e)Predictability of summer rainfall, and influence of initial conditions
Ensemble integrations for the northern summers 1987 and 1988 have been completed using the ECMWF model. These two summers are of particular interest for their contrasting weather in different parts of the world. Over both the African Sahel, and over the Indian subcontinent, rainfall for JJA 1987 was well below average, and both these regions experienced drought conditions. By contrast, the African and Indian monsoons were good (above average for India, close to the long-term average for the Sahel) in 1988. On the other hand much of the central and eastern USA suffered one of its worst droughts during the spring and summer of 1988. During this period, the El Nino evolved from warm to cold conditions.
The JJA accumulated rainfall for each ensemble member for both 1987 (heavy bars) and 1988 (light bars) is shown in Fig 6 for the three regions mentioned above. Ensemble mean values are shown as the letter "M". For the Sahel, nearly all members of the 1988 ensemble show more rainfall than any of the 1987 members. In general terms it can be seen that the Sahel rainfall difference between 1987 and 1988 is highly predictable, and agrees with the observed difference. By contrast, the difference between 1987 and 1988 JJA Indian rainfall is much less predictable (compare, for example, the difference in rainfall for the first and second members of the 1987 ensemble). Nevertheless, taking the ensemble as a whole, there is an overall indication o1 wetter conditions in 1988 (cf the position of the ensemble mean fields), in agreement with observations.
It is generally thought that seasonal predictability is stronger in the tropics than in the extratropics. However, if we compare the Indian monsoon rainfall estimates with those for the eastern USA, we see that the latter shows a highly predictable signal. All members of the 1987 ensemble are wetter than any member from the 1988 ensemble. The relative drought in 1988 has been well simulated. In order to understand the origin of the rainfall predictability over the Sahel and over the eastern USA, a number of additional experiments were performed in which the effects of the initial conditions were isolated from the effects of the SSTs. For each of six initial dates, the following additional experiments were performed: a) with 1987 initial fields but with 1988 SSTs, b) with 1987 SSTs and 1987 upper air initial fields, but with 1988 surface (and subsurface) fields.
Results can be summarised as follows: changing SSTs from 1987 to 1988 (but keeping the initial conditions as 1987) has a reproducible impact on Sahel rainfall. For all six initial dates, the impact of the 1988 SSTs is to increase rainfall quite dramatically. By contrast there is no clearly reproducible impact of changing (only) the initial surface conditions from 1987 to 1988. However, the combination of initial surface and upper air conditions does have a reproducible effect on Sahel rainfall. Essentially the 1987 initial conditions produce more rain in all six cases, than the 1988 initial conditions (both with 1988 SST). In this case, it therefore appears that the influence of the SSTs and the initial conditions are in opposition to one another.
For the Eastern USA, the influence of SST is relatively weak. In 5 out of 6 integrations, 1988 SSTs reduce rainfall compared with the 1987 control, but not by a large amount.
