The Models and Prediction System

The Models and Prediction System The Center for Ocean-Land-Atmosphere studies (COLA) has recently developed an anomaly coupled prediction system, using sophisticated dynamical ocean and atmosphere models, that produces skillful forecasts of tropical Pacific sea surface temperature anomaly (SSTA) up to 1½ years in advance. The atmospheric component is the COLA atmospheric general circulation model (AGCM, Kinter et al., 1988) that includes a state-of-the-art land surface model (Xue et al., 1991) and physical parameterizations. The AGCM is a global spectral model that is horizontally truncated at triangular wavnumber 30 and has 18 unevenly spaced sigma levels in the vertical. The oceanic component is a Pacific basin version of the Geophysical Fluid Dynamics Laboratory (GFDL) ocean model. In the ocean model there are 20 levels in the vertical with 16 levels in the upper 400 m. The zonal resolution is 1.5º of longitude and 0.5º of latitude between 10ºN and 10ºS. Further details of the ocean model are described in Huang and Schneider 1995.

The atmosphere and ocean models have been tested separately in order to evaluate their performance when forced by observed boundary conditions. In the case of the atmosphere, the prescribed boundary condition is the observed monthly mean SST for 1948-present (personal communication C. K. Folland and Reynolds, 1988) and, in the case of the ocean, several multi-year simulations were carried out forced by observed surface wind stress from ECMWF (1985-92) and FSU (1964-Present) analyses and parameterized surface heat fluxes (Huang and Schneider, 1995 and Kirtman et al., 1995). The ocean model does fairly well in reproducing observed interannual equatorial Pacific SST variability when forced by the observed wind stress.

When the atmospheric model was forced with observed SST and the resulting model output wind stress was used to force the ocean model, serious errors were found in the ocean model simulation (Huang and Schneider, 1995; Kirtman et al., 1995). The errors in the atmospheric model wind stress appear to be most severe during boreal spring when the anomaly becomes extremely weak and far too confined to the western part of the Pacific basin (Huang and Schneider, 1995; Kirtman et al., 1995). We have found that these atmospheric wind stress errors are confined to the boundary layer and that the winds at the top of the boundary layer (approximately 850 mb) can be effectively converted into a surface stress that yields a substantially improved ocean simulation and coupled predictions (Huang and Shukla, 1995 and Kirtman et al., 1995).

In order to improve the ocean simulation and initial states for coupled predictions, we have developed an iterative procedure, based on the SSTA error, for adjusting the observed wind stress that substantially reduces the ocean model simulated SSTA error and possibly gives a superior wind stress field (Kirtman et al., 1995). After this iterative procedure is applied the ocean simulation is used to initialize the coupled forecasts.

The primary difference between the prediction system described here and that used by the coupled model project at the National Meteorological Center (NMC) is in the coupling strategy and the development of initial conditions. Where NMC uses a state-of-the-art ocean data assimilation system to initialize the predictions we use the iterative wind stress assimilation scheme described above (Kirtman et al., 1995) which does not include any observed sub-surface or satellite data. In addition to using the 850 mb zonal wind to define the zonal wind stress anomaly the COLA coupled prediction model exchanges 30 day running means of wind stress and SST anomaly once a day. In contrast, at NMC a model output statistical correction based on a singular value decomposition is applied to the wind stress anomaly every time step and the coupled model exchanges fluxes of heat and momentum once every five days. An earlier version of this coupled model described in Kirtman et al., (1994).

Hindcasts

We have completed a set of hindcast predictions for the period 1964-1994 using the dynamical ocean-land-atmosphere general circulation model described above. To date, a hindcast was initialized for each January and July of 1965, 1966, 1970, 1972, 1973, 1974, 1975, 1982, 1983, 1984, 1986, 1987, 1988, 1989, 1991 and 1994. These dates where chosen to sample warm, cold and normal years. The ocean initial conditions for all of these predictions were taken from an uncoupled simulation in which the 850 mb zonal wind was converted into a surface stress and then the iterative wind stress correction was applied so that both the ocean and atmosphere model are close to equilibrium with the wind stress anomaly and the SSTA respectively. Figure 1 provides an overall assessment of the performance of the anomaly coupled model. The top panel of Fig. 1 shows the SSTA correlation in the NINO3 region and the bottom panel shows the root mean squared error (RMSE). For comparison we show the skill of persistence. It is clear that the COLA coupled prediction model is more skillful than persistence and, by these simple metrics, is competitive with any coupled model currently used for prediction experiments. Moreover, with the statistical correction to the wind stress and the iterative assimilation the systematic errors in the predictions are significantly reduced.


Acknowledgements: This research is part of a larger group effort at COLA to study the predictability of the coupled system. Many members (D. DeWitt, M. Fennessy, J. Kinter, L. Marx and E. Schneider) of this group have provided invaluable advice. The SST data were made available by R. Reynolds and C. Folland. L. Kikas assisted in managing the data. This work was supported under NOAA grant NA26-GP0149 and NA46-GP0217 and NSF grant ATM-93-21354.

References


last update: 2 May 1995