Study On Estimation Of Forest Biomass Based On Multi-source Remote Sensing Data

Forest biomass is the base of materials’ circulation in forest ecosystems. It is a key criterion for estimating the productivity of ecosystem. The traditional method of statistic has many weak points, e.g. heavy work, long stage, and difficult to get data of forest in large area. With the rapid improvement of remote sensing, many data of remote sensing have been used for the information detection and extraction of structural characterization of forest. It provides method to biomass estimation and the research on its change. Recently, owing to the diversity of techniques of remote sensing, researchers are raising high expectation to the work on forest biomass in this method.In this thesis, the protected area of Can Yuan County, Yunnan is the research area. Firstly, it discussed the application in forest and current status of remote sensing. Then the data of echo waveform of LiDAR and multi-spectral are preprocessed based on the content and aim of this work. Depending on the data of LiDAR, it estimated calculation of canopy height and the parameters’of multi-spectra, and the factors extracting of horizontal structure of forest. Furthermore, it verified the relationship between the parameters and forest biomass. Lastly, the modules of the multiple linear regression and BP neural network were constructed. It discussed the best one on the research of this thesis.In the present work, we focus on the study in the six points as follows.(1) After the pretreatment of radiation correction, atmospheric correction and orthorectification, author obtains ten kinds of vegetation indexes through the six bands of the Landsat TM data which is acquired in the study area including RVI> NDVI、SLAVI、 EVI、VII、MSR、NDVIc、BI、GVI and WI.(2) Analyzing the distribution situation of the forest, author classifies the forest cover information of the study region. Classifying LiDAR data by using the region growing and clustering algorithm and adding the points of the vegetation into the TM multi-spectrum data to obtain the accurate information for the foundation of the construction of estimating the canopy height and biomass.(3) Using the reflectivity of six bands and ten kinds of vegetation indexes which are obtained from the corrected TM multi-spectrum data, analyzing the relativity of the remote sensing inversion between LAI and every single index, using the model of partial least-squares regression to construct the estimation model which is appropriate for the LAI of the regional scale. Combining the comparison between the sample data and estimation data, author obtains that the correlation coefficient of R2 is 0.91.(4) According to vegetation index of TM data in the study region, using of the pixel unmixing to construct the model between vegetation index and canopy density. Combining the sample data and the estimation data to construct Linear Regression. Author obtains the correlation coefficient of R2 is 0.86.(5) Using of the corrected echo waveform data of LiDAR and Lefskyetal method to construct the estimation model of forest canopy height. Author analyzes the relativity between every independent variable in the equation and average canopy height and explaining ability of the variables. Then author analyzes the prediction precision. As a whole, the estimation precision of forest canopy height relatively high. R2is 0.89. RMSE is around 0.5m. Overall the consistency is obvious overall.(6) Based on the forest canopy height,10 kinds of vegetation indexes, LAI and crown density which are obtained from LiDAR and TM data, using of the multiple linear regression and BP neural network to construct the estimation model of forest biomass. Author finds that spectral information and forest height information of the BP neural network model have the ability of the non-linear processing and estimate the forest biomass better. According to the sample data and estimation data to verify it. Research shows that the R2 of MLR model is 0.78 and the R2 of the BP neural network model is 0.88.So, the BP neural network model is the better model for the estimation of the forest biomass in this study region.Forest biomass is the significant research area of productivity of forest and the distribution of nutrition. The estimation of forest biomass in designated area through multi-remote sensing is affected by the site condition and the Function Group of Plant, the accuracy of the data and so on. It is the next point of research on accurate estimation to the biomass in designated area according to the characteristic of forest.