Cooling is Not Impossible

Recently, while discussing the Berkeley Earth temperature record (BEST), I made the comment it seemed every station showed a similar warming trend in recent times. I decided to test that idea by looking at the last 50 years or so. To do so, I created a map of linear trends from for the 1960-2013 period:


You’ll note, the scale of the map begins at 0. That’s because there isn’t a single point on it below zero. According to BEST, not a single location on the planet has cooled since 1960.

We can confirm this by looking at a histogram of the trends:


Of course, this is just for one period. I tested to see if this pattern held for other periods. Try periods as far back to 1900, I found the same thing. The world simply didn’t cool. However, I did find cooling when I tried going forward in time. Here is a map of trends for 1980-2013:


The colors of the two maps don’t match up as I haven’t worked out the kinks of coloring the plots yet. Still, you can see the scale of this map begins below zero. That means some trends were negative. To see how much of the globe was cooling, we can look at a histogram of the data again:


There are 585 negative trends (out of 24,311 calculated trends), all of which are very small. That shows cooling isn’t impossible in the BEST record. It’s just very unlikely to happen in modern times, and it will only happen with relatively short segments.

Of course, there are always concerns that the periods chosen aren’t representative. To address this possibility, I’ll show the trends for any period people are interested in. Feel free to ask for any you’d like to see.



  1. Thanks!

    I was tempted to compare these to GISS in this post, but I decided against it because I want to give that a post of its own. Before I do that though, I need to see if I can find HadCRUT gridded data for land only. If so, I’ll compare all three data sets. I also want to see if I can find a way to make their maps look better. Because of its larger gridcell sizes and more limited data, GISS’s maps look way uglier. Also, GISS’s data is less smooth so there outliers, and that screws up the levels for coloring and histograms.

    But since I don’t know how long it might be until I make a post for that, I’ve gone ahead and created a couple histograms for comparison purposes. Here are GISS’s trends from 1960-2013. Here they are for 1980-2013. Neither show the more extreme outliers.

    I think later on I’ll try to create images which overlay the histograms on top of one another.

  2. Ah, thanks.

    By the way, you might be interested in two images I just created when testing out how to overlay density histograms. This one shows BEST and GISS trends from 1960-2013. This one shows them for 1980-2013.

    Blue is BEST, orange is GISS, and brown is where they overlap. I don’t care for the color scheme, but I think they are interesting images nonetheless.

    I’m not sure there’s any good way to do that with four data sets though. I’ll probably just compare each one to BEST individually.

  3. I’m sorry; each one to run up to 2013 please.

    But, if it isn’t too much work could you also show 1910 -1940 and 1976 – 2006.

    thank you

  4. Shub Niggurath, ggplot has a lot of great plotting options. I just don’t feel like spending the time to familiarize myself with the package right now. There are so many R packages I’d like to learn. Fortunately, I found a useful alternative.

    Armando, no problem. For the moment, I’m only posting histogram comparisons like what I posted above. I can post maps if you’d like too, but I find these more interesting since they let us compare BEST with GISS. For these, blue is BEST, orange is GISS, and brown is where they overlap:

    Here is 1945-2013.

    Here is 1989-2013.

    Here is 1998-2013.

    I’ll try to get the results for the other two periods calculated shortly. If you’d like the trends of any of these mapped as well, let me know. I have all of these results saved. I’d just need to make/upload the images.

  5. Alright, but I haven’t put any time into getting them to look right. None of the colors match up because it’s a pain to try to align levels and palettes in this. On the upside, revisiting the data made me realize my 1976-2006 image is messed up. I inadvertently hit the Insert key which led to me screwing up a line. That led to me using the 1910-1940 BEST data when doing the 1976-2006 comparison as my variable didn’t get reset. This is the correct version.

    Anyway, onto the maps. Here is 1945-2013.

    Here is 1989-2013.

    Here is 1998-2013.

    Here is 1910-1940 (the missing spots in this one is due to there being less data for the period).

    Here is 1976-2006.

  6. Carrick, time wouldn’t be a problem, but that period is long enough memory is an issue. I should be able to do that whole subset at once if I free up RAM R is currently using, but that’d require dumping some variables. The other option would be to do it in two pieces.

    Neither is a big deal, but they do both take a bit of time, and I have a number of other things I need to get done today. I’ll try to have it done this afternoon.

  7. Brandon

    Thank you very much.
    It would be nice if you would use the same scale (colours) for all maps and draw isothermal lines.
    Tell me if ask too much.
    My aim is to see where the warming was from 1910 – 1940 compared to 1976- 2006.
    Where there is no warming from 1989 on because there is a big jump in temperatures around that time in some regions.
    1945 -2013 Is about Greenland but that should have been 1940 – 2013
    And 1998 is obvious.

    BTW BEST makes maps too:

    And the emperical breaks make me laugh see f.e:


  8. Armando, I think I may have figured out how to use the same scale for all the maps. I don’t know about isothermal lines though. They’d probably be trickier. One thing to remember is before last week, I hadn’t done anything with maps like these. I’ve been learning the code as I go.

    Carrick, I had to do it in two parts, but I got your 1880-2010 period handled. Here is a map of trends. Here is the histogram comparison of them.

  9. By the way, one problem with using the same scale for each map is the different period lengths. Trends calculated over shorter periods can have a lot more variability. That means if you use the same scale for a map for 1998-2013 and 1880-2010, the former will have more spatial information than the latter.

  10. Thanks Brandon.

    I was surprised to see that BEST has a reduced variance in trend compared to GISTEMP.

