Notes On The GISTEMP Ratio Of Land To Sea Surface Temperature Data
Guest post by BobTisdale
Over the past few days there has been some blogosphere buzz about the apparent ratio of Land Surface Temperature (LST) Data used in the GISTEMP combined land and sea surface temperature data with 1200km smoothing. I’ll provide a few comparison graphs to explain.
Figure 1 includes a time series graph of the Hadley Centre’s HADCRUT3 Combined Surface Temperature product, from January 1982 to April 2010. Also included is the weighted average of the two datasets that make up the HADCRUT3 data, with the weighting of 27% LST [CRUTEM3] and 73% Sea Surface Temperature (SST) [HADSST2]. Those weightings were required to match the linear trend of the weighted average to the linear trend of the HADCRUT3 combined product. The weighting makes sense, since the global oceans represent about 70% of the surface area of the globe.
http://i28.tinypic.com/2yoe59f.jpg
Figure 1
Figures 2 and 3 provide similar comparison graphs. Figure 2 shows the NCDC combined surface temperature product and the weighted average of its LST and SST components. To align the linear trends, the weighting required for the components of the NCDC product was also 27% LST data and 73% SST data. Figure 3 shows the GISTEMP product with 250km radius smoothing. For this GISTEMP product, the weighting required for the components was 28.5% LST data and 71.5% SST data. Again, the relationships of the SST and LST data make sense.
http://i26.tinypic.com/2qmi3yh.jpg
Figure 2
##########
http://i25.tinypic.com/1rx6v7.jpg
Figure 3
Here’s the curiosity. It appears in the GISTEMP Product with 1200km radius smoothing when we apply the same component weighting (28.5% LST data and 71.5% SST data) that we had used on the other GISTEMP combined product. The weighted average of the components of the GISTEMP combined product with 1200km radius smoothing has a significantly lower trend than the actual GISTEMP data. This can be seen in Figure 4.
http://i28.tinypic.com/9jot4x.jpg
Figure 4
In order to achieve the same linear trend as the GISTEMP combined product with 1200km radius smoothing, the components have to be weighted with 67% LST data and 33% SST data, almost reversing the ratio of the areas of global oceans and continental land masses.
http://i30.tinypic.com/p9g5d.jpg
Figure 5
Figure 6 is a map illustrating the GISTEMP LST data (trends) from 1982 to 2009. Note how the GISTEMP LST data extends out over the oceans. This is not the case for their combined product, because GISS masks the LST data over the oceans in its combined product. So in order to properly create a weighted average of GISTEMP land and sea surface temperature data with 1200km radius smoothing, the land surface data where it extends out over the oceans would first need to be masked.
http://i26.tinypic.com/4ieop2.jpg
Figure 6
A NOTE ON THE DIVERGENCE BETWEEN GISS AND THE OTHER DATASETS
Much of the divergence between GISTEMP and the Hadley Centre and NCDC combined surface temperature products is likely caused by the fact that GISS deletes SST data in the Southern and Arctic Oceans and replaces it with LST data, which has a significantly higher linear trend than the SST data it replaces. This was discussed in the post GISS Deletes Arctic And Southern Ocean Sea Surface Temperature Data.
ANOTHER CURIOSITY
It can also appear that GISS extends LST data out over the oceans in areas other than those with seasonal sea ice. In fact, I made this mistake in a comment at Lucia’s The Blackboard this morning. Refer to my Comment#49191 at the bottom of her post NOAA: Hottest June in Record. This illusion can be seen in the following .gif animation of GISTEMP trend maps for the period of 1982 to 2009. The April trend is presented in Figure 7. Note how, in the highlighted area of the North Atlantic, there are differences between the SST trend and the trend of the GISTEMP combined product with 1200km radius smoothing. The Faroe Islands are located between Scotland and Iceland, and GISS uses station data there, so that explains the differences in that area. But what of the area of the North Atlantic west of Ireland and south of Iceland, with the approximate coordinates of 50N-60N, 20W-15W? There aren’t any islands there with weather stations.
http://i29.tinypic.com/2i0vhif.jpg
Figure 7
In its GISTEMP LST products, GISS also includes surface station data identified as Ship followed by a letter; that is, “Ship J”, “Ship R”, etc. Refer to Figure 8. These can be found using the station locator feature on the GISTEMP Station Data webpage.
http://i32.tinypic.com/2i09k7r.jpg
Figure 8
Here’s a link to the webpage presented in Figure 8.
