ProbWind in northern Alabama

At the HWT, forecasters working the warning desk in Jackson, MS noted an arc of storms in northern Alabama where ProbWind v3 was much higher than its v2 counterpart. They shared that the greater ProbWind probabilities and the fact that they received sub-severe LSRs, (~40 kt gusts) gave them more confidence that there could be severe-level reports soon. The NWS in Huntsville, AL issued a severe thunderstorm warning at 21:18 UTC and there were indeed trees down at 21:49 UTC, near Athens, AL.

Figure 1: ProbSevere v3 contours, MRMS MergedRef, and official NWS severe weather warnings for storms in northern AL.

Figure 2: ProbSevere time series for the storm highlighted in Figure 1.

ProbWind v3 produced greater and more consistent probabilities than v2 for this storm. A post-mortem analysis showed that the top-5 contributing predictors were:

1. MRMS VIL (23.6 J/kg)
2. 0-3 km lapse rate (7.6 C/km)
3. MRMS 0-2 km AzShear (7 x 0.001 /s)
4. MRMS 3-6 km AzShear (7 x 0.001 /s)
   
5. ABI+GLM intense convection probability (ICP; 69%).

The ProbSevere team has been in active discussions with HWT forecasters regarding explaining and conveying model predictions in AWIPS in near-realtime. 

ProbDryWind?

I was talking with forecasters from Rapid City and Grand Junction offices.  They often experience dry microbursts.  ProbWind often does not do very well with these type of wind events.  This is something us (as developers) are aware of and are working on improving.  The setup and ingredients for these type of wind events are physically quite different than their more moist counterparts.  One forecaster suggested instead of trying to incorporate it into ProbWind and adding to the ProbWind readout, to make a new entry in the ProbSevere family--ProbDryWind.  This is an interesting thought and something we will need to ponder and discuss.  The forecaster added, he couldn't envision a day where you would need both within a CWA--the environments are very different and a separate display/readout would reduce confusion.

-J. Sieglaff

ProbSevere Ingredient Distributions

Many of the NWS forecasters appreciated seeing ProbSevere 1-d and 2-d distributions of the data fields used to calculate probabilities.   Forecasters wanted to see these because they wanted to understand why ProbTor values may be higher one day versus lower the next.  Or why one day may have more 90%+ ProbHail values versus another.

One topic discussed was a 2-d NWP-based predictor in ProbHail (figure below).  The red box indicates a very favorable environment for severe hail (ratios near 10, which act to increase ProbHail values, representative of OUN/SJT CWAs on May 1, 2019) and the yellow box indicates a more marginal environment for severe hail (ratios near 1, which are neutral (do not significantly increase or decrease ProbHail values), representative of OUN/TSA CWAs on April 30, 2019).  This was one reason forecasters were seeing more 90%+ ProbHail values on May 1 compared to April 30, when similar values of MESH were observed.

Forecasters really appreciated seeing this level of detail as a mechanism to better understand ProbSevere and how to interpret and use probabilities from day to day.

ProbSevere-ProbHail 2-d predictor of hail cape (CAPE -10C to -30C) vs precipitable water (PWAT).  The red box represented conditions in the OUN/SJT CWAs on May 1 2019 and the yellow box represents the conditions over OUN/TSA CWAs on April 30 2019.

-J. Sieglaff

ProbSevere Suite Helping on Busy Severe Weather Day

A very active severe weather day continues over the NWS CWAs our teams are covering today (OUN, TSA, SGF).  A couple highlights from quick discussions with forecasters relate to ProbSevere suite of products:

• ProbTor is especially useful to draw attention to storms that might be rapidly increasing their tornadic potential, especially when busy interrogating many storms and issuing warnings.
• ProbHail on a storm over southern OUN CWA was indicating 75% probabilities, but the MESH was low biased (under an inch).  The storm was reported to have produced golf ball sized hail.  The forecaster viewed the high ProbHail value as a demonstration of the value of additional information included in the ProbHail model.  (The forecaster suspected low MESH due to updraft tilt, which is a known source of low bias.)
• One forecaster changed the ProbSevere double contour threshold from 3% to 10% given the very favorable environment for tornadoes in his CWA today (SGF).  The customizability was viewed very positively as that changed greatly decluttered his display.
• ProbTor jumps are proving very useful in highlighting storms with rapidly increasing tornadic threat (e.g., a storm in the SGF CWA had ProbTor jump from 37% to 74% in 8 min as the storm developed a TDS and had ground reports of tornadoes).

