The 1990s station dropout does not have a warming effect
Posted by drj | Filed under announcement
Tamino gives his results for his GHCN based temperature reconstruction. It is well worth reading. He also gives a comparison between stations that are reporting after 1992, and those that “dropped out” before 1992. He concludes that there is no significant difference in the overall trend. In other words refuting the claim that the 1990s station dropout has a warming effect. His results are preliminary and for the Northern Hemisphere only.
Tamino’s analysis use only the land stations; in order to write this blog post I tweaked ccc-gistemp so that we can produce a land index (python tool/run.py -s 1-3,5 now skips step 4, avoids merging in the ocean data, and effectively produces a global average based only on land data).
It is very easy to subset the input to ccc-gistemp and run it with smaller input datasets. So in this case I can split the input data into stations reporting since 1992, and those that have no records since 1992, and run ccc-gistemp separately on each input. I created tool/v2split.py to split the input data. Specifically I ran step 0 (which merges USHCN, Antarctic, and Hohenpeissenberg data into the GHCN data) to create work/v2.mean_comb then split that file into those stations reporting in 1992 and after, and those not reporting after the cutoff. Then I ran steps 1,2,3, and 5 of ccc-gistemp to create a land index:
It is certainly not the case that the warming trend is stronger in the data from the post-cutoff stations. [edit 2010-03-22: In a subsequent post I add trend lines to this chart]
The differences between these results and Tamino’s are interesting. Both show good agreement for most of the 20th century. These data show more divergence than Tamino’s in the 1800′s. Is that because we’re using Southern Hemisphere data as well, or is it because of the difference in station combining? Further investigation is merited.
We hope to make “experiments” of this sort easier to perform using ccc-gistemp and encourage anyone interested to download the code and play with it.
Update: Nick B obliges with a graph of the differences:
February 26th, 2010 at 4:23 pm
Re the 19th century difference: Even in the NH, undersampling will be a bit worse back then. But take the SH data (already undersampled) and split it into two bits, and it could be pretty bad.
So I think the difference will be how the grid boxes are used, along with using SH. Tamino is using coarse grid boxes. GISS is using distance-weighted interpolation to points within smaller grid boxes.
With undersampling, this will likely have some impact. With well-sampled data, it won’t matter.
February 26th, 2010 at 4:39 pm
Thanks for the timely analysis drj.
Any chance it would be easy to whack another mole by generating a similar graph for all urban stations in the U.S. and all rural stations, to tackle the contrarian argument du jour: http://wattsupwiththat.com/2010/02/26/a-new-paper-comparing-ncdc-rural-and-urban-us-surface-temperature-data/ ?
February 26th, 2010 at 4:53 pm
Is that because we’re using Southern Hemisphere data as well, or is it because of the difference in station combining? Further investigation is merited.
Are there fewer thermometers in the 1880? Is mal-distribution more obvious? If the answers are yes, the difference in the mean may just be a consequence of uncertainty in the mean using a smaller number of measurements aggravated by sub-optimal distribution for determining the area averaged temperature.
Also, because you use the anomaly method, unless you did something unconventional, your two versions are forced to match on average during the shared baseline. If there are real differences arising from using different stations, the consequence of those differences will be more visible further away from the baseline periods.
I would not be at all surprised to see the various versions of computing the surface trend diverging as the trends go back in time.
February 26th, 2010 at 5:04 pm
@lucia: I have access to same data as you do when it comes to answering the thermometers question, but I assume you also know the answer.
And yes, of course poor sampling will lead to more divergence, in probability.
GISTEMP uses the anomaly method, and ccc-gistemp is an emulation of it, so yes. And yes, that will force the average over the base period to match. The choice of base period is irrelevant to the trend, however.
February 26th, 2010 at 5:08 pm
@Zeke: Not until Monday, and maybe not then. If however you have a handy list of Rural and Urban stations, do please cut the input data yourself!
February 26th, 2010 at 5:15 pm
Zeke: You might be asking the wrong guys. That particular assertion is about the USHCN adjustments, I’m pretty sure. Those adjustments are already baked in, before GISS or ccc starts doing anything.
You’ll want to go to the USHCN web page, get the raw data, TOB adjusted data, and final adjusted data, and go from there. If you must cheat, the NCDC has made the code available for their adjustments, but I think simple comparisons of the raw and final sets would suffice, for starters.
February 26th, 2010 at 5:38 pm
@drj–
The answer to what?
