Retrieval of Leaf Biochemical Parameters Using PROSPECT Inversion: A New Approach for Alleviating Ill-Posed Problems
Retrieval of Leaf Biochemical Parameters Using PROSPECT Inversion: A New Approach for Alleviating Ill-Posed Problems
复制标题
DOI：
10.1109/tgrs.2011.2109390
复制
发表时间：
2011-03
影响因子：
8.2
通讯作者：
Pingheng Li;Quan Wang
中科院分区：
文献类型：
--
作者：
Pingheng Li;Quan Wang
利用模型反演从反射率测量中提取叶片生化参数通常面临“病态”问题，这极大地降低了逆模型的估计精度。虽然模型反演的标准方法通常只基于一个价值函数同时提取各种参数，但本文提出的新方法为每个提取的参数分配了一个特定的价值函数。每个价值函数是根据在较早的灵敏度分析中发现给定参数对其特别敏感的波长域来指定的。该方法已用现场测量数据集和PROSPECT模型模拟的10000个光谱的人工数据集进行了验证。结果表明，该方法大大提高了模型的精度，基于模拟数据的叶绿素含量、当量水厚度和单位面积叶质量的均方根误差分别为7.12μg/cm~2、0.0012 g/cm~2和0.0019 g/cm~2，而标准方法分别为11.36μg/cm~2、0.0032 g/cm~2和0.0040 g/cm~2。对于野外测量的数据集，该方法的性能也大大优于标准方法，当所有数据合并时，CHL、EWT和LMA的RMSE值分别为8.11μg/cm~2、0.0012 g/cm~2和0.0008 g/cm~2，而使用标准方法时分别为11.84μg/cm~2、0.0020 g/cm~2和0.0027 g/cm~2。因此，本文提出的模型反演方法可以在很大程度上缓解“病态”问题，并可广泛应用于叶片生化参数的反演。
Retrieval of leaf biochemical parameters from reflectance measurements using model inversion generally faces “ill-posed” problems, which dramatically decreases the estimation accuracy of an inverse model. While the standard approach for model inversion retrieves various parameters simultaneously, usually only based on one merit function, the new approach proposed in this paper assigns a specific merit function for each retrieved parameter. Each merit function is specified in terms of the wavelength domains that the given parameter was found to be specifically sensitive to in an earlier sensitivity analysis. The approach has been validated with both in situ measured data sets and an artificial data set of 10 000 spectra simulated by the PROSPECT model. Results indicate that the new approach greatly improves the performance of inversion models, with root-mean-square error (rmse) values for chlorophyll content (Chl), equivalent water thickness (EWT), and leaf mass per area (LMA), based on the simulated data, of 7.12 μg/cm2, 0.0012 g/cm2 , and 0.0019 g/cm2, respectively, compared with 11.36 μg/cm2, 0.0032 g/cm2, and 0.0040 g/cm2 when using the standard approach. As for field-measured data sets, the proposed approach also greatly outperformed the standard approach, with respective rmse values of 8.11 μg/cm2, 0.0012 g/cm2, and 0.0008 g/cm2 for Chl, EWT, and LMA when all data are pooled, compared with 11.84 μg/cm2, 0.0020 g/cm2, and 0.0027 g/cm2 when using the standard approach. Hence, the proposed approach for model inversion can largely alleviate the “ill-posed” problem, and it could be widely applied for retrieving leaf biochemical parameters.