Non-Gaussian Based Buddy Check and Gross Error Check Observational Quality Control Measures and Their Impacts on Non-Gaussian Based Data Assimilation Systems.
Wednesday, 17 December 2014
With the recent derivation of a mixed distribution based incremental VAR data assimilation system, the need to adapt the current Gaussian based quality control measures to allow for the correct assessment of lognormal distribution based observational errors has to be addressed. In this paper we take the “buddy check” system and show that it is possible to derive a similar measure in terms of a linearization with respect to ln x and not x. The gross error check is also discussed in the framework of which statistic, mean, mode of median, is the measure to be “centered” about, given the skewness of lognormal distribution. The final part of this paper is concerned with comparing the impact of using the current Gaussian and the new lognormal based distribution with an incremental mixed distribution based 4DVAR system with the Lorenz 1963 model.