A Timely Release

I'm happy to announce to BEST has finally released some results it gets without adjustments. It's not a full release, but it's still a significant step forward, so for today I'll try not to be snarky. I'll try not to dwell on why they've waited years to release this data, why they falsely told the media they had already released the data, why they refused to correct their false statements or why they now still don't talk about it. I've talked about all that before.

So instead of dwelling on the past, let's instead... look at the past. Yeah, it's not really the best of transitions. Still, an interesting to look at the post I wrote almost exactly one year ago. It began:

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:

7-15-1960-2013

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.

Since I wrote that almost exactly a year before I gained access to this new set of results, ones generated without adjusting the data, obviously I have to try doing the same thing with them.

I couldn't find my old code, but it didn't take very long to recreate it. I was even able to almost perfectly reproduce the above image. I just couldn't get the color scheme to match up quite right. I figured that wasn't important, so I just went ahead and made a higher quality version of it:

7_15_best_adjusted

It could obviously be improved, but it'll do for now. The important thing is having code to make it means having code to make a similar graph for the unadjusted dataset. Again, this is a really crude first attempt at it, and the color schemes don't match perfectly, but the differences are striking:

7_15_best_unadjusted

Before I get to that though, I want to point out these images line up with one of the images BEST has published showing unadjusted result, which is always a good sign.

As for the differences, this shows BEST is not simply adding a bunch of warming to its record willy-nilly. As everyone should have already known, global warming is not purely some artifact of adjustments. On the other hand, the difference in spatial resolution between these two images is remarkable. The rate of warming in the unadjusted results varies quite a bit across continents, but in the adjusted data results, it is far more constant.

For now, I'm not going to argue about which is right. I'm just going to leave this post as an interesting update on where we are, a year later. In that vein, I want to briefly revisit one of my first posts about BEST which led to me discussing this gridded data. That post, titled "Is Best Really the Best?" compared BEST's temperature for the area I lived in to that given by another group, GISS:

7-10-home-trend

Naturally, I decided to update that comparison as well. Here is what I got:

7_15_best_il

It seems I may have been right when I said:

It’s amazing. From month to month and year to year, GISS and BEST look nearly identical. The high frenquency components of their graphs are indistinguishable. The only meaningful differences between the GISS and BEST estimates for my area is BEST adds a huge warming trend.

To be clear, I don’t think this means BEST is fraudulently adjusting the data. I’m not Steven Goddard. I suspect what’s actually happening is BEST is smearing warming from other areas into mine. That is, warming in other states is causing Illinois to seem like it’s warming far more than it actually is. That’s not fraud. That’s just low resolution in the estimates.

But here’s the thing. BEST is supposed to be the best temperature record. It has a website encouraging people to look at data on as fine a scale as individual cities. WHY?! If BEST can’t come close to getting things right for the state of Illinois, why should anyone care at what it says about the city of Springfield, Illinois?

In any event, I'll be looking at this data more in the days to come. I'm not posting a direct link to it though, as it stored in a location with some data for an unpublished paper, and I was asked not to access that data. I'd feel wrong if I gave a link to this data and someone wound up accessing the other data because of it. I'm still baffled they haven't released this data publicly despite all their promises, but... as I said, laying off the snark for today.

Anyway, until the data is publicly available, if there's anything you'd like me to look into, feel free to let me know. One thing I'm particularly interested in right now is how adjustments vary over time. I'm happy to look at other things though.

55 comments

  1. It would be nice to have a difference map and overlay the national borders. Comparing the two graphs, I have the feeling that Angola introduced a large change in its measurements system during this period. Maybe in a difference map one would be able to see more borders, even if the Berkeley Earth fields are very smooth.

  2. "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."

    wrong.

    That is not what the map means or represents.

    There well may be single locations that have cooled.

  3. Victor Venema:

    It would be nice to have a difference map and overlay the national borders. Comparing the two graphs, I have the feeling that Angola introduced a large change in its measurements system during this period. Maybe in a difference map one would be able to see more borders, even if the Berkeley Earth fields are very smooth.

    I could make a difference map very easily. I'm not sure about the national borders thing. I know I could combine these gridded files with map data to overlay the national borders, but I'm not familiar with the map packages for R at the moment. It might be easy, or it might not be. I'll have to look into it.

    Steven Mosher:

    wrong.

    That is not what the map means or represents.

    There well may be single locations that have cooled.

