Seasonal Variation Of Sea Surface Partial Pressure Of Carbon Dioxide In The East China Sea And Its Estimation Using Satellite Remote Sensing

Since the industrial revolution,the global atmospheric carbon dioxide（CO₂）has increased by～50%mainly due to the consumption of fossil fuels and land-use change,the ocean absorbs approximately 26%of the annual anthropogenic CO₂ emissions.Correspondingly,the continuously uptake CO₂ by ocean has resulted in changes to the ocean carbonate system,the partial pressure of CO₂（pCO₂）as one of the important variables of the seawater carbonate system,plays an important role in understanding ocean acidification and estimating the global ocean carbon budget.However,the East China Sea（ECS）as one of the most important marginal seas in China with limited observed data of carbonate parameters,and the physical and biogeochemical mechanisms affecting pCO₂ dynamics and estimated air-sea CO₂ flux（FCO₂）are very complicated.In order to better understand the carbonate chemistry of the ECS,it is critical to improve the coverage of pCO₂ data with sufficient temporal and spatial resolution,on this basis,to analyze the time-series pCO₂ variabilities and their controlling processes.In recent years,there are many studies have been attempted to retrieve pCO₂ of the global coastal oceans based on satellite remote sensing products,because it benefits from the advantages of remote sensing.Although some progress has been made in retrieving surface pCO₂ or other carbonate parameters in the ECS from space,there is still room for studying in interannual variations on the surface pCO₂ and improving the accuracy of estimating FCO₂ in the ECS.Therefore,we developed a multiple nonlinear regression（MNR）model to retrieve the surface pCO₂ from space in the ECS based on observed and synchronized satellite remote sensing data to address the gap of insufficient pCO₂ data coverage in the ECS,expand our knowledge of seasonal and interannual variations of pCO₂,and improve the accuracy of the estimated FCO₂ of the ECS.The following three studies were undertaken in this thesis:（1）The seasonal variations of pCO₂ in the ECS inner shelf are mainly controlled by thermodynamic process,physical mixing,biological activity and Changjiang River discharge.Three field surveys were conducted in the outer Changjiang Estuary on the inner shelf of the ECS in March,July,and October,2018.The surface waters were grouped into three types,namely,Yellow Sea Water（YSW）,Changjiang Diluted Water（CDW）and East China Sea Shelf Waster（ECSSW）.We studied the observations in the surface water of total-scale p H（p H_T）,total alkalinity（A_T）,and calculated total dissolved inorganic carbon（C_T）,pCO₂,and FCO₂,and their relationships with various water masses in different seasons.The results showed that the CDW area was a source of atmospheric CO₂ in July and October(4.97 and 8.67 mmol CO₂ m^-2 d^-1,respectively).The oversaturation of CO₂ was mainly ascribed to the respiration of terrestrial organic and inorganic materials sourced from the Changjiang River discharge,overwhelming the CO₂ uptake due to primary productivity despite the high phytoplankton biomass in summer.The FCO₂ was greater in October than that in July in the CDW,which is attributed to the increasing wind speed.In contrast,the YSW and ECSSW were acted as weak CO₂ sink in March(-0.71 and-2.86 mmol CO₂ m^-2d^-1,respectively)and July(-1.28 mmol CO₂ m^-2 d^-1 in the ECSSW)following the CO₂uptake of phytoplankton production,however,they were a CO₂ source by October(3.30 mmol CO₂ m^-2 d^-1 in the YSW and 1.18 mmol CO₂ m^-2 d^-1 in the ECSSW).The cooling effect during the cold season reduced the sea surface pCO₂,resulting in a CO₂sink in the CDW,YSW,and ECSSW areas in March.However,the regions became a source of atmospheric CO₂ in October,possibly driven by vertical mixing,which brought C_T-enriched bottom water to the surface and increased the pCO₂.The study region was a net CO₂ sink in March and a net CO₂ source in July and October with an average FCO₂ of-1.25,1.71,and 3.06 mmol CO₂ m^-2 d^-1,respectively.It must be pointed out that the sink or source pattern of the outer Changjiang Estuary changes depending on the time and location of sampling.Therefore,the high-frequency and long-term observations of pCO₂ are needed to further identify the outer Changjiang Estuary is a CO₂ sink/source.（2）A random forest model with bagged trees as kernel function（RF＿bagged trees）was developed to retrieve the sea surface salinity（SSS）of the ECS with a spatial resolution of～1 km based on Moderate Resolution Imaging Spectroradiometer（MODIS）products.The study（1）revealed the major controlling processes of the pCO₂ dynamics,closely linked to these processes,several environmental variables（e.g.sea surface temperature（SST）,SSS,Chlorophyll a（Chl a））can be used to estimate surface pCO₂by developing an optimal model.As we all know,the SST and Chl a data with high spatial and temporal resolution are easily available.SSS is a master variable in oceanography and important to understand water mixing processes,however,it is a challenge to obtain SSS data with sufficient temporal and spatial resolution.Therefore,we developed a RF＿bagged trees model to estimate SSS of the ECS with a spatial resolution of～1 km based on a large synchronous dataset of in-situ SSS observations,MODIS-derived remote sensing reflectance(R_rs)and SST,with a root mean square error（RMSE）of～0.84.Meanwhile,the performances of other traditional empirical approaches（multiple