The identification of species is essential to ecosystem monitoring and assessment. Remote sensing can provide robust and repeatable spectral measurements that are spatially continuous across ecosystems. Typically, the remote sensing of vegetation has relied on the measurement of leaf spectra, or the measurement of mixed spectra that include leaves, flowers and other components such as stems and soils. This study compares the ability to discriminate between species using pure spectra of leaves and flowers using herbaceous vegetation from a prairie fen as a case study. Spectral data of leaves and flowers were collected from 22 species using a handheld spectroradiometer with a spectral resolution of 1 nm from 475 nm to 900 nm. The use of continuous wavelet transformation was explored to enhance detection of shape in the spectral signatures. Due to the inherent high dimensionality of spectroscopy data, a forward feature selection algorithm was used to identify the best sets of individual bands and continuous wavelet transformed features. The results show that flowers consistently provide better species discrimination in spectral space than leaves in terms of both overall discrimination and efficiency (i.e. consistently requiring fewer features to achieve maximum spectral discrimination). For both transformed leaf and flower spectra, most of the selected features were found to be between 500 nm and 600 nm, indicating the importance of this region to differentiate between species when measuring both leaves and flowers. These results indicate the potential to map vegetation using the spectra of flowers as the availability of ultra-high-resolution imagery from low-altitude unmanned aerial remote sensing platforms increases.
|Journal||Remote Sensing Letters|
|State||Published - Oct 15 2014|