An Ensemble Kalman Filter Localization Method

Abstract

The invention belongs to the field of measured marine environment data assimilation, and in particular relates to a localization method of an ensemble Kalman filter. The invention includes: preprocessing the measured data of the atmospheric and oceanic environment; calculating the prior observation set members at the observation data for each observation data; calculating the prior set average and variance of the observation; calculating the observation increment of the set average; Observation increment of each ensemble disturbance; project the ensemble average observation increment onto the ensemble average of the mode state; project the observation increment of each ensemble disturbance onto the corresponding ensemble disturbance of the mode state; obtain the ensemble member analysis field. The invention improves the traditional localization method in the ensemble Kalman filter, effectively considers the different spatial scales represented by the ensemble average and the ensemble disturbance, and significantly improves the assimilation precision of the ensemble Kalman filter.

Description

technical field [0001] The invention belongs to the field of measured marine environment data assimilation, and in particular relates to a localization method of an ensemble Kalman filter. Background technique [0002] Ensemble Kalman filtering and four-dimensional variation are two types of advanced data assimilation methods recognized internationally, and they have their own advantages and disadvantages. The biggest advantage of the ensemble Kalman filter over the variational method is that it simulates the prior probability density distribution function of the model state variables through ensemble sampling, and the background error covariance matrix calculated according to the ensemble samples contains the dynamic information of the model, so it is flow dependent. Due to the limitation of computer hardware resources, only a small set of samples (10 2 order of magnitude). For the actual ocean numerical model, the dimension of the state variable is 10 7 , so fewer ense...

AI Technical Summary

Problems solved by technology

Due to the significant difference in the spatial scales of ensemble averaging and ensemble perturbation, traditional localization schemes have obvious limitations

Images