    Um, no. That's exactly what it means, when we're talking about BEST's gridded results. Or BEST's any results. As far as BEST's results predict,* no location has cooled since 1960. That you can point out those predictions are wrong does nothing to defend BEST; it simply emphasizes the point I'm making - BEST's predictions for individual locations do not match what people would expect or find useful.

    *Feel free to replace the word predict with project, interpolate or any other word that you feel fits better.

  4. That I can Armando. I'm going to create a difference map for Victor Venema first, but I should be able to get the others done shortly. I'm running a bit behind on things because of technical difficulties. My phone's charger port died on me a few days ago so I haven't been able to use it. Normally, that wouldn't be a problem, but I use it as my backup internet access point. Naturally, my wireless router decided to die a couple days later. That left me without internet for about 30 hours. The timing of it was very strange.

    Anyway, I found out there are a number of map packages for R that are pretty easy to use, but I haven't found an easy way to overlay them with these maps. I like these contour maps, but because they're generated in a kind of weird way, it turns out they're hard to make work with other things. Even just altering the axes and information plotted along them is a pain. Overlaying things seems out of the question. I should be able to make the maps a different way, but that's more time I have to spend finagling with graphics instead of something more productive.

    But yeah, I'll start running that code now.

  5. I wasn't sure if I should do the difference map as the difference of raw from adjusted or vice versa, so I went ahead and did both. The graphics aren't high quality as I still have a lot to learn about making things look good in R, but I think the results are still rather striking. While the maps in this post show you how smooth the results are, these new maps effectively show you how much smoothing the BEST adjustments introduce, as well as something of the nature of that smoothing. Check them out and tell me what you think:

    Adjusted - Raw
    Raw - Adjusted

    You can see there are large areas with significant warming/cooling trends added to them. Look at the eastern half of the United States, or practically anywhere in Africa. Actually, here, let me change the color scheme. I don't think the neutral (0) color stands out enough in these maps, so the visual effect may not be as clear as it could be. Here is the Adjusted - Raw map again, with a different color scheme. In it, blue is where cooling trends were added while yellow is where warming trends were added. I think it speaks for itself.

    I think I like doing Adjusted - Raw more than the other way around so I'll probably stick with that from here on unless anyone prefers I do the other (or both). I'm also going to go ahead and make these along with any other maps I make when I examine other time periods. If these sort of changes happen in a period with good data coverage, I'm curious what changes happen during other periods. I'll check Armando's 1990-2013 period next.

    (By the way, the new results do go up to 2014 while the other ones only went up to 2013. I'm excluding the last year of data so the comparisons are apples to apples. I don't think that should matter, but I wanted to throw that out there so it was noted.)

  6. Cool news. I figured out how to keep the color scale consistent from map to map. That means there shouldn't be any oddities between adjusted/unadjusted maps for the same periods anymore. It doesn't really work with the differences maps though. I mean, I can keep the same scale, but differences will inevitably be smaller than the values being differenced, so keeping the same scale would make for a hard to read map. I'm still working on what would work best for them. Anyway, here are maps for 1990-2013:

    Raw
    Adjusted
    Difference
    Apparently I adjusted the dimensions of the maps between this set and the last one. I'm not sure when I did that. I should probably hard code those at some point. Oh well. It's not like I'm getting paid for this.

    Next up will be 1900-1950. After that I think I'll do 1950-2000 and 1850-1900. That should give a general idea of the effect of BEST's adjustments. After that, I can maybe look at details/tidying up images.

  7. The maps start getting uglier as you go farther back and data coverage gets worse, but there are definitely some troubling patterns which match what I expected. The enormous drop in spatial resolution alone is enough to question whether BEST's adjustments are actually worthwhile (especially given that BEST encourages users look at data for locations as small as individual cities, which is clearly pointless given the lack of resolution). More troubling, however, is it seems clear patterns show up in the adjustments. Look at the United States in the maps for 1900-1950:

    Raw
    Adjusted
    Difference

    Months back I suggested the BEST adjustment process forces a particular warming trend onto its regression period (1900-2000), adjusting data outside that period to allow for such. I can discuss the details of why I predicted that later, but the key is I said BEST's adjustments would adjust the data outside to show more warming during the 1900-2000 period and less warming outside that period (assuming the area didn't already run hot). We already saw from the earlier maps the United States had data adjusted to warm more in the latter half of the 20th century. Now, in the 1900-1950 segment where there was already a warming trend across the United States (you can see that in both the Raw and Adjusted maps), no adjustment was made. That could be coincidence, but in the 1850-1900 period:

    Raw
    Adjusted
    Difference

    We see a significant portion of the United States was actually cooled. That means in the 1850-1900 period, a region was cooled, then in the 1900-1950 period when it showed warming, it was left untouched, then in the 1950-2000 period when it showed cooling, it was warmed. It could be coincidence, but that is exactly what I predicted would happen if BEST's adjustment process were biased due to it being little more than an overly complicated regression on an arbitrarily chosen time period.

