Re: Why Skeptics Don't See What Climatologists See was Re: [asa] Data doesn't support global warming

From: Dave Wallace <wmdavid.wallace@gmail.com>
Date: Sat Dec 19 2009 - 14:28:08 EST

Rich Blinne wrote:
> Glenn's analysis is sensitive to static bias and thus has a cow when
> he sees air conditioners or techniques that increase static bias.
Yes but only partially. If the air conditioners have been running as
far back as we have records then I can see that anomalies probably fix
that issue. However, if the air conditioners get added and run
sporadically (off some weeks on others as ours does) then I suspect that
still causes a problem. IMO the only way to deal with such heat/cooling
sources is to ignore any months when heating or cooling could cause a
problem. Weather stations in the middle of large parks in urban areas
and not near buildings... probably get corrected just fine by use of
anomalies.

Also depending upon how griding is done if one were simply selecting a
single station to represent all that grid then a bias in such selection
say towards urban areas could well cause problems. Somewhere yesterday
I read that the Russians suspect some of the HadCru data sets of having
had exactly that done???

> *Why use temperature anomalies (departure from average) and not
> absolute temperature measurements?*

Your note explains why we got differences when we looked at anomalies in
the average vrs average anomalies and got very different results.

My net overall from Rich's note is that while I see problems in
averaging temperatures over a large area like Siberia I am not convinced
that looking at anomalies is necessarily correct either.

Read further if you care about why and how I would like to check one way
or the other.

==============
After thinking about use of anomalies and averages, overnight I now
remember that my rules from back in the early 60s when I was working
with a statistician and processing tons of data on an IBM 1410 during my
co-op work term at UniWat. The data was on punched cards so it was in
fact heavy, maybe not tons but weighty so as we got it from the
keypunchers we added new data to our magnetic tapes.

A. never average averages except maybe if the data in each group is very
homogeneous and maybe then you should use weighted averages.

B. another rule was to look up which measure of centrality to use, as
the right mean is data dependent and often means don't work at all and
something else like ANOVA needs to be used. (ANOVA == Fisher's
Analysis of Variance)

C. Also look at numerical calculations using averages with much
suspicion although sometimes they are necessary.

I have spent a couple of hours looking at material on the web about
averages. Summaries follow with more following my signature for those
who care.
> The message here is that you can't average averages.
> If Bill Gates, the founder of Microsoft, walks into a crowded bar ...
> everyone in the bar is instantly a multimillionaire. On average, that is.
One might expect that even the anomaly plots are using averages of
averages in multiple ways. For example, for sites recording more than
one daily temperature, that temp is likely averaged. Possibly multiple
sites within a grid are averaged among themselves and then averaged over
time to create the base anomaly figures, although hopefully each site's
data goes into the average over time for the particular month and then
the total averaged.

 From this discussion above and the web sites on averages it becomes
obvious why an attempt is made to calculate warming trends using a grid
to try to obtain reasonably correlated data with which to compute the
anomaly base and then to compute the trends in temperature. Think about
average computations with one or more bad outliers.

Grids while they might work in relatively uniform areas certainly don't
work well everywhere. For example I lived in a flat part of the Rift
valley that was arid and had moderate temperature (76 F in the day, 55F
at night) yet near us the Rift valley dropped off again to a lower
altitude down by about 300+ meters. The drop off was not gentle but was
a cliff that ran both directions as far as I could see. The lower part
of the Rift valley was much hotter and more humid. Thus a grid cell
covering the boundary would yield very conflicting results. The
escarpment that creates Niagra falls runs for a few hundred klicks in
Ontario and does cause the same issue, we went up and over it very often.

 From the web it also appears that there are different ways of computing
the anomalies, see info below my signature.

In general I would like to see the anomaly approach validated by
1. For each site calculate the linear least squares fit over the period
it has observations
     (or use whatever your favourite curve fitting algorithim that
results in a slope and is also consistent ,as a refinement one could
discard a limited number of outliers and recalculate the fit)

2. Examine the resulting slopes to determine if GW is occurring. The
data which results would a be 365 slopes over n stations.

     -There are brute force and sophisticated ways of looking at large
quantities of data. In the Siberian area Glen was looking at, there are
only about 150 stations so plotting and examining all the slopes is not
a big deal. I have also seen lots of sophisticated 3D stereo data
presentations that would deal with lots of individual points but I
suspect there are good 2d solutions.

But the point being that this is likely simple enough that at least it
is hard to argue over the statistical methods used and at most only one
average was used (other than averaging the daily data from each stations
that record more data). Looking at max and min data in the same manner
would be useful.

