Bill Meier

Green Valley, Arizona sunrise this morning. #azwx #sunrise

@tomhfh @MZHemingway @grok What orbital altitude was obtained during the recent Polaris Dawn mission?

Every single time a human being has left low earth orbit has been between Richard Nixon's first election and his impeachment.

Building overhead of Green Valley. Some thunder. A few rain drops. #azwx

At 2pm.

Cu bubbling up over the Santa Rita Mountains. #azwx

Courtesy of: wxbars.com

Weather Bars gives you a clean, at-a-glance 7-day temperature forecast without the clutter. No ads, no upsells, no noise. The centerpiece is a simple bar chart — each day's high and low floating as a gradient bar on a dark background. You see the whole week's temperature range instantly. Tap any bar for the full NWS forecast text. Tap your current temperature for wind, humidity, dewpoint, and more. Built on the National Weather Service's data. Installs on your phone like a native app. Works great on desktop too. If you're tired of weather apps that bury the forecast under ads and unnecessary features, this one's for you.

Built a simple weather PWA over the weekend with some AI help — clean, no install, no account, no ads. If anyone's tired of bloated weather apps, give it a shot: wxbars.com

@bcherny I like them all. But I tried Dispatch today. That was very cool.

Little known fact, the Anthropic Labs team (the team I joined Anthropic to be on) shipped: - MCP - Skills - Claude Desktop app - Claude Code It was just a few of us, shipping fast, trying to keep pace with what the model was capable of. Those early Desktop computer use prototypes, back in the Sonnet 3.6 days, felt clunky and slow. But it was easy to squint and imagine all the ways people might use it once it got really good. Fast forward to today. I am so excited to release full computer use in Cowork and Dispatch. Really excited to see what you do with it!

Sunset, Green Valley, Arizona. #azwx

Mid to high level clouds illuminated by the last light of the setting sun. From Green Valley, Arizona #azwx #sunset

@Brady_Wx @ameliaUrquhart_ @CameronJNixon @ConvChronicles I've been tracking models for recent storm events. Including your occasional StormNet-v4 glimpses. It has been unusually accurate. Tracking this event, too.

@ameliaUrquhart_ @CameronJNixon @ConvChronicles StormNet-v4 is biting

Thursday, March 5th has started to seriously uptrend on the GEFS, and this is the first setup I've seen this year so far that I really like for the Southern Plains. I'd love to hear any thoughts other forecasters/chasers might have.

@Stormchaser @JAtanackov While I agree somewhat, I do think there is the matter of trust. For instance, if you posted an incredible photo I'd be inclined to trust its authenticity. If some unknown knucklehead did I'd definity question it. Whether others will be as discerning may be the problem.

One of the greatest technological enigmas—and philosophical challenges—in human history is unfolding before us. For photographers, AI may complete the erosion of trust that mass digital media began. Genuine, powerful images are increasingly dismissed as artificial—a trend already underway. It may not be long before AI-generated motion footage becomes widespread, further blurring the line between reality and fabrication. With social media platforms like X heavily promoting AI graphics, meaningful solutions appear increasingly elusive.

I do not want to be a part of a world that has become lobotomized to reality by social media and AI slop. I do not want my children to grow up in a world completely detached from reality.

Of course. But then Hoover was no ordinary pilot. Like being Chuck Yeager's backup pilot in the Bell X-1 program and flying chase for Yeager during the Mach 1 flight. My favorite routine of his was while he was flying the twin engine Shrike Commander. He'd shut down both engines and do acro during the glide to the runway.

@basicbill The problem is it's not legal for most planes to do this, it's not about being first, it's about not crashing when things go wrong.

Today I flew out to Byron to visit the Patriots jet team, they have a bunch of L-39s for display and some other cool aircraft. But they also have a great education program.

@DJSnM Nothing that Bob Hoover didn't do many years ago. youtu.be/4HT2_6BNrdk?si…

They have a North American Sabreliner which they use to give airliner pilots unusual attitudes training, which means they get to do barrel rolls in this business jet.

@NWSTucson Nice chart.

