Estimation of vegetation Equivalent Water Thickness using hyperspectral data and partial least square regression

Abstract

Vegetation water content is an important parameter to evaluate vegetation vigor. Therefore, it is very important to timely understand the vegetation moisture status, especially in the fields of agriculture and forestry. The hyperspectral data can provide continuous spectral information and shows to be a promising tool for precisely describing vegetation water content. Commonly, vegetation water content is often expressed as Equivalent Water Thickness (EWT). In this paper, using LOPEX dataset, we mainly explored the strength of partial least square regression (PLSR) models based on different spectral transformations (original reflectance, logarithmic reflectance and the first derivative reflectance) to retrieve EWT from vegetation reflectance spectra. Also, the performance of different models for EWT estimation was compared. According to the results, all three approaches using PLSR achieve high precision to predict EWT using reflectance spectra. The PLSR models with the first derivative reflectance perform the best with an estimation precision of 0.963 for calibration & 0.938 for independent validation; and the PLSR models with original reflectance follow next with an estimation precision of 0.916 for calibration & 0.859 for independent validation; and the PLSR models with logarithmic reflectance rank last with an estimation precision of 0.914 for calibration & 0.827 for independent validation. From this study, PLSR demonstrates great potential to predict EWT from hyperspectral data and the PLSR models yield the best prediction when combined with the first derivative reflectance.