I suspect we are talking at cross purposes because I used rhetorical questions, which I actually know should be avoided in blog comments. They cause confusion, and that’s my fault.
What I meant to communicate was that to extent that the answer to my rhetorical questions are yes (and I suspect they are even without looking or running any code), I think we should expect see divergences in computed anomalies in the early thermometer record.
I appears we agree on the effect of statistical uncertainty on divergence back in the early record. My take is that statistical uncertainty may be sufficient explanation of the approximately 0.2C difference in temperature anomalies computed using the two different thermometer sets back in 1880-1900. [ed: corrected "1990" to "1900"]
Do we more or less agree on that? Or do you think it’s something else?
I am also saying that — without looking at the code or data — if we look at any T_anomaly vs. Time graph, and there are real differences (or even difference resulting from statistical uncertainty) those will appear larger far from the baseline periods. This effect should be rather data independent because its a direct consequence of the anomaly method.
So, for example, if you happened to re-baselined using 1880-1900, the match would look poor now, and good back then. I think that sort of re-baseline would be misguided because it makes no sense to baseline to a period with larger measurement uncertainties– but that is what would happen if we did pick that period. Anyway: Once again, the reason we see the slight mismatch back in 1880-1900 [ed: and again] is at least partly related to the choice of baseline.
Do we agree?
So, it seems to me the fact that differences between results using pre-cut off and post-cut off thermometers are visible way back in 1900 is both because
a) we know there were fewer thermometers so ‘noise’ alone can cause this to occur even if there is no bias associated dropping out thermometers and
b) the anomaly method does have a tendency to force agreement in the average during the baseline period. So no matter the cause of differences, they will be more noticeable far from the baseline period.
Anyway, I think your post is useful. To my knowledge that makes the fourth post that shows the drop out in thermometers does not cause a bias in the long term trend.
February 26th, 2010 at 5:42 pm
Zeke: although we’re not set up to do experiments with the NCDC datasets, it would be very easy for us (or for you) to do a ccc-gistemp run without any US data. The GHCN data, which is the great preponderance of our input dataset, is raw.
As for the UHI adjustments done in GISTEMP: they do not adjust rural stations at all. Only urban stations are adjusted, and urban stations which cannot be adjusted are dropped. See our current step2.py.
February 26th, 2010 at 5:59 pm
Carrot,
Good catch. I forgot that GISS used adjusted USHCN (but raw GHCN) data as inputs.
February 28th, 2010 at 8:06 am
Off topic, but I noticed that GISS has switched to using night lights for all of its urban adjustments, not just the US48. Are you guys privy to those changes?
They write:
January 16,2010:The urban adjustment, previously based on satellite-observed nightlight radiance in the contiguous United States and population in the rest of the world (Hansen et al., 2001), is now based on nightlight radiances everywhere, as described in an upcoming publication. The effect on the global temperature trend is small, that change reduces it by about 0.005 °C per century.
http://data.giss.nasa.gov/gistemp/updates/
February 28th, 2010 at 8:18 am
Also, I noticed with the newest USHCN_V2 data, GISTEMP’s lower 48 analysis shows a bit more warming that it had previously. In fact, both 1998 and 2006 are now warmer than 1934. Hansen et al. 2001 had stated that they couldn’t declare a warmer year than 1934 until it was exceeded by at least 0.1 degrees, and the new figure for 1998 exceeds that (2006 is 0.1 warmer).
February 28th, 2010 at 9:16 pm
cce@10: Yes, I’d noticed the GISTEMP change, a few lines in STEP2/text_to_binary.py to fish a different field out of the station metadata file. I’ll be making a matching change in our sources before our 0.4.0 release.
cce@11: GISTEMP, and ccc-gistemp, doesn’t include a “lower 48 analysis”. 95% of the world’s population wouldn’t know what a “lower 48 analysis” might be. I do know, and I also know it would cover about 1/60 of the global surface, so temperatures there are not very relevant to our results. When you say “I noticed”, whereabouts did you notice this? Can you provide a link?
February 28th, 2010 at 9:39 pm
@cce, @Nick.Barnes: “lower 48″ may be a reference to GISTEMP’s figure D. In my opinion this is wildly parochial.
February 28th, 2010 at 9:40 pm
The results of the lower 48 analysis is here:
http://data.giss.nasa.gov/gistemp/graphs/Fig.D.txt
It was very relevant during the so-called “Y2K” controversy and is the primary reason why so many people belive the ’30s were warmer than recent years. That is, they confuse the US with the world.
