Wednesday, April 14, 2010

Can't you see the acceleration ?

The current rate of global sea-level rise is 2-3 mm/year. To hit 1 m of sea-level rise in 2100 this rate should  have to accelerate at least to 10 mm/year or faster at some point in th 21st century. While we await this to happen we may have a look at the un-smoothed monthly global sea-level data from satellite measurements. But before watching the data plot, think for a few seconds what you are expecting to see..

 Global monthly mean sea-level from Topex-Poseidon-Jason, corrected for barometric effect and isostatic adjustment

I must confess that at first sight I was a bit surprised, and maybe you as well. I had seen the multi-yearly smoothed data often, and I did not expect that by eyeballing the acceleration in this period seems to be negative. A linear fit to a parabola yields a negative value of the quadratic term. The bootstrap confidence interval (5%-95%) of the linear fit suggest a possible development for the global sea-level in the next 10 years. Or perhaps the up-tick in the past 4 years will continue. 
Do not take this too seriously. It is just a pastime while we wait.

The lines represent the bootstrap 5%-95% confidence interval of the mean response (no including residuals)  . The calculation of the confidence intervals takes into account the full serial correlation structure of the regression residuals.

For those of you who may be wondering, this fit implies a 95% probability for 35 cm or less of sea-level rise in 2100.


P Gosselin said...

As I've written before, I have yet to find a single scientist who is willing to bet on 6mm/yr for the next 10 years - let alone 10mm or 16mm or 20mm! Not even Stefan Rahmstrof, who has already bet "big money" on temperature rise for the next 10 years.

Many of these projections are made I think because it attracts a lot of publicity and attention.

Donna Laframboise has just announced that she'll be releasing the results of her comprehensive audit of the IPCC 2007 AR4 references today! (I know the results already, but I'll let Donna have the fun of releasing them).

P Gosselin said...

Somewhere I recall reading that Hans von Storch gave a range of 40-80 cm, if I'm not mistaken. That should allow him to bet another bottle of wine.

ghost said...

hm, 2100 is maybe a bit late for deciding the bet...

one remark: there are interesting papers about the recent sea level... at least for me as lay person: eg.

interesting is to know, what slowed the rise. In the linked paper showed the thermal expansion or its reduction could be the main reason. Seeing La Nina/El Nino as well as the temperature development of the last years, this could explain it... not sure.

Another figure in this paper was also nice and some fun: showing the MSL + R07 projection + AR4 projection. MSL is on the lower end of R07 and on the upper end of the AR4 projection.

PS: did you imply, smoothing was "misleading" visually or worse: smoothed data may caused false trends in computations?

Marco said...

Hi P Gosselin: You did some great work. Too bad you labeled all books as "not peer reviewed". Wrong!

Back to the drawing board!

isaacschumann said...

I came across this:

'”Instead of using temperature to calculate the rise in sea levels, we have used the radiation balance on Earth – taking into account both the warming effect of greenhouse gasses and the cooling effect from the sulfur clouds of large volcanic eruptions, which block radiation”, explains Aslak Grinsted, PhD'

I was only able to see the abstract, but I would be interested to hear your commentary on this when you get a chance to read it.

It predicts at least roughly 70cm of sea level rise with aggressive mitigation, 1.2m without. These claims seem too deterministic to me, but then again, I'm no expert. How does this analysis compare to the temperature to sea level model discussed in your last post?

P Gosselin said...

So much for the Bible of AGW:

Marcel Severijnen said...

The rate of change of the sealevel is an interesting phenomenon. Simon Holgate from Proudman Oceanographic Laboratory wrote about it in GEOPHYSICAL RESEARCH LETTERS (doi:10.1029/2006GL028492) in using a selection of stations worldwide to follow the rates of change. I took the opportunity to mirror sealevel rate change at one Dutch station (Den Helder, annual mean sealevel rise 1,46 mm) with one of Holgate's selection, UK station Newly in a historical reconstruction from 1916. Complete with rate changes of Jason satellite measurements, which fit quite well in the picture.
In Dutch, but that is easily understood in Hamburg......

Marco said...

@P Gosselin:

Is there a grade lower than F? It would be the grade you guys and gals would get for that shoddy piece of work. The very first part I checked (1a) not only completely ignored books and agency reports (often these ARE peer reviewed), it even does not label IPCC reports as peer-reviewed. A howler of a mistake!

eduardo said...

@ 3


thank you for the link. Indeed, sea-level is a very complex and interesting topic.
Another paper worth reading is the one by Walter Munk, A sea-level enigma, although some of the discrepancies outlined there have been in the meantime lessened.

