As many know, I’ve been testing the central tendency of the IPCC projection from the AR4 to see if it’s consistent with observed temperatures since 2001. The question I ask is: Is 2 C/century consistent with measurements of the earth’s surface temperature? Since the beginning of the year, the answer has been “No, 2 C/century is inconsistent with the actual trend based on data since 2001.” The rejection is done to p= 95% confidence.”
However, Gavin has been suggesting that, the rejection doesn’t hold because the true variability of 8 year trends for the current period should be estimated from models, not based on the observations of real earth weather.
In comments, gavin suggested justified larger uncertainty intervals that I obtain using OLS and Cochrane-Orcutt fits to observed data this way:
However, for the case that is closer to what I did with the models, I calculated 9000 7-year AR(1) time-series with an underlying trend=2 C/century and with p=0.1. The distribution of the resulting OLS trends is N(2.0,2.2), but the range of the s was from 0.16 to 3.8 (mean was 1.7). The chances of getting within 25% of the s.d. in the trend distribution (i.e. an s between 1.65 and 2.75) is just under 50%. The chance of getting something less than half as big, is ~15%. Therefore there is a significant uncertainty in what the real s is given only one measurement of it.
Oddly enough, I agree that if the real earth “weather noise” were described by the models, or this particular AR(1) process that, evidently mimics the models, then we couldn’t say the 2C/century trend is falsified by a 0C/century 7 year trend.
Unfortunately for Gavin’s argument, that if is a very big if. I am simply not persuaded by any argument that suggests we should use results of any model of any sort if that result is contradicted by observations.
So, not withstanding Gavin’s 9,000 simulations based on his “closer” AR(1) process, if the variability of earth data we have contradicts his AR(1) process there is no reason at all to expect the uncertainty intervals for 7 year trends based on 9,000 (or even a billion) simulations will apply to the real honest to goodness physical earth.
To settle this issue (for myself, if not for Gavin) I felt the test whether Gavin’s AR(1) process is consistent with real earth observations.
We will see that Gavin’s closer AR(1) process, which evidently, reproduced the variability of “model weather noise” to some degree is found to be inconsistent with real earth weather. In other words, this process does not correctly reproduce the statistical properties of “real earth weather”. To the extent that it is similar to the model weather, the models also don’t reproduce “real earth weather”.
How can we test whether the AR(1) process is consistent with real earth weather?
If Gavin’s proposed AR(1) process describes real earth weather, then we should find it correctly predicts the statistical properties of earth weather. Specifically, the statistics for the real earth weather shouldn’t fall outside the range typical of the proposed AR(1) process.
So, if I generate statistics describing the behavior of Gavin’s AR(1) process, I can then test whether the recent earth data falls inside the range for Gavin’s process.
For today’s post, the statistical property I chose to test is the standard error in 7 year trends, sm, as estimated red noise corrected OLS fit to 89 monts of data. Given this choice, I test the “closer” AR(1) process as follows:
- Assume the AR(1) process Gavin says is “closer” to replicating model behavior describes the weather.
- Generate 12,000 time series with 89 months.
- For each of the 89 month series, compute the standard error in the 89 month trends, sm using the “red-noise” correction for OLS, and also compute the trend “m”.
- Find the distribution of all 12,000 sm resulting from the simulation.
- Determine if the value of sm determined based the available 89 month observation falls inside the ±95% confidence intervals for the distribution of the sm based on Gavin’s “closer” process
If the real observation does fall inside the ±95% confidence intervals for the distribution of sm for Gavin’s “closer” AR(1) process, then we will fail to reject his closer process as describing real weather noise. That is: we will accept that Gavin’s “closer” process may be true.
This would put my past falsification of 2C/century in doubt, as it could now fall inside the larger uncertainty intervals for the models.
However, if the single real earth observation falls outside the ±95% confidence intervals for Gavin’s “closer” process we will consider that “closer” process falsified. This is because, even though we only have a single 89 month observation, we find that if Gavin’s closer process is true then the probability of getting one this far away from the “mean” of his process is less than 5%.
Why choose the estimate for the uncertainty in the trend as a test variable?
Initially, I was going to test the data by comparing estimate of the lag-1 correlation, ρ or the residuals sT to the fit. These have more direct connection to the parameter used to generate the AR(1) process. For the AR(1) process to be supported by data, both of these must survive a hypothesis test not only individually, but in combination.
