Avaliação de alterações no desvio padrão do erro de background em um sistema de assimilação de dados oceânicos 4D-variacional

Abstract

The standard deviation of a long run without data assimilation is commonly used to obtain the background error covariance matrix in 4D-variational data assimilation modeling systems. Proper determination of this matrix has a fundamental importance because it determines how observational information will be propagated in the model. The background-error standard deviation (Ʃ) is to ensure that the control variable corrections do not exceed the magnitude of the typical variability. The purpose of this dissertation is to update the Ʃ matrix from the prior solution of an assimilative simulation, assuming that it is a more adequate representation of the background field. A three-month experiment with the modified standard deviation was compared with a simulation in the standard configuration (i.e., the standard deviation calculated from a free run). The evaluation was based on the comparison of the simulation results with dependent and independent observations. The lowest values for the error metrics considered were generally observed for the simulation with the new approach. Differences between simulations were significant in reducing errors in relation to satellite data of sea surface temperature and height, as well as the surface velocity component of ocean drifters.