iiinova — where your digital self keeps evolving

Tim Supinie @plustssn.autumnsky.us · 13h

After some tweaks to the precision for various fields, I managed to get the file sizes down quite a bit. Over 4.5 GB/run previously to just above 3 GB/run now. That's a nice savings for the data download, but it's also great for long-term storage, since these add up quickly.

031

Tim Supinie @plustssn.autumnsky.us · 14/10/2026

This morning's 12Z OUN sounding has a higher precipitable water than any other sounding taken outside the June 26-September 23 period. Hence the swath of very heavy rainfall in western OK overnight.

A precipitable water climatology for the Norman, OK site. The value from 12Z October 14 is labelled.

0185

Tim Supinie @plustssn.autumnsky.us · 14/10/2026

A much cooler and drier air mass moves in for Friday. Given cold advection, waning sun angle, and perhaps some residual cloud cover, temperatures may struggle to rise much above 60ºF on Friday.

Mean surface temperatures, winds, and PMSL from the REFS initialized 00Z 10/14, valid 12Z 10/16.

Mean cloud cover from the REFS initialized 00Z 10/14, valid 12Z 10/16.

110

Tim Supinie @plustssn.autumnsky.us · 14/10/2026

The 00Z REFS has a lot of water around here over the next 48-ish hours. PM mean QPF has a broad swath of 4+" through central OK. Given the extremely moist air mass (2+" precipitable water, among the highest ever observed in October) and potential for training storms, someone is gonna get wet.

Probability-matched mean QPF from the REFS initialized 00Z 10/13, valid from 12Z 10/13 through 12Z 10/15.

Precipitable water with ensemble max composite reflectivity overlaid in purple. REFS initialized 00Z 10/13, valid 10Z 10/13.

170

Tim Supinie @plustssn.autumnsky.us · 08/10/2026

Well, that's one variable that has all members, so that's part of it. Also, you can see a lot of noise in the RRFS/member cref, which is tough to compress. I've had some luck with basically doing cref = max(9, cref), which smashes any noise that isn't plotted. Not fully implemented yet, though.

RRFS member 5 reflectivity from the 18Z 10/7 run

110

Tim Supinie @plustssn.autumnsky.us · 08/10/2026

Also on the subject of bugs, I realized earlier this week that I accidentally dropped the encoding on some of the fields in the zarr stores backing the page. The encoding specifies the quantization, so those variables are full-precision in the zarr stores, meaning they don't compress as well.

Stacked bar chart with connectors between the two bars. The left is with the original (broken) encoding, and the right is with the corrected encoding.

100

Tim Supinie @plustssn.autumnsky.us · 06/10/2026

Here's an autumnplot-gl bug I hadn't found. A quirk of floating point numbers is that -0 is distinct from 0. Turns out, if you pass u=-0 to the autumnplot-gl barbs, MacOS incorrectly rotates the barb 180°. Or, it did until I fixed the dev branch just now. Still an "Easter egg" in the REFS page.

A screenshot of the REFS page containing a field of barbs mostly from the north/northwest, but a few conspicuously rotated 180 degrees from where they should be.

2181

Tim Supinie @plustssn.autumnsky.us · 06/10/2026

Also, some strange behavior when looking at upper tilts from live data. The most recent volume is incomplete, so I'd like to look at the most recent complete volume. But when I go back one volume scan, it can jump to the incomplete volume. I assume this is due to having to handle the SAILS cuts.

210

Tim Supinie @plustssn.autumnsky.us · 06/10/2026

There are a whole bunch of radials that look like they're duplicates of the ones next to them. I think you said you're doing some regridding to keep a common autumnplot grid in the background. Is this a result of that?

210

Tim Supinie @plustssn.autumnsky.us · 26/09/2026

Big REFS Transition Project 2 (SPC REFS viewer v1) is functionally complete. I sent a version around to SPC forecasters for a final pass at bugs this afternoon (though I realized I had a few more 5-minute tweaks I wanted to make when grabbing these screenshots).

