It is generally well accepted that mesoscale model forecasts of precipitation benefit from the addition of mesoscale detail in the initial state of humidity. This is particularly true for tropical regions that are characterized by a lack of reliable conventional atmospheric humidity measurements. Satellite cloud data can provide indirect information on the humidity field, that are not taken into account by the current large scale analysis. This indirect link between the satellite imagery and the humidity field is balanced by good spatial and time samplings (5 km and 30min for the METEOSAT over Africa). A method that uses a satellite-based cloud classification to provide vertical profiles of humidity has been developed. Those profiles are assimilated owing to the use of an Optimal Interpolation scheme in order to modify the initial field of humidity for mesoscale simulations. This approach resulted in some improvements of the precipitation forecast for some mid-latitude systems. Here we present the results of applying this method to a tropical convective event. It corresponds to a day of the JET2000 experiment conducted over Western Africa in August 2000 to study the West African Monsoon (Thorncroft et al. 2001). For this case a cloud classification produced by EUMETSAT has been used to generate the profiles of humidity. The initial fields are provided by the operational ECMWF analysis. The impact of using the "METEOSAT" profiles are tested with the Méso-NH model initialized at two horizontal resolutions, namely 30 km and 5 km. The evaluation of the simulations is based on the comparison of IR synthetic radiances, computed from the model forecasts, with those observed by METEOSAT. The main results can be summarize as following: At meso-beta scale (30km resolution) the convection is parameterized. The inclusion of the humidity profiles within the simulation allows the simulation of the major convective system to occur at the correct location as compared with simulations without the additional humidity profiles. For the simulation without the additional humidity profiles the convective system is generated at the wrong position. Nevertheless the method is sensitive to the local time due to the strong diurnal cycle associated with convection, with the best results obtained when the simulation is started at midday. Also the largest systems collapse during the night contrary to observations and their propagation speed is underestimated. For high resolution runs (5 km) the convection is explicitly resolved and these errors do not occur.