High mercury (Hg) affects biochemical-physiological characteristics of plant leaves such as leaf chlorophyll, causing refractive discontinuity and modifications in leaf spectra. Furthermore, the hyperspectroscopy provides a potential tool for fast non-destructive estimation of leaf Hg. However, there are few studies that have investigated Hg for wetland plants via hyperspectral inversion. In this study, reeds (Phragmites australis) leaf Hg concentration and hyperspectra were measured under different soil Hg treatment. Hg-sensitive parameters were identified by basic spectral transformations and continuous wavelet transformation (CWT). Inversion models were developed using stepwise multiple linear regressions (SMLR), partial least square regression (PLSR), and random forest (RF) to estimate leaf Hg. The results indicated that CWT improved the correlation of hyperspectra and leaf Hg by 0.020-0.227, and R2 of the CWT-related model increased by 0.0557-0.2441. In addition, Hg-sensitive bands were predominant at 600-750 (visible region) and 1500-2300 nm (mid-infrared), and Hg might modify leaves spectra primarily by affecting chlorophyll and water contents. Of the studied models, SMLR using normalized transformation (NR) and CWT (NR-CWT-SMLR) model (R2 = 0.8594, RMSE = 0.0961) and RF using NR and CWT (NR-CWT-RF) model (R2 = 0.8560, RMSE = 0.1062) suited for leaf Hg inversion. For Hg content < 1.0 mg kg-1, the former model was more reliable and accurate. This study provided a method for the estimation of Hg contamination in wetland plant and indicated that model-based hyperspectral inversion was feasible for fast and non-destructive monitoring.
Keywords: Estimation; Hyperspectral; Mercury; Model; Reed.