I have found little to no information on these GHCN “ship stations”. Are they presenting SST or Nighttime Marine Air Temperature? Dunno. They may have served a purpose when GISS first prepared their GISTEMP product due to the sparseness of SST data in the early SST datasets, but now, these “ship stations” only add an unknown bias to well-documented optimally interpolated SST data. (And if they don’t add a bias either way, then there’s really no reason to have them. All they do is add confusion.)
How about someone answering a more basic question –what is the legitimate scientific need to have two different products, 250km vs 1200km? To me, granularity is a very important topic, and you should drive to the highest degree of granularity you can reasonably produce. This is probably part of what’s wrong with Mann’s hockey stick, btw –even Briffa seems to recognize that.
If there are significant differences (as there appear to be) between the 250km smooth and the 1200km smooth, then to me that is merely evidence of the advisability of ash-canning the 1200km smooth.
Excellent analysis, but you missed something important. You cannot simply blow off as inconsequential the bleeding of LST data over large swaths of SST data. Everyone knows the oceans are the dominate temp regulator, not the land. All over the Earth you can see higher LST’s over riding lower SST’s due to the GISS sloppiness.
Everyone knows large bodies of water will attenuate large changes in temp. Regions next to oceans tend to be warmer in winter and cooler in the summer.
So I see no scientific (let alone logical) reason to bleed LST’s over water where the SST’s directly measure a lower temp. Is it a surprise the shifts on the GISS chart appear mainly towards hotter temps?
You nailed the problem when you showed the slight shift in percentage of LST verses SST.
What you should have also nailed was the error in measurements of both data sets going back in time. There is NO WAY SST data has any accuracy prior to satellite coverage. While I would guesstimate the LST to have errors of +/-3°C over 1200 km up until the age of satellites, the SST data prior before the 1980’s is garbage.
Which means 70% of the historical data from which we determine whether it is warm now or not is garbage.
These ‘scientists’ who claim they can confidently detect a sub °C increase in global temp since 1900 are rank amateurs, slavishly lost in statistical noise and seeing ghosts.
Peter_DTM has it exactly right. These are Ocean Station Vessels that were discontinued decades ago.
See: http://www.uscg.mil/history/webcutters/rpdinsmore_oceanstations.asp
For a map showing where these ships maintained stations and took weather observations including two radiosonde runs every day.
davidmhoffer: Sorry. I didn’t have time to finish my earlier reply.
The bottom line: Here’s a comparison graph of the GISTEMP combined land plus sea surface temperature (1200km radius smoothing) with the global data (90S-90N) and the global data minus the Antarctic and the Southern Ocean (60S-90N):
http://i30.tinypic.com/11awlqb.jpg
Note that there is no difference in the linear trends. So after all is said and done, all of your speculation that the addition of the Antarctic data after 1950 changes this and that is really meaningless.
AJStrata says: “You cannot simply blow off as inconsequential the bleeding of LST data over large swaths of SST data.”
I didn’t. The GISTEMP LST data extends out over the oceans in the GISTEMP land surface data only. They mask the LST data over the oceans when they combine it with SST, as I noted in the post.
Regards
Rod Smith: Thanks for the link to the Coast Guard Weather ships. The coding appears to be the same as those used by GHCN and, in turn, by GISS.
Bob Tisdale says:
July 18, 2010 at 1:16 am (Edit)
Steven Mosher wrote: “WRT ships. In my code these ships recieve ZERO weighting, they are dropped.”
But they appear to influence the trends.
***************************************************
I think you are missing the point. in MY CODE I handle ship reports differently than GISS. It’s a methodological choice.
I regard every record in GHCN as a point source. Those point sources are allocated
to an area cell. In the case of a ship it is allocated to a cell that has ZERO LAND.
To calculate the temperature for that cell I then do the following:
Multiply the temperature by the LAND FRACTION in that Cell. if there is no land
in the cell it doesnt get weighted. no value.
Then I test the sensitivity of this decision by Not using land fraction. In this case
the ship cell recieves a FULL weight. Based on area.
The difference between these is real but minor. Its a methological uncertainty.
According to World Weather Records, mean temperature computations for Ship Stations are “based on 8 observations per day, made at 3-hourly intervals”. So in GHCN, they are neither SST nor NMAT.
Hi Bob!
My article that Davidmhoffer refers to is this:
http://wattsupwiththat.com/2010/07/17/tipping-point-at-giss-land-and-sea-out-of-balance/
As Ron has pointed out and as I pointed out to someone who raised this issue to me a while back this phenomena is not the result of GISS throwing any data away or sneakily weighting data. Its a result of a METHOD. a method of creating a global average when you have land data missing. The SST datasets that everybody uses
are infilled so that they are complete. every 5 degree cell is present. For the land,
thats not so. There is another issue. Land FRACTION. CRU cover this in detail in their paper. We have noted this as a difference between CRU and GISS over TWO YEARS
ago. There is no “right” way to handle this issue in constructing an average. So what we do ( zeke, chad, myself) is explore the different ways of handling this issue. Yes, you get different answers. marginally different answers.
AJ
“So I see no scientific (let alone logical) reason to bleed LST’s over water where the SST’s directly measure a lower temp. ”
It depends. If what you are trying to do is capture the TREND, then what you want is a consistent method OVER TIME. For coastal cells there are several ways to approach the problem. GISS is just one. as a METHOD it is brittle WRT changing frequency of reports from coast stations.
That’s one reason why I prefer to use land fraction. I think land fraction is the “right” approach, but GISS approach does “work” to capture changes in trend. I dont think it works as well as others, however
Bob,
As other folks have mentioned, the GISTemp land record is not actually a land record; rather its an attempt to estimate global temps using only land stations. This means it ends up using a zonal weighting (and no land mask) such that it runs considerably colder than other land series. Linearly combining published GISTemp land record with the ocean record is not strictly meaningful.
As a quick experiment, try examining the land fraction implied by combining NCDC land with Reynolds instead of GISTemp land. You will see that interpolation plays a role, but the largest factor in the odd 70% land finding is simply that the GISTemp land record is being used inappropriately.
Thanks Bob, that looks like a lot of work. The bottom line wrt temperature measurement seems to be that lacking enough data points, the numbers can be fiddled with to look like anything. I have to agree with AJStrata that parsing a fraction of a degree of ‘climate’ change/century from this mess isn’t science.
Steven Mosher says: July 18, 2010 at 8:10 am: Thanks for the clarification. I had assumed you were discussing how GISS handled the Ship Stations.
Frank Lansner says: “My article that Davidmhoffer refers to is this”
Is it the same as the one at Jo Nova’s?
Going from Fig 1-3, 1998 goes from highest temp to being surpassed in 2007. I recall many months ago 1998 was heralded as the highest temp on record only to be corrected to second highest after the 1930s high. We certainly aren’t informed as to the rubbery nature of these curves. Would it not be best to stick to land temps if we are trying to guage warming – its where most of the disasters are supposed to occur. We could have separate trends for land and sea, especially since the SST is basically only a few millimetres deep.
One fine day Pea met
Thimble, Thymble, and Thuemble.
You pick the vowel.
============
Is it the same NASA´s GISS where J. Ali Mustafa Califa Ak Bhar Hansen, a.k.a. the famous death trains´Emir works?
Now, they must and should care to reach, ASAP, the Crescent Moon. LOL!
And….no amount of intelligent and elaborated statistic corrections will ever convince the followers of the most holy Global Warming religion, of changing their faith, but the chilling reality of shorter summer seasons and earlier and longer winters.
Steve Mosher, you wrote:
“As Ron has pointed out and as I pointed out to someone who raised this issue to me a while back this phenomena is not the result of GISS throwing any data away or sneakily weighting data. Its a result of a METHOD.”
Steve, do you think that if someone creates a “method” i cannot be .. well … rotten? A method cannot be developed with “so-so” intensions?
