SYTYCF

So you think you can forecast?

The atmosphere always has a surprise in store. After much debate about the goodness of the previous days crazy heat burst forecast and the subsequent realization that one occurred in Madison WI, we had some good discussion about what makes a good forecast. These issues always arise. Define good? Is there really one metric that just sums it up? What do people expect from the forecast? Does the forecast add value? Add value relative to what?



Th easy answer is ... there aren't easy answers. Especially for the types of rare event forecasts that we make in the experiment. We have ways of making it easier, like using long time periods, or forecasting for any severe weather. We have neighborhood methods that give credit for being close, where close is within XX km. But these are difficult problems and we seldom have the right tools for the job. We make do with the best tools we have and struggle to make better ones. Including models!

The real story with forecasting, for me, is always about crafting the right probabilities and having that make sense for the scenario that I think will unfold. We can use summary information all we want, but it can be a distraction away from the scenario. This leads to the classic data overload problem. If I have 15 ensemble members, and 2 hours, can I faithfully say that i can identify the scenario's and understand them all? Can I do that and make a really good forecast?

So fast forward to todays event which didnt muster a slight risk on either day 2 outlook. But as we started digging deeper into the detailed hi res models we did see a very good environment. Can I say we totally predicted this event? Nope. In fact after staring at the radar screen for hours I am probably biased.

Our afternoon briefings trended in the right direction. I briefed the EWP folks about he possibility that multiple rounds of storms were possible in North Texas. The environment seemed to get really good between 22-03 UTC when the low level flow switched to southeasterly and easterly, a low level jet extended westward, a 345K PV feature approached along with a 500 mb compact short wave led to surface pressure falls into the area. Deep layer shear was already good all day and would increase from 40 to 55 knots. All indications pointed to supercells. Shear vector orientation was perpendicular to the dryline indicating a discrete mode.

the caveat were robust storm coverage further east. This meant the possibility of too many storms in an otherwise uncapped atmosphere. At least as far as the models could tell. The hi res models were robust with storms and showed north texas having possible multiple rounds of initiation in multiple areas along the dryline prior to the dry line retreating westward. Thus models were a bit all over the map, literally, but all pointing to the end game of upscale growth into possibly multiple MCSs or squall lines. That transition was relatively quick. And this brings me back to model capability.

Do models have the capability to represent the smallest of storms? No. A 4km grid spacing WRF model strongly filters below 7 delta X. So a giant supercell can be represented in the model. We are fortunate for this. But not all supercells are giant and given a great environment, they can and will grow upscale. So we have a problem. We cant know for sure, no matter the signals the models reveal, if the transition time will be accurate. We also dont have much faith in model convection initiation timing. It could be off by 2 hours. Thats forever in a situation where tornadoes are concerned.

that said, we made decent slight risk forecasts, going 30 percent in the full 16-12 UTC period and then upping our 3 hour probabilities to 20 percent in the 00-03 UTC [I need to verify this] time period. i think we had the signals but we didnt exactly have the scenario. The EWP forecasters asked me what I thought would happen in our first briefing and I said "I dont know". It could be real messy and it could turn into a "mesoscale accident" (MA). Pattern recognition or intuition was at work. MA is a term to describe the unpredictable. A whole host of things happen to come together at the right time to produce something unexpected or extraordinary. You cant really forecast it, because the uncertainty is just enough to cast doubt on going big. Seldom in these events will you ever have enough time to truly know all the details you need to know in advance. Let alone rely on these signals that you dont know enough about to call them reliable.

The last mesoscale accident, was 24 May 2011. Coincidentally and perhaps causatively the last time Lance B was visiting, we had an OK outbreak. A crazy long lived left moving supercell dodged every right moving tornado warned supercell until it hit the Norman area and messed up the supercell that was producing the Goldsby tornado. Nothing happened in Norman, though it rained debris in the NWC parking lot.

Todays MA was the opposite. It meant that 4 discrete tornadic cyclic supercells developed 30 minutes early than our earliest guess and lasted well into the night before finally achieving the MCS/squall line end state. It appears many were hurt, some killed. Some communities fared better than others despite impressive radar signatures. I dont know how many tornadoes occurred but 3 reports are present and many others yet to be entered. Dallas Forth Worth and surrounding communities have endured yet another tornado event. I am confident I speak for everyone when I say that our thoughts and prayers are with you. We wish you the speediest of recoveries.