| Forest Biomass is a key feature for forest ecology system. As remote sensing develops, multi-source remote sensing data has been applied to estimate biomass on various scales. In this study, high resolution optical remote sensing and microwave remote sensing data are used. High resolution optical data has less spectral information but it has more spatial information. So it can reflect objects’ structure characteristics and discipline which contributes to analyze forest structural parameters. SAR can observe objects regardless of time, weather and external influence. The SAR with C band can response tree crown, which is of great help for forest estimation.This study is conducted in Tahe county, Daxinanling area in Heilongjiang province with Wordview-2 and Radarsat-2 data. Firstly, spectral grayscale values, vegetation indice and textual eigenvalues were extracted from Worldview-2. Second, backscatter signal and textual eigenvalue were extracted from SAR. Third, the texture variables, band spectral features, vegetation indice from optical image were combined with the backscatter and texture eigenvalue of different polarized SAR to build biomass model. It is possible to explore potentiality of multi remote sensing data in forest biomass estimation and provide data support for local carbon management through this study.The main contents and results include:1. Extract Worldview-2 image’s original 4 band spectral grayscale values, PCA band grayscale value,8 texture eigenvalues of different window sizes,6 vegetation indice.2. Extract Radarsat-2 backscatter of polarization HH, HV, VH, VV, HV, HV/HH, VH/HH, HV/VV, VH/VV. Extract 8 texture eigenvalues of 3×3 window size.3. The textual eigenvalues of Worldview-2 high resolution image are of significant correlation to above ground biomass. In broadleaf forest, the biomass is more related with correlation and variance. In coniferous forest, the biomass is more related with second moment, homogeneity and dissimilarity. The different textual signals between broadleaf and coniferous forest maybe due to the different structure of trees.4. The backscatter of the polarization SAR is also of significant correlation. In broadleaf forest, the biomass is more related with HV/VV and VH/VV. In Coniferous forest, the biomass is more related with HH contrast and HH dissimilarity. The cause maybe the coniferous forest has less leaf area so need more fine textual eigenvalues.5. Multivariate nonlinear and SVR models were built. In both models, broadleaf biomass’s accuracy (multivariate nonlinear 0.5757, SVR 0.7904) is higher than coniferous (multivariate nonlinear 0.4198, SVR 0.7684). And for both broadleaf and coniferous forest, SVR’s accuracy is higher than multivariate nonlinear model. The reason maybe C band SAR is more capable to reflect tree’s crown with more leaves.6. Terrain factors are analyzed with biomass values that SVR model predicted. The terrain factors are elevation, slope and aspect. The results show broadleaf forest biomass distribution is gathering in direction of north, northeast and south, in elevation of 500-700m and in slope of 5°-25°. For coniferous forest biomass, aspect distribution and elevation distribution are same with broadleaf forest’s, and slope distribution is between 15° and 25°, which is a little higher than broadleaf’s. The reasons of distribution are (a) the place in north, northeast direction has better water storage capacity and the place in south direction has more sunshine, (b) the place in 500-700m elevation has less human impact and more fitted temperature, (c) coniferous forest in higher slope is more adaptable to poor site conditions than broadleaf forest. |