What I learn is that when one talks about accelerating sea-level, one compares the present rate with the one at the beginning of the 20th century, for when only a few records are available. From 1950 onwards it is very difficult to detect an increase in the rate of change

'PS: did you imply, smoothing was "misleading" visually or worse: smoothed data may caused false trends in computations?'

Of course not! climate scientist always present the data in the most objective way possible.

P Gosselin said...


The process has to be rigorous and it simply cannot count everything under the sun with a copyright symbol as "peer-reviewed".

Even peer-reviewed is not a guarantee in any way. The process has been corrupted more than once, and "science" like the Hockey Stick have been stamped as settled science. Peer-reviwed, and flawed.

If anything should come out of this debate, it is that the peer review process and criteria have to be more clearly defined and adhered to. The IPCC wnet over the line in selling the report to the public. Look at the quotes:

They exaggerated and were less than truthful. Now the price has to be paid.

Marco said...

In short, P Gosselin, you created a strawman: you made up a new definition of "peer reviewed" without telling anyone, then claimed a lot was not "peer reviewed" (but only based on that new definition that was not described), and when caught start complaining that peer review is not even a guarantee. Oh, and that's actually what scientists are saying anyway. Nonetheless, peer review is peer review, and by definition also includes book chapters and reports and whatmore that have gone through a formal review by an expert and judged on its merits before publication.

Oh, and I guess the price you believe the IPCC should pay is people like you and your fellow 'auditors' making things up. Great. Attacking what you believe to be untrue by being less than truthful.

Rob Maris said...

Marco: I'd suggest you to ignore P. Gosselin's post. This helps lowering stress levels. I already ignore his posts since two weeks, and as long as his posts are not too frequent, this works OK for me.

Note: it is said that austrian people know a level beyond ignoring: "I don't even ignore him!".

P Gosselin said...

Jo Nova:
"Every time the IPCC have spat on a scientist with “that’s not peer reviewed”, they have set themselves up to look like duplicitous fools when caught relying on student theses, magazine articles, and boot cleaning guides."

Boot cleaning guide?

I couldn't have said that better.
The IPCC is a victim of its own arrogance. They've tried ignoring us for years, and now it's catching up to them.

Rob Maris said...

eduardo: In Werners "Chill-out" posting (april 4th) you noted sea-level that "It attained a maximum at the so called Midholocene High Stand, about 5000 years ago 2 meters higher than today, and has been declining, at these long time scales, since then.".

Can you provide some links to up-to-date sea-level information that covers this time span?

ghost said...


thank you very much... a bit sarcastic... hm. ;)

hm, what I always find really frustrating in "your" research area is the combination of missing "perfect" data from the past and the long timescales that are needed for validating results. of course, there are really great, very impressive works that try to improve this situation. But still...

I mean, for the last 17 years one can see a slight decrease in the rise rate (several times are flat periods), but compared to the beginning of 20th there is an acceleration (probably). What is the noise, what is the signal? Is both noise? I assume, that is one problem of semi-empirical models (what about the other, not VR09/R07 approaches, they show similar results?), while the other (physical) projections do not know all processes or processes in detail. Despite the progress, still a bit frustrating for a lay person like me.

Actually, for me, 50cm look bad enough and the rise won't stop in 2100, I think 1m or 2m rise or more will also cause problems in the 22nd or 23rd century, I assume... well, except the Vulcans are coming.

Marco said...

@Rob Maris,

Don't worry, I'm not stressed that much. But I think someone needs to point out, for those less aware of the facts, that there are other 'opinions'.

_Flin_ said...

Please correct me if I am wrong:
Isn't the key issue concerning sea levels the stability of the big ice shelfs (Greenland, WAIS, EAIS)?

And isn't our knowledge concerning these rather limited?

So if the shelfs stay stable, the whole sea level issue will be existent, but non-catastrophic, because we can easily adapt to a meter higher sea (at least those of us not living on the Maledives).
But if the shelfs do not stay stable, our whole projections are going down the drain.

Currently we think that the ice shelfs are not an issue. But recent measurements can be interpreted that we might have underestimated a few things.

Or did I severly misunderstand something here?

eduardo said...

@ 14


Depending on your background these references could be useful

eduardo said...