    Ideally, what I'd do is create a new data file which was just a difference file of the two we currently have, then turn it into a movie. That would let us see how the data was altered by month rather than looking at how estimated trends change based on those adjustments. That'd be a lot of work though. I think what I'll try doing next is just examining individual areas in greater detail.

  8. I have to agree with Steve Mosher. The temperature trend after homogenization is the temperature trend of the region. This region seems to be larger for Berkeley Earth than for other homogenization methods, but in all cases homogenization aims to estimate the regional climate signal not the local one.

  9. Victor Venema, that doesn't agree with what Steven Mosher said. Or rather, that doesn't contradict what I said, which Mosher said was wrong. I'll repeat what I had said a year ago:

    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.

    Claiming that is only true because homogenization reduces the spatial resolution of the data set does nothing to rebut the point. All it does is add the qualifier, "At the scale the data can be resolved to" - a qualifier which ought to be implicit in all statements regarding the data set in the first place.

    But again, I'll point out BEST created a website which encourages people to look at its results on a relatively fine scale. Not only did it publish its results in 1x1 degree gridded files, it also created a website designed to allow users to select any location or even city to browse the temperature data for. By doing that, BEST portrayed its results to the public as having an incredibly fine resolution.

    Even if one thinks my statement was too broad because I failed to qualify it by explicitly stating the resolution in question (even though nobody else in these discussions ever does), worrying about that would be looking at the speck in my eye while ignoring the plank in BEST's.

  10. For one quick example of what I'm talking about:

    If you want to know what the temperature change has been in your city, your state, or even your country, you can now find this online at BerkeleyEarth.org" says Rohde. He adds, "We hope people will have a lot of fun interacting with the data." This feature should be available to the public by Monday, July 30.

    It's pretty cheeky for BEST to tell me I can look up "the temperature change" in my city on their website then turn around and criticize me when I refer to their results as being for individual locations rather than regions. And by cheeky, I mean... I shouldn't say more because I'm trying not to be snarky on this page.

  11. Yes, that quote at Science Daily is somewhat unfortunate, where a softening factor is that they are talking to the public. Better would be: "If you want to know what the temperature change has been in your region, you can search here by city. We also provide averages by state or country".

    As a variability freak, I read your "not a single location" as a claim about local temperatures, not about regional temperatures. Location is the smallest spatial scale.

  12. Somewhat unfortunate...? BEST has made the same sort of remark multiple times in multiple locations. It's explicitly stated it wants people to use its results for that purpose, actively encouraging people to use a tool it created to do so. And it's not like this is a matter of mixed signals - BEST has never said a single word to distance its product from that usage (unless you want to count the needlessly cryptic droppings of Steven Mosher like those on this page).

    If you really want to get down to it, BEST provided its results at a particular resolution. I discussed the results at that resolution. There's nothing wrong with that. If BEST doesn't feel its results can be resolved to the scale it provides them at, it should not provide them at that scale. It shouldn't complain when people discuss its results at the scale it provides them at.

    I guess one could argue "locations" could cover finer scales than 1x1 grid cells, but that just seems like being intentionally obtuse. If someone points to gridded data and refers to what it shows about locations, it's kind of obvious they're talking about what the map shows >.<

  13. Have you used the latest GISS numbers? There was a big change for the June numbers. Although it is in the change of ocean data, I wonder if they somehow smear that as well. They have now doubled the change from 1998 to 2014.

  14. Nope. I'm using the exact same data files I used last year. I only used GISS's land data last year though, so I doubt any change to the ocean data they use would affect things. They actually have two different gridded files up right now, one with the new ocean data and one with the old. I haven't downloaded either, but I suspect the land portion will be the same in both. GISS actually states how much they smooth (smear) their results (250km for land-only, 1200km when including ocean). I haven't checked it, but I suspect it's reasonably accurate. GISS definitely has much better spatial resolution than BEST does.

    By the way, I'm only using data up through the end of 2013. That's how far the BEST data went when I did the comparisons last year, so that's when I cut it off. I'm cutting it off at the same point now just to keep things consistent. The extra year or so of data shouldn't affect anything.

