P65 MPAS-A sensitivity to floating-point precision, mesh configuration, and interpolation scheme for weather forecasting in western Canada.
Chui, Timothy C. Y., and Roland Stull, University of British Columbia, Canada
Sensitivity
studies for the Atmospheric core of the Model for Prediction Across Scales (MPAS-A)
are conducted, to determine its potential as a dynamical core for medium-range
weather forecasting over the complex terrain of western Canada. Sensitivity to
the choice of floating-point precision for computation and mesh refinement
configuration is initially explored using a baroclinic jet simulation.
Baroclinic instability is found to be affected more by mesh quality and
resolution than the choice of floating-point precision.
Model performance is further explored using three case studies over British
Columbia: a summer high-pressure system; a fall low-pressure system; and a
record-breaking early-winter wind storm. Forecasts are verified against surface
observations and Global Data Assimilation System reanalyses, and the results
are compared against control simulations produced by Version 4.0.3 of the
Weather Research and Forecasting (WRF) model.
The MPAS-A forecasts with variable-resolution meshes of a given fine-region
resolution perform similarly to the corresponding WRF nest with a similar grid
resolution. MPAS-A deterministic forecast errors are found to be most sensitive
to the choice of mesh refinement configuration, and less on the choice of
interpolation scheme or floating-point precision. Single-precision forecasts
offer a 40% reduction in wall time compared to double-precision forecasts on
average.