multi-source remote sensing; Spectral reflectance