摘要
为了提高森林蓄积量估测精度,以福建省三明市将乐县国有林场中杉木林作为试验区,选择资源3号卫星多光谱高分辨率影像及Alos Palsar影像为数据源,将相关性较高的极化雷达参数与最优窗口下的纹理参数相结合,协同两种遥感数据反演蓄积量。利用灰度共生矩阵分别提取高分辨率影像在3×3、5×5、7×7、9×9和11×11的5组窗口大小下8种纹理特征信息,提取Alos Palsar影像双极化方式下后向散射系数并进行比值运算。采用多元逐步回归分析方法,分别利用5组纹理特征信息反演杉木林蓄积量,找出最优窗口;检测不同极化方式下后向散射系数与蓄积量之间相关性。结果表明,单数据源反演蓄积量模型中,5×5窗口反演效果最好,模型复相关系数R=0.869,均方根误差σRMSE=23.38 m^3/hm^2,蓄积量总体的估测精度为80.32%;多数据源反演蓄积量模型中,两种极化方式下的后向散射系数比值与高分影像纹理特征参数结合后,反演模型的效果更好,模型中R=0.901,σRMSE=22.32 m^3/hm^2,蓄积量总体估测精度达到85.42%。研究表明,基于多数据源数据的森林蓄积量反演精度更高,结果更准确。
In order to improve the precision of forest volume estimation, the Chinese fir （ Cunninghamia lanceolata） stands of state-owned forest farm in Jiangle County, Sanming City, Fujian Province were selected as the study object, and the high resolution images of ZY-3 and the images of Alos Palsar were selected as the remotely sensed data sources. The polarization radar parameters with high correlation and the texture parameters of the optimal window were combined for the volume inversion. Eight texture features of ZY-3 high resolution image were extracted by the gray level co-occurrence matrix under 5 kinds of window sizes including 3×3,5×5,7×7,9×9 and 11×11 pixels. Meanwhile, the backscatter coefficients in HH and HV polarization modes were derived from Alos Palsar images. Furthermore, ratio of the two baekseatter coefficients above was computed. The texture features from 5 different windows were used as independent variable in the inversion of forest volume respectively by using stepwise regression analysis to find the optimal window. Then, the correlation between backseatter coefficients from different polarization modes and the forest volume was computed. The results showed that for the inversion model based on ZY-3 images, the optimal window was the size of 5×5, the value of multiple correlation coefficient reached to 0.869, with a root mean square error 23.38 m^3/hm^2 and a total estimation accuracy of 80.32%. While for the inversion model integrating the ratio of baekseatter coefficients from Palsar with the texture features of optima/window from ZY-3, the value of multiple correlation coefficient reached to 0. 901, with the root mean square error 22.32 m^3/hm^2 and the estimation accuracy of total forest volume 85.42%.The resuhs suggests using hi-source remote sensing data can produce a higher precision of volume estimation on average.
出处
《南京林业大学学报（自然科学版）》
CAS
CSCD
北大核心
2016年第5期107-114,共8页
Journal of Nanjing Forestry University：Natural Sciences Edition
基金
国家高技术研究发展计划(2012AA102001)
