Identifying Hyperspectral Characters of Wetland Species Using In-situ Data

Wetland species identification is a very important process when determining wetland ecosystem types and wetland composition. Hyperspectral data with the merit of high spectral resolution have become a powerful tool for ground feature identification compared with multi-spectral data. This paper investigated the hyperspectral characters of wetland plants by considering the two prominent attributes of hyperspectral data: continuity and hyper-bands. In term of the continuity attribute, Derivative Reflectance (DR) and Continuum Removal (CR) method were used to extract key characters from the original spectra. In order to reduce spectral bands, firstly four factors such as slope (K), area of absorption curve (A), area of left wing (AL), and symmetry (S) were constructed, and secondly Mahalanobis Distance (MD) method was used to select the best bands for identifying plant species. The results showed that the First Derivative Reflectance (FDR), Second Derivative Reflectance (SDR) and Continuum Removal (CR) curves illustrated the variances of plant spectra from various ways. There are 5 bands for the original spectra, 5 bands for FDR curves, 6 bands for SDR curves, and 4 bands for CR curves were selected as better bands for species identification. All the selected bands are mainly located in the typical chlorophyll absorption and water absorption region. The constructed factors such as K1, K3, S3, Al, AL1, A2 and AL2 varies for different species which indicated that they can be used for species identification. However, for factors like K2, A3, AL3 can only be used to identify emerged, floating and submerged plant types. * Corresponding author. Tel: +86 10 6279 4119; Fax: 86 10 6279 4119; Email: [email protected]