Sensitivity of atmospheric transport model performance to numerical advection schemes and resolution

We have tested four newly developed 3D advection schemes named Mol-rg, Split-u, Split-us and Split-rg. We compared these schemes with Slopes and Second Moment. Mol-rg and Split-rg make use of a reduced grid, a grid with less cells near the poles, to overcome the well known pole-singularity. Two tests were performed with all schemes: the solid-body rotation test and a radon test. The radon test uses 3D meteorological input for the month January 1992 from a numerical weather prediction model, together with parametrizations for sub-grid scale vertical transport. We compared model results with measurements on three islands in the Indian Ocean. From the solid-body rotation test we learn that all new schemes do not generate undershoot and overshoot and are mass conservative. Split-us and Split-rg are very cheap in terms of CPU time and memory requirements and give accurate results. The model results for the radon test give good predictions for the background concentration. However, the correlation between measured and simulated radon concentration peaks is poor for the simulated period. The model results are found to be almost independent of the numerical scheme used, but depend mostly on the resolution and the quality of meteorological input. Therefore it is important to use cheap advection schemes such as Split-rg to be able to perform model calculations on high resolutions. The new advection schemes are available through Internet.