Multivariate statistical methods for estimating grassland leaf area index and chlorophyll content using hyperspectral measurements

Grassland habitat covers about one-quarter of the Earth’s land surface, providing significant contribution to the world’s total agricultural production, plant biodiversity, and carbon sequestration. The advent of hyperspectral remote sensing and the future launch of planned spaceborne hyperspectral missions will open up new possibilities over conventional multispectral RS to better quantify grassland characteristics. Hyperspectral data, while rich in information, presents a challenge for analysis due to its high dimensionality and multicollinearity. This present study investigated three promising high dimensional multivariate regression models namely partial least squares regression (PLSR), regularization and shrinkage method Lasso, and nonparametric Random Forest (RF) regression, to estimate grassland leaf area index (LAI) and chlorophyll using field canopy hyperspectral measurements (n=185). For each regression model, three spectral transformations namely continuum-removal, firstderivative, and pseudo-absorbance were evaluated.