2265

Tim Supinie @plustssn.autumnsky.us · 20/09/2026

Oh, I see it's "convective line develops just east of us" season already. It's never gonna rain here again, is it?

3161

Tim Supinie @plustssn.autumnsky.us · 19/09/2026

👀👀👀👀 (I mean, this is still basically average for the time of year, but man, it's gonna feel good after 2 months of 100 ºF almost every day.)

020

Tim Supinie @plustssn.autumnsky.us · 14/09/2026

Text I sent to the wife on August 27. Okay, it was three weeks, but close enough.

A text that reads:
Guarantee within the next week, maybe 2, there's gonna be a all-hands, hair-on-fire effort to scrub "Lake Ontario" from our website.

0110

Tim Supinie @plustssn.autumnsky.us · 14/09/2026

Huh. Well, I guess that explains the lack of Internet since early this afternoon.

A text that reads:
OEC and OEC Fiber
We are experiencing a large outage in Noble due to a grass fire that damaged main fiber lines. Repairs are expected to continue into tomorrow. No ETA is available, but we will keep you updated as best we can.

060

Tim Supinie @plustssn.autumnsky.us · 13/09/2026

The subtropical ridge remaining over Oklahoma well into September, combined with the antecedent very dry soil conditions, led to yesterday's 106ºF reading at OKC. My guess is the frequency of this kind of persistent September heat will increase as the climate warms.

GFS forecast for 7pm CDT Thursday, September 17. The subtropical ridge is still hanging around the south-central US, with a 597 dam ridge centered near Ft. Smith, AR.

020

Tim Supinie @plustssn.autumnsky.us · 13/09/2026

Summer 2026 in Oklahoma in 3 graphics. (Reposted to correct the caption on of the graphics.)

Persistent ridging over the central US during July and August 2026 ...

... led to generally below-average rainfall across OK, with much of the state being at or below 50% of normal rainfall for the period July 15-September 12 ...

... leading to very depleted deep soil moisture across much of the central and western portions of the state.

120

Tim Supinie @plustssn.autumnsky.us · 31/08/2026

Man, I wouldn't want to be Future Tim. That guy's really got a mess to clean up.

A comment in some Python code that reads, "fmax can't handle pint quantities, so strip and reapply units. Checking to make sure the units of da1 and da2 are compatible is an exercise for future Tim."

4210

Tim Supinie @plustssn.autumnsky.us · 29/08/2026

The top of the boundary layer is visible on KTLX. The patchy distribution of scatterers implies convection within the boundary layer. Contrast with the smother distribution in the free atmosphere. I think there's a local max in temperature near Wichita Falls that accounts for the deeper PBL there.

090

Tim Supinie @plustssn.autumnsky.us · 28/08/2026

Another piece of guidance forecasters use for day 3 thunder is from the SREF, but that will be retired. So we're adding GEFS Calibrated Thunder (GEFSCT), developed by @wx-pythonista.bsky.social, to cover out to day 8. This is also our first foray into guidance for AK, HI, and PR.

Sample day 3 GEFS Calibrated Thunder forecast

150

Tim Supinie @plustssn.autumnsky.us · 28/08/2026

Another major replacement was HREF/GEFS severe guidance with GREFS, put together by Chris Karstens. GREFS is one of the guidance products forecasters can use to draw their outlooks, and it's also an input to the National Blend of Models.

Sample 24-hr GREFS tornado probability forecast graphic

Sample 24-hr GREFS wind probability forecast graphic

Sample 24-hr GREFS hail probability forecast graphic

150

Tim Supinie @plustssn.autumnsky.us · 28/08/2026

One of the major replacements was HREF Calibrated Thunder (HREFCT) with REFS Calibrated Thunder (REFSCT), both developed by David Harrison. This provides guidance for us to draw the thunder line on the days 1 and 2 convective outlooks. This will also be available on Big REFS Transition Project 2.

A sample full-period REFS Calibrated Thunder forecast graphic.

240

Tim Supinie @plustssn.autumnsky.us · 25/08/2026

Ah, late August, when the "cold" fronts take you from 105°F/55°F to 95°F/70°F.

