Thoughts From Today's Look at UAHuntsville's SATCAST_v2 CI Nowcasting Product

From a look at the real-time runs of SATCASTv2 over the past few days, it appears that there are a few issues in terms of performance with respect to, both, false alarms and missed events.

False Alarms:
It seems that early in the day in
stable environments, the algorithm tends to flag developing boundary layer cumulus clouds, forecasting them to convectively initiate, when all they end up doing is growing up to the level of a lower tropospheric temperature inversion and stop development (see Figures below).




Sounding from Springfield, MO
valid at 1200 UTC, 05/20/2010.
Note the low-level temperature inversion.

















GOES-13 visible satellite imagery valid at 1432 and 1445 UTC, respectively. Note the rapid early growth of boundary layer fair weather Cu clouds (Above).











SATCAST_v2 CI Nowcast from 1445 UTC imagery (Above).


In some ways, this is discouraging (because they obviously aren't going to produce rainfall anytime soon), but in other ways it is encouraging to see that the system is sensitive enough to detect growing clouds, even in the earliest stages of growth. In order to reduce the number of false alarms in these instances, we've tried to tune some of the CI interest fields (remember, there are 6 of them used for current GOES, and 5 out of 6 of them must "pass" in order for a cloud object to be flagged for a CI forecast) so that they would become less sensitive to very early, lower level cloud growth.

Misses:
On the other hand, SATCASTv2 tends to have issues in highly unstable, uncapped environments, such as the "soupy" airmass that usually is encountered across the southeast U.S. in the warmer months. Often, in these environments, as was the case today (see Figures below), the algorithm becomes much more of a diagnostic tool than a prognostic tool, flagging cloud objects for future CI, right around the same time the corresponding radar scans detect echoes of 35 dBZ or greater... What we consider a "miss". Or.. It simply misses CI altogether. There are, perhaps, two issues that plague the algorithm in these environments:
1) As soon as surface-based air parcels begin to ascend in the unstable airmass, clouds grow very rapidly and become very efficient at producing rainfall, even with relatively narrow initial updrafts. Therefore, tuning the CI interest fields to become less sensitive to early cloud growth in order to decrease our false alarm rates (as mentioned above) actually severely limits us here.
2) Cumulus clouds in these environment sometimes possess very narrow updrafts that remain smaller than the current 4km resolution IR channel pixels can resolve (sub-scale pixels). So, the small-scale objects that we ARE able to track possess pixels that contain sub-scale growing Cu clouds that appear warmer in the respective IR pixels than the cloud tops actually are. There isn't really much we can do about this problem... It is a known limitation of our current GOES satellite instrument... but higher spatial resolution IR data in the coming years aboard GOES-R should dramatically assist with this issue.

SATCASTv2 Forecast at 1415 UTC
05/20/2010

Radar Base Reflectivity at 1416 UTC
05/20/2010



As a team, (myself... John Walker, Wayne Mackenzie, and John Mecikalski) are all working to embrace these challenges and to come up with solutions to these current limitations in the product. One potentially viable solution we may have is to create two sets of CI Interest fields... one set that is more sensitive to early cumulus cloud growth that can be used in highly unstable, uncapped environments (this would at least help with problem #1 under the "Misses"section)... and another set of fields that is less sensitive, to be applied in more stable environments where mainly boundary layer cloud development only is expected (this might help to decrease our number of false alarms).

Of course, we'd need some sort of intermediary input into the algorithm that would divide our domain into regions of "highly unstable and uncapped" locations and regions that are stable with some sort of low-level inversion to prevent much more than very early stage cumulus growth. Perhaps this can be accomplished with high resolution model data (the HRRR maybe)... but the better route would probably be to make use of vertical atmospheric columnar information generated from geostationary sounder data... the higher the vertical resolution the better (only then could subtle low-level inversions be consistently and accurately detected on the fly). Anyway, as reads the title, these are just a few thoughts and observations from the current experiment.