What I found doing early tests was for the set of parameters that results in variability of 7 year trends that match Gavin’s, AR(1) processes with ρ that were large-ish, were falsified based on data based on ρ by itself. In contrast, those with lower ρ falsified based on sy by itself. All falsified — but variability in the order means that as different value of lag 1 correlation (ρ) are proposed, to save time, I would either vary which analysis I showed, or I would sometimes need to explain two.
In contrast, the estimate for the variability of 89 month trends, sm, tends to falsify on the first pass for the few choices I picked, and so makes checking a bit quicker. ( Basically, I don’t need to show two steps!)
Preliminaries:
Before running the simulation, I needed identify the parameters for the monthly simulation, and also check that I reproduce Gavin’s distribution of the trends. I began by identifying his “closer” process.
During conversation in blog comments, Gavin described an AR(1) time series that is closer to his models. He described his series of annual average values this way:
However, for the case that is closer to what I did with the models, I calculated 9000 7-year AR(1) time-series with an underlying trend=2 degC/century and with p=0.1.
After I asked him, Gavin said the he used a standard deviation of σYEAR=0.1C for this year. (Note, gavin uses ‘p’ for the lag one residuals, I will use “ρ”, with a subscript of either “month” or “Year” to indicate the specific time period. I use “m” for the trend.)
In yesterday’s post, I translated this into monthly values, resulting in:
First, to test that I reproduce the variability in 84 month (7 year) trends Gavin reports in his comment, I ran 12,000 simulations of an 84 month process. I wasn’t sure what the precise usage was for σ, but sort of guessed how it must be used when describing generated AR(1) processes. I obtained:
- The averaged trend: m=1.99 C/century (which is pretty close to 2 C/century)
- The standard deviation in the trends: σm=±2.23 C/century. (This represents the standard error in the 7 year trends for Gavin’s “closer” process. As you see, this monthly process reproduces the ±2.2 C/century Gavin reported for his annual average process.)
- The average correlation in residuals: Ïmonth=0.663. (This is a bit low, as expected.)
- The average standard error in residuals to the fit, sT=0.16C.
Based on the statistics of each fit, I also estimated the standard error in the trend estimate based on OLS for each of the, ‘i’, 12,000 cases and corrected for red noise using:
Afterwards, I averaged over the corrected standard errors, sm,i, and obtained an average for the sample of sm = 0.0203 C, which is slightly smaller than σm=2.23 C, as expected. (BTW, using the known Ï, C-O obtains matching values for the average of sm and σm, but that’s not the topic of this post.)
Comparison of Distribution of sm,i to experimental values
After verifying my monthly process did reproduce Gavin’s variability in 7 year trends, I ran 14,000 simulations with 89 month trends, permitting comparison of observations from Jan 2001 through May 2008.
- The averaged trend: m=2.00 C/century (which is pretty close to 2 C/century)
- The standard deviation in the trends: σm=±2.01 C/century.
- The average correlation in residuals: Ïmonth=0.667.
- The average standard error in residuals to the fit, sT=0.16C.
Based on the statistics of each fit, I also estimated the standard error in the trend estimate based on OLS for each of the, ‘i’, 12,000 cases and corrected for red noise using:
The average over 14,000 samples sm=0.0188 C/year (1.88 C/century). The distribution is shown below:
click for larger.Figure 1: The distribution of the estimate for the variability trends, sm, computed using 89 monthly values.
After computing the distribution of the estimate for the variability trends, sm, I used the “histogram” function in EXCEL to determine the ±95% confidence intervals for this value. The upper and lower bound were found to be 0.011 C/year and 0.028 C/year respectively.
I compared the values based on the single 89 month observations, examining those from three different reporting agencies: GISS Land/Ocean, HadCrut and NOAA NCDC, and computed the value for a merge based on a simple average of the three measurements. All four cases fell below the lower ±95% confidence intervals. This means, that we must reject the hypothesis that Gavin’s “closer” model describes weather data at a confidence of 95%.
