Case Study: Winter Storm Reconnaissance Program 2005, flight date 01/20/05

Part 1: Fixed observation time for increasing verification times

In order to gain a better understanding of the evolution of the ETKF signal variance (SigVar) I examined a 2005 WSR case where the signal due to the targeted observations appeared to remain coherent over a 144 hour period as it propagated from the western Pacific into the North Atlantic, north of eastern Europe. This case had been run for a 72 hour verification time for an eastern U.S. verification region. First I looked at the ETKF s predicted signal variance due to targeted observations (summary maps) and the ensemble spread(perturbation energy) to see the location and dynamic characteristics of the target regions this method produces. For this purpose I created summary maps for verification regions centered at 40N starting at a westernmost longitude of 120W 24 hours after the observation time and ending up at 0W 120 hours after the observation time. The verification regions were chosen with the assumption that the significant weather patterns would propagate at a speed of approximately 25 degrees in longitude per day. The verification regions extended from 25N to 55N in order to contain the general storm track latitudes.

At the observation time the most obvious weather features are a surface cyclone and an upper level trof co-located at 50N, 150W; a small cut off low or eddy that is apparent both at the surface and 500mb, located at 27N, 115W centered on the Baja peninsula ; a surface high at 40N, 115W; high pressure at the surface and an upper level ridge over Canada; a large anti-cyclone centered at 25N, 15W; and an upper level trof and surface low situated between the two areas of high pressure , just off the N.E. coast of the U.S.

For all verification regions and times there were two main target regions. The first is a large region over the northeast Pacific that extends from 150W to 120W and from 35N to 50N with maximum values in a fairly large area centered in the vicinity of 45N 130W and in a smaller area at about 37N, 155W. The larger region of maximum values is associated with the leading edge of the upper low and the frontal region of the surface cyclone while the smaller region is at the bottom of the trof on the downstream side. Both target regions coincide with the strongest pressure gradients. The ensemble spread is also large in this region, greatest in the eastern region. The ensemble spread and target regions in this location are most likely associated with uncertainty in the location of the frontal zone and the strength of the front as indicated by the southern extent of the trof.

There second main target region is in a much smaller area centered at about 25N 115W. This area is co-located with the absolute maximum of ensemble spread . This target region is associated with the low over the Baja. The large ensemble spread is this region is probably due to the difficulty in resolving such a localized feature. It is not clear what the dynamical importance of this area might be as a target region as the feature seems to be pretty much stationary. It is possible that this region may have importance as an energy source for downstream development.

The maximum signal in these regions is a maximum of 1.6 for a 24 hour forecast and decreases as the forecast time is increased to .8 for the 120 hour forecast. This would suggest that the impact of the sondes decreases with time as as more recent observations are assimilated. Also the impact of the sondes would be expected to be less for verification regions over the U.S and to the east since these areas have far more routine observations than the region over the Pacific.

The signal maximum in the region around 130W spreads out in latitude along the frontal zone 24 after the ensemble initialization then spread out longitudinally over the ridge and along the it's leading edge as the forecast time is increased. The signal in the smaller regions near 155W and over the Baja stay the same for all forecast times. It is reasonable to assume that the target regions in the trof and frontal zones would be dynamically important to the model analysis and subsequent forecasts since the location and strength of these dynamic features would be expected to impact the forecasts for downstream regions. The dynamic connection is not as clear for the Baja target region. For forecasts 96 hours or more past the ensemble initialization time there is a weaker target region at 35N 35W co-located with a stationary cut-off low and another in the subtropics at 25N 25W. located on a small ridge. The ensemble spread is moderately large in these areas. These target region are partially inside the verification regions. Perhaps observing these features would have an impact on later forecasts since they remain in the verification region until the verification time.

All of the ETKF target regions except the one near the Baja were co-located with the mid-latitude jet stream as inferred from the 200mb wind reanalysis. The jet stream appears to split into a mid-latitude and a subtropical branch just off the eastern U.S. coast. The south Atlantic target regions are associated with the sub-tropical branch. Only the westernmost target region was associated with the highest winds or jet streaks the other regions were located in lower wind regions downstream. The target regions are all positioned on the northern side of the jet. The largest target region was located in the exit region of an upper level trough and might be associated with uncertainty in the location of the jet stream and the possibility of growing instabilities.

There were two regions of high ensemble spread for the later forecast that did not coincide with target regions. One in a small pocket of low pressure off the coast of Maine and a second along the upstream side of a surface low at 65N 20W. Neither region would be expected to impact the verification region since the first is such a small local feature and the second is well to the north of the verification regions.

Except for the Baja region the ETKF picked out regions of high ensemble spread that would be expected to have a dynamic connection with the verification regions. Further investigation is necessary to determine whether the Baja region is source of energy for downstream development. It is encouraging that the this method did not target the regions of high spread that did not appear to have any dynamic connection. Also the decay of the signal with time was also to be expected, however a significant signal is present for later forecast times indicating that the ETKF may be useful in targeting for longer term forecasts.

The actual region chosen for targeting in this case was the westernmost trof region with a flight path that expended northeastward along the leading edge.

Prompted by a colleague's questions regarding target regions I produced a summary map for the easternmost region for a verification time that coincided with the observation time. This should give the effect of targeted observations on the background error. It was expected that the maximum signal variance would occur in the verification region since the targeted observation data does not need to propagate into the verification region. The summary map actually was almost identical to the ones produced for forecast 24 or more hours after the targeted observations. This would be surprising except that the verification region used was the easternmost region over northern Europe which is extremely well observed. It is reasonable to assume that the ensemble spread is very small in this area further observations would not improve the background state... Must run again for a different verification region !!!!