HR: 1330h
AN: H23A-12    [Abstracts]
TI: Satellite Rainfall Probability and Estimation. Application to the West Africa During the 2004 Rainy Season
AU: * Chopin, F
EM: chopin@lmd.polytechnique.fr
AF: Laboratoire de Meteorologie Dynamique IPSL/CNRS, Ecole Polytechnique, Palaiseau, 91128 France
AU: Berges, J
AF: PRODIG Universite Paris 1, 191 rue Saint Jacques, Paris, 75005 France
AU: Desbois, M
AF: Laboratoire de Meteorologie Dynamique IPSL/CNRS, Ecole Polytechnique, Palaiseau, 91128 France
AU: Jobard, I
AF: Laboratoire de Meteorologie Dynamique IPSL/CNRS, Ecole Polytechnique, Palaiseau, 91128 France
AU: Lebel, T
AF: IRD, LTHE, BP 53, Grenoble Cedex 09, 38041 France
AB: The international program AMMA (African Monsoon Multidisciplinary Analysis) is in intensive phase from the beginning of 2005 until 2007, over West Africa. It has a crucial need of precipitation estimations at scales ranging from the small basin to the regional scale, and from instantaneous values to monthly totals. This need includes estimations of the errors corresponding to each scale. Moreover, no sufficient ground data are available in West Africa to satisfy the AMMA time and space scale needs. To fulfil those, a satellite precipitation algorithm is developed. In order to get relevant information in tropical regions with very sporadic rainfall, the use of a high time sampling, which can only be provided by the geostationary satellites, is required. Unfortunately, although statistical information on rainfall occurrence can be obtained from these geo-satellite data, instantaneous rain rate intensity cannot be derived from the available infrared or visible channels. Because of their close relationship with rainfall phenomenon, active or passive microwave data from low orbit satellites are necessary. In this study the TRMM Precipitation Radar (PR), which provides instantaneous values of ground rainfall intensity, is used. The training of a neural network system is performed using six months collocated data during the 2004 rainy season of Meteosat-8 (MSG) infrared data and TRMM PR algorithm rain estimates. Several Meteosat-derived parameters, including radiances and space-time characteristics, constitute the entries, while Precipitation Radar 2A25 rain information (rain/no rain) are used to train the feed forward neural network. The outputs are considered as Rainfall Probabilities. Rainfall estimations from these outputs are obtained by multiplying these Rainfall Probabilities by a Potential Rainfall intensity. Considering that the Potential Rainfall intensity depends on the geographical localisation and the phase of the seasonal cycle, its value has to vary in space and time. These Potential Rainfall intensities are calibrated by a reference dataset of rainfall estimations, and have been calculated thanks to an upscaling formula. The reference dataset used for this study was the 1dd GPCP product because of its global geographical cover and its appropriate temporal resolution. From these two products (Rainfall Probability and daily Potential Rainfall intensity) estimated rainfall intensity fields are provided at the fine geostationary satellite time and space resolutions. By taking into account the Rainfall Probability, the algorithm presented here may be interpreted as a way to improve and downscale the rainfall estimations deduced from the reference dataset.
DE: 3314 Convective processes
DE: 3354 Precipitation (1854)
DE: 3374 Tropical meteorology
SC: Hydrology [H]
MN: 2005 Joint Assembly