Monday, January 5, 2015

Monckton and Goddard - O Lord!


Viscount Monckton of Brenchley has produced yet another in his series of the "Great Pause" - now 18 years 3 months. He uses only the troposphere average RSS - to quote Roy Spencer  on how RSS is differing from his UAH index:
"But, until the discrepancy is resolved to everyone’s satisfaction, those of you who REALLY REALLY need the global temperature record to show as little warming as possible might want to consider jumping ship, and switch from the UAH to RSS dataset."

Lord M heard. But in his latest post he is defensive about it. He says:
"But is the RSS satellite dataset “cherry-picked”? No. There are good reasons to consider it the best of the five principal global-temperature datasets."

There is an interesting disagreement there. Carl Mears, the man behind RSS, says
"A similar, but stronger case can be made using surface temperature datasets, which I consider to be more reliable than satellite datasets (they certainly agree with each other better than the various satellite datasets do!)."

You can see in this plot how much an outlier RSS is. The plot shows the trend from the date on the x-axis to present. You can see the blue RSS crossing the axis on the left, around 1996. That is Lord M's Pause. No other indices cross at all until UAH in 2008. In the earlier years, UAH often has the highest trend.



Anyway, Lord M cites in his defence "The indefatigable “Steven Goddard” demonstrated in the autumn of 2014 that the RSS dataset – at least as far as the Historical Climate Network is concerned – shows less warm bias than the GISS [3] or UAH [2] records."

He shows this graph:



No details on how HCN is done, but certainly there is no TOBS adjustment, which for USHCN is essential. That is the main problem, but the clearly wrong averaging contributes. In the past, Goddard has vigorously defended his rights as a citizen to just average all the raw data in each month (eschewing anything "fabricated"), and I'm sure that is what we see here.

So what is wrong with it? We saw the effects in the Goddard spike. The problem is that in each month, a different set of stations report. SG is averaging the raw temperatures, so what kind of stations are included can have big differences in average temp, without any actual change in temp. If a station in Florida drops out, the US average (SG-style) goes down. Nothing to do with the weather.

NZ Prime Minister Muldoon understood this. When the NZ economy hit a rough patch, he was scornful of locals leaving for Australia. But he took consolation. He said that this would improve the average IQ of both countries. It helped me - I can now figure out what he meant.

I wrote at some length about the Goddard spike issues here. But this example gives a simple case of the problem and an easy refutation of the method.

Every month, a different group of stations reports. Suppose we switch to a world in which temperatures do not change from year to year. Each reporting station reports the long term average for that month. So there is no change of actual weather (except seasonal). But the population of stations reporting varies in the same way as before.

For a fixed subset of stations, the average would be constant, as it should. But here it isn't. In fact, over time, the average goes down. That is because the stations dropping out (as they have, recently) tend to be warmer than most. I don't know why, but that is what the graph shows. It covers the period from 1979 to 2013, and shows the Goddard average raw in blue and the average of averages in red. It also shows the trends over this time, with slope on the legend in °C/century.



And that is the key. The cooling (in long term average) of the set of reporting stations induces a spurious cooling trend of 0.33°C/cen. That isn't large relative to the actual warming trend, but it makes a significant difference to the plots that Lord M showed. And it is simple error.


25 comments:

  1. Thanks for this, Nick. I was wondering what Christopher was talking about but couldn't bring myself to visit Steve Goddard's blog.

    I did think it funny when Christopher argued that all the other different data sets must be wrong because they showed a warming bias. (If you plot all RSS it has a warming bias too, and not all that different from the others - 0.122C a decade from 1979 to the present on a linear trend.)

    ReplyDelete
    Replies
    1. Hi Sou,
      Well, there's no enlightenment to be found at SG's blog. He's a man of few words when explaining what he does.

      It's silly of M of B to be calibrating with ConUS data. It isn't global, and it imports all the baggage of non-anomaly averaging, which is tricky to get right.

      Delete
  2. Just in case you were unaware, "Goddard" is now out of the closet. His name is Tony Heller.

    ReplyDelete
    Replies
    1. Thanks, Rob. I knew, but I generally respect pseudonyms. Or, more negatively, tag writers with their reputations.

      Delete
  3. "O Lord!" says it all. How Monckton would claim that picking the low-ball number isn't cherry picking, well it just makes my head hurt.

    I've suggested in the past that one way to handle the problem, when you aren't sure which series is more likely correct, just use the average of the series…

    Here's what I get for satellite only comparisons:

    Figure

    The earliest year for which the trend of the average of the two series is negative is April 2001.

    Note I think this is an absolutely terrible way of defining a "pause", but it's clear that many of the climate science critics seem to like this definition and appear hostile to attempt to revert to a more sensible definition and terminology.

    After repeatedly butting my head against Brandon, I've resolved to use the term "global warming slowdown" to describe the putative phenomenon. I've also suggested using significance testing to see whether it's even a measurable phenomenon, and various people have objected to that too.

    So at the moment, it seems a bit like "p*rnography", nobody can agree on a definition, but everybody still knows what it is when they see it.

