In article ,
Len Wood writes:
What chance of winter mean temp going below zero C in
Wanstead?
-----------------------------------------------------------------
Zero if you ask me.
snip

Well it very nearly happened in 1962-3, so though very unlikely it's not
impossible.
--
John Hall "He crams with cans of poisoned meat
The subjects of the King,
And when they die by thousands G.K.Chesterton:
Why, he laughs like anything." from "Song Against Grocers"

On Tuesday, December 17, 2013 7:23:56 PM UTC, John Hall wrote:
In article , Scott W writes: Further to Dave Cornwell's comment a few weeks back that people mostly remember a winter through the amount of days with snow lying I decided to use the data I produced for my winter forecast and try to find out what snow cover has been like in my area going back to 1946/47 - the first year of the original snow survey. I then divided the snow lying days by the winter mean to give the index. I realise there is the work of Bonacina to consider but as this is national I wanted to look more indepth That's very interesting. Thanks for going to the trouble of producing it. If snow cover is the main interest, then why divide the number of days with lying snow by the mean temperature? Because the mean for 1962-3 was 0.2C and that for 1946-7 was 1.3C, the division massively inflates the index for 1962-3 compared to the earlier winter. (And what would you have done if the mean for 1962-3 had come out negative, as it very nearly did? [Added later: Having now read the follow-ups, I see that Norman has made the same point, only rather better.] There was snow lying through much of the first two weeks of March in 1946-7. It looks as though that is included in the number of days of snow lying, but how is the winter mean value defined (and the winter rainfall total)? Was that the mean for the "traditional" three months of DJF? If so then it seems inconsistent. I'm surprised that 2010-11 doesn't make the top 20, given how cold and snowy December 2010 was over much of the country. -- John Hall "He crams with cans of poisoned meat The subjects of the King, And when they die by thousands G.K.Chesterton: Why, he laughs like anything." from "Song Against Grocers"

Hello John,

I have modified the spreadsheet and used Kelvin instead of mean temp. I have divided snow lying by average mean Kelvin then multiplied by 100 to give it a chunky figure.
The temperature stats refer to the meteorological winter - December, January, February. The snowfall stats refer to October to May. I realise this is not consistent but I was merely trying to give a 'perception' of each winter. One could argue that I should include temp stats and rainfall to go with the snowfall data ie October to May - but then this would introduce its own inconsistencies in not being 'meteorological winter'. With a couple of exceptions all my snowfall occurs December to March. Further feedback welcomed.

Here is a link to the new spreadsheet
http://sdrv.ms/1hZwtkG

With a couple of exceptions all my snowfall occurs December to March. Further feedback welcomed.



Sorry, make that December to February

--

------------------------------
This email was sent by a company owned by Pearson plc, registered office at
80 Strand, London WC2R 0RL. Registered in England and Wales with company
number 53723.

In article ,
Scott W writes:
I have modified the spreadsheet and used Kelvin instead of mean
temp. I have divided snow lying by average mean Kelvin then
multiplied by 100 to give it a chunky figure. The temperature stats
refer to the meteorological winter - December, January, February.
The snowfall stats refer to October to May. I realise this is not
consistent but I was merely trying to give a 'perception' of each
winter. One could argue that I should include temp stats and
rainfall to go with the snowfall data ie October to May - but then
this would introduce its own inconsistencies in not being
'meteorological winter'. With a couple of exceptions all my
snowfall occurs December to March. Further feedback welcomed.

The trouble with dividing by degrees Kelvin is that there will then be
very little difference between cold winters and mild ones. Dividing by
274 (1C) in one case and 278 (5c) in the other won't affect the result
that much. So whereas dividing by the Celsius temperature meant that
temperature tended to dominate over snowfall, dividing by the Kelvin
value will result in the reverse.

