Estimation Of Key Vegetation Parameters Base On Hyperspectral Remote Sensing Data

Vegetation can convert solar energy into chemical energy, carbon dioxide in the atmosphere into organic matter, providing mankind’s most basic material and energy sources. All kinds of ecological processes of vegetation, such as evaporation, transpiration, primary production, waste decomposition, and biochemical parameters are closely related of vegetation body materials such as chlorophyll and physical parameters such as leaf area index (LAI). How to calculate the ability of the forest carbon sequestration, estimate the forest productivity and biomass, and evaluate the ecological benefits of forest value became a hot issue for national scholars. Remote sensing technology for its large area, fast, dynamic advantage can be obtained forest canopy information in different time and spatial scales without destroying the forest, hyperspectral remote sensing as another milestone in the development of remote sense, can provide more abundant spectrum information, and have greater advantages in recognition of plant, inversion plants in physical and chemical information. In this paper, typical trees in central China and typical vegetation types of American were chosen. The inversion of three key vegetation parameters:vegetation chlorophyll,LAI and gross primary productivity(GPP) using hyperspectral was analysed. The main research work including:(1) The influence of chlorophyll content and LAI on spectral were analysed from the leaf and canopy scale respectively. The results showed that the most important bands were400-500nm,525-500nm,525-675nm,720-730nm and800nm when estimating the chlorophyll content and LAI. Using wheat as the research object, the chlorophyll content of every piece of leaf and canopy spectra were gathered during the whole growth. The leaf of wheat was stratified. The relationship between the sum of chlorophyll content of each leaf layer and typical band reflectance and vegetation index was analysed. The most penetrating band was red-edge band. Green band sensed slightly deeper than red band and near infrared band sensed shallowest.(2) The results showed that the three vegetation indices based on red edge band (CIrededge, NDRE, MTCI) showed higher precision at both leaf and canopy levels. At leaf scale, physical model showed similar precision than vegetation index model. But at canopy level, the precision of the physical model decreased because of the lack of the measured parameters.(3)The chlorophyll content was estimated using continuous wavelet transform method in two levels based on in-situ measurement and simulated data. The most suitable wavelet scale and band were found and the most accurate wavelet transform model was built up for the purpose of comparing with vegetation index models. Results showed that high correlation of wavelet coefficient areas are mainly distributed in the area near720nm and780nm at both leaf and canopy level, and the wavelet coefficient model were more accurate than vegetation index model. Through cross validation, we found that the models built up using simulated data can be applied to the measured data, which shows that the wavelet coefficient model has certain adaptability between different data sets.(4)Based on MODIS surface reflectance products, combined with the measured LAI, the vegetation index models and BP neural network models were built up for the estimation of LAI. Results showed that CI and NDVI model was more accurate than SR and EVI model. Because of joining information of the rest band, the precision of BP neural network model is higher than the vegetation index models.(5)Based on the MODIS surface reflectance products and flux meaurements of the study sites in Amecrican, the result of three models commonly used in GPP inversion were compared. The reason of the decreasing of accuracy in drought sites was analysed. An improved PAR model is established and verified, and the results showed that this model could improve the precision of the estimation of GPP in drought sites while the precision of the estimation of GPP in the other sites remained the same.