ANOVA might probably also do as validation but the data sets would by
very large.

One could argue that this is a lot of extra work but it is the kind of
thing I did for at least limited samples to check the data I was
analyzing in the 60s. If we got things wrong back then, pilots would
tend to crash fighter aircraft, especially if super sonic at say a
1000ft above the deck.

As I said earlier if we get it wrong as to whether or not AGW is
occurring we are likely to kill people, say in the Rift valley. Either
set of policy decisions, if based on wrong science, could have needless
very bad results. This is why Glenn and I care.

Dave W

PS
On averages
> google either average of averages or statistics average of
> averages or Simpson's Paradox. I have just spend an hour and a half
> reading up on this on the web. Some good samples follow:
>
> is average of average accurate
> http://wiki.answers.com/Q/Is_an_average_of_averages_accurate
>
> different kinds of measures of centrality and when to use them...
> http://en.wikipedia.org/wiki/Average
>
> The message here is that you can't average averages.
> http://plus.maths.org/issue33/features/stickland/index.html

  On Anomalies

> Note that an anomaly is a numerical value which indicates how far a
> measurement varies from the average. The average is determined for a
> significantly large sample of data and is then used to calculate the
> anomaly for a specific measurement in the data set. For example the
> base average used to calculate the anomalies for the sea surface
> temperature data set was calculated using the data from 1950 to 1979.
> *average temperatures are reduced to anomalies from the period with
> best coverage (1961-90)*.

http://www.appinsys.com/GlobalWarming/GW_Part2_GlobalTempMeasure.htm
>
> *Calculating Global Averages*
>
> Different agencies use different methods for calculating a global
> average. After adjustments have been made to the temperatures, the
> temperatures at each station are converted into anomalies – i.e. the
> difference from an average temperature for a defined period. In the
> HadCRU method (used by the IPCC), anomalies are calculated based on
> the average observed in the 1961 – 1990 period (thus stations without
> data for that period cannot be included). For the calculation of
> global averages, the HadCRU method divides the world into a series of
> 5 x 5-degree grids and the temperature is calculated for each grid
> cell by averaging the stations in it. The number of stations varies
> all over the world and in many grid cells there are no stations. Both
> the component parts (land and marine) are separately interpolated to
> the same 5º x 5º latitude/longitude grid boxes. Land temperature
> anomalies are in-filled where more than four of the surrounding eight
> 5º x 5º grid boxes are present. Weighting methods can vary but a
> common one is to average the grid-box temperature anomalies, with
> weighting according to the area of each 5° x 5° grid cell, into
> hemispheric values; the hemispheric averages are then averaged to
> create the global-average temperature anomaly.
> [http://www.cru.uea.ac.uk/cru/data/temperature/]. The IPCC deviates
> from the HadCRU method at this point – instead the IPCC uses “optimal
> averaging. This technique uses information on how temperatures at each
> location co-vary, to weight the data to take best account of areas
> where there are no observations at a given time.” Thus empty grid
> cells are interpolated from surrounding cells.
> [http://www.cru.uea.ac.uk/cru/data/temperature/#faq] Other methods
> calculate averages by averaging the cells within latitude bands and
> then average the latitude bands.

http://www.metoffice.gov.uk/climatechange/science/explained/explained5.html
>
>
> Anomalies
>
> Absolute temperatures are not used directly to calculate the
> global-average temperature. They are first converted into ‘anomalies’,
> which are the difference in temperature from the ‘normal’ level. The
> normal level is calculated for each observation location by taking the
> long-term average for that area over a base period. For HadCRUT3, this
> is 1961–1990.
>
> For example, if the 1961–1990 average September temperature for
> Edinburgh in Scotland is 12 °C and the recorded average temperature
> for that month in 2009 is 13 °C, the difference of 1 °C is the anomaly
> and this would be used in the calculation of the global average.
>
> One of the main reasons for using anomalies is that they remain fairly
> constant over large areas. So, for example, an anomaly in Edinburgh is
> likely to be the same as the anomaly further north in Fort William and
> at the top of Ben Nevis, the UK’s highest mountain. This is even
> though there may be large differences in absolute temperature at each
> of these locations.
>

It is not clear that how the GWS site calculate anomalies. I send them
a note asking, and have a response but I probably need a clarification
Monday when I have time to think about this topic again.

To unsubscribe, send a message to majordomo@calvin.edu with
"unsubscribe asa" (no quotes) as the body of the message.
Received on Sat Dec 19 14:28:47 2009

This archive was generated by hypermail 2.1.8 : Sat Dec 19 2009 - 14:28:47 EST