Apocalyptic weather

Southwell downgraded to a balmy 38 now.

I'm quite jealous, we only get a pretty puny 25 this week.

I've cancelled my office days this week. There were due to be a terrifying three of them. Would have been proper sweaty. The Met Office has put us under a red alert. 37C, with the possibility of 38C to 40C. Im going to sit it out in my underpants in the home office instead. 40C is as hot as Thailand would get in the hottest season. Anyone without a fan is going to die, or have to sit in a cold bath all day.

I know how much you love weather chat. Anyway, the apps all use different forecasting models to produce the temperatures. A shift of 100 miles or so in weather systems has a big impact on the temps forecast. Obviously the Met Office app uses the Met Office Model, but a lot of the other apps use the European Centre for Medium Range Weather Forecasting (ECMWF) for their forecasts. That's why you'll see variations in temperatures.

Also, all apps use unmoderated raw output for their forecasts (hence why they seem to change so often), whereas TV and radio forecasts are based on a blend of the most respected models (The UK MO, ECMWF and US GFS). The MO has the best high resolution system for the UK, but due to the he complexity and detail it can only forecast out to 3-5 days. ECMWF is also based in Reading; we're a bit of a weather hub here with Reading Uni having one of the top rated meteorology courses in the world, plus a top rated climate science department. I didn't really stand a chance of not being interested in weather, especially as the Met Office also used to be just down the road.

There's a saying in meteorology that it is possible to have a perfect forecast for tomorrow, but it wouldn't be available until three days later as computers cannot cope and data cannot be provided fast enough.

I don’t know about all that but I’m at my parents in the South of France and every other channel is a panel discussion about “LE CANICULE*.”

*’Ow you say, ‘eat wave?

"AI generated forecasting" meaning using proper AI to gather data and produce a forecast, with an LLM to make it read/sound nice I hope?
Because LLMs suck at analysis, and only get worse as their inputs deteriorate in quality due to LLMs.

LLMs are bad at analysing large data sets. Or small ones. And bad at trends.

Odd. because that is not what they are designed to do. Just to look like they are.

Had my annual discussion with my elderly mother when i nipped round after work yesterday that, in this heat, keeping your front door, and patio doors and all the curtains and windows wide open in this heat is a bit mental. She might as well put the heating on.

She won't listen though, obvs.

Send KM round to explain it all to her.

One of the biggest drawbacks is lack of data points. We have good data in certain areas; weather stations, flight data etc. but large gaps elsewhere, particularly over areas like Greenland (which impacts on forecasts for the UK) and areas with low population. They get around this by running the same model multiple times with tiny tweaks to the starting data to produce ensemble forecasts and determine forecast confidence.

As for AI, you can view the model output online and it is ok to a point, but the deterministic models still outperform AI significantly. I'm not sure what you mean by trends? Patten matching or observing if larger scale patterns are trending in a certain way?

I'm on your mums side with this one. We've spent all winter begging for warm sunshine, when it finally shows I'm inviting it in 😎