Research On Sea Surface Wind Retrieval Method And Its Application Based On Sentinel-1 SAR

The sea surface wind plays a very important role in meteorological and marine forecasting,maritime action support and wind energy resource assessment.Compared with the traditional on-site direct observation method,the satellite remote sensing method has more advantages in the rapid acquisition of the global sea surface wind,and it has become one of the important means of sea surface wind detection.Among them,the spaceborne synthetic aperture radar(SAR)can obtain sub-kilometer spatial resolution sea surface wind,which greatly meets the demand for high-resolution sea surface wind in research and application.However,the following problems still exist in the current research on sea surface wind retrieval based on C-band SAR: the variational method of VV polarization sea surface wind retrieval needs to be improved;the retrieval performance and differences of different variational method schemes need to be systematically evaluated;the sea surface wind retrieval method suitable for Sentinel-1 IW cross-polarization data needs further exploration;the application of dual-polarization data in the typhoon sea surface wind retrieval needs further study.Focusing on the above problems,this paper carried out research on the sea surface wind retrieval method and its application based on Sentinel-1 SAR.The main research contents are as follows:(1)In view of the problems existing in the variational method of VV polarized sea surface wind retrieval,the quasi-Newton method is proposed to solve the cost function of the variational method,and the treatment of local optimization and saddle point is introduced according to the characteristics of the cost function.Through simulation and case tests,and compared with CMOD direct retrieval method,optimal interpolation method and damped Newton variational method,the results show that the quasi-Newton variational method(QNVAR)has relatively higher wind speed retrieval accuracy.The QNVAR retrieved wind speed RMSE is 1.24m/s and MBE is-0.11m/s,which is significantly improved compared with the background wind speed(RMSE is 1.86m/s and MBE is-0.43m/s).Compared with the damped Newton variational method,the QNVAR calculation efficiency is significantly improved,and the calculation time is reduced by more than 70%.(2)Based on Sentinel-1 VV polarization data and NDBC buoy data,the errors of different variational method schemes composed of two different observation error values and three different geophysical model functions in the application of SAR wind retrieval are systematically evaluated and analyzed.For Sentinel-1 VV polarization data,when the observation error is constant 0.1d B and the geophysical model function is CMOD7,a relatively higher wind retrieval accuracy can be obtained.The retrieved wind speed RMSE is 1.11m/s and MBE is-0.12m/s;The wind speed RMSE retrieved by variational method is significantly lower than that of the CMOD direct retrieval method,and the maximum decrease is about 0.2m/s(CMOD5.N).(3)Based on the Sentinel-1 IW VH polarization data,a new cross-polarization wind speed retrieval model based on subswath and wind speed branch(C2WSM-S2)was constructed.The RMSE of the wind speed retrieved by the C2WSM-S2 model is1.86m/s,and the MBE is 0.47m/s.The retrieval accuracy is significantly better than that of other existing models;the retrieved wind speed error of the C2WSM-S2 model decreases significantly as the wind speed increases.The retrieved wind speed RMSE in the high wind speed section of the some subswaths is as low as about 1m/s,indicating that the cross-polarization data is more suitable for wind speed retrieval under relatively high wind speed conditions.(4)Based on Sentinel-1 IW VH polarization data,the influence of wind direction on C-band cross-polarization backscattering is discussed.The experimental results show that wind direction will affect the cross-polarization backscattering.In subswaths 2 and subswaths 3,the influence of wind direction is small at low wind speeds;when the wind speed is 8 ± 1m/s,the influence of wind direction is the strongest,and the upwind-crosswind difference in cross-polarization backscattering is about 2d B;with the increase of wind speed,the influence of wind direction will weaken.(5)Considering the complexity and uncertainty of the influence of wind direction on C-band cross-polarization backscattering,a machine learning algorithm named random forest is proposed for wind speed retrieval.The noise-removed VH polarization backscattering,relative wind direction,radar incidence angle and noise equivalent sigma zero are selected as input characteristics to train and obtain the random forest wind speed retrieval model.The RMSE of retrieved wind speed by the random forest model is 1.70m/s and the MBE is 0.07m/s,and the overall retrieval accuracy is better than the C2WSM-S2 model.However,in the high wind speed section,the retrieved wind speed error of random forest model is significantly higher than the C2WSM-S2 model.At the same time,the evaluation results of the importance of the random forest model on the input characteristic further confirm the conclusion that the wind direction will affect the cross-polarization backscattering.(6)The application performance of dual-polarization data in typhoon sea surface wind retrieval is systematically analyzed.Combined with infrared cloud images,the different characteristics of SAR co-polarization and cross-polarization typhoon images are compared and analyzed,which intuitively shows the influence of precipitation on different polarization backscattering and the possible differences in sea surface wind retrieval.Different typhoon sea surface winds are obtained by using co-polarization,cross-polarization and dual-polarization SAR data,and their differences are systematically compared.Considering the accuracy and spatial continuity of retrieved wind speed,the dual-polarization sea surface wind retrieval variational method is more suitable for typhoon sea surface wind retrieval,and its existing shortcomings can be improved by adjusting the weight coefficient of cross-polarization term.Using stepped frequency microwave radiometer wind speed and precipitation data,the influence of precipitation on C-band SAR backscattering is analyzed,which confirms and supplements the previous research conclusions.