-J. Sieglaff

NWS Forecasters Data Usage and Access

An interesting point that has repeatedly come up, both during the EWP debrief as well as during discussions on the forecast desk, is related to how NWS forecasters view and access data pertinent to their jobs.  A number of forecasters have stated they often use the software GR2 Analyst to interrogate radar data and display ProbSevere data.  Additionally, forecasters have discussed how they often use the web for a variety of data--including NWP data and ProbSevere time series for storms of interest (Figure 1).

Figure 1.  ProbSevere time series for a storm of interest over Oklahoma available from the ProbSevere website

The ability to display ProbSevere time series information has been a common feedback point from previous experiments.  The ProbSevere team is working on adding this ability within AWIPS-II.  The group this week is excited to hear about the capability will be added to AWIPS-II for ProbSevere, however they also expressed some concerns.  Many forecasters shared experiences of AWIPS-II slowing down with so much data loaded and expressed the need and desire for capable displays being available on the web, etc.  While this is a larger topic that crosses the intersection of policy, user needs, and technology, among many other aspects--it seems forecasters often need to balance AWIPS-II data load (for usability/speed concerns) and using alternative data display methodologies (e.g., GR2, web-based, etc.) to achieve that balance.

J. Sieglaff

ProbSevere Double Contour

A quick post regarding the ProbSevere 'double contour'.  New to the ProbSevere AWIPS-II display is an outer contour that indicates the ProbTor value for a storm outside of the inner contour (the inner contour is the maximum of ProbHail/ProbWind).  This double contour idea came from the forecaster feedback from the 2018 HWT experiment--as a means to have both severe thunderstorm warning and tornado warning guidance in a single display.  Today, forecasters have very active weather, with many storms (Figure 1).  The default for the double contour to appear is ProbTor >= 3%.  Forecasters have stated today they appreciate the ability to change the threshold for the double contour to appear.

Figure 1.  ProbSevere contours from a forecaster's AWIPS-II screen.  The inner contour is the maximum of ProbHail/ProbWind for the storm and the outer contour is the ProbTor value (displayed when ProbTor >= 3%).

ProbSevere running hot in Iowa

ProbSevere has been running hot, with probability of hail and probability of wind, pulsing into the 30-50% and 60-80%, respectively for many storms across the DMX WFO.  Thus far no severe reports have been received.  In discussing the environment with the DMX team, we are wondering if the shear is too high for the amount of instability, which is shearing the tops off developing storms, preventing storm organization.  The tops being sheared off can be seen in 1 minute GOES-16 visible imagery (not shown).  One silver lining would be ProbSevere Version 2 probabilities are 10-40% lower than the ProbSevere Version 1--illustrating the better calibration of probabilities in ProbSevere Version 2 relative to Version 1 (although still too high for the small sample size of a few hours this afternoon in the DMX region).

-J. Sieglaff

KOUN Experimental Tornado Warning

The forecaster covering the southern Norman WFO noticed the tornado probability from ProbSevere Version 2 (All Hazards) rapidly increasing between 2114 and 2124 UTC Wednesday afternoon (Figue 1).  The rapid increase in tornado probabilities was due to rapidly strengthening low-level circulation.  ProbSevere here aided the forecaster to issue an experimental tornado warning--increasing lead-time and confidence in his decision making.  Storm chaser live video showed a tornado developed at 2139 UTC.

Figure 1. KFDR storm-relative velocity and ProbSevere Version 2 tornado probability valid 2114 - 2124 UTC 02 May 2018.

