DATA SAMPLING STRATEGIES FOR GENERAL-CIRCULATION MODELS

For climate study purposes, complex general circulation models are often run for extended periods; the model state is sampled regularly and various diagnostics calculated. A convenient and often used sampling period is 1 day, but in this case diurnal and semi-diurnal variations, for example those associated with the atmospheric tides, will alias into the time mean. Using the UK Universities Global Atmospheric Modelling Project model, adapted from the ECMWF forecast model, an attempt is made to quantify this aliasing in various diagnostics. A variety of data sampling/processing strategies are investigated, for example time-averaging the model fields before writing them out, in an attempt to eliminate the aliasing problem. It is found that sampling the model state every 3/4 day or every 5/4 day is a simple and effective solution.