基于数据同化的珠江河口悬沙浓度多模型协同反演

Muti-model collaborative retrieval of suspend sediment concentration in the estuary of Pearl River based on data assimilation

以珠江河口为研究区,根据2014-2015年珠江河口野外实验数据,构建了8个悬沙浓度的反演模型,并对各模型反演结果进行了比较评价。基于数据同化思想,利用此模型集进行了多模型协同反演算法的构建。结果表明:构建的悬沙反演模型中近红外与蓝波段比值的指数函数模型整体精度较高,平均绝对误差MAPE为27.1%,多模型协同反演算法能够有效融合不同反演模型的优势,改善单模型反演精度较低区域的反演结果,且随着模型个数的增加,多模型协同估算精度越高,较单模型提高了整体反演精度及稳定性,最优模型组合的MAPE为25.8%。该算法为悬沙浓度大范围、高精度业务化监测提供了一种新的思路。

Accuracy and applicability of different retrival models of suspended sediment concentration(SSC)vary with the complexity of water body.Eight models were established and evaluated by observed data during 2014 to 2015.Then,we tried to Multi-model collaborative retrieval algorithms by these models were established and verified based on data assimilation.Results show that accuracy of exponential function model using ratio of near-infrared and blue reflectance was the best with the MAPE error of 27.1%.The multi-model collaborative retrieval algorithms effectively blend the advantages of different retrieve models and improved the accuracy of the single model for the lower retrive accuracy region.The more models participating were in the multi-model collaborative retrieval algorithm,the accuracy was better.Compared with the single model,the accuracy and stability of the multi-model collaborative retrieval algorithms were better than that of any single model,and the MAPE of optimal combination was 25.8%.The proposed algorithm provides a new way for large scale and high accuracy operational monitoring of suspended sediment.