@ 15


yes, it is sometimes frustrating, because the conclusions are often soft, based on assumptions that are difficult to check. In some sense it is even worse than astrophysics. In astrophysics they cannot do experiments but they can observe many stars or galaxies. We have just one object, long time scales and many confounding factors. This is the main reason for all these endless debates. There are mostly no debates in numerical weather prediction, as they can test their results everyday.
However, when you gloss over the political implications, it is very attractive science because you are forced to think from many different angles. The discussion about sea-level variations and its possible causes is a nice example.

eduardo said...

@ 17

I would say you are mostly right. However 'recent measurements' are exactly that, recent, and lack the long-term perspective. Low and Gregory mention this in their commentary. Measurements of outet glacier velocities in Greenland are very variable showing some years with high velocities followed by stand-still. It is difficult then to estimate a long-term mean because the measurements cover barely 10 years or so and only some locations.

Anonymous said...

This is a nice academic debate but in practice, we need better downscaled models and more local studies. Local sea-level rise can vary a lot from global average and engineers need to have numbers for their area.

I work with a city engineer in a coastal town where the dikes are already close to their limit. It is a highly controversial topic, therefore, I chose the anonymous option. On top of global sea-level rise, the engineer has to take into account:
- geological processes
- tides
- low-frequency climate cycles
- episodic storm events such as storm surges
- surface waves

Even if global sea level rise might be limited to 35cm in 100years (what I strongly doubt), it might still lead to a dike breach in combination with the other factors. For example, it has been suggested that climate change will also lead to a higher frequency of storm events.

Most areas of that town are under sea level already and a dike breach could have severe consequences. Therefore, the engineer argues that we need adaptation measures now. Problem is: Local residents are confused by the climate change debate and don`t believe in sea-level rise anymore - therefore, they object all adaptation measures. In particular, they don`t want higher dikes blocking their views and they don`t want to participate in the costs. Local politicians now stopped work on the dikes because they don`t want to upset the public.

Major concern of the city is responsibility and liability in case of a dike breach. Insurance companies pulled out long ago - they won`t pay anybody... In another municipality that was actually hit by a small flood, residents sued the city. Who is liable? The city? The scientist?

This is the practical dilemma beyond the academic debate. And the stakes are high - a flood or dike breach can ruin a community and even lifes may be at risk.

Rob Maris said...

eduardo: thanks for your pointers. I have read the abstracts, found 1 interesting quote "Early Holocene eustatic rise in sea level and late Holocene hydro-isostatic decline in sea level combined to produce a regionally variable mid-Holocene highstand in tropical Pacific sea level that stood 1.0–2.6 m above modern sea level.",
which coincides with that what you said.

However, when I look at other places (via google search), e.g.
- danish location research yield monotonous rise figure.
- A readable article for an engineer like me makes the impression to present scientific data.

Regarding my "Wattenmeer"-Buch reading: there it is suggested that the Waddensea area has grown to its today's "shape" thanks to relative steady sea level rise.

Footnote: I'd consider eustatic long term sea-level change over the past 2000 years as helpful to better interpret what is going on the past 100 years.

eduardo said...

@ 21

Anonymous , would you like to pen a longer blog post on this particular case? I think the interaction between climate projections and local public policies would be of interest for many here

P Gosselin said...

This is interesting:

Anonymous said...

@ 23 eduardo

I really would like to but this particular case is highly contentious at the moment. Until summer, a new zoning plan is debated with the participation of local citizens and I don`t want to get my engineering colleague into trouble.

I could either write a longer report on this case later, when the plan process is finished. Or I could write about another planning and climate change case where I was involved and which was less contentious. However, that one took place in the interior and sea-level rise didn`t play a role.

P Gosselin said...

Last week some hecklers in the blogospehre saw fit to try to trash the excellent results of Donna Laframboise's Citizen's Audit.
Well, don't take my word, read Richard Tol's comments here.
Perhaps his comments ought to be posted here at Klimazwiebel.

Anonymous said...

@Eduardo Zorita

could you do an update of your post? Maybe, you could also discuss the newest developments in the "sea level" part of the climate science and in the politics.


Bam said...

Ghost, my prediction with the newest data (that would have to be the AVISO data, which for some reason is more up to date than the more commonly used UC dataset) will give a 95% interval of _at least_ 30 cm of sea level rise, and a possibility of (wild guessing at the moment) 60 cm.

Anonymous said...

Thanks, Bam. I am wondering if Zorita will answer and update his post in a honest way.

Anyway, today Jason-3 will start. Let us hope that everything will work fine.


eduardo said...

Ghost, Bam,

using monthly data until October 2015, the central estimation of the acceleration (=the quadratic term in the fit) turns positive, but the 5%-95% range still includes comfortably zero acceleration.
Translating this 5%-95% range of the acceleration into implied sea-level rise by 2100 relative to 2000, I get a 5%-95% range of 25.2 - 48.9 cm .

