[conspire] [OT] weather prediction (was: Re: Electronics Flea Market 2019)

[Michael Paoli]

> From: "Rick Moen" <rick at linuxmafia.com>
> Subject: Re: [conspire] Electronics Flea Market 2019
> Date: Mon, 4 Mar 2019 17:24:34 -0800

> With all well-owed respect for Wunderground.com, it's not actually
> possible to reliably predict weather five days in advance, as weather is
> a chaotic system (in the mathematical, Bufferfly Effect sense of that
> term).

Well, yes (mostly) and no (a bit - under certain specific circumstances).

Many moons ago (think early 1970s or so and before), weather prediction was
pretty much limited to about 3 days out.  Anything beyond that was
mostly highly inaccurate and typically educated guesses, at best.

But ... enter lots of data (satellites, telecommunications, etc.),
and advancing mathematical models.

So, by, oh, maybe very roughly about 1972 or so, *theoretically* significantly
longer better forecasts were possible.  The problem was technological
limitations ... lots of data, really good modeling, but, uhm, ... yeah,
computer processing power.  I mean if you wanted a pretty accurate
4 or 5 day forecast ... sure, ... would just take 2 weeks (or much more!)
to run it on the computer cluster to get the forecast - not so useful ...
if you could even manage enough storage for the computer to handle it.

But technology evolved ... most notably the computers (and also still lots
more data).  As time progressed, it became possible to make *fairly*
accurate predictions out to about 5 days, and those became pretty
common ... still far from perfect, and a non-trivial percentage
turning out "wrong", but at least much closer statistically to actual
weather than the near wild guesses of years earlier.

And technology (computers, modeling, data, etc.) continue to advance.

Forecast out to 7, sometimes 10 days became semi-common.  Okay, sure,
beyond 5 days, not all that accurate, but still significantly better
than totally unavailable before, and a large percentage of the time
fairly accurate - at least to a first order approximation.

And, along with that, and weather science, and loads of historical data,
etc. ...

Pretty unusual/rare (at least for *most* locations), but once in a
great while, certain climactic conditions / weather patterns will
develop that are *highly* predictable.  So, *sometimes* they can in fact
accurately give a forecast that's good for up to as much as 14 days ...
but that's much more the exception than the rule.  So, yeah, once in
a (relatively rare) while, I've seen well qualified meteorologists on the
forecast that goes something like:
"... and that's our forecast for the next 5 (or maybe 7) days, however, the
particular pattern we're in right now is highly well known and predictable,
so you can expect for up to at *least* the next 14 days we'll have" ...
So, yeah, don't hit those patterns that are *that* predictable *that*
often (at least around here - maybe some other places on the planet ...
monsoon season anyone?), but, *sometimes*, for some locations and
conditions, there are accurate forecasts that go to up to about 14
(or more?) days ... but, at least all I've ever seen, that's the rare
exception, rather than the rule.

And I don't think I've yet seen any reputable source stick their necks out
*ever* for anything beyond 14 days under any circumstances (at least that's
the longest I recall ever seeing ... so far).

However, in the vast majority of cases and circumstances, forecasts quickly
degrade in accuracy and only go downhill from there, somewhere around the
3 to 7 day range, with about 5 days +- being typical.
National weather service does a 5 day forecast ... if you ever dig into the
details, you'll also see that forecast gets fuzzier as you get into the
last couple days of it (essentially the error bars on everything, including
the when, start spreading significantly).

Wunderground.com ... 10 days? ... All the time?  Paint me kind'a skeptical.
Perhaps it would be more accurate/useful, if, on each of those days, they
gave a % confidence level.  If they did so, maybe around 8-10 days it would
typically look increasingly like or close to a wild *ss guess.  Even
around 5 to 7 those typically aren't very accurate (but still typically
better than a mere guess or simple historic climactic guess).