Verification continued

Yesterday I showed the full period (roughly a day 1 convective outlook) of the SSEO. Since we will test the ensemble at finer time scales, down to 3 hour periods, here are the three 3hr periods under consideration.
18-21 UTC

The figures presented here should be used with caution. The models do not always produce reliable or skillful forecasts of the initiation or evolution of convection or convective mode. Thus using them as I did before depends crucially on being good enough in the time period under consideration to provide useful but not specific guidance.

This first thing you may notice is how circular all the probabilities look due to the use of a gaussian smoother. Thus there are both very few reports and model simulated reports in these areas. There is very little overlap between reports and model reports.
21-00 UTC

In this second period, model reports increase dramatically for some members. The amount of overlap increases for NSSL-WRF, NMMB, and 3/4 of the HRW members. Not bad. I would hope that the members can capture some of the severe weather scenario that played out, including getting close to the proper location. NSSL-WRF does well in this period in SW OK as do the HRW-NMMs. If you look to NE MO then those members plus an HRW-ARW member cover that maximum pretty closely.
00-03 UTC

In the 3rd period, the ARW's, NSSL-WRF, and NMMB and HRW-NMM all contribute (glancing blow for some of them) in some way to the maxima in SW OK. The same holds for MO.

For this case at least, the models appear to be able to simulate at least 3 hour probabilities of total severe weather. As I have indicated elsewhere, the use of UH to match against all severe reports appears, again for this case, justified. Such will be one of my foci for the upcoming Hazardous Weather Testbed Spring Forecasting Experiment as part of the Experimental Forecast Program. Testing this out for a bunch of cases this year, and extending it back in time will be a goal of mine moving forward.

Verification issues: severe wx and fronts

I have been computing total severe probabilities from the SSEO and the event on the 17th offered a nice opportunity to do "verification". The idea is to use object based Hourly Maximum Updraft Helicity and treat local maxima as storm reports. This method allows us to extract information from the 4km pseudo convection allowing ensemble in a way that is comparable to storm reports (at least at the level of the grid used for verification, wind reports are scrutinized for speed).

Here is the verification, and of course this is all experimental:

Getting attention

What seems to garner attention these days? It certainly isnt success. More often than not it is failure. Failure it seems, is quite the motivating force. Politicians use the perception of failure in concise talking points. Sometimes they use absolute failure but more often than not they never let the truth get in the way of a good story. This attracts both kinds of attention: The people that agree with you love your refreshing honesty and the people that disagree with repeat your talking point and explain why it is wrong in some nuanced fashion. Double the message at half the price!

Even in social media on a personal level. Speaking about something calmly and rationally gets some attention. But making a scene, getting "facts" "wrong", well that is an entirely engaging flame war for the masses. Everyone wants to take their turn at the pinata that you just happened to leave near a giant baseball bat.

The other ways of getting attention usually involve solid communication skillz. Relating on a personal level, unleashing some passion, being vulnerable, expressing yourself on a deep emotional level. These garner a different kind of attention, perhaps. They seem to be places that resonate from within, in a way that is just as expressive but can also be deeply constructive. The "I have been there" attitude, "I know what that is like".  You know these conversations. they occur on topics outside of controversy, or perhaps we practice them outside of controversy and dont bring them to the flame war party.

We have to choose to behave like this. To attract attention to those things we care about on a personal level, to treat other people as if we share a perspective and thus can relate. To become "friends" before we hammer out our discourse. After all it isnt the perspective that blocks us from acheiving this, its the pollution from "I am right and you are wrong" attitude. A contamination of the communication environment. Its hard to make any point when you have to absorb all the nastiness in your first breathe before you get your words out.

So ask yourself: Are you contributing to the pollution? Are you selling your point of view or are you collecting information to update your perspective or perception?

We cant be perfect in every conversation. So dont expect other people to be. But at least give a listen and try to understand. Make them identify their assumptions and you should identify yours. After all the point isnt to get attention in the first place. Its to communicate.

Risk A Vision

Enough nonsense crosses my path that I have been thinking about the ways in which innovation happens. Its usually an outsider, a person who doesnt think too hard about rules, a rebel. Someone who implicitly knows THE assumptions of the day are actually holding progress hostage. The intuitive feeling for most people would be to simply press ahead, work from within the system, to effect incremental change.

Yet innovation is almost always something larger than incremental. It opens up possibilities where there were none before. It naturally makes you consider other perspectives, options, etc. It comes through revolution -  overthrowing a system of control that was suffocating. It is always a counter intuitive example.

Yet somehow, as a society, we have found ways to control or manage such things. That some process completely understood by people could lead one to innovate. Yet with all this control, we seem to be rather adept at running things into the ground. We keep rediscovering what it means to fail and yet not understanding how total control, perceived or otherwise, led to failure. Some might have you believe that this type of failure is actually a part of the process of success.

What I think marks innovation is perspective not clouded by limitation. Sure you have to fail before you succeed. Failure isnt the goal however. Its the knowledge gained, the perspective altered or rearranged, and the process sharpened. Innovation is very much a step function improvement. A rare event.