Anonymous said...


not enough. You claimed that smoothed data is a misleading, fraudulent data presentation for the public. So, show us the smoothed data, the monthly data, and explain it to us. Details please.


Anonymous said...

Mir scheint, hier werden alte Rechnungen beglichen. Einen Peak nach unten zum Anlass zu nehmen die Frage zu stellen, ob sich der Meeresspiegelanstieg beschleunigt, hat auch wenig wissenschaftlichen Wert, das war eine reine Polemik.

Wäre ich an einer ernsthaften Diskussion dieser Frage interessiert, würde ich die Daten zur Massenbilanz des grönländischen und antarktischen Eisschildes anschauen. Die Entwicklung, die da erkennbar ist, wird sich meiner Meinung nach zwangsläufig in einer Beschleunigung des Meeresspiegelanstiegs niederschlagen.

Wann diese Beschleunigung in den Meeresspiegeldaten signifikant nachweisbar sein wird, das entzieht sich meiner Kenntnis. Eduardos Artikel klingt so, als hätte es dazu Vorhersagen ergeben. Ich kenne keine.

Sollten wir es nicht dabei belassen? Ich kann mir nicht vorstellen, dass Eduardo heute glücklich ist über seinen Beitrag von 2011, nachkarten bringt nichts.


eduardo said...


I cannot see why I should feel sorry of that post. I used that the data available until that moment and calculated the acceleration (according to one possible definition) and the confidence intervals. It turned to be negative. Now, with more data, it turns to be positive. It may trun negative again, because the dynamics of polar ice-sheets is very uncertain. Antarctica may gain ice, Greenland ay lose, with further warming and increased precipitation. Now, can you explain me what is morally wrong with those calculations ?
The acceleration of sea-level is very difficult to demonstrate, and many scientist still think sea-level rise is not accelerating - whereas others do think so. One reason is that there are no homogenous record of global sea-level over the whole 20th century. Other reason is that signal to noise ratio is very small. Even taking the simulated global sea-level until 2100, one cannot detect an acceleration in scenarios other than RCP85, because the rate of sea-level rises very slowly. The acceleration caused in this scenario is mostly due to the (quite uncertain) estimation of land-ice melting and very little to the thermosteric contribution.

I would have liked to answer other comments here if they had been politely formulated, but honestly I do not feel any need to engage in any discussion with fundamentalists. The data are available here, and I am sure that most of our readers are intelligent enough to download the data themselves, produce any plots they like and calculate the acceleration by the method of their choosing.
But perhaps I am wrong.

Anonymous said...

Unmoralisch ist eine Kategorie, deren ich mich nicht bedienen würde. Für mich klingt ihr Beitrag von 2010 so, als hätten Sie sich damals über etwas geärgert und ihrem Ärger mit einer Provokation freien Lauf gelassen. Und die beiden Vorkommentatoren sind nicht "fundamentalistisch", die scheinen sich damals nur mächtig geärgert zu haben über Sie.

Sie sehen ja selbst mit den Daten von heute, dass ihre Methode nicht robust war (was damals schon klar war), und das ist nichts, wo man als Wissenschaftler mit Stolz zurückblickt. Im Grunde haben Sie etwas getan, was ich als "den Lomborg machen" bezeichne:
Nämlich Fragen gerade zu dem Zeitpunkt aufwerfen, wo die Antwort einem gefällt. Und dann noch Datenextrapolation bis 2100, wo doch die Antwort auf die Frage von der zukünftigen Stabilität der Eisschilde abhängt, und nicht von den Daten vergangener Jahrzehnte.

Das sind Dinge, die ich mit Skeptikerblogs verbinde, nicht aber mit dem Namen Eduardo Zorita. Dass Sie ihren Beitrag nun verteidigen, überrascht mich jetzt. Ich dachte, wir wären in Zeiten angelangt, wo die Kriegsbeile begraben sein sollten. Dabei belasse ich es, ich habe kein Interesse, die alten climate wars aufleben zu lassen.

Viele Grüße,

hvw said...


do you think that fitting a polynomial to GMSL is scientifically more meaningful than, say, fit an exponential function?

From my limitied knowledge of what we know and don't know about the driving processes, both appear equally meaningless, scientifically. Then, however, your article might be interpreted as transporting a non-scientific message dressed up with a statistical method for sciency appearance. And that emits funny odor. But I'd be happy to be corrected by some theory that says sea-level rise ought to be polynomial.

