Storm-Scale Numerical Rainfall Prediction for Five Precipitating Events over France: On the Importance of the Initial Humidity Field

This paper presents results of numerical experiments on five precipitating events: two cases of convective systems over the northern French plains that induced localized flooding, two cases of quasi-stationary systems over the southern French mountainous areas with recorded precipitation above 100 mm in less than 3–4 h, and the extreme flooding case of 12–13 November 1999 with daily precipitation having reached 550 mm, which led to more than 30 deaths. The ability of a high-resolution (2.5 km) model to forecast the quantitative precipitation associated with these five events is evaluated. Both qualitative evaluations that compare observed and modeled reflectivities or surface precipitation, and quantitative evaluations based on classical scores are performed to assess the quality of the numerical experiments. Starting from the same analysis as the operational models, the high-resolution model improves the results for one of the cases, the extreme flash flood case. For the four other cases, higher resolution and more advanced physics than those currently used in operational models do not improve the results. In order to verify that the failures of the high-resolution simulations are due to the initial conditions, the initial state of the simulations are modified for these four cases by using more mesoscale observations, such as mesonet surface observations, radar reflectivities, and Meteosat data. With these modified initial conditions, the high-resolution model produces more realistic precipitation fields. The description of the humidity field in the initial state is crucial for the forecast of the convective systems. Information, such as the presence of a mid- to upper-tropospheric saturated area associated with the developing convective clouds, or nearly saturated low levels, is essential.