The results are show in the last column in table 1 below.
| Surface Based Data Set | ρ (α) | sT (α) | sm (α) |
| GISS | 0.493 (3.3%) | 0.121 (1.4%) | 0.0102 (1.7%) |
| HadCrut | 0.535 ( 6.92%) | 0.093 (0.01%) | 0.0084 (0.2%) |
| NOAA | 0.361 (0.09% ) | 0.106 (0.07%) | 0.0076 (0.02%) |
| Merged of GISS, HadCrut, NOAA | 0.474& (2.1%) | 0.102 (0.01%) | 0.0084 (0.2%) |
| The quantity “α” in brackets represent the proportion of simulated weather events that exhibited parameters below the values shown.& Gavin’s ‘closer’ process is rejected at p=95%, two-tailed, when α < 2.5%. It’s rejected one tailed when alpha < 5%. |
|||
Based on the merge of all three data sets, we find estimate for the variability in the 89 month trend was 0.0084 C/year (or 0.84 C/century). If Gavin’s closer AR(1) process described real earth GMST, we would expect to get a value this low or lower 0.2% of the time. So, using p=95%, we reject the hypothesis that Gavin’s closer process describes real earth weather.
The most favorable results for Gavin’s AR(1) process happen to be those from GISS Land/Ocean. For this case, we find estimate for the variability in the 89 month trend was ±0.0102 C/year. If Gavin’s AR(1) process were true, variabilities this low or lower would occur in 1.7% of observations for the earth.
The chosen cut-off was 5%–but two tailed. So, if the earth’s observation fell below the 2.5% or above the 97.5% bound for Gavin’s process, Gavin’s process must be rejected.
Based on GISS, we reject the hypothesis that Gavin’s closer AR(1) process describes “weather noise” for the current period.
Since GISS represented the most favorable case of the four, Gavin’s AR(1) process is rejected based on all four possible data sets shown in table 1. (I didn’t bother to check the satellite measurements.)
What about the lag 1 autocorrelation or the standard error?
Are you wondering whether we can reject specific magnitude of the lag-1 autocorrelation or variability for 1 month weather noise individually? It turns out:
- We get mixed results for rejecting the lag-1 autocorrelation for the AR(1) process. It’s rejected based on some of the data sets, but passes for others.
- We must reject the variability sT for monthly data based on each of the four data sets. If Gavin’s process did describe “weather noise” and the one month lag-1 autocorrelation was ρ=0.728, then we find the real earth weather noise is too small to be considered consistent with weather noise expected for the simulated process. We would reject Gavin’s model as requiring excessive variability for 1 month averaged data.
The falsification is more profound than I’ve made it sound.
Believe it or not, while the falsification is already proven, further analysis would show it to be even more profound. Recall that, since the models is supposed to be based on correct physics, and this AR(1) model is supposed to mimic the results of that model, the full claim is that the trend is m=2C/century and ρmonth=0.728 and &sigmamonth=0.175C. When I tested each of the value above, I found for example, that the values of parameters falsified individually. It is also possible to screen the data to find the probability that that the observation was 0 C/century and the estimate for ρmonth was less than 0.5 and the residuals to an OLS was less than 0.121 C. I didn’t actually do the test, but rest assured the probability this would occur is much, much less than 1.4%!
So, when compared to data, the hypothesis that the AR(1) process proposed by Gavin describes the data appears well and truly dubious.
Conclusions
Gavin suggested an AR(1) process as being closer to model results than the best fit AR(1) process based on 89 months data. If we assume Gavin’s AR(1) process describes “weather noise”, the current flat trends in GMST would not falsify to 95%. However, comparison of Gavin’s “closer” AR(1) process to data shows that particular AR(1) process is inconsistent with the properties of the “weather noise” experienced since 2001.
So, the hypothesis that, Gavin’s “closer” AR(1) process describes real weather variability must be rejected at p=95%.
Caveat: Gavin didn’t say this process describes the properties of model weather perfectly– just that it’s an AR(1) process that gives a closer fit to models than obtained using the best fit to weather data. obtained when we assume the data are AR(1) and fit it to data. However, the process Gavin described is the AR(1) process that replicates the variability of his eight year trends. Other processes either give higher variabilities or lower variabilities. So, if this AR(1) doesn’t describe the weather, we can say a) if the process is AR(1), then the models get the incorrect values for the time constant and the variability or b) the process is not AR(1).
Checking the second requires looking at more earth data and/or model data. I plan to look at both, but in the meantime, I’m sticking to the uncertainty intervals based on real earth data rather than those suggested by models whose weather noise appears entirely inconsistent with earth weather!