If you want both snowfall and temperature to have roughly equal weight,
then I think you need to experiment until you get something that "feels"
right. I'm not sure that dividing one value by the other is the way to
go. If you find that the median number of days with snow lying is 10,
say, and the median (or mean - it won't make much difference here)
Celsius temperature over all winters is 4, say, then how about an index
L-(10/4)*T, where L is the number of days with snow lying and T the mean
Celsius temperature of the winter in question? That looks as if it might
give reasonable weightings to the two factors, and will mean that the
median winter will have an index of zero. (I think that for days of snow
lying the median value is more "typical" than the mean, as the mean is
bumped up by the occasional very snowy winter like 1962-3.)

On second thoughts my suggestion would probably still over-emphasise
snowfall in relation to temperature, as the number of days of snow lying
can be anywhere between zero and 60-70, whereas the limits on winter
mean temperature are probably between about zero and 6-7. So the range
of values of the former is about ten times that of the latter. To
compensate for that, a possible index might be L - 10*T. So a very mild
and non-snowy winter would have a value of about -60 or -70, and a very
cold and snowy one like 1962-3 would have a value of +60 or +70. An
average winter would be about 10 - 10*4 = -30. If you'd rather the
average index was close to zero you could use 30 + L - 10*T.
--
John Hall "He crams with cans of poisoned meat
The subjects of the King,
And when they die by thousands G.K.Chesterton:
Why, he laughs like anything." from "Song Against Grocers"

John Hall wrote:

In article ,
Scott W writes:
I have modified the spreadsheet and used Kelvin instead of mean
temp. I have divided snow lying by average mean Kelvin then
multiplied by 100 to give it a chunky figure. The temperature stats
refer to the meteorological winter - December, January, February.
The snowfall stats refer to October to May. I realise this is not
consistent but I was merely trying to give a 'perception' of each
winter. One could argue that I should include temp stats and
rainfall to go with the snowfall data ie October to May - but then
this would introduce its own inconsistencies in not being
'meteorological winter'. With a couple of exceptions all my
snowfall occurs December to March. Further feedback welcomed.

The trouble with dividing by degrees Kelvin is that there will then be
very little difference between cold winters and mild ones. Dividing by
274 (1C) in one case and 278 (5c) in the other won't affect the result
that much. So whereas dividing by the Celsius temperature meant that
temperature tended to dominate over snowfall, dividing by the Kelvin
value will result in the reverse.

If you want both snowfall and temperature to have roughly equal weight,
then I think you need to experiment until you get something that "feels"
right. I'm not sure that dividing one value by the other is the way to
go. If you find that the median number of days with snow lying is 10,
say, and the median (or mean - it won't make much difference here)
Celsius temperature over all winters is 4, say, then how about an index
L-(10/4)*T, where L is the number of days with snow lying and T the mean
Celsius temperature of the winter in question? That looks as if it might
give reasonable weightings to the two factors, and will mean that the
median winter will have an index of zero. (I think that for days of snow
lying the median value is more "typical" than the mean, as the mean is
bumped up by the occasional very snowy winter like 1962-3.)

On second thoughts my suggestion would probably still over-emphasise
snowfall in relation to temperature, as the number of days of snow lying
can be anywhere between zero and 60-70, whereas the limits on winter
mean temperature are probably between about zero and 6-7. So the range
of values of the former is about ten times that of the latter. To
compensate for that, a possible index might be L - 10*T. So a very mild
and non-snowy winter would have a value of about -60 or -70, and a very
cold and snowy one like 1962-3 would have a value of +60 or +70. An
average winter would be about 10 - 10*4 = -30. If you'd rather the
average index was close to zero you could use 30 + L - 10*T.

Sorry John, I can't agree with that. Using the temperature expressed in deg C
in this way is akin to saying that a day with a max of 4 deg C is twice as warm
as a day with a max of 2 deg C which, I think you would agree, is nonsense.
Using deg K is the only valid method. The new spreadsheet that Scott has
devised does appear to show a realistic spread in the relative severity of
individual winters.