linear regression,MNR）,machine-learning based empirical approaches with different kernel functions（decision tree,random forest,support vector machine regression,and multilayer perceptron neural network）and semi-analytical method were also tested,results showed that the RF＿bagged trees showed the best performance compared to the other tested methods with the same input variables.The accuracy of the SSS model was examined using an independent dataset during the period of 2020 to 2022 with an RMSE of 0.66.The SSS model was applied to retrieve the distribution patterns of monthly SSS for the ECS from August 2002 to July 2022 based on MODIS-derived R_rs and SST.The results showed that the lowest SSS generally dominated the Changjiang Estuary and nearshore coast,the intermediate SSS is mainly distributed in the transitional region and the high SSS is distributed in the open water influenced by the high-salinity Kuroshio（KS）,indicating that the distribution patterns are closely affected by the circulation of water currents.The Changjiang discharge and the East-Asian monsoon are important drivers for the formation of Changjiang low-salinity plume.The seasonal cycle of SSS in the ECS showed that the SSS gradually decreases from spring to summer,reaching a minimum value in August,then slowly increases from autumn to winter,remaining almost constant in winter.These dynamics were in good agreement with the Changjiang River discharge.The RF_<sub>bagged trees-based SSS model is a robust approach for SSS modeling from space in the coastal oceans,but it relies on the feature of the training dataset,the RF＿bagged trees model can be applied to other coastal oceans retraining the model using local dataset.（3）A MNR model was developed to estimate the pCO₂ for three different physical-biogeochemical domains of the ECS and retrieved the monthly pCO₂ for the whole ECS based on MODIS-derived products and RF_<sub>bagged trees-derived SSS from 2002 to 2022.A MNR model was developed to estimate pCO₂ for three different physical-biogeochemical domains of the ECS by employing in-situ hydrographic（SST and SSS）and carbonate system observations combined with satellite observations of Chl a,with an RMSE of 45.19μatm and R² of 0.87 in Domain I,an RMSE of 11.59μatm and R² of 0.92 in Domain II and an RMSE of 3.73μatm and R²of 0.97 in Domain III.The reliability of the MNR model was also validated using an independent dataset with a good accuracy（RMSE of 9.15μatm for Domain II and RMSE of 7.71μatm for Domain III）.Meanwhile,the MNR model had the better performance compared with the semi-analytical algorithm based on the independent dataset.The MNR model was applied to estimate the surface pCO₂ based on ten cruise observed SST,SSS and Chl a in winter,spring and summer from 2013 to 2018,these spatial distributions of estimated pCO₂ were consistent with previous studies reported.The high pCO₂ was found near the Changjiang Estuary and Hangzhou Bay,while low pCO₂ was occurred in the southeast corner of the survey area.We further calculated FCO₂ based on MNR-derived pCO₂,the results showed that the ECS inner shelf was a CO₂sink in winter and spring,and a CO₂ source in summer,this agreed with our study（1）.In addition,the MNR model was used to simulate the monthly pCO₂ of the ECS inner shelf based on the FVCOM outputs from 2000 to 2016,the seasonal cycles suggested that pCO₂ gradually decrease from winter to spring,and display its minimum in May,mainly attributed to spring bloom under increasing SST;the pCO₂gradually increase from spring to autumn and display its maximum in September,mainly due to the enhanced vertical mixing in autumn,which caused the CO₂-enriched subsurface or bottom water mixed into the surface.From the time-series of surface pCO₂ and FCO₂ on the ECS inner shelf showed an increasing trend of 3.32±0.79μatm yr^-1 and 0.018±0.0125 mol CO₂ m^-2 yr^-2,and the region was a CO₂ sink with an CO₂ uptake of 0.12±0.48 Tg C yr^-1.Lastly,the MNR model was also used to retrieve the monthly pCO₂ for the whole ECS based on MODIS-derived products and RF_<sub>bagged trees-derived SSS from 2002 to 2022,showed reasonable distribution patterns,and suggested that the Changjiang discharge may affect the distribution of pCO₂ in Domain II during wet season,the seasonal variations of pCO₂ in Domain III are mainly controlled by thermodynamic process.Moreover,the area-averaged monthly time series of surface pCO₂ in the entire ECS and three domains all showed clear increasing trends of 2.46±0.23,2.22±0.42,2.40±0.20 and 2.73±0.25μatm yr^-1,respectively.The MNR model can be used as an effective way to improve the coverage of pCO₂ data in the ECS to a certain extent.Satellite remote sensing has become an important tool for synoptic estimation of sea surface pCO₂ due to its advantages of spatial and temporal resolution and coverage.Developing an optimal model to retrieve carbonate parameters from space is a cost-effective way to address the limitations of ocean observations from ship-based measurements.The simulated pCO₂ from space improves the coverage of pCO₂ data in the ECS,and also can be used as initial conditions or verification data for ocean carbon system models.Meanwhile,our study also contributes to the understanding of carbon cycle and carbon budget in the global coastal oceans.Moreover,the accuracy of pCO₂ model in the ECS needs to be further improved,and the pCO₂ model can be extended to other coastal oceans by retraining the model using local dataset.