Use of Ensemble Anomaly Data
Computer models will, at times, attempt to forecast rather extreme
events – anything from a major winter storm or flooding rains, to
tropical cyclones and extreme heat. Sometimes these occur, but often
they do not. How can a forecaster get increased confidence that a
model forecast of an extreme event is correct? Anomaly fields can
help forecasters put certain model-forecast weather events in perspective.
These plots were used with success operationally early on during the
August 6-10, 2007, heat wave.
The image to the right (from the Pennsylvania State University’s “Eyewall”
website at
http://eyewall.met.psu.edu/ensembles/) shows the mean
850 mb temperature (green contours) from the Short
Range Ensemble Forecast (SREF) modeling system
(see
http://www.hpc.ncep.noaa.gov/ensembletraining/ for details).
An ensemble mean solution is essentially the average solution from
several different independent solutions (often called the ensemble “members”).
The SREF includes members with differing initial conditions as well
as members from different model packages. The result is the
Short Range Ensemble Forecast (SREF) system. By using ensemble
mean forecasts, rather than just using one particular model
(which may or may not be correct by itself), the forecaster is
essentially seeing the solution produced by the greatest number
of model members, which oftentimes makes it the most likely
solution. Those member solutions that are “outliers”, or greatly
different from every other member solution, are dampened out or
eliminated by viewing the ensemble mean. Ensembles do not and
are not intended to produce a specific YES or NO forecast.
Instead, they should be used to see a range of possibilities
and the most likely outcome.
Also shown on the above map is the mean 850 mb temperature’s departure
from the normal value for this time of year (look at the scale on
the left). This departure from normal is given in terms of
standard deviations above normal, which in very simple terms
is a measure of how unusual a certain value (in this case,
850 mb temperature) is as compared to what it should be this
time of year (see
http://www.robertniles.com/stats/stdev.shtml
for an expanded yet basic explanation of standard deviation).
The higher the number of standard deviations above or below
normal, the more unusual the forecasted event is. In this case,
the 850 mb temperatures forecast by the SREF were near 22 C,
2-3 standard deviations above the normal – in other words,
highly unusual, or anomalous.
After viewing this data, the forecaster might assume one
of two things: (1) the model is forecasting an unusual
event and is totally wrong, or (2) the model is forecasting
an unusual event and is correct. Even with the filtering out
of individual member solutions that differ too greatly from
its fellow member solutions, the SREF mean temperature was
still a large departure from normal. This gave forecasters
the confidence to believe that this very unusual solution
was probably correct.
Forecasters were also able to look back to past extreme
heat events to see how this particular event would stack up.
The map to the left (taken from the Hydrometeorological Prediction
Center’s Reanalysis data, at
http://www.hpc.ncep.noaa.gov/ncepreanal/)
shows the atmospheric pattern in place at 850 mb on August 18, 1988,
when RDU last reached its all-time record high of 105 degrees.
On this day, the 850 mb temperature (yellow contours) were 2-3
standard deviations above normal (image; scale on the left) – similar
to the SREF mean forecasts. By using the forecast anomalies, and
by comparing them to a past event, forecasters had increased
confidence to forecast record-breaking high temperatures during
the heat wave. Indeed, high temperature records were broken on
August 9 and 10, both with highs of 104°F.
|
|
