联合GF-3 PolSAR数据与Landsat-8 OLI数据的森林地上生物量估测

Estimation of forest above-ground biomass based on GF-3 PolSAR data and Landsat-8 OLI data

【目的】森林是陆地生态系统的重要组成部分,精确估测森林地上生物量对森林资源的经营管理具有指示作用,对研究全球碳循环具有重要意义。为了改善单一来源遥感数据估测森林地上生物量的不足,探讨了联合高分三号(Gaofen-3,GF-3)全极化(Polarimetric synthetic aperture radar,PolSAR)数据极化分解参数和Landsat-8 OLI数据估测森林地上生物量的可行性,并针对多源遥感数据的冗余问题优化特征组合。【方法】以广西南宁市高峰林场为研究区,结合森林样地调查数据,提取GF-3 PolSAR数据的后向散射系数、极化分解参数和Landsat-8 OLI数据的光谱信息、植被指数、纹理,使用基于序列前向特征选择的K最近邻法(K-nearest neighbor based on sequence forward feature selection,KNN-SFS)估测研究区的森林地上生物量,以留一法交叉验证得到的森林地上生物量预测值和实测值之间的均方根误差(Root mean square error,RMSE)最小为原则,对比验证使用多源遥感数据和单一来源遥感数据时的估测结果,寻求估测森林地上生物量的最优特征组合,基于最优特征组合绘制研究区的森林地上生物量空间分布图。【结果】结合GF-3 PolSAR数据和Landsat-8 OLI数据估测研究区森林地上生物量的精度为RMSE=21.05 t·hm-2,R2=0.75,优于仅使用GF-3 PolSAR数据估测的精度(RMSE=28.38 t·hm-2,R2=0.47)和仅使用Landsat-8 OLI数据估测的精度(RMSE=29.52 t·hm-2,R2=0.42)。【结论】多源数据协同反演森林地上生物量可以提高估测的精度,基于KNN-SFS方法联合GF-3 PolSAR数据与Landsat-8 OLI数据可以较好地估测森林地上生物量。

【Objective】Forest is an important part of terrestrial ecosystems. Accurate estimation of forest aboveground biomass is of great significance to the study of global carbon cycle.In order to overcome the defects of estimating forest aboveground biomass based on single source remote sensing data, we explored the feasibility of estimating forest aboveground biomass using the polarization decomposition parameters of GF-3 PolSAR data and Landsat-8 OLI data. And in order to solve the redundancy problem of multi-source remote sensing data, feature combination is optimized.【Method】The research area is Gaofeng planted forest in Nanning city, Guangxi. We extracted backscattering coefficient and polarization decomposition parameters from GF-3 PolSAR data and then extracted spectral information, vegetation index, texture from Landsat-8 OLI data. Combining theextracted parameters and forest sample survey data, we estimated forest aboveground biomass in the study area using k-nearest neighbor method based on sequence forward feature selection method(KNN-SFS). Based on the principle of minimizing the RMSE calculated by the predicted and measured values of Forest Aboveground biomass, we compared and validated the inversion results of multi-source remote sensing data and single remote sensing data andfound the optimal feature combination for estimating forest aboveground biomass. Based on the optimal feature combination, the spatial distribution map of forest aboveground biomass in the study area was drawn.【Result】The accuracy of estimating forest aboveground biomass by combining GF-3 PolSAR data with Landsat-8 OLI data is RMSE=21.05 t·hm-2, R2=0.75, which is better than that by using GF-3 PolSAR data(RMSE=28.38 t·hm-2, R2=0.47) and by using Landsat 8 OLI data(RMSE=29.52 t·hm-2, R2=0.42). 【Conclusion】The accuracy of estimating forest aboveground biomass can be improved by using multi-source data. The model based on KNN-SFS method, which combines GF-3 PolSAR data with Landsat-8 OLI data, can better estimate forest aboveground biomass.