主被动遥感特征优化的天然林森林蓄积量估测——以新疆巩留县为例

Estimation of Forest Stem Volume of Natural Forests based on the Optimization of Active and Passive Remote Sensing Features——A Case Study over Gongliu County of Xinjiang

结合国产主被动遥感数据高分六号(GF-6)PMS和高分三号(GF-3)双极化PolSAR估测森林蓄积量,并针对多源遥感数据的冗余问题进行特征组合优化。以新疆巩留县天然林地为研究区,提取GF-6 PMS数据的光谱信息、植被指数、纹理以及植被覆盖度信息和GF-3 PolSAR数据的后向散射系数、极化分解参数,结合地形因子,在森林样地调查数据的基础上,利用快速迭代特征选择的K最近邻法(K-Nearest Neighbor with Fast Iterative Features Selection,KNN-FIFS)估测研究区的森林蓄积量。对比国产主被动遥感数据和单一遥感数据源时的估测结果,基于最优特征组合反演研究区的森林蓄积量,结果表明:联合GF-3 PolSAR和GF-6 PMS数据估测研究区森林蓄积量的精度为R2=0.72,RMSE=92.48 m3/hm2,相比于仅使用GF-6 PMS数据估测的精度(R2=0.56,RMSE=118.8 m3/hm2),R2提高了0.16,提高了28.6%,RMSE降低了26.32 m3/hm2,降低22.2%。说明主被动遥感数据协同反演可以提高森林蓄积量估测精度,KNN-FIFS方法可以有效地估测天然林森林蓄积量。

We estimated forest stem volume using domestic active and passive remote sensing data GF-3 PolSAR and GF-6 PMS. And in order to find a way out of the redundancy problem of multi-source remote sensing data,feature combination is optimized. The research area is the natural forest land in Gongliu County,Xinjiang.We extracted spectral information,vegetation index,texture,vegetation coverage from GF-6 PMS data and then extracted backscattering coefficient and polarization decomposition parameters from GF-3 PolSAR data.Combining the extracted parameters,terrain factor and forest sample survey data,we estimated forest stem volume in the study area using K-Nearest Neighbor with Fast Iterative Features Selection(KNN-FIFS)method.Comparing and validating the estimation results when combined active and passive remote sensing data and a single remote sensing data source,we inverted the forest stem volume in the study area based on the optimal feature combination. The results show that the accuracy of combining GF-3 PolSAR data and GF-6 PMS data to estimate the forest stem volume in the study area is R2=0.72 and RMSE=92.48 m3/hm2,which is compared with the accuracy estimated using only GF-6 PMS data(R2=0.56,RMSE=118.8 m3/hm2),R2 increased by0.16 with an increase rate of 28.6% and RMSE decreased by 26.32 m3/hm2 with a decrease rate of 22.2%. It indicated that the cooperative inversion of active and passive remote sensing data can improve the estimation accuracy of forest stem volume,and the KNN-FIFS method can effectively estimate the forest stem volume of natural forests.