Abstract
In ensemble data assimilation systems, the impracticalities of full sampling and systematic error often lead to spurious correlations between two variables with low actual correlations. To solve these problems, researchers have previously proposed a covariance localization (CL) method, which mainly involves the Schur product between a state error covariance matrix and a distance-based correlation matrix. Although this CL method can reduce spurious correlations to a certain extent, observational data remain difficult to be used effectively, which results in unreasonable assimilation. In this study, we develop a new CL method coupled with a fuzzy logic control algorithm, which we call the covariance fuzzy (CF) method. The proposed CF method is a distance-based localization method with “fuzzy” vanishing correlations in data assimilation (DA) systems. To verify the effectiveness of the new algorithm, we conducted a set of experiments using an ensemble Kalman filter (EnKF) that combines the nonlinear Lorenz-96 model or the quasi-geostrophic (QG) models. First, the performances of the CL and CF methods are discussed with respect to different strength forcings, ensemble sizes, and covariance inflation factors. The experimental results show that the proposed CF method can obtain a more effective observation weight than the CL method and can reduce the errors caused by spurious correlations. Additionally, using power spectral density (PSD) as a performance evaluation index, the robustness of the proposed fuzzy logic localization method is demonstrated. However, the application of the fuzzy logic-based localization methodology to a real atmospheric model remains to be tested.
Highlights
Data assimilation (DA) engages a set of specialized techniques that combine geoscience model forecasts with available observational data to determine the best estimates of the corresponding geoscientific state
To eliminate the spurious correlation phenomenon caused by sampling errors when estimating the background error covariance under a small ensemble number condition, we propose a new localization method coupled with fuzzy logic in the ensemble Kalman filter (EnKF), namely, the covariance fuzzy (CF) method
The similarity of the results showed that the fuzzy logic assumption can efficiently deal with spurious correlations, confirming the findings of Bai et al [26], and that the fuzzy control methodology was an efficient strategy for improving estimations of the error covariance matrix
Summary
Data assimilation (DA) engages a set of specialized techniques that combine geoscience model forecasts with available observational data to determine the best estimates of the corresponding geoscientific state. On the basis of ensemble transform Kalman filters (ETKFs), Bai et al [26] developed a fuzzy control function that performed a series of fuzzy inferences on the input distance variables to produce a more accurate observed weight coefficient than that of smoother covariance localization functions such as the GC function [18,27]. These experiments were only conducted with toy models such as the Lorenz-96 model, and the proposed methodology needs to be tested with more models.
Sign-in/Register to view highlights & summary 
Published Version (
Free)
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call