I cannot begin to express my total astonishment that apparently the Climate Change Community as represented by GISS/NASA employees and RealClimate can propose replacement of observed data with numbers from models.
No wonder their models/codes have yet to be validated; they have the processes accepted by the remainder of the science and engineering worlds backward.
Until the models/codes/applications are Validated, the numbers obtained from them will forever be suspect relative to representations of the Earth systems.
If the states of the climate system produced by the models/codes are not consistent, to some degree, with the states of the climate system during the times of interest, there might be no basis at all for attempting comparisons of data with model results. The use of a global solution-meta-functional tends to make these kinds of comparisons somewhat iffy under the very best of circumstances.
Lucia- This is a very informative analysis, and I urge you to condense and submit for publication (such as in GRL). We need these detailed quantitative analyses of the model predictions such as you have completed. My one comment is that one next step in this type of comparison between models and observations should be on land maxuimum and minumum temperature trends. We have found and reported on several warm biases in the observed data set (e.g. see – http://climatesci.colorado.edu/publications/pdf/R-321.pdf and http://climatesci.colorado.edu/publications/pdf/R-333.pdf), which should be considered in such an analysis (as well as for your current assessment).
Aside from the fact that 2C/century may or may not be occurring right now, it should be noted that such temperature change rate estimations cannot be constant. There’s imbalance right now, but as temperature progresses towards the balance temperature, the rate of temperature change would have to decrease. I assume this works in a way similar to Newton’s Law of Cooling, so the rate of change will be proportional to the difference between the equilibrium temperature and the actual temperature. The concentration of CO2 in the atmosphere increases in a way that is linear at best at the moment. Yet, the equilibrium temperature is proportional to the log of the CO2 concentration. Therefore it seems that the rate of temperature change would have to be either decreasing or start decreasing in short order.
If I made a wrong assumption in the argument, do let me know. I’m kind of new at this.
Joseph,
The models make all of these assumptions already. If they are not correct for the last 7 years, there is only random chance that they could be correct further into the future. They must not represent reality, some physical process must be either not accounted for, or others accounted for incorrectly, or a combination of these. If some single future model prediction is correct, it’s pretty similar to a broken watch being correct twice a day (12-hour clock).
Joseph,
the models faux physics, I believe, are currently based on CO2 warming driving the H20 which does the heavy lifting. They consider the current flat trend “internal variability.”
Lucia, you might want to complete this sentence: “When I tested each of the value above, I found for example, that the values of parameters individually.”
comment 4192 Joseph (2 comments.) July 15th, 2008 at 8:18 pm
The atmosphere, ocean, land, biological, chemical, and geo-bio-chemical sub-systems that make up the Climate System have never been in the past, and will never be in the future, in equilibrium. Steady states nor stationary states have never occurred and will never occur.
Attempts to assign some ill-defined Global Average Temperature of the atmosphere and oceans near the Earth’s surface to represent a physical realization of an equilibrium state are wrong.
This is not to say that changes to the boundaries of the subsystems and changes internal to some of the subsystems will not occur, but the physical phenomena and processes that govern these are functions of the local states of the subsystems and the states of the boundaries surrounding the subsystems. Not all of these changes will be important to anything.
Mother nature does not operate on the Global Average of any state variable. Changes in states work with local gradients of driving potentials. The enormous range of the heterogeneity of important local physical phenomena and processes presents significant difficulties relative to attempting development of meaningful averages.
Thanks Neils!
James H— I don’t think the models are useless. I think they have utility relative to no models. The difficulty is that “useful” doesn’t necessarily mean “accurate”.
Engineers use approximate methods frequently. Two circumstances that come to mind immediately are:
a) when one needs an answer quickly, and doesn’t have time to do something precisely and
b) when the tools to get a precise, accurate value aren’t available.
In the second case, we often draw from a range of approximate tools to figure out what might be the best path to take. When you do this, you need to remember initial analysis was heurisitic– that is it’s useful for guidance, but not necessarily accurate. Then, you take care to constantly check if things are on track based on observations of whatever it is you happen to be doing / predicting /watching.
Right now, the models are off track.
