Reliability of HRRR and short-term models

[Andrew Constans]

Well, I decided to chase SE Nebraska/SW Iowa for the 3-30 event. Every run of the HRRR had cells firing around Nebraska City and moving NE through SW IA, but those cells never initiated and resulted in a blue sky bust. How reliable are the HRRR and RAP? In contrast, the HRRR accurately predicted a linear mode storm system moving through the same area a week prior. Were there elements to the setup on 3-30 that these models overlooked or could not see?

 

[Jason Boggs]

Remember that they are just models, and nothing more. It's ok to look at the HRRR/RAP (we all do), but don't ever take them as written in stone. If every run of the model puts out storms, ask yourself why, and also ask yourself if they could be wrong. Dig deeper into your forecast and look at realtime data first, and then look at models.

 

[Jason Boggs]

Oh...I forgot to add that sometimes models get screwed up when there is early morning convection in the risk area. Not sure if that was the case in your area or not.

 

[Paul Knightley]

'All models are wrong but some are useful' is a good mantra to live (and bust!) by. What this essentially means is that no model can simulate the atmosphere perfectly but some will yield good information. The key thing is to make your own forecast based on a detailed analysis of the current data and an assimilation of the various model output - if you can't find a reason that storms should form where models simulate them then you have to question why the models have simulated them.

 

[Quincy Vagell]

The HRRR is one of many tools to use with severe weather/convection, although it is not always correct. While it tends to handle convection fairly well, there is often a bias to over-convect/prematurely initiate storms. This seems to happen quite a bit during the peak season (May/June), as perhaps it is heating the boundary layer too fast and/or it has more to do with the convective scheme within the model. On the other hand, it has also initiated storms in the wrong areas at the wrong times. For example, the HRRR picked up on the Pilger area supercells several hours out, but the HRRR struggled very much (recently) inside of three hours in Oklahoma on the day of the Tulsa tornado, with respect to convective placement.

I always start with the current surface analysis, mesoanalysis and latest observations. Use those tools and see how the models are initializing. Does the 1 or 2 hour forecast from HRRR line up closely with observations? How does forecast radar look compared to actual radar?

As with all models, although they can be useful, one must remember that meteorology comes first. Models can and are sometimes very wrong. Don't get suckered into modelology and relying too heavily upon the models. With that said, I use the HRRR a lot to get a general idea of what may happen. Watching run-to-run trends can also give clues to how a setup may be evolving. If the HRRR is very consistent with storm mode and convective placement, for example, and the model progs make sense based on observational trends, then I'll probably place my bets on the HRRR being close to correct.

The RAP is lower resolution, but I do like to look at it in the short-term. Assuming it seems to be initializing decently, the mesoanalysis RAP progs are very helpful to get a quick projection into the future.

 

[Larry Bell]

Don't forget to look up, into the sky.....I see so many people so concentrated on looking at their laptop/cell phones, they forget to look up. Models give ya a good "general" idea of where to position yourself to start, but look at the sky....you see towers building about 50 miles away....I'd work my way there

Sent from my SM-N920V using Stormtrack mobile app

 

[Jeff House]

I recommend making a forecast before looking at hi-res reflectivity output. Normally our group looks for the outflow boundary and dry line intersection. Sometimes we go warm front dry line. If the outflow gets south into more unstable air, and still quality shear, we go there. Only after establishing our broad thinking we look at hi-res output. Is it showing a hard right moving beast on the boundary intersection we are eyeing? If so great! Does the hi-rest stuff favor a different boundary intersection? Hey, maybe it just messed up the placement and we should stick to our thoughts. If after drilling down we feel the guidance has a point we may reconsider a new boundary intersection. Note, still an intersection. Fundamentals do not change.

All the while we keep up on surface trends and visible satellite. Boundaries will shift and we shift with them. Sometimes the hi-rest models catch up. Sometimes they choke. Either way hi-res should become more useful around 14-15Z, sometimes not until 16-17Z; earlier runs just do not have late enough data. Mid-morning and later runs also get more specific (not necessarily better) at local inflow environment and/or updrafts. Use more than just reflectivity simulation. Still just take it all as guidance. Chaser should focus on the boundaries and track them with great care. Yes, with a visual keep looking at the sky too!

 

[John Wetter]

Ensemble information is important. Also, I like to look at the experimental HRRR runs much more than the operational as the pace of improvement with the operational HRRR slowed a lot when it went onto WCOSS. I've found the CAM's to be outstanding forecast tools though, you just need to look at the entire forecast... Did it capture the morning convection, is it seeing the warm front/dry line/cold front/whatever surge or retreat like you're seeing in observations? They're incredibly useful tools but need to be used as a tool like everything.