So I'm looking at the hourly UKMO forecasts for tomorrow (for Ely, East
Cambs) and currently it's showing a 3o% rain probability for each hour
from 1200-2000 inclusive (but by 2100 it's miraculously dropped to 5%).

I assume that this means that in any one of these hours there's a 30%
probability of more than a token amount of rain (though I don't know
what the threshold might be). So in that 9-hour afternoon stretch then
surely that means that - barring statistical flukes and recognising that
each hour is nominally independent - there's something approaching 100%
probability of some rain. Or am I misinterpreting the way the forecast
is presented?

In message , JGD
writes
So I'm looking at the hourly UKMO forecasts for tomorrow (for Ely, East
Cambs) and currently it's showing a 3o% rain probability for each hour
from 1200-2000 inclusive (but by 2100 it's miraculously dropped to 5%).

I assume that this means that in any one of these hours there's a 30%
probability of more than a token amount of rain (though I don't know
what the threshold might be). So in that 9-hour afternoon stretch then
surely that means that - barring statistical flukes and recognising
that each hour is nominally independent - there's something approaching
100% probability of some rain. Or am I misinterpreting the way the
forecast is presented?

You're assuming that each hourly event is totally independent of the
others. You're not allowing for the possibility of a single outbreak of
rain lasting long enough to cover several successive hours. In the
extreme case, a single outbreak with a 30% probability of occurring
could last from before 12:00 till after 20:00.

But even allowing for that, the forecast does seem surprising.
--
John Hall
"Hegel was right when he said that we learn from history
that man can never learn anything from history."
George Bernard Shaw (1856-1950)

On 26/07/2018 08:57, John Hall wrote:

You're assuming that each hourly event is totally independent of the
others. You're not allowing for the possibility of a single outbreak of
rain lasting long enough to cover several successive hours. In the
extreme case, a single outbreak with a 30% probability of occurring
could last from before 12:00 till after 20:00.

But even allowing for that, the forecast does seem surprising.

Yes, I guess you're right. But a statistician looking at those numbers
at face value might say that a 30% probability of rain is a 70% chance
of no rain. And so the probability of no rain in every one of the 9
consecutive hours is 0.7^9 or around 4%. So rain at some point had a 96%
probability. Admittedly my stats is a bit rusty nowadays :-(

NB This wasn't having a go at UKMO - I was genuinely wondering how to
interpret the way the forecast for rain was presented.

Of course, as of the 0700 forecast update today, the picture has changed
again with a 60% rain probability but only for the 1400 and 1500 hours
(so 84% probability of rain in one or both of those hours) plus 30%
remaining for 1800-2000 or 66% rain probability later. So, again a 95%
probability of rain sometime today.[But still with an overnight low of
20C :-( :-( :-( ]

JGD wrote:

On 26/07/2018 08:57, John Hall wrote:

You're assuming that each hourly event is totally independent of
the others. You're not allowing for the possibility of a single
outbreak of rain lasting long enough to cover several successive
hours. In the extreme case, a single outbreak with a 30%
probability of occurring could last from before 12:00 till after
20:00.

But even allowing for that, the forecast does seem surprising.

Yes, I guess you're right. But a statistician looking at those
numbers at face value might say that a 30% probability of rain is a
70% chance of no rain. And so the probability of no rain in every one
of the 9 consecutive hours is 0.7^9 or around 4%. So rain at some
point had a 96% probability. Admittedly my stats is a bit rusty
nowadays :-(

NB This wasn't having a go at UKMO - I was genuinely wondering how to
interpret the way the forecast for rain was presented.

Of course, as of the 0700 forecast update today, the picture has
changed again with a 60% rain probability but only for the 1400 and
1500 hours (so 84% probability of rain in one or both of those hours)
plus 30% remaining for 1800-2000 or 66% rain probability later. So,
again a 95% probability of rain sometime today.[But still with an
overnight low of 20C :-( :-( :-( ]

Without quantifying the amount of rain these probability forecasts are
not at all helpful. For example 'rain' can be just a few spots of
drizzle or it can be 30 mm in half an hour. To be useful the
probability forecasts need to be worded along the lines of 'probability
of 5mm of rain in previous 3 hours', for example

--
Norman Lynagh
Tideswell, Derbyshire
303m a.s.l.
https://peakdistrictweather.org
Twitter: @TideswellWeathr

On 26/07/2018 10:16, Norman Lynagh wrote:

Without quantifying the amount of rain these probability forecasts are
not at all helpful. For example 'rain' can be just a few spots of
drizzle or it can be 30 mm in half an hour. To be useful the
probability forecasts need to be worded along the lines of 'probability
of 5mm of rain in previous 3 hours', for example


Agreed and shame that this isn't done - I guess it wouldn't fit on the
current graphic presentation.

In the event, there's been no rain here today at all other than a
handful of heavy drops at one point) and no sign of any for the rest of
the day. To be fair, a line of thundery showers did pass 10-30 miles
west of here mid-afternoon.

Interesting that a new UKMO local forecast seems to be appearing almost
every hour. Is this a new feature that I missed or are they working
overtime because of the very hot spell? It's virtually nowcasting though
sadly still not very accurate for rainfall prediction even one hour ahead!

The frequent updates are also interesting in seeing just how much the
forward detail (eg 3-24 hours) changes from one hour's forecast to the
next, especially again for cloud & rain predictions, though temperature
is rather more stable.

In message , JGD
writes
The frequent updates are also interesting in seeing just how much the
forward detail (eg 3-24 hours) changes from one hour's forecast to the
next, especially again for cloud & rain predictions, though temperature
is rather more stable.

The thing I notice most is how different the temperatures in those
hourly computer-generated forecasts are from the maxima and minima
included as part of the textual human-generated forecasts. In hot summer
weather, such as we are having now, the computer-generated temperatures
seem to be substantially lower. In cold winter weather they seem to be
higher. The human-generated figures normally seem to be far closer to
reality.
--
John Hall
"Hegel was right when he said that we learn from history
that man can never learn anything from history."
George Bernard Shaw (1856-1950)

Another curiosity I've just noticed for this afternoon is a predicted
rainfall probability comparison for Ely and Cambridge (just 15 miles
apart, both of course well inland) and both on the 1200 forecast

Time Ely Cambridge

1500 10% 50%
1600 50% 5%
1700 50% 5%
1800 50% 5%
1900 60% 60%
2000 60% 30%
2100 60% 10%
2200 70% 10%
2300 20% 10%

(Sorry don't know how well the formatting will hold up!)

I guess the storms will be coming broadly from the S/SW and travelling
at eg 10mph so Cambridge might be one hour ahead of Ely for a given
storm cell. NB Cambridge is approx 15 miles SSE of Ely.

But it's odd - apparently high resolution predictions but not too much
correlation between two places that are quite close neighbours.