Anonymous said...

Fundamentalisten... so einfach kann man eine Diskussion beenden.

Jeder hier sieht, dass Zoritas Polemik, nicht mehr oder weniger war es, eben ein wenig nach hinten los ging. Und nun ist Zorita eingeschnappt, dass es nicht vergessen wurde. Das ist eben das Risiko einer Polemik. Was hat denn Zorita gedacht, dass wir den Post nach 5 Jahren vergessen haben? Nö. Das ist ein Hobby von mir in Klimawandelfragen Posts von vor 5 Jahren oder länger mal wieder anzugucken. Wenn es den Blog in 5 Jahren noch gibt, hole ich den Post noch mal vor :)

Das zeigt auch, dass ein Blog keine perfekte Plattform für eine solche Diskussion ist. Viele Themen sind langfristige Diskussionen mit Änderungen der aktuellen Lage, der Erkenntnisse oder gar der eigenen persönlichen Einstellung dazu. Das liegt in der Natur der Sache: Klimawandel.

Ich finde es allerdings allgemein daneben, dass Zorita einfach Menschen unterstellte, dass sie absichtlich die Öffentlichkeit irreführen und betrügen. Das ist unnötig und vor allem: falsch. Es ist einfach nur miese Nachrede ohne Beweis, Details, Namen und Motive. Macht man einfach nicht. Tja, da bin ich ein wenig fundamentalistisch.

Und nun die Pointe: ich bin nicht mal ein "Fan" der "semi-empirischen" Modelle, auf die Zorita sicher anspielte. Die Kopenhagen Diagnose war aber nicht so schlecht, obwohl ich den IPCC-Report zum Thema klar bevorzuge. Die "Diagnose" war ein anderes Ziel der "Polemik" Zoritas. Nun ja... was soll's. Ich bin nur ein Fundamentalist.

Beste Grüße,

Hans von Storch said...

First, it may be useful to explain why the posts here show up somewhat delayed. If a thread has reached a certain age, not sure,maybe 30 days or so, all incoming comments are supposed to be moderated - and indeed many of the "late" comments are unrelated to the topic of the thread. Therefore it may take a few hours before the valid comments show up here

Second, I suggest people to read the original piece of Eduardo, and then ask if the actual comments here are related to the original text, which Eduardo qualified as "Do not take this too seriously. It is just a pastime while we wait."

But, there was one comment which I thought was interesting - it was by hvw on how to describe "trends" - "do you think that fitting a polynomial to GMSL is scientifically more meaningful than, say, fit an exponential function? From my limitied knowledge of what we know and don't know about the driving processes, both appear equally meaningless, scientifically." The request for something being "scientifically meaningful" sounds good but is a rather empty request. It makes sense, however, to use function families which are (after suitable normalization) orthogonal; if the series are stationary, trigonometric ones may be good; if they are not, polynomials make sense - but one may think of others as well.

It makes certainly sense to use as first approximation a linear function, and then consider if the "difference to linear" is associated with a tendency "in/decreasing positive/negative difference to linear"; if you wish to do that with a quadratic or an exponential - fine: the choice would be arbitrary - and I would expect the result to be similar. The shorter the period, the bigger the sensitivity to new data points.

Maybe we should start a new thread discussing this issue, which certainly depends on what we know, or assume, about the drivers, the internal dynamics, the time periods and the time scales of both etc. The issue of trends is a difficult one, among others because everybody believes to know exactly´what it is, and because of the big trap of "significance of trends".

Another problem is that in everyday language a "trend" is expected to describe changes in the foreseeable future; that it would continue on into the next years. This connotation is not included in the statistical meaning of the term. We have a trend towards higher temperatures in Hamburg from January to August, but this does not imply that September is warmer than July.

Anonymous said...

@ HvS

Ich denke, Fits an bisherige Daten anzulegen ergibt hier überhaupt keinen Sinn. Die Prozesse, die für den Meeresspiegelanstieg für die nächsten Jahrhunderte entscheidend sind, hängen mit der Stabilität der Eisschilde zusammen. Regionale Instabilitäten können plötzlich und abrupt erfolgen. Extrapolationen mit Daten stabiler Eisschilde ignorieren diesen Prozess.