    ReplyDelete
    Replies
    1. Pausography?
      I'll call the top plot the pausograph :)

      Delete
    2. "Note I think this is an absolutely terrible way of defining a "pause", but it's clear that many of the climate science critics seem to like this definition and appear hostile to attempt to revert to a more sensible definition and terminology."

      Predictions and declarations have been made by the IPCC. No further increase in CO2 emissions was to give 0.1°C per decade of warming. We increased our use of CO2 dramatically and its less than this. Its not about it certainly being 0, debunking global warming nor predictions of an ice age. Its about how much faith to put in predictions of 2°C plus by the end of the century.

      15 years was supposed to make it a significant trend. I think that 2001 is a more appropriate start year than 1998 and 15 years is only a year away. The trend in UAH so far is 0.02°C per decade. A sensible estimate of the error gives±0.06&deg per decade so its still only consistent with the lowest estimates of global warming, just. There should be no talk of more than 1°C increase by the end of the century unless something dramatic happens.

      That is the best that you'll get for a sceptical position. Everybody speaks for themselves.

      Delete
    3. Actually, the IPCC uses the term "hiatus" rather than pause . Here is their definition:

      In summary, the observed recent warming hiatus, defined as the reduction in GMST trend during 1998–2012 as compared to the trend during 1951–2012, is attributable in roughly equal measure to a cooling contribution from internal variability and a reduced trend in external forcing (expert judgment, medium confidence).

      I like 2001 as a better starting year, since 1998 includes a major episodic event. And I don't understand why they choose 1951, since we know there was a cooling period from 1951-1975, which the models once upon a time claimed was understood in terms of an increases in aerosols.

      So I'd suggest 1975-current against e.g., 2001-current, since 1975 is supposedly (as of IPCC AR4 anyway) the start of the period where human forcing is required to explain the magnitude of the observed warming.

      I think a better way to test is temperature acceleration", which is pretty easy to analyze including the uncertainty associated with internal variability, though it takes a bit to wrap your head around as an alternative.

      Delete
    4. I don't know if that is the correct way to estimate the acceleration. Could you integrate the results twice (adding a constant to the first from averaging the difference between months in consecutive years ie month n+12 - month n) to compare with the data.

      I like the results. Not because its what I want to see but it does show what's clear in the data. It has decelerated for a long period since fossil fuel use was supposed to have accelerated warming.

      Choosing 1951 to 2012 and the beginning of 1998 - 2012 is to reduce the difference. Now if the IPCC can cherry pick, why can't Lord Monkton and pick the middle of 1998?. Personally, I would reduce the 8 months of high warmth to what is given by a line of best fit from 1990 to 2000 (about 0.1 for RSS, I think) and show that it makes it even more obvious that the trend has changed around 2001-2003.

      Delete
  4. Anonymous: I don't know if that is the correct way to estimate the acceleration

    It's just the second derivative of a polynomial fit to temperature evaluated at the geometrical center of a window of data. It's a correct way and in a very technical sense it's an "optimal" method. I'm sure many improvements could still be made (e.g., using a tapered rather rectangular window of data). At the moment, I'm more interested in instrumental questions (how you measure a slowdown).

    Could you integrate the results twice (adding a constant to the first from averaging the difference between months in consecutive years ie month n+12 - month n) to compare with the data.

    That's not going to yield anything interesting without more work. The functional fit of the polynomial over a window of data, gives you a low-pass filtered version of the original time series. The second derivative throws away the offset and trend of the data. These serve as initial conditions that need to be enforced in reproducing the smoothed series. If you use the known values, you just get the original polynomial (before differentiation). If you don't, there are two constants you have to infer from the data. But there's no reason to do that, since by construction we know that this is the formal second derivative of that smoothed curve.

    Choosing 1951 to 2012 and the beginning of 1998 - 2012 is to reduce the difference. Now if the IPCC can cherry pick, why can't Lord Monkton and pick the middle of 1998?.

    Neither should do it, and both should be criticized for it. It was a mistake when the interval was selected in the IPCC AR5, and it's a mistake now.

    Personally, I would reduce the 8 months of high warmth to what is given by a line of best fit from 1990 to 2000 (about 0.1 for RSS, I think) and show that it makes it even more obvious that the trend has changed around 2001-2003.

    IMO, the question shouldn't be whether the trend has changed (it practically is always changing due to natural variability). Rather it should be "is the amount of change since e.g., 2001, inconsistent with the amount of change expected from historical variability in climate?"

    We can use Monte Carlo-based analysis of the effect of natural fluctuations to test that last question. On my list to play with, but I've a few papers I need to finish before I'm going to invest any more energy on this.

    ReplyDelete
    Replies
    1. I'm not going to find the time to look up the method (it could have been useful to me about 18 years ago). My point was to confirm for people like me that you do have a good estimate.

      I think I'll stop here. I've had a go at estimating the 2nd derivative of GISS LOTI by looking at the difference between the same months of consecutive years and then 15 year linear regression on that. Its still too noisy but comes out similar to your plot. Its just enough for me to doubt the work of Cahill (?) that CPA doesn't show a break point around 2000.

      http://s5.postimg.org/ec3gvxhpj/2nd_d_LOTI.jpg

      I don't have the skill or software to stay in this conversation (i only have basic Open Office) but let me know where you will post what you find.