The other things is, looking at this, I’m suspect the reason modelers may think the models give appropriate magntidues of “weather noise”, is that past comparisons were made mostly during periods with large variability induced by volcanic eruptions. In that case, it would be difficult to detect a mis-match between a models ability to predict “weather noise” due purely to internal variability and “weather noise” that is induced by the dramatic variations in the rapidly changing extrenal forcing.
Dan
We can always define averages of any sort we like. We can also test whether the models happen to predict them. We can do this even if the particular averaged “thing” isn’t particularly meaningful or at steady state.
We do this all the time in engineering, so I don’t know why it’s not permitted for climate science.
The atmosphere, ocean, land, biological, chemical, and geo-bio-chemical sub-systems that make up the Climate System have never been in the past, and will never be in the future, in equilibrium. Steady states nor stationary states have never occurred and will never occur.
I can imagine that’s the case. But if we consider everything else being equal as a simplification, i.e. we only care about CO2 as the primary driver of climate at the moment, there is a theoretical equilibrium temperature. The number that I see often thrown around is 3C for a doubling of CO2.
At this sensitivity, the current equilibrium temperature is not that high relative to the actual temperature anomaly. It might be between 0.7C and 0.8C. The current temperature is probably 0.6C. The equilibrium temperature shouldn’t be increasing too rapidly, because of its log relationship to the CO2 concentration. If the actual temperature were 0.4C, I believe the temperature change rate would be roughly twice as fast.
Lucia,
I should have focused on the, ” … even if the particular averaged “thing†isn’t particularly meaningful … ” aspect.
And I think we don’t spend very much time on the approach if, ” … the particular averaged “thing†isn’t particularly meaningful … “.
Instead we go looking for better things that properly characterize the phenomena and processes of interest.
Dan–
As a practical matter, we test models using data exist. Thermometers have been deployed on the surface of the planet. So, that data is available. I’m not entirely sure why you think the surface measurement is entirely un-meaningful.
Are there inaccuracies in the measurements? Sure. All measurements of anything contain inaccuracies.
Does this particular measurement give the best possible estimate of the change in heat content of the planet? No. It would be better to have more information, including samples that permit us to integrate over some depth of the lower atmosphere and some depth of the ocean.
Nevertheless, I would expect the temperatures at the surface of the earth to give some information about whether or not the planet is warming or cooling. Also, if models– based on physics — have reasonable fidelity, they should get all these things correct, including the time evolution of the earth’s surface temperature.
This is true whether or not one believe the earth’s surface temperature is “meaningful” with regard to some particular question related to the physics of the planet.
On top of all that, the only consistent, concrete quantitative projections made by the IPCC relate to the temperature of the surface of the earth.
That’s why I have been examining this particular value.
I’m very interested in other data too. But, for blog purposes, I’m drawing from data others have process that are readily available to the public. I don’t have ARGOS data! etc.
RogerSr.
Thanks!
I think this need to be more complete to turn into any sort of paper. I’ve relied on Gavin’s description of the “closer” AR(1) process. For a paper, I think I need to:
1) Get the actual time series.
2) Find the “best” AR(1) processes, and see if they fit at all.
3) Repeat what I did here.
4) And possibly repeat for the relatively longer period with no volcanos in the 30s.
Maybe I’ll call to chat about what you think. 🙂
Lucia
Certainy! I would be glad to further encourage you to publish your research! 🙂
Lucia,
I didn’t mean to say that models have no use. They have a certain purpose, such as studying the effects of changing certain parameters on the system. But the result of this is that you only know what the effect was of the parameters that were varied, not of other parameters that may have unknowingly changed. This is where model validation is so important, to check that some physical process or property was not missed. Your results suggest that something has been missed.
The big problem here is that the models may be effective at understanding the effects of changing some things while keeping others constant, but nature is not keeping other things constant. The second problem is that there are probably many natural processes and responses that are not understood and therefor not accurately represented in the model.
All this would be fine if the results of the models weren’t used to:
1) try to predict temperature and weather events 100 years into the future
2) to form policies to curb CO2 emissions which could be very costly to the world, especially developing countries
My personal opinion is that the modeling should of course continue, and be refined over time as more understanding is realizedm but not be used for long-range predictions unless they demonstrate a high skill level comparing them to actual observations as you have.
This is great work that you are doing, I really have been interested in your blog postings!
Lucia:
Your analysis has been very helpful.