    Now if I had all the time in the world, I'd be look at the new ocean data too, but there are only so many hours in a day. I'm still looking to see how much difference there is in BEST's resolution on the horizontal as opposed to the vertical scale. I think there's a meaningful difference in how much BEST smears information depending on if it is north-south/east-west, but I can't quite pin it down.

  15. "Um, no. That’s exactly what it means, when we’re talking about BEST’s gridded results. Or BEST’s any results. As far as BEST’s results predict,* no location has cooled since 1960. That you can point out those predictions are wrong does nothing to defend BEST; it simply emphasizes the point I’m making – BEST’s predictions for individual locations do not match what people would expect or find useful.

    *Feel free to replace the word predict with project, interpolate or any other word that you feel fits better."

    Nope.

    He is what a grid value MEANS.

    A grid value represents the best estimate for ANY random location in that grid.
    It is does NOT mean that no actual location in that grid has no cooled.

    In operational terms it means that if you sample unsampled locations in that grid that the estimate that provides the smallest error
    if the grid value.

    So a grid value may be .1 per decade plus uncertainty. The uncertainty may of course include values that are negative.
    Actual locations may be negative. BUT the estimate that minimizes the error for all unsampled locations is positive.
    That is of course logically distinct and different from the statement that ALL LOCATIONS are positive or that NO LOCATIONS
    are negative. The grid value does not ENTAIL that all locations are the same value as the grid value, rather it says that IF you
    sample the unsampled locations you will find that the estimate that minimizes the error is postive.

    A simple example: I have 1000 stations. 999 have a value of .1C per decade. 1 has value of -.1C per decade
    The grid value will be positive. BUT that positive value does NOT entail or mean that NO LOCATION is negative

  16. Brandon you have some pretty fundamental mis understandings of what the various data products are and what they represent and what purposes they are intended for.

    Lets just start with gridded data. When a grid cell has a value of X.. that does not mean that every location in that cell has a value of X.

    That's just fundamental. Instead of temperature suppose I gridded the weight gain of people in illinois and the average weight gain was 10 lbs per decade. Would that average for the state entail that every person had gained 10 pounds and that NO person had lost weight?

    Nope. It means "go sample a bunch of people in Illinois, and your best estimate of weight gain will be 10 lbs."

    That is what a grid cell MEANS.

    It doesnt mean that there are NO PEOPLE who lost weight.

    In spatial stats the value of the cell is the prediction of unsampled locations. and it doesnt entail that every sample will have the value of the prediction. In fact the opposite. Its highly unlikely that ANY location will match the prediction perfectly. The same way if I tell you the average weight gain will be 10 pounds.. you might find very few people with EXACTLY 10 pound weight gain, but 10 pounds will be the estimate that minimizes the error of prediction.

  17. Brandon, GISS's numbers for 2013 have changed substantially, at the global level.
    Land only has cooled about .01C every year compared to the last update, so no big changes like the oceans.

    Where is GISS publishing both new and old ocean data?
    I would like to be able to keep comparing with the old numbers.

  18. Steven Mosher, your responses are, as ever, completely unhelpful. You write:

    A simple example: I have 1000 stations. 999 have a value of .1C per decade. 1 has value of -.1C per decade
    The grid value will be positive. BUT that positive value does NOT entail or mean that NO LOCATION is negative

    But this is based upon a completely uncharitable interpretation of my post. I provided data in a gridded format as a map which cannot possibly be examined in a finer resolution then and referred to "locations." You had two options: 1) You could assume the word "locations" referred to the gridcells I was showing; 2) You could assume the word "locations" referred to something the data I provided couldn't possibly show. You chose number 2.

    You chose to interpret my word choice in a way that made my post sound stupid instead of pointing out something simple and obvious. I don't know why. I clearly intended meaning 1. Regardless, I'll be explicitly clear. When I referred to BEST saying no locations had cooled, I was referring to the gridcells it estimates. I suspect the same would hold true if one examined the data in a non-gridded fashion, but that's not the format I have the data in.

    So when you then say:

    Brandon you have some pretty fundamental mis understandings of what the various data products are and what they represent and what purposes they are intended for.

    Lets just start with gridded data. When a grid cell has a value of X.. that does not mean that every location in that cell has a value of X.

    That’s just fundamental. Instead of temperature suppose I gridded the weight gain of people in illinois and the average weight gain was 10 lbs per decade. Would that average for the state entail that every person had gained 10 pounds and that NO person had lost weight?

