I ended up in a brief Twitter exchange yesterday with Roger Pielke Jr. He had responded to a tweet from Jeffrey D. Sachs with the following
Wrong @JeffDSachs Attached figure from paper you (mis)cite for WNP basin (where Haiyan hit) showing trends, see any? pic.twitter.com/662iPYohmW
— Roger Pielke Jr. (@RogerPielkeJr) November 12, 2013
The figure is from Elsner et al. (2008). The numbers a the top are maximum wind speeds (m s-1) and represent the values at each quantile (0.2, 0.4, 0.6, 0.8). The numbers at the bottom are for the (unseen) figure below, so ignore them. The y-axis is the trend (m s-1 yr-1). What this is telling us is the rate at which the maximum wind speed at each quantile is changing per year. This is explained quite nicely in the paper – when discussing global trends – when it says
For example, at the 80th percentile, on average 17 cyclones globally exceed 49 m s-1. With a 1oC rise in SST, the 80th percentile increases to 51 m s1. At this threshold level, on average 13 cyclones per year are observed. So the increase in SST of 1 oC results in an increase in the global frequency of strong cyclones from 13 to 17 cyclones (31%) per year.
Now, there is much more that could be said about this paper. I, however, don’t have much time to write this post and Phil Plait (the Bad Astronomer) has already covered it in quite some detail.
What I thought I would do here is pose a question for discussion. It seems that Roger Pielke Jr has interpreted the analysis for the Western North Pacific (WNP – the figure in his tweet, and the region relevant for Typhoon Haiyan) as implying that there is no trend. This is because the 90% confidence interval spans 0 m s-1 yr-1. Is he correct to interpret it in this way? Alternatively, considering only the data for the WNP, can one actually say more? For example, could one conclude that the analysis of the WNP implies that a positive trend is more likely than a negative trend (or no trend)? I know what I think, but I’d be interested in what others thought. As usual, please bear the moderation policy in mind, especially as I may be a bit too busy today to due much moderating.
Jim Elsner could perhaps be asked. But re Haiyan specifically, there seems to be little question that it blew up while crossing over an area of much-enhanced temps just below the surface. The same thing happened to Katrina. All else equal, TCs respond to such conditions very reliably.
Re the WNP analysis, had temps there increased as much as in the North Atlantic during the period analyzed? If not, that might explain a reduced trend.
The paper actually says pretty much that
So, when considering all the information, one could explain why some basins show weaker trends than others. The question in the post, though, was more to do with whether or not – given the WNP data only – that one would conclude that there is no trend, or would one could conclude that although one cannot rule out that there is no trend (at the 90% confidence level) that a positive trend is more likely than a negative (or no) trend.
I wonder if SST is quite the right metric given the importance of sub-surface water temps.
I see that Aslak Grinsted just updated the WNP analysis, although I don’t know enough about quantile regression to understand the meaning.
Yes, I saw Aslak’s tweet but don’t quite understand it either.
Quantile regression is actually pretty simple. It finds the best straight line with e.g. 5% of the data above it and 95% below it. You don’t really need to understand more than that to interpret the figure.
It turns out that the median of a sample and the value with least absolute dev. is the same. And least absolute deviation regression is the same as quantile regression for the 50th percentile. For other quantiles you simply modify the least absolute deviation with a weight. You can see the weighing specifics in my code here: http://www.mathworks.com/matlabcentral/fileexchange/32115-quantreg-m-quantile-regression/content/quantreg.m (the weighing of the residuals in the misfit function is done in the one liner function called “rho”)
Pingback: When is it ok to discuss climate change? | quakerattled
More interesting than the chart is table 1 from which we learn that the 95th percentile shows a trend of +0.29 m sec^-1 year^-1 in wind speed, with a standard error of 0.116 and a P value of 0.012. I believe that to be statistically significant in any man’s language, and you have to wonder why Pielke Jr would draw attention to the 90th percentile (the highest percentile in the graph) to dispute claims about Haiyan, which definitely fell well above the 90th percentile. (Actually, long experience leaves me not wondering at all.)
