tag:blogger.com,1999:blog-7729093380675162051.post4311522487941067324..comments2024-03-28T13:56:47.604+11:00Comments on moyhu: Detecting periodicityNick Stokeshttp://www.blogger.com/profile/06377413236983002873noreply@blogger.comBlogger26125tag:blogger.com,1999:blog-7729093380675162051.post-48042758922483151342014-06-03T18:05:24.454+10:002014-06-03T18:05:24.454+10:00PS the distorted leakage is the lesser part of the...PS the distorted leakage is the lesser part of the problem. Any aliased trends will come from the part between 2mo (required filter cut-off) and 1mo (first zero in RM). <br /><br />This is a significant signal.<br />Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-56966997122357703492014-06-03T16:49:06.081+10:002014-06-03T16:49:06.081+10:00You are correct. The 12th harmonic can not slip ou...You are correct. The 12th harmonic can not slip out of phase with the ( albeit irregular ) 12 per year sampling. <br /><br /><br />I think his fig 4 and fig 8 which he labels as showing that the data IS aliased is a misinterpretation. It is simple that part of the spectrum below Nyquist is meaningless and should be ignored.<br /><br /><br />His figure 7 does show the leakage and distortion of the running mean filter below the sampling frequency. This WILL cause aliasing. Though it is relatively small.<br /><br />What he does not seem to say explicitly is that taking monthly means is identical to monthly re-sampling of the running mean, and as seen the running mean is not an adequate anti-alias filter. <br /><br />A subsequent point is that the leakage is in fact inverted as well. <br /><br />I have not tried to reproduce his histogram but he seems to have a point about the possibility of spurious decadal scale trends being created. <br /><br />His synthetic data seems to shows a magnitude that is significant. His description of his method seems reasonable. This should be reproduced and verified. <br />Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-56172508202780740832014-06-03T08:20:17.228+10:002014-06-03T08:20:17.228+10:00Greg, "fictitious"
I disagree; the annu...Greg, <i>"fictitious"</i><br />I disagree; the annual cycle is the real physical thing. Another real cycle could provide beat frequencies, but monthly can't; it is locked to 12/yr. If a beat freq is to be significant it has to be maintained, in phase, over several years. But even where months vary, after one year it's in sync. OK except for leap years, but that's a tiny effect.<br /><br />Why tiny? Leap years represent the difference between the daily and annual cycle, both real. 365 + 0.25. But if you're looking for beats there, you are starting at the 365th harmonic of annual.<br />Nick Stokeshttps://www.blogger.com/profile/06377413236983002873noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-72633787861552176922014-06-03T07:52:22.228+10:002014-06-03T07:52:22.228+10:00The "annual cycle" is a fictitious simpl...The "annual cycle" is a fictitious simplification, it is not a "cycle". <br /><br />There is roughly repetitive variability each year, which has significant energy at 12th and below. This is not a fixed cycle it changes from year to year. This implies the magnitude of this 12th changes each years. <br /><br />But when does it change?? The 1st Jan each year? No, it is constantly changing. You could recalculate the "annual cycle" each month (Jan-Jan; Feb-Feb.....) and get a different value for the 12th.<br /><br />Now, monthly sampling is an irregular mix of two sampling frequencies, each of which could cause aliasing. <br /><br />E&OE this gives 8.3y and 3.65 y<br /><br />3.7y is something that I come across very often in a wide range of data.<br /><br />Maybe this merits a closer look.<br /><br />Greg.<br /><br />Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-52081568459817945282014-06-03T06:45:07.484+10:002014-06-03T06:45:07.484+10:00Goodman, Your criticism is only precautionary. ...Goodman, Your criticism is only precautionary. Here is a very simple proof. If somehow the sun doubled the radiative forcing from what it currently is, I would EASILY be able to pick up that correlation in CSALT. However, it wouldn't be accurate because the resultant temperature is a nonlinear function of radiative forcing -- S-B law. Now consider that since the actual radiative forcing is incremental, I can always linearize the result as a Taylor series approximation. Add in a simple lag and voila, we have an estimate of the transient effect. We all realize that the fat tails of the equilibrium response are buried in the huge heat sink of the ocean, so we can safely ignore that for the time being. <br /><br />So the proof is that the limiting case makes sense and the incremental view for transient response makes sense.<br /><br />The physical view is that this is no different than a variational thermodynamics approach, which is well accepted. <br /><br />I do realize that you are creating a "just-so" story about how volcanic eruptions have long term effects on the climate but I ain't buying that. The effects of the volcano condense out within a few years. CSALT picks that up amazingly well.<br /><br />The fact that CSALT is a transient analysis should be pretty clear and it thus shows a lower-bound estimate to how much warming has occurred immediately due to the thermodynamic factors.<br />@whuthttps://www.blogger.com/profile/18297101284358849575noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-23673569345668011162014-06-01T20:36:15.153+10:002014-06-01T20:36:15.153+10:00Greg,