--
Norman Lynagh
Tideswell, Derbyshire
303m a.s.l.

"Norman" wrote in message ...
John Hall wrote:
On second thoughts my suggestion would probably still over-emphasise
snowfall in relation to temperature, as the number of days of snow lying
can be anywhere between zero and 60-70, whereas the limits on winter
mean temperature are probably between about zero and 6-7. So the range
of values of the former is about ten times that of the latter. To
compensate for that, a possible index might be L - 10*T. So a very mild
and non-snowy winter would have a value of about -60 or -70, and a very
cold and snowy one like 1962-3 would have a value of +60 or +70. An
average winter would be about 10 - 10*4 = -30. If you'd rather the
average index was close to zero you could use 30 + L - 10*T.

Sorry John, I can't agree with that. Using the temperature expressed in deg C
in this way is akin to saying that a day with a max of 4 deg C is twice as warm
as a day with a max of 2 deg C which, I think you would agree, is nonsense.
Using deg K is the only valid method.

I don't agree, Norman.
I think John's method is perfectly valid, as the 'index' (like a temperature scale) has an
arbitrary zero point which can be chosen at will and only differences in its value (not ratios)
are meaningful. A max of 4C compared to a max of 2C will just subtract '2 degrees worth' of
severity from the index, ie 20 points on his scale.

And this will work exactly the same whether you use deg K or deg C - it just changes the zero
point of the index.

Subtracting (some multiple of) the temperature is much better than dividing, since it makes the
index linear in both L and T.

Gavino wrote:
"Norman" wrote in message ...
John Hall wrote:
On second thoughts my suggestion would probably still over-emphasise
snowfall in relation to temperature, as the number of days of snow lying
can be anywhere between zero and 60-70, whereas the limits on winter
mean temperature are probably between about zero and 6-7. So the range
of values of the former is about ten times that of the latter. To
compensate for that, a possible index might be L - 10*T. So a very mild
and non-snowy winter would have a value of about -60 or -70, and a very
cold and snowy one like 1962-3 would have a value of +60 or +70. An
average winter would be about 10 - 10*4 = -30. If you'd rather the
average index was close to zero you could use 30 + L - 10*T.
Sorry John, I can't agree with that. Using the temperature expressed in deg C
in this way is akin to saying that a day with a max of 4 deg C is twice as warm
as a day with a max of 2 deg C which, I think you would agree, is nonsense.
Using deg K is the only valid method.

I don't agree, Norman.
I think John's method is perfectly valid, as the 'index' (like a temperature scale) has an
arbitrary zero point which can be chosen at will and only differences in its value (not ratios)
are meaningful. A max of 4C compared to a max of 2C will just subtract '2 degrees worth' of
severity from the index, ie 20 points on his scale.

And this will work exactly the same whether you use deg K or deg C - it just changes the zero
point of the index.

Subtracting (some multiple of) the temperature is much better than dividing, since it makes the
index linear in both L and T.

--------------------------------------------------------------------
I can't tell you how pleased I am to see people once again thinking
about the weather in an intersting way and coming up with some varied
and interesting ideas on UKSW. Having tried some similar things I
realise how difficult it is. I agree with Scott that there is more to a
severe winter than the superficial perceptions people have and also how
difficult it is to represent those perceptions mathematically.
Dave

Gavino wrote:

"Norman" wrote in message
...
John Hall wrote:
On second thoughts my suggestion would probably still over-emphasise
snowfall in relation to temperature, as the number of days of snow lying
can be anywhere between zero and 60-70, whereas the limits on winter
mean temperature are probably between about zero and 6-7. So the range
of values of the former is about ten times that of the latter. To
compensate for that, a possible index might be L - 10*T. So a very mild
and non-snowy winter would have a value of about -60 or -70, and a very
cold and snowy one like 1962-3 would have a value of +60 or +70. An
average winter would be about 10 - 10*4 = -30. If you'd rather the
average index was close to zero you could use 30 + L - 10*T.