But you can not control for it or put rules around it or dedicate a fraction of time to it. Those rules are meant for focus, iterative development, and incremental advances. Maybe you want to challenge me here at this point. "Doesnt that sound like innovation?" It is perhaps counter intuitive but no it isnt innovation. Innovation is all about changing that process from making some thing to exploring the vision of that idea. The result of innovation might be a thing, but that thing was born out of a new way of thinking.

Most of our scientific organizations rely on this. Your proposal to NSF might require you change the way the problem is thought about. To develop new ways of thinking. To tackle problems that are elusive and demand new solutions; to push the envelope of knowledge.

Innovation seldom comes from resource constrained environments. In fact I might argue that it necessarily does not come from there. It will evolve externally, never even making its way into that environment ... why would it? It would be killed in that hostile environment. We speak about topics like critical mass, institutional memory, great people that drive innovation. At times this is true. But seldom does this come at a time when the belts are tightened. It almost always happens in spurts when there is aplenty. In fact it doesnt even have to mean money. It could be people or resources. Any constrained environment, cut off from perspective becomes stagnant.

But the major factor in innovation, and probably the most important player: Risk. Identifying it, seeking it and taking it. The people that innovate arent good at taking risk, they are good at ignoring the risk they are taking. In fact, I think when people look back at it, they say "oh yeah we were taking a big risk but...".  The biggest risk is that you dont take a risk. And that is what our culture has nurtured. We identify the risk and play it safe ... so called risk aversion. More like innovation aversion.

4/7/13 - All about the details

Today is the beginning of multi-day severe weather episode. Thus, it is a good opportunity to take a look at the Storm Scale Ensemble of Opportunity (SSEO) and do a #science. The SSEO is a 7 member WRF-ARW (NSSLWRF 4km, 2 5.1 km HiResWindow) , WRF-NMM (1 CONUS, 2 HiResWindow all 4km) , and NMMB (New NAM 4km) convection allowing ensemble. My recent work (unpublished) suggests that the NMMB is similar in effective resolution to the 5.1 km ARW membership.

Todays forecast features the not-so-mythical cap: that love-to-hate it feature which keeps storms from forming OR makes them isolated and pristine to photograph. So it is not surprising that the first take-home message is that not all members even initiate storms in western OK. NSSL-WRF has a hard time generating storms for the southern half of OK. It appears it develops at least isolated, shallow convection: the models equivalent to turkey towers. Two NMMs develop convection further south, one of which has an isolated storm down by the Altus area. And this is the second take-home message: This behavior in the individual members suggests the models are picking up on storms that they themselves cannot resolve (i.e. storm coverage is low). In situations like this it may mean that isolated storms will be the only mode, and that they will be short lived the further south you go. It is my interpretation that at least one good storm will develop off the dryline by the surface low in SW OK, though it might be short-lived by our standards, long enough to pose a solid hail threat if it is supercellular (sometimes it takes time to ramp up these storms in tough to thrive in capped environments).

The result of this limited storm coverage is reduced probabilities for severe weather proxies. In this case, we use model derived Updraft Helicity as a signal for storm intensity. Our sensible threshold for stronger than most is UH > 25 m2s-2. To calibrate you, storms that get up to 100 are considerably stronger. Storms in NSSL-WRF have generated upwards of 400, but those tend to be single point maxima.

The forecast UH > 25 m2s-2 probabilities (3 hour variety) show the maximum chance for severe weather in the models is northern OK and southern KS. Quickly tonight the threat shifts to MO for any line segments along a front.

Storm motion estimates (southern OK) for right moving supercells are slow and to the southeast, while left moving storms will be quick to the northeast. For the north, somewhat slow and to the east, with left splits quicker to the northeast.

So the uncertainty is relatively large in OK, but I guess I would say that is normal in a capped environment. That models have any signal at all suggests I should favor it. But being wrong by one or two storms means ... well it may mean that nothing happens (i.e. risk is there but not realized) or that something does indeed happen (but is no severe because it doesnt last long enough).

Best chances for today appear to be in KS, with isolated supercells, eventually merging along that boundary and moving eastward. No doubt KS has the bigger tornado threat but you cant rule it out in OK.

The bottom line is that, on a day like today, it is highly unlikely that I could draw for you what the Radar Screen would look like. The devil is in the details and today is ALL about the details in OK.

UPDATED 4/8/13:

 The above is verification for yesterday from my experimental processing of the SSEO relying on Updraft Helicity tracks essentially using 4 passes. The strongest tracks (UH > 75) get the new fancy magenta contours (same levels as the color bar). Looking back at the evolution in the NSSL WRF this morning, it was apparent that it did a fantastic job with both mode, location, and evolution of the storms. It had some flaws in timing across northeast KS and overdid the MO MCS. I also looked at the experimental 12UTC version of the NSSL WRF and it had some differences but generally looked pretty similar. I guess that means I have to start processing the 12Z experimental SSEO version!

The ensemble mean probabilities takes on the shape of the dominant contributing members of the NMM variety but does bot beat the skill of the NSSL WRF at the 15 percent probability level. Looking carefully, each member overlaps with, and thus contributes to, at least one of the storm report local maxima.