Here is a puzzling comparison of two zonal averages from Phil Jones’ CRUTEM3 gridded land data. Red shows the average from 20S to 20N and black shows the average of the 20-30S band (both N and S). These are calculated from gridded data at http://hadobs.metoffice.com/crutem3/data/CRUTEM3.nc.
I did this comparison because I noticed a difference between my own average of 20S-20N gridded data and the archived “low latitude” average (which was 30S to 30N)
Figure 1. CRUTEM3 Zonal Averages. Black 20-30 N and 20-30S. Red. 20S-20N.
The range of differences goes from -1.7 to 4.6 deg C. Explanations welcome.
Climate Audit 
42 Comments
The Hadley cells descend around 30 degrees. Maybe that’s warmer than the air at 20 degrees?
Recalling the paucity of stations in equatorial Latin America and Africa, my guess is that the absence of actual stations causes more reliance on the smearing algorithms 20S to 20N. 20 – 30 N and 20 – 30 S.
Using the GISS station inventory for 1880 only six land stations existed in the S. Hemisphere out of the total of 51 available between 30N and 30S. Four of those are between 20S-30S. N. Hemisphere had 22 stations between 20N-30N.
S. Hemisphere stations.
WALGETT POST OFFICE -30 148.1
GOYA -29.1 -59.3
ALICE SPRINGS -23.8 133.9
RIO DE JANEIR -22.9 -43.2
JAKARTA/OBSER -6.2 106.8
ZANZIBAR/KISA -6.2 39.2
Dear Bob, I initially came with the same explanation – too little data in the South. However, when you look at the numbers, it just doesn’t seem to work too well. The annual local fluctuations are only by about 1 deg C, see e.g. Central England
http://www.junkscience.com/MSU_Temps/HadCET_an.html
By using four of them, this noise gets averaged out to sqrt(1/4) times that i.e. 0.5 deg C. Moreover, this noise should have a pretty high frequency because the “weather” doesn’t stay on the same region for three decades. So a systematic, 30-year-long difference by 2-5 deg C seems just way too much for your explanation.
Re: Luboš Motl (#4), your probably right. Just figured I’d post the information so people could get a better visualization of the data distribution.
Possibly the prevalence of one nation’s maritime fleet for SST accounts for the difference.
This is land not SST.
Wouldn’t it be nice if these things agreed for once? The math can’t be that hard. It shouldn’t take someone as smart as Lubos Motl (I’ve read his math on string theory–makes my head hurt) to get these things right.
It must be an “adjustment” inconsistently applied; the lines have broadly the same fluctuations but displaced until they eventually meet again in 1890. BUG!
Forgive me if this is too off-topic, or too old hat. What is the point of interpolating observational data? Presumably the observational data is to be used for comparison with the output of modelling runs. But it is dead easy, and not at all arbitrary, to interpolate the model predictions to accord with the time and place of observations – there can be no need to fanny about with the observations at all.
I believe I can explain that deep dip back in the 1865-67 period:
http://www.avo.alaska.edu/volcanoes/eruptsearchresults.php?fromsearch=1&yearstart=1865&yearend=1868&year=&volcano=-1
five alaskan volcanos erupted multiple times over those three years.
My first thought is to compare the 0 to 20N and the 0 to 30N sets and the 20S to 0 and 30S to 0 sets.
GHCN has older data than the 1880 start by GISS. I checked GHCN 1850-1880 and found no data for 0-30S prior to 1866 and just five stations up to 1880. 43 stations for 0-30N during 1850-1880.
Here are the SH stations with starting date.
1866 503967450000 JAKARTA/OBSER -6.2 106.8
1871 303837430000 RIO DE JANEIR -22.9 -43.2
1877 501945890000 YAMBA -29.4 153.3
1877 301872700020 GOYA -29.1 -59.3
1878 501957150010 WALGETT POST OFFICE -30 148.1
Re: Bob Koss (#12),
Phil Jones’ data set is not the same as GHCN though he’s falsely stated in an FOI refusal that his data is available at GHCN. I looked at the gridcells 0-30S with values before 1875. There’s a very interesting QC difference between CRU and GISS that jumps out. If there’s a scruffy bit of intermittent data, in a lot of cases, Hansen doesn’t use it. From looking at a few sites.
For example, Jones has some spotty values from an island in the gridcell 177.5E 22.5S in the 1860s and then nothing until 1951-1970 and nothing since 1990. The values in the 1860s were a lot higher than values in the 1980s, which likely indicates some sort of inhomogeneity. Given the lack of continuity in the record, I don’t see that there’s any objective way to quantify the displacement and my guess is that a lot of this scruffy data is unusable. As always, the moral is that the people involved need to document the stations that in some sense are “best quality” and use those as a framework, rather than throwing everything into a hodgepodge and relying on questionable software to come out the other side.
