Retrieval Study On Sea Surface Salinity For Microwave Imager Combined Active/passive

Ocean salinity relates the Earth global water cycle to ocean circulation.Thus,the Chinese Ocean Salinity Satellite Planning is proposed in 2015.The microwave imager combined active/passive(MICAP)is a candidate payload for Chinese Ocean Salinity Satellite,atming at obtaining global sea surface salinity(SSS).Refer to the technology experience of the ESA's SMOS and NASA/CONAE's Aquarius/ASC-D,MICAP is a combined L/C/K band one-dimensional microwave interferometric radiometer and L-band digital beam forming scatterometer for the first time,and has the ability of multi-bands and active/passive detection.Considering the different configurations of instruments,the existing SSS retrieval algorithms can not be used for MICAP.This thesis mainly focuses on the SSS remote sensing algorithm development and validation for the application of the future Chinese Ocean Salinity Satellite.Firstly,based on the microwave radiative transfer equation and geophysical model function(GMF),the brightness temperature(TB)models of rough sea surface and backscatter coefficient models for MICAP are built.The dielectric constant models and its impact on the TB of flat sea surface are compared.The L/C/K bands TB models of rough sea surface for MICAP are built based on the selected dielectric constant models.The L band GMF is derived to calculate the backscatter coefficient between 30° to 55° incident angle for MICAP by combining extrapolation method and L band GMF of PALSAR.In addition,the effect of spectrum models and foam models to sea surface TB are analyzed respectively,which provide references for the selection of wave spectrum and foam models.Secondly,the sea surface salinity retrieval algorithms without rain for MICAP are studied on the basis of the forward models.Based on Monte Carlo simulation method,the SSS,SST,and WS retrieval performance of MICAP is estimated using the combined active/passive salinity retrieval algorithm.The results show that the root mean square errors(RMSE)on retrieved SSS,SST,and WS are about 0.6 psu,1.2 °C,and 0.8 m/s during only one satellite pass for the low/mid latitude regions respectively.The RMSE of monthly average SSS is below 0.13 psu over 200 km × 200 km and 30 days for most the low/mid latitude regions when the uncertainties of the sensor calibration are assumed to be equal to the sensitivities of radiometers.In addition,the retrieval accuracies of SSS,SST,and WS with different instrument configurations for MICAP are also evaluated.It shows that without the 23.8 GHz channel,the retrieved SSS,SST,and WS RMSE almost do not increase compared with default configuration.This study result provides a reference for the final band configurations of MICAP.Thirdly,the performance of combined active/passive salinity retrieval algorithm is evaluated using the Aquarius and AMSR2 measurements.The regression is used to correct the systematic bias between measured TB and modeled TB using the selected models.The results show that the RMSE on retrieved SSS,SST,WS and cloud liquid water are about 0.63 psu,0.73 °C,0.90 m/s,and 0.038 mm during only one satellite pass by comparing with the interpolated monthly Scripps Institution of Oceanography Argo salinity and SST,WS and cloud liquid water products from the RSS respectively.This indicates that the effectiveness of the proposed algorithm is validated.Meanwhile,this also illustrated that the SSS,SST,WS and cloud liquid water can be retrieved simultaneously using the proposed algorithm in this paper.Finally,SSS and WS retrieval methods are studied by using four machine learning methods,including the Deep Neural Network(DNN),Gaussian Process Regression,Support Vector Machine Regression,and Kernel Ridge Regression.Based on Aquarius measurements in the South China Sea and ancillary data,the SSS and WS are retrieved with trained models to validate the feasibility of machine learning methods,and the retrieval accuracies of SSS and WS are compared for the above four methods.Then,the SSS retrieval accuracies are futher compared between DNN method and retrieval algotihms.The results show that compared with HYCOM SSS,the bias and RMSE on retrieved SSS of the former are better than the latter.Compared with Scripps SSS,the results of the former are still better than the latter except the retrieval bias.Compared with Argo SSS,the bias on the retrieved SSS of the former is better than the latter.The SSS RMSE of the former is better than the CAP algorithm.However,the DNN method strongly depend on the quality and number of reference samples.Thus,the retrieval accuracy of DNN is not ideal for small sample.