Significance testing for variational assimilation
Julia C. Muccino, Norma F. Hubele, Andrew F. Bennett
Abstract
Julia C. Muccino, Norma F. Hubele, Andrew F. Bennett
Abstract
Abstract The hypothesis test associated with a variational data‐assimilation algorithm is examined in detail. It can be shown that the test statistic, Ĵ, should have a$\chi^{2}_{M}$ distribution, whereMis the number of scalar data assimilated,ifthe hypothesis regarding the model and data residuals is consistent with the true physical system. The skill of the Kolmogorov–Smirnov (KS) test in identifying whether the hypothesis used in the assimilation is ‘true’ (that is, the hypothesis correctly reflects the residuals in the model and the data) or not (that is, the hypothesis has either incorrect parameters in the covariance or an incorrect bias) is evaluated. It is shown that the KS test does indeed have some skill in identifying a false hypothesis. However, as one might expect, the skill is limited to cases where the particular component of the hypothesis that is incorrect has an impact on the test statistic and therefore on its distribution. Copyright © 2004 Royal Meteorological Society.
OpenAlex reports 12 citations for this work. Citation counts describe recorded attention and do not establish research quality.
A contribution statement is not available in the OpenAlex record.
Method details are not available in the OpenAlex metadata.
Findings are not separately available in the OpenAlex metadata.
Limitations are not available in the OpenAlex metadata.
Application details are not available in the OpenAlex metadata.
Abstract The hypothesis test associated with a variational data‐assimilation algorithm is examined in detail. It can be shown that the test statistic, Ĵ, should have a$\chi^{2}_{M}$ distribution, whereMis the number of scalar data assimilated,ifthe hypothesis regarding the model and data residuals is consistent with the true physical system. The skill of the Kolmogorov–Smirnov (KS) test in identifying whether the hypothesis used in the assimilation is ‘true’ (that is, the hypothesis correctly reflects the residuals in the model and the data) or not (that is, the hypothesis has either incorrect parameters in the covariance or an incorrect bias) is evaluated. It is shown that the KS test does indeed have some skill in identifying a false hypothesis. However, as one might expect, the skill is limited to cases where the particular component of the hypothesis that is incorrect has an impact on the test statistic and therefore on its distribution. Copyright © 2004 Royal Meteorological Society.
Key concepts: Test statistic, Statistical hypothesis testing, Statistic, Covariance, Statistics, Mathematics, Alternative hypothesis, Scalar (mathematics)