Sensitivity of ENSO Forecasts to Ocean Initial Conditions
J. O. S. Alves, D. L. T. Anderson, T. N. Stockdale, M. A. Balmaseda and J. Segschneider (Seasonal Forecasting Group, European Centre for Medium-Range Weather Forecasts, Shinfield Park, Reading, RG2 9AX, England)
email: nea@ecmwf.int (Oscar Alves)
ABSTRACT
ECMWF has developed a global operational coupled model system to produce seasonal forecasts. One of the components of this system is a global ocean data assimilation procedure which assimilates into the ocean model all available sub-surface ocean temperature observations.
One of the main contributors to predictability on the seasonal timescale is ENSO. The impact of ocean data assimilation on ENSO forecasting is studied by looking at forecasts started from initial conditions produced by the data assimilation system compared to those produced by an ocean model constrained only by the surface forcing and observed sea surface temperature. In general, the forecasts are significantly better when sub-surface temperature information is taken into account. The statistics of the forecasts errors over this period show that, with data assimilation, the forecasts (of NlNO 3 SST anomaly) clearly beat persistence over all lead times (1-6 months), which is not the case when sub-surface ocean observations are not assimilated.
Errors in the forecasts can arise from uncertainties in the atmosphere model, ocean model and initial conditions (ocean, atmosphere, land). Errors due to uncertainties in the ocean initial conditions are examined by comparing coupled forecasts started from initial conditions using different initialisation methods (with and without ocean data assimilation and different assimilation set ups). Results show that ocean data assimilation can have a large positive impact on the coupled model forecasts of ENSO. High correlations are found between these forecasts (of NINO 3 SST anomaly) at short lead times and uncertainties in the ocean initial state locally in the central/eastern equatorial Pacific. At longer lead times high correlations are found from differences in the initial state at a remote region, i.e. north western tropical Pacific, outside the TOGA-TAO array region.
INTRODUCTION
The importance of ENSO (EI Nino Southern Oscillation) for variability of the earth's climate on seasonal to inter-annual timescales has lead to a development of many systems to forecast SST anomalies in the tropical Pacific. These very widely in complexity, ranging from purely statistical methods to fully coupled dynamical models of the ocean/atmosphere system. Latif et al. (1998) review these methods and their performance for hindcasts of periods in the 1980s and 1990s. Trenberth (1998) reviews operational and near operational forecasts of the 1997/98 El Nino. He concludes that, at least for the 1997/98 El Nino, the systems that performed best were based on coupled dynamical models of the ocean and atmosphere.
ECMWF has set up its own operational seasonal forecasting system, one of its main aims being the forecasting of ENSO up to six months lead time. It is based on a fully coupled ocean atmosphere model. Some results have already been reported in Stockdale et al. (1998). Forecasting the SST in the tropical Pacific using a fully coupled model is essentially an initial value problem; any predictability can only come from information contained in the initial state of the system (ocean, atmosphere, land, etc). For the atmosphere and land the ECMWF system uses initial states from the NWP analyses. For the ocean, an optimum interpolation data assimilation scheme is used, which is the subject of this paper. The ocean and atmosphere models are fully coupled (the atmosphere model sees the full SST), flux correction is not used.