Steve, in CRU program code released in climate gate its a permanent method that temperature proxy graphs are cut of in 1960, and the same can be read in their mails. So they have “a method” to cut of data most often in 1960, in then the sceptics cannot complain?
Just because “its a method” then no critique? No?
In the present case for example we have a GISS LST+SST that is the same as the SST (HADISST) 1900-1920. The GISS LST is higher. So the weight of land data is zero. Obviously this is build into a method , a program etc. and then its suddenly ok??
If you think so, you are giving the Alarmist such an incredible easy time to get away with anything.
The argument: “Its a METHOD” changes absolutely nothing, zippo zilch nada.
A land% = zero in year 1900-20 appears rotten even though its obviously result of a GISS “method”.
This reminds me of H.C.Andersen, the emperor has no clothes on. Oh.. but its a “method”!!! Then its ok.
The Ships are unmanned buoys which measure air temperature. they use some from Canada too.
Bob says:
July 17, 2010 at 8:36 pm
“If the data shows warming, attack the data, and the methods used to collect and analyse them. If the data shows cooling, no problem then!”
And if the land data is suspect (Refer to the Surface Stations Project) and is applied to the oceans (read the post above), then publish it and scream “WE’RE ALL GONNA DIE!!!!!!”
(Thanks Mr. Tisdale. Good post.)
Gary Pearse says: “I recall many months ago 1998 was heralded as the highest temp on record only to be corrected to second highest after the 1930s high.”
Wasn’t that the U.S. surface temperature record?
Bob Tisdale;
The bottom line: Here’s a comparison graph of the GISTEMP combined land plus sea surface temperature (1200km radius smoothing) with the global data (90S-90N) and the global data minus the Antarctic and the Southern Ocean (60S-90N):
http://i30.tinypic.com/11awlqb.jpg Note that there is no difference in the linear trends. So after all is said and done, all of your speculation that the addition of the Antarctic data after 1950 changes this and that is really meaningless.>>
It isn’t the specific data that is added it is the manner in which it is handled in the analysis that matters. Sorry, but I am far more familiar with the 250 km smoothing than the 1200, so this is from the 250 km data |(downloaded in answer to your question from NOAA’s site because they provide it in NetCDF format). In 1880 there are just over 8000 grid points with data in the combined data set. There are about 1100 land and almost 500 of them overlap with ocean. The official GISS combined anomaly runs very close to the SST anomalies. By 2009 there are over 5000 land grid points and a total of over 12000 in the combined set. As you go from one end of the graph to the other, the combined anomaly is increasingly biased toward land and diverges from SST while in Hadcrut it stays about the same between SST and Land. (Thanks to Steven Mosher for advising that this was a known issue, wasn’t aware of that, only started looking at this stuff 6 mos ago).
Now I don’t know what the right ratio is, I think Frank Lansner suggested 30% land 70% ocean. For trending purposes, I just don’t see how you can change the weighting between land and ocean of the combined anomaly based on the number of places you have weather station data for. Suppose the number of land cells doubled, but their new ones had the exact same average anomaly as the original ones. In Hadcrut the combined anomaly would not change, but in GISS it would because a larger fraction of the data is now Land.
Measuring something in more places shouldn’t change the final result if the anomalies are the same.
Zeke “As other folks have mentioned, the GISTemp land record is not actually a land record; rather its an attempt to estimate global temps using only land stations.”
Very true, Hansen et al (1999) was still praising the use of Land Surface Tempertures only as a means of estimating global temperatures. They may also have done it in later papers.
You continued, “This means it ends up using a zonal weighting (and no land mask) such that it runs considerably colder than other land series.”
Some of this should also result from the use of smaller island (and ship) stations, which will track closer to the cooler SST anomalies. There are very few islands used in the CRUTEM and NCDC land surface data, and those islands are chunks (technical term) of land mass such as Madagascar, Great Britain and Iceland.
You wrote, “Linearly combining published GISTemp land record with the ocean record is not strictly meaningful.”
Which was why I wrote this post.
BTW, disregard the note I left for you over at Lucia’s. As noted in the post, I discovered my error. I had thought GISS extended land surface data over the oceans, but what I was seeing was the impacts of the Ship stations.
Regards