"There will not be a sharp "peak&q...Greg,<br /><i>"There will not be a sharp "peak" at 30.3 but there will be the 12th harmonic of the very strong annual cycle."</i><br /><br />No, that can't alias with the monthly cycle. The reason is that there are exactly 12 months in the year. So you can't have a beat frequency over years.<br />Nick Stokeshttps://www.blogger.com/profile/06377413236983002873noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-55840038759908625522014-06-01T18:17:19.604+10:002014-06-01T18:17:19.604+10:00Actually he refers to looking at individual "...Actually he refers to looking at individual "stations" so I think like so many others he seems to be confusing Had and CRUT. AFAICT he is looking at CRUTEM. Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-14600038819472913612014-06-01T18:06:21.460+10:002014-06-01T18:06:21.460+10:00Thanks NIck, it was Saumarez and I think this is t...Thanks NIck, it was Saumarez and I think this is the article (IIRC he did a similar one earlier). He makes some good points and some silly mistakes (like we all do). I believe no one till I've checked.<br /><br />I think he presents a good case that there is evidence of aliasing being present. Tough it would be better to look at hadSST and CRUT separately.<br /><br />There will not be a sharp "peak" at 30.3 but there will be the 12th harmonic of the very strong annual cycle. Looking at FT of daily arctic ice, for example, you can see all these higher harmonics quite clearly. <br /><br /><br />The problem is that climate science has yet to discover d.p. basics like the need for anti-alias filtering before resampling data.<br /><br />Greg.<br />Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-6490524152299329272014-06-01T09:14:47.021+10:002014-06-01T09:14:47.021+10:00Greg,
I looked again at Saumarez article. I though...Greg,<br />I looked again at Saumarez article. I thought then and think now that it was absurdly theoretical. I gave up on RS when he seemed to think that HADCRUT was used to drive a model. I queried with no response. But he's looking, in this context, at the possibility that there might be a sharp peak with period 30.3 days, which after 30-day averaging, would contribute to an apparent 10-year cycle. By sharp I mean that it stands apart from, say, 30 day period.<br /><br />Apart from the implausibility of that, it would be hopelessly confounded with the different month lengths.<br />Nick Stokeshttps://www.blogger.com/profile/06377413236983002873noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-20377291332845441712014-06-01T08:57:18.295+10:002014-06-01T08:57:18.295+10:00Greg,
I think you were thinking of this:
http://ju...Greg,<br />I think you were thinking of this:<br />http://judithcurry.com/2011/10/18/does-the-aliasing-beast-feed-the-uncertainty-monster/Nick Stokeshttps://www.blogger.com/profile/06377413236983002873noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-10049854681990058622014-06-01T08:54:11.941+10:002014-06-01T08:54:11.941+10:00Greg,
Well, 0.5 is the Nyquist frequency. It's...Greg,<br />Well, 0.5 is the Nyquist frequency. It's saying in general that you'd bring in (negatively) a part of that, except that those frequencies can't be properly represented on the sampled grid anyway.<br /><br />It may be significant when the filter is translated; then you do alias in some near-Nyquist frequencies as low freq. Again, I don't know how to detect that.<br />Nick Stokeshttps://www.blogger.com/profile/06377413236983002873noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-68405226788777870242014-06-01T04:50:08.827+10:002014-06-01T04:50:08.827+10:00Thanks Nick. What I meant was that in your first g...Thanks Nick. What I meant was that in your first graph above, the second peak at 0.5 is inverted. That presumably means that side-bands are deviations from an inverted signal not form an in-phase copy. <br /><br />I have a feeling the must be a way to detect that sort of pattern in the FT. <br /><br />I'm annoyed I cant find the article at Judith's site, it was quite good. <br />Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-39084746931158166012014-06-01T04:37:10.669+10:002014-06-01T04:37:10.669+10:00Webby, you seem to have some good maths knowledge ...Webby, you seem to have some good maths knowledge so I hope you can see the following point. <br /><br />A change in radiative forcing , whatever its origin, will produce an in-phase response that is a change in dT/dt, not a change in T. Flux is a power term not energy. That means that changes in F and T are orthogonal.<br /><br />Once the system has had time to settle, at least nearly, to equilibrium you can look for a final change of T (delta T) that matches the new level of radiative forcing. <br /><br />Now since F and T are orthogonal the correlation between them is ZERO. <br /><br />What any regression analysis seeks to do is to find the correlation between the variables. That is not possible if the two quantities are orthogonal. Indeed this is sometimes used in an attempt to separate the two responses. eg Forster & Gregory.