© Charles Chandler
I'm going to keep "My Great Title" as the subject line on my first post here, because I like it.
I'm still participating in the contest, and I enjoy it. It would be cool if we could get more people to participate somehow.
I just had one question right now. Is there a slight discrepancy in the map and the actual coordinates of a box you draw and posted tornado reports? Each time I draw an outlook, the actual point that is plotted is slightly north just slightly west of where I clicked to plot a point. Over time, I have learned to click a bit southeast of where I want my points, or manually alter the coordinates for a point such that it is in the correct location on the map.
That said, even if my outlook is where I want it to be when I see it on the map, I do not know, unless I plot the exact coordinates somewhere else, if my outlook is where it should be, and thus whether tornado reports I may expect to be inside my box are actually outside. Tornado reports do not seem to plot correctly on the map either (again, plotting a bit southeast of where they should be).
For example, on my most recent outlook, valid Monday, April 4, my outlook when viewed on the map clearly extends south of the Mississippi coastline, as I intended. A tornado was reported very near the Mississippi coast (see SPC's reports on their website) near Ocean Springs, MS (see report at end of this message). This is clearly on land, but the Tornado Outlook Contest maps show the report to be in the Gulf of Mexico, off the coast and outside my outlook box. I'm not sure if the map is incorrect, or the report is plotted in the incorrect location, but this is a report that should be in my outlook (on land) but is in the water instead. How is this happening?
0145 5 E OCEAN SPRINGS JACKSON MS 3040 8872 ROOFING PEELED OFF OF A COUPLE METAL COMMERCIAL BUILDINGS IN FOUNTAINBLEAU AREA.WINDOW BLOWN OUT OF TWO HOUSE. LARGE SECTION OF TWO FENCES BLOWN DOWN. SEVERAL MEDIUM TR (LIX) (http://www.crh.noaa.gov/arx/Stormready/WFOlist.php)
Thanks for the fix. I'm not so sure that it didn't move past outlooks themselves, but that's a minor issue. For example, my last outlook for April 4 I know was south of the Mississippi coast. While the tornado report was moved north to the coast, my outlook box was also moved north to not include the entire land area of the state. Again, minor issue, but if it's fixed for future outlooks, then I won't complain too much.
I don't use the Unisys maps. I have seen them on there, and looked at them, but I don't use them at all. I generate my outlook area based on different places I look at (I weigh heavily on the SREF I told you about earlier) and have that in mind before I go to this site to draw an outlook. Also, if the pick-points aren't exactly correct, that's no big deal either as I have been manually altering the coordinates to correct for this when I do the outlook. So don't worry about that, at least for me.
I've encountered at least one such outlook myself that you mention with a tornado on the other side of the country, notably one in which there was a weak tornado in Oregon (tornadoes here in the Pacific Northwest are rare and always weak!) and I got a high dst missed score. Disregarding a certain number or percentage or tornadoes may fix that issue, but may cause other issues (?). I've had some thoughts about scoring change proposals as I've gone through this, but I'll have to remember what they were. Many of them though were "this doesn't seem right; I'll have to think of a way to make this make more sense" - so that I didn't have a specific change yet in mind.
Actually, limiting the distance is a good idea... maybe even less than 1500 miles away, especially if it's in an east-west direction, because of the different system thing and the fact that you can only draw one polygon.
If this was a contest really intent on predicting tornado locations, there would also be negative points for tornadoes that occur when no outlook was made at all, but I know this is no such contest, and many people don't dedicate their lives to it either! haha. Take this week for example - tornadoes are occurring across the southern Plains and southeast today and tomorrow, but I'm at a conference this week and have had no time to look at this recently. I could do an outlook for tomorrow, but I have found it more beneficial point-wise to do the outlooks further in advance due to the lead-time factor.
One thing I did think about was single tornadoes vs. outbreaks. If you draw a small outlook polygon and a single tornado falls in the area - for example, my February 1 polygon if it was moved one state east, or my May 7 outlook from last year, you get 100% of the tornadoes in the polygon and a very good score with a small area. Fine, that makes sense. But lets say you expect a large number of tornadoes, like an outbreak. If it's something like the Super Outbreak of April 3-4, 1974, you would have had to draw a large polygon from the Gulf Coast to Michigan to capture every single tornado, and you would get a lot less points because your polygon was so big. But it had to be big. Even with not so large an outbreak, take my most recent polygon from April 4. I expected a significant event, and I got the main area of tornadoes pretty accurately, but in order to have captured all of them, I would have had to draw a fairly large polygon from Mississippi/Louisiana north to central Ohio and east to North Carolina, a polygon that would have covered parts of at least 11 states and likely a few more unless I knew exactly where to cut the line to the west to avoid Indiana and Arkansas. I have to wonder if the points lost for increase in area and perimeter would have been worth the points gained for percentage of events and dst missed. Probably, but even so, it's really tough to draw such a large area knowing you are starting at at a lower value.
