Case 2: Forecast verifying 12 UTC 16 April 2002

 
Discussion

All of the answers are correct, but none of them alone are sufficient to really get a handle on what the ensemble forecast is telling us. We'll illustrate the desirability of examining each ensemble tool in the next four sections. You can compare the graphics in these sections to the 500-hPa and MSLP verification, which can be found here.

Ensemble means

The animated graphics here show ensemble mean forecasts from 00, 12 UTC 8 April and 00 UTC 9 April 2002 for 500-hPa heights and sea level pressure valid 12 UTC 16 April 2002, in three frames.

The ensemble means in panels one and two look pretty similar, with weak troughing over the west coast of the US and the intermountain West, respectively, and with weak lee-side troughing in the Great Plains. We can also see from panel one that the 00 UTC 8 April operational forecast was pretty far afield from the ensemble mean, with its low pressure area in the Great Lakes and high pressure over the Great Plains and intermountain west.

Panel three, on the other hand, represents a dramatic shift, particularly in the upper air pattern, with a fairly deep trough in the intermountain West and a strong downstream ridge over eastern North America. Gulf inflow is increased with a stronger southerly to southwesterly sea level pressure gradient. This certainly raises a medium- to long-range red flag for the middle of the country with respect to severe weather possibilities, although forecast skill at these time ranges is quite low.

Ensemble spread

The graphic below shows the ensemble mean and spread of 500-hPa heights for the northern hemisphere from the 00 UTC 9 April 2002 run valid 12 UTC 16 April. (NOTE: Current graphics may be found here.) The larger the spread as measured by the standard deviation about the ensemble mean, the warmer the shading. The color bar at the right gives values for each color.

Note that the spread is low in the ridges off the west coast and over the eastern US. Greater uncertainty exists, however, in the western US trough itself.

"Spaghetti" diagrams

Ensemble mean and spread may hide the existence of clusters of similar forecasts, in which case the ensemble mean may mislead the forecaster with respect to the most likely solution, and the spread is merely measuring the distance between two or more clusters of ensemble members. Spaghetti diagrams can help with examining the ensemble forecast for the possibility of ensemble clustering. The graphics found here show the 5640-m height contour at 500-hPa for each ensemble member, the ensemble mean, and for operational forecasts from around the ensemble run time.

In panel 1, we see that two of the 00 UTC 8 April ensemble members (n1 and n2) indicate a deep western US trough. Other ensemble members have a trough further west (p3, n4, p1, ensemble control), a weaker trough (n5, p2, p4, p5), or no trough at all (n4). The forecasts from 12 and 24 hours later verifying at the same time are also shown and, while very different from each other, generally fall within the range of solutions in the western US. However, in eastern North America, the forecast from 24 hours later shows a very strong ridge, stronger than any of the ensemble members from 00 UTC 8 April.

In panel 2 (12 UTC 8 April 2002 ensemble run, 192-hr forecast), a weak ensemble mean trough is on the west coast rather than inland over the intermountain West. This seems to be a net result of superpositioning of many ensemble members with short-wave trough axes from the eastern Pacific to the intermountain West. Only a couple of ensemble members agree on the location of the trough. The operational forecast in bold red presents an additional solution that is faster than almost all of the ensemble members. Over the eastern US, the forecast of a potential trough is highly uncertain, though 4 members support it to varying degrees. Some members have a ridge instead. The current operational forecast has a weak trough/ridge pair, while the previous operational forecast is rather flat and the forecast from 12 hours later is far more amplified than the ensemble mean or its members.

Panel 3 indicates the same dramatic shift as is shown in the ensemble mean forecast. Many members agree with the trough in the western US and ridge in the east to varying degrees of amplitude. The operational forecast is in the middle of the pack in the west, but more amplified with the ridge in the east than any of the ensemble solutions. We can also see that the previous operational forecasts from 12 and 24 hours ago are generally at the extreme end of the current ensemble forecast range.

The ensemble runs over the period 00 UTC 8 April to 00 UTC 9 April emphased different forecast possibilities from each other. This should not be surprising given

• The time of year (transition seasons are typically times of relatively poor performance and high run-to-run inconsistency for NWP models),
• The lead time to verification (7-8 days), and
• The flow regime (the global forecast model has difficulty with development of troughs in the western US).

The last point is emphasized in the final section below. Relative measure of predictability

Finally, we examine the relative measure of predictability for the 00 UTC 9 April 2002 ensemble run valid at 00 UTC 16 April. Recall that most of the ensemble members from that run indicated a western US trough of significant strength. However, agreement among ensemble members doesn't always mean a good chance of verification, as we see in the graphic below:

Note the blue colors in the Great Basin and Southwest US. These indicate that in spite of many members of the 00 UTC 9 April 2002 ensemble agreeing on a trough in the western US, the measure of predictability is only on the order of 20-40%, indicating that the current forecast's level of agreement among ensemble members, though appearing to be significant, is not enough to render the ensemble mean highly predictable. Additionally, the likelihood of a good forecast is only in the 10-13% range, at or only slightly better than the 10% climatological likelihood. Further north, the probability of verification increases to 39% or better, and the regime is highly predictable (90% or better compared to ensemble forecasts from the last 30 days). Also note the high predictability and relatively high probability of verification for the eastern North American ridge, especially in New England and eastern Canada.

Recall that the verification had a strong trough in the western US and an equally strong ridge in the eastern US, where an historic April heat wave was taking shape. The 00 UTC 9 April 2002 ensemble run, and the operational run from the same forecast cycle, were all heading in this direction at 168 hours, although the inherent predictability and chance for ensemble mean verification at this lead time were low for significant areas affected by these features.