Models: signal or noise?

Todays forecast challenge had to do with storm location and initiation. From the TX panhandle northward thru KS into NE and CO. It seemed fairly evident that storms would form in this deep and relatively dry boundary layer (at least for May). With the PBL depth approaching 3km, models were not shy in breaking out storms to enjoy the roughly 40-50 kts of vertical shear and lovely quarter circle hodographs. It was relatively obvious that the PBL would control where and when storms broke out.



During the overnight and morning, a strong (Bonner class 3 at AMA) Low Level Jet was transporting moisture from DRT thru AMA into DDC. Models were consistent with increasing dew points just up until peak heating before drying out again at the surface. This process would lead to the maintenance and slight production of CAPE, albeit relatively dependent on your model grid spacing and PBL physics. We took a look at the 1km vertical velocity fields from the CAPS control member and it showed a rather robust set of horizontal convective rolls in the dry air leading up the dry line.

Now here is where the story got interesting. The 00 UTC ensemle was not shy with storms in the TX panhandle but the 12 UTC ensemble was and the corresponding control members emulated this. In the 12 UTC simulation the rolls were absent in large sections of the northern panhandle and there appeared to be two different PBL regimes. I am no PBL expert but I didnt notice anything different, so we will leave this for future investigation. However, further north in KS we could relate the PBL features at 18 UTC to observed visible imagery cloud features.

Furthermore storms initiated on the strongest of roll circulations, something that would appear to be necessary to reach the 3km LFC height before realizing much CAPE. While rolls appeared in the warm sector, they also were partially resolved probably arranged at roughly 4-5 delta X while in the dry air they were 7-8 delta X. WRF has been shown to filter kinetic energy to about 7 delta X.

It appears we did well in anticipating storms though timing of initiation continues to be a large issue. There is a big difference between simulating realistic looking features, relating it to observations, and having that same forecast verify with storm initiation timing.

I let the CAPS data anchor our discussion/briefing with the EWP forecasters. But a focus of the conversation was the relative uncertainty. Would there really be a gap in the TX panhandle? Would the dry line generate a few storms all at once or generate them in rounds? How many storms would initiate?

Of course as 4pm came along, confidence was still in the middle since cloud bands that formed around 1pm had yet to yield storms. But soon thereafter they did. One fundamental challenge in severe storms forecasting always relates to how much moisture you need to initiate storms. Today it was down in the upper 40's with temps near 80F. 80F sure seemed like a stretch in the morning and I held on to that reasoning throughout the day. Splotchy 81F areas along the dryline may have been the extra bit to get storms going, but thats just a bit of speculation.

The threat for todays supercells appeared to be a hail/wind mix in the south (according to the max 10m wind speeds from the ensembles) and more hail in KS. We have been playing with variables for a while now using vertically integrated graupel and new this year is 0-5km vertically integrated graupel. It turns out these variables are strongly related to Updraft Helicity (via the updraft). The hope was that if graupel was falling out, there might be some bottom heavy cores that may lead to at least a proxy for actual hail. Stay tuned as we investigate this further.


Yesterdays forecast in North Carolina and Kentucky did exhibit some signals of stronger storms with signals of 10 m/s updrafts, Updraft Helicity tracks, and decent graupel from the models. Of course, seeing these signals and interpreting how they relate to what may actually occur at the ground is of course a great challenge, especially in an uncapped atmosphere!

Models can be aggressive in producing storms but these forecasts are still just realizations of the risk given a certain environment. We rely on good environment forecasts, then move on to interpreting storm forecasts but the uncertainty is still high. Its high because we dont know when or how to trust in the number of storms (integrated over time) being generated and the timing of such storms. The predictability of the storms is much less than the predictability of the storm environment. Ultimately making better forecasts hinges on our ability to understand as much about model capability as model skill.