SmartModel and Severe Weather Ingest/Output Process.

[RMacDonald]

Hello all, per request and good suggestions, Hope this is the right section to post this. I want to explain how my model output works and how the graphics I post are built for future reference. To Start I ingest the NAM, GFS, WRF text data from Surface to 300mb. I take the current run and the previous run. Then it grabs the 24 history of surface observations, upper air (if available) and some PIREPS and ArtyMET. Then I compare the actual observed data with the previous and current model output to identify trends and how much each surface parameter was off. For example if the NAM was going for winds 120 at 8 knots but the observed wind was 190 at 13 kts. The model was off by -60 degrees and -5 knots. I do this comparison for Winds, Temperature, Clouds, Visibility, precipitation amounts, altimeter, present weather. Then based off of performance and establishing overall under or over forecasting of each parameter, I take that value and apply it to the current model run and then output the data into a spreadsheet, which is broken down hourly. These calculations take place every hour as I get new surface observation, Upper Air, and PIREPS/Artymet. So the output changes hourly, I only post when significant changes occur.

As for the severe weather graphics. I use the same NAM, GFS, WRF, and RUC SVR, current observation trends for temps, dew point and pressure, and Pireps. It compares 28 environmental indicies and synoptic conditions, and each is valued differents, and produces a threat value, with 1.00 meaning that ll the indicies and synoptic conditions are met for a significant Tornado threat. This is updated hourly as surface conditions changes and as weather reports come in. The graphical map outlines areas in Green (1-1.5" Hail, 50-60kts, F1 Tornado threat) that show at .3 or higher value, Orange (1.5-2", 60-75kts, F2 Tornado Threat) for .5 or higher, .7 or higher for Red( 2-3" Hail, 75-85kts, F3/F4 Tornado Threat, and .9 for Purple. (3"+ Hail, 85kts+, F5/F6 Tornado Threat). Every 10 days I compare how the values related to what the storm reports showed to fine tune the numbers. This is a general scope of what it does, any specific questions please let me know, still have plenty of work left to do on it. And would love any additional guidance or performance critique. Thx.

 

[rdale]

I'll leave the non-existing F6 classification to another post but what makes a .7 lead to the conclusion that we're in for a F4 tornado or 3" hail? I guess that's too much of a simplification to combine all those into on parameter. How does it verify?

 

[RMacDonald]

Either or on the F5/F6, it is minute at that point. The values such a .7 is just a first guess I will be able to refine it further once I get actual occurrences to what the values were at that time. I rely on shear values and forecasted synoptic conditions. tried to use case studies on past F4 torndoes to see what the environment was when it happend.

 

[Bill Tabor]

This sounds cool. I've often pondered creating my own model, or weather algorithms, or forecasting tools. Certainly this must be fun to play with especially for you as you create it and compare it to results.

What rdale mentioned also occurred to me as I read your description as generic severe parameters which indicate large hail may not always indicate strong tornadoes or even any tornadoes I believe based on the winds primarily.

What are you using to convert your coordinates into visual representations plotted on a map?

I now see your website. I'll head over to check it out cause I might like to get my hands on a running copy if that is how you have it set up, -or perhaps you just plot the graphics to the web for everyone. I'll take a look.

PS: It occurs to me you combine all the models for results but perhaps eventually you may wish to be able to toggle on/off various ones to see how it affects the results.

 

[RMacDonald]

Thanks yeah it is alot of trial an error. The graphic is the overall threat areas across the regions. The actual site specific cities forecasts break it down by convective gust, hail size, tornado/supercell threat. As for all the models, if one exceeds a correction value it is not used and will just use the remain model runs.

 

[Bill Tabor]

My guess is one particular issue for accuracy in determining threat areas will be based on the probability of precipitation. I've noticed for other composite parameter's developed by SPC and those of John Davies it is always an issue. No precipitation - then no severe weather so of course that then influences the probability of severe weather. I've noticed the new HRRR is nice because not only does it show and map the intensity of rainfall in an area and may include increased resolution to at times see discrete cells, but it also has a product called the 'probability of precipitation' which I think can be very helpful as another source to go to, another opinion if you will. I particularly like one of Jon Davies' tornado parameters because it takes into account 700mb temps when determining precip chances.

 

[Bill Tabor]

Additional notes / questions: I notice your website says Smartmodel can be run every hour as opposed to IS run every hour. Shouldn't the website always be updated for the latest as you are taking in current obs? I would expect to see it up to date similar to SPC's mesoanalysis otherwise it will obviously be full of errors when somebody pops over there to take a look. For instance I just checked and it is displaying March 5th for 03z but the time is currently almost 18z. Additionally I can't tell if that map is for just that hour, or if it is an 'outlook' for severe like SPC puts out for say a 12 or 24 hour period.

How does the whole timing with your model work?

 

[RMacDonald]

Thanks Bill for the opinion and critique's. Right now, I can only run it manually when I am at my Laptop or PC. I try to update it as many times as i can, until I can gain enough knowledge to have it run automatically on the hour. As for the graphic map, I will add a valid time on the reference map. The map references severe weather threat based on current hour Plus 6 hours out. The snowfall is for current plus 12 hours out. Again the whole knowledge of getting it to plot as I used the values on the map to manually draw the shaded areas, trying to gain more knowledge to automatically do this.

 

[Bill Tabor]

Ok, makes sense...thanks for the clarification. I look forward to the regularly running automated version. Will be interesting how will it predicts events.

 

[RMacDonald]

Thanks, I am would love the feedback if I hit or miss on the events. I will start doing statewide charts instead of US nation charts to add more detail in the analysis.