@ ghost

Ich hatte diesen Post vergessen und war dankbar dafür, dass Sie darauf hingewiesen haben. Mich überkam beim Lesen eine gewisse Nostalgie. Das waren die Zeiten, als Streit über die richtige Klimapolitik auf dem Felde der Klimawissenschaft ausgetragen wurde. Gott sei Dank sind diese Zeiten vorbei, heute stehen die wirklichen politischen Fragen im Vordergrund (Wirtschafts- und Sozialwissenschaften sind z.B. bedeutsamer geworden).
Selbst der Temperaturrekord 2015 interessiert hier niemanden mehr, aber aus Interesse habe ich die erregte Diskussion von vor einem Jahr hier noch einmal nachgelesen.

Die Zeiten haben sich verändert. Interessant für mich an den letzten Kommentaren war nur zu sehen, wer diesen Wandel auch selbst mitgegangen ist, und wer nicht.

Grüße an alle,

Hans von Storch said...

Andreas, ob es Sinn macht, Fits an Zeitreihen zu legen, hängt davon ab, was man fragt, mit welcher Hypothese man sich beschäftigt. Man kann zum Beispiel fragen, ob wir langsam (oder auch abrupt) aus einem Regime hinausgehen, in dem die Eisschilde überwiegend stationär waren (sofern man das mal annimmt) oder ob wir weiterhin den bisherigen linearen Anstieg (den ich nicht verstehe) sehen.

Frage ist auch, ob die in den letzten jahren/zehnten etc. gesammelte Evidenz eine Bedeutung für unsere Deutung und Erwartung hat, oder ob wir uns auf theoretische Konzepte verlassen, deren Validität wir - wenn wir die vergangene Änderung nicht zu dekonstruieren suchen - wir nicht bestimmen können?

eduardo said...


können Sir bitte uns sagen, wo ich in meinem damaligen Beitrag 'Menschen unterstellt habe, dass sie absichtlich die Öffentlichkeit irreführen und betrügen' ?

Ich sehe es nirgendwo. Das haben Sir einfach selber ausgedacht.

Sie haben mir gebeten, neue Berechnungen mit aktuelleren Daten durchzuführen. Das kostet mich etwas Zeit. Trotzdem, wenn die Ergebnisse anscheinend Ihnen nicht passen, antworten Sie mit weiteren Forderungen und Unterstellungen. Sie gehen davon aus, dass jede Aussage gegen eine Beschleunigung von einem Skeptiker kommen muss. Wahrer ist es , dass es ist noch zu früh, um eine Beschleunigung im Meeresspiegel detektieren zu können:

Wenn man anderem etwas unterstellt,sollte man sich vorher besser informieren.

Andreas, hvv
das war und ist nicht 'meine Methode'. Ich habe auch nicht geschrieben, diese Methode wäre robust. Eher im Gegenteil, ich habe darauf hingewiesen, dass die Berechnung nicht sehr ernst genommen werden soll. Es überrascht mich auch, dass Sie mir das unterstellen.

Hier haben Sie einen Überblickartikel, wo diese und andere Methoden beschrieben werden, und wo Sie auch lesen können, dass diese Frage weiterhin diskutiert wird.

eduardo said...


I have just read all older comments in this thread again. I can only see a nice exchange of ideas, with links to papers, explanations, discussions, etc..

Nothing about fraud, lies or similar terms.

What is this now coming from ? I have no idea.

hvw said...


let me take back the "funny odor" remark. I must admit that I didn't go back to the original post but rather commented on your #30. Where there you seem to predict GMSL in 2100 from the quadratic fit. Which of course would be nonsense. But scrolling up I see that you made it quite clear that this is just a little mathturbation while we wait and the "prediction" just tacked on as decoration.

Thanks for the review paper. It is quite boring and has a highly irritating statement:

"2.1 Why Estimating Trends?
Second, trends may be viewed in terms of prediction. The estimated trend is conceived as that part of the series which, when extrapolated, provides the clearest indication of the future long-term movements in the series."

This is what I wanted to attack. In the context of a trend derived from the historical record of the variable, without covariates, without considering driving processes, and without the ability to assess model performance (as in NWP) -- I think this is complete and utter bullshit.

Hans von Storch said...

Yes, opinions, hvw.

I am highly irritated because of your language and intention.

You are "attacking" - that means you want to destroy something or somebody; your intention is not to exchange ideas, with the purpose of testing other peoples' ideas and views, but also to reflect upon your own positions and the possibility of error, unintended limitations or incompleteness. The purpose of Klimazwiebel is not "war", not "attacking" but exchanging, building a discourse.

Your quote about the utility of trends, I find also not adequate, but why labelling it as "utterly bulllshit". The most frequent reaction to such insults is "counterattack", is the interruption of exchange. Why not saying "is inadequate" because of the good reasons you have listed?

