HRRR Composite Reflectivity

[Peter Wharton]

I've seen comments on this forum and other forums from time to time, noting that the HRRR Composite Reflectivity charts "show a supercell over" somwehere at "5 pm" - and drawing the conclusion that this should be a specific chaser target for the day.

From what I understand the state of severe weather forecasting is still limited to probabilities of severe weather over a reasonably large area - predicting with any certainty where actual storms will initiate and move is still out of reach. Targeting on any day is at best limited to a 100 mile by 100 mile area in the morning, with more precision as conditions develop, and final precision when you move to "visual/radar flight rules" late in the afternoon.

However I'm not professionally trained , so would someone here more qualified be able to comment - can a forecast model actually predict with precision the initiation and location of a single supercell at a point in time?? Or are the commentators using these chart outputs assigning too much precision to the prognosis? Are the models still just a probability statement of severe weather in an area?

 

[Jeff Snyder]

In my experienced the convection-allowing models such as the HRRR and the 3km and 4km WRF runs (either NSSL or EMC configurations that are available online) do a pretty darn good job at storm mode forecasts. Now, they certainly are not perfect -- we are still talking about numerical weather prediction, often out to as much as 36 hours in advance. However, compared to the lower-resolution, convection-parameterized models like the NAM (WRF-NMM at this time) and the GFS, it's a major improvement. The high-res models have also shown utility, in my eyes, when it comes to forecasting whether or not convective initiation will occur. Now, exactly WHERE a model has a storm doesn't really matter much to me, or at least certainly not in terms of pinpointing down to the mile exactly where a storm should be in XX hours. I think it's unwise to pay too much attention to the spatial location of convection in the model, at least at precision below ~50-100 miles.

Personally, I use the 1 km reflectivity (composite reflectivity will make storms look huge by comparison, by the very nature of composite reflectivity) along with various updraft helicity products. Of course, the latter is used to determine which convective "blobs" in the model are actually more likely to be supercells vs. other modes such as a multicell or "ordinary" cell.

Above all, everyone should remember that these are just TOOLS. Sometimes a particular model will have an incredibly accurate 12-24 hour forecast; other times, that same model may end up verifying terribly. When used with other tools, both of forecast nature (e.g other model forecasts) and of "nowcast" nature (observations, radar and satellite data, etc.), I think including the high-res models often can improve forecasts. Well, they've at least helped improve my own forecasts.

 

[Stan Rose]

Ive been watching the HRRR pretty closely this season, and i must say ive been extremely impressed, especially compared with other high-res models. It's definitely been out to lunch a few times (especially with initiation) but when it's been right, it's really been right, and more often than not it has gotten the basic idea pretty well. For example, for the OK tornado outbreak May 24, the HRRR not only pegged the initial (Canton) supercell, it also progged the line break that allowed the storms south of OKC to really dominate. That was the main reason my initial target was just south of OKC (and i should have stuck with that game plan, lol). Another example, just last week Jun 20 with the moderate in NE--the HRRR totally pegged the initial tornadic cell (coming out of Hill City). It was a bit too far north at first (as all the models were with the surface low) but once it caught on--it caught on! Not as good with the storms further east, but pretty darn good overall. It may be a ways off, but this is the first season where i can honestly forsee a time when we (SPC/NWS) are able to delineate a higher threat area within an outlooked region, based on these high res outputs.

 

[Jeff Duda]

I don't use HRRR simulated reflectivity forecasts to tell me exactly where to go on a given chase day, but I do use the product to hone in on my target region and/or check whether I'm thinking the same thing the model is thinking. Usually there will be a surface feature (warm front, cold front, outflow boundary etc.) that will provide the focus for storm initiation. Thus seeing where the model initiates storms is also one way of seeing where the model has the feature or where else it thinks convection may fire if it isn't right on the boundary. Thus, when chasing, that makes me think, "are there any other boundaries I should be paying attention to?"

Convection allowing modeling in 3D using real data is pretty new, new within the last 8-10 years of published scientific articles at least. It's something I will be getting into very heavily when I start my PhD work at OU in the fall. Even in my masters work at Iowa State I am already looking at the skill of the WRF using 3 km grid spacing in terms of convection initiation. The model has gotten initiation very well in some cases, with initiation being less than an hour off and less than 50 km off in distance (granted, some of the skill depends on how you define convection initiation), but very poorly in other cases. However, just know that the HRRR is a single deterministic model run and given the spatial variability of fields at high resolutions, you will probably begin to see more focus on convection allowing model ensembles that use probability of convection rather than a binary "yes/no" on convection in the future. In fact, some of that is already being done with time-lagged HRRR ensembles (see http://ruc.noaa.gov/hcpf/hcpf.cgi, for example). Also, as of the last AMS meeting that took place in Seattle in January, plans are already in the works for expanding the HRRR and SREF into some very high resolution ensemble forecast systems. Last I heard these were slated to come around in the 2014-2015 time frame. The future is bright and interesting for explicit convection allowing modeling.

 

[Peter Wharton]

Thanks for all the learned answers - this forum continues to be a valuable educational resource for me.

 

[Zack Hargrove]

Three specific days I seem to remember the HRRR being spot on were the super outbreak II, the eastern NC outbreak, and the Iowa outbreak on 4/9/2011. The Mapleton, IA tornadic supercell was almost exactly where the HRRR said it would be. The best way to approach the HRRR and the other Hi-Res models with simulated reflectivity is to use it as a tool. It is more important to read the sky, use your instincts as a forecaster (practice practice practice.... read read read), and get to know the surface obs and trends.

 

[Adam Lucio]

As has been stated, the HRRR model has really nailed certain initiation days like April 3rd and April 9th. There have also been times where that model shows a nice big supercell, causing a chaser convergence in whatever town it was shown to be near, only to have nothing happen [May 8th and 9th come to mind here.] So like any other model it has its good days and its bad days. To me it is just another tool, what matters more to me is whether or not it shows initiation at all.

 

[Zack Hargrove]

As has been stated, the HRRR model has really nailed certain initiation days like April 3rd and April 9th. There have also been times where that model shows a nice big supercell, causing a chaser convergence in whatever town it was shown to be near, only to have nothing happen [May 8th and 9th come to mind here.] So like any other model it has its good days and its bad days. To me it is just another tool, what matters more to me is whether or not it shows initiation at all.

Hehehe... the 5/8 and 5/9 days made me mad... we should have just had a townwide chaser football game in Murdo, SD.

 

[Mikey Gribble]

The HRRR has been failry impressive on occasions this year, but like Jeff said you really can't take any of that literally. I think the HRRR is fairly reliable with storm mode and convective evolution, but the timing, extent and exact placement of convection can't be trusted a lot of times. I use it more for getting a general idea of how things will unfold. The 4km WRF is pretty good in that regard too. Even with these newer tools predicting when/where storms will fire and how convection will evolve is still one of the hardest aspects of forecasting IMO.