GISS Infilling the true Hypothetical Cow
March 9, 2010Posted by on
UPDATE #2 Added in an image of the Grid Cell. Grid Cell Center and Alert’s location according to GISS via Google Earth
UPDATE: Took out the none functioning links and just made tables from the GISS data page for the referenced data and put a link into the main GIStemp map making page where people can set the parameters themselves and get the data from GISS.
The debate on does losing thermometers cause a bias to show in trends continues to rage across the blogosphere and the Statistical Judo is astounding. I as well have been looking at this but I take a different approach. Instead of looking at “cartoon” worlds and other artifical constructs I rather just look at what GISS tells us with their own computations, data, maps and graphs made from their data. You see NASA GISS has this neat utility on their website that lets you make gridded Anomaly Maps with all the neat colors corresponding to how much warmer or colder it was based on a selected baseline. What is less well known is that you can look at Trends and make maps of those, again in nice colors. What is even less well known is you can download the trend or anomalies for each of the 2° x 2° grid boxes.
Sidenote here for some that refer back to the Hansen et al 1987 paper, there is an update to that paper Hansen et al 1999 and there is now 16,200 2° x 2° grid boxes not the 8,000 others have mentioned. I found this out the easy way of actually reading that paper where Dr. Hansen states the size of the grid boxes and then after downloading the data for a GISS Gridded Map into a spreadsheet took the row count, which tells you how many grid boxes there is and it also confirms the size of the boxes.
Now the thing that trips everyone up is that NASA GISS does infilling where they don’t have actual thermometer readings and then compare that derived data against historical readings. Now this infilling does induce a radical bias into the data and it is dependent on the infill radius or as GISS calls it the “smoothing” and I will demonstrate this by using a grid box on the edge of nowhere, that had only 1 station in the grid for its history. The Grid Box that I’m using is centered on 83N by 63W or in the GISS data -63 long by 83 lat. This box holds the station of Alert Canada:
Acording to GISS the useful data runs from 1951 to 1991 (however there is no data for one whole season (SON) in the 1991 data so I usually drop it). So lets start by showing the Adjusted station data according to GISS for the 51-80 baseline:
Now as you can see there is a slight cooling trend for Alert during that time period. Now you would expect that the GISS Trend Map would show roughly the same thing at the 250km infill radius during the same period and that is what you see here:
Now to see where Alert is look way up north just to the west of Greenland, but we don’t need to rely on that imprecise color scale at the bottom of the map. You see on the GISS website ( http://data.giss.nasa.gov/gistemp/maps/ ) you can access the data to make that map. Now on that page scroll down to you get to the right co-ordinates which is bass ackwards from regular use so look for -63 Long and 83 Lat. Once you found that you will see the trend is listed at -.4184 which corresponds to that sickly green colored boxes up there. Here is a table of data taken for that map:
also of note notice that there is more then one box colored that green including some ocean boxes. That is because the data from the box holding that one thermometer is being used to determine the temperature trend for those other boxes which have no thermometers in them.
So lets see what happens when we put things back to the 1200km “smoothing” radius that is the “official” one used by GISS:
Again we don’t need to try and guess what the trend is from the color (-.2 to -.5) we can look at the data:
and according to the data the trend is still -.4184. So far things are looking ok but did you notice that the color of the boxes to the southwest of Alert changed color? That’s right the trend changed in those boxes to a cooler trend because when we went to 1200km smoothing we averaged in 10 different stations that at least partly over laps the time span of the Alert Station.
That brings us to what happens after 1990, and does GISS infill the box that used to hold the Alert Station?
The answer to that is yes they do, so lets expand the time frame out to 2005. (I know you are asking why not out to 2009, you will see why 2005 later).
So lets look what the GISS Trends are at both 250km and 1200km (I’m not going to show the pretty pictures this time).
At 250 km the Trend is: -.3246
At 1200km the Trend is: 1.1616
Now that is a huge change of almost 1.5° C just by increasing the number of thermometers being used to infill box -63 Long by 83 Lat and GISS no longer having data available from that station. Or do they? The answer to that question is yes and no.
No in the sense that the source that GISS uses GHCN doesn’t have the data, yes for the fact that the Canadian Government (Environment Canada) has made their data readily available for downloading ( ftp://arcdm20.tor.ec.gc.ca/pub/dist/CDCD/ ) and what do you know there is data from Alert up to 2005 (this is why I choose that year earlier). So lets take a look at what the GISS adjusted data looks like compared to the Canadian data:
Now there you can see that the Canadian Data and the GISS Adjusted data are in very close agreement during the years 1951-90, so lets add in the Canadian data for the Alert Station post 1990 and see what trend that gives us.
Look at that the trend when using actual data for that entire time period according to the Canadian Data is about .4 which is only about a third of what the 1200 km infill trend is.
As an added bonus I went and downloaded the Gridded Anomaly data for every year from 1880 to 2009 for 250 km infill so that you can graph that compared to station data. So here is the graph in figure 5 with that added in:
As you can see the 250km gridded anomalies from 1951 to 1990 alignment with the GISS Alert Station data is visually exact and very close to the Canadian data. So when we had an actual thermometer being used the gridded data for that box matched it, but once you lost that thermometer depending on how much infilling you do you get two wildly different trends and when you compare it to the data that is on the Canadian site you see that both the 250 km and 1200 km infill data trends for grid -63 Long by 83 Lat is wrong.
So as of 2005 the best trend is based on the EC data of .4° C from 1951 to 2005. So if the infill in GISS from 1990 to 2005 for that grid was that far off, how accurate is the trend of 1.6057° C from 1951 to 2009 with 1200km infill or 2.1158° C from 1915 (first year they could infill) to 2009 with infill both before the 1951 to 90 period and after?
My answer is slim to none and slim just bought a plane ticket to the US Virgin Islands.