Equatorial Pacific SSTA forecasts

Steve Zebiak

If you're familiar with the differences between the LDEO and LDEO2 products you can go directly to the analysis or data. Otherwise read on.

Forecast sea surface temperature fields are produced on a monthly basis using the Zebiak and Cane (1987) coupled atmosphere-ocean model.

At present, two forecast products are available: The standard forecasts (consistent with those issued in the NOAA Diagnostic Bulletin over the past several years), and a newer product, referred to as LDEO2, that utilizes a different forecast intialization scheme.

In the standard initialization, FSU pseudo-stress anomalies (with some filtering and smoothing as described in Cane et al, 1986, Nature) are used to "spin up" the ocean model from 1964 until the beginning of the forecast period; the atmosphere model is similarly spun up using the simulated SSTA from the forced ocean run. Thus initialized, the coupled model runs ahead in time with no data insertion, producing a forecast for the evolution of the coupled atmosphere-ocean system.

For the standard product, SSTA fields from individual forecasts are composited by lag-averaging the results from six consecutive monthly initialconditions. The composited forecasts for lead times of 3,6,9, and 12 months are then post-processed using a singular value decomposition analysis, to remove systematic spatial pattern errors and to map the idealized model domain onto the real one. The results are archived from January 1972 (verification time) to the most recent month available. An analysis of the results in terms of an ENSO (El Niño-Southern Oscillation) index is here. The forecast results are here.

For the LDEO2 product, a nudging algorithm as described in Chen et al, 1995 Science, is used to initialize forecasts. In effect, the ocean is spun up with a weighted average of FSU and model-derived winds. This scheme results in much reduced "initialization shock", considerably improved forecast skill (.1 - .2 increase in correlation at all lead times from 0 to 18 months) and reduced seasonality of skill. Fore the LDEO2 product, no lag-averaging is used, although small corrections for systematic errors are made using a value decomposition analysis. Results are available at all lead times from 0 - 18 months. An analysis equivalent to that above for LDEO2 can be found here. LDEO2 forecast results will be included ASAP