We are interested in the large-scale environmental phenomenon of West African monsoon. It is the major atmospheric phenomenon, which drives the rainfall regime in Western Africa. The causes of spatio-temporal variability in monsoon rainfall have not yet been determined in an unequivocal manner. However, there is a considerable body of evidence suggesting that spatio-temporal changes in sea surface temperatures in the Gulf of Guinea and changes in the Saharan and sub-Saharan albedo are major factors. The aim of the study is to analyse the sensitivity of the rainfall simulated by a regional atmospheric model(MAR) to the variability of these inputs parameters. Precipitation from MAR is compared with severals precipitation data sets and the MAR reasonably simulates the Westy African monsoon. This model is expensive in computing time, the sensitivity analysis cannot be realized by running the MAR, as we work on discretization grids in space and time, that is with huge dimensions. Hence an important preliminary step is the construction of a stochastic spatio-temporal metamodel approximating the MAR. The main properties required for this metamodel is the ability to be ran in a reasonable time and the consideration of the spatio-temporal dynamic of the underlying physical phenomenon. In this study we neglect the effect of albedo and focus our effort on regressing the rainfall on the sea surface temperature (SST). An important point in this study is that the numerical storage and processing of model outputs, as far as the statistical description of the data, requires considerable computation resources. A grid environment can provide the required resources. Nevertheless, one main difficulty of the grid platform is the resource provisioning. How to find the best resource at a given time and the best amount of theses resources? The answer should come from the middleware designed with an efficient scheduler. Moreover the middleware can give a transparent access to a distributed and heterogeneous platform as a Grid. We used DIET,we will see how it schedule a huge number of computation requests according to the data management.