OK, so I'm definitely not going to make a habit of tracking down EF ratings for tornado reports, if I have to poke around for them. Sometimes I have the time to futz with this, and sometimes I don't, so I have to set up the site such that it will run itself in my (sometimes protracted) absence. Anything that requires ongoing consistent manual labor on my part is just not something that I can commit to. So we're not going to weight reports on the basis of EF ratings. :(

 

The outlooks for 2010-05-07, 18:00 CST are a good example of how the density factor would work. There was only 1 tornado, in Ohio. It was fully contained in your poly (100% credit), and only 11 miles outside of mine, so I got 89% credit (since the credit fares from 100% in or on the poly, to 0% at 100 miles from the poly). And my poly was slightly smaller than yours.

 

The formula is (tornadoes inside/nearby) / (poly area)...

 

yours: 1 / 142138 = .000703541629 (~ .00070)

 

mine: .89 / 126084 = .000705878621 (~ .00071)

 

So this looks like it is correct, even though it's a rare case where I got a better score than you. :)

 

As concerns multiple areas of interest, you're right — there's no good solution if you can only draw one poly. Eventually I'll provide support for multiple polys. I was even thinking about having multiple polys, where you could assign different probability factors. So you could do a 2% poly, and a 5% poly, etc. Then you would get weighted credit for each poly. That would give you the ability to express your intent a lot more specifically. So I'll see what I can do. In the meantime...

 

Let's think about doing away with all of the existing penalties (poly area, perimeter, and dst missed). Rather, I think it should be like this:

 

(lead time bonus) + (density bonus) + (percentage inside/nearby bonus) = (score)

 

Sort the maps by density, and look at how it works out. You can see visually that the outlooks that really had the right idea score the highest, with tight polys that enclosed a lot of tornadoes, while the huge polys that missed most of the tornadoes — like mine usually :) — got bad scores. So density should definitely be the key factor.

 

The reason for the lead time bonus is obvious — it makes you gamble on model maps that are further in advance.

 

The reason to also factor in the percentage inside/nearby is less obvious, but if you think about it, it makes sense. The density factor already takes this into account, because that's events / area. But if that's all you had, people would just select a point in the middle of the highest activity area. That would be cool, but there's already a contest like that (the Virtual Storm Chase, which I think you participate in). So the difference with the Outlook contest is that you have to define the region, not just the highest activity point. So you still have to give extra credit for the percentage of events inside/nearby, forcing people to select a region instead of a point.

 

So I think that these are the factors to look at. Then, we just get rid of the poly area penalty, because it's already taken into account in the density factor. We also get rid of the perimeter and dst missed penalties. So if there are two different areas of activity, you just pick one, and neglect the other, and it won't hurt your score if you missed a bunch of events, nor will you have to draw a big or weird poly to enclose both areas of interest.

 

I'll work on getting this all hooked up, so you can see.