| With the generation and development of hyperspectral remote sensing, higher spectral resolution datum are provided for the application of remote sensing survey of forest resources, forest health research, forest biomass and other forest parameter estimation. Forest biochemical parameters, especially chlorophyll and nitrogen content in the forest directly reflect the health and stress of forests. It is a significance of studying global carbon cycle model and evaluating the role of forest on terrestrial ecosystems by using remote sensing technology to estimate forest biochemical parameters quantitatively.Leaf biochemical parameters, especially quantitative estimation methods and models of chlorophyll and nitrogen content retrieval were studied in this paper, combining with the field measurements of leaf chlorophyll and nitrogen content and the corresponding leaf spectral data. Multivariate statistical model, neural network model, support vector machines (SVM) and physical optics model, four methods were used for estimating chlorophyll and nitrogen content. The results showed that the improvement of error feedback neural network model Erf-BP is the best among the chlorophyll content estimation models on leaf-scale, of which contain 13 neurons in three layers has the highest accuracy, with the fitting accuracy of 95.28%, test accuracy of 94.46% and the RMSE of 3.321μg/cm2 respectively. The Diff(R535) variable is used for nitrogen content retrieval on leaf-scale. When Sigmoid=1, the support vector machine model of radial basis function kernel is better, with the correlation coefficient between the measured data and simulation results of 0.984.Chlorophyll and nitrogen contents estimation models are established basing on the leaf spectral and those models can not be directly used to inverse the forest canopy biochemical parameters from remote sensing imagery. Therefore, the geometric optical 4-scale model and looking up table (LUT) method were used to change canopy spectrum from Hyperion image to leaf spectrum, which make the parameters of chlorophyll and nitrogen content retrieval model and image the same scale. And the inversion model was last used for estimating leaf chlorophyll and nitrogen content. Leaf-level chlorophyll and nitrogen content were converted to canopy-level combining with leaf area index (LAI).Statistical method was used for leaf area index calculation. Model of leaf area index was established by hyperspectral remote sensing vegetation indices based on field measurements. R2, RMSE and the accuracy of LAI inversion model were 0.8156,0.5426 and 83.02% respectively.Canopy scale mapping of chlorophyll and nitrogen content in study area were achieved by leaf area index from Hyperion image, combining with leaf scale chlorophyll and nitrogen content mapping. The results show that the estimation of the chlorophyll and nitrogen content were consistent with the actual situation. |
