Interested in making cool maps, working with big datasets, and helping drive big weather decisions? We're hiring a Meteorological Developer. DM me for more info. #mdwx umd.wd1.myworkdayjobs.com/UMCP/job/Unive

Seriously, stuff like this was prohibitively difficult before Tim’s autumnplot library, and if you want to read grib2 in the browser, his other library is clutch. Foundational infrastructure doesn’t always get the attention it deserves, but a few years ago this stuff was basically not possible.

I'd like to introduce you all to a personal project I've been working on recently. It's called Heimdall, and it's an MRMS + ProbSevere situational awareness monitor for CONUS scale severe weather. It's got roaming, alerts, archive mode, and is fully client side. heimdall.stormscale.io

Whatever you do, do NOT get the chicken curry from Flame Grill & Mediterranean Bakery in Kansas City. Unless you enjoy food poisoning. In which case, go right on ahead.

Collecting hazard pay and a stiff drink on my way to Kansas City for a work trip this week.

On my day off I thought I’d stop being a curmudgeon and try Claude Code so that I can at least have informed opinions. It one shotted a feature in my BUFR library in 25 minutes. So. That’s terrifying.

Spring is doing its thing in the southern plains next week

Good Morning! We’re starting off Friday with some exciting news. We’re expanding to Atlanta! We are excited to bring our brand of severe and winter weather coverage to the Atlanta metro area. Tell your friends, tell your mom and them, we're in Atlanta now. #atlantaweather #atlantawx #atlwx

I’m very excited to announce the latest release of SHARPlib, v1.4.0. This might be one of the coolest updates yet (in my opinion) and results in dramatic performance improvements through the use of pseudoadiabatic lookup tables. Available on PyPI/conda-forge. github.com/keltonhalber...

I’m very excited to announce the latest release of SHARPlib, v1.4.0. This might be one of the coolest updates yet (in my opinion) and results in dramatic performance improvements through the use of pseudoadiabatic lookup tables. Available on PyPI/conda-forge. github.com/keltonhalber...

New paper: "Evaluation of Planetary Boundary Layer Parameterization Schemes Using WoFS Ensemble Members and Observations from TRACER"—led by Francesca Lappin, with appearances by Petra Klein, @tmbell05.bsky.social, @ksquared812.bsky.social, and yours truly. 🧪🌊 journals.ametsoc.org/view/journal...

If you’re a software dev, & especially so in open source, now is as good of a time as any to check your security posture. I just went through & audited my GitHub workflows & security settings, & you should too. Open source supply chain attacks are getting more frequent. youtu.be/yiLIZLPNEm8?...

That was so freaking awesome

Hearing “Full Send” on the NASA stream was incredible

SHARPlib received its first unsolicited LLM agent contribution in the form of an unsolicited PR review. Does that mean my project has made it now? I guess at least it identified something somewhat useful.

Always so much fun getting to give the annual tornadogenesis lecture for the SPC elective class. Good group of students this year!

Last night was the final shift before retirement for SPC forecaster Mark Darrow. He’s been with SPC since it was called SELS in Kansas City, & retires holding the record for most SPC convective outlooks issued. Congratulations on retirement Mark! It’s been a pleasure working evenings with you.

Only a few days left to apply for student travel grants! Students in undergrad and graduate school are highly encouraged to apply if you're planning on attending the Joint Summit.

If you’re a student, I highly recommend checking out these opportunities. Not only is SLS a great conference, but Madison is one of the best cities ever and I can’t wait to (hopefully) go back

Version 1.3.1 has been pushed to PyPI and conda-forge. It’s primarily a bug fix for the detection of the dendritic growth zone and hail growth zone, where an edge case was found to return bad data with extremely shallow layers. Version 1.4.0 is coming soon, and it’s a big one.

My latest project in autumnplot-gl has been to add a feature that @stormscale.io suggested: the ability to do simple computations on the GPU. This alleviates some issues with having to compute and/or download additional fields. (1/3)

Sure, the world is falling apart but have you tried coding for fun while having a gin & tonic while listening to the Goat Rodeo Sessions?

I need to get more proficient with the tenderizer and get the chicken thinner next time, but made some spicy truffle katsu chicken sandos for dinner and it definitely goes hard.

I can now post process the entire RRFS North American domain in ~45 seconds for a single fcst hour. HRRR CONUS domain takes 4 seconds. ~120 convective variables including effective inflow layers, SB/MU/ML parcels, and all sorts of derived parameters. Timings use a 128 core node with OpenMP.

