Integration Of Field Hyperspectral Data And Satellite Imagery In Monitoring The Invasion Of Spartina Alterniflora In Chongming Island

Along with the development of globalization, the invasive plant has dramatically reduced the biodiversity of estuarine wetland ecosystems. It is thus important to monitor the invasive plant in order to maintain the biodiversity and nature resources of estuarine wetlands. While it is not an easy job to study the distribution of invasive plant on landscape scale through the traditional investigation methods, especially for monitoring the vastly spread invasive plants, fortunately, remote sensing has the potential to update surveys by repeatedly covering large areas, especially for areas with difficulty to access. Therefore, great progress of remote sensing applications in wetland has been made in the past decades. Meanwhile, there rises a problem: how to choose the appropriate remote sensing data and the algorithms since choosing appropriate imageries and algorithum is very important for successful invasive plant detection. In other words, how to incorporate different remotely sensed data and make the best use of their advantages to solve the problem of identify plant in species level.This study is an attempt to incorporate field hyperspectral data, high spatio-resolution data and Landsat TM data with abundant historical accumulation for monitoring invasive plant in estuarine wetland. Specifically, in order to have a more clearly understanding of the vegetation spectra in different season, we employed spectrometer to measure the vegetation reflectance spectra as a support for choosing bands and algorithmum of satellite imgaries on large scale. Meanwhile, SMA (Spectral Mixture Analysis) was applied to Landsat TM images and to study the potential of TM imaging for determining the distributions of invasive plants at the species level. Moreover, our work provides a cross-validation technique for comparing the unmixing results derived from TM data with high spatial resolution imagery.The results show that:1) The three dominate species in the study area have different reflectance spectra on different phenological stages in one year. During December to March, the reflectance spectral curve of vegetaion is similar to that of soil, and there is no obviously difference between the spectra of Spartina and Phragmites. While in late April and May, the reflectance spectra of Phragmites and Spartina are distinct, and the statistics prove that there are significant differences between the two at many wavelength positions. The spectra in July, August and Spetember are less distinct between Phragmites and Spartina, indicating that summer is an improper time to identify Spartina through satellite imageries. Meanwhile, the soil background has influence on the spectra reflectance of vegetation, especially for short plants. Applying continuum remove to the spectra can enhance the separability of different spectral curve, but the statistical results are different from date without continuum remover.2) The proper endmember combination could generate satisfactory SMA results, and had a relatively good capacity to predict vegetation distribution. A positive correlation between the Phragmites fraction predicted with SMA models and the NDVI value also validated the capacity of SMA to investigate the spatial distribution of vegetation at the species level. However, the assessment of accuracy showed that certain errors occurred in the SMA fraction images. Most of these inaccurately predicted how the pixels were distributed in areas filled with water or near water, indicating that tidal water has a great effect on SMA modeling.