I noticed the change when I looked at the updated graph of US temperature:
http://data.giss.nasa.gov/gistemp/graphs/Fig.D.lrg.gif
1998 and 1934 have been statistically tied for a number of years (I believe since the TOBS improvements incorporated in 2001), but now 1998 is back to being significantly warmer than 1934.
March 1st, 2010 at 6:54 am
Hey now, Figure D was rather useful as a validation tool in testing my recent foray into gridded temperature analysis :-p
http://i81.photobucket.com/albums/j237/hausfath/Picture68.png
Comparing GISS-temp to adjusted GHCN temp for the globe is less interesting, since the methods are rather different.
March 1st, 2010 at 9:55 am
@Lucia: I do not think we are talking at cross purposes.
It’s fine to ask rhetorical questions (although I do have a tendency to answer rhetorical questions). But since you and I both know the answers to your rhetorical questions, I wasn’t sure what the point was.
You ask if I agree “that statistical uncertainty may be sufficient explanation of the approximately 0.2C difference in temperature anomalies computed using the two different thermometer sets back in 1880-1990″. Well, yes; I already said “poor sampling will lead to more divergence, in probability”.
You also ask if I agree that “the reason we see the slight mismatch back in 1880-1990 is at least partly related to the choice of baseline”. Well, yes, partly, but that doesn’t seem to be a very interesting statement. Plainly the divergence between the two series is greater in the 1800′s than the other parts of the series and that is not entirely a choice of base period. Offsetting the shorter series so that two series have the same average over their common period we get:
and we can see that the 1800′s still have greater divergence.
Of course we are in broad agreement, but I still think there is room for further investigation. In particular I think we can learn a bit more than “it’s just statistcal noise”: How much the difference between this analysis and Tamino’s is down to difference of station combination method? How much is down to different data? (and how much is essentially uninteresting variation due to poor sampling). A northern hemisphere graph from ccc-gistemp will help with that.
March 1st, 2010 at 12:46 pm
cce@14: Ah, yes, figure D. That was originally in Hansen et al 1999. I am reading that paper now. It describes in great detail an algorithm similar, but not identical to the GISTEMP algorithm (the main differences I see are that it uses a 2-degree grid, and has less flexible UHI adjustment). I imagine it then area-weights the series for grid squares which include parts of the “lower 48″.
Because this figure is so very parochial, it’s not very high on my list of priorities to reproduce it. I might send email to GISS to ask how it is done. It would be the work of a few minutes to generate a US-only graph, by running gistemp using lower-48 data only, but that isn’t what Hansen et al 1999 describes.
Without running any code at all, I can speculate that changes in this chart over time might be caused by the switch from USHCNv1 to USHCNv2 input datasets.
March 1st, 2010 at 3:01 pm
Nick,
I’m pretty sure the USHCNv2 changes are the cause. They began using it in November. I don’t think they updated the figure D graphic and table until February.
Also be sure to read Hansen et al. 2001 if you haven’t already, “A closer look at United States and global surface temperature change,” which is the most recent paper to describe GISTEMP.
March 1st, 2010 at 6:03 pm
[...] stations with records post-1992 to those without, to quickly replicate the work that Tamino and the CCC folks did on addressing E.M. Smith’s whole station dropout [...]
March 1st, 2010 at 9:13 pm
Sorry if this is mentioned elsewhere, but why are you referring to station drop out, when in fact it is actually old station data being added?
http://www.ncdc.noaa.gov/oa/climate/ghcn-monthly/images/ghcn_temp_overview.pdf
See section 2, Sources “When possible, we tapped related projects for potentially
useful data. For example, NCDC recently
collected and processed station normals for the period
1961–90 as a contribution to WMO (WMO 1996a).
On occasion, a WMO member country supplied year/
month sequential data in addition to the 30-yr means
and other statistics…”
March 2nd, 2010 at 12:02 am
Turboblocke: sensible people know the history of how the GHCN was built. But the blog terminology was set by certain conspiracy theorists, so ‘drop’ is what sticks. ‘Drop’ is also less cumbersome than other semantic constructions.
March 2nd, 2010 at 10:27 am
[...] tym, po za autorem bloga OpenMind, czy DoskonaleSzare zajęli się również autorzy CCC-GISTEMP. Co jest tu istotne? Nałożono na siebie wykresy pokazujące przebieg odchyleń globalnych dla [...]