3191

Tim Supinie @plustssn.autumnsky.us · 16/08/2026

BLTs (with market Ts) and market peaches: the only reasons to look forward to summer.

050

Tim Supinie @plustssn.autumnsky.us · 14/08/2026

Somebody @wikipedia.org had fun with "from Siberia to Iberia."

A screenshot from Wikipedia's "In the news" section. The top item is "A solar eclipse (animated path shown) causes totality from Siberia to Iberia."

020

Tim Supinie @plustssn.autumnsky.us · 10/08/2026

Another thing I've been working on is the ability to plot quantized fields, say using a UInt8Array with a scale and offset. This example uses quantized u and v fields to create a vector field and then reconstructs the original values using a scale and offset. Missing values are respected throughout.

The first function call constructs a quantized vector field, where each component is a UInt8Array, and the values are integers from 0...254. The missing value in the original field is set to 255.

The second function call reconstructs the original values using the given scale and offset. It also says that the missing value for the output field should be -999.

160

Tim Supinie @plustssn.autumnsky.us · 10/08/2026

For this example, I've set the missing value for the wind vector field to be -999, and, it's properly handled in the GPU computations (which compute the wind speed field), the plotting (ignoring those barbs and that section of the wind speed field), and the sampling.

The readout from a region of the field where the height field is NaN, the wind vector field is [-999, -999], and the speed is computed from the vector field.

130

Tim Supinie @plustssn.autumnsky.us · 10/08/2026

For autumnplot-gl, NaN is kinda the de facto "missing value," but certain machines/OSes don't handle that like you'd expect (see the pic for an example). So I added a missing value option. The idea being that you can specify a value that isn't plotted and will be ignored in all calculations.

An MRMS radar reflectivity field with NaNs plotted. The machine has picked the first value in the color map to use as the color for the NaN regions, whereas I think most people would expect that to be transparent.

170

Tim Supinie @plustssn.autumnsky.us · 10/08/2026

One of the better iterations of a burger and sweet potatoes I've made, and I really like this meal.

080

Tim Supinie @plustssn.autumnsky.us · 04/08/2026

I'm working evenings tonight and tomorrow, so I made dinner early. Lightly modified the Moroccan spice blend from another recipe and put it on chicken and rice.

Moroccan spiced chicken and rice in the pan.

And the same thing portioned into 4 meals.

080

Tim Supinie @plustssn.autumnsky.us · 27/07/2026

Made stromboli for dinner this evening.

0110

Tim Supinie @plustssn.autumnsky.us · 10/06/2026

Finally got around to getting the 2025 environment data around for @tornadoarchive.com. Here's a preview: the March 14 mid-south tornado event.

150

Tim Supinie @plustssn.autumnsky.us · 31/05/2026

Watermelon mojitos was the original thing, and then I ran out of limes.

000

Tim Supinie @plustssn.autumnsky.us · 31/05/2026

We got bequeathed a watermelon, so I'm coming up with creative ways to use it up.

Watermelon Jolly Rancher
2 oz watermelon syrup
2 oz rum
1.5 oz club soda

Stir over ice

260

Tim Supinie @plustssn.autumnsky.us · 11/05/2026

Chickpea and squash curry for dinner tonight. Squash (as a new world ingredient) is nontraditional, but it still works. Nice vegetarian meal.

160

Tim Supinie @plustssn.autumnsky.us · 09/05/2026

Inflow band on this storm coming into Norman makes it look like a shark.

0304

Tim Supinie @plustssn.autumnsky.us · 20/04/2026

Hahaha, imagine getting lectured by That Fucking Guy about sexual assault. Irony is alive and well. (Predictably, the statement not really about sexual assault, but the usual schlock about immigrants.)

Statement from the White House dated April 7, 2026 titled "Presidental Message on National Sexual Assualt Awareness Month."

091

Tim Supinie @plustssn.autumnsky.us · 16/04/2026

Mammatus show this evening in south Norman.

