Abstract Deriving inherent optical properties (IOPs) from multispectral imagery of shallow water environments using physics-based inversion models require prior knowledge of the spectral reflectance of the bottom substrate. The use of an incorrect bottom reflectance adversely affects the IOPs and, in part, the depth derived from inversion models. To date, an operational approach that determines the bottom reflectance from multispectral imagery is lacking; development in this area is especially paramount for locations that exhibit temporal variability in the spatial distributions of submerged aquatic vegetation and benthic microalgae. In this work, we develop a multispectral implementation of the HOPE-LUT algorithm (Hyperspectral Optimization Processing Exemplar with benthic Look Up Table), and apply the approach to MERIS imagery of the Great Bahama Bank (GBB). Overall benthic classification accuracy of this approach was 80.0%, revealing the areal coverage of benthic flora can range from 1052.3 km2 to 6169.3 km2 between years in the Exumas, GBB. Comparison of HOPE-LUT IOP retrievals to common inversion model implementations (particularly HOPE, with its default sand endmember) shows that using an incorrect bottom reflectance can lead to over-estimations in aphy(443) (absorption coefficient of phytoplankton at 443 nm), of up to 95%, under-estimations of adg(443) (absorption coefficient of detritus and gelbstoff) up to 50%, and over-estimations of depth up to 20%. In addition, the HOPE-LUT parameterizations generate IOPs within the range of those measured in situ. We demonstrate that, at the scale of a MERIS pixel, the dominant substrates of seagrass, unattached bottom macroalgae and benthic microalgae are spectrally unresolvable at the depths that these classes occur in the GBB. Lastly, we evaluate the performance of commonly used atmospheric corrections algorithms for bathymetry estimation and benthic classification accuracy. The combined benthic classification and inversion scheme presented here is autonomous, i.e., it does not require scene-specific thresholds or modifications. Thus, it should be portable to Sentinel 3 OLCI and potentially MODIS Aqua imagery to obtain a continuous time series of changes in IOPs and benthic cover for the shallow waters over the Great Bahama Bank.

中文翻译：

---

使用 MERIS 图像在浅水环境中进行底栖分类和 IOP 反演

摘要 使用基于物理的反演模型从浅水环境的多光谱图像中推导出固有光学特性 (IOP) 需要了解底部基底的光谱反射率。使用不正确的底部反射率会对 IOP 产生不利影响，并在一定程度上影响从反演模型得出的深度。迄今为止，缺乏从多光谱图像确定底部反射率的操作方法；对于在沉水植物和底栖微藻的空间分布上表现出时间变化的地点，该领域的发展尤其重要。在这项工作中，我们开发了 HOPE-LUT 算法（具有底栖查找表的超光谱优化处理示例）的多光谱实现，并将该方法应用于大巴哈马银行 (GBB) 的 MERIS 图像。这种方法的总体底栖分类精度为 80.0%，揭示了 GBB 埃克苏马斯底栖植物群的面积覆盖范围为 1052.3 平方公里至 6169.3 平方公里。HOPE-LUT IOP 检索与常见反演模型实现（特别是 HOPE，具有默认的沙子端元）的比较表明，使用不正确的底部反射率会导致 aphy(443)（浮游植物在 443 nm 处的吸收系数）的高估，高达 95%，低估 adg(443)（碎屑和凝胶的吸收系数）高达 50%，高估深度高达 20%。此外，HOPE-LUT 参数化会在原位测量的范围内生成 IOP。我们证明，在 MERIS 像素的尺度上，海草、未附着的底部大型藻类和底栖微藻的主要基质在这些类别出现在 GBB 的深度上在光谱上无法解析。最后，我们评估了常用的大气校正算法在测深估计和底栖分类精度方面的性能。这里提出的组合底栖分类和反演方案是自主的，即它不需要特定场景的阈值或修改。因此，它应该可移植到 Sentinel 3 OLCI 和潜在的 MODIS Aqua 图像，以获得大巴哈马银行浅水区的 IOP 和底栖覆盖物的连续时间序列变化。我们评估了常用的大气校正算法的性能，用于测深估计和底栖分类精度。这里提出的组合底栖分类和反演方案是自主的，即它不需要特定场景的阈值或修改。因此，它应该可移植到 Sentinel 3 OLCI 和潜在的 MODIS Aqua 图像，以获得大巴哈马银行浅水区的 IOP 和底栖覆盖物的连续时间序列变化。我们评估了常用的大气校正算法的性能，用于测深估计和底栖分类精度。这里提出的组合底栖分类和反演方案是自主的，即它不需要特定场景的阈值或修改。因此，它应该可移植到 Sentinel 3 OLCI 和潜在的 MODIS Aqua 图像，以获得大巴哈马银行浅水区的 IOP 和底栖覆盖物的连续时间序列变化。