Thoughts on Experimental Models?

Mar 21, 2005
Kearney, NE
As a newbie, I'm curious about the track record of the experimental models out there. Are there some that appear to be more successful (and useful) than others? I realize that most models will be less accurate the farther out they forecast, but how do they perform as you get into Day One? I'd appreciate knowing how you throw them into the mix on chase day (or before).

Earl Barker's Experimental Products page is where I look at most of them. Are there other pages/sources for experimental products worth considering?

Also, I notice that there is an old and a new STP definition being used for separate STP models. Not sure how new the "new" Rich Thompson definition is, but I was curious as to whether there was any data on whether or not the new definition was outperforming the old one. There is definitely a difference between them on the model maps right now (check 48 hours out or so).
Hmmm. I _just_ realized that Rich Thompson of the SPC is a forum member!

As are Roger Edwards and Jon Racy (I believe), both SPC forecasters as well. Additionally, the parameters being questioned are not really experimental models, but just experimental parameters, whose components are from various models (RUC/NAM output). :)

At any rate, see "An Update to the Supercell Composite and Significant Tornado Parameters" from Thompson, Edwards, and Mead (2004) at . Rich has also addressed some of the changes in a current thread at . As far as statistics for STP and SCP, the only real numbers I've seen are in both the original publication on STP and SCP as well as the update linked to above. For other parameters, there are other threafds that include links to statistics/graphs evaluating the ability of various parameters (e.g. LCL, LFC, 0-3km CAPE, etc) to discriminate non-supercell, supercell, and tornadic supercell storms.

The "accuracy" of the parameters from any current numerical model output is entirely great, but of course that's all relative. Like other model forecasts, the forecasts for SCP/STP/others likely won't verify exactly, with, as you noted, inaccuracy becoming more severe with forecast time. These composite parameters may be ever more prone to inaccurate forecasting, owing to the fact that they include the output of several model forecasts (for example, STP includes several instability and shear parameters), meaning that an error in one of the components could significnatly effect the whole (STP in this example).

For the non-composite-type of parameters, such as 0-1km helicity and similar quantities, other problems arise as well. For the case of 0-1km quantities (VGP, EHI, SRH, etc), the forecasts are heavily dependant upon the accuracy of a relatively small atmospheric layer (0-1km), which is highly variable in both temporal and spatial extents. Additionally, for 0-1km SRH, the values obtained are a function of storm motion, and thus an error in forecast storm motion can yield highly erroneous values.
Originally posted by Darren Addy
I'm curious about the track record of the experimental models out there......Earl Barker's Experimental Products page is where I look at most of them.

For the record, I'd like to point out that these are not experimental models, but experimental products derived from operational models, namely the NAM, RUC and AVN. As for gaining a better understanding of what all these fancy products are, most are available on the SPC mesoanalysis page. If you then select a region, and then a product from the menu on the left, the main frame will have a "Description" tag at the top to the right of the named product. Clicking this brings up a short description - and at the bottom of the short description, if that isn't enough to satisfy your curiosity, is a link to a more detailed description, which typically is a paper describing the index and showing performance statistics.

Jon Davies gave an interesting presentation this afternoon as part of NSSL's seminar series regarding a new tornado forecast parameter he calls Effective Stretching Potential (ESP). His theory behind its development is to identify likely environments for non-mesocyclone tornadoes. From analysis of his archive of RUC-proximity soundings he has found a relationship between high 0-3 km CAPE and steep (> 7 C/Km) lapse rates from 0-2 or 0-3 km. Lots of experimentation needs to be done yet to test its effectiveness but it looks to have potential after some more refining. Most notably, researchers are concerned that this will be another dimensionless parameter not based on any physical fundamentals or meaningful units. Definitely an interesting concept and will have me paying more attention to intersection areas of high low-level CAPE and lapse rates that would otherwise be overlooked thanks to their high LCLs and minimal SRH. In almost all cases these environments were rooted along or nearby a boundary noted by strong surface vorticity.

More info can be found at Jon's website at