I noticed when putting out my initial tornado warning, that TORP was highlighting a circulation slightly southwest of the more apparent circulation. Not sure if this is due to some setting I had toggled on, but I was a bit shocked to see that TORP did not highlight an area of very strong rotation.

Throughout this case, I did notice that divshear did a phenomenal job at detecting and highlighting circulations off shore. There were several times that I was made aware of a circulation from both the Az and Div Shear products before velocity.

There was one instance where Az and Div Shear swayed my warning decision. The above image shows that in reflectivity, a beam blockage kept imagery relatively weak. However, the Az and Div shear products are highlighting an area offshore. Because I saw this, I was inclined to investigate my SRM product, which showed strong rotation. I found this to be a very useful instance of the tools working to influence decision making.

-Fujitareno

I was impressed to see the evolution of AzShear and DivShear during a supercell cycling mesos. The classic AzShear dipole and DivShear clover appearance are evident to start, along with notable convergence along the leading edge of the rear flank gust front just beyond the clover. As the cycling occurs, two distinct dipoles form in AzShear, with the new one strengthening and the old one weakening. Meanwhile, the clover look in DivShear became less organized as it attempted to resolve two nearby circulations. While this process can be observed in base velocity, that may not always be the case and AzShear/DivShear proved helpful in visualizing this occurring.

TORP probabilities dropped a bit quicker than I would like to see as the cycle occurred. I don’t know if it struggled with tracking given the two nearby circulations both changing in intensity (in opposite ways – one decaying and one strengthening). In this case both weakened so the lowering probabilities were alright, but often a tornado could still be ongoing underneath the decaying meso (or rapidly spinning up under the new meso) so I would prefer to see at least medium probabilities continue a bit longer until the cycle is complete.

– Mr. Peanut

I am from an inland office, so my marine experience is limited. But based on a few couplets with persistent TORP probs, went for a warning. The TORP probs helped draw my eyes to problematic areas.

AzShear and DivShear did well with the circulations out over the water. It even did well where the radar power was greatly reduced

TORP probs have been oscillating quite a bit, which hinders confidence some. However, the first few cases of the experiment TORP was more steady. Maybe these oscillations are actually more common, and the steady probabilities for Days 1 and 2 of the H.W.T. were not the normal performance of TROP?

The SRM Vrot didn’t look that great, but the combination of higher TORP probabilities along with persistent mid-level meso led me to reissue a tornado warning at 0843z near Little River, SC. The TORP and AzShear help to augment this decision.  (see image below)

TORP continues to be useful for Special Marine Warning and waterspout conditions. Keep in mind, it doesn’t take as much to create problems over water than it does for land, so it is easier to have confidence. In addition, you don’t need to have a waterspout to create problems for marine, so having the marine warning out won’t result in false alarm problems like they can on land. AzShear and DivShear imagery seems to help “explain” TORP output as well. (image below).

I find it interesting that in this area of Z dropout from the tree blockage, TORP tends to increase and makes sense with the velocity data. I wonder what would happen if you removed Z and ZDR from the algorithm and only used Velocity, Sprectrum Width, and CC, and maybe KDP in the TORP learning.

This case has had the most persistent TORP probabilities above 80%. Is there something about tropical cyclones or environments that help to drive this and allow the algorithm to have higher confidence?

Here on Day 4, subjectively, this has been the most robust performance of AzShear and DivShear. Today, AzShear/DivShear signals and TORP trends have been very consistent. For example, the first storms that produce tornadoes and waterspouts had TORP with 50% or greater and had enhanced Az/Div Shear values. Subsequent tornadoes/waterspouts had very similar values. This repeat in values in the same environment helped to drastically increased confidence in these fields. These provided similar lead time to the development of notable Vrot in the V/SRM fields. Operational forecasters love persistence. And through 0938z, we have had that with the TORP and Az/Div Shear parameters.

Example below of how AzShear and TORP produced higher values, similar to previous storms, and storms behaved in the same way. This persistence increased confidence.

