大气校正模型对多光谱水深反演影响的多维度分析

大气校正是水体定量遥感的基础与前提。本文从大气校正模型、大气校正模型参数、水体组分差异以及水深反演波段组合方式4个维度探讨大气校正模型对水深反演的影响。研究采用6S、FLAASH、ACOLITE与QUAC 4种大气校正模型，选取大陆型、海洋型与城市型气溶胶模式，以瓦胡岛西北侧与谢米亚岛周边浅水作为清洁水体研究区，以辽东浅滩与槟城海峡作为浑浊水体研究区，基于Landsat-8多光谱影像开展大气校正，并采用8种波段组合方式进行水深遥感反演。研究结果表明：（1）4种大气校正模型均可在一定程度上削弱大气对水体信号的影响；因参数选取以及研究区水体组分的不同，不同模型的校正结果存在一定差异；两类水体反射率峰值分别出现在蓝波段与绿波段；（2）6S大气校正模型鲁棒性较强，该模型因研究区水体组分发生变化导致对应的水深反演结果与其余模型相比波动较小；FLAASH模型在海洋型和城市型两种气溶胶模式水深反演结果在浑浊水体存在较为明显的差异，辽东浅滩浅水区平均相对误差相差7.9%；ACOLITE模型受水体类型影响显著且对浑浊水体具有优越性与稳定性，平均相对误差较FLAASH降低5.6%；（3）多波段水深反演精度普遍优于单波段，但反演精度与波段数目之间无显著的相关性；水深反演波段组合方式对不同研究区敏感性不同，清洁水体三波段模型的反演精度较好，浑浊水体中四波段模型的反演精度最优，平均相对误差较三波段模型降低达5.6%。

Multi-dimensional analysis of atmospheric correction models on multi-spectral water depth inversion

Atmospheric correction (AC) is the basis and premise of quantitative remote sensing of water column. The effects of different AC models on water depth inversion from the four aspects of AC model, AC model parameters, water component differences, and water depth inversion band combination are discussed in this paper. The research uses 6S, FLAASH, ACOLITE and QUAC four AC models, select continental, marine and urban aerosol patterns, and the shallow waters around the northwest side of Oahu Island and Shemya Island are used as the study area of clean water, while the shallow waters around Liaodong Shoal and Penang Strait are used as the study area of turbid water. AC is performed based on Landsat-8 multispectral images, and eight wavebands are used for bathymetric remote sensing inversion. The results show that: (1) all the four AC models can weaken the atmospheric influence on the water signal to some extent; the correction results of different models are somewhat different depending on the parameter selection and the components of the water column. And the peak reflectance of the two types of water column occurs in the blue and green bands, respectively. (2) The 6S model is more robust, and the bathymetric inversion results of this model are less volatile than the rest of the models due to the changes in the components of the water column. The water depth inversion results of the two aerosol models of the FLAASH have more obvious differences in turbid water, and the difference of MRE in shallow water of Liaodong Shoal is 7.9%; the ACOLITE model is significantly influenced by the water column type and has superiority and stability for turbid water, and the MRE is 5.6% lower than that of FLAASH. (3) The accuracy of multi-band water depth inversion is generally better than that of single-band, but there is no significant correlation between the accuracy of inversion and however, there is no significant correlation between the inversion accuracy and the number of bands; the combination of bathymetric inversion bands has different sensitivity to different study areas, the inversion accuracy of the three-band model is better in clean water, and the inversion accuracy of the four-band model is optimal in turbid water, and the MRE is reduced by 5.6% compared with the three-band model.