0151

Tim Supinie @plustssn.autumnsky.us · 15/04/2026

I think that storm I mentioned earlier is now this storm northeast of Duncan, OK. It's still relatively discrete. Seems like long-term, that has the best shot at producing a tornado as anything in OK/TX. Gonna need some stronger low-level flow before that happens, though.

Side-by-side reflectivity and radial velocity from KFDR at 2339 UTC (6:39pm CDT) 4/14.

VWP from the KTLX radar valid 2350 UTC (6:50pm CDT) 4/14.

130

Tim Supinie @plustssn.autumnsky.us · 14/04/2026

VWP hodographs in the vicinity of the storm in western north Texas are a dead ringer for that HRRR forecast hodograph from last night. Blobbification incoming.

Hodograph from the Frederick, OK radar valid 2002 UTC (3:02pm CDT) 4/14. Surface wind is taken from a nearby Oklahoma Mesonet site.

150

Tim Supinie @plustssn.autumnsky.us · 14/04/2026

On the other hand, this profile from a little farther north would probably pose a bigger overall tornado risk if the early stages of convective development were in this environment. Better low-level storm-relative flow to keep the outflow at bay until the shear picks up closer to 00Z.

100

Tim Supinie @plustssn.autumnsky.us · 14/04/2026

00Z CAMs seem pretty solid to get convection in OK tomorrow, my main question is the wind profiles. With this kind of surface pattern, lots of veered flow in the low levels means weak low-level shear until evening decoupling + mass response to the synoptic-scale lift.

2 m dewpoint from the 14/00Z HRRR, valid 21Z 4/14.

100

Tim Supinie @plustssn.autumnsky.us · 14/04/2026

FWIW, a lot of the missing sites in the central/western US launched at 18Z. I guess we'll unfortunately probably never know if having FGZ and ABQ balloons in the 12Z modelling suite would have provided a better constraint on, for example, the KS convection.

1144

Tim Supinie @plustssn.autumnsky.us · 14/04/2026

The northern two storms at least formed on the dryline, so that makes sense (even if lift was questionable). The southern storm currently west of Butler, MO formed way east of the dryline with no obvious boundary. So I have no idea how that storm is there. Terrain shenanigans?

130

Tim Supinie @plustssn.autumnsky.us · 12/04/2026

The dryline circulation ought to be fairly well-developed, with strong heating on the dry side (the HRRR's attempt is here). But the strong cross-dryline flow means parcels may be quickly removed from the zone of ascent. This plus lack of synoptic forcing makes me skeptical of sustained convection.

A cross section from this morning's 12Z HRRR perpendicular to the dryline valid at 23Z 4/13.

170

Tim Supinie @plustssn.autumnsky.us · 12/04/2026

I'm not surprised SPC dropped the slight risk for the southern Plains tomorrow. Completely straight mid-level flow and neutral height tendencies over the dryline suggest little synoptic forcing for ascent, so you'd be relying completely on the dryline circulation to get updrafts going.

SPC outlook issued 1727Z 4/12 valid 12Z 4/13 through 12Z 4/14.

This morning's 12Z GFS forecast valid 00Z 4/14.

160

Tim Supinie @plustssn.autumnsky.us · 12/04/2026

In honor of the successful return of Artemis II, I'll call these Orange Spacesuits. Kinda the same vibe as a whiskey sour, but sweeter since I used mandarin oranges instead of lemons.

An Orange Spacesuit cocktail (makes 2 drinks)
2 oz orange syrup (see below)
4 oz mandarin orange juice (~2-3 oranges)
7 oz whiskey

Stir together and serve over ice.

Orange syrup:
Zest from 2 mandarin oranges
1/4 c sugar
2 tbsp water

Add zest to sugar and mash it around with your fingers. Let it sit for a few minutes. Add water, microwave for 10 seconds, and stir to dissolve. Strain out the orange zest to get the syrup.

070

Tim Supinie @plustssn.autumnsky.us · 09/04/2026

Here's my shot. Not confident in Dean, Felix and Ivan. I think Beryl might also have another segment somewhere in the Caribbean, but not sure which. Maybe the one just west of Jamaica? (I spent way too much time on this.)

130