Estimation of Rice Aboveground Biomass by Combining Canopy Spectral Reflectance and Unmanned Aerial Vehicle-Based Red Green Blue Imagery Data

Zhonglin Wang; Yangming Ma; Ping Chen; Yonggang Yang; Hao Fu; Feng Yang; Muhammad Ali Raza; Changchun Guo; Chuanhai Shu; Yongjian Sun; Zhiyuan Yang; Zongkui Chen; Jun Ma

doi:10.3389/fpls.2022.903643

Estimation of Rice Aboveground Biomass by Combining Canopy Spectral Reflectance and Unmanned Aerial Vehicle-Based Red Green Blue Imagery Data

Front Plant Sci. 2022 May 27:13:903643. doi: 10.3389/fpls.2022.903643. eCollection 2022.

Authors

Zhonglin Wang^{1

2}, Yangming Ma^{1

2}, Ping Chen³, Yonggang Yang^{1

2}, Hao Fu^{1

2}, Feng Yang³, Muhammad Ali Raza³, Changchun Guo^{1

2}, Chuanhai Shu^{1

2}, Yongjian Sun^{1

2}, Zhiyuan Yang^{1

2}, Zongkui Chen^{1

2}, Jun Ma^{1

2}

Affiliations

¹ Rice Cultivation Laboratory, Rice Research Institute, Sichuan Agricultural University, Chengdu, China.
² Crop Ecophysiology and Cultivation Key Laboratory of Sichuan Province, Chengdu, China.
³ Sichuan Engineering Research Center for Crop Strip Intercropping System, Chengdu, China.

Abstract

Estimating the aboveground biomass (AGB) of rice using remotely sensed data is critical for reflecting growth status, predicting grain yield, and indicating carbon stocks in agroecosystems. A combination of multisource remotely sensed data has great potential for providing complementary datasets, improving estimation accuracy, and strengthening precision agricultural insights. Here, we explored the potential to estimate rice AGB by using a combination of spectral vegetation indices and wavelet features (spectral parameters) derived from canopy spectral reflectance and texture features and texture indices (texture parameters) derived from unmanned aerial vehicle (UAV) RGB imagery. This study aimed to evaluate the performance of the combined spectral and texture parameters and improve rice AGB estimation. Correlation analysis was performed to select the potential variables to establish the linear and quadratic regression models. Multivariate analysis (multiple stepwise regression, MSR; partial least square, PLS) and machine learning (random forest, RF) were used to evaluate the estimation performance of spectral parameters, texture parameters, and their combination for rice AGB. The results showed that spectral parameters had better linear and quadratic relationships with AGB than texture parameters. For the multivariate analysis and machine learning algorithm, the MSR, PLS, and RF regression models fitted with spectral parameters (R² values of 0.793, 0.795, and 0.808 for MSR, PLS, and RF, respectively) were more accurate than those fitted with texture parameters (R² values of 0.540, 0.555, and 0.485 for MSR, PLS, and RF, respectively). The MSR, PLS, and RF regression models fitted with a combination of spectral and texture parameters (R² values of 0.809, 0.810, and 0.805, respectively) slightly improved the estimation accuracy of AGB over the use of spectral parameters or texture parameters alone. Additionally, the bior1.3 of wavelet features at 947 nm and scale 2 was used to predict the grain yield and had good accuracy for the quadratic regression model. Therefore, the combined use of canopy spectral reflectance and texture information has great potential for improving the estimation accuracy of rice AGB, which is helpful for rice productivity prediction. Combining multisource remotely sensed data from the ground and UAV technology provides new solutions and ideas for rice biomass acquisition.

Keywords: aboveground biomass; machine learning; rice; texture; unmanned aerial vehicle; vegetation indices; wavelet features.