Kelton Halbert
banner
stormscale.io
Kelton Halbert
@stormscale.io
Personal account, views my own, you know the drill. Techniques Development Meteorologist for the Storm Prediction Center.

See more of my photography:
https://www.instagram.com/keltore

My blog:
https://www.stormscale.io
Idk, as of now we play a Notre Dame that hasn’t beaten a ranked team. I think our defense can keep us competitive with most of the field.
November 22, 2025 at 9:47 PM
I will mention, that’s mostly experience with Gemini, since that’s what NOAA has access to.
November 19, 2025 at 7:01 PM
Then there’s the code churn. I’ve learned I don’t want AI assistants working on entire files or projects, because it just starts changing things that are unrelated and I don’t like that. I actually try to restrict the amount of context it has and hold the bigger picture in my head.
November 19, 2025 at 6:59 PM
My experience is that it’s fast and efficient at things it’s seen before. So, I kind of treat it like a smarter google search for functions and algorithms I want to use. Once you get into domain specific stuff, you have to keep your eyes peeled. It’ll confidently give you wrong stuff.
November 19, 2025 at 6:58 PM
40+ days.

That was awful.

I dare someone to send one more “welcome back to work” email.

I BEEN HERE.
November 13, 2025 at 4:41 PM
Man, surprised to hear multiple people talking about this chase negatively. We had great lighting and visibility, and hardly had to move at all.
November 9, 2025 at 7:52 PM
Great shot!
November 9, 2025 at 5:48 PM
postprocessing time**

Still learning how to work with observation buckets to get the actual analysis part working.
November 5, 2025 at 8:12 PM
Works properly, too!

The processing time for sfcOA-HRRR (full 3km, native level resolution) on WCOSS2 on my first attempt was 11 seconds 🥳
November 5, 2025 at 8:12 PM
Reposted by Kelton Halbert
Another big finding was that QLCS events have changed drastically in the last decade and are not distributed uniformly. This is likely the result of dual pol implementation, different warning/reporting philosophy, changes in population/urban bullseye and some decades variability.
November 5, 2025 at 3:18 PM