Theory And Application Research Of Oceanic Four-dimensional Variational Data Assimilation

The uncertainty of initial field has always been one of the main sources of forecast error in the numerical forecast model.Whether the initial field that meets the quality requirements can be obtained is one of the bottlenecks restricting the development of current ocean numerical forecast.Four-dimensional variational assimilation method,that provides an initial field for numerical prediction models,not only overcomes the linear limitation between observation variables and model variables,but also makes full use of past and present observations,and has some characteristics different from other data assimilation methods.But there also are a series of problems that need to be solved and studied.This paper combines the knowledge of atmospheric science,marine science and computational mathematics,and studies the construction of background field error covariance matrix,optimization algorithm and computational efficiency in four-dimensional variational assimilation systems.Then a four-dimensional variational assimilation system is established in the South China Sea,and some related experiments were carried out.In this paper,the background field error covariance matrix is decomposed into three parts for construction,that is,the standard deviation is counted by the climatic variance method,the spatial correlation matrix is simulated by the solution operator of the diffusion equation,and the relationship between the ocean elements is introduced as the balance operator.Then,through the single point assimilation experiment,it is verified that the background field error covariance matrix,constructed by present paper,can realize the propagation of observation information in three-dimensional space and multi-variate.As a four-dimensional variational assimilation method solving in the observation space,the Physical Space Analysis System can greatly improve the computational efficiency.However,present paper finds that during the iteration process,the cost function has artificial oscillation,this instability phenomenon is mainly caused by the non-monotonic reduction of the gradient norm of the cost function.By introducing Minimal Residual algorithm,the monotonic reduction of the gradient norm is controlled,which effectively improves the convergence property and assimilation effect of the Physical Space Analysis System during the iteration process.Based on the construction scheme of background field error covariance matrix,Physical Space Analysis System and Minimal Residual algorithm studied by this paper,and the Regional Ocean Model System,a four-dimensional variational assimilation system is established in the South China Sea.Then continuous assimilation experiments were carried out using oceanographic observation data,and the 2013 reanalysis field data of the South China Sea region was obtained.Finally,the comparison between the reanalysis data and the observation data shows that its have a high degree of space-time matching,which verifies the accuracy of the assimilation system.