Data assimilation with a 1000-member ensemble at 1-km horizontal resolution

JUAN RUIZ; GUO-YUAN LIEN; KEICHI KONDO; SHIGENORI OTSUKA; TAKEMASA MIYOSHI

Florianopolis

Workshop; 7th WMO Data Assimilation Symposium; 2017

WMO

In this work, a 1000 member ensemble data assimilation experiment isperformed using the SCALE-LETKF mesoscale data assimilation system at 1km horizontal resolution and assimilating radar observations from theOsaka University phased array radar every 30 seconds. The 13th July 2013heavy rain case over Kyoto is used to run the large ensemble experiment.Firstly, the large ensemble experiment is used to quantify the spatialdistribution and temporal evolution of non-Gaussianity in the ensemble.The mechanisms associated with non-Gaussian PDFs are also explored aswell as the dependency of non-Gaussianity with the length of theassimilation window. The large ensemble experiment revealed that strongdepartures from Gaussianity are strongly linked to fast instabilitieswhich can be detected and quantified using high resolution bred vectorsand that even a 30 second assimilation cycle is not fast enough to keepthe error distribution close to a Gaussian in convective areas.Secondly, the 1000 member ensemble assimilation experiment, is used tostudy the spatial structure of the background error covariance matrixand its dependence with the state variables. It was found that spatialdecorrelation scales are significantly shorter in convective areas andthat they are a function of height.