From rainfall analysis over West Africa (rotated empirical orthogonal functions, Varimax criterion), significant leading modes are identified: i.e., along the Gulf of Guinea, the Senegal/west Mali area, and the central Sahel. Using the moisture convergence/divergence (MCD) parameter derived from ERA-40, three leading empirical orthogonal function modes are extracted. They are respectively linked to the Intertropical Convergence Zone position, the equatorial Atlantic upper-ocean thermal conditions (center and west), and significantly correlated with identified rainfall modes. Linkages between temporal variability of the above modes and that of the tropical Atlantic upper-ocean thermal are highlighted. A new statistico-dynamical probabilistic model for West African rainfall prediction is thus built using MCD as predictors diagnosed from the DEMETER hindcasts. The model takes advantage of both identified statistical linkages, and conditional probabilities between precipitation occurrence and MCD values. It is found that new probabilistic scores for precipitation predictions, are significantly improved when compared to those which were obtained directly from DEMETER precipitation forecasts.