EXAGGERATIONS about model predictions or about ‘short’ anomalies in the observational record don’t help to set useful social policy.
GCMs are admitted to have AT LEAST the following residual weakness:
Their dynamical ocean sub-modules still poorly predict ENSO, AMO and PDO with respect to onset, intensity and duration. Therefore, at present, OTHER one-time or long-period perturbations of average mixing of surface and deep waters that might result in SMALL increase AVERAGE upwelling and result in drops in average SSTs, are not in the models. If such exist, theoreticians believe they probably are perturbations of the ‘chaotic’ hydrodynamic system, and will average out over the long term. Thus they believe the current 8-year-flat in global surface temperature is not to be taken too seriously – WITH RESPECT TO LONG TERM WARMING WITH BUSINESS AS USUAL. Maybe they’re right?
Lucia has made an excellent case that the current flat trend in ‘warming’ is incommensurable, at the 95% confidence level, with the projections of the current GCMs.
What does it mean?
The inexorable rise of the Keeling Curve at least assures us that although the Biome may be adapting somewhat to increasing CO2, such effect is so far marginal enough, that it is unlikely to spontaneously cancel global warming (otherwise the Keeling Curve would flatten). And the acidification of the oceans will continue.
Tipping point MAY be delayed by the current flat, perhaps giving us a little more time to change business, but business has to change! We MIGHT have enough time to use up all the (nonrenewable) oil, gas – and even coal. But there’s no doubt that their deposits are limited, and ‘we’ll’ have to switch to other sources before those supplies run out. So why not, for our descendants sake, switch as quickly AS POSSIBLE. And, maybe the tipping points are closer than even Jim Hansen thinks!
Note: Every so often, Lucia reminds us that ‘falsifying’ Gavin’s ‘predictions’ doesn’t mean warming has been falsified.
Len-
Warming itself is not falsified by any means. There is plenty of room between 2 C/century and 0 C/century!
The century long trend of warming is certainly right there in the data. It would take a huge cooling spell to cancel that. I don’t think it will happen.
And you’re right– I think it would be very wise to switch to energy sources that don’t generate GHG’s. I favor encouraging them all, and in particular nuclear to bring up baseload.
Len,
I’m not a stats pro, but the trend stated for AR(1) is 2C with a standard error of 2.2C. Doesn’t this mean that even the model thinks it is possible that there is no warming or even cooling? Do you think it is possible that warming is not being masked by some unknown short-term event, but that the CO2 increases aren’t really affecting the earth as much as (or in the same way as) the models were coded – or perhaps more dominant factors than CO2 are exerting their influence?
You state that business has to change because of non-renewable resources running out, and that we should just switch now for our descendant’s sake, supporting the argument only with the precautionary principle. I think that many folks may have an issue with paying much more for energy than we need to just to possibly avoid warming that may or may not happen, especially those that have a hard time making it with today’s energy prices or that want to be able to retire one day.
I agree that we need to continue developing alternatives, but we should not jam them down people’s throat unless they are viable. Artificially making conventional energy more expensive to advance alternatives is not acceptable. If we are going to mandate cap-and-trade or carbon taxes, we should be darn sure about the benefits they could provide vs their cost.
If the models have a variability of 0.84 degrees per century and are predicting a net of only 2 degrees change per century and if, in addition, available measurements may be off by a significant fraction of a degree and if climate history shows unpredictable significant swings over varied time periods for unknown reasons, why would we expect any kind of mathematical precision in these ranges and measure sets? Is there any point in ever making predictions of net changes smaller than 4-5 degrees?
What about ocean heat content? It’s been rather steady lately, so the question to be asked is do the models allow this much variability? Especially considering that no volcanic eruption could have caused this recent radiative balance. So, essentially, my question is how much internal variability in the global heat budget do the models allow for?
You got your teeth sunk deeply in his achilles tendon.
Carl–
That’s a good question. But, I don’t know how much variability the models permit for ocean heat content and I don’t have easy access to a data set describing this.
The ARGOS buoys are designed to give key data, but they were only deployed recently. I’m not aware of online data set of widely accepted data for OHC, so that limits my ability to test that as a “blog level” effort.
steven mosher, I’d say she had them sunk firmly in, but somewhat higher!
Nice analysis, lucia, not that I followed the details too closely, not my field.
