We describe forecasts for tropical Pacific sea surface temperature (SST) made by combining the forecasts produced by multiple independent coupled atmosphere ocean general circulation models (CGCMs). This initial forecast framework uses two CGCMs. The two CGCMs utilize the same ocean component model and are initialized using the same ocean data assimilation product. This work is an extension of the combined CGCM forecasts described in Schneider et al. (2003). Differences between that work and the current forecasts include use of an extended period for developing prior statistics, use of fewer models, and employment of a different technique for combining the SST forecasts from the different models. This manuscript documents forecast skill for April 1 initial condition retrospective forecasts and gives the real time forecast starting from April 1, 2004.

The model descriptions have been given in previous versions of the Extended Long-Lead Forecast Bulletin (ELLFB) and are skipped here for brevity. The atmospheric GCM (AGCM) component models used are the ECHAM4.5 AGCM of the Max Planck Institute for Meteorologie (Roeckner et al., 1996) and the Center for Ocean-Land-Atmosphere (COLA) studies AGCM version 2 (Schneider, 2002). The ocean GCM (OGCM) component model is version 3 of the Geophysical Fluid Dynamics Laboratory (GFDL) modular ocean model (MOM) (Pacanowski and Griffes, 1998). The ocean initial conditions are taken from an ocean data assimilation system produced at GFDL using a variational optimal interpolation (Derber and Rosati, 1989).

Procedure for Producing the Forecast

Retrospective forecasts using 7 ensemble members for the period April 1982 to present have been made using the coupled system with ECHAM AGCM component.

Similar retrospective forecasts using 5 ensemble members have been made using the coupled system with COLA AGCM component. The variance of SST anomalies differs between the two models and is also not the same as found in nature. The retrospective forecasts from both of the coupled systems have been normalized by the observed variance. At each point a bootstrap cross validation procedure is applied to bring the forecast variance closer to observed. In this calculation, the forecast for a particular year is not used to compute the mean or the standard deviation for either the model or the observations. The forecast anomaly for that year is taken as the difference from the model climatology made without using that year multiplied by the ratio of observed to model standard deviation also calculated without the data from the year of the forecast. The final SST forecast is the simple arithmetic mean of the normalized forecasts from the two coupled models.

Retrospective Forecast Skill

An estimate of the skill of the combined coupled models is provided by the retrospective forecasts for the period 1982 to 2003. Here the forecasts for April 1 initial conditions are shown. It is known that the skill for tropical Pacific SST forecasts is a strong function of season (Balmaseda et al., 1995) so it is important to examine forecast skill as a function of initial condition date. As a comparison forecast we use persistence of the SST anomaly at the start of the forecast. The persistence forecast has been processed using the same bootstrapping methodology to bring the variance closer to observed as described for the coupled forecast above. Figure 1 shows the anomaly correlation coefficient (ACC) and root mean square error (RMSE) for the combined coupled model forecast and persistence for SST averaged over two regions. The regions shown are Nino3.4 (170°W to 120°W, 5°S to 5°N) and Nino3 (150°W to 90°W). The combined coupled model forecast beats persistence for all leads in both Nino regions. This is true both in terms of the ACC and the RMSE. Also shown in Figure 1 are the skill scores for the individual coupled model systems. It can be seen that the multi-model combination skill basically tracks that of the higher skill model, ECHAM4.5 in this case. This means that the multi-model forecast is not always an improvement on the individual component model forecasts for this initial condition.

Current Forecast

The current forecast for the Nino index averaged SST anomalies made from April 1, 2004 is shown in Figure 2. The combined coupled forecast calls for both Nino regions to have very weak cold anomalies trending toward even weaker cold anomalies for Nino3 and weak warm anomalies for Nino3.4.

Acknowledgements

These forecasts were made possible due to help from several institutions. Matt Harrison and Tony Rosati of GFDL developed the ODA system and ran the 1980 to 1999 period. Ben Kirtman and Duhong Min of COLA have ported the ODA to their system and run the 2000 to present period. Bohua Huang of COLA implemented the lower resolution of the OGCM used here. Max Planck has kindly provided the ECHAM4.5 AGCM to the IRI. David DeWitt was supported by a grant from the National Oceanic an Atmospheric Administration: (NA07-GP0213). Support for Schneider and Hu was provided by the National Science Foundation (ATM 98-14295,ATM01-22859), the National Oceanic and Atmospheric Administration (NA 96-GP0056), and the National Aeronautics and Space Administration (NAG 5-8202).

References:

Balmaseda, M. A., M. K. Davey, and D. L. T. Anderson, 1995: Decadal and seasonal dependence of ENSO prediction skill. J. Climate, 8, 2705-2715.

Derber, J. and A. Rosati, 1989: A global oceanic data assimilation system. J. Phys. Oceanogr., 19, 1333-1347.

Pacanowski, R. C., and S. M. Griffes, 1998: MOM 3.0 Manual, NOAA/Geophysical Fluid Dynamics Laboratory, Princeton, USA 08542.

Roeckner, E., and Coauthors, 1996: The atmospheric general circulation model ECHAM4: Model description and simulation of present day climate. Max-Planck-Institut fur Meteorologie, Rep. 218, 90pp. [Available from MPI fur Meteorlogie, Bundesstr. 55, 20146 Hamburg, Germany.]

Schneider, E. K., 2002: Understanding the differences between the equatorial Pacific as simulated by two coupled GCMs. J. Climate, 15, 449-469.

Schneider, E. K., D. G. DeWitt, A. Rosati, B. P. Kirtman, L. Ji, and J. J. Tribbia, 2003: Retrospective ENSO Forecasts: Sensitivity to atmospheric model and ocean resolution. Mon. Wea. Rev., 131, 3038-3060.