<br /><br />Once surface temperature starts to change ( the integral of the induced dT/dt change ) there will be a radiative feedback. This will be in-phase with the instantaneous T. <br /><br />So if you get a finite result when regressing T and Rad terms what it reflects is the climate feedback responses NOT the long term temperature increase delta T induced by changes in the forcing. <br /><br />Do a multivariate regression and you have the same problem in spades. <br /><br /><br />Now not all your CSALT terms are rad forcing. ENSO is temperature but likely has non zero phase with global averages so you will not get the correct magnitude there either. Anyway, most are rad terms, so the whole exercise is doomed from the start. <br /><br />Yes, it sort of looks OK because you have lots of variables with about the right mix of frequencies but that does not mean it's physically meaningful.It's just like regressing a bunch of suitable AR1 series would also "explain" most of the temp. record. <br /><br />You are not alone, there are several published papers based on equally incorrect correlation analyses.<br /><br />Lagged regression is similarly flawed because it fits neither one nor the other and as soon as we depart form orthogonality with one or the other, the correlation gets diluted and the derived regression coeff is an under-estimation. <br /><br />Spencer & Braswell note the problem without finding a solution for the correct result.<br /><br />I attempt to derive a more correct correlation by anticipating a simple "relaxation the mean" response here:<br /><br />http://climategrog.wordpress.com/?attachment_id=884<br /><br />Perhaps you could adapt that to CSALT <br /><br />There is also the much overlooked problem of regression dilution due significant error and non linear variation in the x-variable. This means most cases of regressing one climate variable against another one will be fundamentally wrong also. Again under-estimating the correlation:<br /><br />http://climategrog.wordpress.com/2014/03/08/on-inappropriate-use-of-ols/<br /><br />Once you get into multivariate regression you have the problem in spades,again. The regression will be incorrectly fitting one variable and falsely attributing correlations to one or more others to minimise the overall residual. <br /><br />Any resemblance of the result to the physical reality is purely fortuitous. <br /><br /><br />===========<br /><br />Best regards, Greg. <br /><br />Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-87948499053901522552014-05-31T10:00:04.452+10:002014-05-31T10:00:04.452+10:00Sure Nick, and Willis was pleased to find that som...Sure Nick, and Willis was pleased to find that someone had apparently done (in the 1980's) exactly what he has been doing himself. I do not dispute that what Willis is doing could, at least in theory, be learned from various earlier publications (eg. Tamino's references to 1980's astronomical research). But come on Nick, Willis is working on his own, without funding, and more to the point, without free unlimited access to existing published work. I have plenty of my own experience 'discovering' something that was already (somewhere) known long before I developed it. That is OK. What is not OK is stuff like "My impression of all this is that WE (and its a general failure at WUWT, and similar) are doomed because they don't know any of the prior art." Doomed? Really? Doomed to what? I'll take Willis every day over a jaundiced and unpleasant individual who's only goal is to force his green tinted vision of the future on everyone else. I'll take Willis on a general intelligence test over 99% of his critics, including WC... and I'll make money for sure.Anonymoushttps://www.blogger.com/profile/08873551929362479719noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-53338054873373822792014-05-31T08:00:05.787+10:002014-05-31T08:00:05.787+10:00I probably didn't say clearly, but you can act...I probably didn't say clearly, but you can actually obliterate it your self - there is an option at the bottom of the window. And you can edit while you are writing.<br /><br />But yes, it's not a great writing environment. I always write in a local editor, then paste. I think Blogger may restrict pasting unless you log in with an identity, eg Google or Wordpress.Nick Stokeshttps://www.blogger.com/profile/06377413236983002873noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-21478129323664867352014-05-31T07:47:54.976+10:002014-05-31T07:47:54.976+10:00I think pointing out that Willis is doing Fourier ...I think pointing out that Willis is doing Fourier analysis with a limited knowledge of prior art is fair comment. It was made plentifully on the WUWT threads, and Willis generally acknowledges that. In fact, he was quite chuffed when Tamino identified his method as DCDFT. If you check the latest thread, I'm mainly defending him against accusations of imperfect Fourier decomposition. And earlier, trying to put what he's doing in context. He didn't see that as hostile.<br /><br />I'm actually very sympathetic to anyone who wants to work out their own mathematics, and I think Willis does that very well. I bristle when inadequate maths is used to attack what other scientists are doing.<br />Nick Stokeshttps://www.blogger.com/profile/06377413236983002873noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-40074236655042930432014-05-31T07:46:32.977+10:002014-05-31T07:46:32.977+10:00Nick,
