Construction And Application Of Sea Surface Salinity Inversion In The Sea Area Adjacent To The Changjiang Estuary From Muti-satellite Data

Sea surface salinity is one of the most important basic parameters in the field of oceanographic research,and plays an important role in the study of ocean circulation,land-sea interaction,marine environmental monitoring,and marine biological monitoring.The cost of obtaining sea surface salinity data by traditional navigation is high and the timeliness is poor,which cannot meet the needs of dynamic ocean observation.In recent years,the rapid development of remote sensing technology has provided a new opportunity for sea surface salinity observation.Under the guidance of effective models,dynamic ocean observation needs with a wide range and strong timeliness can be realized.In this paper,based on SMAP(Soil Moisture Active and Passive)and MODIS(Moderate Resolution Imaging Spectroradiometer)satellite data and various measured data,a combined inversion model of sea surface salinity in the Sea Area Adjacent to the Changjiang Estuary is constructed,which effectively solves the data time span of salinity satellites.shortage problem.The main work is as follows:(1)A machine learning inversion model of SSS in the Sea Area Adjacent to the Changjiang Estuary based on SMAP and MODIS data was constructed.Considering the possible mapping relationship between sea surface temperature,remote sensing reflectivity and sea surface salinity,this paper takes the SMAP satellite level 3 monthly average sea surface salinity product as the true value,and the MODIS level 3 monthly average sea surface temperature and sensitive reflectivity band combination as the Input,use random forest,particle swarm optimization support vector regression(PSO-SVR)and automatic machine learning(TPOT)to model and verify the effect.The experimental results show that:in the flood season of the Yangtze River,the model constructed by random forest has the best performance,the test set accuracy R~2 is 0.870,MSE is 0.694;in dry season,the limit random tree output by TPOT has the highest test set accuracy,R~2is 0.871,MSE is 0.298.The optimal model was applied to the 4km daily average MODIS data for inversion and matched with the SSS measured by the Korea Fisheries Center.The verification results showed that within the range of 30～34psu SSS measured data,the MAE in the flood season was 0.503psu,and the MAE in the dry season was 0.263 psu.(2)A nearshore supplementary inversion model based on MODIS data and measured data from the East China Sea Station is constructed.Aiming at the problem that the nearshore error of SMAP data leads to the poor performance of the machine learning inversion model in the low salinity area,this paper uses the measured SSS data of the Yangtze Estuary buoys and the MODIS2-level daily average reflectivity to construct a supplementary statistical model for the nearshore waters of the Yangtze River Estuary.The experimental results show that the ratio of the wavelength band 667nm and 488nm has the highest correlation with the measured data,and the verification accuracy is R~2=0.843 and RMSE=1.29psu.(3)The model combination inversion process is constructed.In this paper,the area below 30psu in the sea surface salinity results obtained by the machine learning inversion model constructed based on SMAP and MODIS data is replaced by the SSS obtained from the nearshore supplementary statistical model,and the basic process of the dual-model combined inversion is constructed.After comparing with the Copernicus Ocean reanalysis data,the correlation coefficient between the two can reach 0.764,and the average deviation is-0.86psu,which solves the problem of poor inversion accuracy of the machine learning model in the low salinity area.(4)Using the combined inversion model,the temporal and spatial variation of sea surface salinity was analyzed in the Sea Area Adjacent to the Changjiang Estuary.The seasonal differences of sea surface salinity in the study area are in good consistency with the movement differences of the circulation;the interannual variation of the average SSS in the study area from March and August from 2003 to 2020 showed a downward trend as a whole.Compared with the flow of Datong Station,the correlation coefficient between the average SSS and monthly runoff in August is-0.755,and it is only-0.340 in March.The results of spatial variation analysis show that the area of the sea area where the sea surface salinity tends to decrease in both the flood season and the dry season is more than70%,and some sea areas where the Taiwan Warm Current is located have a significant downward trend,which indicates that the sea surface salinity of the Taiwan Warm Current may have decreased in recent years.The waters near Jeju Island showed a slight to significant downward trend in the dry season,but a slight upward trend in the flood season;the core area of Changjiang Diluted Water showed a significant downward trend in the flood season,which indicates that the diffusion range of Changjiang Diluted Water may appear in recent years.expansion phenomenon.During the summer of 2003-2020,the expansion patterns of the Changjiang dilute water can be divided into the northeast expansion type,the northeast expansion type,the north east expansion type,the multidirectional expansion type,the northeast-southeast expansion type and the uniform expansion type.The first three types are more common.The latter three types are more common in special years with abnormal salinity distribution.