Hopefully the marine data is of better quality ?

Hmmm, see http://www.climateaudit.org/?p=1276

Given that the paper states that all of the station data has been adjusted, I’m still puzzled as to why both raw and adjusted measurements for each station haven’t been stored. I guess after reading this blog for a while, I shouldn’t be

Hopefully the marine data is of better quality ?

Assuming that the stations for which the homogenisation correction is known can be considered a random sample from the total number of stations, why isn’t the distribution normal and not the bimodal distribution as plotted?

That’s because of undetected adjustments, here’s how it goes

1) There is a 1940-60 period where thermometers were moved to out of town airport. That necessitates negative adjustments for earlier record (take the UHI effect out).

2) There are some positive adjustments (probably due to recent rural cooling, hard to tell because only histogram is shown).

3) As you all know, the central limit theorem requires zero-mean bell-shaped distribution for everything. Thus, the bimodal histogram of known adjustments represents such distribution, but small values are missing. The bimodal histogram is not balanced, but

this asymmetry is small compared with the typical adjustment, and is difficult to quantify

That is, take hypothesized adjustment histogram, subtract observed adjustment histogram, make it zero mean, and you’ll get the uncertainty. All clear?

Assuming that the stations for which the homogenisation correction is known can be considered a random sample from the total number of stations, why isn’t the distribution normal and not the bimodal distribution as plotted?

Or am I missing something basic?

I have not had time to read all of the thread above, but will repost an email to me from Phil Jones from 2006. His email follows:

—– Original Message —–
From: Phil Jones
To: sherro@bigpond.net.au
Cc: Sheppard Sylvia (SCI) ks918
Sent: Saturday, March 25, 2006 4:20 AM
Subject: FW: Early global temperature data
Dear Geoffrey,
We no longer have the Australian station date we were using in the
early 1980s. At that time we had a limited network. In the 1990s,
the Australian Bureau of Meteorology began issuing a lot more
station data each month. Up to that time it had been about 40
stations internationally. Through contacts with personnel in Melbourne,
we got access to the back data from all the new stations, so added
these. In order to use temperature data, we need historic series
with at least the 1961-90 base period. We now have access to over
100 stations from BoM in real time.
I wish more countries would release more data in real time like
Australia. Some have, but not that many – and none release extra
data in Africa, South America and southern Asia. We have
managed to get extra historic data though in South Africa, Argentina
and Brazil. We got the latter, though, on the agreement that we
didn’t pass the data onto others, but we can use them in our
gridded data. This condition also applies to a number of
European countries – again only historic data, no real time. Australia
is the only country to make additional data (additional to the about 1500
exchanged by Met Services) to us in real time.
I have looked back at a publication where we adjusted station
records for homogeneity in the mid-1980s. We didn’t omit any
Australian series then, but adjusted the following sites:

Darwin, Townsville, Thursday Island, Gladstone, Forrest,
Adelaide, Sydney and Norfolk Island. We still have these adjustments.

In the mid-1990s, we compared our Australian temperature with
a series developed by Neville Nicholls (then of BMRC, he has retired
in the last few months and now works at Monash) and a student
he had. Over the period from 1901-1992 we had a correlation of
our two series of 0.92 and with no difference in trend over these
92 years. This work was done before we got the additional
station data (so in effect we had about 40 stations, whereas
Neville and the student had almost 200). I can’t recall the student’s
name (Simon ?).

So, I can’t send you anything. We don’t have the earlier station
data now.

Talk about throwing away information!

Yep, take N samples from i.i.d gaussian, sort the data in ascending order, sort the data in descending order.. Histogram wont change. Adjust ’40s data down, and some other period data up, and you’ll get a bimodal histogram. I’m a bit worried about this step:

So the distribution of adjustments is bimodal, and can be interpreted as a bell-shaped distribution with most of the central, small, values missing.