Wrecked in a Hurricane

Guest Post by Willis Eschenbach

Bemused by claims about connections between solar radiation, shipwrecks, and hurricanes in the pre-publication press release of a paywalled study highlighted by Anthony here on WUWT, I thought I’d take a look. I got a copy of the study thanks to my undersea connection, much appreciated … they’re working with three datasets. One dataset, called “TCsupp”, uses years of “suppressed” tree growth in the Florida Keys as a proxy for local hurricanes. It covers the period from 1708 – 2009. Here’s the map of the location of the trees, along with recent (post-1850) hurricanes, from their Supplemental Information, which also contains their data in table form.

Figure 1. Original Caption: Geographical location of the BPK tree-ring site (red dot), HURDAT-derived (1851–2010 CE) (17) category 1–5 Tropical Cyclones (white lines) that tracked within 100 km (red buffer ring; n = 23) and 300 km (dashed red ring; n = 60) of the site.

The second dataset is a subset of the HURDAT historical hurricane dataset for 1851-2009, giving the hurricanes in the local area around the trees.

The final dataset, called “TCship”, covers a period starting before the Maunder minimum, from 1495 to 1825. It is a dataset of the numbers of selected shipwrecks per year as a proxy for the number of hurricanes. The problem, of course, is that the time period for which we have shipwreck data doesn’t overlap with the HURDAT hurricane records. Well, that’’s not the only problem with the shipwreck dataset. They say:

The TCship time series thus consists of the number of Spanish ships that wrecked per year in the Caribbean region due to storm activity or for unspecified causes.

Now, including “unspecified causes” is a worrisome choice. Particularly back then, ships went down for many reasons other than hurricanes. First, many square-rigged ships of the day couldn’t go to windward well at all, so if they got caught on a lee shore they often couldn’t stay off the rocks. In addition, the navigation of the 1500s, 1600s and 1700s was very primitive—no accurate charts, no accurate clocks, few sextants and fewer trained to use them, no lighthouses, buoys, or other aids to navigation. And setting aside the hurricanes, there were no weather satellites warning of small craft advisories that could easily force a ship into danger, or of periods of reduced visibility that could result in the loss of the ship, or of predicted squalls that could topple a mast.

As a result, a ship foundering on an unsurveyed reef with the loss of all hands or going down in a sudden squall was a common occurrence back in the day … and that doesn’t even count the wrecks from non-weather-related causes like fire and shipworms and inept commanders and pirates and poor maintenance and sleepy lookouts and lightning strikes. The list is long.

Finally, the causes of shipwrecks would assuredly have changed in the three plus centuries from Columbus’s time at the start of the record to 1825, and thus the composition of the reasons for the wrecks would also change.

So when I see that they are using storm-caused wrecks as a proxy for hurricanes, I scratch my head. Even losses in a storm are often for other reasons, most storms are not hurricanes, and most years don’t even have hurricanes in a pre-selected location.

But when I see they’ve included wrecks from “unspecified causes”, the hair on the back of my neck stands up. That’s special pleading, and something like that is rarely done without the reason being that it “improves” the analysis … but I digress.

Let me start by complimenting the authors for including the datasets in question. Without them, it would have been impossible to analyze their results. Well done, would that more authors followed the practice.

Their claim rests on three testable legs. The first is that suppressed tree rings are a good proxy for local hurricanes. The second is that shipwrecks are a good proxy for tree ring data. The third claim rests on the other two. It is that IF shipwrecks are a good proxy for tree ring suppression data, and IF tree ring suppression data in turn is a good proxy for hurricanes, then it follows that shipwrecks are a good proxy for hurricanes. This is how they say they get around the fact that they have no contemporaneous hurricane and shipwreck data.

Let me start with the first leg of the tripod, the idea that tree rings are a good proxy for local hurricanes. It turns out that yes, they are a good proxy for hurricanes in the sense that there is a low p-value, less than 0.001 … but what makes them a bad proxy is that the tree ring data only explains about 10% of the variance in hurricanes.

Let me give an example. There are 159 years in the overlap period of the hurricane records and the tree ring data, 1851-2009. Of those, 125 years had no hurricanes. The problem is that of the 125 hurricane-free years, no less than 80 of them had suppressed tree rings. This means that at a bare minimum about two-thirds of the suppressed tree rings occurred in years without hurricanes. And even when we discount the year after the hurricanes, there are still 50 years which are neither the hurricane year or the year after, but still have suppressed tree rings. This is not good news for the use of suppressed tree rings as a proxy for hurricanes.

