The high resolution simulations (horizontal mesh of 2.5 km) have been carried out with the 3D mesoscale non-hydrostatic model MESO-NH, allowing a two-way interactive grid-nesting fashion. With the intention of improving the numerical simulation of these convective cases, a particular care is focused on the model initialization. In this study, we assess different initialization methods: a static fine-scale physical one and several dynamic initializations.
For the static fine scale initialization, the meso-beta scale description of low-level conditions is improved by the use of a mesoscale surface data analysis as initial state. Moreover, a meso-gamma scale information about the presence of a developing convective system is supplied. Based on a simple cloud analysis from conventional radar and infrared satellite data, this step adjusts the humidity and hydrometeor fields of initial state. This improved initial state produces well developed systems and leads to better precipitation forecasts.
For dynamic initializations, we have developed a sequential data assimilation based on the nudging technique. The basic idea is to relax during a pre-forecast period some prognostic model variables (wind, temperature and humidity) towards their observed or analyzed values. For meso-beta scale, these variables are relaxed during a 3-h period towards mesoscale analysis. For meso-gamma scale, hydrometeor fields are relaxed during a shorter period towards some estimated values from conventional radar reflectivities and infrared satellite data.
Comparisons with observations and between the different initial states, by means of scores, backward and forward trajectories will be present at the conference as well as a discussion about the initialization strategy for fine scale numerical simulations.