Remote Sensing Estimation Of Dissolved Organic Carbon And Its Temporal And Spatial Dynamics In The Yangtze River Estuary And Adjacent Seas

Dissolved Organic Carbon（DOC）is the largest active carbon reservoir in the ocean and has significant impacts on global carbon cycling and climate change.As a key link for land-ocean DOC exchange,the Yangtze River Estuary（YRE）has multiple complex dynamic processes and is a hotspot for marine carbon cycle research.Satellite remote sensing technology has advantages of large area coverage and long-term observation compared to traditional in situ surveys.Currently,the research on DOC satellite estimation in the complex water bodies of the YRE is relatively limited.Therefore,optimizing and constructing a remote sensing estimation algorithm for DOC based on in situ survey data can help to achieve satellite monitoring of surface DOC distribution in the YRE,which contributes to a better understanding of the mechanism of carbon cycling in marginal seas.Colored Dissolved Organic Matter（CDOM）is closely related to DOC in estuarine waters and can serve as a proxy for DOC remote sensing.In this study,based on 232 water parameters and 71 remote sensing reflectance data obtained from five cruises in the YRE in 2021,a remote sensing estimation algorithm of CDOM parameters(CDOM absorption coefficient at 300 nm,a _CDOM（300）,and spectral slope between 275～295 nm,S_275-295)was developed by comparing various algorithms.On this basis,a two-step algorithm based on CDOM was constructed to investigate the spatiotemporal dynamics and driving mechanisms of DOC using the GOCI（Geostationary Ocean Color Imager）images in 2020.The main conclusions are as follows:（1）A machine learning-based estimation model for CDOM in the YRE,which is suitable for a small number of sample datasets,was improved.In the variable analysis,it was found that the red and green spectral bands were more sensitive to changes in a _CDOM（300）and S_275-295 than the blue spectral band.In the analysis of model structure,the performance of five machine learning models（Support Vector Regression,Nu Support Vector Regression,Random Forest,Extreme Gradient Boosting,and Backpropagation Neural Network）and traditional algorithms（multiple linear regression and Quasi-Analytical Algorithm）in CDOM remote sensing estimation was compared.It was found that the Nu Support Vector Regression algorithm had the best retrieval accuracy for a _CDOM（300）and S_275-295,with a mean absolute percent difference（MAPD）of 32%and 8.6%,respectively,and the CDOM distribution depicted was relatively reasonable.（2）A CDOM-based estimation model for DOC in the YRE was developed.Three CDOM-DOC empirical equations were established based on measurements from multiple seasons,including a non-seasonal a _CDOM（300）-DOC model,a seasonal a _CDOM（300）-DOC model,and an S_275-295-a_CDOM^*（300）model based on the CDOM-specific absorption coefficient(a_CDOM^*（300）,representing the ratio of a _CDOM（300）to DOC).The results showed that the seasonal a _CDOM（300）-DOC model accurately retrieved DOC concentrations,with MAPD values of 11.5%and 13.8%in the measured validation set and satellite validation set,respectively.In addition,a comparison was made with a multiple linear regression algorithm based on remote sensing reflectance,which had lower accuracy than the two-step algorithm（MAPD=20.4%）.（3）The distribution of surface DOC concentration in the YRE exhibits significant spatiotemporal heterogeneity.In terms of spatial distribution,DOC concentration shows a characteristic of high nearshore(≥1.2 mg·L^-1)and low offshore(≤1.2 mg·L^-1),and the greatest variation occurs in the maximum turbidity zone with a coefficient of variation of 25%～27%.In terms of seasonal variation,the DOC concentration is higher in summer(1.4 mg·L^-1)and autumn(1.27 mg·L^-1),and lower in spring(1.11 mg·L^-1)and winter(1.24 mg·L^-1).In terms of diurnal variation,the variation of DOC in nearshore waters with high concentration(≥1.5 mg·L^-1)is greater than that in shelf waters with low concentration(≤1.3 mg·L^-1).（4）Runoff,tides,and winds have varying degrees of influence on the distribution of surface DOC concentration in the YRE.Yangtze River runoff has a significant monthly variation effect on DOC concentration（correlation coefficient r>0.54,p<0.001）,with an increase in concentration during the flood season（July to September）and a decrease during the dry season（January to April and November to December）.Tides have a significant negative correlation with diurnal variation of DOC concentration,especially in the mouth（r<-0.89,p<0.001）.Winds have an impact on the distribution of surface DOC concentration.In winter,the nearshore waters are significantly affected by DOC under the influence of north winds（r=-0.64,p<0.001）,while in summer,the shelf waters are significantly affected by DOC under the influence of south winds（r=0.98,p<0.001）.