The interannual variability as a test ground for regional climate simulations over Japan

The validation of regional climate models is usually based on the intercomp arison of the model's mean climate with the observed climatology. Albeit a prerequisite for the use of the model in a predictive mode, a successful va lidation of this type does not strictly test the model's ability to simulat e anomalous conditions as might be associated with anthropogenic climate ch ange. Here, we explore an alternate strategy, whereby the model's ability t o reproduce the observed interannual variability is tested. The model utili zed is an operational numerical weather prediction model of the German Weat her Service, and it is tested for its use over East Asia and Japan in a ser ies of 5 month-long January simulations. The model is used in a domain of 5 100 x 5100 km(2), has a horizontal resolution of 56 km, and 20 levels in th e vertical. It is driven at its boundaries by the European Center for Mediu m-Range Weather Forecast (ECMWF) analysis. In validating the integrations, particular emphasis is put on the precipita tion fields. For validation we use three different observational data sets: a terrestrial analysis from rain gauges, including the Automated Meteorolo gical Data Acquisition System (AMeDAS) data of the Japan Meteorological age ncy, the gridded data set of the Global Precipitation Climatology Project ( GPCP), which over sea is largely based upon satellite information, and the ECMWF Re-Analysis (ERA) data set, which is produced by a model in an assimi lation mode. It is demonstrated that the synoptic-scale evolution of individual low-pres sure systems within the modeling domain is deterministically controlled by the lateral boundary conditions. Precipitation - spatially averaged over se lected subdomains - compares remarkably well with the observations, both in terms of the monthly amounts and of the temporal evolution throughout the integration period. Using the strategy of a previous study, we analyze the year-to-year variations of the model results, both for the dynamical and pr ecipitation fields. It is found that the modeling error is substantially sm aller than the typical year-to-year fluctuations of the interannual variabi lity. Implications of this result, concerning the model's use as a tool for down-scaling climate change, are also discussed.