The Japan Meteorological Agency (JMA) operates a Coupled ocean-atmosphere General Circulation Model (CGCM) as an El Niño prediction tool.
Operational information of the El Niño outlook has been produced with the new prediction system since March 2010. The new system improved the calculation method for initial perturbations, with perturbations for the atmosphere being introduced in addition to those for the ocean. Moreover, the ensemble size for the calculation of ensemble statistics was increased from 30 to 51.
Since this improvement, the CGCM has also been applied to seasonal weather forecasts.
The oceanic part of the model is known as MRI.COM (Ishikawa et al., 2005), and is identical to the ocean general circulation model used in the Multivariate Ocean Variational Estimation/Meteorological Research Institute Community Ocean Model-Global (MOVE/MRI.COM-G) system.
The atmospheric component is a lower-resolution version of the Global Spectral Model (GSM0603) used by JMA for operational numerical weather prediction (JMA, 2007).
To improve the features of heat, momentum and fresh water fluxes on the sea surface, several physical parameterizations in the atmospheric component were modified from those in the higher-resolution model.
Coupling between the ocean and the atmosphere is performed every hour.
The oceanic component of the CGCM supplies sea surface temperatures (SST) to its atmospheric component, while the atmospheric component provides hourly-mean heat, momentum and freshwater fluxes to the ocean component.
Simulated fields in a coupled model are liable to approach the model climate state, which differs substantially from the averaged state of the real ocean and atmosphere.
In order to suppress this climate drift, adjustment is made to both the heat and momentum fluxes.
The adjustment amounts are statistically calculated for each initial month and for each lead time from hindcast experiments:
the momentum-flux adjustment is estimated as the difference between JRA-25 analysis and atmospheric model prediction, while the heat-flux adjustment is estimated from the thermal amount required for relaxation of the prediction to the analysis of SSTs.
The specifications of the CGCM are summarized in Table 1.
The Ensemble Prediction System (EPS) with the CGCM for El Niño prediction adopts a combination of the initial perturbation method and the Lagged Average Forecasting (LAF) method.
Nine members are run every five days, and the EPS consists of fifty-one members for the latest six initial dates.
Initial perturbations are generated both for the atmosphere and for the ocean.
The results of analysis by the JMA Climate Data Assimilation System (JCDAS) are adopted as the atmospheric initial conditions for the CGCM.
Atmospheric initial perturbations are obtained using the Breeding of Growing Modes (BGM) method.
Oceanic initial perturbations are estimated through the ocean data assimilation sys
tem (MOVE/MRI.COM-G) forced with surface heat and momentum fluxes in the atmospheric initial perturbation fields.
A set of 348 hindcast experiments of 15-month prediction from January 1979 to December 2007 was performed.
Figure 2 shows all predicted sea surface temperature (SST) deviations from the 1971 - 2000 means for the NINO.3 region (5ºN - 5ºS, 150ºW - 90ºW) since 1979 together with the observed values.
JMA/MRI-CGCM appropriately predicts most cases with large deviations for the NINO.3 region.
In particular, the occurrence of the 1997/98 El Niño event is well predicted more than six months in advance, although the predicted amplitude is a little smaller than the observation.
Figure 3 shows the performance of JMA/MRI-CGCM in terms of anomaly correlations and root mean square errors (RMSE) for NINO.3 SST anomalies for the 348 hindcast experiments from 1979 to 2007.
The anomaly correlations for persistence prediction and the RMSEs for persistence and climatology predictions are shown for reference.
The anomaly correlations for the predictions with 8- and 12-month lead times are nearly 0.6 and over 0.3, respectively.
It is shown that the model performs well up to a 9-month lead time in terms of RMSE compared to climatological and persistence predictions.
(1.0º (lon) x 0.3º (lat) near equator)
50 vertical levels
Asterisks (red), open squares (green) and solid squares (blue) denote the model's prediction, persistence prediction and climatological prediction, respectively.