      Delete
  5. Sorry, left out a phrase: These serve as initial conditions that need to be enforced in reproducing the smoothed series when you integrate the series twice.

    ReplyDelete
  6. By the way, the smoothing filter I described above goes under the name "Savitzky–Golay filter". In their original 1964 paper, in their Appendix I, they given an equivalent expression for the second derivative to the one I gave on Lucia's blog. You can find a nice discussion of this filter (as it is usually implemented as as digital filter) here.

    ReplyDelete
    Replies
    1. Carrick,
      I would do it this way (I probably will). You want a second derivative. OK, take the second difference. It's hopelessly noisy, so smooth by, say, convolution with a Gaussian. There is the usual compromise between smoothing and smearing.

      But then you can do summation by parts. It's the same (except for endoints). You're convolving the original data with the second derivative of Gaussian (Hermite polynomial). In fact, all the derivatives are related to Hermite summations.

      The relation to what you do is that it is linear. You are separating a smoothed part and differentiating that. This gets that derivative, with the smoothed residuals added. The idea is that the residuals should smooth to nothing. No need to separate.

      Delete
    2. Nick, did you get a chance to play with this?

      I've looked at it from a signal-processing perspective, and, because it's such a basic operation, there doesn't seem to be any real advantage to simple differencing the data before filtering.

      There's a substantial improvement using Savitzky-Golay over Gaussian for derivatives, because the error (for uncorrelated measurement noise) scales as $latex \sigma_n /N^(\mu + 1/2)$, where $latex \sigma_n$ is the measurement noise, $latex \mu$ is the order of the derivative and $latex N$ is the filter length.

      For simple convolution of Gaussian with first differencing, each order scales as $latex \sigma_n /\sqrt{N}$.

      I was able to confirm that the second-derivative filter that I described on Lucia's blog does much better than any simple filter you might obtain by differencing then filtering that.

      To me this isn't very surprising, because OLS polynomial models are globally optimal filters, where as differencing filters are neither optimal, plus by construction they act locally.

      Delete
    3. Hm..the latex didn't expand to the math expressions. Did I do something wrong or is it broken?

      Delete
    4. Noticed an error:

      Should have said "For simple convolution of Gaussian with successive differencing" rather than "first differencing".

      (Hopefully this was obvious anyway.)

      Delete
    5. Carrick,
      Yes, I've been doing a lot on it, and reaching similar conclusions, I think. Yes, it doesn't make much difference whether you just difference and smooth, or smooth and difference, or make a combined op. I did the latter (Savitsky style), because I could plot the spectrum, which for derivative rises and damps; so it has the character of a band-pass and this shows.

      I agree that the reason for using these operators is that ordinary trends leave in high frequency info, which is why we argue about charrypicking start dates etc. Since we urge using a long enough trend period to see climate effects, the smoothing should be consistent.

      I drafted a post - it was too long, so I'm splitting it into two. The first is just on trends as we know them - the good and the bad. The good is that they are, for a set of N data, the minimum variance estimate.

      I'm not sure about Latex - it shouldn't be broken, because it loads (and delays).

      Delete
  7. Nick, as long as the operations are both linear (and they usually are), the operations commute and you can perform the operations in either order. Of course you know this, just stating it for any lurkers.

    What I like about Savitzky–Golay is it gives you the function and its derivatives in one feel swoop: That is the smoothed value, the trend, the acceleration etc. So this can be a fairly time-efficent approach.

    The advantage of your approach is it is easier to customize the smoothing filter.

    I'll be interested in seeing what you come up with if you get a chance to play with it.

    ReplyDelete
  8. The good lord is infamous for switching temperature anomaly series in mid blog to select the outcome he is looking for. Consistency is not his hobgoblin

    ReplyDelete
    Replies
    1. It's not just the Noble Viscount. I see Tallbloke is using UAH to assert that 2014 was not the warmest year on record: http://tallbloke.wordpress.com/2015/01/09/uah-confirms-2014-was-not-hottest-year

      A true sceptic would have written "UAH indicates 2014 may not have been the hottest year across the lower troposphere".

      Delete
    2. Eli "Consistency is not his hobgoblin

      Eli, that's a really good line. Thanks.

      Delete
  9. Monckton mentioned that the UAH has a warming bias and will soon be adjusted. However, NOAA and NASA both use terrestrial data, so the global temperature derivation is determined by less than 30% of the earth surface. (Much less, actually, because there are no thermometers in large uninhabited regions, and in some cases one thermometer represents the temperature in the surrounding 1200km area.) But it gets worse, most of the thermometers are located within UHIs, so RADICAL adjustments must be made to the raw data. (In South America, several thermometer readings,each covering a large area, had raw data showing a 60 year cooling period, but the "revised" input used to determine global warming showed a 60 year warming period. Now that more than borders on poetic license.

    ReplyDelete
    Replies
    1. I don't think you know much about surface land/ocean indices. They all use an SST dataset to cover the ocean.

      Delete