Remote Sensing Inversion In The Yangtze Estuarv And Adjacent Waters And Its Diurnal Variation Analysis Of CDOM Based On COCI Image Data

Colored dissolved organic matter(CDOM)is one of ocean color components.The research of distribution,migration and transformation of CDOM in estuarine and coastal waters are not only important for the studying of the ocean color remote sensing,but also have great biological and spectroscopy significance.In this paper,absorption coefficient of CDOM is used as the CDOM concentration index to study the CDOM retrieval.And on this basis,a primary study about the CDOM distribution and its daily variation of the Yangtze River estuary and adjacent waters is resented.For the high turbidity water of mouth of the Yangtze River,the inversion model is not sensitive to the change of CDOM and the retrieval accuracy is low.So the BP neural network model of which the failure data is relatively few comparing with other algorithms,was chosen as the research method.As seen from the precision results,the accuracy of BP neural network model is higher than that of the Moon algorithm and YOC algorithm provided by GOCI standard software GDPS.However,the BP neural network algorithm also has the disadvantage that people need to participate in the study.This study selected the measured data of the Yangtze River Estuary and its adjacent waters field as the research data,and a BP neural network algorithm based on the QAA-E algorithm was established to inverse the relationship of bbp(555)and ap(443).The newly established algorithm applied to the inverse model of GOCI satellite data.The model was validated by the synchronous data of sunstar in April 26,2012,and the results showed that it could be applied to invert absorption coefficient of CDOM of the GOCI satellite data.Based on the above analysis,the distribution of CDOM absorption coefficient and the daily change of CDOM in the Yangtze River Estuary and its adjacent waters were analyzed.The results demonstrate that:1)The BP neural network method based on QAA-E algorithm is a good method for the inversion of CDOM absorption coefficient,which is suitable for CDOM inversion of Yangtze River Estuary and its adjacent sea area.Overall,CDOM inversion accuracy in the high turbidity waters still needs to be improved.Because the high content of suspended matter concentration in Yangtze River Estuary and its adjacent waters has a large effect on the backscattering spectrum and the spectra of chlorophyll and CDOM,thus weakening the correlation between CDOM and the spectra of scattering backward,resulting in lower inversion algorithm effect in complex water.2)The absorption coefficient of CDOM in the Yangtze River Estuary and its adjacent sea area is retrieved by the GOCI image in March 15,2014,and the temporal and spatial characteristics of the diurnal variation are analyzed.By using the QAA-E-BP neural network model,the daily variation of CDOM in the Yangtze Estuary and its adjacent area is retrieved and analyzed,and the daily variation of CDOM is as follows:The daily variation of Yangtze River Estuary and its adjacent waters of CDOM is mainly affected by tide,and the Yangtze River Plume.The concentration of CDOM in the Yangtze Estuary in flood period is higher than the low tide period.Because of the Yangtze diluted water,CDOM from the estuary to the sea area showed gradual decreasing trend.At the same time affected by the mixed dilution of seawater and fresh water,the spatial distribution of CDOM present a pinnate distribution pointing to the northeast.3)Based on the high temporal resolution of GOCI data,it can capture variation of CDOM within a day.Besides,it is helpful for the real-time monitoring of the CDOM cycle process,which provides an important observational data for further study on the diurnal variation characteristics of CDOM and its driving mechanism and estuarine evolution law in the Yangtze River Estuary and its adjacent waters.4)The error of classification,correction and inversion model in quantitative remote sensing needs to be evaluated in order to determine its performance and effectiveness.The approximate distribution of errors is simulated by computer,and the relationship between the sample size n and the statistical indexes RMSE,MAE and UA are studied.The results showed that the variation trends of RMSE,MAE and UA with the sample size of N were different:when n was less than about 40,RMSE and MAE tended to increase with the sample size increasing,then it tended to be gentle;UA always smoothly decreased.It can be seen that in the case of a small amount of sample,UA is more suitable than RMSE and MAE to evaluated the uncertainty(reliability)of remote sensing model,because the larger the sample size,the more reliable the model is(uncertainty less).