Modeling Interannual Variations of Summer Monsoons

Results from a set of 90-day integrations, made with a T42 version of the ECMWF model and forced with a variety of specified sea surface temperature (SST) datasets, are discussed. Most of the integrations started from data for 1 June 1987 and 1 June 1988. During the summer of 1987, both the Indian and African monsoons were weak, in contrast with the summer of 1988 when both monsoons were much stronger. With observed SSTs, the model is able to simulate the interannual variations in the global-scale velocity potential and stream-function fields on seasonal time scales. On a regional basis, rainfall over the Sahel and, to a lesser extent, India showed the correct sense of interannual variation, though in absolute terms the model appears to have an overall dry bias in these areas. Additional integrations were made to study the impact of the observed SST anomalies in individual oceans. Much of the interannual variation in both Indian and African rainfall can be accounted for by the remote effect of the tropical Pacific SST anomalies only. By comparison with the effect of the Pacific, interannual variability in Indian Ocean, tropical Atlantic Ocean, or extratropical SSTs had a relatively modest influence on tropical large-scale flow or rainfall in the areas studied. Integrations run with identical SSTs but different initial conditions indicated that for large-scale circulation diagnostics, the impact of anomalous ocean forcing dominated the possible impact of variations in initial conditions. In terms of local rainfall amounts, on the other hand, the impact of initial conditions is comparable with that of SST anomaly over parts of India and Southeast Asia, less so over the Sahel. While this may suggest that a nonnegligible fraction of the variance of month-to-seasonal mean rainfall on the regional scale in the tropics may not be dynanamically predictable, it is also quite possible that the disparity in the apparent predictability of rainfall and circulation anomalies is a reflection of model systematic error.