| 摘要: |
| 介绍了一种基于辐射传递模拟和人工神经网络技术(ANN)的二类水体水色要素(CHL,SPM,CDOM)的反演算法.在辐射传递模拟计算中,纯海水吸收和散射、浮游植物吸收的数据或模型是已发表的被广泛采用的结果.黄色物质和非浮游植物颗粒吸收以及海洋颗粒物的散射模型从COASTLOOC数据中导出.另外,还利用了一个新的海洋颗粒物后向散射概率模型,在该模型中颗粒物后向散射概率是颗粒有机物与SPM比值和波长的函数.把上述定义的固有光学性质作为输入,经过辐射传递模拟得到海表面以下辐照度反射比数据集,然后将该模拟数据集用于训练不同的人工神经网络,获取水色和水色要素浓度之间函数关系的最佳近似.利用以上建立的基于人工神经网络的算法,把COASTLOOC数据集和PMNS数据集的辐照度反射比作为输入进行水色要素反演,通过比较反演值和真实测量值来评价算法性能.结果显示,建立的基于ANN的二类水体水色要素反演算法具有很好的性能. |
| 关键词: 海洋辐射传递模拟 人工神经网络 光学组分的反演 二类水体 |
| DOI：10.18307/2009.0204 |
| 分类号: |
| 基金项目:国家自然科学基金项目(40876005)资助 |
|
| Algorithms based on artificial neural network for retrieval of oceanic constituents in Case II waters |
|
ZHANG Tinglu, QIU Guoqiang
|
|
Ocean Remote Sensing Laboratory of Ministry of Education of China, Ocean University of Qingdao, Qingdao 266100, P. R. China
|
| Abstract: |
| In the present paper, we report an algorithm method to retrieve the oceanic constituent concentrations (CHL, SPM, and CDOM) in Case II waters. The method is derived from the radiative transfer simulations and is subsequently applied in the Artificial Neural Network (ANN) techniques. Information on absorption and total scattering of pure sea water, as well as absorption of phytoplankton and associated particles are taken from measurements or parameterisations in published literatures, and information on absorption of coloured dissolved organic matter and nonagal particles, as well as scattering of marine particles were derived from the COASTLOOC data set. Additionally, a new model on the backwards scattering probability model is used, of which probability is a function of the organic particulate matter and the total particulate matter (SPM) ratio and wavelength. Such defined inherent optical properties are input as a radiative transfer code in order to generate a synthetic data set of hemispherical reflectance spectra, subsequently used for the training of various ANNs to find the best approximation of the functional relationship between ocean colour and oceanic constituent concentrations. The performance of the ANN-based retrieval schemes is assessed by applying it to the hemispherical reflectance spectra contained in the COASTLOOC data set and PMNS data set, and comparing the retrieved oceanic constituent concentrations to those actual measurements. The results show that the ANN-based algorithms have good performance in retrieval of oceanic constituents for ocean colour measurements in Case II waters. |
| Key words: Ocean radiative transfer simulation artificial neural network retrieval of oceanic constituents case II waters |