Nor does it get much better if we just look at the years with lots of suppressed tree rings. There were 41 years when over half of the tree rings were suppressed … but 24 of those occurred in years with no hurricanes. So even when we only consider years with over 50% suppressed tree rings, more than half of the suppressed tree rings occurred in years without hurricanes. And even when we discount the year after the hurricanes, there are still 12 years which are neither the hurricane year or the year after, but still have suppressed tree rings. That’s about 30% of the suppressed tree ring years that are not caused by hurricanes.

So despite a very good p-value, there is a big problem with using the tree rings as a proxy—with only 10% of the variance explained by suppressed tree rings, there is just not enough of an effect for the suppressed rings to be a useable proxy for hurricanes. The method throws up far too many false positives to be useful.

Let me move on to the second leg, the idea that shipwrecks are a good proxy for tree-ring data. The overlap period of those two datasets is from 1708 to 1825. Bear in mind that the later part of the shipwreck dataset is likely to be much more accurate than the earlier part, both because of better record keeping and less lost records, and also because of less navigation related wrecks in the later years as charts and navigation improved greatly. With those provisos (which on their own might be enough to sink the ship of state, it’s unknown), here’s the good and the bad news.

The p-value of shipwrecks and tree rings is 0.005, again very good … but once again the same bad news. The shipwrecks only explain a small amount of the variance in tree ring suppression, in this case about 7% (R^2=0.065). And this low explanatory value is what we’d expect, given that the tree rings represent hurricanes in only one small area, but they’ve counted shipwrecks all over the Caribbean, viz:

We compiled all Spanish shipwreck events that were recorded to have occurred (i) due to storm activity (66%) or for unspecified causes (34%), (ii) during the hurricane season (July–November) or with unknown seasonality, and (c) in Florida (21%), on the Atlantic Coast of Mexico (25%), in Hispaniola (22%), Cuba (13%), the Lesser Antilles (5%), and Puerto Rico, Texas, South Carolina, Louisiana, Mississippi, Jamaica, the Cayman Islands, Bermuda, and the Bahamas (less than 5% each.

Why would we expect hurricanes in the Florida Keys to correlate well with shipwrecks in e.g. Bermuda?

So … now that we have the data, let’s look at the numbers. Given the fact that shipwrecks only explain 7% of the variance in tree ring suppression, and given that tree ring suppression in turn only explains 10% of the variance in hurricanes, how much of the variance in hurricanes would we expect the shipwrecks to explain?

I believe the correct answer is … nowhere near enough to draw any serious conclusions.

And note that this does not include any problems with changes over time in the underlying datasets, or problems in the selection criteria for the wreck data, or issues relating to the threshold chosen for inclusion as a “suppressed” tree ring, or increased uncertainties due to the small number of trees in the early period of the record, or …

My conclusion?

Not.

Not enough data. Not good enough data. Not enough correlation. Not enough explained variance. Not enough overlap between datasets. Not. That’s all. Not.

Anyhow, that’s the news from my delightfully soggy part of the world … the El Nino has indeed brought the rains. The cat mopes around the house looking for the door into summer.

Best to all,

w.

My Usual Request: Clarity is crucial for scientific discussions. If you disagree with me or anyone, please quote the exact words you disagree with. I can defend my own words. I cannot defend someone else’s interpretation of some unidentified words of mine.

My Other Request: If you think that e.g. I’m using the wrong method on the wrong dataset, please educate me and others by demonstrating the proper use of the right method on the right dataset. Simply claiming I’m wrong doesn’t advance the discussion.

My Formal Educational Qualifications:  For some reason this became an issue in my last post, gotta love trolls. Regarding educational qualifications, I have none. Get over it. My formal scientific education is a year each of college freshman physics and chemistry, period. Despite that, I now have five peer-reviewed papers published in the scientific journals, including a 2004 peer-reviewed “Brief Communications Arising” in Nature magazine (my conclusions were finally upheld more than a decade after publication by two other studies), and over fifty citations to my work. In addition, my posts and ideas have been discussed in the New York Times, the Sydney Morning Herald, the UK Telegraph, and other papers around the world. Heck, I even inadvertently and unknowingly set the stage for Climategate by making the very first Freedom of Information Request to Phil Jones. Go figure. But setting all that aside, the scientific issue is never qualifications—the only valuable question is, are my claims valid or not? If someone’s scientific claims are valid, their qualifications are immaterial … and if their claims are not valid, all the qualifications in the world won’t help their claims in the slightest. So please … can we skip the ad hominem attacks on my education, my ancestry, and my manifold sins of omission, commission, and emission, and stay focused on the science?