    A few suggestions (please feel free to ignore):

    I do wish there was a way to more easily distinguish > 0 from < 0 in your color scheme. You might want to multiply numbers by "10" to get °C/decade so you don't get numbers clipping off the right. Also indicating the scale in your label, e.g., "Linear trend from 1880-2010 (°C/year)" is a good idea.

    One trick people as the scale changes is keep the same color scheme but add banding or contour lines (if contour lines, divide the actual range shown in an image, then draw one every 1/10th of that range. See e.g., this for an example in R.

    I thought of another way of displaying your second figure: Scatter plot. Horizontal axis is trend in GISTEMP, vertical axis is trend in BEST.

  11. On the scatter plot…. what I meant was to plot each grid point as BEST trend (vertical) versus GISTEMP (horizontal).

  12. Carrick, I wasn’t surprised to see that. I figured it was inevitable once I saw how much spatial smoothing there is. If I hadn’t known about that though, I definitely would have been surprised.

    As for your suggestions, I agree with you about the color scheme and scale issues. I think I understand the syntax for modifying color choices and scale levels, so I think I should be able to improve those. It’s just a matter of taking the time. I’ve been spending so much time working on the underlying stuff I haven’t had the time to work on improving the presentations.

    For contour lines though, I don’t know how I’d add them. It might be worthwhile, or it might make things too cluttered. Either way, it’ll probably be a while before I try since it’d take longer to figure out how to do it.

    On the scatterplots, I do like your idea. I’m just not sure how I’d implement it since BEST and GISS have different gridcell sizes. That means I can’t directly compare their values. It also means if I use all the BEST values, each GISS value will be quadruple counted. I’m thinking I’ll try it out using one BEST gridcell per GISS gridcell. With the amount of smoothing in the BEST data set, I don’t think that should be a problem.

    But as I said, time. Yesterday I had a lot of it, today not so much.

  13. Brandon—good point on the effect of smoothing. Can you link the text versions of the data for 1880-2010? I’d like to play with that a bit if you don’t mind.

    As to scatter plots.. a couple of things to consider:

    • collapse the BEST trends into averages over the GISTEMP grid cells.
    • use the centered BEST cell for each GISTEMP cell.
    • plot the BEST cell versus the GISTEMP cell that it is contained inside of.

    Regarding contour lines… I was just brainstorming there. Where I to put labels on a figure like this, I wouldn’t use labels for the contours, and I’d probably use e.g. another color than black for the line color for that.

  14. Try again:

    Where I to put contour lines onto a figure like this, I wouldn’t use labels for the contours, and I’d probably use e.g. another color than black for the line color for that.

  15. Carrick, text versions of what data? I’ve never made any text versions of any of this data, but I could output the data used for the maps to a text file. I don’t think I could do that with the temperature data itself though. There’s just too much of that (360 x 180 x 101 + 180 x 90 x 101 points).

    For the scatter plots, I would normally do what you describe in your first suggestion. The problem in this case is just writing the code. I believe it’d ultimately make my code use four layers of nested loops. That’s not appealing. What I’ll probably do is select one the southeast BEST gridcell within each GISS gridcell and use it. If that becomes an issue or I want to pursue the topic further later on, then I’ll do something more complex.

    On contour lines, it seems some packages for R make them fairly easy to add. I could probably fit in small black lines on the borders between levels if I wanted. I’ll try to find some time to improve the maps’ visuals tomorrow. No promises though. I just found out the bathroom next to my room is being remodeled this weekend. It looks like they may even have to tear out a couple feet of the wall of my room.

  16. Brandon, I was thinking text files of just the trends from 1880-2010 for each cell, not the raw data.

    Something like this:

    write.table(data, “data.txt”, sep=”,”)

  17. Carrick, that’s no problem. You did just make me realize something though. My previous comment said there were 101 years worth of data. That’s obviously not right. 1880-2010 is 131 years worth of data. I’m not sure why I thought you requested I use a different amount. To be clear though, I used the right data in making the map. I just had a brain fart when making that last comment.

    Anyway, here is a file with the BEST trends. Here is one with the GISS trends. WordPress doesn’t allow me to upload .txt files so I used some random web service I found. I’d recommend renaming the documents. Also, I didn’t include row/column names so you may want to add those. Coordinate values for GISS begin at -89 x -179 and go up by two. Coordinate values for BEST begin at -89.5 x -179.5 and go up by one.

    Let me know if any of the formatting got messed up.

  18. No prob.

    By the way, I managed to get things worked out so I can have those maps all use the same scale. I still have all the data saved for the maps I’ve made thus far, so I can easily go back and redo them if people would like. If anyone would like one redone, let me know. I might be a bit slow to respond though. The bathroom remodel just started. I’ve decided to help out even though I don’t have to. The opportunity to use a sledgehammer on walls is just too good to pass up.

  19. Brandon,
    It’s a bit late, but I just noticed that your trend scale seems off by a factor of 10, assuming you intended the units to be K/year. E.g., for the 1960-2013 plot, .001 to .0015 as the mode doesn’t make sense. On the other hand, if the units are K/month, then the scaling is probably correct.

  20. HaroldW, the data I used is monthly data, so I used K/month for the trends as that was the default. I could have multiplied the values by 12 to get K/year or 120 to get K/decade, but it didn’t seem important. All that’d change is the numbers displayed on the scale.

    I should have clarified the dimension of the scale though.

  21. Brandon –
    Thanks for the clarification. That makes sense and I should have assumed that was the case. As you say, K/month is a natural unit for trend if you’re processing monthly data. Since you’re comparing two metrics using the same scale, it doesn’t really matter which units you’re using. I only noticed it because I happened to contrast your histograms with some posted at Lucia’s (which had units of K/decade).

Leave a Reply

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

You are commenting using your 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