-J. Sieglaff

ProbSevere Aids in Experimental Tornado Warning

The forecasters in the Wichita, KS CWA noted ProbSevere tornado probability rapidly increased on a supercell thunderstorm that developed just ahead of a large severe squall line.  Figure 1 shows KICT 0.5 degree storm-relative velocity and ProbSevere probability of tornado contours rapidly increasing from 13% at 2048 UTC to 70% at 2056 UTC.  The velocity data shows a very strong, tight circulation.   The jump of the probability of tornado alerted the warning forecaster and after a very quick radar interrogation, he issued an experimental tornado warning at 2100 UTC.

Figure 1.  KITC 0.5 degree storm relative velocity and ProbSevere Version 2 tornado probability, highlighting the supercell ahead of the squall line valid 2048-2056 UTC 02 May 2018.

-J. Sieglaff

Spin up of another tornado from a long-lived supercell

A long-lived supercell thunderstorm transversed across the northern Wichita, KS WFO and into the Topeka, KS WFO Tuesday evening.  This supercell had exhibited ProbSevere All Hazards probability of tornado values that cycled with the intensity of the low and mid level circulation.  Between 2342 UTC and 2358 UTC the probability of tornado increased from near 50% to over 90% as both low and mid-level rotation quickly strengthened (Figure 1).  This storm produced at tornado at 0004 UTC 02 May 2018.  Forecasters have noticed with many of the long-lived supercells, the cycling of probability of tornado values related to cycling of mesocyclone strength.  This is giving forecasters good confidence in how to use the probability of tornado information in their decision making.

Figure 1.  KICT 0.5 degree storm relative velocity and ProbSevere All Hazards probability of tornado 2342 - 2358 UTC 01 May 2018.

-J. Sieglaff

Possible Tornadic Circulation in southern Dodge City, KS WFO

Early Tuesday evening, 0-1 km storm relative helicity had increased over the Central and Southern Plains.  The DDC crew noticed a rapid jump in ProbSevere All Hazards probability of tornado between 2328 and 2336 UTC--going from 10% to 70% in the 8 minute time period (Figure 1) in Comanche county, Kansas.  The forecast crew said the rapid increase alerted them to a rapidly intensifying low-level circulation and prompted an experimental tornado warning at 2339 UTC.

Figure 1.  KDDC 0.5 degree storm relative velocity and ProbSevere All Hazards probability of tornado contours 2328 - 2336 UTC 01 May 2018.

-J. Sieglaff

A few quick ProbSevere All Hazards Comments from a Forecaster

With such a busy afternoon of severe weather over the WFOs of interest, I was able to ask a forecaster if he had any comments on performance of ProbSevere All Hazards so far today.  He said, "it is performing very well and I am very happy to see ProbSevere All Hazards performing well with left moving supercells after storm splits, as often algorithms struggle with these types of storms".  He also added, "I was going to severe warning on this right moving supercell, and noticed ProbSevere All Hazards probability of tornado was up to 40%, while I wasn't confident enough to go tornado warning, I added the tornado possible tag to my severe warning".

-J. Sieglaff

Experimental Tornado Warning in Hastings, NE

One forecaster in the Hastings, NE WFO issued an experimental tornado warning based on his interrogation of base velocity data at 2137 UTC 01 May 2018.  The forecaster noted the ProbSevere All Hazards probability of tornado was low (9%) at the time he decided to issue the warning (2134 UTC), but then jumped to 40% in the following scan (see Figure 1).  The probability of tornado values continued to increase into the 60% range.  The forecaster commented, he realized the rotation increase he noted in base data took a few minutes to be captured in the MRMS AzShear fields and hence ProbSevere All Hazards.  In this case he said the probability of tornado confirmed his decision to warn and it gave him confidence in the probability of tornado, but to keep aware of the latency in base data into derived products.  This storm did produce a brief tornado shortly after the warning issuance.

Figure 1. KUEX 0.5 degree storm-relative velocity and ProbSevere All Hazards probability of tornado contours over Nebraska 2132-2142 UTC 01 May 2018.

-J. Sieglaff

ProbSevere All Hazards Alerts Dodge City Forecaster

One of the forecasters covering the Dodge City, KS noticed a rapidly developing storm just ahead of the main line around 2050 UTC.  The base reflectivity was only near 50 dBZ and MRMS MESH was less than 0.20", but the forecaster noted the appearance of a ProbSevere AllHazards contour of 48% at 2046 UTC.  The ProbSevere All Hazards probability then rapidly increased to 84% only 6 minutes later--due to rapidly increasing total lightning, strong satellite growth rates, and very favorable thermodynamic environment for severe hail (see Figure 1).  The forecaster said, "this caused me to interrogate the storm further using all-tilts and noticed a strengthening trend aloft".  The forecaster added, "ProbSevere All Hazards drew my attention to this storm, that combined with the ProbSevere All Hazards reaching high probabilities with other storms before MRMS MESH was exceeding 1.00" and getting reports of severe hail gave me the confidence to issue an experimental severe thunderstorm warning earlier than I would have using only an all-tilts interrogation of the storm".

Figure 1.  MRMS Composite reflectivity and ProbSevere All Hazards contours 2046 - 2052 UTC 01 May 2018.  The storm of interest AWIPS read-out demonstrates the jumps in probability of severe hail as well as rapidly increasing total lightning flash rates.

-J. Sieglaff

ProbSevere All Hazards Providing Lead-Time in Kansas

Severe thunderstorms explosively developed along the dry line over central Kansas Tuesday afternoon.  The HWT forecasters are centered on the Hastings, NE and Dodge City, KS WFOs.  Forecasters from both WFOs have noticed ProbSevere All Hazards was exhibiting rapid jumps to to probabilities in excess of 90%, while MRMS MESH values remained in the 0.50" to 0.75" range (see Figures 1 and 2 below).  The forecasters remarked they were pleased with the lead-time ProbSevere All Hazards was providing, despite the MESH still being below severe size (1.00").  The jumps in the probability of hail from ProbSevere All Hazards can be attributed to 1) intense GOES-16 vertical growth rates, 2) rapid increases in total lightning flash rates, and 3) the environmental parameters from Numerical Weather Prediction (NWP) used are extremely favorable for severe hail.

Figure 1.  Storm over Hastings, NE CWA at 2012 UTC 01 May 2018.  Notice ProbSevere All Hazards is 88%, while MRMS is only 0.37".  This example illustrates the power of utilizing other observations of deep convection, specifically satellite derived growth rates and total lightning flash rates, as well as environmental conditions from NWP data.

Figure 2.  Storm over Dodge City, KS CWA at 2014 UTC 01 May 2018.  Notice ProbSevere All Hazards is 95%, while MRMS is only 0.73".  This example also illustrates the power of utilizing other observations of deep convection, specifically satellite derived growth rates and total lightning flash rates, as well as environmental conditions from NWP data.

-J. Sieglaff

ProbSevere at HWT Spring 2018 Experiment

The NOAA/CIMSS Probability of Severe (ProbSevere) model is being evaluated by forecasters at the 2018 Spring Experiment at the Hazardous Weather Testbed (HWT).  For 2018 forecasters are evaluating an improved version of ProbSevere known as ProbSevere AllHazards.  The existing ProbSevere model provided the probability a thunderstorm would produce any severe weather (severe hail, severe wind and/or tornado) in the 0-90 minute timeframe.  Based on feedback from previous experiments, ProbSevere All Hazards provides probabilistic forecasts that thunderstorms will produce specific hazards--severe hail, severe wind, and tornado--in addition to the overall probability of severe (which is the maximum of the 3 hazard models).  Starting in January 2018 ProbSevere, both the version feeding NWS WFOs and the experimental ProbSevere All Hazards version are using GOES-16 data over the entire CONUS.  GOES-17 data will be incorporated into ProbSevere for the western US in late 2018.

Today forecasters are becoming familiar with the experimental products they will evaluate and developing AWIPS procedures they will use all week.  The initial WFOs of interest today are Lubbock, Texas and Cheyenne, Wyoming.  I will provide some examples of ProbSevere All Hazards via blog posts, in addition to characterizing some discussions with partipicants related to ProbSevere All Hazards.

-J. Sieglaff

Experimental warnings in PDT

Forecasters working the Pendleton, OR CWA this afternoon have noted jumps in GLM data and ProbSevere helped issue their first warning (Fig. 1). The ProbSevere jumped up form 26% to 46%, largely due an ENI total lightning jump of 19 to 41 flashes/min. ProbSevere later jumped from 51% to 76%, with MRMS MESH nearing 0.9".

Reports of 0.5" were received.

Fig. 1: ProbSevere, MRMS composite reflectivity, and EWP experimental warnings.

Later, a golfball report was received for another storm, further west. This storm had strong satellite growth rates, increasing MESH, (up to 1.2") and an increasing flash rate (nearly 30 flashes/min) combining for a ProbHail value of 93% (Fig. 2). A third storm was also warned on, but with no reports, currently. It should be noted that GLM data are non-operational at the time of this posting.

Fig. 2: ProbSevere contours (curved contours), HWT experimental warnings (yellow boxes), and GLM flash density (shaded boxes).

NWP fields in ProbSevere

A conversation with a forecaster working in the Goodland, KS CWA yielded some good points about the NWP fields in ProbSevere. ProbSevere currently uses data derived from the Rapid Refresh (RAP). ProbSevere uses several analysis/forecast hours to do a 'compositing' for most fields, usually taking the maximum over that time, followed by a spatial smoothing. It uses the F-01, F00, F01, F02, and F03 fields (i.e., the previous analysis, 'current' analysis, and the 1, 2, and 3 hour forecast fields). In this way, when new data arrives, it is 'centered' on the 1-hour forecast time. This is because RAP data typically has about a 1-hour latency from the analysis time. This temporal compositing and spatial smoothing help mitigate timing/placement errors of features or gradients in the NWP data.

The forecaster noted that it might be a good idea to put analysis time in the ProbSevere readout, so forecasters know how old the data are. This would also help inform forecasters if there was a jump in the probabilities, they could discern if it was just new NWP data being incorporated, or if the jump was from changes in the observed radar, satellite growth rates, or total lightning fields. We also discussed the advantages of grouping all the NWP fields together in the readout.

A severe storm in western Kansas, with hail reports of at least 1.75" diameter.

Testbed observations yields ProbSevere training point

The post about a severe wind gust in the Chicago WFO (http://goesrhwt.blogspot.com/2017/06/prob-severe-severe-outflow-wind-report.html) offers insight on a subtle, yet important point that should be communicated to ProbSevere users in the training module.  In this example, a storm produced strong outflow winds that raised away from the reflectivity core far enough such that it no longer was encompassed by the ProbSevere object.  As a result, the ProbSevere (ProbWind) probability was 0%.  For storms were outflow is expected to race away from the reflectivity core, users should be aware the ProbWind model may underestimate (possibly by a large amount) the probability of severe winds.  Additionally, this is another example of where maximum velocity magnitude could be a very useful observation for the automated prediction/detection of severe winds.

-Sieglaff

Severe hail from storm in Dodge City WFO

A severe thunderstorm had developed over the northwestern Dodge City, Kansas CWA Tuesday afternoon.  The NOAA/CIMSS ProbSevere values cycled between 45 and roughly 70% from 1925 until 1952 UTC; then radar observed reflectivity core rapidly increased, causing the ProbSevere values to correspondingly jump from 71% at 1952 UTC to 94% at 1956 UTC.  The jump in ProbSevere values as well as other features on radar (very strong storm top divergence and 3-body scatter spike (see post: http://goesrhwt.blogspot.com/2017/06/probsevere-compared-to-classic-radar.html)) warranted the DDC team to issue an experimental severe thunderstorm warning.  Shortly after the warning, 2.00" hail reports were received.

Figure 1. KDDC 0.5 degree reflectivity and ProbSevere contours valid 1926 - 1958 UTC 20 June 2017 for select times.

Another valuable discussion between myself and the forecaster was had where we examined the storm top divergence with this storm.  Strong storm-top divergence is something that often precedes a rapid intensification of a thunderstorm (and often severe reports).  Storm-top divergence is something a forecaster can quickly identify by analyzing all-tilts of base velocity data, but an automated storm-top divergence field is unavailable from the MRMS suite of products.  Such a field being made available in the MRMS suite could would be valuable for evaluation and possible inclusion into the ProbSevere model.

-Sieglaff