Retrieval of betalain contents based on the coupling of radiative transfer model and SVM model

Abstract

Betalain (Bt) is as a class of edible natural pigments and important biochemical aparmeter can explain the physiological response and the resistance of plants caused by different environmental stress factors. Hyperspectral spectroscopy has been widely used to estimate plant pigments contents accurately and non-destructively. However, there is very little research about hyperspectral remote sensing estimation and inversion of betalain contents. The algorithm of Bt content inversion with viable universality is the key to improving the practicality of quantitative remote sensing. Therefore, in this study, based on the radioactive transfer mechanism and SVM model, the optical characteristics of Bt and other factors of Suaeda salsa are analyzed, and a coupling model (PROSAIL + SVM) of simulated canopy reflectivity and factor contents is established. The model was then applied to the remote sensing inversion. Results demonstrated that Bt content was sensitive to the spectral range between 460 and 592 nm, particularly between 530 and 550 nm where the sensitivity index reached 0.7. Optimized indices; NDSI473nm,475nm, RSI473nm,475nm and NDPI473nm,475nm, calculated by simulated spectral reflectance had a significant correlation with Bt content (R= ±0.80, p < 0.001). Moreover, simulated Sentinel-2A reflectance of the Blue (B2) and Green (B3) bands showed sensitivity of 66% and 68%, respectively. Spectral indices NDPI b2, b3, NDSI b2, b3, RSIb2, b3 showed correlation of ± 0.71. The PROSAIL + SVM model developed from simulated hyperspectral reflectance, estimated Bt content with high R2 (0.82), indicating that the precision of the model was higher and the universality was stronger. A model composed of optimized indices (NDSI473nm, 475nm, RSI473nm, 475nm and NDPI473nm, 475nm) showed promising estimation ability with low RMSE (0.39 μg·cm−2) with R2 = 0.78 and RPD = 1.99. When the PROSAIL + SVM model was extended to multispectral images (Sentinel-2A), the model estimated the Bt content at moderate to good levels (R2 = 0.68). The results indicated that optimized hyperspectral and multispectral NDPI, NDSI, RSI have significant correlation with betalain content of S. salsa, suggesting these indices might be used to quantify Bt content with high accuracy, thereby alleviating the ill-posed inverse problem and improving sensitivity. Not only would this enhance the application of quantitative remote sensing technology in betalain research, but the results also pose positive application for precision agriculture in detecting Bt pigments in crops, as a surrogate for evaluating the health condition of crops.