Thanks for the obliteration of the multiple ...Nick,<br />Thanks for the obliteration of the multiple deleted comments. I will try to type much more carefully on your blog. <br /><br />I was in Turkey for several days until yesterday morning. 18 million people in Istanbul, and all wanting to increase their carbon footprint. Any approach to control fossil fuel use is going to have to address the desires of the billions who are now poor (or semi-poor), but who want to be rich.<br /><br />SteveAnonymoushttps://www.blogger.com/profile/08873551929362479719noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-2461143703824970372014-05-31T07:35:36.135+10:002014-05-31T07:35:36.135+10:00Stephen,
Edit functions are rare on blogs. I think...Stephen,<br />Edit functions are rare on blogs. I think Lucia's may be the only one. Blogger (this host) does allow you to obliterate your comment completely. I've done that with the ones you removed.Nick Stokeshttps://www.blogger.com/profile/06377413236983002873noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-4154145068324059772014-05-31T07:28:11.747+10:002014-05-31T07:28:11.747+10:00Nick,
You need an edit function. Really.Nick,<br />You need an edit function. Really.Anonymoushttps://www.blogger.com/profile/08873551929362479719noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-63651192276022333642014-05-31T07:27:08.071+10:002014-05-31T07:27:08.071+10:00William Connolley,
Is your critique of Willis so h...William Connolley,<br />Is your critique of Willis so hostile because you disagree with him politically or because what he is doing is incorrect? My guess is the former. Nick, to his credit, at least acknowledges that what Willis is doing is giving accurate results. Your approach toward all 'skeptics' is symptomatic of the poisoned pool of climate science; a pool in which you seem to so very comfortably swim. I do wonder if your goal is to advance understanding or to force arrival at your preferred political outcome. All evidence indicates the policy outcome is all that matters to you; sad.Anonymoushttps://www.blogger.com/profile/08873551929362479719noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-34123122354029058042014-05-30T00:48:42.447+10:002014-05-30T00:48:42.447+10:00The CSALT model that I developed uses a similar le...The CSALT model that I developed uses a similar least-squares spectral analysis method of estimating Fourier components. <br />http://contextearth.com/context_salt_model/<br /><br />I add an oscillating factor with frequency w as an unknown amplitude A*cos(wt) + B*sin(wt) to a multivariate regression analysis and include the A, B parameters to the model if the R2 error is reduced markedly.<br /><br />Of course this has the problem of over-fitting unless one is committed to restricting to only those sinusoidal factors that make physical sense. <br /><br />Wondering Willis became very upset with me when I pointed out how I applied his "discovery". What a nimrod.<br /><br />btw, the outcome of the model is that CSALT does pick up the TSI oscillations, but they are at the 0.05 C level in amplitude, which is well below the secular trend of 1C due o global warming.<br /> <br />@whuthttps://www.blogger.com/profile/18297101284358849575noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-67822256231971046782014-05-29T21:02:22.686+10:002014-05-29T21:02:22.686+10:00> his methods are adequate for the task
I only...> his methods are adequate for the task<br /><br />I only skimmed. I wouldn't be too surprised to find that they aren't; that they won't find more subtle signals. But its all so pointless; there is so much work on this topic. For exampe:<br /><br />* http://www.cawcr.gov.au/staff/jma/vanloon_solar.pdf<br />* http://depts.washington.edu/amath/old_website/research/articles/Tung/journals/coughlin-tung04strat.solar.pdf<br />* http://strat-www.met.fu-berlin.de/labitzke/signal/<br /><br />Well, you know how this stuff goes.William M. Connolleyhttps://www.blogger.com/profile/05836299130680534926noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-47203744456655724512014-05-29T19:10:57.219+10:002014-05-29T19:10:57.219+10:00Well, yes, but he's currently being sensibly s...Well, yes, but he's currently being sensibly sceptical about solar cycles etc, and his methods are adequate for the task. And well, I guess he would say that he's having fun.<br /><br />Actually, I was quite surprised to see Nick Lomb's name turn up on a periodogram. I've been aware of him for a long time as a fairly practical astronomer (and a very good curator). I was actually quite surprised that the prior art of the 70's was seen as new at the time.<br />Nick Stokeshttps://www.blogger.com/profile/06377413236983002873noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-4852706524140794032014-05-29T18:42:56.503+10:002014-05-29T18:42:56.503+10:00My impression of all this is that WE (and its a ge...My impression of all this is that WE (and its a general failure at WUWT, and similar) are doomed because they don't know any of the prior art.William M. Connolleyhttps://www.blogger.com/profile/05836299130680534926noreply@blogger.comtag:blogger.com,1999:blog-7729093380675162051.post-2759760487959952962014-05-29T18:00:42.884+10:002014-05-29T18:00:42.884+10:00Greg,
The denominator sinc doesn't affect any...Greg, <br />The denominator sinc doesn't affect anything much until toward Nyquist. So it's basically just the sign of the sinc numerator. I've done the arithmetic signed, plotted as abs. I'll see if I can add a good plot.<br /><br />As for testing aliasing, I don't know how to do that.<br />Nick Stokeshttps://www.blogger.com/profile/06377413236983002873noreply@blogger.com