Analysis and estimation of spectral response of biochemical components in vegetation leaves

Traditional chemical analysis of vegetation biochemical components is destructive and labor-intensive, and hyperspectral techniques can provide an effective means for rapid and non-destructive estimation of vegetation biochemical components.

Based on the LOPEX’93 data set, four transformed spectra (first-order differential DR, reciprocal 1/R, logarithmic logR, reciprocal logarithmic log (1/R)) were performed on the original spectra, and the trilateral parameters, spectral indices, and full-band spectral indices were extracted and the correlations with chlorophyll content (C_ab), dry matter content (LMA), and equivalent water thickness (EWT) were analyzed, and the extracted spectral features with high correlations were used as input variables, and the models were built and predicted by partial least squares regression (PLSR), support vector machine (SVM), and random forest (RF), respectively. The decision coefficient R² and root mean square error (RMSE) were used as indicators to determine the quality of the models.

The most effective method among the four spectral transformations was the first-order differentiation, in which the first-order differential bands D₇₄₂ and D₅₃₅ had strong spectral responses to C_ab; D₁₇₀₀, D₁₂₂₉, and D₂₁₉₂ had high responses to LMA; D₁₁₄₅, D₁₃₀₂, and D₉₅₅ had high responses to EWT. Although the trilateral parameters improved the correlation with chlorophyll to some extent, the correlation was not as good as the first-order differential spectra and spectral indices. Among the four full-band spectral indices constructed, the correlations of RI, DI, NDVI, and TVI with C_ab were 0.64, 0.66, 0.64, and 0.67, and the correlations were significantly enhanced compared with the conventional spectral indices. The above 8, 5, 5 spectral features about C_ab, LMA, and EWT were involved in modeling and estimation, and the best estimation models were SVM, PLSR, and RF with R² of 0.73, 0.75, and 0.93, and RMSE of 9.62, 0.001 274, and 0.000 971, respectively.

The first-order differential transform and full-band spectral index can improve the spectral response to biochemical components, and the constructed SVM, PLSR, and RF models can generally achieve the estimation of vegetation leaf C_ab, LMA and EWT, which provide a reference for refining the inversion of vegetation biochemical components from hyperspectral data.