Tony– Watch out. This blog is rated PG! After all, I was a Catholic schoolgirl once. 🙂
To someone like Gavin, it’s very possible that there is a deep-rooted confusion between “Real Earth” and “Model Earth”. That’s why he seems to assume that if he matches the noise properties of his model, then those are the noise properties of the “real” climate (ah! ah! involuntary pun!…)There is a fine sociological analysis of how modelers tend to confuse models and reality here: http://sciencepolicy.colorado.edu/admin/publication_files/resource-1891-2005.49.pdf
Lucia, this is getting better and better. I also think you should publish this. I know you can always improve and extend the analysis, but you have enough already to write a fine little paper. This is not so much about “falsifying the IPCC” as it is about how does one estimate the confidence intervals on observed and predicted climate trends. As Roger pointed out, there are just not enough of those analysis out there.
From the abstract of Domingues et al 2008 (http://www.nature.com/nature/journal/v453/n7198/abs/nature07080.html),
“On average, the decadal variability of the climate models with volcanic forcing now agrees approximately with the observations, but the modelled multi-decadal trends are smaller than observed.”
This study differs from Ishii et al (2006) and Levitus et al (2005).
http://atm-phys.nies.go.jp/~ism/pub/ProjD/doc/Ishii-Kimoto-2008.pdf
ftp://ftp.nodc.noaa.gov/pub/data.nodc/woa/PUBLICATIONS/grlheat05.pdf
So, we have three different ocean heat content datasets, and Domingues et al implies that large multidecadal observations (as seen in Domingues but moreso in Ishii and Levitus) are not predicted by models.
Lucia how many years of current trend would it take for you to say warming is falsified? Just a hypothetical, maybe an interesting exercise for your blog?
Tony,
worst dog bite I ever had was in the back of the knee. That’s high enough.
hurt like hell. never jog past an unleashed pit bull.
My mini pinscher scratched my eyeball the other night, begging for belly rubs.
but he is so cute I cant be mad.
The chocolate Lab tried to lick it all better. sweet girl, she.
The Liver Dalmation ran and hid.
Other than that I love my 3 dog nights.
Vincent,
I can do that mid-month. Of course, it depends on the variability, and looking forward I need to assume we will experience stratospheric volcano eruptions. So, the assumed variability would be longer. (In contrast, using existing data, the variability is whatever it was. We’ve had lull in stratospheric volcanic eruptions.)
Can you point to a synthesis or to poststhat you stand behind. Mosh pointed to your work as notable,but then when I read some of your stuff from several months ago, there seemed to be several flas that you later corrected. Is there some way to efficiently read your insights withouth having to chronologically read through all the posts and comments from March?
TCO–
What flaws have I corrected?
People ask if certain issues reverse the result of the hypothesis test. I look at those issues, often doing back of the envelop type calculations to see whether the order of magnitude of the effect could affect the results. The results of these investigations have, so far, indicated the methodology is sound, with the same caveats I’ve always metioned. People also ask me to look at things using different methods to see if that makes a difference.
I am, over time, trying to address the issues I’ve always admitted were there. (For example, in the first post on the hypothesis test, you will see I discuss the difficulties associated with the possibility there are large amounts of variability at low frequencies.) That issue is open.
I also update with new data each month– often showing things in a slightly different way. But the major results aren’t changing: 2C/century is falsified.
As for summaries: The name of this blog is “The Blackboard”. The results evolve. There is no “final” post. (In that regard, this is not particularly different from peer reviewed literature. People constantly publish something, and then refine the methodology. Of course, as this is a blog, you see more prolific postings, with smaller increments of refinement.)
Yes, you have to read chronologically. Or, just read the more recent stuff. Whichever you prefer.
It would be interesting to see it written in synthesis rather than the gore of chronological investigation. This would not prevent further experiments or new reports, but would be easier on the reader than meandering through the slate tiles of the blackboard chronicles.
TCO, then don’t and leave it to others. Lucia is taking her readers on a journey. She is not obligated to provide the casual passerby continuous Cliff Notes for the trip. You can wait until she feels she’s completed as much of the investigation as she wants and either writes or doesn’t write one or more summary papers. Or maybe Mosher can fill you in. And if you think you spotted a flaw, think twice before writing once to ask about it. She invariably answers intelligent questions, and sometimes even silly ones.