Ensemble forecasts from 00 and 12 UTC 8 April, and 00 UTC 9 April 2002

The three ensemble runs from 00 UTC 8 April to 00 UTC 9 April show that the AVN/MRF forecast runs from this time period were all more or less within the envelope of potential solutions indicated by the ensemble spread. Additionally, the ensemble forecasts were all indicating more uncertainty than usual for the 12 UTC 11 April 2002 valid time.

Ensemble forecasts comparison to operational forecasts

We show the first point by showing the three consecutive ensemble forecasts for the 500-hPa 5640-m height contour in a three-frame animated graphic. The thin lines represent the 11 ensemble members, and the thick lines are the ensemble means and operational forecasts from 00 UTC 8 April through 00 UTC 9 April 2002.

Ensemble forecasts, ensemble mean, and operational forecast for 500-hPa 5640-m height contour

In frame one, the 00 UTC 8 April 2002 ensemble forecasts indicate a large amount of uncertainty regarding the location and strength of the western US short-wave trough. Some members resemble the operational forecast (see operational forecasts, frame one for reference) for 12 UTC 11 April 2002. Note that the ensemble mean indicates less trough with its axis further east than about 1/2 of the ensemble members. The other half of the members are faster and weaker with the trough; there seems to be a tendency for ensemble members to cluster around slow/deep and fast/weak solutions. The thick purple (06 UTC), blue (12 UTC), yellow (18 UTC), and green (00 UTC 9 April) forecasts in the western US generally are within the very wide range of solutions, but grouped with the faster, weaker solutions.

In frame two, the 12 UTC 8 April 2002 ensemble mean is faster and weaker with the western US trough than the previous ensemble run. The degree of uncertainty, while still significant, is considerably less than in the previous ensemble run. One member still shows a deep western trough and resembles the 00 UTC 8 April 2002 operational forecast in bold blue contour, indicating that there is still a small probability for such a verification. Note also the trend from 00 UTC (blue) to 06 UTC (purple) operational global model forecasts to weaken and speed up the western trough. Upstream over the Pacific, there is considerable uncertainty as to the position of the long wave trough and embedded short wave disturbances. The trend in the ensembles and in the operational forecasts is to position the axis further west to near the dateline, rather than at 170° to 165°W.

Frame three shows the 00 UTC 9 April 2002 ensemble mean with its western trough in about the same place as in the 12 UTC 8 April 2002 run. Note that the six and 12 hour prior operational forecasts are closely aligned with the new ensemble mean with respect to the trough position. However, the west coast ridge is now flattened in some ensemble members (which is trended toward by the consecutive operational runs), with increased uncertainty both here and in the fast Pacific flow upstream. The central Pacific trough seems to have better agreement among the members, however, with a position near the International Dateline.

To summarize:

• In almost all cases, operational model forecasts both previous and subsequent to the ensemble forecasts fall within the range of ensemble solutions.
• If we consider the operational forecasts an additional ensemble member (as we should), one can say that the operational forecasts over consecutive runs merely represent different ensemble members.
• The ensemble mean generally gives us a much better indication of forecast trends, probabilities for different forecast outcomes, and better verification than the operational forecasts.

Relative measure of predictability

Another tool that illustrates the large uncertainty in the forecast can be found in the NCEP Medium Range Ensemble forecast system graphics on the web (click on Relative Measure of Predictability link in table of Forecast Plots for the Continental US). We examine the relative measure of predictability (RMOP) for the ensemble mean forecast (click here for details on interpretation of RMOP graphics), valid at 00 UTC 12 April 2002, 12 hours beyond the time of interest. Since the predictability measure changes relatively slowly with forecast verification time, these graphics are a useful indicator of the predictability 12 hours earlier. The ensemble forecasts are from consecutive 00 UTC ensemble runs starting at 00 UTC 6 April 2002. Cool colors indicate areas of relatively low predictability, while warm colors represent areas of high predictability. Pay particular attention to the western US.

Measure of predictability for 00 UTC 12 April 2002 from consecutive global ensemble runs for 500-hPa heights

As we step through the consecutive forecasts in the animation, we see that:

• Features over Alaska, the central Pacific, the tropical eastern Pacific, and eastern North America are highly predictable.
  • At 144 hours, 48% of the ensemble forecasts with 90% relative predictability verified over the last 30 days, which certainly beats the climatological odds of 10%.
  • As the ensemble run starting time gets closer to the verification time, the 30-day verifications for high predictability increase, as expected.
  • At 96 and 72 hours (from the 00 UTC 8 and 9 April 2002 runs which we have been discussing), the ensemble mean forecast in areas with 90% predictability verified 72% and 79% of the time, respectively. These are shown on panels 3 and 4 in the animation.
  • At 24 hours, the value is 92%.

• The western US (where operational forecast from 00 UTC 8 April 2002 showed an amplifying short-wave trough), the area between the tropical east Pacific ridge and Alaska low (where same operational forecast showed an amplified upstream shortwave ridge), and parts of the east-central Pacific (upstream long-wave trough) have low predictability.
  • The ensemble mean forecast shows less amplitude to the features of interest than the 00 UTC 8 April 2002 operational forecast (animation panel 3).
  • For some areas, very few (in some cases, less than 10%!) of ensemble forecasts over the last 30 days had as few ensemble members in the same climatological bin as the ensemble mean. These results may indicate either/both of the following:
    • A wide spread exists in the individual members in these areas without any groupings around multiple solutions.
    • Two or more clusters of forecast solutions exist in the ensemble members, in which case the ensemble mean can actually be misleading.
  • Further examination of the ensemble forecast through the use of "spaghetti diagrams", for example (see previous section), can help with these more subtle interpretations of the relative measure of predictability.

• For the 96-hr ensemble forecast whose operational counterpart indicated the deepening trough in the western US, only 20% or less of the previous 30 days of ensemble forecasts exhibited less predictability than this forecast. Conversely, 80% or more ensemble mean forecasts were more predictable.