March 3rd, 2010 at 3:03 pm
[...] Clear Climate Code, Feb. 26, 2010 compares GISSTemp type calculations of global surface temperature anomalies based on the “full” and “cut-off” thermometer set. They find no major differences between the two traces. [...]
March 5th, 2010 at 10:29 pm
[...] concluded that the bias is actually toward slightly under-reporting the warming. See Open Mind, Clear Climate Code, The Blackboard and Menne, 2010 (described at Skeptical [...]
March 6th, 2010 at 5:40 pm
[...] more vehement in recent months even though the code base and data have been available for years and clearly demonstrate that the criticisms are [...]
March 7th, 2010 at 8:34 am
[...] jak metoda CAM stosowana przez CRU). Można o tym przeczytać tutaj, a także tutaj, tutaj, tutaj i tutaj. Jasne jest, że Watts i D’Aleo się mylą, a oskarżenia pod adresem NOAA są [...]
March 7th, 2010 at 11:26 am
[...] A replication from clear climate code [...]
March 7th, 2010 at 7:26 pm
[...] more vehement in recent months even though the code base and data have been available for years and clearly demonstrate that the criticisms are [...]
March 8th, 2010 at 2:13 pm
[...] drj @ Clear Climate Code http://clearclimatecode.org/the-1990s-station-dropout-does-not-have-a-warming-effect/ [...]
March 8th, 2010 at 9:22 pm
[...] more vehement in recent months even though the code base and data have been available for years and clearly demonstrate that the criticisms are [...]
March 8th, 2010 at 10:10 pm
[...] without the actual programs, and with the same or different data sets as has been done by Tamino, clearclimatecode, The Blackboard, The Whiteboard and others. woodbe. __________________ Admission of the week: [...]
March 9th, 2010 at 5:46 am
This is great.
Can you guys either post the data tables or calculate
trend lines for both cases.
March 9th, 2010 at 8:03 am
@Steven: secret fact about Google charts: the data is in the URL for the image. Right click on the graph and select “Copy Image Address” (or whatever is equivalent in your browser), then paste that into a document:
http://chart.apis.google.com/chart?cht=lc&chds=-100,100&chtt=Global+Temperature+Anomaly|Land+Stations&chdl=post-cutoff|pre-cutoff&chdlp=t&chd=t:9,-7,27,9,15,1,5,-32,-13,22,-9,-29,-33,-28,-15,-16,-5,-1,-14,-21,1,1,-21,-29,-36,-25,-5,-32,-25,-26,-14,-19,-29,-23,1,3,-12,-37,-27,-8,-13,6,-10,-13,-11,-14,6,-9,-2,-23,-1,3,1,-11,2,-6,6,16,21,-1,9,7,9,5,12,-1,-2,10,-1,-9,-18,-3,2,11,-8,-8,-17,9,11,7,1,11,5,2,-24,-17,-9,-2,-13,0,4,-9,-4,19,-6,-1,-21,16,6,13,26,38,7,33,13,11,17,32,41,28,48,43,12,18,31,45,37,39,70,45,41,56,68,65,58,77,66,73,55,73,-999%7C-25,-9,-14,-17,-55,-26,-37,-64,-41,-6,-41,-72,-55,-60,-65,-54,-53,-22,-14,-32,-9,-9,-33,-42,-48,-35,-21,-49,-43,-41,-28,-35,-28,-28,0,11,-27,-65,-36,-9,-19,-12,-13,-16,-16,-19,3,-6,2,-23,-3,2,5,-14,5,-10,0,9,12,2,20,16,14,2,11,2,2,12,-4,-5,-9,-2,6,13,-7,-8,-27,4,13,6,1,8,5,6,-28,-12,-4,-3,-5,10,9,-21,-12,16,-14,0,-19,19,6,13,25,35,9,37,3,18,22,48,46,28,54,46,-999,-999,-999,-999,-999,-999,-999,-999,-999,-999,-999,-999,-999,-999,-999,-999,-999,-999,-999&chxt=x,y,r&chxl=0:%7C1880%7C%7C%7C%7C%7C%7C%7C%7C%7C%7C1890%7C%7C%7C%7C%7C%7C%7C%7C%7C%7C1900%7C%7C%7C%7C%7C%7C%7C%7C%7C%7C1910%7C%7C%7C%7C%7C%7C%7C%7C%7C%7C1920%7C%7C%7C%7C%7C%7C%7C%7C%7C%7C1930%7C%7C%7C%7C%7C%7C%7C%7C%7C%7C1940%7C%7C%7C%7C%7C%7C%7C%7C%7C%7C1950%7C%7C%7C%7C%7C%7C%7C%7C%7C%7C1960%7C%7C%7C%7C%7C%7C%7C%7C%7C%7C1970%7C%7C%7C%7C%7C%7C%7C%7C%7C%7C1980%7C%7C%7C%7C%7C%7C%7C%7C%7C%7C1990%7C%7C%7C%7C%7C%7C%7C%7C%7C%7C2000%7C%7C%7C%7C%7C%7C%7C%7C%7C%7C2010%7C1:%7C%7C-0.5%7C+0.0%7C+0.5%7C%7C2:%7C%7C-0.5%7C+0.0%7C+0.5%7C&chco=ff0000,0000ff&chs=440×330
See that list of mostly 2 digit numbers separated by commas? That’s the two data series (separated by %7C in the middle). The numbers are the actual anomalies in centikelvin (or 0.01 °C if you prefer). -999 is “no data”.
