Study On Forest Type Identification Based On Spaceborne Large-footprint Lidar Data

Human activities make the carbon dioxide (CO2) and other gases in atmosphere continue to increase, so that the environment around us has taken place a series of changes. Forests are the largest carbon reservoir in terrestrial ecosystems.and forests can decrease the carbon dioxide in atmosphere through photosynthesis. So accurate estimate of forest carbon storage has an important effect on the environment. Forest carbon storage is converted by forest biomass. It is about47%of forest biomass. Forest biomass is closely related to the height and the type of forest. Traditionally, the forest canopy height is mainly obtained through field investigation. This method is not only costly and time consuming, but also difficult to obtain large area data in a short time. A lot of research has been done on forest applications in abroad and made great progress, especially in tree height detection. Many researches show that Lidar system as an active remote sensing technology has a great advantage and potential in inversion forest canopy height. But there is only a few paper used lidar data to class the forest types. In order to obtain forest biomass reasonable we should also do research on forest classification.In this paper, we used ICESat-GLAS (Ice, Cloud and Land Elevation Satellite-Geoscience Laser Altimeter System) data to class forest types. And it will be useful to study forest biomass and the change of carbon dioxide in the atmosphere. So in this paper we made these studies.(1) First download the ICESat-GLAS data, converting the data from binary to ascii, then connecting the GLAO1data with GLA14based on the same part. Using Gaussian Filter to reduce the noise of the ICESat-GLAS return waveforms and making Gaussian decomposition with waveform. Finally we make a complete preprocessing method of ICESat-GLAS data.(2) After Gaussian decomposition we class the waveform curve into vegetation echo and surface echo. And we extract waveform information of each Gaussian curve. In each Gaussian curve we extract the effective echo time; the energy value corresponding to the peak; the time corresponding to the peak.(3) Based on the waveform information extracted from the Gaussian curve, we made classification parameters conversion, after conversion, we got the classification parameters. Parameters for classification of forest types are as followed. The slope of the first Gaussian curve after the Gaussian decomposition; the mean slope of Gaussian curve corresponding to the vegetation canopy; the improvement mean slope of Gaussian curve corresponding to the vegetation canopy; the standard deviation of Gaussian curve corresponding to the vegetation canopy; the improvement standard deviation of Gaussian curve corresponding to the vegetation canopy. (4) In this paper, K Nearest Neighbors method was used to class two kinds of forest types using different parameters groups which were extract from broad-leaved forest and coniferous forest. In the study I found that when using parameters group two to class forest type, the classification accuracy of broad-leaved forest is91.67%, the classification accuracy of coniferous forest is33.33%, the overall classification accuracy of test date is77.78%; the Kappa coefficient is0.2899. When using parameters group five to class forest type, the classification accuracy of broad-leaved forest is95.85%, the classification accuracy of coniferous forest is40%, the overall classification accuracy of test date is82.54%; the Kappa coefficient is0.4268. When using five groups of parameters to class two kinds of forest types, the worst classification result is parameters group two, the best classification result is parameters group five.When C-SVC(Support Vector Machine) method was used to class two kinds of forest types using different parameters groups which was extract from broad-leaved forest and coniferous forest. I found that when using parameters group two to class forest type, the classification accuracy of broad-leaved forest is95.83%, the classification accuracy of coniferous forest is33.33%, the overall classification accuracy of test date is80.95%; the Kappa coefficient is0.3571. When using parameters group five to class forest type, the classification accuracy of broad-leaved forest is97.92%, the classification accuracy of coniferous forest is40%, the overall classification accuracy of test date is84.13%; the Kappa coefficient is0.4643. When using five groups of parameters to class two kinds of forest types, the worst classification result is parameters group two, the best classification result is parameters group five.The research results provide a new method of ICESat-GLAS waveforms data processing and also provide scientific references to class the forest type. The research is helpful to use ICESat-GLAS data for classification forest type and precise estimation of forest biomass and carbon storage.