Of course, the 99th percentile shows a positive trend, which is not statistically significant in the WNP, but that is almost certainly due to the low number of storms sampled at that wind speed (just 21 for global data, and presumably much fewer for the NWP).
Finally, directly answering the question in the OP, even from the graph is is quite clear that for the upper 50% of typhoons in the NWP, there is a positive trend of wind speed with increasing temperature; but that that trend is not statistically significant (in most cases).
Tom, yes I agree that if one considers table 1 there is much more information and that if you consider the 95th percentile (for example) there is a statistically significant trend.
As you say
Some people seem to want to condense that to no trend which – to me at least – is ignoring a lot of the information that such data is providing.
Aslak, thanks – I just wasn’t quite sure what your lines were representing (my own lack of knowledge more than anything else). I now see what your figure is showing. Very interesting.
This seems quite similar to my observation from a 2009 publication by the Taiwanese EPA, which included (if you can forgive their own mis-labelling of Figure 2 on page 3) the startling fact that the number of extremely wet typhoons per decade has increased from 4 in the 1970s to 14 in the 2000s. As implied by my recent tweets, this suggests 10 out of 14 extremely wet typhoons would not have occurred if the climate had not been changing.
Of course, as with every other piece of inconvenient evidence, deniers accuse me of cherry-pickng my data… And then, without any apparent trace of irony, go back to asserting that a 15 year hiatus in surface warming means we can ignore all the change that is still accelerating (e.g. 80% loss of sea ice in the Arctic since the 1980s; a sixfold increase in the rate of Greenland ice sheet melting since 1990; and a tenfold increase in rates of glacier retreat in the high Arctic in the last Century)…
Clearly, cherries are a dish best served warm – with a generous topping of hypocrisy.
My main issue is this idea that we all have to accept the basic statistical argument that, given the data/analysis shown in the figure, if p > 0.05 then there is no trend. Well, as far as I can tell from the analysis shown in the figure, beyond the 50th percentile it is more likely that there is a trend than there isn’t. Claiming, based on that data, that there is no trend seems more incorrect than claiming that there is (which I’m not by the way). Certainly, we can’t rule out – with high confidence – that there is no trend. Also, we can’t claim – with high confidence – that there is a trend. The analysis does, however, tell us the likelihood of a positive trend and completely ignoring that seems wrong. Of course, I’ve just had a brief Twitter discussion with Roger Pielke Jr and apparently there is no trend and it’s not worth debating.
I should add, however, that I’m only referring in this post to the Elsner et al. study. I believe there is more recent work that considers the WNP some of which conclude that there is a trend and some of which conclude that there isn’t. So I’m certainly not suggesting in this post that there is a trend, I’m simply suggesting that this simply binary statistical analysis (p > 0.05 no trend, p < 0.05 trend) ignores much information that this type of analysis can provide.
Martin Lack, I do not think you can make that simple extrapolation in that neither the Taiwanese EPA publication nor you provide us with number of Typhoons per decade. Thus we cannot determine what part of the increase in very wet typhoons in 2000-2009 is due to an overall increase in the frequency of cyclones, and what portion is due to an increase in wet cyclones specifically. The publication indicates that there has been a 100% increase in rainfall in extreme wet events, which is likely to be a more reliable statistic. On that basis, 7 of the 14 very wet typhoons would not have occurred without global warming, plus or minus some undetermined, but probably large relative to the effect, error margin.
Wotts, interesting that the “update” to which Pielke refers you reports only trend per decade, and not trend per degree C increase in SST. With a dominant pattern of El Ninos over much of the period reducing SST in the NWP region, it is not clear that they will correlate well.
Thanks Tom, I hadn’t noticed that. Haven’t really had a chance to read it properly yet.
The notorious David Rose on “Typhoon Haiyan”.
“How the BBC turned a catastrophic crisis into a drama about global warming”
That article is terrible, Lars. But I really don’t expect anything less from David Rose.
Roger Pielke jr has an article in the Guardian playing down the evidence for increases in extreme weather events and the case for a global climate fund.
I’ve just seen that. It’s quite a remarkable article, but will need to give it some thought before writing more. This blog post is well worth a read and says some things I had thought of writing. I think the last paragraph is particularly significant.