Abstract:

Grassland ecosystem plays an important role in the development of animal husbandry, soil and water conservation, and maintenance of the ecological balance. Therefore, it is of great significance to perform real-time monitoring of degradation of grasslands. The technology of hyperspectral remote sensing can greatly improve precision of the identification of degraded vegetation structure in the process of grassland degeneration, and opens up a new field in the study of grassland degradation. In using the technology to identify degraded vegetation structure, it is very important to choose and extract characteristic bands. To that end, based on the hyperspectral data measured in field, spectral reflection curves of three species of degradation indicator plants and two dominant species in Bashang Region were processed with the logarithmic transformation method. Then bands were selected out of the original spectra and transformed logarithmic spectra with the confidence interval of mean for extraction of spectral characteristics of the degradation indicator species. And the selected bands were identified and validated with the Manhattan distance method. Results of the study show as follows:(1) Compared with the two dominant species, namely Carex pediformis and Leymus chinensis, the degradation indicator species Stellera chamaejasme featured at 402-412 nm, Artemisia frigid at 627-689, 715-929 and 929-1 033 nm and Potentilla acaulis at 705-721 nm; (2) In the above characteristic bands, the Manhattan distance of the vegetation homogeneous in plant species was obviously smaller than that of the vegetation heterogeneous in plant species. And the Manhattan distance of Stellera chamaejasme, Artemisia frigida and Potentilla acaulis was 0.006 6, 0.310 1 and 0.385 5, respectively; (3) No big difference was found, in the visible band between degraded vegetation and vegetation of dominant species. After logarithmic transformation, the difference was amplified and made easy the extraction of characteristic bands. The use of the original spectral curve in combination with its logarithmic spectral curve based on the confidence interval of mean made the extraction of characteristic bands more accurate. The eventually defined characteristics band for Stellera chamaejasme is 402-412 nm, for Artemisia frigid 627-689, 758-924 and 940-1 033 nm and for Potentilla acaulis 705-721 nm.

Key words: hyperspectral remote sensing, characteristic band, degradation indicator species, the confidence interval, Bashang Region of Hebei Province