The article is "quite boring". Is it - in absolute terms? Do you have the authority to determine if something is boring or exciting? For whom? You are entitled to find it boring for you, but on the other hand, I find it boring for me to learn what you consider boring.

But, often enough, your input is very useful and well founded. Thus, why behaving as a bully? - behind the secure curtain of anonymity. Just try to behave, please, respect those, whom you may want to label "enemies", and which I would call opponents.

May I suggest that you have a text here on Klimazwiebel on "trends" - such a text may be more convincing that your claims of "bullshit". From what I read, you have something to say on this topic.

hvw said...

Dear Hans von Storch,

thanks for your opinion. Seems I did it again - improper language on the Klimazwiebel, sorry.

"The article is "quite boring". Is it - in absolute terms?"
For me it is self-evident that such a statement can only be completely subjective. In fact that paper is surely highly interesting for people who are involved with such methods and even would have been for me, at some point in the past. Yes superfluous sentence that doesn't contribute anything. A little allowance for literary freedom perhaps? It's a blog, not Nature.

"You are "attacking" - that means you want to destroy something or somebody"
Yes, that is correct. I am attacking a misconception and if that could contribute only a little to destroy that misconception in a person's mind, I'd be happy. That doesn't preclude the possibility to accept that my thinking is erroneous which is, in my mind, always a possibility for anybody, self-evidently.

The use of overly strong language ("bullshit") maybe happened in the hope to provoke any reaction at all - which then could lead to some serious discussion and exchange of ideas. Not working? But conjunctives and euphemisms also don't work here.

I don't want to "bully". In fact I took back my inappropriate offensiveness towards Eduardo. If Hans Visser showed up here, he would have something to say why this is not "bullshit", I would concede and we could sort out how and how not this paragraph makes sense. I'd also have a chance to compliment him to good (subjective) parts of that work, such as advocating for the use of structured time series models in climate science. I regard a blog comment as a part of a conversation, not a document that ought to be correct and balanced and stand by itself for eternity.

[absolute terms used for readability, insert relativizations and conjunctives at your own discretion]
Back to the topic: The misconception a am talking about is the belief that the past of a time series, by itself, in the absence of other sources of information, can tell you anything about its future. This should be common sense and is at least clear for anybody who took an introduction to applied stats. Yet, people ruin themselves with elaborate roulette-systems and "technical analysis" in the stock market. And if I got a cent for any paper where lack of actual scientific hypotheses is hidden behind a hierarchical Bayesian, neural-network, random forest with boosting, improved lasso, VAR, GLM whatever (literary freedom) analysis to come up with a result that is exactly as meaningful as an ordinary linear regression, I'd buy the US presidential election.

Hans von Storch said...

hvw -
I think you summarized your point about extending trends into the future very well, and I wholeheartedly agree to you position. It makes very much sense to me. And I would expect that your calmly presented argument will be much more convincing for others than using rowdy language.
I would also agree that many scientistts use in their scholaraly articles technically demanding methods for extrapolating into the future, without a basis in a persistent or evolving forcing or without exploiting -to the extent possible- the inherent memory of the system. Since it it a widespread phenomeon, we need more exchange about this misconception, but attack will not help but arguments. Detection & attribution is a meaningful replacement of this misconception.

hvw said...

Hans von Storch,

nice to hear that we are on the same page on that subject. I guess climate science is in comparatively good shape because domain expertise meet statistics skills often enough. And I find it increasingly difficult, in a climate of science abuse, not to question the motives of people who commit such childish mistakes and always come up with their results pointing to the same direction. I know that such an attitude is contrary to the good and well established conventions of conduct in the profession, but I am not sure whether they apply under post-normal conditions.

Better ignore that and look at the other opportunity to make mistakes: If you want to incorporate a big chunk of process knowledge and hypotheses into the statistical model, you quickly reach the point where the most advanced methods barely suffice and where exchange with experts surely has a good effect.

Hans von Storch said...

hvw - for me, in a postnormal situation political utility is sometimes more important than scientific methodical rigour - on both sides (or all sides, if there are more than two). Becoming angry and using strong, polarising or even insulting language does not help, since "the others" do it also.

Your last paragraph I do not share. If you refer to formally assimilating data into a model, you may be right. But in general, additional dynamical knowledge allows the usage of simple methods. And I would believe that the real experts in data assimilation are in meteorology and oceanography.

In the process of the International Meeting on Statistical Climatology IMSC, next one in June in Calgary!, we have tried to bring in "statistical experts" from outside for a long time. It was rarely succesful - with one reason being that these smart professionals are unaware of the dynamcial knowledge, which you referred to.

