Dr. Roy Spencer's Sea Surface Temperatures

Gneiss says:
March 6, 2013 at 10:04 am

Willis writes,

“Dr. Roy’s results agree pretty well with the Reynolds OI data, and that there’s been no significant increase or decrease in SST in the last 15 years or so …”

The “or so” part is not true. The increase in Reynolds SST (your data) appears not significant if you pick 1998 as the start year, or 1996, 1997, or 2000. But it is significant if you pick 1999, 2001, 2002, or 2003 as start years (or anything before 1996). This leads into that whole discussion about cherry picking, and why it’s a poor way to draw conclusions.

I was being quite conservative. If you adjust for autocorrelation using the method of Nychka, there’s been no significant increase in the Reynolds SST database since just after the beginning in 1981.
Cherry picking? Sorry, no way, I don’t engage in that kind of thing. The problem is the autocorrelation of the SST database is very, very high (0.94). As a result, we expect it to contain quite wide swings compared to a random walk … and that’s exactly what it’s done. The lowest p-value for the entire period is 0.048, barely significant at the weak level (p less than 0.05), and that is the only significant value for the entire dataset. If we start with the second month, the p-value is already rising, it’s 0.502, not significant, and it just gets worse from there.
So yes, what I said is not only true for the last 15 years—it’s true for the whole Reynolds dataset.
w.

kadaka (KD Knoebel) says:
March 6, 2013 at 11:02 am

Let’s see, run the checklist:
1. Willis
2. Ocean temperatures
3. Nothing significant happening globally (no sign of global warming)

I’m not clear what this “checklist” is supposed to represent … other than your ideas. Totally unclear, sorry. Am I supposed to have said there’s no sign of global warming? If so, over what time period? Or is that your claim?
Useless.

Yup, we’re about ten posts from the first commenter with some version of “Well Willis, if you were smart and/or educated enough to understand the uncertainties/error margins and/or properly compute the (statistics, trends, differentials) then You would know the real truth is…”

Whose post are you talking about, and why on earth aren’t you QUOTING or LINKING to whatever you’re on about? This is just a pathetic attempt at character assassination.

Bonus points for bringing up the ARGO data and trying to reduce his obvious glaring overwhelming ignorance of how the floats operate and the quality of those results, and for mentioning how the truth of the oceans is clearly revealed by Levitus 2012, of which perhaps someday Willis will make the tremendous personal sacrifice of someday schlepping down to the public library often enough that perhaps he could review and acquaint himself with that fine scholarly peer-reviewed expert document.

You are the first person to discuss either Argo or Levitus on this thread, which is about the Reynolds and the Spencer SST measurements … so I have to ask, are you drunkblogging? Have you posted this on the wrong thread? What does the Argo data have to do with this thread?
As to Levitus, if you have a page number and a quote I’m happy to discuss exactly what you are referring to. At present, your pointing at the entire document and saying that the answer to all mysteries is in there somewhere is useless handwaving and political posturing.
If you object to my words, QUOTE THEM and tell me exactly where I’m wrong. Don’t bother me with your bullshit checklists and vague claims, that kind of thing is as useless as a trailer hitch on a bowling ball.
I know you can do much better than that whole meaningless farrago, KD.
All the best,
w.

kadaka (KD Knoebel) says:
March 6, 2013 at 1:12 pm

Re Willis Eschenbach on March 6, 2013 at 12:45 pm:
Oh Willis, you’re wrapped entirely too tight these days.

Clearly I totally misread your intent, my apologies.
You might, however, profit by [sarc] tags in the future. I get attacked relentlessly for things people imagine I said. The problem is, no matter how outrageous and unbelievable you think you are being, I’ve likely been attacked by someone seriously saying the same thing …
Again, my apologies for my response.
w.

Gneiss says:
March 6, 2013 at 1:00 pm

Willis writes,

“I was being quite conservative. If you adjust for autocorrelation using the method of Nychka, there’s been no significant increase in the Reynolds SST database since just after the beginning in 1981.”

Your adjustment increases the standard errors so much that nothing is significant over this time frame. The regression trends remain the same. As does the visually obvious increase in that graph.
Why not take a modern approach and model autocorrelation directly? Doing so gives a different result.

Mmmm … thanks, Gneiss, nice screen name by the way. The problem is that claiming “a different result” doesn’t tell me much. Actually, it tells me nothing.
More to the point, I’ve looked at that. In the past, I’ve run a number of tests comparing Nychka’s adjustment versus the actual variations in ARIMA random autocorrelated data. I investigated various combinations of AR (auto regression) and MA (moving average), particularly in the range common with climate variables. Curiously, these often run high AR (≈ 0.8) and negative MA (≈ -0.3).
I haven’t run this particular dataset, but I’ve never found Nychka’s algorithm to be too far out, it gives good results.
Finally, I have provided my data and spreadsheet above in the head post. If you think there’s a better way, I invite you to do what I did. Use that same data, do your own analysis, and post the results on the web.
All the best,
w.
PS—Assuming I haven’t considered something is not infrequently a mistake …

Gneiss, I did the analysis you suggested. Remember that Nychka indicated, the full length of the data (375 months = 31+ years) is significant, but not by much.
Here’s the usual measure of whether the trend is significant. Below is looking at the last 300 months (25 years) of data, since the start of 1988. Temperatures and dates are in two variables cleverly named mytemps & mydates.

lm(formula = mytemps ~ mydates)
Coefficients:
            Estimate Std. Error t value Pr(>|t|)
(Intercept) 5.042264   2.387385   2.112   0.0355 *
mydates     0.007675   0.001193   6.432 5.01e-10 ***
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1 

