Simulation of a convective rainfall event and associated water budget over West Africa : an intercomparison of mesoscale models

Simulation of rainfall over West Africa with large-scale models suffers from major weaknesses, in both numerical weather prediction (NWP) systems (Agustí-Panareda et al. QJRMS 2009) and climate general circulation models (Hourdin et al. BAMS 2010). Indeed, rainfall is actually provided, for the most part, by mesoscale convective systems (MCSs), which are still poorly handled by large-scale models, while at the same time, they are responsible for devastating floods every year.

From this perspective, mesoscale models could help improving the simulation of rainfall. This raises other issues though, as discussed for instance by Moncrieff and Liu (2006) for simulations of MCSs over North America using horizontal resolutions on the order of 10-30 kilometers, thus allowing MCSs to be parly captured. In fact, it is currently unknown how accurately mesoscale models depict rainfall and the associated water budget over West Africa, and how they compare to each other in the simulation of individual MCSs. The present study investigate this issue with results from an intercomparison of mesoscale models conducted with the African monsoon multidisciplinary analysis (AMMA) program (see Guichard et al. WAF 2010).

In practice, an evaluation of precipitation (P) and evapotranspiration (E) simulated by mesoscale models is carried out with high-resolution satellite rainfall products (EPSAT-SG, TRMM and CPC-RFE2) and outputs from offline land surface models (LSM) (see Boone et al. BAMS 2009). Six mesoscale models (BOLAM, COSMO - formerly known as the Lokal Modell, MesoNH, PROMES, MOUM and WRF) performed simulations of a MCS observed to cross part of West Africa during the core of the monsoon season, the 28 August 2005. This case is characterized by strong synoptic activity, in the form of African easterly waves. Results from the ECMWF integrated forecast system are also considered because the ECMWF analysis was also used to prescribe initial and boundary conditions for all mesoscale models.

Indeed, initial and boundary conditions are found to significantly control the location of rainfall at synoptic scale as simulated with either mesoscale or global models (ARPEGE, ECMWF, and UM). When initialized and forced at their boundaries by the same analysis, all models forecast a propagating rainfall structure, as observed by satellite products. However, rainfall is also forecast at other locations where none was observed, and the nighttime northward propagation of rainfall is not well reproduced. There is a wide spread in rainfall rates across simulations, but also among satellite products. (This indicates the need of more accurate rainfall products for a proper evaluation of simulated rainfall rates at the mesoscale.)

The range of simulated meridional fluctuations of evapotranspiration (E) appears reasonable, but E displays an overly strong zonal symmetry. Offline land surface modeling (ALMIP, Boone et al. BAMS 2009) and surface energy budget considerations show that errors in simulated E are not simply related to errors in surface evaporative fraction, and involve the significant impact of cloud cover on the incoming surface shortwave flux. The use of a higher horizontal resolution (a few km) enhances the variability of precipitation, evapotranspiration and precipitable water (PW) at the mesoscale. It also leads to a weakening of daytime precipitation, less evapotranspiration and smaller PW amounts.

Thus, the simulated MCS propagates further northwards and somewhat faster within an overall drier atmosphere when the resolution is increased and the convective parametrization is switched off. These changes are associated with a strengthening of the links between PW and precipitation.