HadCRUT/GISS/NCDC Global Comparison

Ok I have finally tracked down all the Surface Datasets to be able to put them all into one graph. Up until now I only had NASA GISS for surface data, because to be frank GISS is the easiest one to find, you just punch GISTEMP or just GISS into the search box on the NASA site and bam you get the links which are easy to understand and follow to get the data. So hats off to Dr.’s Hansen and Ruedy for making their data easy to find and download. However here is the link to the GISTEMP page: http://data.giss.nasa.gov/gistemp/

The CRU data can be found here: http://www.cru.uea.ac.uk/cru/data/temperature/#filfor

Their CRUTEM3 product and combined Hadley Center product called HadCRUT3 are just anomalies only based on the 1961-90 time period. Unlike GISS which on their annual anomaly product page gives an absolute temperature value for the 1951-80 time period to turn their anomalies back into absolute temperatures, CRU does not so far as I have found. If anyone knows where it is on that CRU page or any other I would appreciate it in the comments that way I can put them onto the same baseline with satellite anomalies.

The NCDC analysis can be found here: http://www.ncdc.noaa.gov/cmb-faq/anomalies.html#grid

NCDC does give an absolute temperature to add back in to change base lines to the Anomaly products they have. They also are the only ones that use a long term (century length) baseline they compute anomalies from.

So lets start comparing the datasets. All will be based on the 1961-90 baseline.

Here we have the Global Trends (Land + SST) for all three major dataset. Notice that HadCRUT has two versions, HadCRUT3v is variance adjusted according to the CRU website.

What you can see is that both HadCRUT3 trends are virtually identical at about .75° C. Also notice that NCDC and GISS trends are slightly less then HadCRUT 3 at about .73° to .74° C. Now from this you might get the impression that the analysis methods that each uses, even thought they are different from each other, shows that that they are all “robust”. That is a false impression which can be shown in the next graph:

Now the above graph is land only trends so that means you are looking at only the CRUTEM3 portion of the HadCRUT3 product. Notice that the the CRUTEM3, both versions, have a trend of almost 1° C. NCDC has a trend of about .93° C and GISS has one of about .76° to.77° C.

Now what is causing all this?

Well it is the way that each agency uses the base product GHCN. According to Dr. Phil Jones of CRU he claims that 90% of the CRUTEM3 product is from GHCN data and we know from the GISS page that they use the GHCN dataset and the same for NCDC (hey still got to check even though you would think they would use it since they are the ones compiling it). The way they put all the data together shows that they do not support each other, NCDC and CRU might be called in close agreement but the GISS trend is way off by almost a quarter of the CRU trend. (note GISS uses USHCN adjusted data in place of GHCN in their analysis for the area covered by the United States)

So why does all of them come to such close agreement in the Land+SST sets?

Well one reason would have to be that they all pick and choose which SST dataset to use, well I should qualify that and say GISS picks and chooses. The HadCRUT3 product is the combined product of the CRU and the Hadley Center and since Hadley is the compiler of various SST products you could reasonably expect them to use one of their own products. In this case it is the HadSST2 product. The same can basically be said for NCDC since they are part of NOAA who compiles the ERSSTv3b dataset which is used in their analysis. Now GISS the one that varies the most in the land trends from the other two, switches the SST dataset they use. They use the HadISST1 dataset for the years 1881 to 1981 and the OISSTv2 from 1982 to the present. Now these datasets are basically the SST versions of CRU, NCDC and GISS land data analysis, they are all based on the ICOADS data compiled by NOAA. So the combined dataset isn’t based on just one analysis of all the data but 3 different analysis of land and 4 different ones for the SST datasets used. The reason that is important is that the SST data is over 70% of the data in the combined dataset since the oceans cover over 70% of the earth so by changing which SST dataset you use you will change your global trend.

You can check and see how the different SST datasets effect the analysis by going to this page on the GISS site:

http://data.giss.nasa.gov/gistemp/maps/

There you can set the analysis to no land and select the SST data that GISS uses in its analysis and generate a trend map. When you do that you will get a trend # in the upper right hand corner. Here is the map for the SST datasets that GISS uses:

Note that the trend for that is .58° C

Now go back and set the SST data to the NOAA ERSSTv3b SST dataset and make a map for the same time frame. This give this map:

Note that the trend changed to .62° C. Unfortunately I can’t make a map like that for the HadSST 2 dataset on the GISS site, but you can get the idea that there is differences in those datasets as well.

What this all shows is that the agreement between HadCRUT, NCDC and GISS Global land + SST is a product of the way they pick and chose which SST dataset they use and how they stitch them into their land analysis. When you just do an apple to apple comparison by looking at all three independent analysis of basically the GHCN dataset it shows there is no agreement between all three, just a general one between NCDC and CRU.

Advertisements

2 responses to “HadCRUT/GISS/NCDC Global Comparison

  1. Chris Poynton March 8, 2011 at 6:08 pm

    Can you advise if your graph comparing the various temperature data sets can be automatically updated as each months data comes in? Could you set up a link where such an updated graph could be viewed? (I am helping to run a climate commentary website in Australia called http://www.thelongview.com.au and we are trying to source good transparent information sources such as the one you have provided). Many thanks, Chris Poynton (VIC, Australia)

  2. boballab March 8, 2011 at 9:16 pm

    @ Chris Poynton

    The answer to the question “can they be automatically updated” is Yes and No, it really depends on what you mean by updating as the new month’s data comes in.

    If you mean you download the entire dataset each month, then the answer is yes. You just need to write up a program to go to each of the sites and download the set and put it into a spreadsheet program. This is basically doing what CRU and NASA GISS does with the “raw” data that NCDC acquires except they then do their own analysis of it. Each month CRU and GISS automatically download the latest copy of the GHCN “raw” dataset and run it through their computer programs. Once that is done they publish the result of their new analysis.

    If on the other hand you just wanted to download the latest months data only and add it on to what you already have, the answer is no. The reason for that is that the older data changes every now and then as the GHCN file is updated. Sometimes the reason for this, is that NCDC receives late data from the previous months, even to the point that some stations do not report every month as they should but once every 3 months or longer. To put that into context when NCDC, CRU and GISS update for March there is no guarantee that any previous months anomaly will stay the same. Another factor that changes the past is old records are found and added into the GHCN record and again changing the anomalies of the past, and to be honest there has been some data that should be in the GHCN file but for whatever reason isn’t. A classic example of this is that NCDC chops off data in GHCN for the station Alert, Canada. You can go to the Canadian governments own website and find temp data from 1951-2005 but for some reason you find only up till 1991 in GHCN “raw” and they chop it off earlier in the GHCN adjusted series. So even though CRU, GISS and NCDC start out with the exact same data (which in itself is truncated) GISS uses the entire Raw dataset while NCDC prunes it even more.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: