The Study Of Estimation Models Of Chinese Fir Chlorophyll Content Based On Red Edge Parameters

With the development of hyperspectral remote sensing technology, the diversity of data processing algorithms and the urgent need of grasping forest ecosystem, using hyperspectral remote sensing data to estimate the content of biochemical components in the forest is becoming valuable. Chlorophyll is a main physiological characteristic of the vegetation. The accurate estimation of the content of chlorophyll will indirectly control the health of the forest. The estimation models for estimating vegetation chlorophyll content based on hyperspectral feature are established by using hyperspectral remote sensing technology. They can improve the traditional determination methods of forest canopy chlorophyll content. They can give service to the large-scale, real-time, accurate, non-destructive determina-tion methods of monitoring changes of forest canopy chlorophyll content and the building and researching of the forest ecosystem.In this study, young, middle and near mature Chinese fir in Huang Fengqiao state-owned forest farm of Hunan Province were selected as interesting object. The spectral reflectance of canopy was measured by ASD FieldSpec and the chlorophyll content was measured by spectrophotometry at the same time. The relativity between eleven red edge parameters extracted by spectral derivative technique and leaf chlorophyll content were analyzed. The estimation models based on unary regression method were established by using the red edge parameters that were highly significantly correlated with chlorophyll content. The estimation models based on stepwise regression analysis were established by using all the red edge parameters. The dimensionality of the eleven red edge parameters was reduced by means of principal component analysis. Single hidden layer BP neural network was established for estimating vegetation chlorophyll content. Weight value and threshold value were initialized by using genetic algorithm and principal component scores were the input eigenvectors. The fitness between the predicted value and the measured value was tested by the determination coefficient, the lowest root mean-square error and the average relative error. The results are as follows. The first, the red edge position, the red edge amplitude, the red edge area, the anisotropy index of red edge amplitude and the red edge width are highly significantly correlated with fir canopy chlorophyll content. The second, the power function based on the red edge position has a highest prediction accuracy of 99.765%. The model is y= 3.23×10-52x18.14703. The third, multiple linear regression model based on the red edge position, the red edge width and the red edge amplitude has a prediction accuracy of 91.9%. The model is y=-20.984+0.031x1+0.009x2+29.619x3. (Note:x1, x2 and x3 - the red edge position, the red edge width and the red edge amplitude.) The fourth, when the number of hidden layer neurons is ten, single hidden layer BP neural network has a prediction accuracy of 97.372%, the determination coefficient is 0.976. This result is better than other cases. The fifth, using fir red edge parameters to fast estimate chlorophyll content is feasible and all the models have a prediction accuracy of above 91.0%. The power function based on the red edge position has the highest prediction accuracy, so it is selected as the fittest model.This paper was supported by National Natural Science Foundation of China (NO.30871962) and Research Fund for the Doctoral Program of Higher Education of China (NO.200805380001).