On a similar note, I'm guessing it's probably a lot of work to work tornado strength in the scoring. Many of the Enhanced Fujita ratings aren't determined until NWS survey teams go out into the field to do the surveying, I know, and are often not added to the local storm reports themselves. But there could be one isolated "weak" EF0 tornado within a small outlook polygon that could give a very high score, while in an outbreak, you could have drawn a polygon in an area that captures every single EF2, EF3, EF4, and EF5 tornado, with some weaker ones missing the polygon (for example), that would score lower even though you were able to pinpoint the strongest tornadoes during a large outbreak that may have been difficult to cover the entire extent.
That said, I worked for the National Weather Service briefly in the past, and from experience I know that though most are, not 100% of tornado reports that are sent out are 100% reliable. There are also multiple reports sent out for the same tornadoes, especially the longer-track ones, near larger population centers, or where several chasers have converged on the same tornado. There are tornadoes that go by with no reports. There are funnel clouds that never touch the ground, or maybe even barely come out of a cloud, that are called in by the public to NWS offices as tornadoes, and thus are sent out as reports (those are not too common because of the training of storm spotters, however). Of course, most of those are things that wouldn't affect the scores too much, and are things I try to take into account when I do outlooks (for example, on my most recent outlook, I intentionally cut out most of the entire southern Louisiana marshland/Mississippi delta south of the New Orleans area because of the scarcity of population there).
Those are some of the thoughts I've had. When I remember more I'll include them.
Most of the time you might have to visit the websites of the National Weather Service area where the tornado occurred, following their survey of its track, to find out the rating. That can definitely get cumbersome, and not all offices do this for all weaker tornadoes. Sometimes they will also go back and add it to the text of the storm report, but that's unfortunately not always the case. At some point the tornado ratings do get to SPC, but I'm not sure if there is a place to find that data; I have not yet found one. Here's an example (from April 4 this month) of EF-ratings that have been added to many, but not all, of the tornado reports if you look at the "Comments" column: http://www.spc.noaa.gov/climo/reports/110404_rpts.html. Even some from last year haven't been added. One comment about basing everything on major events (EF2+) though - even if you were to do that, I think past polygons would have to keep all tornado reports in the scoring because the polygons themselves might have been created differently if the forecaster knew that only the stronger tornadoes would be counted.
Question about the density factor - how do you think it might be used in the scoring? And does it only count the events that fell in the polygon, or does it count those outside it as well (maybe you answered this that it counts those outside it within the buffer given partial credit down to 0 at the blue line)? I've look at examples such as yours and my polygons for May 7 (and same time) last year and try to figure out which would get the higher score if the percentage factors are replaced by density, since your outlook polygon, which didn't include the one single report, has a higher density (albeit only slightly) than mine, which did include it.
I have another qualm about the density factor - well I guess even the way things are right now. What if you think there are two areas of possible tornado maximum chances (caused by the same system) with not much of a chance between, say one in western Ohio and another in northern Alabama (things like this happen, such as in the big outbreak I was living in northwest Ohio for on November 10, 2002 - http://www.spc.noaa.gov/climo/reports/021110_rpts.html - but say this isn't an outbreak). Because you can only draw one polygon, you draw one to encompass the entire area between northwest Ohio and Montgomery. That's a fairly large polygon, but you might not be expecting a large number of tornadoes in that area, just a few in northwest or west central Ohio and a few in northern Alabama. They aren't far enough apart to fall under your proposed distance rule (and in fact they could be caused by the same system as in that 11/10/02 case). Do you draw a large polygon and risk a small density score? Do you draw a smaller polygon around one of the two areas and risk a large dst missed score? Or maybe you draw polygons around the two areas connected by a very narrow sliver between the two such that it is one polygon with not too large a density score (but that would increase the negative perimeter score)? This kind of situation is not limited to the proposed density score, it's an issue right now as well with the area and perimeter scores.
I have another suggestion - how hard would it be to include the year in the dates for each outlook?
One more thing. My browser remembered my password for this site for well over a year, and today it wasn't there anymore. For the life of me I couldn't get it right, apparently, so I created a new user name, cumulonimbus, as I really wanted to get an outlook out for Tuesday (I would have liked to do it a day ago but was too busy unfortunately). Is there any way to put this new outlook under my nje310 name and secretly get me my password? haha!
