Using mountain flows to evaluate GRID technology for meteorological models of varying complexity

In project MeteoAG, a cooperation between the Distributed and Parallel Systems Group (DPS) and the Institute for Meteorology and Geophysics (IMGI) in Innsbruck, a GRID workflow was established using the highly complex, numerical meteorological model RAMS. GRIDs are a relatively new technology providing high performance computing on many different, spatially distributed computer facilities.

The MeteoAG workflow is used to to build a very high resolution heavy precipitation reanalysis database over the mountainous western part of Austria, utilizing five nested grids to zoom down to the Arlberg Region with the highest being 500~m in the smallest domain. Data from ECMWF is used to initialize the simulations. Based on experiences from this first phase of the AustrianGrid project, a less complex model is used in the second phase of AustrianGrid. We utilize the linear model of Smith and Barstad (2004) to investigate orographic precipitation and probability forecasts.

As GRIDs are a heterogeneous computing environment, tests need to assure that model results are independent of the system they have have been computed on. Using RAMS with an idealized mountain flow simulation reveals differences up to 15-20 percent in the pressure perturbation of a linear hydrostatic mountain test case. This is only dependent on the utilized computing system as the input and setup is kept exactly the same. To investigate whether this is a general problem with different hardware or code specific, a similar test with the linear model was performed. With this much simpler code no differences were found suggesting that the differences in RAMS results are not a consequence of the usage of different hardware but are instead code-dependent.