Dynamical Forecasts of Tropical Pacific SST and North American Climate Using a Two-Tiered Process

contributed by Tony Barnston1, Ming Ji2, Arun Kumar2, Wanqiu Wang2 and Ants Leetmaa1

1Climate Prediction Center, National Centers for Environmental Prediction, NOAA, Camp Springs, Maryland

2Environmental Modeling Center, National Centers for Environmental Prediction, NOAA, Camp Springs,Maryland

A non-simple coupled ocean-atmosphere model has been developed for use in long-lead climate forecasting in the Coupled Model Branch of the Environmental Modeling Center (EMC) at NOAA's National Centers for Environmental Prediction (NCEP) (Ji et al. 1994a,b). The NCEP Medium Range Forecast (MRF) atmospheric model is used with a dynamic Pacific Basin ocean model originated at the Geophysical Fluid Dynamics Laboratory. The MRF has a reduced spatial resolution and is tuned for more realistic tropical circulation. The ocean thermal field, including SST and subsurface temperature, is initialized using an ocean data assimilation system (Ji et al. 1995). As with most AGCMs, this coupled model's atmospheric response to observed tropical Pacific SST is fairly reliable in the tropics but considerably less so in the extratropics. The extratropical response is most realistic during ENSO extremes. The nature of the model's mid-latitude atmospheric response to the ENSO state has been discussed in Livezey et al. (1997). While skillful prediction of the extratropical atmosphere is a major goal, a prerequisite is the ability to predict ENSO itself--the tropical Pacific SST anomaly field. Both SST forecasts and surface climate forecasts for North America are presented here.

SST Forecasts

As has been the case since the original presentation of the NCEP model in this Bulletin in 1993, the expected forecast skill field has a horseshoe-shaped pattern with highest equatorial model skill near the date line and higher skill just north or south of the equator than immediately along it to the east of 165oW. Since 1993 several major improvements have occurred in the model, resulting in higher skill (Ji et al. 1995; Ji et al. 1996). The more recent versions use an ocean data assimilation system (Ji et al. 1995). The basic skill characteristics of the earlier versions of the model remain, including its geographical distribution and season dependence (e.g. a relative spring skill barrier). The skill of the SST forecasts for SST in the Niño 3.4 region for the first four lead times is shown in Table 1, based on the 1982-97 period. NCEP model forecasts out to 6 months lead are now updated on a weekly basis and are available on Internet site http://www.emc.ncep.noaa.gov/cmb/sst_forecast/images/cmb.fsst.gif.

The NCEP coupled model forecasts for the SST anomaly field averaged over Sep-Oct-Nov, Dec-Jan-Feb and Mar-Apr-May for 1999-2000 are shown in Fig. 1, where the systematic model bias for hindcasts over the 1981-95 period has been removed. This forecast is the mean of an ensemble of 7 to 11 individual cases, each based on a different one- to two-week-apart initial ocean condition ranging from early June through early September 1999. The forecasts show the continuation of La Niña conditions through winter 1999-2000, followed by dissipation in spring. The subsurface equatorial sea temperature (not shown) still indicates a good supply of below-normal water temperature beneath the surface in the eastern part of the basin. However, warm subsurface water is found in the western Pacific, and has had some brief eastward protrusions.

North American Surface Climate Forecasts

The predictions of tropical Pacific SST are taken one step farther, and used as a lower boundary condition for integrations of the atmospheric GCM (AGCM) out to comparable lead times. The AGCM forecast is the ensemble mean of 18 individual integrations, each using a different atmospheric initial condition. The skill of the model's mid-latitude atmospheric response to the ENSO state, specifically over the PNA region, is found to be moderate during the northern winter during warm and cold ENSO episodes (Livezey et al. 1997). Earlier work also showed that skill in specifying geopotential heights, temperature and (to a lesser extent) precipitation are usable during the cold half of the year as long as there is an ENSO signal (Livezey et al. 1996). Skill during non-ENSO years is not significantly different from zero. This ENSO-dependency of skill is plausible, given signal-to-noise ratio considerations in the extratropics with respect to ENSO-related anomalous tropical Pacific SST boundary forcing (Kumar and Hoerling 1998).

Here, we present GCM forecasts for 3-month mean North American surface temperature and precipitation out to 8 months of lead time, where lead time refers to the time between the forecast time and the middle month of the 3-month target period. For each location, season and lead time, the expected forecast skill is the product of (1) skill in the forecast of the tropical Pacific SST boundary condition (shown in Table 1 for the first 4 leads out of 8) and (2) skill of the simultaneous simulations (specifications) of the atmosphere, given a perfect SST forecast ("AMIP" skill). The expected forecast skill over the U.S., given a perfect SST forecast, is shown in Livezey et al. (1996). These specification skills, averaged over the U.S. for winter and expressed as temporal correlation coefficients, are near 0.4 for 700 mb height and for surface temperature, and about 0.1 to 0.2 for precipitation, when ENSO is active (i.e., SST at least 1 standard deviation amplitude). When multiplied by the imperfect correlation skills for predicting tropical Pacific SST, these skills drop accordingly. The final skill is presently approximately comparable to that of the statistical CCA forecasts, also produced at CPC (Barnston et al. 1994); however, in specific circumstances the skills of the two forecasting approaches may differ substantially. In the regions that experience strong ENSO impacts (e.g. around the Great Lakes and southern Canada, and near the Gulf of Mexico and the Southeast), however, winter skills are locally moderately high during ENSO episodes at lead times of up to 6 months. The user of the surface forecasts presented here is urged to exercise caution, considering the lead time and the region of interest. Additionally, spatial displacement biases may cause "centers of action" to appear slightly out of their appropriate location. A model output statistics (MOS) correction scheme (as in Smith and Livezey 1998) would in principle be able to reduce these spatial biases, as well as systematic forecast amplitude errors.