    This is just insulting. I've distinguished between station data and data for areas multiple times in the past. You know I'm fully aware of the difference between station data and estimates for an area. You had absolutely no reason to assume my use of the word "locations" was intended to refer to specific temperature stations. You've written two lengthy (for you) comments to me which are nothing more than you taking advantage of the ability to read the word "location" in more than one way.

    Coming here and copping an attitude because you can play stupid semantic tricks rather than try to understand people is just pathetic. Is that really how BEST is going to handle its public relations?

  19. MikeN, GISS has a number of NetCDF files available on this page. That includes land-only and the different land+ocean data set it's used. That's gridded data though. I don't know that it's publishing the results you'd get if you took an average across the data set. You could always do it yourself though.

  20. This isn't the first time Mosher has purposely misinterpreted your words, if I felt he was worth the time I'd find the posts on Climate Etc.

  21. MIchael Schonewille, I don't know if it is intentional or not, but Steven Mosher does misinterpret me a lot. Often in ridiculous ways which let him mock me for completely idiotically untrue things. For instance, when BEST published some results showing, for the first time, some of what effect its adjustments had, one of the posts it did this with was at Judith Curry's place. The first image in the post had BEST's global results back to 1850 with and without adjustments. With adjustments was plotted in red, without adjustments was plotted in blue. During a discussion of the post, I said:

    This post says:

    As Figure 1 illustrates the effect of adjustments on the global time series are tiny in the period after 1900 and small in the period before 1900. Our approach has a legacy that goes back to the work of John Christy when he worked as state climatologist: Here he describes his technique.

    But if you look at just the blue and red lines, you can see a .2 to .3 degree difference in the earlier portions. That means there’s a 15-20% change introduced by these adjustments. And that’s for if we only go back to 1850. The BEST temperature record extends another hundred or so years back. We can’t see how much of a difference its adjustments make in that period.

    Pretty simple and straightforward. This is how Mosher responded:

    This is precisous

    “But if you look at just the blue and red lines, you can see a .2 to .3 degree difference in the earlier portions. ”

    Red is Africa
    Blue is the US

    when we talk about the adjustments being inconsequential GLOBALLY
    we mean the BLACK line

    The red line is Africa– 20% of the land
    the blue line is the US 5% of the land.

    So yes if you look at 5% of the data ( blue) you see a .2 to .3 degree difference

    the POINT of showing people continents and how they differ is so that people will AVOID the kind of mistake Brandon just made.

    GLOBALLY ( we are estimating the GLOBAL average) the adjustments are mousenuts..

    BUT because people can cherry pick ( the US) they can show BIG differences.. BUT they also IGNORE big differences in the other direction.

    It turns out Figure 4 in the post has results for individual areas plotted in different colors, with the United States plotted in blue and Africa plotted in red. So Mosher decided to act as though I was referring to it. Why? I don't know. Nobody with basic reading skills could possibly have come up with that interpretation.

    And let's not forget, Mosher frequently postures about how good his reading skills are. He often talks about how people can choose to interpret what others write in different ways, depending on how charitable they want to be. One would presume he would be able to understand when text explicitly refers to Figure 1, it is not referring to Figure 4.

    So maybe this is all intentional. Maybe Mosher is just a huge jerk who likes sabotaging honest attempts at discussions. Maybe BEST's public spokesperson is someone who actively tries to prevent people from understanding BEST. Or maybe Mosher is just a hypocritical idiot. I don't know. I can't say I care either way.

    As you point out, he feels it is appropriate to label people he disagrees with Holocaust deniers. If I didn't know enough about his character already, that would tell me plenty. For people who haven't seen it, here's a post showing what we're referring to. I'm still kind of annoyed about it. Not because of what Mosher said, but because Judith Curry let his comment stand yet moderated my response. Because apparently calling a person a Holocaust denier is okay, but saying:

    Wow. Steven Mosher compared me to a Holocaust denier.

    Is not.

  22. Location and spatial averaging scale are two different things. You can average an area around your house of 1, 10, 100 or 1000m. You can do the same for your neighbor. For the larger averaging scales the answer will be very similar, but that does not mean that it is not a permitted and often insightful computation.

    Kenneth Fritsch, you mean the benchmarking of homogenization methods in the International Surface Temperature Initiative? The statistical modeling of the homogeneous data is ready. We have agreed on how to introduce the inhomogeneities into that. This August we will start implementing those inhomogeneities.

  23. When Thomas Fuller's bet with Joe Romm was being discussed at the BlackBoard, I said that he could lose on adjustments alone. Mosher then criticized me for thinking adjustments are biased, and said I should analyze them. Since then there have been two major adjustments on top of the minor adjustments, both of which had the effect of increasing the trend(although the first actually helped Fuller a bit).
    I wonder if Mosher thinks the latest adjustment is valid.