The increase in AzShear and TORP probs resulted in the warning issuance of at 0954z from Lumbergh. The vrot increased after the enhanced Azshear and TORP. Again, persistent behavior increasing confidence!  (image below)

One thing to note, is that I am not a tropical forecast office meteorologist. I have some Great Lakes experience. Therefore, my mental conceptual model of tropical cyclone tornadoes and supercells is limited, which at first may have hindered my ability to “calibrate” TORP to what I was seeing on base products and dual-pol. However, persistent signals are a forecaster’s best friend. But I do think it is important for meteorologists using these algorithms to have a conceptual model. If you don’t apply critical thinking to these, you will be chasing after a lot of false alarms.  These tools are very fascinating. But they take time to learn. The NWS currently has a paradigm where 4,000+ meteorologists are to have a baseline training in severe storms radar interrogation, as well as forecasting, hydrology, and Impact-based Decision Support Services. My fear is that if these tools (e.g. AzShear, DivShear, TORP) were to be deployed, it may get negative feedback despite positive feedback in HWT and OPG experiments prior to deployment. I feel that NSSL and the greater research community really needs to advocate for putting this work in the hands of dedicates subject matter experts in an operational environment. I think advocating for this will aid in expediting the advancement of these tools.

I find myself often lowering object filter to 0% to see every object. I can quickly flip through and triage this way, and also the higher probability show up as thicker circles anyway.
-dryadiabat

Using prob trend graphics to see upward trend in rotation. Going for lead time with the Tornado Warning based on persistence and trends.

-dryadiabat

Watching an area of deeper rotation immediately east of the radar site that does not currently have a TORP object. I was surprised to see it does not, even with object filtering at 0%.

-dryadiabat

UPDATE: It appeared with a history trend, but was absent from the display for a total of 11 minutes in PHI.

Explored the direction trend option and saw small changes. I thought this would be a good case to test its usefulness since storms weren’t deviant but changed path gradually with time given the larger scale circulation. If the degree range were narrowed, this would stand out more. Just a suggestion.

-dryadiabat

The future torp probabilities here do not match my conceptual model given the current probability of 78%. I would think probabilities would remain high in the near term, and then possibly steadily decrease later on. Here, they decrease through 10 min and then increase again at 15 min. Not sure why.

The above two screenshots were taken one scan apart. In the time series, the probabilities remain high, but the red torp circle drops off from the screen as if the torp probabilities rapidly tanked.

Not that it was really needed, but divshear did show a strong signal with this TDS confirmed tornado.

Not much time to blog, but overall thoughts are I find myself defaulting to TORP and prob trends the most. Did notice a strong divergence signature in velocity and here is how it showed up in DivShear.

-dryadiabat

I issued a Tornado warning based on a sudden spike in TORP via KHPX, which contrasted with lower TORP values on KOHX. The history is short, too. It’s broad but strong and increasingly cyclonic-convergent. I might’ve held off without TORP but we’ll see what happens. 88% to start and now up to 94%. Of note, the velocity pattern was way less concerning on KOHX, and again TORP from OHX was relatively much lower.

-dryadiabat

Noticed TORP is having a hard time keeping the same tracking number on the same circulation in PAR. There was a storm where there was a consistent weak circulation and TORP should have been able to keep the same tracking number the whole way. Attached are two screenshots that are one scan right after the other (1 min difference). To a human the same circulation can easily be seen, but TORP assigns a new tracking number. The KTLX TORP on the same storm was able to keep the same tracking number. This in turn changes the history trend graphs and made it more difficult to follow.

-Gary

Quick initial thoughts on the mesoscale environment with limited data to analyze. Satellite shows residual PBL stability present with billow clouds immediately ahead of the convection that has developed, likely tied to residual capping EML. However, surface heating is resulting in steep low level lapse rates below this although midlevel lapse rates are poor at the moment, they should steepen and large scale ascent should begin to help lift/erode the cap soon. As we crest the peak of diurnal surface diabatic heating and enter early evening, the low-level jet should increase, just conceptually given upstream mid-latitude system over the central Rockies. I’m guessing (without model data) that as the trough approaches the hodographs should elongate in both the low and mid levels, resulting in greater shear values and more organized convective cells soon.

