Estimation of goji berry (Lycium barbarum L.) canopy water content based on optimal spectral indices

Crop canopy water content (CWC) is a crucial parameter in determining crop photosynthesis rate, growth and yield. In comparison with traditional chemical analysis methods, hyperspectral remote sensing technology has the advantage of non-destructive and rapid acquisition of crop phenotypic parameters. To assess the applicability of spectral indices in monitoring crop water conditions in arid areas, this study focused on the Zhongning county and Huinong district in Ningxia Hui Autonomous Region, China. The main research objective was to develop a rapid and non-destructive technology for assessing the water status of the main cultivation of goji berry (Lycium barbarum L.) varieties (i.e. 'Ningqi No 1 ', 'Ningqi No 7 ' and 'Ningqi No 10 ') in Ningxia. Therefore, we constructed optimal spectral indices using hyperspectral data from field experiments conducted between June and September 2022, along with laboratory-determined leaf water content data. A simple linear regression algorithm was employed to establish models for estimating goji berry CWC. In addition, for the full spectra, the modelling accuracy of four multivariate statistical analysis algorithms (i.e. ridge regression, least absolute shrinkage and selection operator, principal component regression and partial least squares regression) was evaluated. The results showed the following: (1) compared with the original spectra (OS), the first-order derivative spectra (FD) exhibited a stronger correlation with goji berry CWC. Single-band models constructed based on the FD showed a more significant improvement in accuracy across different datasets, with a reduction in the root mean square error of prediction (RMSEP) by 8.838 %-35.494 % in the validation sets. (2) Compared with single-band models, dual-band spectral indices had higher accuracy in estimating goji berry CWC. Among them, the model based on the optimal FD spectral index outperformed the models based on the OS in the entire sample, Zhongning, Ningqi No 1 and Ningqi No 10 datasets, exhibiting a reduction in RMSEP ranging from 7.483 % to 45.424 %. We also observed that the optimal spectral indices constructed using visible light bands such as blue (446 and 448 nm), green (502, 504 and 518 nm), yellow (577, 578 and 580 nm) and red edge (692 and 704 nm) can effectively estimate the CWC of goji berry. (3) In terms of modelling with full spectra, the coefficient of determination (R2) values for the models of the four multivariate statistical analysis algorithms were all above 0.599. Among them, the Lasso algorithm exhibited the highest modelling accuracy for the OS of the Ningqi No 7 dataset (R2 = 0.773 and RMSEP = 0.883 %), surpassing the dual-band spectral indices. However, it was observed that in modelling for the other five datasets, the accuracy of multivariate statistical analysis algorithms was not as high as that of the dual-band spectral indices. The results of this study have significant implications for the efficient utilisation of agricultural water resources, irrigation decision-making and crop growth monitoring in arid regions.