  24. What introduces a distinct methodological bias in BEST's results is the reliance upon the bald assumption of spatial homogeneity of temperature variations over far larger distances than manifest in nature. Subsequent kriging to establish a continuous spatial field almost invariably winds up spreading UHI effects from urban data throughout the field. In many regions where virtually all stations are urban, the illusion is created of highly trending time series throughout the countryside. Lacking any thorough vetting to recognize UHI-corrupted records and removing them from the sampling scheme, the same problem is encountered in gridded treatments.

  25. Brandon,

    Could you scale to 0.01 the maps for 1990-2013?
    I mean one colour for every 0.01 degree.

    thanks

  26. Victor Venema:

    Location and spatial averaging scale are two different things. You can average an area around your house of 1, 10, 100 or 1000m. You can do the same for your neighbor. For the larger averaging scales the answer will be very similar, but that does not mean that it is not a permitted and often insightful computation.

    I have no idea why you say this. It's true, but nobody had suggested otherwise. I guess you can chime in to say random things that are true but have no connection to the discussion at hand, but... I don't get the point.

    Yes, there can be reasons to smooth data to all sorts of different scales. There can be reasons to look at any given point in the resulting smooth (which is what calculating the average centered on your house or your neighbor's house effectively is). That has nothing to do with whether or not BEST's results have very poor spatial resolution for a temperature record - which they do. It also had nothing to do with whether or not BEST's poor spatial resolution may introduce biases into its final results - which is quite possible. And finally, it has nothing to do with whether or not I'm right to suspect BEST's choice of period for its regression is introducing biases into its final results (some of which BEST sort of admitted when pressed by me).

    But if you want to talk about it anyway, go ahead. I don't mind people talking about random tangents.

  27. Armando, I can see about doing that. I think I know how to do it, but I haven't tried manually setting the levels for colors in these before. I've only set the range of levels. It'll be a few hours though. I don't have the data saved from your run, so I'll have to rerun the code, and I'm heading out for dinner in about an hour.

    By the way, I can use different colors if you prefer. I'm using default palettes in R, but it's fairly easy to substitute in other colors. I'd just need to know what range. That goes for anybody.

  28. I didn't redo the difference map because I obviously couldn't do it on the scale you asked for (since the differences in trends are far smaller than the trends themselves). Here are the other two maps though:

    Raw
    Adjusted

    As you can see, limiting the levels like you asked impacts the adjusted map noticeably more than the unadjusted map. That's because of the reduction in spatial resolution BEST's adjustments cause. Before, I had set the levels as high as possible to pick up as much variance as possible. That favored the slight changes in the adjusted data set. I felt that was appropriate because it painted BEST in the best light possible (please forgive the pun).

    That's particularly relevant because of the short time period you chose. We expect a lot more "noise" with shorter periods for our trends than longer ones, so that's another choice which may well make BEST's spatial resolution appear greater than it is.

    If you'd like anything else made, feel free to let me know. Right now I'm going to go back to trying to figure out how to get those numbers on the bottom and left side not to show up. I know one way to remove them, but when I do, it removes other stuff I like. I really do not like the filled.contour command in R.

  29. I wrote: "Location and spatial averaging scale are two different things."

    Brandon Shollenberger: "I have no idea why you say this."

    That was a response to what I interpreted as an objection that BEST has a tool that lets people see the temperature signal for the region around their city:

    Brandon Shollenberger: "But again, I’ll point out BEST created a website which encourages people to look at its results on a relatively fine scale. Not only did it publish its results in 1×1 degree gridded files, it also created a website designed to allow users to select any location or even city to browse the temperature data for. By doing that, BEST portrayed its results to the public as having an incredibly fine resolution. "

  30. Brandon,

    I made a mistake.
    Could you plot a map with only two colours: one for below 0,001 and one for above 0.001?
    It should show some regional differences more clearly.
    thanks

  31. Victor Venema, please don't pretend BEST is only offering a tool to let people see the temperature signal for a large around their city. Or rather, that that is how they are portraying it. I agree that is all they are doing, but they routinely describe it as being far more. I quoted Robert Rohde above; now I'll quote the BEST website:

    Did you know? Berkeley Earth gives you historical temperature data for your home town, state, and country. Check out our Results by Location page to learn more.

