Interesting GISTemp Trend Maps

There has been much talk across the blogosphere about what happens in GISTemp when station data “drops” out. Such questions as how does it effect the gridded anomaly maps and does it effect the trends are being asked and people are attempting to answer. 

Now so far I have seen a few suggestions and attempts to prove or disprove what station drop out does. One suggestion was to prune the GHCN “raw” dataset of stations that don’t make it through the “Great Dying of Thermometers” and then run the version of GISTemp published on the GISS website and see what it spits out and compare it to what it gets when you run the full GHCN dataset. Now this is logical and should be the way it works but according to EM Smith, who has a copy of GISTemp up and running, GISTemp gets cranky when you prune the dataset it uses and crashes.

Now this then leaves us with various other attempts including some making up their own statistical method that is NOT the way GISTemp is suppose to operate (at least according to Dr. Hansen’s published papers and I’m not talking about RomanM). By changing the way the stations are put together in their grid cells then trying to see how loss of stations effect the output does not tells us how that effects GISTemp. If your method shows no difference, congratulations, publish it and maybe GISS will switch to your method. However until such time we need to know how loss of data effects the GISS method not yours, so you didn’t prove a damm thing about GISS. 

Now every one of these attempts are all looking at trying to simulate what GISTemp does in one way or another, instead of actually seeing what GISTemp really does do. That was when I had a bright idea: Why not see what GISTemp does by changing the settings on the GISTemp gridded map maker program on the GISS site.

Now this program allows you to change baselines, change the length of series, change if you use SST anomalies or not, see an Anomaly map or a trend map and what is critical for this to change the amount of infill from 1200km to 250km.

Why is the infill radius critical?

Because what are you doing by taking the infill out but simulating the loss of station data in GISTemp. 

Keep this in mind that at the 1200km range you get a lot of “estimated” data thrown in and at 250 you are closer, but not exactly to, seeing straight up data from a thermometer. Now awhile back I did a short animation called “March of the Thermometers”. What this was is GISS generated Anomaly maps at the 250km infill level, one frame per 10 years. This lets you see how thermometers spread and contracted around the globe, however this time around we want to see not the spread or loss of thermometers but the change in trends for each grid cell over time. Now the only way that GISS can do this is by comparing a temperature anomaly at one time to other temperature anomalies in the past for the same grid cell.

Now the contention is that in the 1950’s, 60′ and 70’s it gave one anomaly made up of x amount of reporting stations that were combined, now here in the 2000’s you no longer have x amount of stations in that grid reporting; you got y and that if you still had x amount the anomaly value for that grid cell would change, therefore it would change the trend. This is then compounded by the GIStemp Infill function, because in some grid cells you have actual observations from the 50’s and 60’s and now you got “estimated” data being compared to it and that is not a good comparison.

 At least that was the conjecture.

By playing around with the GISS map maker program and if it is an accurate reflection of the program GISS uses for their official gridded maps, I found something quite startling to say the least.

First here is the set up: 

GISS land analysis selected 

Ocean: None  

Map Type: Trend 

Time Interval: 1880 to 2009 

Base Period: 1951 to 1980 

Smoothing Radius: 1200km 

Projection: Regular 

That is what we are starting with and here what you get: 

Figure 1

Notice that most of the land area is covered except for one area that we know there has been observations from since at least the 1950’s: Antarctica. So right off the bat we got a mystery that I believe I solved shortly thereafter when I changed the Smoothing Radius aka the Infill from 1200 to 250km and got this:

Figure 2

Holey Toledo look at that most of the land area went to “No Data” gray!

Now this begged the question of why this happened and at first I didn’t know but there was something awfully familiar looking about that map. Then it hit me: this 250km “1880 – 2009 trend” map looked like one of the early 1900 anomaly maps. So I went and pulled them up until I got this one:

Figure 3

Notice this map is Anomaly not Trend, it is on the same baseline but it is only for 1900 – 1901, but doesn’t the coverage area look mighty close to being the same?

This led me to my own hypothesis: GISTemp only makes a trend for any given cell as long as it has data either by observation or Infill to compare the present or end time selected to some pre determined time period where GIStemp has data in the past.

