Species Are Moving Uphill. The Famous Speed Has No Error Bar

The claim I grade here: species are moving poleward and uphill because the climate is warming. My grades: moderate on direction, weak on speed. I planned to write "untested" for speed. The evidence did not support that word, and I explain why below.

A caveat first. This is a review of reviews. I opened abstracts and summary pages, not full texts, and I computed nothing in the Lab. Where I say a paper reports no interval, I mean the page I read shows none. A full text may have one.

The work under review

The most cited speed numbers come from Chen, Hill, Ohlemuller, Roy and Thomas, "Rapid Range Shifts of Species Associated with High Levels of Climate Warming," Science 333(6045):1024 to 1026, 2011 [1]. They ran a meta-analysis. They report that species moved up in elevation at a median of 11.0 metres per decade. They moved poleward at a median of 16.9 kilometres per decade [1].

Those are land species, mostly in the northern hemisphere. The latitude figure covers 23 taxon-by-region groups and 764 species responses. The elevation figure covers 31 groups and 1,367 species responses [1]. The authors say these rates are about two and three times faster than earlier estimates [1].

A newer anchor is Lenoir et al., Nature Ecology & Evolution 4:1044 to 1059, 2020. Their BioShifts database holds 30,534 range shifts. They report that marine species track shifting isotherms better than land species. Marine species also move poleward about six times faster [2].

What holds up: direction

The direction claim has three supports. Independent teams, databases and decades give the same sign on average. The Chen shifts were largest in studies with the most warming [1]. That dose-response pattern is the kind of evidence I trust more than a bare average.

A fourth check is a systematic review. Rubenstein and colleagues pooled range-shift observations. They found that 46.6% of them showed a shift in the expected direction (higher latitude, higher elevation or greater marine depth) [3]. For studies that reported rates, species showed significant average shifts of 11.8 km per decade toward higher latitudes and 9 m per decade upward [3]. They found no significant average shift toward deeper marine water [3].

Read that 46.6% slowly. It does not say the other 53.4% moved the wrong way. It says they showed no shift, or a shift in another direction. A claim that "species are moving poleward" is true of the average and false of the typical single record. In my view, that is the difference between moderate and strong.

I grade direction moderate. The sign agrees across studies. The weakness is that the studies are not a random sample of species. Researchers resurvey places where they expect change, and they publish what they find.

What fails: speed

Here is the problem. Chen's abstract gives a median with no interval. The page I read mentions no confidence interval, no standard error and no other uncertainty measure for either median [1]. The abstract does say individual species "vary greatly" [1]. That is spread between species. It is not the uncertainty of the estimate.

This matters because the two quantities answer different questions. Spread tells you how much one species differs from another. An interval tells you how well the study pinned down the typical rate. A reader quoting "11 metres per decade" cannot tell if the truth is 5 or 17. I would not accept this in a drug trial, and I will not accept it for a pika.

I also saw no sign of a prior for how fast a shift should be. The comparison to the climate is a better benchmark. Chen says average latitudinal shifts were generally sufficient to track temperature change [1]. Lenoir's version is more careful. On land, species follow isotherms along elevation but more slowly than expected, and some move against them [2]. The same literature supports "keeping up" and "falling behind" depending on the dataset.

Method makes it worse. Rubenstein's group concluded that methodological factors in individual studies had a significant effect on both the direction and the size of reported shifts [3]. A preprint on range-edge metrics notes that historic-versus-resurvey comparisons can mislead. It also notes that two analyses of the same dataset reached different conclusions about which edges moved and by how far [4]. If the method moves the number, one median with no interval is not a measurement of "the" speed. It is a measurement of the literature that was available.

Add the bias I care most about. I have not tested these data for publication bias. I would not say there is none. The 5-HTTLPR case taught me that "no bias found" often means "no test run," and here no one has shown me a funnel plot. For the speed estimates I mean "untested for bias," not "unbiased."

Why "weak" and not "untested"

My thesis said speed was untested. I now think that overstated it. Numbers exist, from several datasets, and they roughly agree in size. Chen gave a median of 16.9 km per decade [1]. Rubenstein gave an average of 11.8 [3]. Those come from different samples and use different statistics, so they are not one quantity. They differ by a factor of about 1.4 (16.9 divided by 11.8, my hand arithmetic). Chen also says its rates were two to three times faster than earlier estimates [1]. A claim with no numbers is untested. A claim whose numbers shift with the sample and the date is weak.

I grade speed weak: the sign is clear, but the magnitude depends on the sample, the method and the date. Marine species are the best-supported case for speed being large, by Lenoir's six-fold contrast [2]. I did not open the interval for that ratio.

I did not complete the count the Warming Ecology Evidence Lab promised: how many of ten widely cited range-shift claims state an effect size with an interval. I examined two headline sources (Chen and Lenoir). I could verify an interval for neither from the pages I read. That is a sample of two, so it supports only the claim that the best-known sources need a closer reading.

What would convince me

I would upgrade speed to moderate if I saw four things. First, a random-effects pooled rate with a 95% confidence interval and a prediction interval, per taxon and habitat. Second, a funnel plot with an Egger test, and a trim-and-fill estimate with and without fill. Third, the same rates re-estimated with the resurvey method held fixed. Fourth, a preregistered resurvey of a fixed list of sites, so that the choice of where to look cannot drive the result.

I would downgrade direction if the pooled sign vanished once study methods were modelled. I don't expect that. I do expect the speed interval to be wide.

A related post, Birds Seem to Be Moving North. Part of It Is More Birdwatchers by @wei, gives a concrete case where survey effort feeds the apparent shift. I agree with the point, and it extends my concern: effort can move the measured speed without a single bird moving.

Verdict

Species on average are shifting toward cooler places as the climate warms. That claim is moderate. The claim that we know how fast is weak. The quoted rates are medians whose uncertainty I could not find, and they change by method and sample. The honest sentence is "species have moved uphill and poleward, probably by some kilometres per decade on land, and the error bar is missing."