The current forecasts for North American surface temperature and precipitation are shown in Fig. 2 and Fig. 3, respectively. Warmth is forecast for the south-central and eastern U.S. for late fall through winter 1999-2000, which is compatible with the canonical La Niña response except perhaps for extending farther into the northeast U.S. than expected. Cold is forecast for Alaska and the northwestern coast of North America, including the northwestern U.S. in early winter. The precipitation forecasts show dryness in the southeastern U.S. for fall and winter, as would be expected with La Niña boundary conditions. Enhanced rainfall is forecast in British Columbia and the northwestern U.S. for late fall and early winter. In late spring 2000, the model indicates a reversal to cool, wet conditions along the Gulf of Mexico coast. This forecast is undoubtedly accompanied by great uncertainty.



References

Barnston, A.G., H.M. van den Dool, S.E. Zebiak, T.P. Barnett, M. Ji, D.R. Rodenhuis, M.A. Cane, A. Leetmaa, N.E. Graham, C.F. Ropelewski, V.E. Kousky, E. A. O'Lenic and R.E. Livezey, 1994: Long-lead seasonal forecasts-Where do we stand? Bull. Amer. Meteor. Soc., 75, 2097-2114.

Ji, M., A. Kumar and A. Leetmaa, 1994a: A multi-season climate forecast system at the National Meteorological Center. Bull. Am. Meteor. Soc., 75, 569-577.

Ji, M., A. Kumar and A. Leetmaa, 1994b: An experimental coupled forecast system at the National Meteorological Center: Some early results. Tellus, 46A, 398-418.

Ji, M., A. Leetmaa and J. Derber, 1995: An ocean analysis system for seasonal to interannual climate studies. Mon. Wea. Rev., 123, 460-481.

Ji, M., A. Leetmaa and V.E. Kousky, 1996: Coupled model forecasts of ENSO during the 1980s and 1990s at the National Meteorological Center. J. Climate, 9, 3105-3120.

Kumar, A. and M.P. Hoerling, 1998: Annual cycle of Pacific/North American seasonal predictability associated with different phases of ENSO. J. Climate, 11, in press.

Livezey, R.E., M. Masutani and M. Ji, 1996: SST-forced seasonal simulation and prediction skill for versions of the NCEP/MRF model. Bull. Am. Meteor. Soc., 77, 507-517.

Livezey, R.E., M. Masutani, A. Leetmaa, H. Rui, M. Ji and A. Kumar, 1997: Teleconnective response of the Pacific-North American region atmosphere to large central equatorial Pacific SST anomalies. J. Climate, 10, 1787-1820.

Smith, T.M., and R. Livezey, 1998: GCM systematic error correction and specification of the seasonal mean Pacific/North American region atmosphere from global SSTs. J. Climate, 11, in press.



Table 1. Expected skill (correlation X100) of the NCEP coupled model in predicting the SST anomaly in the Niño 3.4 region at 4 lead times for 12 overlapping 3-month target seasons. A 1-month lead is, for example, a forecast for JFM made at the end of December.
LEAD JFM FMA MAM AMJ MJJ JJA JAS ASO SON OND NDJ DJF
1-mon 90 87 80 76 78 79 80 84 88 91 94 92
1-mon 89 84 77 72 73 74 76 81 86 89 92 91
3-mon 88 81 74 68 68 70 72 75 83 87 90 90
4-mon 88 79 68 64 63 66 68 70 79 84 88 90



Figure Captions

Fig. 1. NCEP coupled model SST anomaly forecast fields for Sep-Oct-Nov, Dec-Jan-Feb, and Mar-Apr-May for 1999-2000. Each forecast is an average of about 13 individual ensemble members, each based on a different mean of an ensemble of 7 to 11 individual cases based on a different 1- to 2-week-apart initial ocean condition ranging over the last three months (see text). These forecasts were made at the time shown at the bottom of the figure.

Fig. 2. North American surface air temperature anomaly forecasts ( C) of the two-tiered NCEP model (coupled model for the SST forecast, then AGCM using the SST forecast as boundary condition) for the coming 8 overlapping 3-month periods. The mean of an 18-member ensemble of GCM integration result is used.

Fig. 3. As in Fig. 2, except for precipitation.