As tempting as it is to tell Judith Curry to “Do the math!” — I don’t think she can handle it.
A recent post at WUWT quotes her saying this little gem:
It is possible that a shift to the cold phase of the AMO is underway, which would extend the warming hiatus for ~2 decades.
That made me wonder, what does she think a “hiatus” is? What’s her evidence for it?
I managed to find a post on her blog where she tells us. A hiatus, according to Judith Curry, is when
1) the rate of warming over a particular period of at least 10 years is not statistically significant from zero (with the context of a nominal 0.1C uncertainty)…
It’s almost impossible to believe that anyone who seriously claims to be a scientist would accept (let alone propose) such a ridiculous definition. Yes, she really means that any 10-year or longer trend estimate below 0.1 °C/decade is a “hiatus.”
If you’re wondering what “nominal 0.1C uncertainty” means, it should mean that the standard error (the likely difference between estimated and true trend rates) would be about 0.1 °C/decade. It turns out that for 10-year time spans, that’s quite close. It’s about the only thing she gets right.
Since Judith Curry’s post includes “show me the data” in its title, let’s look at some data. She shows graphs of data from NASA, from HadCRU, from NOAA, from the Berkeley Earth Surface Temperature project, from reanalysis (both ERA-interim and CFS), and — of course — troposphere temperature from UAH (but not from RSS). All we’ll need is one of them, so I’ll use the data from NASA. I’ll use yearly averages rather than monthly data, because the autocorrelation in monthly data is so strong you can’t afford to ignore it, and I don’t think Judith Curry can handle that. Here it is since 1975, with the estimated trend using least squares regression:
The estimated warming rate is 0.18 +/- 0.03 °C/decade, and that’s a 2-sigma error range for roughly 95% confidence. We conclude that the increase from 1975 to the present (the final figure for 2018 is year-to-date) is statistically significantly different from zero.
Now let’s use Judith Curry’s definition to look for a “hiatus.” I’ll test every 10-year time span to see whether or not the estimated rate is below her 0.1 °C/decade limit. Here are the results, in graphical form:
Every place one of the circles (the estimated trend rate) dips below the blue dashed line (the 0.1 °C/decade limit) meets Judith Curry’s definition of a “hiatus.” By her standard, there are 10-year “hiatuses” centered on 1982, 1983, 1985, 1991, 1992, 1993, 2006, 2007, 2008, 2009, and 2010. That’s a lot of hiatuses.
But wait — maybe she would insist that they’re not all separate “hiatuses,” they should be combined into longer “hiatuses.” After all, she did say “at least” 10 years. We can do that. Let’s look for even longer time spans with an estimated trend rate below 0.1 °C/decade. We can combine all those “hiatuses” into just two of them: an 18-year span from 1979 through 1997, and a 16-year time span from 1998 through 2013. If we plot those two “hiatus” trends we get this:
There’s a sizeable “jump” from the one to the other. I wonder whether Judith Curry actually believes this is a realistic portrayal of the temperature trend since 1975. If so, it might remind you of something.
There’s also the fact that longer time spans result in smaller standard errors of regression analysis (smaller “nominal uncertainty”). For the 18-year time span the standard error is less than half her 0.1 °C/decade “nominal uncertainty” so the trend actually is “statistically significant from zero” (she really means statistically significantly different from zero). The same is true for the 16-year time span. But by Judith Curry’s standard they’re not. Her standard doesn’t seem to involve actual statistics — at least, not done right.
If Judith Curry can declare a “hiatus” for any 10-year time span with an estimated trend rate just 0.083 °C/decade less than the overall 1975-present rate, then using the same “logic” why can’t I declare an acceleration for any estimated 10-year trend rate more than 0.083 °C/decade more than the overall 1975-present rate? Doing so would allow declaring acceleration centered on 1980, 1987, 1997, 1998, 1999, 2012, and 2013. That’s a lot of accelerations.
But of course I should combine some of them into longer accelerations. That leaves three: a 10-year span from 1975 through 1984, a 16-year span from 1992 through 2007, and a 13-year span from 2006 through 2018. Those trends look like this:
Seriously — if I went about declaring regular episodes of acceleration in global warming based on that (which is Judith Curry’s logic), how would Judith Curry respond?
And then there’s the fact that using Judith Curry’s logic, some of the “acceleration” episodes overlap with some of the “hiatus” episodes. In fact two of the “acceleration” episodes overlap with each other. How is it that by Judith Curry’s statistical standard, a given point in time can be in both an “acceleration” and a “hiatus”? How can the years 2006 and 2007 be part of two different “accelerations” and a “hiatus”?
Of course it’s all nonsense. There is no evidence of either “acceleration” or “hiatus” that stands up to scrutiny, when you do the statistics right. Judith Curry didn’t. My opinion: she can’t.
What she did is pick a ridiculously short time span to allow herself to declare a “hiatus” because the uncertainty in such short periods is so large. As for even doing basic statistics right — let alone any of the more sophisticated stuff that really is needed to get it right — I don’t think she can handle it.
She also gives an alternate definition of what makes a “hiatus,” a “2)” which has to do with model output. That’s just ignorant; there was or there wasn’t a “hiatus,” regardless what model results say. She further adds a “3)” about how anything exceeding 17 years is really truly definitely a “hiatus” — but she doesn’t require that, 10 years is long enough, and any trend estimate less that 0.1 °C/decade is low enough, for her to declare “haitus.” The “model output” thing is irrelevant. The “17-year” thing is based on faulty analysis by others, which ignores the multiple testing problem. But I don’t expect Judith Curry to get that either.
This blog is made possible by readers like you; join others by donating at My Wee Dragon.