Using Nonlinear Theory To Study Forest Stock Qantitative Estimation Based On "3S

In this paper, with the modern spatial science and computer science as the basis, according to the RS and GIS information of a few ground sample plots, adopting nonlinear theory, the forest canopy density and stock quantitative estimation equation and neural network models based on the unit of pixel are established. By means of the approach of simulation, the stock of all pixels of non-sample plots can be estimated. By way of pixel integration, the total forest stock estimation can be fulfilled. The main research substances, methods and conclusions are as follows.The chief research substances:(1) With the RS and GIS information as the basis, determining the arguments setting on forest stock quantitative estimation, screening the principal arguments influencing stock estimation.(2) The defects analysis of traditional least square principle and linear model used in forest stock quantitative estimation.(3) The unfavorable influence of abnormal value and Multi-collinearity to stock estimation, and the overcoming methods.(4) The influence of remote sensing region size to spatial calibration and stock estimation, the function of GPS positioning technology in stock estimation.(5) The extracting method of RS and GIS information of ground sample plots.(G) The determination of the minimum ground inventory work load needed for the forest stock quantitative estimation based on "3S".(6) The transformation models and accuracy between the coordinates of GPS and GDZ80 and BJZ54.(7) The application of neural network in stock quantitative estimation.(8) The forest stock estimation on segments. The main research methods:(1) Screening the principal arguments influencing stock estimation by utilizing ridge trace analysis and the principle of residual mean squares.(2) Forest canopy density and stock estimation by using ridge estimation and robust estimation.(3) Forest stock quantitative estimation by utilizing mended BP neural network.(4) Forest stock quantitative estimation on segments by means of radial basis function network.The main conclusions:(1) With the RS and GIS information of a few ground sample plots as the basis, byestablishing forest canopy density and stock estimation equation, the total stock and canopy density can be forecasted effectively.(2) Among the RS and GIS information influencing canopy density and stock estimation, the RS information is essential, the GIS information such as stand type, aspect, elevation etc. plays an important role. Only integrating the information of RS and GIS, the forest total stock and canopy density can be predicted effectively.(3) The problem of screening the principal arguments influencing canopy density and stock estimation is solved validly by adopting ridge analysis and principle of residual mean squares. The chief variables can be obtained from quite a few RS and GIS arguments that may have effect to the estimation of canopy density and stock. The stability, reliability and forecasting accuracy of the estimation equation are improved greatly when the arguments that have no or few effect to estimation are rejected.(4) The real examples demonstrate that forest canopy density and stock can be estimated effectively by only sampling 30%~40% of traditional sample plots.(5) Regardless of adopting ridge trace analysis or the principle of residual mean squares, diere may exist a certain degree multi-collinearity among screened arguments. When there are some multi-collinearities among the chief arguments influencing stock estimation, the equation of forest canopy density and stock had better be established through ridge estimation take the place of LS estimation.(6) The ridge estimation is a bias estimation, it may be used only when there are some serious multi-collinearities among main arguments influencing canopy density or stock estimation, or the estimation accuracy will reduce. Using ridge estimation, ridge trace method should be a main method for the selection...