Impact of the fine-scale initialization on mesoscale simulated precipitation over mountainous areas

On the 8th and 9th of September 2002, a quasi stationary convective system leaded to damaging flash floods in the Cévennes-Vivarais region, in the South-East of France. This case is one of the cases studied in the European project INTERREG IIIB - Hydroptimet which aims to optimize hydrometeorological forecast tools for flash flood warning. A control simulation is performed, with the non hydrostatic mesoscale model MESO-NH, (Lafore et al, 1998) on this case, using as initial conditions, the French operational analysis. The maximal cumulated simulated rainfall do not exceed 341 mm in 24 hours, whereas during the same period, the maximal observed cumulated surface rainfall had been recorded at 690 mm. Moreover, the simulated convective system is shifted northward, over the Massif Central crests, whereas the most active convection was observed mainly over the upwind lower mountainous areas.

Several sensitivity simulations on different initialization methods (Ducrocq et al, 2000) have been performed to improve the control simulation. The first method uses a mesoscale analysis of surface observations. The second one uses the same mesoscale surface observations analysis and adds a moisture and microphysics adjustment from radar reflectivities and IR METEOSAT brightness temperature.

The first initialization method leads to more realistic simulated precipitation fields. The heaviest precipitation is now located over the upstream lower mountainous areas. However, the maximal quantitative precipitation forecast is a bit weaker than in the control simulation. A detailed study performed on the mesoscale simulated fields shows that the location of the maximum observed precipitation is directly linked to the low-level mesoscale patterns even if the convective system is forced by synoptic conditions.

Also, the second initialization method allows simulating a convective system over the upstream lower mountainous areas. Moreover, with this initialization method, the quantitative precipitation forecast is stronger than with the control simulation.

Then, to evaluate the contribution of these methods on the quantitative precipitation forecast for hydrological purposes, a hydrological criterion has been used. The simulated rainfall fields force a hydrological model from the TOPMODEL family. This model has been run on several mountainous subcatchments of four of the main catchments in the region, ranging from 200 km² to more than 2 000 km².

The hydrological simulations highlight better convection localization in simulated rainfall fields from both initialization methods. The total simulated flow from control simulation rainfall fields on northern (/southern) catchments are weaker (/stronger) than those simulated from control simulation rainfall fields. Also, with both initialization methods, flash-flood time evolution is better reproduced on all the catchments.

In conclusion, we see that mesoscale analysis and information from radar and IR satellite data can help to improve significantly the quantitative precipitation forecast in mesoscale simulations.