This is true for most (all?) of the charts I’ve prepared for this blog. They’re made with Clear Climate Code’s vischeck.py tool.
March 9th, 2010 at 8:08 am
Trend calculation was always on the todo list; now I’ve created Issue 49 so that we have a permanent reminder.
March 12th, 2010 at 7:45 am
@Steven: Trends calculated, updated chart in a newer blog post
March 18th, 2010 at 6:23 pm
[...] przypomnieć, że chłopaki z ccc-gisstemp potwierdzają dokładnie to samo w stosunku do analizy GISS. Wygląda więc na to, że problem ów wylągł się jedynie w głowach [...]
April 19th, 2010 at 11:20 am
[...] that better statistical methods would have changed the results – NASA, NCDC, and now multiple different amateur reconstructions have all replicated the basic results of CRU using the same raw data CRU [...]
May 16th, 2010 at 5:20 pm
[...] greining var einnig gerð af Clear Climate Code, sem bar einnig saman hitastigsgögn fyrir stöðvarnar sem duttu út og þær sem haldið var [...]
May 25th, 2010 at 4:40 am
[...] Clear Climate Code, Feb. 26, 2010 compares GISTEMP type calculations of global surface temperature anomalies based on the “full” and “cut-off” thermometer set. They find no major differences between the two traces. [...]
June 25th, 2010 at 3:01 am
[...] results have been published recently at Clear Climate Code and by Tamino at “Open Mind”. This post provides further detail of the effect at [...]
July 18th, 2010 at 6:42 pm
[...] kompletnie się mylili. Więcej na ten temat tutaj. Na stronie projektu ccc-gistemp także można znaleźć dowody, że skasowanie znacznej ilości stacji nie miało żadnego wpływu na obserwowane trendy. [...]
September 12th, 2010 at 7:08 pm
This post was referenced in an investment research paper called “Climate Change: Addressing the Major Skeptic Arguments”, by Deutsche Bank Climate Change Advisors.
November 23rd, 2010 at 1:07 pm
[...] has about the station drop out. First of all, lets remember that overall the station drop out had no effect on global temperatures: So having just debunked their entire argument, I could just leave it as it were and call it a [...]
January 3rd, 2011 at 3:06 am
Sorry to comment to such an old post. Referencing my comments from last February and March, below is an example of how GISTEMP’s analysis of the US lower 48 states (Figure D) becomes fodder for “skeptics” and is thus relevent to the “debate.”
http://climateaudit.org/2010/12/26/nasa-giss-adjusting-the-adjustments/
January 6th, 2011 at 9:17 am
@cce: Hmm. What do you suggest we do? I suppose a ccc-gistemp version of Figure D is vaguely on the cards, and I’ve created Issue 101 so that we don’t forget about it completely.
January 12th, 2011 at 5:06 am
Perhaps an email to Reto Ruedy to see if he could make the source code available.
Somewhat related, a cool feature for ccc-gistemp to have would be the ability to generate time series for individual countries. I wonder if you could pull country masks out of Google Earth or similar mapping software.
January 12th, 2011 at 10:07 am
@cce: Yep. I’ve added a note to the issue so that we don’t forget to make it flexible enough to do other countries.
February 1st, 2011 at 6:14 pm
[...] [...]
October 17th, 2011 at 4:18 pm
[...] not– of if it did materialize, it appears small. The exact same analyses by Zeke, Tamino,, CCC, The Whiteboard etc. all suggest that any biases that might have been introduced by “the [...]