东中国海表层悬浮体浓度卫星遥感反演研究进展

用卫星遥感手段反演海洋表层悬浮体浓度(Suspended Sediment Concentration,SSC)来研究其分布和输运的方法已经被广泛使用。东中国海属于水文和光学性质较为复杂的二类水体,表层悬浮体浓度的分布规律和水体的固有光学特性时空变化大,增加了遥感研究的难度。在对前人的研究进行比较和总结后发现,根据实测SSC数据对不同区域、不同时间段(季节、潮汐周期)建立分段模型可以提高整体反演精度。在选择参与反演的波段时,河口和近岸等高SSC海域以及远岸低SSC海域有各自不同的最优波段组合。高SSC海域常使用水体反射率第二反射峰、第一反射峰前波段作正比波段组合参与反演,低SSC海域常使用水体反射率第一峰波段作正比、峰前波段作反比参与反演。同时,在反演模型中考虑泥沙粒径的影响可以显著提升反演精度,并且也有可能在浅海区突破现有遥感研究手段的水深限制。目前模型精度评价标准使用较为混乱,平均相对误差、平均绝对误差和均方根误差等可以作为综合精度评价指标,模型的稳定性则可以用误差敏感性分析方法验证。高时空分辨率的海色卫星传感器的出现使得海洋短时间尺度事件的研究成为海色遥感研究的趋势之一。

Progress on Retrieval Models of Suspended Sediment Concentration from Satellite Images in the Eastern China Seas

Remote sensing has been widely used to research suspended sediment concentration on sea surface. The hydrology and inherent optical properties of sea waters are very complex in the eastern China seas, which makes building retrieval models from satellite images more difficult. By comparing and summarizing former researches, some conclusions and suggestions about establishing inversion models have been offered. It can improve accuracy to build models separately in different time and regions identified by in-situ data. Areas containing different SSC have their own optimal bands combination to be used in models. In the coastal areas with high SSC, using the combination of the second peak of water reflectance and the increasing part before the first peak as proportional input factors and choosing inverse proportional input factor based on specific spectrum feature of water can offer the best bands' choice. In offshore waters with low SSC, optimal bands' combination are the first peak of reflectance of water and the increasing part before it. And taking sediment grain size into models can also increase accuracy and may break the depth limitation of remote sensing in shallow sea. It's better to use determination coefficient, mean relative error, mean absolute error and root mean square error as the assessment criterion of models' results and its stability can be certified by error sensitivity analysis. Though semi-analytical and neural network models have more explicit physical foundations, empirical models have better precision and applicability. There is a promising trench using remote sensing to study instantaneous oceanic events due to the advancement of high spatial-and-temporal resolution satellites.