Weather forecasts - or not?

And after every program thanking us for "supporting" the channel, with money!

I've lost count of the times when I could look across to the other side of the valley (maybe 1/2 a mile away) and see it is raining heavily there whilst it remains dry here. I assume there are plenty of times when the opposite must be true, but if it's raining heavily here, I can't see across the valley. When weather is so localised, it is impossible for a forecast to be anything other than a set of probabilities.

Thank you for that image.

Usenet _is_ the indoor equivalent of listening to drunks yelling nonsense at passers by.

Hexactly. What we're dealing with is unimaginable amounts of data, being processed at **literally unimaginable** speeds.

The meteorological models can't keep up, hence our "weather people" (especailly in the BEEB's Long Range forecast which is my own favourite forecast -- 'Weather for the week ahead" they call it) *emphasise* that "one model says this, another says that, so we don't yet really know what's going to happen 3 or 4 days from now." (Such frank uncertainty makes Louise and Elizabeth so much more acceptable compensation!)

John

Terse summaries do not capture what is going on.

You need to look at the simulations, the Doppler radar, to get a better picture.

This is how I cross-check my Canadian weather, with Norwegian-made forecasts. The current day shows "dry", but one thing about the Norwegians is they don't mind a bit of mist, so a 1mm forecast means rain, 0mm could be mist or clear.

If two terse summaries are written from the "same" simulation model reported results, then you expect the textual forecasts to match. But if one model was done on a Norwegian computer, another model on a UK computer, the textual summaries can and will be entirely different on "sensitive days" where bad weather is only a scant few miles from your location.

Also, there is some amount of AI being used today, for the summaries. It has been discovered that the AI training data, with the "hidden patterns" in it, can be used to augment the simulation. The AI has been awarded a "more accurate" designation for the summaries it writes by looking at the simulation results. Humans try to do this too (use local knowledge to correct a simulation), but the AI is better at this game.

Paul

I like the Norwegian service, myself. Example.

Plug your location in here.

If two summaries come from two different simulations, well of course the summaries won't agree.

If divergent summaries come from the *same* sim, that's bad!

You should examine the American practice of putting up six different simulations (different sources) of Hurricane Tracks and landfall determinations. They then use a Ouija board to establish a final result. "Well, I like the Blue curve because it matches my suit." Science.

Weather is a three dimensional chaotic system with insufficient recordings of initial conditions. What is not to like, exactly ?

Paul

Weather forecasters use different models which give different results. This page allows you to compare some of them. They rarely agree completely.

Some models not only use different calculations but their grid size may be different and this can make a forecast more/less specific for a given locality.

The time slice that a given forecast covers may vary too.

In addition, different forecasters will recalculate at a different frequency and I vaguely recall Apple Weather recalculates every minute for rain in the next hour.

I read a press story that the BBC/Met Office used to have a policy of deliberately exaggerating forecast rainfall. It was something to do with the greater upset viewers got into if it rained when it wasn't forecast than the other way round.

I think the point is that persistence forecasting for the following day was on average more accurate in those days than a forecast based on a model. I doubt that's been true for some time now.

Sorry. Sent that twice(ish)

The main problem with modelling is that crap in = crap out but because the answers has been generated by computer many people believe it to be true and shouldn't question it.

Crap in = crap out can already be seen with AI search engines. If some complete BS is shared enough on social media AI will repeat it as the correct answer to a search question.

AIs are basically gigantic expert systems, which are unfortunately not trained by experts.

... and as applied to weather forecasts we have fewer actual observations being fed into the forecasting computers. There used to be light vessels and weather observation ships all aound the oceans but no loner. It feels as if there are fewer on land as well but I'm not sure about that.

...but no longer...

I see auto weather stations located alongside major roads such as the A1. Does anyone know if the auto-observations can be viewed online?

I believe the accuracy has lessened since they changed from whom they sourced their data about 3 (or more) years ago. Now they concentrate on where the weather changes rather than where the population is. Then the forecast spends the most amount of time concentrating on what might happen in 2,3,4 or more days time which inevitably changes. Annoyingly the weather presenters spend a lot of time talking to the map rather than facing the camera.

But the Met Office doesn't seem to use them. They have only three (I think) places in East Anglia which they use.

They have been in partnership with MeteoGroup for a few years but are going back to the UK Met Office shortly. Will forecasts improve?