Scattered valley rain & mountain snow showers (snow level 7000 feet) this morning as a cold front moves through. Mostly sunny to partly cloudy this afternoon with high temperature 4 to 8 degrees below normal. #azwx

I tracked the forecasts for yesterday's Texas / Louisiana Severe Weather Event for 7 days. Here is the summary. A significant severe weather event struck East Texas, Louisiana, and southwest Mississippi on Saturday, February 14–15, 2026, producing tornado warnings across multiple counties, a peak wind gust of 63 knots (72 mph) at Lake Charles, and widespread damaging winds along a fast-moving squall line. The event was tracked for seven days using multiple weather models. WINNER: ECMWF IFS ECMWF correctly identified Saturday as the main threat day from Day 5-6, held that solution steadily through the forecast period, and accurately predicted inland mid-60s°F dewpoints — the critical moisture element that other models placed too conservatively offshore. GFS flip-flopped through three different timing solutions before finally correcting. AIGFS (AI-GFS) inherited GFS’s conservative bias throughout, providing no independent predictive value. StormNet-v4, a purpose-built ML tornado model, earns special recognition for its early tornado probability signal in the exact corridor where warnings were ultimately issued. Key Takeaway: In this event, the traditional European model (ECMWF) significantly outperformed the US AI-enhanced model (AIGFS) in medium-range skill. However, a purpose-built AI tornado detection model (StormNet-v4) showed genuine value that traditional guidance lacked. What Happened A strong positively-tilted shortwave trough moved rapidly eastward across Texas and Louisiana on Saturday, February 14, dragging a cold front through the region. The system tapped abundant Gulf moisture with surface dewpoints in the mid-60s°F — exceptional for mid-February. A pre-frontal tornado threat window opened during the afternoon (approximately 3-7 PM CST) as discrete cells formed ahead of the main line, followed by an intense squall line passage during the evening and overnight hours. Confirmed Impacts Tornado Watch issued for SE Texas until 9 PM CST; extended to Louisiana/Mississippi until 2 AM Sunday Tornado Warnings confirmed in: Gregg & Harrison counties TX (Longview/Hallsville corridor), Nacogdoches & San Augustine counties TX (Melrose/Chireno area), Galveston & Brazoria counties TX, Bienville & Red River parishes LA Peak wind gust: 63 knots / 72 mph at KLCH Lake Charles (0253Z Feb 15) — exceeds severe criteria KIAH Houston: 34-knot peak gusts during +TSRA squall passage, 1-1.5” rainfall KSHV Shreveport: Squall line arrived near midnight, +TSRA with 0.5SM visibility in fog Dense fog preceded event at all three stations, confirming saturated boundary layer. Notable Moments in the Model Evolution Day 5-6 (Feb 9): The critical divergence point. ECMWF placed its solution firmly on Saturday with inland moisture. GFS placed the threat on Friday with moisture remaining offshore. This split defined the rest of the forecast period. Day 4-5 (Feb 10): GFS/NWS Houston cleared Friday entirely, shifted threat to Sunday. ECMWF did not move. StormNet-v4 independently showed 20-40% tornado probabilities for Saturday evening SE TX/SW LA corridor — holding firm when traditional guidance was retreating. Day 2 (Feb 12): SPC still at 'Predictability Too Low' despite both models now converging on Saturday. Conservative approach appropriate for public messaging, but ECMWF had been right about the Saturday solution for four days at this point. Day 1 evening (Feb 13, 1730Z): SPC upgraded from Marginal to Slight Risk — the final pre-event category. Tornado language strengthened on Day 1 to 'strong tornado or two possible.' This validated StormNet-v4's signal from 4 days prior. A complete copy of this report is available here: docs.google.com/document/d/1Fn…

Is it available (or will be ) to the public? I've been tracking and scoring wx models recently for storm events. After your post a few days ago regarding tomorrow's TX , LA storm system I threw that post into the mix. The result: ECMWF got the timing, moisture placement, and storm mode right from 5-6 days out. GFS was initially too dry and too fast (Friday vs Saturday). AIGFS inherited the GFS bias. StormNet-v4 gets credit for the early tornado signal that SPC eventually acknowledged in their Day 3 text.

With the v4 release, we'll also be releasing a site with extensive evaluation stats & 2024-2025 validation runs of the latest version!

This event is one of the events we focus on when evaluating future StormNet versions.. The soon-to-be-released StormNet-v4 nailed this thing from 5+ days out. I'm so excited about this model (v4). It feels special. Will be available to all very soon! 🙂