Hyperspectral Characters And Species Identification Model For Wetland Plants

Remote sensing has been widely used for wetland plants protection and monitoring. Even though the traditional multi-spectra data could identify ground objects such as bare-ground, water and plants with high precision, it cannot meet the requirement to identify wetland plant species or community because of the poor hyperspectral resolution. Therefore, the hyperspectral data, with the merits of high spectral resolution, hyper-bands numbers and huge numbers of data, are possible to be used to identify wetland plants and its communities.As a new type of data applied for wetland research, the hyperspectral data of the popular wetland plants around Beijing has been analyzed from different aspects. First of all, making use of the continuity attribute of hyperspectral data, Derivative Reflectance (DR) and Continuum Removal (CR) methods were used to highlight the changing speed, curvature and absorption attribute of the original reflectance spectra curves. Secondly, Mahalanobis Distance (MD) and Principal Component Analysis (PCA) methods were applied to reduce the dimensions and redundancy of hyperspectral data. And finally, nine scenarios were designed based on the parameters extracted from hyperspectral data, and Discriminant Method was used for building identification models. The precisions of models were compared and discussed.The results showed that the spectra of wetland plants were greatly influenced by the water absorption from background. Except the typical"two-convex-three-concave"character of plants'spectral curves, more attributes could be highlighted by using DR and CR methods. For example, the highest speed of the reflectance increasing is located in 520nm and 710nm, and the typical smaller concaves are in 510nm, 580nm, 690nm and 970nm. Besides, the second absorption band (550-730nm) has the strongest absorption strength. Water absorption influenced the hyperspectral curves of wetland plants a lot, especially in Near Infrared (NIR: 800-1000nm) where the reflectance is significantly different between emerged species and submerged species. Both MD and PCA methods can efficiently reduce the dimensions of hyperspectral data, and the parameters extracted from the results of MD and PCA methods could separately constructing a good identification model for wetland plants with high precision (more than 90%). However, the parameters extracted from CR curves which standing for the absorption attribute of spectra, built better models than only making use of the original reflectance curves, which indicated that the water and chlorophyll absorption influenced the hyperspectral curves of wetland plants.In general, the analysis of the hyperspectral data of wetland plants helped us to better understanding the spectra of different wetland plants. And, the constructed identification models for wetland plants could also be used in the future, for supervising the wetland mapping and monitoring using space based remote sensing data.