Figure 3 shows the 4-6 month average ensemble mean forecast anomaly for the NlNO 3 region for forecasts starting from the initial states produced by A_ERA and A_FSU. The differences between the two sets of forecasts are generally quite small and, with one exception, less than 0.5℃. In many cases the forecasts are virtually identical (less than 0.1℃). However, there are several cases where these differences are greater than 0.2℃. In general, the differences between the forecasts from the two assimilations are significantly smaller than differences between forecasts from the two controls or differences between forecasts from a control and an assimilation.
The impact of assimilation differences on forecasts can be investigated, in the first instance,by looking at how differences in say the 20℃ isotherm depth in two different analyses (A_ERA and A_FSU) impacts on the NlNO 3 SST anomaly forecasts. This is done here by calculating the correlation coefficient between differences in the two analyses (at each horizontal grid point) and differences in the NINO 3 SST anomaly forecasts, as a function of lead time, ove rall the cases in 1991-96. The absolute value of this correlation is shown in fig. 4. High correlation values indicate regions where differences between the A_FSU and A_ERA initial states are strongly correlated to differences in the NlNO 3 ensemble mean forecasts from each of these analyses.
Figure 4 Absolute value of the correlation coefficient between differences in the 20℃ isotherm depth in the assimilation initial states (A_ERA and A_FSU) and the differences in the NlNO 3 SST forecasts starting from these initial conditions (5 member ensemble forecast), as a function of lead time.
At lead times of one and two months there are correlations exceeding 0.6 in the equatorial central/eastern Pacific (NINO 3 region). These decrease with lead time so that at four months lead time and beyond there is little correlation. This strong local correlation at short lead times shows that the representation of the thermocline structure in the NlNO 3 region is important for the subsequent short range forecasts. This is also the region which is strongly wind driven, so that errors in the wind field would result in differences in the analyses even if this is a relatively data rich region covered by the TOGA-TAO array.
