With the longest archive of satellite remote sensing images, the Landsat series of satellites have demonstrated their great potential in aquatic environmental studies. However, although various atmospheric correction (AC) methods have been developed for Landsat observations in water color applications, a comprehensive assessment of their accuracies across different AC methods and instruments has yet to be performed. Using in situ spectral data collected by Aerosol Robotic Network-Ocean Color (AERONET-OC) sites, the performances of five types of AC methods over three different Landsat missions (i.e., Landsat 5/7/8) were evaluated. The Landsat 8 Operational Land Imager (OLI) showed more accurate AC retrievals than the other two instruments, and the results for its green and red bands appeared more reliable than those for the other wavelengths (uncertainty levels of ∼30 %). The iterative NIR algorithm with 2-bands (NIR-SWIR2) model selection embedded in SeaDAS showed the best performances for OLI in two blue bands. Moreover, larger residual errors were found for most Landsat 5/7 bands regardless of the AC methods and spectral bands employed with an uncertainty of >50 %. Interestingly, a simple aerosol subtraction method over the Rayleigh-corrected reflectance (Rrc) outperformed the exponential extrapolation (EXP) algorithms, especially for Landsat 5/7. Neither the image-based AC algorithm nor the surface reflectance (SR) products provided by the United States Geological Survey (USGS) showed acceptable performances over coastal environments. The uncertainties in the various Landsat reflectance products over water surfaces could be associated with a relatively poor signal-to-noise ratio (SNR) in addition to radiometric calibration uncertainties, imperfect aerosol removal methods. Future research is required to collect in situ data across a wider range of water optical properties (particularly more turbid inland waters) to examine the corresponding applicability of Landsat-series observations.

拥有最长的卫星遥感图像档案，Landsat系列卫星已证明了它们在水生环境研究中的巨大潜力。但是，尽管已经为水彩应用中的Landsat观测开发了各种大气校正（AC）方法，但尚未对各种AC方法和仪器的精度进行全面评估。使用Aerosol机器人网络-海洋颜色（AERONET-OC）站点收集的原位光谱数据，评估了五种类型的AC方法在三种不同的Landsat任务（即Landsat 5/7/8）上的性能。与其他两种仪器相比，Landsat 8实用陆地成像仪（OLI）显示出更精确的AC检索，并且其绿色和红色谱带的结果似乎比其他波长的结果更可靠（不确定度约为30％）。嵌入SeaDAS的带有2波段模型选择的迭代NIR算法（NIR-SWIR2）在两个蓝色波段显示了OLI的最佳性能。此外，无论采用哪种交流方法和所使用的光谱带，对于大多数Landsat 5/7波段，都发现较大的残留误差，不确定度> 50％。有趣的是，在瑞利校正反射率（Rrc）上使用简单的气溶胶减法的效果优于指数外推（EXP）算法，尤其是对于Landsat 5/7。美国地质调查局（USGS）提供的基于图像的AC算法和表面反射率（SR）产品均未在沿海环境中表现出可接受的性能。除辐射定标不确定性，不完善的浮质去除方法外，水面上各种Landsat反射率产品的不确定性还可能与相对较差的信噪比（SNR）相关。需要进行进一步的研究以收集更广泛水光学特性（尤其是内陆水域更浑浊）的原位数据，以检验Landsat系列观测的相应适用性。