Research On The Thinning Of Ocean Observation Data In Data Assimilation

With the advent of satellite remote sensing observation technology,it is necessary to study how to better apply high spatial and temporal resolution satellite remote sensing observation data to oceanographic research.Satellite remote sensing observation data has a large amount of data and high spatial and temporal resolution.Due to the limitation of error correlation and calculation cost,data thinning must be needed.The purpose of this project is to improve the ability of the thinning algorithm to sparse satellite remote sensing observation dataset.In this paper,several classical thinning algorithms are studied and evaluated,and a more effective thinning algorithm and evaluation indexes are proposed to improve the characteristic resolution of the thinned dataset.Firstly,this paper briefly introduces the background and research status of the thinning algorithm for satellite remote sensing observation dataset.By analyzing and comparing the methods and characteristics of common satellite remote sensing observation,the basic dataset of this paper are selected.The redundancy of the dataset is analyzed,and the dataset preprocessing scheme is designed for the possible abnormal problems of the dataset.Secondly,this paper studies and simulates the satellite remote sensing sea surface temperature observation dataset by using three mainstream thinning algorithms: stepwise thinning algorithm,thinning algorithm based on estimation error analysis,and thinning algorithm based on clustering.Under the condition of controlling the observation numbers of the results,the three algorithms all reduce the redundancy of data.Among them,the first method is simple and fast,but the similarity is low.The second method has high similarity but low computational efficiency.The third method is the closest to the information of original dataset and retains the effective information more accurately.Thirdly,in order to study and evaluate the thinning methods in many aspects,this paper designs the evaluating indicator of the estimation error signal-to-noise ratio and the evaluating indicator of the cluster average gradient.Several algorithms are simulated and calculated at different thresholds and parameters respectively.The analysis results show that the evaluating indicators can effectively evaluate the thinned dataset.At the same time,it is found that the clustering based thinning method has the highest similarity,retains the most effective information and has the best overall performance.However,the feature resolution should be further improved.Finally,in order to solve the above problems,an improved clustering based thinning algorithm is designed.The measurement of gradient is added to the improved algorithm to recognize and retain detailed features.Through simulation and comparative analysis,the number of reserved observation points in the improved algorithm results is slightly increased,but the data redundancy is basically unchanged compared with the results before the improvement.The effective information retained is increased,and the data similarity and feature resolution are both improved.Compared with the original algorithm,the improved algorithm has better thinning effect.