Estimating leaf mercury content in Phragmites australis based on leaf hyperspectral reflectance

• Main Authors: Weiwei Liu, Mengjie Li, Manyin Zhang, Daan Wang, Ziliang Guo, Songyuan Long, Si Yang, Henian Wang, Wei Li, Yukun Hu, Yuanyun Wei, Hongye Xiao
• Format: Article
• Language: English
• Published: Taylor & Francis Group 2020-12-01
• Series: Ecosystem Health and Sustainability
• Subjects: mercury pollution; reed leaf; hyperspectrum; inversion model; nondestructive monitoring
• Online Access: http://dx.doi.org/10.1080/20964129.2020.1726211

Introduction: High mercury (Hg) concentrations affect the chlorophyll in leaves, thereby modifying leaf spectra. Hyperspectra is a promising technique for the rapid, nondestructive evaluation of leaf Hg content. In this study, we investigated Hg contents and reflective hyperspectra of reed leaves (Phragmites communis) in a gold mining (Jilin province, China). Spectral parameters sensitive to Hg content were identified through basic spectral transformations, continuous wavelet transformation (CWT), and spectral indices techniques. Leaf Hg inversion models were developed using stepwise multiple linear regression, partial least squares regression, and random forest algorithms.Outcomes: The results indicated that: 1) leaf Hg content decreased with increasing distance from the mine: Jiapigou (JPG) > Erdaocha (EDC) > Laojingchang (LJC) > Erdaogou (EDG) > Lingqian (LQ) > Weishahe (WSH). 2) Hg–sensitive wavelengths were primarily in the visible region; CWT increased the correlation between hyperspectral data and leaf Hg content, and improved the regression and accuracy of inversion; 3) the continuum removal–CWT–stepwise multiple linear regression was better for estimating low leaf Hg content; while the differential spectral index–partial least squares regression was better for estimating high leaf Hg content.Conclusion: These hyperspectral inversion methods could be used for rapid, nondestructive monitoring of wetland plants.