To be clear, it is my opinion that it is warmer now than before 1880; I’m just curious as to how much weight we can put on each data set.
Long Lat Location
-57.5 -27.5 From info above, I presume that this is Goya, which is near Paraguay. 1874 to present; very steady
-67.5 -22.5 perhaps Chuqicamata: a few in 1874, spotty in 19th century
-42.5 -22.5 Rio 1851 on
57.5 -22.5 island in Indian Ocean SE Madagascar. 1853-1861, then 1910 on
177.5 -22.5 island in Pacific OCean: a few values from 1860s,gap to 1951-1970; some in 1980s; none since 1990
107.5 -7.5 Djakarta fairly steady since 1866. Gap in 1980s.
Re: Steve McIntyre (#17),
This is either Reunion or Mauritius. The precise location given is open sea. Both islands are to the northwest, and the only nearby inhabited land.
Re: John (#21), both are in the 57 22S gridcell. Perhaps someone can check GHCN.
Re: Steve McIntyre (#17),
“To be clear, it is my opinion that it is warmer now than before 1880”
I think the cities are warmer. I think UHI and land use changes can explain all warming.
There were 23 volcanos erupting around the globe in 1866 and 29 in 1867:
http://www.volcano.si.edu/world/find_eruptions.cfm
Open your computer and vacuum around the CPU. There may be layers of Bristlecone pine fibers which have been known to distort data related to older temperatures.
It may just be a matter of where those few SH stations were in hte late 1800s. You have Rio and Goya in Brazil and Argentina, Waglett and Alice Springs in Australia. I don’t know whether satellite data would show these 4 locations Alice Springs is in the middle of the outback so probably pretty hot relative to the rest of the SH. If these 4 are to represent the whole of the SH, having 1/4th come from the outback may really skew that data. Maybe cooler stations were added to the SH later on? Were the NH and SH 20s-30s equally hot in the 1800s? Hard to know from current data but that may be telling.
S. Hemisphere stations.
WALGETT POST OFFICE -30 148.1
GOYA -29.1 -59.3
ALICE SPRINGS -23.8 133.9
RIO DE JANEIR -22.9 -43.2
The data is pretty sparse down there, so pretty noisy.
I would think data should be sparse from 177.5 E 22.5 S. That is deep ocean with no islands south of the Fijis. I think it is actually Tonga at 177.5 W.
57.5, -22.5 is either Mauritius or Réunion, more likely the former.
Re: tty (#19), Thinly dispersed buckets? 😉
Perhaps not number of stations but number of gridcells. Smear the gridcells into a signal for each hemisphere and average the two, and you get a different answer from averaging all the gridcells in both hemispheres.
Steve, I agree with you that data showing that it was warmer in the 1880s than today has problems. You, Anthony and others have identified many stations with problems. I am amazed that so many people in the climate science community believe that they can ignore GIGO by using intricate software solutions to manipulate flawed data.
Given the small number of stations and the colonial nature of many of those stations, might it have something to do with the diffusion of the Stevenson Screen? The small number of stations would lead to a bigger “instrumentation” effect. After a few years when most adopted the SS as the standard the records would come into greater alignment.
P.S. The Stevenson Screen was designed by Thomas Stevenson (1818-1887), a British civil engineer. Perhaps Anthony can provide more information on the rate and manner of adoption of the screen I recall he had a few postings on this topic before he started his terrific audit of sites.
Re: bernie (#26),
That’s one issue. The USHCN experience shows that in homogeneities can be introduced through relatively small station moves. If there is a 70 year gap in a record, there is no basis for assuming that the thermometers are in the same spot.
Re: Steve McIntyre (#27), Yep, plus some of these were naval bases and any expansion of bases to accomodate larger vessels with the new coal fired boilers would have sent the thermometers to somewhat higher elevations (though confounded with going inland as well). All this suggests that some of this data is very suspect without a full and complete station history. The small number of stations means that the various micro-climate effects cannot be assumed to cancel each other out.
However, the fact that the two groupings all seem to be in step after about 1880 suggests that some new commonality emerged. That commonality could be instrumental, sitings, observational, methodological or any combination thereof.
Re: Steve McIntyre (#27), The way you wrote this comment suggests that you actually have an answer. Do you?
Has the air temp been compared to the more-than-four-station SST? There probably were many more SST readings taken at latitudes similar to the land stations. If there is a relationship to air temp, after compensating for ocean current temp differences there should be a similar latitude temp difference.
