Bayesian identification and validation of a daily rainfall model under monsoon conditions

A model for daily rainfall occurrence and daily rainfall amount is fitted t o the data of Dedougou, in the Sudano-Sahelian zone. The fitting is perform ed in a Bayesian framework using the Hastings procedure which allows for a full representation of parameter uncertainty. The method is based on a comp arison of any observed characteristic of the rainy season to its predictive law. If and only if the observation falls into the central part of the pre dictive distribution the model is judged to describe the characteristic und er scrutiny. Using this method, several records were studied. It is shown t hat the model preserves the commonly used hydrological records like daily a nd annual precipitation, number of rainy days per year, rainy sequence leng th. The model depicts the climatic change which occurred in the late sixtie s. The Markovian hypothesis, on which the model is based, is validated. The independence hypothesis between rainfall amounts in two successive days do es not hold. The extreme daily rainfall values are consistent with the larg est values generated by the model.