The NOAA/OAR federal positions are now open! Applications close in *one week* on March 25th: www.usajobs.gov/job/861651700 For more details on GSL’s opening in AI4NWP and related areas, please see: gsl.noaa.gov/employers/jobs (refresh page if needed - we’re dealing with some pesky server issues)

On a completely different note; is making it on Ryan Hall’s twitter feed the modern equivalent of making it on the Weather Channel?

Days like today are the good reminders that the work and persistence are worth the challenges and frustrations. Solving real and concrete problems, strong positive feedback, and a goal/direction/horizon to keep aiming towards.

“Babyman” knows irony when he sees it. I still accept and appreciate that he concedes it was in poor taste. Frankly, my responses were pretty measured in the face of uninformed hyperbole from a snowflake. Direct and candid? Absolutely. Arrogant? Up for debate, but nuance is hard on the internet.

It’s me! 👋🏻 I’m the genius mentioned below responsible for implementation. Did I achieve “stable genius” status?!?!

A few days of bashing my head into a wall later, I’ve cut down the filter-and-read time for RRFS North America domain native-level files from 2-3 minutes to 5-10 seconds. Had to use every trick in the book. Index files, memory mapped files, and parallel decompression.

“AI is so great at coding!” *Gemini 3.1 pro proceeds to hallucinate every single GRIB2 API call*

HRRR public cloud data: BUFR profiles in a directory RAP/GFS public could data: BUFR profiles in a tar.gz -_- HOW IS A COMPRESSED ARCHIVE CLOUD FRIENDLY ACCESS

Fighting with GRIB2 Is just the greatest.

Ah crap, I’m gonna have a busy midnight - 2 AM stretch tonight -_-

So no discussion on conditional intensity today? We’re only going to do that on the busts, aren’t we….

As someone who has dozens of unique problems brought to me daily due to my position (not my intelligence), it is helpful from time-to-time to remember that problems are easy. Solutions are hard.

I have publicly criticized my fair share of the weather enterprise as I tried to make a name for myself, especially when I was younger. I wish I knew then what I know now. And this not-so-vague vague thread is not directed at any one person or entity. But rather a reflection come full circle.

With this paper about to be formally published, I thought for funsies I’d look at the right-moving supercell tornado environments ala Thompson et al. 2012. The increased sample size shrinks many of the distributions, and shows a small/narrow sample of EF4+ 2012-2022

Apparently folks need to be reminded that the Michigan tornadoes would not have been hatched in either system of intensity outlooks. So.

Today is SPC's first explicit forecast of significant tornadoes possible (CIG1) in a tornado probability less than 10%. Based on radar data and environment a significant tornado may have occurred in parts of that forecast area. SPC could not have explicitly forecast this kind of scenario previously.

I think today was the perfect example of the value added with conditional intensity outlooks. First potential significant tornado in a 5% CIG 1.

Some exciting news to share: very soon, NOAA Research will be hiring new federal employees into science positions across the agency! A couple of notes: 1) They will only be open on USAJobs for seven days. 2) Multiple positions at different sites may be in one posting. (1/2)

A new era for SPC Outlooks & severe storms forecasting begins today! If you haven’t seen the new outlook (severe & fire weather) graphics yet, check out the SPC website. www.spc.noaa.gov Increasing severe risk in the coming days will give a first live look at the new conditional intensity levels.

The end of an era: with today’s SPC Conditional Intensity update comes new graphics. For the first time ever, the main convective outlook graphics will not be plotted by GEMPAK.

Be sure to check out the short explainer from SPC on what conditional intensity is, and what it means for our outlooks!

The NWS Storm Prediction Center's severe weather outlooks will be changing slightly next week. Meteorologists Evan Bentley and Liz Leitman, our experts at SPC, provide an update about the new conditional intensity outlook.

SHARPlib 1.3.0 is available via PyPI, conda-forge, & source distributions. This update brings a fix to a longtime issue present since SHARPpy w/ how the dendritic growth zone is calculated. New features include PyroCB Firepower Threshold, PBL top detection, & more. github.com/keltonhalber...

Duck fat crispy potatoes Oven roasted baby carrots Rack of lamb with red wine and chimichurri sauces

SPC Conditional Intensity teaser video dropped on YouTube. We’ll have another explainer video (around 3 min length) coming soon! Outlook changes coming March 3rd. youtube.com/shorts/uWzMP...