Growing stock volume estimation from L-band ALOS PALSAR polarimetric coherence in Siberian forest
Growing stock volume estimation from L-band ALOS PALSAR polarimetric coherence in Siberian forest
复制标题
DOI：
10.1016/j.rse.2014.05.007
复制
发表时间：
2014-12
影响因子：
13.5
通讯作者：
T. Chowdhury;C. Thiel;C. Schmullius
中科院分区：
文献类型：
--
作者：
T. Chowdhury;C. Thiel;C. Schmullius
本文研究了 L 波段 ALOS PALSAR 全极化雷达数据在林业应用中的潜力。为了将偏振响应描述为西伯利亚森林蓄积量 (GSV)、天气条件和不同树种的函数，分析了 HH 和 VV 偏振之间的偏振相位差以及偏振 HHVV 相干性的大小。此外，还评估了森林蓄积量（GSV）的反演精度。调查针对林分水平进行，最大 GSV 为 350 立方米/公顷。使用多时态 L 波段 ALOS PALSAR 极化相干性数据和林分水平的大量森林清查数据，我们观察到极化相干性随着 GSV 的增加而明显下降，皮尔逊相关系数在 − 0.74 和 − 0.79 之间。对于相对蓄积量> 70%（即更均匀的林分结构）的林分，相关性提高到-0.87，并且极化相干性对GSV的敏感性增加到~250 m3/ha GSV。偏振相位差（PPD）也可以将森林与非森林地区区分开来。图像采集时解冻和未结冰天气条件下 PPD 的比较表明，在后一种条件下体积散射的贡献更大。此外，还观察到偏振相干性和相位差的差异作为树种的函数。特别是，与白杨、桦树和松树相比，落叶松林在未结冰的条件下通常表现出 + 0.1 的偏振相干性。与其他树种相比，无论是在解冻还是未冻结的条件下，PPD 的变异性也较小。使用指数模型来表征作为 GSV 函数的偏振相干行为。在未冻结条件下，GSV 的恢复效果更好，RMSE 为 33 m3/ha，R2 为 0.73。解冻条件下，反演误差稍高（RMSE= 51 m3/ha，R2= 0.61）。在林分面积大于 30 ha、相对蓄积量至少为 70% 的情况下，GSV 反演误差分别降至 RMSE=16 m3/ha 和 RMSE=19 m3/ha。
This paper investigated the potential of L-band ALOS PALSAR full polarimetric radar data for forestry applications. In order to characterize the polarization response as a function of growing stock volume (GSV), weather conditions and different tree species in Siberian forests, the polarization phase difference between HH and VV polarizations and the magnitude of the polarimetric HHVV-coherence have been analyzed. Moreover, the retrieval accuracy of forest growing stock volume (GSV) has been evaluated. The investigations were conducted on forest stand level and maximum GSV was 350 m3/ha. Using multi-temporal L-band ALOS PALSAR polarimetric coherence data and extensive sets of the forest inventory data at a stand level, we observed a clear decrease of polarimetric coherence with an increase in GSV, with Pearson's correlation coefficient between − 0.74 and − 0.79. The correlation improved to − 0.87 for the forest stands with high relative stocking > 70% (i.e. more homogeneous forest stand structure) and the sensitivity of the polarimetric coherence to GSV increased to ~ 250 m3/ha GSV. The polarization phase difference (PPD) also allowed forest to be discriminated from non-forest areas. Comparison of PPD during thawing and unfrozen weather conditions at the time of image acquisition revealed a higher contribution of a volume scattering at the latter conditions. In addition, differences in polarimetric coherence and phase difference were observed as a function of tree species. In particular, stands of larch typically exhibited a + 0.1 polarimetric coherence under unfrozen conditions compared to aspen, birch and pine. There was also less variability in the PPD compared to other tree species, both under thawing and unfrozen conditions. An exponential model was used to characterize the polarimetric coherence behavior as a function of GSV. The retrieval of GSV was better under unfrozen conditions, with anRMSEof 33 m3/ha andR2of 0.73. The retrieval error was slightly higher (RMSE= 51 m3/ha andR2= 0.61) under thawing conditions. In the case of the forest stands with the area larger than 30 ha and the relative stocking of at least 70%, the GSV retrieval error decreased toRMSEof 16 m3/ha and toRMSE= 19 m3/ha respectively.