Sorry John, I can't agree with that. Using the temperature expressed in deg
C in this way is akin to saying that a day with a max of 4 deg C is twice
as warm as a day with a max of 2 deg C which, I think you would agree, is
nonsense. Using deg K is the only valid method.

I don't agree, Norman.
I think John's method is perfectly valid, as the 'index' (like a temperature
scale) has an arbitrary zero point which can be chosen at will and only
differences in its value (not ratios) are meaningful. A max of 4C compared to
a max of 2C will just subtract '2 degrees worth' of severity from the index,
ie 20 points on his scale.

And this will work exactly the same whether you use deg K or deg C - it just
changes the zero point of the index.

Subtracting (some multiple of) the temperature is much better than dividing,
since it makes the index linear in both L and T.


You're right, Gavino. I obviously didn't read John's post correctly. I thought
he was still suggesting dividing by the temperature but he is not. I agree that
in John's proposed method it doesn't matter whether deg C or deg K are used -
apologies John :-)

--
Norman Lynagh
Tideswell, Derbyshire
303m a.s.l.

Hi

I too applaud Scott's attempt to produce a Winter Index.

But does anyone have the article in the Weather magazine (maybe from the 1970's) on the same topic of "Winter Index"? I'm sure that it followed a similar article on a "Summer Index".

As a Summer Index would require temperature, precipitation and sunshine, then a winter index can only work if you include both precipitation (ideally snowfall) and temperature anomalies.

I tried to do this earlier this year in a blog I called "Central England Snowfall" in which I attempted to link CET and EWR. What it showed was: If you like snow 1946/47 is the winter to beat (although 1978/79 came close especially in the north), and if you like cold 1962/63 was intensely cold and rather dry.

Bruce.

PS Here's a link to my "Central England Snowfall" blog.

http://xmetman.wordpress.com/2013/08...land-snowfall/

In article ,
Norman writes:
John Hall wrote:

On second thoughts my suggestion would probably still over-emphasise
snowfall in relation to temperature, as the number of days of snow lying
can be anywhere between zero and 60-70, whereas the limits on winter
mean temperature are probably between about zero and 6-7. So the range
of values of the former is about ten times that of the latter. To
compensate for that, a possible index might be L - 10*T. So a very mild
and non-snowy winter would have a value of about -60 or -70, and a very
cold and snowy one like 1962-3 would have a value of +60 or +70. An
average winter would be about 10 - 10*4 = -30. If you'd rather the
average index was close to zero you could use 30 + L - 10*T.

Sorry John, I can't agree with that. Using the temperature expressed in deg C
in this way is akin to saying that a day with a max of 4 deg C is twice as warm
as a day with a max of 2 deg C which, I think you would agree, is nonsense.

I agree if you consider the temperature in isolation. And the effect if
you divide one value by the other would clearly be totally wrong. But I
don't think it's a problem with the formula above, because the constant
of 10 has been chosen to be compatible with the possible range of values
and because the lower end of the range of possible values happens, very
conveniently, to be zero. When I have time I'll use Scott's values of L
and T with my formula to see what the table would become and if it looks
reasonable.

I now see that Gavino has said that much more clearly than I've managed.
My thanks to him.

Using deg K is the only valid method. The new spreadsheet that Scott has
devised does appear to show a realistic spread in the relative severity of
individual winters.

I haven't seen the results using K temperatures yet, but it seems to me
that they would lead to the importance of temperature to be
underestimated compared to the number of days with snow lying. If
there's a problem it will only show up for the occasional winters that
are mild overall but still have a large number of days of snow and those
that are cold but non-snowy.
--
John Hall "He crams with cans of poisoned meat
The subjects of the King,
And when they die by thousands G.K.Chesterton:
Why, he laughs like anything." from "Song Against Grocers"