If you go back and look at Fig’s 2 there is Antarctica again with no trend for it, even though we know there is data from the 1950’s for that area. The same applies for all of Central America, a lot of the missing area in South America and other places on that map. To show this we will change the parameters to this for Fig 4:

GISS land analysis selected

Ocean: None

Map Type: Trend

Time Interval: 1950 to 2009

Base Period: 1951 to 1980

Smoothing Radius: 1200km

Projection: Regular

Figure 4

Now what do we see? Areas that in Figure 1 that were gray now have trends in them including Antarctica and all I did was change the start date from 1880 to 1950, everything else is the same. Now we switch to 250 Km in Fig 5:

Figure 5

Now this time unlike what happened when the start date was 1880 I didn’t lose that much, you don’t see half of the land area turn gray.

So that only can lead to one conclusion that GISS will only have a gray “No Data” area on its gridded trend maps if they can’t infill from what they have both in the present and in the past. That is why there was so much more gray areas in the 1880 – 2009 250km trend map: GISS had nothing to compare too in the 1880/1900 time range. That is why Antarctica is blank on that map even though there is data from 1950 to the present for that area.

So bottom line according to the 1880 – 2009 GISS 250km trend map the only grid cells that matter are the ones that have observations back in the 1880/1900 time frame or can be infilled from observations during that time. Everything else is not used.

Oh Btw for those wanting to know if a change in data will change the trend the answer is yes. Go back to Figure 1 and look in the middle of the 1200km infilled area of Russia. Do you see any white in there or only oranges and reds? Now look at Figure 2 and you will see a section that is white where before it was orange. In that case more data (ie more infill) caused a warming bias in that grid cell. Now can it happen the other direction yep take a look at Florida in Figure 2 there is a band of yellow running through it and it’s solid white (ie cooler) in Fig 1.

UPDATE: For those interested you can download the gridded data in text form from the map pages on the GISS site. Now I went and did a quick check on 2 Grid cells that are lat 73 N by 73 E and 73 N by 75 E the trend for both those cells in the 1880-2009 1200km setting is a whopping 1.4509. When you change the setting to 250 km the trend for those cells drops to -.6114. In this case a loss of data causes cooling. OTOH when you look at cell 41 S by 145 E which at 1200km is .0405 and cell 41 S by 147 E also at 1200km is .0319. When you switch to less data by going to 250 Km the trend jumps up to .7877 and .4595 respectively showing warming.

I have now shown that a change in data does change the output and it can go either way and the change is quite large in the first example it was a decrease of 2.1° with a loss of data and in the second example there was increases of .76° and .42° with loss of data.

So GISTemp is effected by Station dropout.

Advertisements

3 responses to “Interesting GISTemp Trend Maps

  1. E.M.Smith February 26, 2010 at 3:59 pm

    Very well done.

    Using only GISS tools, you’ve shown that “The Anomaly WIll Fix It” is a broken argument for GIStemp.

    So in addition to my benchmark showing GIStemp lets station change bias show through, we now have it without the need to change the input data at all.

    Nice.

    Now if only there was some way to sort out what causes them to swap the trend around so much based on start of time. Perhaps for another day… At this point I don’t know if that will be in the GIStemp STEP3 code, or if it is another artifact of their web graphics presentation engine. But it is there in the stuff being shown to the public in either case.

  2. Ruhroh February 26, 2010 at 7:00 pm

    I will not be surprised when they disable the 250km ‘infill’ parameter which seems more like an outreach parameter.

    I saw somewhere that it was intended as a ‘debug’ feature.

    I just don’t think they wanted to have users finding warts with it…

    I expect you could get some variably odd results by sweeping the end year or beginning year to see things come online or go away.

    That was what I was doing with the ‘self-anomaly’ thing when they coincidentally decided to disable it.

    Remember to scroll down and snag at least the PDF of each interesting map, if not also the data.
    Fortuitously the PDF filename does fully capture the relevant variable parameters.
    You never know when they will decide to disable a insight-enabling feature.

    RR

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: