ProbTor model on June 28, 2017

A shortwave at 500mb and a negatively tilted trough at 800mb with a strong jet helped force severe weather yesterday along and ahead of a cold front in Iowa and Minnesota, and along a warm front in Wisconsin.

Figure 1: Surface analysis Jun 28 2017, 2100Z

Participants in the GOES-R Proving Ground / HWT Experimental Warning Program were able to investigate the NOAA/CIMSS ProbTornado model for mainly the first time this week. The image toggle below (Figure 2) is a synopsis of ProbTor for the event, showing correspondence between ProbTor centroids (boxes, colored by probability), NWS warnings, and local storm reports (LSRs). Subjectively, elevated ProbTor values corresponded fairly well with NWS tornado warnings and reported tornadoes. There are of course misses (e.g., in northwest Wisconsin [not shown in Figure 2] and northwest Missouri. There are also false alarms (e.g., one storm in central Iowa north of Des Moines, and a couple in southern Wisconsin). These false alarms had severe wind and hail reports associated with them, but no tornado reports.

Forecasters using ProbTor should always keep interrogating base radar data, but we hope ProbTor output can help give forecasters a heads up on developing tornadic threats and add confidence to their warning decisions.

Figure 2: Accumulation of NWS severe weather warnings (red=tornado, orange=severe tstorm), reports from the SPC log (red=tornado, green=hail, blue=wind), and ProbTornado storm centroids > 5% (filled boxes).

I created post-mortem time series of ProbTor and some of its predictors for several storms. Figure 3 is for a storm that traveled east of Minneapolis/St. Paul, MN, into western Wisconsin. You can see the ProbTor value increase to > 50% at 21:22Z, owing to a sharp increase in 0-2km MRMS AzShear. Total lightning activity stayed fairly constant (20-30 fl/min), after an initial surge up to 40 fl/min. The environment was good (35 kts effective shear; 35 kts MeanWind 700-900mb; 2000 MLCAPE; 200 m2/s2 0-1km SRH), helping push the probabilities to 30-50% for about 40 minutes. Two tornado touchdowns were reported.

Figure 3: Time series for a tornadic storm east of Minneapolis/St. Paul, MN.

The next three time series show the trends for storms south and west of Des Moines, IA. This environment is characterized by higher effective shear and MLCAPE, but comparable MeanWind and 0-1km SRH to the storm in Figure 3. The storm portrayed in Figure 4 attained a 40% value 16 min before the first report of a touchdown. The value then dropped considerably, as the AzShear values diminished. Forecasters have noted the value of seeing trends in the probability and predictor values for ProbSevere and ProbTor in AWIPS-2, and we will make that a future priority. Forecasters should also note that there is no magic probability threshold -- storms with 80% ProbTor and 20% ProbTor will produce tornadoes, as well as false alarms. The environment the storm resides plays a large role in the overall probability, so forecasters are encouraged to interrogate the storm that "sticks out" from the rest (e.g., 80% ProbTor near storms with 30-40%; 30% ProbTor near storms with 0-10%). Anecdotally,  this seems to be true for many cases, but of course not every case.

Figure 4: Time series for a tornadic storm west of Des Moines, IA

The next two storms had higher MeanWind 700-900mb and AzShear values, spiking the ProbTor values into the 80-90% range. Large increases and decreases in ProbTor may represent storm cycling. We are beginning to look at storm path length, width, and duration, and their relation to ProbTor values.

Figure 5: Time series for a long-lived storm south of Des Moines, IA.

The storm in Figure 6 below attained ~90% ProbTor 9 minutes before the initial tornado report. It is uncertain how long the tornado lasted. The ProbTor decreased markedly down to 25%, before jumping back up to almost 80%, with no additional reports. The character of the rotation evident in the radial velocity (and thus MRMS AzShear products) should play a key role in warning decision making. For instance, perhaps the second spike in AzShear, while still strong, may have been broader than before, and did not require warning.

Figure 6: Time series for a tornadic storm southeast of Des Moines, IA. 

Future work on the ProbTor model will also include pinpointing areas of rotation and assigning separate probability values, as opposed to just contouring the entire storm, since this doesn't adequately identify where the threat is, especially in squall line or messy situations.

EWP Week 2 End of Day 3...Thoughts and Feedback

Participants ended their day working in the Des Moines, IA and Kansas City/Pleasant Hill, MO CWAs.  A line of supercells developed this afternoon and tracked southeastward across Iowa and northwest Missouri.

• ·       GOES-16 ABI Imagery and products

o   Fire products were useful, good for doing DSS, more effective for messaging and briefing

o   Help sheets are good, really key

o   RGBs will like be even more useful for general, daily forecasting

o   Saw -66 to -77 values in the 1 minute IR imagery, unreal

o   1 minute imagery was also useful to see areas where they was a lack of convergence, knew the development wasn’t going to continue along the boundary, good to see relative to the major areas of population

• ·       GLM

o   Using every day, love it.  Spikes in lightning were evident, have confidence with the lightning jump that you could warn on the storm right away…every tornadic storm had a lightning jump and a cooling top

o   Like the gridded GLM best overlaid with radar or MESH, not with satellite

o   Still would like a centroid, a way to calculate the lightning jump or a ProbSevere for lightning

o   Like the GLM combined product, tried the 1 minute vs the 5 minutes…not sure anything was gained by using the 1 minute

o   Would not release to local offices until the mapping is fixed

o   For training, going to need case studies, lots of concrete examples instead of job sheets.  Provide procedures.  Could even do live webinars.  Felt the training module for this project was too slow, equation/laws information were too in-depth  

• ·       NUCAPS

o   Like the gridded versions a lot, there is some real potential value there

o   Gridded data is really blocky

o   Latency is still an issue

o   Could give confidence to your forecast, or give you pause…maybe need to re-asses your forecast

o   If you aren’t near an upper air site, the data has use even if it’s not perfect

o   Make sure it makes sense with its location in the AWIPS menu, and provide a way to know when a satellite pass occurs…otherwise it’s easy to forget to look at the data

• ·       ProbSevere

o   Why isn’t ProbTor higher than 90% when there is a confirmed tornado?

o   Trends were good, helpful to get a warning out right away…especially on shift changes

o   Weird times when ProbTor would go up, but the AzShear would go down

-Holly Obermeier

ProbTor92%

EAX above

 DMX below

ABI and GLM helped again in taking a look

 

 Here's an example where the the sandwich product and lightning jump on the GLM grabbed our attention to look at a storm in our northern CWA. We were focused on the middle and lower part of our CWA so this was a great situation awareness tool today.

Moisture/Dry Line Retreat

Dryline/airmass retreat in western KS/OK/TX. Benjamin

 Channel Difference

Derived Wind pictures

Worlds greatest upper divergence in far southeast NE? Flow around updraft? Flow outside of anvil into central IA?

Benjamin

speed max through west central Nebraska help storm development and sustenance. Upper level derived winds on water vapor.

Highest TorProb

We saw the highest values for TorProb so far at 71% with a storms southwest of Des Moines, IA. We actually tor warn long before this values was reached but it still gave us confidence in the warning we issued. Funnel cloud reported with this storm.

ProbTor Large area

Same area of storm for  prob severe parameters over 2 states and 2 cwfas. Different areas of circulation with a combined 6% prob tor.

Benjamin

added another probsevere with 3 circulations...1 prob for the whole things. 80+ percent.

Gidded NUCAPS - spotty coverage

500MB temperature from gridded new caps. Spotty nature due to lack of good data. Some areas of green dots have no gridded data.

Does the gridded data have to be contoured?  What about just an array of color coated numbers or something. The contoured analysis gets kind of goofy looking at times.

Benjamin

GLM

Here's looking at various tornado warned storms many with confirmed tornadoes. Just with today's storms we could see a correlations to tornado warned storms and lightning spikes.