Our Stormeyes sites (including Image of the Week, East Norman Rainfall, this BLOG, and various non-SkyPix web pages, etc.) have been down most of the last few days while Cloudflare and HostPapa point fingers at each other. [SkyPix is on a different host.] The problem, whatever it was, seems to have gone away for now…so, I actually have a backlog of entries. First, a reasoned defense from an inside perspective of perceived lapses in SPC outlook and watch performance so far this year…

———————–

A story from NBC, and others on the less-than-2% outlook’s tornado occurrences in eastern Kansas, some of which also note the watchless tornadoes in Lower Michigan earlier this year, grossly oversimplify reality on the SPC forecast desk. The NBC story in particular relies on a lot of idle speculation by people who haven’t worked the SPC outlook desk, and as such, just don’t know.

I’ve done SPC outlooks for decades, and as usual, will be brutally honest here. I know how it goes there, from the inside. Read and learn. These insights don’t lend themselves to 10-second attention spans nor quick sound bites. So this is long. Don’t “TLDR” this post if you really care about actually finding out how it works.

I’ll put the bottom line here, near the top: Chances are, there is no smoking gun. That may not suit rage-baiting and click-baiting, but it’s simply reality. That’s the lede. Here’s the rest…

Why no smoking gun? Far too much goes into a forecast to lay “blame” at any one factor. Like it or not, bad forecasts happen. They always have and always will. The aim is to reduce them over years, knowing that some events are so localized and/or extreme that both human and computer forecasts can’t always nail them down. That’s reality.

Until forecasters have extremely high-resolution sampling of the real atmosphere on scales storms form and operate (a few miles), even the most sophisticated models, both from traditional, physics-based and AI/statistical packages, will suffer sometimes with localized subtleties. Guess what’s involved in forming a dryline storm here vs. somewhere else, amidst capping and modest broader lift? You got it, friends, localized subtleties.

Yes, observational balloon data were missing and are, in bulk and when quality-controlled, important to models. Several scientific papers have shown this. Maybe that mattered here. Maybe it did not. Satellite-derived data matter too, but radiosondes can influence model performance by several relatively simple situational measures (e.g., this paper, or this one, or this one).

How important was the lack of radiosondes to this case on this day? We don’t know. As Alan Gerard alludes in his article on the radiosonde-loss issue: that needs to be studied (using data-denial experiments). Until then, it’s speculation to say how much that altered output at any of many levels of the atmosphere, from models that *variably* and *incompletely* influenced this particular outlook’s positioning.

*Numerous* models are examined every forecast cycle, especially early-arriving deterministic ones, ensembles, and newer/quickly computed AI packages that work off historic pattern recognition. How they may be affected by missing input data can vary from model to model and by data type. It can be such a dense black box that such effects are simply unknown to the forecasters. We’re not, and cannot be, privy to every nook and cranny of their physics or statistical equations. Forecasters often notice and account for model biases, but where they come from can be quite complex and not just tied to one factor.

Between that and diagnostics that should precede models, it can be a veritable firehose of information, on deadline. With time, experience, on-shift mentorship of the leads and senior outlookers, and training, forecasters get better at situationally prioritizing what to drink from that firehose, when, how, and why. It is simply impossible to examine every possible diagnostic and prognostic detail from every data source.

Models are not all that go into a forecast. So do diagnostics: analyses of surface and upper-air data. The latter factually do have holes that may cause analysis to miss subtle features, but was that true here? We don’t know yet. Other diagnostics, such as satellite and radar-indicated features, and intangibles such as reading, research, forecaster experience, and intuition with specific situations, also play a role.

It’s even more speculative, and likely inaccurate, to say the lack of greater staffing affected the outlooks in these cases so far. [That isn’t to say it can’t, or won’t, the rest of the season.] Though I recently retired, and was not a participant in these forecasts, I do know the principals involved. Everyone who did the outlooks for the KS day were working normal 8-hour shifts and not overtime. If “exhaustion” or “fatigue” were factors, it comes simply from the nature of rotating shift work, which is documented to be unhealthy mentally and physically, and a known carcinogen. Don’t knock it ’til you’ve done it.

Yes, with two retirements last month, 5 openings (out of 10 positions or 50% vacant) are on the SPC outlook/mesoscale desk as of this writing. That is unprecedented. They need to be permanently filled with full-time forecasters, stat! A lot of fill-in shifts by both managers (one of whom is an extremely sharp and highly experienced forecaster), and less-tenured forecasters, will be needed until those are filled. Results may vary.

That won’t help, and yes, it might hurt! But it’s premature and speculative to pin any single forecast performance so far, or the rest of this season, just on that. Again, forecasts sometimes simply miss. SPC has a well-earned reputation for, and internally motivated standard of, excellence. Excellence is not synonymous with perfection. Even I had some bad forecast decisions I’d like to have back. 😉

Outlooks at SPC do not happen in a vacuum. One or two names may appear thereon, but it’s a team forecast. Internal collaboration is required. External coordination with involved local NWS offices (WFOs) is strongly encouraged if major changes are being made to a previous outlook. Otherwise, there is not enough time to coordinate every part of every outlook line with every involved WFO, who themselves also are busy with other tasks. Every minute doing that is a minute not spent doing meteorology. So there must be a balance achieved on deadline.

I don’t know for sure here, but it is possible that the 2% and 5% tornado lines that drove the “MRGL” and “SLGT” areas were suggested by, or compromised with, the WFOs serving eastern KS. Only they and the SPC forecasters on duty could verify either way.

And even everything I’ve typed is just a superficial, condensed summary of the outlook forecast process. I did thousands of them, both the graphics and long-form text discussions, so give me the benefit of the doubt here.

[Addendum] From a subcomment I made on one of Alan Gerard’s BLOG posts:

Even if the 5-forecaster deficit on the SPC outlook/mesoscale desk didn’t directly influence this poor forecast, that doesn’t preclude its becoming a real problem as the season wears on and staff start wearing out.

Managers and less-senior fill-ins have scrambled to get the shifts covered creatively, without much OT as it turns out, and at the expense of granting leave. As you know from being an MIC, that’s just a very short-term band-aid. At some point, leave will pile up and have to be granted (or lost!), somebody or multiple somebodies will get sick with no shift room to spare, managers will have to do their day jobs eventually too, and with half the outlook positions still empty, something will have to give. I hope it’s not forecast quality.

If and when we start seeing consistent misses/overforecasts, and minor but normally uncommon errors becoming common in the products, that’s the sign that things are starting to go poorly, and that SPC is on the brink of disaster on the forecasting floor.