The Research Of Regional Forest Above Ground Biomass Inversion Combining ICESat-GLAS Waveform And HJ-1A Hyperspectral Imageries

The traditional way of forest metrics estimation is depends on the field investigation which requires large amount of manpower, material resources and time. Besides, this mathod can only access the point-to-point data, and it is unable to achieve the spatial distribution of forest metrics of large area in short period of time. The emergence and development of remote sensing technology makes up for the shortcomings of traditional estimation method, and it can quickly achieve forest parameters of large area without any damage, improving the efficiency of forest parameters estimation and satisfying the requirement of forest inventory.In the research, taking Wangqing forest area of Jilin province as the study area, some metrics(W, TS, I, ec) were extracted from GLAS(Geoscience Laser Altimeter System) waveform, among of this metrics, W represents GLAS waveform length, TS is a metric related to terrain slope,I is the ratio of forest canopy energy and total energy and ec is the differential between forest canopy energy and total energy. All of these metrics were used to estimate forest parameters such as maximum forest height, forest canopy density and forest aboveground biomass(AGB) for three forest types(broad-leaved forest, coniferous forest and mixed forest). Considering that the spatial distribution of GLAS footprints are discrete, which has no ability of imaging and cannot be implemented to regional estimation, the paper attampts to combine GLAS waveforms with HJ-1A/HSI hyperspectral images to finish regional estimation and imaging by building the ralationship between spectrum information and GLAS estimated forest parameters based on the Support Vector Regression(SVR) machine method. The conclusions are as follows:(1) Three best bands(band2, band52 and band107) of HSI images were selected by the segmented principal component analysis method and the band index method. Due to the low spatial resolution of HSI images, the classfication result is not ideal. When the HSI images were fused with CCD images by wavelet transform method, the spatial resolution was improved and the classification accuracy increased to 85.3% using the band information and texture characteristic by the support vector classfication (SVC) method.(2) There is a strong linear relationship between TS and terrain slope. The maximum forest height model with W and TS as variables is better than that built with W. It is also superior to the model built with W and terrain index. The map of maximum forest height was producted by the SVR model between GLAS estimated maximum forest height and the three MNF components of HSI images using the support vector regression (SVR) method. The maximum and minimum error of the maximum forest height map was 7.1m and 0.07m, respectively. And 25% of estimation error were lower than 0.75m and 50% of estimation error were between 0.75m and 2.31m.(3) For GLAS waveform, the accuracy of forest canopy density multi-variable model with I75 and ec as variables was superior to the single variable model with I75 and ec as variable, respectively. For HSI images, the performance of the multi-variable model was better than that of single-variable model. Besides, the HSI multi-variable model was superior to GLAS model. So in the research the HSI multi-variable model was used to map forest canopy density of the whole study area. There are about 25% estimation error of the forest canopy density map were lower than 0.031, and 50% estimation error rangs from 0.031 to 0.126.(4) The forest AGB model built with GLAS estimated maximum forest height alone was not ideal. However, the estimation accuracy of the forest AGB was improved when GLAS estimated maximum forest height and HSI estimated forest canopy density were combined to produce the model using linear regression method and SVR method, and the performance of the SVR model was more better than linear regression model.