AUTHOR=Wang Ning , Yang Guang , Han Xueying , Jia Guangpu , Li Qinghe , Liu Feng , Liu Xin , Chen Haoyu , Guo Xinyu , Zhang Tianqi
TITLE=Study of the spectral characters–chlorophyll inversion model of Sabina vulgaris in the Mu Us Sandy Land
JOURNAL=Frontiers in Earth Science
VOLUME=10
YEAR=2023
URL=https://www.frontiersin.org/journals/earth-science/articles/10.3389/feart.2022.1032585
DOI=10.3389/feart.2022.1032585
ISSN=2296-6463
ABSTRACT=
As the dominant shrub community plant in the Mu Us Sandy Land, S. vulgaris is the key factor of ecological environment restoration in the Mu Us Sandy Land, It is of great significance to explore the estimation and inversion of content based on spectrum for ecological environment evaluation and intervention in Mu Us Sandy Land. The SVC HR-1024 portable feature spectrometer and SPAD 502 chlorophyll meter were used to study Mu Us Sandy Land of S. vulgaris. The best band is screened by correlation matrix method, the best vegetation index is screened by Structural Equation Modeling model, and then the best inversion model is established by different mathematical modeling methods. Results revealed that the vegetation indices and chlorophyll content were correlated, combining the six vegetation indices revealed that 610–690nm and 700–940 nm were the bands with the highest correlation. In the selection of optimal vegetation index, NDVI, ratio vegetation index and mNDVI perform best and are suitable for subsequent modeling. Of the four models, the partial least squares model had the best fitting effect (R2 > 0.91). The univariate linear regression model had the simplest processing procedure, but its accuracy was unstable (R2 = 0.1–0.9). multivariate stepwise regression accuracy is also appropriate (R2 > 0.8). The stability of BP neural network modeling is not high. Compare the four methods, PLS and multivariate stepwise regression have their own advantages, and the accuracy is higher, you can make a choice according to the demand as the late modeling method.