Outstanding Operational Forecast Challenges Board

Difficulties in Prediction of ENSO Phase Changes

     and Impact on Outlooks of 2006 North Atlantic Hurricane Season & 2006/07 DJF US Drought 


In addition to interdecadal variability/trend, the predictive ENSO condition is very important information for forecasters when making seasonal outlooks.  Usually, the prediction skill of tropical SST is high due to persistency after April for next winter forecast.  The persistence is low from fall into the next year past April.  Forecasters found that all tools are too much like persistence, according to which too late phase transition would also occur in the forecast, the events starting too late and then lasting beyond the time they should.  Following is a recent case, showing forecasts of ENSO phase transitions in late 2006 and early 2007 and the influence on the outlooks of 2006 North Atlantic hurricane season and 2006/07 DJF US precipitation. The focus is on the NCEP Climate Forecast System (CFS), though other dynamical and statistical models have similar problem as seen in the ENSO prediction plume graph produced routinely by the International Research Institute for Climate and Society (IRI).

To view the figure, move cursor on the link.

To enlarge, just click it!

2006 North Atlantic Hurricane Season Outlook


i)    The 2006 May outlook called for a good chance of an above normal season (Fig.1) mainly based on prevailing ENSO neutral condition (Fig.2 upper panel) and warming trend.

ii)   In early August forecast update, the interpretation of already less favorable atmospheric conditions continued to be possibly due to unfavorable intra-seasonal activity with at best a weak El Nino development (Fig.2 middle panel), hence reluctant to downgrade the earlier active season forecast (Fig.1).  

iii)   In mid September NOAA declared that El Nino has already developed and is likely to continue.  (Fig.2 lower panel)

iv)   The 2006 hurricane season was finally classified as a near-normal season.  (Fig.1)

2006/07 DJF US Drought Outlook


El Nino faded out rapidly in 2006/07 winter.

CFS ensemble forecast didn’t catch such change until January (Fig.3), causing problems in DJF precipitation forecast (Fig.4), which degraded the seasonal drought outlook.

(The above two cases were provided by Muthuvel Chelliah and Douglas Lecomte of CPC/NCEP.)

~~~~~ ~~~~~ ~~~~~ ~~~~~ ~~~~~ ~~~~~ ~~~~~ ~~~~~ ~~~~~ ~~~~~

Summary of Discussion Points

1. Systematic bias

The model behavior toward elongating ENSO events (both cold and warm) in time has also been seen in the past in the CFS hindcasts.  When the initial state is, say, cold, the CFS tends to persist it and misses the transition to warm. The same problem occurs in the opposite transition direction. Figure 5 from an analysis of CFS hindcasts shows how cold states and warm states tend to be overly persistent.

2. Cases exploration

Figure 6 shows Hovermoeller diagrams of SST along the equator in the Pacific for a case study of the 1983 event for two different models, namely, CCSM on the left and CFS on the right.  The CCSM hindcast ensembles have been run from real ocean initial states (MOM3 ODA) initialized with AMIP atmosphere and land initial states in Jan 1983. The OISSTv2 was used as verification for both CFS and CCSM and the systematic error of both models has been corrected, based on the cases in the set (19 cases for CCSM and 25 cases for CFS).  The two models each have 8 panels with the observed (verification) in the upper left, the ensemble mean in the upper right and six individual ensemble members below.  (The CFS picked the first six members from its hindcast set of 15 members.). This figure indicates that in the 1983 case, the CCSM did a more creditable job of predicting the transition not only by the ensemble mean but also by individual members.

Figure 7 is the same as Figure 6, except for the 1997 case with Jan 1997 ocean initial state.  The CCSM also did a fairly good job, although it underestimated the amplitude and displayed the characteristic excessive westward propagation of anomaly phase that is a systematic behavior of CCSM in ENSO events.  Since the six CFS members selected by the designated way could be underrepresented, an additional check of the latest 10 members ensemble mean was made and shown in Figure 8.  Again, the forecast held on to the initial cold anomaly too much and significantly delayed the onset of El Nino.

Figure 9 shows the CCSM predicted the transition from cold to warm in the middle of 2006.  This admittedly limited set of results suggests that the multi-model approach could be invaluable for the purpose of providing better information to forecasters whose concern is, say, six-month lead forecasts of ENSO that are used as guidance for hurricane and/or drought outlooks

3. Prediction obstacle

One should not fault CFS for not predicting a transition at all times. The plume diagrams of Figure 10 show that CFS did a fine job when the initial states passed the spring predictability barrier for the four years of 82, 83, 97 and 98.  However, the critical point of such prediction obstacle can occur in any seasons (such as that between July and August as well as that between November and December in 2006).  Besides, forecasts of weak events continue to be difficult by any verification scores that are a signal to noise ratio measure.  The weak El Nino of last summer could be just something with a name to blame the failed hurricane forecast on. There is more to it as other researches indicated.








(The discussion points were supplied by James Kinter of COLA/GMU and Huug van den Dool of CPC/NCEP)