Sea Surface Temperature Reconstruction For FY-3C Satellite Data

This paper focus on sea surface temperature?SST?reconstruction by combining with the remote sensing retrieval SST from Fengyun-3C?FY-3C?satellite of China and the in-situ SST from the iQuam system.After the quality control to these data and the bias correction for the remote sensing retrieval SST,the SST quality should be improved and the original SST field is built,then the no-gap reconstructed SST can be estimated by using the objective analysis methods.According to this principle,the SST products from the Visible and Infrared Radiometer?VIRR?and Microwave Radiation Imager?MWRI?aboard the Fengyun-3C?FY-3C?satellite are evaluated against the measurements of in-situ SST,then the outliers of SSTs are rejected and the biases in satellite-retrieved SSTs are removed,and the optimum interpolation?OI?and Kalman filtering method are used to reconstructing these SST data.Moreover,a radius basis function network?RBFN?model is made to improve the quality of reconstructed SST that estimated with limited SST samples.Firstly,the SST quality of VIRR and MWRI in various latitudes and periods are evaluated against the measurements of in-situ SST.According to the SST bias performance of each product for different periods and latitudes,it is clear that the biases from both the daytime VIRR SST?VIRRD?and nighttime VIRR SST?VIRRN?are the largest at low latitudes and are relatively lower at mid-high latitudes and that negative SST biases in the VIRR increase with varying time.Statistical results for 2015 reveal a mean bias ± standard deviation of error?STDE?for daytime MWRI SST?MWRID?and nighttime MWRI SST?MWRIN?of 0.6044 ± 3.9064°C and 0.7653 ± 3.7307°C,respectively.And large abnormal SST biases from MWRI products occur in Time III?the 211 th to 319 th day of 2015?,and the average values of the SST biases are 3.6010°C and 3.7822°C for the MWRID and MWRIN,respectively.Furthermore,our error analysis of each SST product confirms that the errors from SST retrieval algorithm result in a difference in the SST product quality at various periods and latitudes and that the brightness temperatures?BTs?of sensors have a significant effect on the accuracy of SST products,especially for MWRI products,which present large errors in terms of BTs and cannot meet the requirements of application.Secondly,a bias correction method based on the piecewise regression algorithm for VIRR SST products of 2015 is proposed in order to correct the biases in satellite-retrieved SSTs.Compared with the probability density function?PDF?matching technique for bias correction,the proposed method not only considers the related variables of SST,but also choose the optimal matchup samples that close to the variables for regression,such as the climatology SST and the view zenith angle.And the optimal local matchup data are selected for each SST observations based on the distance to each part of regressors.Using the proposed method for bias correction,the average biases and SDs are decreased from-0.5877? and 1.3544? to 0.0173? and 0.4965? for VIRRD,and from-0.4801? and 1.2840? to-0.0177? and 0.4920? for VIRRN,respectively.And the quality of VIRR SSTs in both time and space are improved significantly after bias correction.In addition,after the procedures of quality control and bias correction,we separately used the optimum Interpolation?OI?method and Kalman filtering method to analyze these different sources of SST data,and the rectangular correlation scales and oriented elliptic correlation scales are selected and calculated to make the effective associated regions for SST analysis.The statistical results from OI analysis indicate that the quality of reconstructed SSTs using the oriented elliptic correlation scales?OIellipse?are much better than those using the rectangular correlation scales?OIrect?,and the average RMSEs of 2015 for OIrect and OIellipse are 0.4194? and 0.3816? respectively,which are much smaller than 0.4775? of originate SST data.While the reconstructed SST from Kalman filtering method using the rectangular correlation scales(Kalman_rect)and the oriented elliptic correlation scales?Kalmanellipse?are much better than those from OI analysis.The statistical results show that the average RMSE of Kalman_rect and Kalmanellipse are separately 0.2921? and 0.2869?,which are quite close to the value of 0.2780? from the OISST products.Thus the Kalman filtering method that using the dynamically estimated errors for observations and background fields can obtain a high accuracy of SST than the OI method that only use the fixed data to guess error ratios for SST observation and background field.Finally,since both the OI method and Kalman filtering method are not effective when the samples are limited,a radial basis function network?RBFN?model is built to reconstruct daily SSTs.The statistical results show that the RBFN method has a better performance of reconstructing SST than the OI method in the study.Additionally,the RBFN methods with different basis functions and clustering algorithms are tested.Compared with the other clustering algorithms as K-means algorithm and Kohonen-map algorithm,the improved nearest neighbor cluster?INNC?algorithm can obtain more hidden knots from the SST samples,which will contribute to acquiring high quality of reconstructed SST.And using the multi-quadric function as basis function is more effective than the Gaussian function to avoid the SST increment anomaly on the regions of hidden knots.Thus the INNC algorithm with multi-quadric function is quite suitable for the RBFN method to reconstruct SSTs in the study.