In summary, without data assimilation the performance of the forecasts is sensitive to the surface wind or wind stress forcing. Those forecasts from control using FSU wind forcing are better than those from the control using ERA/Ops forcing. Data assimilation constrains the statistics of the forecasts significantly, making them more or less independent of the wind forcing used in the assimilation, at least in terms of the mean statistics for 1991-96. Furthermore, forecasts initiated from data assimilation are in general better than those without data assimilation over the period 1991-96. All forecasts (with and without data assimilation, and different wind forcing) underestimated the strong warming around April/May 1997. This suggests that the problem may not be related to errors in the initial conditions, at least not in the temperature field.
SENSITIVITY TO INITIAL CONDITIONS
Figure 3 Ensemble mean 4-6 month forecast SST anomaly for the NlNO 3 region. Line-observed (with 3 month running mean), star - A_ERA, diamond - A_FSU.
The differences in the coupled model forecasts described in the previous section are simply due to differences in the initial states. This section investigates how differences in the initial states impact on the forecasts. Differences between the two assimilations A_FSU and A_ERA are studied. These were chosen rather than, say, differences between the controls or differences between an assimilation and its corresponding control, although these would be of interest in their own right. This choice is because these two sets of initial conditions will have the smallest differences and hence the differences that evolve in the forecasts are less likely to be influenced by non-linear effects. Also a significant part of the difference between say a control and assimilation would be the difference in the mean state. Choosing the two assimilations will focus on differences due to the assimilation procedure rather than systematic error. Furthermore, it is interesting to see where the data assimilation has not overcome differences due to different wind fields which subsequently impact upon the forecast.