There also could be a difference between the equipment at 20S vs 30S, due to the background of the countries. 20S has numerous Spanish, Portuguese, and French colonial countries. 30S has 2 Spanish countries in South America, and the different backgrounds of South Africa and Australia.
Also, as I mentioned before, the subtropical high pressure zone is around 30S, while 20S has the trade winds.
Steve, which way did you figure your averages? Dunno about GISS, but HadCRUT does not average e.g. the entire globe. They average the NH and the SH separately, and then average the two …
GISS might do the same with e.g 30N to 30S. They may average 30N-0 and 0-30S, and then average them. This might account for some of the difference.
w.
Re: Willis Eschenbach (#32),
I downloaded the gridded data and did zonal averages,
The problem of the early days of land temperatures arises from modern rejection and adjustment of data.
According to Australian Bureau of Meteorology records, these stations started operating in these years:
1855 Melbourne Central
1857 Armidale NSW, Goulburn NSW
1858 Casino, Bathurst, Sydney Observatory, Deniliquin, Beechworth, Murrurundi
1859 Echuca, Gabo Island, Port Augusta, Mount Gambier, Robe?, Longerernong
Of these, Casino at 29 deg 51S is the most northerly and only one included in the band being discussed. But others closer to the Equator followed. Darwin in 1869 at 12 deg 27S, Alice Springs in 1873 at 23 degS, etc.
Some of these are still in use today as rural sutes (like Gabo Island). Others have been overcome by UHI (like Melbourne and Sydney).
In recent years, the BOM has been reluctant to use station data later than about 1910. Some years ago it was decided that a change to Stevenson screens was mostly complete by 1910, so data before then were rejected in bulk, e.g. in Della-Marta et al, 2004. (This is a general statement and there could be instances of other treatment).
When the BOM sends its data to GISS or CRU (in those days) for compilation into a global average, it is already homogenised in ways taking man-years to reconstruct. What GISS or CRU do in terms of further rejection or further sdjustment is something of a mystery.
Hence part of the reason for the graph above. My guess is that it’s mainly poor housekeeping of data.
Again, I caution against getting too sophisticated with mathematical/statistical treatment of data that has been hacked around to an unknown extent. These early stations were mainly set up to help farmers and travellers and it is a bit much to expect the extraction of tiny trends from them today, no matter how good the maths package is.
If GISS or CRU have a paucity of stations from pre-1900, maybe they would be the people to ask why so many were rejected and on which criteria. I’ve just found 17 old stations in Australia in a few minutes. There are plenty more. If the global gurus claim the existence of only a few early stations, then it’s their shortcoming through not looking hard enough. Don’t believe their counts of available stations.
It’s not rocket science. It 1, 2, 3, 4, 5, 6,……
My guess is that the difference in land mass between the Nothern and Southern hemisphere, combined with the much later move of western science into the Southern hemisphere has in some way skewewd the data. I doubt that there were many weather station inland in Southern Africa before about 1900.
The data for South Afirca before 1900 would largely be biased towards the 30S zone. I suspect the same would apply to Australia.
Just a guess.
What are the counts of the number of grid boxes with valid data for all three bands, over time?
I think that the first 25 years of the red line were inadvertantly inverted. Oh wait, I forgot, that processing algorithym is only allowed for bristlecones and select Anartic surface stations!
Bill
RE 36. You can head over to http://hadobs.metoffice.com/crutem3/data/download.html
and have a look
http://hadobs.metoffice.com/crutem3/diagnostics/time-series.html
links to http://hadobs.metoffice.com/crutem3/CRUTEM3_accepted.pdf ,
I guess they mean Brohan et al. , a paper that IMO has some problems with bias-like errors. But reported bias uncertainties seem to be quite wide for the series in question.
Re: UC (#39),
Seems the filename is wrong, this one works though:
http://hadobs.metoffice.com/crutem3/HadCRUT3_accepted.pdf
In #4 (Motl) the comment “ . . . difference by 2-5 deg C seems just way too much . . . ”
caused me to look at the graph again. Y-axis is in tenths below the zero and in full digits above. Odd.
Maybe its not why the difference but what’s the difference
… who’s gonna miss us in year or so, so whats the difference if we go.
The problem I have, amongst many others, is the idea of a global temperature.
Where ever there are accurate temperature measurements, these should be observed and lets assume that the temperature record in the US shows a dramatic increse in temperature which correlates with a dramatic increase in Europe then I suppose one could say there is an increase in global temperatures.
The use of anomalies is lovely when the data is complex however if one looks at the graph of temperature/against time for central united kingdom since 1650 then what’s the difference?
Steve if off post please delete.