    That is the BEST website, explicitly saying you can get the temperature data for your home town from the tool which really just gives you results for a very large area centered around whatever location you pick. In other words, that is BEST explicitly telling its users something untrue.

    My criticism been quite clear. I have not criticized BEST for providing a tool which lets the user find results for a large area centered on the location of their choice. I have criticized BEST for having far worse spatial resolution than other groups (e.g. GISS), and I have criticized BEST for portraying their results as having far greater spatial resolution than they actually have.

    Your commentary on their tool does nothing to address any of my criticisms of BEST. That their tool could be defended for a use other than the one they advertise it for does not make their statements regarding it less false. Nor does it make the fact BEST has terrible spatial resolution less damning.

  32. Armando, I can do that. I doubt there will be much difference between the raw and adjusted results though.

    I'll try to get it done after lunch.

  33. Alright, that took a bit longer than I expected due to a weird bug. It took me a while to realize what was happening. I had written some code to handle the level settings automatically so it would be consistent when I made multiple maps for one period, and it worked fine before. I had never thought to try it with only two levels before, and it turns out one of the lines in it will end up dividing by zero if you only use two levels. That led to a variable not getting changed when I reran the code, keeping the value from a previous run and causing weird issues I couldn't identify at first.

    Anywho, I updated the code to handle unusual values, and here are the maps for 1990-2013, with only two colors. Red is warming, yellow is cooling. Different colors probably would have made more sense, but these were the default ones for my code, and you hadn't requested other ones. I can redo it with other colors if you'd like:

    Raw
    Adjusted

  34. Brandon,

    something went wrong I think
    a trend of 0.01/month is about 2.9 degrees difference in 24 years
    as you're at it you may use 4 colours 🙂

    I should learn this stuff myself (sorry)

  35. Oh. Apparently I set the breakpoint at 0 not 0.01. I can rerun it if you'd like. It won't be until tomorrow though. I have another project I'm working on in R right now, and I don't have the memory to spare to load the BEST data while I'm working on it.

    For what it's worth, this is actually really quite easy to do in R. I can provide some simple code that'd walk you through it.

  36. Brandon,

    The scale has to be from - 0.002 to + 0.002 with four colours please.
    When I have time I'll learn R.

  37. Brandon,

    Just a thought, since I know you try to be a stickler for accuracy.

    I checked the entire GHCN-M raw database for duplicated data during the same year across multiple stations within the same country. I found hundreds of occurrences of 7 or months in a year being duplicated. A couple of the many stations I found in GHCN with a decade of duplicate data I manually looked up in Berkeley and found identical data indicating it had been sourced to GHCN-M. (GHCN-M mostly reports only 1 decimal precision)

    When you have time you might consider doing similar station comparisons with Berkeley raw data. Probably take a couple hours to run through the 32000 stations. Age is slowly ossifying my brain, so I'm not really into taking on and coding for another database.

    I don't know if it makes much difference globally, but months ago I told GHCN about it and their new V3.3 data still has almost all the same duplications. They did condescend to correct one of the most egregious(10 years) duplications between two stations 100 miles apart and a 3000m difference in elevation.

    Had a rather unsatisfying discussion about duplicate data with Mosher a few months ago in this thread.
    http://judithcurry.com/2015/02/09/berkeley-earth-raw-versus-adjusted-temperature-data/#comment-673206

    Here are two station pairs in Berkeley with decade long duplication.
    Berkeley ID#: 151423, Berkeley ID#: 151425 1951-1960
    Berkeley ID#: 153938, Berkeley ID#: 158939 1961-1971

  38. Bob Koss, thanks for letting me know about that. I know I've heard about such duplication problems before, but I don't think I've ever looked into them at all. Unfortunately, the material I'm working with won't let me look into that. What I'm looking at is the gridded results you get when BEST finishes its calculations, both with and without having adjusting its data. Gridded results obviously won't let me see duplicated stations.

    I do, however, have the raw station data BEST uses so I can look for duplicated data. It would probably be something I'd want to do a bit later though. I'm a bit behind on doing stuff with this data set (I haven't forgotten you Armando). I want to get a file together with a difference map showing how much BEST's estimates for the temperature of the planet, laid out on the map, change due to their adjustments. And I want the file to show it for the entire record. That's not hard to do, other than being a huge memory sink, but then I want to make a video so people can see the changes visually. I'm hoping to get a first pass of it done by the end of the month, but that might be overly optimistic.

    But once I do that, I'll probably be done with looking at these files for a while, so looking at the raw data could be next.

  39. At your convenience, Brandon. It is just a suggestion. I don't work with Berkeley, just figured I'd give you a heads-up about the problem.