Note the claim that the result is very significant. Also, the slope is + 0.0076°C/year.
To do the analysis, I first determined the best fit AR and MA coefficients for the full Reynolds data.
These turned out to be AR = 0.76 and MA = 0.64, viz

Call:
arima(x = thetemps, order = c(1, 0, 1))
Coefficients:
         ar1     ma1  intercept
      0.7590  0.6358    20.3639
s.e.  0.0352  0.0330     0.0248
sigma^2 estimated as 0.005084:  log likelihood = 457.14,  aic = -906.28

Then I generated a thousand random ARIMA datasets with the standard deviation, AR, and MA, of the Reynolds OI data. I looked at the slopes of the 1,000 datasets, to see how common it was to find a slope of 0.008°C per year (the slope of the Reynolds global SST data).
I found that for a p-value less that 0.05, the usual climate science standard for kinda significance, you need to go back about 300 months. That is 25 years … anything less than that, according to the Monte Carlo analysis, isn’t significant.

> datacount=1000
> recordlen=300
> myrand=arima.sim(list(ar=.759,ma=0.6358),n=recordlen*datacount)
> mym=matrix(myrand,nrow=recordlen)
> mym=mym*sd(thetemps)/sd(myrand)
> mydata=apply(mym,2,getslope) # mydata is 1,000 slope results
> mean(mydata)
[1] 9.498117e-05
> round(sd(mydata)*1.96,4)
[1] 0.0074

The final number is the p = 0.05 slope. If the test slope is higher than that 0.0074°C/yr, it’s significant.
Now, notice that the slope over that 25 year period (shown above) is 0.0077°C/yr … just barely significant. And as you go shorter, things get worse. So rather than the trend since 1988 being very, very significant as we’d calculate without autocorrelation, that trend in the Reynolds data of twenty-five years is just achieving significance.
So indeed, I was right that there is no significant trend going back 15 years or so … to get a statistically significant trend you need to start about 25 years ago.
All the best, thanks for the suggestion.
w.

kadaka (KD Knoebel) says:
March 7, 2013 at 4:08 am

Re Willis Eschenbach on March 7, 2013 at 12:06 am:
So let’s just agree, if we were in the same physical location, to each have their beverage of choice and laugh the whole thing off, put it behind us, life’s too short.
Deal?

Deal anytime, my friend. Totally seriously, let’s have a good laugh. I’m buying …
James Allison says:
March 7, 2013 at 1:21 am

kadaka (KD Knoebel) says:
March 6, 2013 at 4:18 pm
A fun and well written spoof Kadaka – I recognise the skill, thank you.
Willis is either a wound up personality, tight, like a spring – as you suggest – which slights the ebb and flood flows of his incessant pacific stories. Or perhaps he’s simply gotten too enthralled in his own cleverness to see the lightness around him. Willis I recommend the old Czech movie “The unbearable lightness of being”

“Spoof”?
Listen, James, when a man is apologizing to you, the idea that you should “spoof” him is pathological.
Kadaka got into trouble the first time by being too foolish to put in a “sarc” tag. Yes, I was “spoofed”, and it damn near cost kadaka my friendship. Great joke, right?
You seem to think that taunting or spoofing a bear is funny. Heck, bear-baiting used to be a sport in England, you were just born too late.
James, I get attacked by ambulatory rectums every day of the week, who at present are totally indistinguishable from you. Yeah, go ahead, see if you can spoof me into thinking you’re one of them, that will be great fun …
I can assure you. The bear doesn’t see the joke, and it might cost you dearly. You’ve already lost stature on my planet, at this point your opinion is valued right up there with my dog’s opinion … and I don’t have a dog.
Ha ha ha, yeah, James, that spoof was damn hilarious, nothing like poking a man with a stick, and then when he jumps, tell him he’s wound too tight … you’ll make a friend for life that way.
James Allison.
No question … I’ll remember that name …
w.

James Allison says:
March 7, 2013 at 10:41 pm

Willis Eschenbach says:
March 7, 2013 at 9:38 am
Nice thought you wanting to remember me Willis but I real can’t fit your ego into my life.
All the best
J

I do love these tragic announcements about how someone is going to take their ball and go home, and it’s supposed to be this heavy thing, them depriving us of their priceless company and witty repartee. But they have to do it, you see … because I’m so unbearably bad.
Me?
James, you are the one who thought it was some kind of fun joke to “spoof” me. You are the one who thought it was great when KD poked the bear with a stick, and now that the bear bit your hand bit off as well, somehow you’re the noble guy and the problem is the bear’s ego is too big?
Really? That’s your excuse for your pathetic behavior, the bear’s ego?
And yet despite the fact that you can’t stand me, here you are, still barking at my ankles and disturbing respectable people, another man once again threatening to leave … but who hasn’t left quite gotten up the courage to leave yet, it’s dark outside, and besides, you gotta get in one more bark.
If you want to go, James, then why are you still here? Don’t stand on the order of your going. Don’t bother with final farewells, they just make you look operatic and ridiculous. Nobody cares if you leave, this is the internet. Someone equally unpleasant will show up soon, you won’t be missed.
Now, please be clear about my position in this, which is that you’re more than welcome to stay. If you want to go that’s your choice. If you stay and can get over your upset, you might learn something.
So the choice is yours, but if you go away, don’t go away mad, it’s hard on your liver … just go away. And if you stay … don’t stay mad either. Just remember your lesson about interfering when two men have a disagreement and a misunderstanding—if you want to take a hand in that kind of circus, the bear may bite it off …
Regards …
w.