We present a three-step inverse model (3SAA) for estimating the inherent optical properties (IOPs) of surface waters from the remote sensing reflectance spectra, R_rs(λ). The derived IOPs include the total (a(λ)), phytoplankton (a_phy(λ)), and colored detrital matter (a_cdm(λ)), absorption coefficients, and the total (b_b(λ)) and particulate (b_bp(λ)) backscattering coefficients. The first step uses an improved neural network approach to estimate the diffuse attenuation coefficient of downwelling irradiance from R_rs. a(λ) and b_bp(λ) are then estimated using the LS2 model (Loisel et al., 2018), which does not require spectral assumptions on IOPs and hence can assess a(λ) and b_b(λ) at any wavelength at which R_rs(λ) is measured. Then, an inverse optimization algorithm is combined with an optical water class (OWC) approach to assess a_phy(λ) and a_cdm(λ) from a_nw(λ).The proposed model is evaluated using an in situ dataset collected in open oceanic, coastal, and inland waters. Comparisons with other standard semi-analytical algorithms (QAA and GSM), as well as match-up exercises, have also been performed. The applicability of the algorithm on OLCI observations was assessed through the analysis of global IOPs spatial patterns derived from 3SAA and GSM. The good performance of 3SAA is manifested by median absolute percentage differences (MAPD) of 13%, 23%, 34% and 34% for b_bp(443), a_nw(443), a_phy(443) and a_cdm(443), respectively for oceanic waters. Due to the absence of spectral constraints on IOPs in the inversion of total IOPs, and the adoption of an OWC-based approach, the performance of 3SAA is only slightly degraded in bio-optical complex inland waters.

我们提出了一个三步逆模型 (3SAA)，用于从遥感反射光谱R _rs (λ)估计地表水的固有光学特性 (IOP )。导出的 IOP 包括总量 ( a (λ))、浮游植物 ( a _phy (λ)) 和有色碎屑物质 ( a _cdm (λ))、吸收系数以及总量 ( b _b (λ)) 和颗粒物 ( b _bp (λ)) 反向散射系数。第一步使用改进的神经网络方法从R _rs估计下流辐照度的漫衰减系数。a (λ) 和b _bp(λ) 然后使用 LS2 模型（Loisel 等人，2018 年）进行估计，该模型不需要对 IOP 进行光谱假设，因此可以在R _rs (λ) 的_任何波长处评估a (λ) 和b _b (λ) ) 进行测量。然后，逆优化算法用光学水类（OWC）的方式相结合，以评估一个_PHY（λ）和一个_CDM从（λ）一个_NW(λ)。使用在开阔的海洋、沿海和内陆水域收集的原位数据集对所提出的模型进行评估。还进行了与其他标准半分析算法（QAA 和 GSM）的比较以及匹配练习。该算法对 OLCI 观测的适用性是通过分析源自 3SAA 和 GSM 的全局 IOP 空间模式来评估的。3SAA的良好性能是由13％，23％，34％和34％为中位数绝对百分比差异（MAPD）表现b _bp的（443），一个_NW（443），一个_PHY（443）和一个_CDM(443)，分别为海洋水域。由于在总 IOPs 的反演中没有对 IOPs 的光谱限制，并且采用基于 OWC 的方法，3SAA 的性能在生物光学复杂的内陆水域中仅略有下降。