Numerical errors above steep topography: A model intercomparison

This paper compares the behaviour of three mesoscale models (MM5, Penn State University/NCAR, LM, German Weather Service, and KAMM2, IMK Karlsruhe) in the vicinity of a steep isolated mountain under zero-wind conditions. The setup of the simulations is such that the model atmosphere would remain at rest for all times in the absence of numerical errors. Our results show that the errors occurring in the models under these idealized conditions differ strongly between the models. In the MM5 and the LM, the isentropes become heavily disturbed above the mountain after a few hours of integration, and peak vertical motions of several m s(-1) are encountered. In KAMM2, however, isentropes remain virtually undisturbed and vertical motions do not exceed 5 cm s(-1) except close to the upper boundary. Sensitivity tests indicate that the numerical errors in the MM5 disappear almost completely when the horizontal diffusion of temperature is computed truly horizontally rather than along the sloping coordinate surfaces. Experiments with moist physics show that the numerical errors over steep mountains can induce substantial amounts of spurious rain. The largest values are found for the LM, followed by the original MM5. The KAMM2 and the modified MM5 version produce only negligible amounts of rain.