One of the new concepts, which ws introduced from outside (here: theoretical physics) was "long memory" - but it was mostly the determination of the long tail of the autocovariance function (and the consequences for, say, confidence intervals), but not how it is dynamically established. (My own explanation: presence of many components with very different "short" memories.)

hvw said...

Hans von Storch,

I agree that polarizing and insulting each other doesn't help. But I feel that trying to engage with some varieties of "others" with the hope of having a rational, calm, scientifically minded, fact-driven argument, where everybody is ready to acknowledge own mistakes, is hopeless, a waste of time and effort. Because there is only a rational facade. And in pointing out such cases, a good measure of ridicule is OK.

> But in general, additional dynamical knowledge allows the usage of simple methods.

Maybe an example, where additional dynamical knowledge has simplified the statistical treatment, would help me here.

I guess it can go both ways: Each additional process taken in into account will add to the deterministic part of the model, with more parameters (degrees of freedom) and assumptions about the error structure. In case your dynamical knowledge tells you that the deterministic part is non-linear and has non-Gaussian error structure, I can't help but feel seriously under-powered to tackle such modeling. But this is an environmental scientist speaking, Physics-educated people (like meteorologists and oceanographers have different standards ...).

Formal assimilation (you refer to 3/4D-Var, EKF, ... for, e.g reanalysis?) I did not refer to, but that ain't easy either from where I am looking ...

> we have tried to bring in "statistical experts" from outside for a long time. It was rarely succesful - with one reason being that these smart professionals are unaware of the dynamcial knowledge, which you referred to.

Yes, but that is where the cake is, no? Either the stats people learn some climsci, or the climate scientists learn some (more) stats, or both. Interdisciplinary is not easy.

>long memory ..
I remember that to be difficult too: Estimate of Hurst coefficient (or similar) sensitive to wrong method and also to inhomogeneities ...

Hans von Storch said...

HvW: in the detection and attributrion business - if we have qualified "guess patterns", which suggest "directions" of strongest or best signal-to-noise signal, the statistical hypothesis tests become manageable with a reasonable power. Then, you can employ straight Hotellling or even t-tests.

"Either the stats people learn some climsci, or the climate scientists learn some (more) stats, or both. " - ideally, true. But practically this was rarely the case (in my perception). "Interdisciplinary is not easy." True, but it is not suffficient to bring them into the same lecture hall; if they start a joint project, it may be better.

When you refer to a "model", you mean a statistical model, with the selection of predictors being physically motivated (or by the opportunity of available data) and coefficients fitted to data by optimizing the performance of the full model to determine the predictand?

hvw said...

Hans von Storch,

"When you refer to a "model", you mean a statistical model, with the selection of predictors being physically motivated (or by the opportunity of available data) and coefficients fitted to data by optimizing the performance of the full model to determine the predictand?"

Exactly, the common idea of time series analysis that I had in mind in these comments. I think that is a different category than "detection & attribution" methods (we talk about optimal fingerprinting, right?) you are bringing up.

I guess I see what you mean with simpler statistcs through dynamics knowledge in that context. In these methods you make a lot of "soft" choices, based on hard to formalize domain knowledge to assure that assumptions necessary for a simple statistical treatment are fulfilled.

E.g. the choice of the "guess pattern" (indicative of the forcing-effect of interest (bit circular, no?); enough good obs; aggregated enough to make it Gaussian, assumed to be modelled reliably enough by the AOGCM(s); at time & space scales that allow for a good estimate of internal variablility; and "interesting", which these days probably means small-scale & other variables than temperature, which in turn constitutes a tradeoff with respect to the other requirements above). Another soft tradeoff is the choice of the dimension reduction of the phase space. And you somehow need some confidence that the external forcing effect just adds linearly to what you identify as internal variability. And then some freedom in choosing the AOGCM (ensemble)...

Unless you write papers on that yourself, you have to have quite a lot of trust in the judgement of those who do. Which I have, certainly more than in an arcane statistical method that promises to deal with weak assumptions rigorously but escapes my direct judgement as well.

Yet, I feel the takeup of what the stats-people produce into (climate science) - applications could be more efficient than it is. Because of the difficulties of true interdisciplinary work, which you also observe. Which, I believe, have something to do with the incentive structure in our academic environment. If your H-Index is what your are after (and who doesn't want tenure?) then you avoid spending time to learn something completely new and the risk of irritating reviewers with stuff they don't know about.