From the December 2012 issue of the JCSDA Quarterly
Figure 1.Northern hemisphere dropouts. Control analysis state of
400 hPa geopotential heights (black contours every 120 m) for initialization
of day-7 forecast dropout events on (a) 08 February 2012, and (b) 15 January 2012.
For each case, the normalized root of the squared error in control forecast 400
hPa geopotential height versus a verifying analysis is provided for the
24 hour (blue), 48 hour (red) and 72 hour (green) forecast, with arrows
depicting the movement of error-features. (c) Day-7 500 hPa geopotential
height anomaly correlation scores for control (blue) and experiment (red),
with verification of day-7 dropout events in panels a and b highlighted.
Figure 2. Southern hemisphere dropouts. Control analysis state of 400
hPa geopotential heights (black contours every 120 m) for initialization of
day-7 forecast dropout events on (a) 25 January 2012, and (b) 18 February 2012.
For each case, the normalized root of the squared error in control forecast 400
hPa geopotential height versus a verifying analysis is provided for the 24 hour
(blue), 48 hour (red) and 72 hour (green) forecast, with arrows depicting the
movement of error-features. (c) Day-7 500 hPa geopotential height anomaly
correlation scores for control (blue) and experiment (red), with
verification of day-7 dropout events in panels a and b highlighted.
Note: Dropout 'a' was chosen because the difference in AC scores was much larger
than others in this time period in the Day-5 and -6 forecasts.
The impact of the MODIS polar Atmospheric Motion Vectors (AMV) in global numerical models
has been historically neutral to slightly positive in mid-range forecasts, as measured
by northern and southern hemisphere Anomaly Correlation Coefficient (ACC) at 500 hPa.
However, several NWP centers have also found that the MODIS winds can improve the ACC
scores in some forecast busts, also known as dropout events. We have discovered additional
examples of dropout-improvement cases in our experiments to refine the quality control
procedure for MODIS AMVs. We describe these below and attempt to explain these improvements
in terms of flow regimes.
A two-month trial was carried out to test the feasibility of
applying the Expected Error
(EE) to the quality control of MODIS polar AMV in the hybrid ensemble/3D-VAR Global
Data Assimilation System (GDAS) and Global Forecast System (GFS) between January
and February 2012. The control forecast includes a dynamical rejection criterion:
winds are excluded when they express an innovation in zonal or meridional flow
greater than a threshold value, typically 7 m/s. This criterion was changed for
the trial forecasts to exclude winds with a ratio of EE to observation
wind speed in excess of 1.3367 - a value chosen (based on two weeks of assimilation data)
to exclude roughly the same number of observations as those rejected by the
original dynamical rejection criterion, while allowing for a level of context
sensitivity. Light winds are more likely to be rejected with relatively low EE,
while strong winds are more likely to be retained even with large EE.
The goal is to allow winds into the analysis that have a greater than 7 m/s
deviation from the background when the observed wind speed is sufficiently
large (e.g. winds sampling the polar jet).
A positive
impact was observed in some but not all forecast dropout events; performance on
non-dropout days was largely unchanged. Furthermore, it was observed that for two dropout
events in the northern hemisphere (Fig. 1c), in which one was improved while the other wasn?t,
both conformed to previously observed behavior for northern hemispheric dropouts: Errors
originate near the eastern end of the Pacific jet (blue contours in Fig. 1a, b), with errors in
the geopotential height field corresponding to the development of waves across the continental
United States downstream. In the improved case, the analysis initializing the day-7
dropout event was significantly modified in the Pacific jet region -
an area far-removed from the relatively few MODIS winds that were changed for
that particular analysis period. Modifications to the jet were smaller in the
analysis that did not lead to dropout improvement.
We hypothesize that modifications to the background produced by the experiment
may carry downstream over several analysis cycles. In a regime with enhanced
cross-polar flow (Fig. 1a) these perturbations may be able to cut across the arctic
and feed directly into the Pacific jet (see block arrows on Fig. 1a), creating
the potential for large impacts on dropout events. On the other hand, when this
cross-polar flow is suppressed, modifications to the background from the
experiment cannot efficiently influence this important region (Fig. 1b).
In the southern hemisphere, dropout improvement was observed when wave activity
was confined to high latitudes, where modification of the Antarctic environment
could have a direct impact (Fig. 2a). On the other hand, when low wavenumber
wave-activity is amplified, errors often originate in the deep troughs at
more equatorward latitudes and propagate into the Antarctic (Fig. 2b),
thus limiting the potential for polar wind observations to affect the
forecast bust. Once again, the flow-regime may be playing a role,
a finding consistent with other research on the subject that has been
done at the University of Wisconsin.
One last point: The experiments reported on here do not compare forecasts
with and without MODIS winds, but rather forecasts with different quality
control algorithms for the MODIS winds. In all cases the new QC algorithm
performed at least as well as the control algorithm, and, as discussed,
alleviated several forecast busts.
Several other dropout events occurred in the southern hemisphere
(Fig. 2c). We plan to perform similar analyses
on these to relate improvements (or lack thereof) in forecast busts to the flow regime.