Micro-meteorological simulations with high-resolution near the ground are of great interest to study pollutant dispersion phenomena and to get accurate information on the lower atmosphere dynamic and thermal fields necessary for predicting long-range noise propagation (typically between ground and 50 meters high). The present study whose aim is to estimate the terrain-influenced meteorology is in line with the second topic. The small-scale complex terrain of St-Berthevin (Mayenne, France) - a 80 meters deep valley- has been chosen for this numerical investigation partly due to the availability of wind and acoustic fields measurements recorded during a two weeks experimental campaign.

Accounting for the influence of large-scale meteorological fields on smaller ones requires the extension of the computational domain quite far from the area of interest that must be accurately described by the grid. Using a fine mesh over the whole domain is presently not feasible due to the computer capacity. To overcome this problem, a now currently used method is the grid-nesting approach. The finite-difference non-hydrostatic compressible 3D atmospheric model SUBMESO and a technical module (Debreu & Blayo, 1999) able to manage the Adaptive Mesh-Refinement (AMR) method developed by Berger & Oliger (1984) have been coupled together. The grid-nesting module was developed to be adaptable to any finite-difference oceanographic or atmospheric model, with a minimum of changes in the single-grid model. In order to focus on an a priori defined limited-range area, the adaptive aspect of the method has been left aside while a particular effort was concerned with the improvement of the grid-nesting boundary conditions which were initially restricted to the interpolation of coarse-grid fields at the fine grid boundary. Among all the boundary conditions suggested in the literature, the most effective in avoiding the reflection of numerical waves at an outflow boundary seems to be the radiative condition proposed by Carpenter (1982) applied to all the prognostic variables except the pressure for which Dirichlet conditions is less disturbing. The inflow boundaries are treated with the simple Dirichlet conditions. The method was first validated against the case of a stably stratified atmospheric flow over a 2D 1-m-high hill that provides a direct comparison with analytic solution, and against the 3D case of neutral flow over flat terrain where streaky structures do appear.

This method is applied to large-eddy simulations of the atmospheric flow above the real topography of St-Berthevin crossed from northwest to southeast by the valley of the river Vicoin. In order to estimate the accuracy of numerical results at several locations in the valley where instrumented masts were installed, three flow conditions were chosen among the experimental data obtained upstream from the region of interest : a stably stratified flow blowing from the west with a weak intensity, and two convective flows with either a north-westerly high wind or a south-westerly moderate wind. The numerical results clearly demonstrate that channelling effects in the valley, accelerations over the bumps and decelerations in the holes depend on the atmospheric stratification and on the upstream wind direction.