On Saturday, December 7, 2013 7:35:05 PM UTC, Togless wrote:
"exmetman" wrote:



...

I would have thought that the UKMO would have to be extremely

careful about how the latest data is derived and have plumped

for plotting the 10 year average (2003-2013) at the end of the

10 years i.e. 2013.



Perhaps its worth getting in touch with them to find out.



I did that some time ago and the answer was:



"The handling of 'tails' in smoothing algorithms varies. In this case, the

mean of the final few values of the series is used to extend the series so

that the Gaussian-weighted filter can be applied to the end of the original

series. The principal effect of this is to prevent the smoothed curve from

ending on an anomalously high or low value (since it is tending towards the

mean of the final values).



Running means are sometimes shown ending short of the end of a series, and

this is being considered here. But as I'm sure you understand, in this

situation such an action could be misinterpreted."

================================================== ===========================
I too have come in late here, but it seems obvious to me (and the UKMO I hope), that if you are taking 10yr running means then you cannot use any number other than 10. Your last 10 year period is 2003-2012. So the last running mean on your graph would be centred on 2008. Nothing after that.
Similarly at the beginning of the record. Say for example the record began at 1950. The first 10 year period is 1950-1959, so the first running mean on the graph would be 1954.

Len
Wembury

Len Wood wrote:
On Saturday, December 7, 2013 7:35:05 PM UTC, Togless wrote:
"exmetman" wrote:



...

I would have thought that the UKMO would have to be extremely
careful about how the latest data is derived and have plumped
for plotting the 10 year average (2003-2013) at the end of the
10 years i.e. 2013.
Perhaps its worth getting in touch with them to find out.


I did that some time ago and the answer was:



"The handling of 'tails' in smoothing algorithms varies. In this case, the

mean of the final few values of the series is used to extend the series so

that the Gaussian-weighted filter can be applied to the end of the original

series. The principal effect of this is to prevent the smoothed curve from

ending on an anomalously high or low value (since it is tending towards the

mean of the final values).



Running means are sometimes shown ending short of the end of a series, and

this is being considered here. But as I'm sure you understand, in this

situation such an action could be misinterpreted."

================================================== ===========================
I too have come in late here, but it seems obvious to me (and the UKMO I hope), that if you are taking 10yr running means then you cannot use any number other than 10. Your last 10 year period is 2003-2012. So the last running mean on your graph would be centred on 2008. Nothing after that.
Similarly at the beginning of the record. Say for example the record began at 1950. The first 10 year period is 1950-1959, so the first running mean on the graph would be 1954.

Len
Wembury

--------------------------------------
How do you determine the significance of pairs of data that run at
alternatively opposite extremes? Does the sequence end at infinity or
does one of the values come to an end. e.g the sequence:,
L,D,L,D,L,D,L,D................ infinity?

On Saturday, December 7, 2013 9:58:10 PM UTC, Dave Cornwell wrote:
Len Wood wrote:

================================================== ===========================

I too have come in late here, but it seems obvious to me (and the UKMO I hope), that if you are taking 10yr running means then you cannot use any number other than 10. Your last 10 year period is 2003-2012. So the last running mean on your graph would be centred on 2008. Nothing after that.

Similarly at the beginning of the record. Say for example the record began at 1950. The first 10 year period is 1950-1959, so the first running mean on the graph would be 1954.



Len

Wembury



--------------------------------------

How do you determine the significance of pairs of data that run at

alternatively opposite extremes? Does the sequence end at infinity or

does one of the values come to an end. e.g the sequence:,

L,D,L,D,L,D,L,D................ infinity?

--------------------------------------------------
Can't see where infinity comes into what I was referring to Dave.

One is working on a data series in this case, which is a well defined time series. For example mean annual temps between 1950 and 2012.


Len

On 07/12/2013 18:11, exmetman wrote:
Hi

I know I have come in a bit late on the debate about the Met Office graph but my take on it is that if it were a 10 year centre moving average, then the smoothing line would have stopped in 2008, and because it dosent it strongly suggests that the latest values are for the period 2003-2013, and as such are as valid as the rest of the moving average series.

No. That isn't what they have done. The smoothed line is a weighted low
pass filter kernel applied to the data combined with some sort of ad hoc
extension to the end of the series based on recent means.

Since there was one aggressively low spike recently this biasses the
results. I get more or less the same graph but slightly spikier using a
basic boxcar average and 11 bins. If I use a Gaussian filter 1/e at 5
years exp(-t^2/5) I could reproduce their curve more or less exactly.

Their full dataset for seasonal and monthly is available online back to
1659. One interesting observation is that the years around 1730 were
actually as warm as it is now if the series is accurate back then.

http://hadobs.metoffice.com/hadcet/data/download.html

I would have thought that the UKMO would have to be extremely careful about how the latest data is derived and have plumped for plotting the 10 year average (2003-2013) at the end of the 10 years i.e. 2013.

No. That isn't what they have done. The values plotted for years closer
to the end of series than 5 years are biased (compromised) by the
unstated assumptions they have made to extend the time series. This is
actually rather bad practice given that the deniers will attempt to put
their own twisted spin on it to pretend that AGW isn't happening.

Perhaps its worth getting in touch with them to find out.

Bruce.



--
Regards,
Martin Brown

On 09/12/2013 08:38, Martin Brown wrote:

basic boxcar average and 11 bins. If I use a Gaussian filter 1/e at 5
years exp(-t^2/5) I could reproduce their curve more or less exactly.

Ooops typo ^^^^^^^^ should read exp(-(t/5)^2)

--
Regards,
Martin Brown