-dryadiabat

We noticed that there is a lot of “noise” in the velocity data from PAR. I think Charles mentioned this is due to a different dealiasing scheme relative to what we’re used to in the 88D. We don’t notice the noise in KTLX. This is impacting TORP, with values as high as 37% indicated on the back side of the storms associated with this signal, that clearly is false. Analysis of PAR and 88D data and conceptual models from storm structure give us confidence these probabilities are far too high and there’s no immediate tornado threat.

-dryadiabat

Quick note to say in the early stages of storm development, AzShear and DivShear seem noisy and not adding a lot of value without more clear mesocyclones and storm-scale velocity features. I’m sure that will change as the case evolves.

-dryadiabat

Continued jumping around of TORP objects chasing the bad data quality velocity due to the dealiasing strategy (re: post above). Meanwhile, we’re monitoring trends in intensity and character of the midlevel mesocyclones which are more steady using base data analysis to understand storm organization and as an indication of tornado protential trends.

-dryadiabat

I’m not sure I could’ve identified storm top divergence with DivShear along if just looking at that product without looking at velocity. In velocity it’s fairly clear although only a small area of outbounds, whereas in DivShear is more noisy. With time, I’m sure I could learn to pick this signature out more easily in DivShear, but for now it seems relatively difficult for me.

Also, another limitation is sometimes the (-) and (+) components are separated by one elevation, and not on the same scan, so it may not be as apparent using DivShear as compared to just standard base velocity.

-dryadiabat

Highest TORP probs of the day so far in northern Grady County seem to be associated with a storm that has anticyclonic rotation. But this may be more associated with noisy velocity data since TORP isn’t set to detect anticyclonic rotation, if I understand correctly.  Unsure what to make of this. Top image is PAR, bottom image is KTLX (slightly higher tilt).

-dryadiabat

Really liking the temporal resolution of PAR, especially trying to analyze/assess the northern Grady County supercell which displayed some weak low-level rotation embedded within the front flank of the cell to its south. Evolution on rotation magnitude and width were more easily seen in PAR than KTLX, even with SAILS3 enabled.

This has made the Prob trend graphics more complete and useful too (see below):

-dryadiabat

TORP probs increased from 38% to 63% as the tornado report came in and about the time the Tornado Warning decision was made. It peaked at 79% when Vrot was strongest, but as Vrot decreased TORP values fell quickly, despite likely tornado still ongoing given character of the velocity couplet. Probs dropped completely and the TORP object disappeared at 0007z despite weak couplet still present. Tornado threat seemed to completely end shortly after, but perhaps persisted a minute or two longer after TORP dropped.

-dryadiabat

Unsure if it’s just the color tables or other factors but the AzShear seems more useful in PAR than in the WSR-88D.

-dryadiabat

Noisy velocity not associated with tornadic rotation has 38% prob.

-dryadiabat

The “one more scan” dilemma is easier with the rapid updates of PAR. Rather than waiting for a new full volume of data, or even another SAILS cut, we have quicker updates to see the evolution of the RFD.

-dryadiabat

Noticed 87% TORP but there is some bad velocity quality within the TORP circle separate from the weak couplet that I’m wondering if may be contributing to these probabilities, since the couplet is fairly weak. It is tight though so that may be the primary impetus for the higher probabilities.

-dryadiabat

Also noticed higher (+) velocity on the north side of the circulation in KTLX compared to PAR. KTLX at the top, and PAR at the bottom below.

-dryadiabat

Stronger couplet and better AzShear and DivShear signal noted simultaneously.

-dryadiabat

~30 knot Vrot in central McClain County on KTLX. Stronger inbound maxima noticed on PAR too. Probs on TORP seem to be holding in mid-range a little more than I would’ve expected. 55% peak so far.

-dryadiabat

This storm started with what appears to be side lobe  contamination. TORP did identify this feature and put out low probabilities. I can see this as a positive since it does keep probabilities relatively low for sidelobe. Curious to see how it would react if the sidelobe contamination appeared stronger?

As the storm progressed, a well defined meso formed over southern Michigan. TORP recognized this relatively quickly and began putting out 70-80% probabilities.

A CC drop was noted at 2119Z for this storm. Here is the Az and Div shear values at that time. Overall, not a great response for a confirmed tornado.

-FujitaReno