  40. Brandon,
    I made maps of GHCN v3 station trends (and ERSST) here. They are raw stations, linearly interpolated with shading, so of course there are lots of negatives. The US is particularly variable, which I think is partly the TOBS effect. Switching to adjusted makes it smoother.

    Re Bob Koss, I have found lots of junk in GHCN unadjusted - eg La Paz, June 2010, 86°C. But they are mostly flagged. I remove all with QC flags; it isn't a lot, and I haven't had problems with the rest.

  41. Thanks for the comment Nick Stokes. I've never been big on visuals, so that's a much better display than anything I'm likely to create.

    I have some other thoughts on the topic, but I have no motivation right now. I've stayed offline for most of the day because I'm still flabbergasted over the recent situation with Retraction Watch, and I just don't feel like doing much of anything.

  42. Armando, I promise, I haven't forgotten! I'll get that done this evening. Just to make sure, those numbers are monthly trends, right? I think my previous ones may have been yearly. I'd have to check. I've been using whatever R sets for its default when doing linear regressions, and if I remember right from the last time I was making these maps, it doesn't default to monthly for some reason.

  43. So Armando, I was messing around with this, and I realized there's a bit of a problem with your request. If I only use the range you request, there's a huge amount of white space because there are a lot of areas outside the ranges you specified. I did some spot checking to confirm, and it turns out there really are that many areas with that much warming/cooling. To show what I mean, I made maps with the the range twice as wide as you requested:

    Raw
    Adjusted

    I can reduce the range further, to as small as you had requested, but... it's not going to look pretty. I'm looking at it on my screen right now, and half of Europe, Asia and Africa are all missing. A large part of Canada too, as well as a bunch of other places. You can envision what it'd look like by imagining if two of the colors in the images above were erased. That's what you'd get.

    I think you might want to tweak your request.

  44. Brandon,

    It's o.k, but could you do these maps with 8 colours between -0.004 and 0.004?
    I'd like to see the change in steps of 0,001 degree.

  45. Armando, you're in luck. You caught me just as I was packing it in for the night (I got an alert just as I was about to turn off the computer). I still had the data from when I had run the code for the last maps loaded, so all it took to make the new ones was about 60 seconds. I think it probably took longer to get them saved and uploaded than it did to get them made. Anyway, here you go:

    Raw
    Adjusted

  46. No prob Armando. Let me know if there's anything else you'd like me to do. Things have calmed down for me now, so I should have more time to work on projects of my choice. There has actually been a bit of a story I've been working on behind the scenes on top of everything going on with Retraction Watch, but I think the most time consuming aspects of it are about wrapped up.

  47. Brandon

    You will remember my remarking that some 5 years ago my self and a colleague identified hundreds of counter cyclical stations showing a statistically meaningful cooling trend

    http://wattsupwiththat.com/2010/09/04/in-search-of-cooling-trends/

    This was just before BEST released their data and I forwarded the results to Richard Muller who confirmed that he agreed that one third of stations are cooling. In recent times Mosher then put on lots of caveats minimising any cooling.

    Perhaps I should ask Dr Muller again directly for his current position on the significance or not of any cooling? Statistically meaningful cooling seems to me to be a very important aspect of Global warming.

    tonyb

  48. tonyb, I think it would be interesting to get his position on any number of things given his public statements regarding BEST have had a tendency to be, shall we say, not accurate. But seeing as BEST is perfectly aware of his misstatements yet chooses not to correct any of them, I wouldn't count on getting a straight answer from him.

    I mean, seriously, if I hadn't pressured BEST, how long would it have taken them to release the results I discuss in this post? These aren't even their complete results sans adjustments, and yet, this partial release came months after BEST told everybody the full release had already been made? And it only came after they were pressured to? What kind of nonsense is that? If Hadley had done it, Mosher would have included it as an example of dishonesty in his book. Instead, he'll defend it even though it's BEST flat out lying.

    I still can't believe BEST has the audacity to go around telling everyone they're completely open and transparent while intentionally deceiving the public about what it has released. I'm sure Muller believed what he said when he said untrue things about BEST, but it's hardly a good sign leader of the project doesn't know basic details about his own project that he's publicly talking about. Even worse, when his mistakes were pointed out, BEST refused to correct them, leaving false claims on the public record. (And quite likely giving a completely bogus answer to the person who wrote the story.)

    But... bleh. I don't feel like dwelling on that right now.

Leave a Reply

Your email address will not be published. Required fields are marked *