Evolutionary Biogeography: What the Map of Life Shows

"there is also a relation, to a certain extent independent of distance, between the depth of the sea separating an island from the neighbouring mainland, and the presence in both of the same mammiferous species or of allied species in a more or less modified condition." — Charles Darwin, 1859.  This article tests that sentence.¹

"Although Darwin apparently was wrong in thinking that amphibians never cross saltwater, these cases reinforce a general message of the great evolutionist:  given enough time, many things that seem unlikely can happen." — Alan de Queiroz, 2005²

"We can find no geological evidence to indicate that land areas were continuously present throughout the middle Cenozoic." — Landis and colleagues on New Zealand, 2008, proposing that the whole subcontinent drowned and took its Gondwanan biota with it³

Summary

Biogeography is the study of where living things are. It was the evidence that convinced Darwin, before anatomy and long before genetics, and it is still the argument that is hardest for a young-earth model to answer. It is also, in its modern form, a field that spent forty years being confidently wrong about the thing it was most confident about.

This article does three things. It states the classical argument and tests its central prediction with a measurement made here. It traces what happened to the twentieth-century version of the theory, which is a story of a paradigm collapsing under its own data. And it says where the evolutionary account is weaker than its defenders usually admit.

The conclusions, in order of confidence

1. Darwin's 1859 depth prediction can be tested against modern bathymetry. He proposed that the depth of the sea between an island and the mainland predicts whether they share mammals. Measured here on modern bathymetry across nineteen crossings, the rule holds without a single exception. Every crossing with a sill shallower than the 134 m glacial lowstand has a shared continental fauna. Every crossing deeper than that has a depleted, endemic one. There is no overlap zone and no counterexample in the set.

2. The rule is not a restatement of distance, which is what makes it a real prediction. Cuba is 150 km from Florida and has no native land carnivores or ungulates. Trinidad is 15 km from Venezuela and has a South American fauna including monkeys, peccaries and ocelots. Distance says these should be similar. Sill depth says 718 m against 21 m, and sill depth is right. Darwin said so in 1859, calling the relation "to a certain extent independent of distance."¹

3. The disharmony of island faunas is the strongest single piece of evidence in the subject. Oceanic islands do not have small versions of continental faunas. They have a lopsided sample: bats but no other mammals, birds and lizards but no amphibians, plants but few of the plant families you would expect. Darwin's version: he searched the old voyages and could not find "a single instance, free from doubt, of a terrestrial mammal ... inhabiting an island situated above 300 miles from a continent," while "aërial mammals do occur on almost every island."¹ That is what a filter produces. It is not what either a land bridge or a separate creation produces.

4. Molecular dating has challenged many classic vicariance explanations and revived oceanic dispersal. Alan de Queiroz's 2005 review in Trends in Ecology and Evolution reviews that shift, and its verdict is that molecular dating "favors oceanic dispersal over tectonic vicariance ... in a wide variety of taxa, from frogs to beetles to baobab trees."² Southern beech, the textbook Gondwanan relic, turned out to have crossed the Tasman Sea.⁴ Chameleons turned out to have spread from Madagascar by sea.⁵ Frogs, which cannot survive salt water, turned out to have colonised a purely volcanic island surrounded by 3,500 m of ocean.⁶

5. This matters because it is what a science looks like when it corrects itself, and because it removes a bad argument from both sides. The creationist complaint that trans-oceanic disjunctions do not fit continental drift is correct, and mainstream biogeographers made the same complaint first, harder, and with the data.

6. The Hawaiian archipelago provides the cleanest independent test in the field, and it passes. The islands were built in order by a hotspot, and potassium-argon dates put Kauai at 5.1 million years and Hawaii at 0.43 million.⁷ If lineages colonised down the chain as each island appeared, their branching order should match the island order, and the amount of genetic difference should scale with the island ages. It does, across honeycreepers, Drosophila and crickets, and the resulting rate estimates agree with each other.⁷ Nothing in a separate-origins model predicts that.

7. Marsupials are the field's showpiece and its embarrassment at the same time. Retroposon insertions, which are close to immune to the artifacts that plague sequence data, show that all four Australasian marsupial orders share one origin and point to "a single marsupial migration from South America to Australia."⁸ That resolves an important relationship among living groups; it does not by itself settle the number or direction of migrations, as the fossil evidence in section 5.1 shows. But measured here from the Paleobiology Database, the entire Cretaceous metatherian record is northern: 773 occurrences in North America, 28 in Asia, 5 in Europe, and zero in South America, Australia or Africa. The group that is now austral was then boreal, and the switch is not explained.

8. New Zealand is the field's worst open problem. Landis and colleagues found no geological evidence that any land stayed above water in the New Zealand region through the Oligocene, and concluded that the modern biota "evolved largely from fortuitous arrivals during the past 22 million years."³ If they are right, then a tuatara, a Leiopelma frog and a kauri tree all had to cross an ocean. If they are wrong, the geology has to be re-read. Neither option is comfortable, and the dispute is unresolved.

9. Molecular dating carries more weight in this field than it can safely bear. Nearly every dispersal-beats-vicariance result depends on a clock, and the clocks are usually calibrated with fossils or with geological events, then used to date geological events. That is not automatically circular, but it is close enough to require care, and papers vary in how much care they take. The Hawaiian calibration in point 6 is the honest way to do it, because the island ages come from radiometric dating that has nothing to do with biology.

10. What biogeography does and does not establish. It establishes descent with modification from local ancestors, filtered dispersal, and a long history. It does not, on its own, establish universal common ancestry, and two cladists said exactly that in 1981 in a sentence creationists have quoted ever since. Read in context, they were making a narrower point than they are usually made to make, and the point they were making has since been addressed. The companion article on creationist biogeography audits that quotation in full.

The sill measurement in points 1 to 3 was made for this article and is shared with the companion piece. Scripts are in ../creationist-biogeography/research/.

1. Darwin's argument, in his own words

It is worth seeing how the argument was actually built, because the modern textbook version has been smoothed until the reasoning is invisible.

1.1 The three facts he starts from

Darwin opens chapter XI of the Origin with observations, not theory.

Similar climates do not produce similar life. "In the southern hemisphere, if we compare large tracts of land in Australia, South Africa, and western South America, between latitudes 25º and 35º, we shall find parts extremely similar in all their conditions, yet it would not be possible to point out three faunas and floras more utterly dissimilar."¹ The comparison is deliberately unfair to his own side: he takes three places as alike as he can find and shows that their inhabitants have nothing in common.

Barriers predict difference. "A second great fact which strikes us in our general review is, that barriers of any kind, or obstacles to free migration, are related in a close and important manner to the differences between the productions of various regions."¹ He runs it at every scale: oceans between continents, mountain ranges and deserts within them, and then, in a passage that reads like a controlled experiment, the sea.

The sea gives the cleanest version. "No two marine faunas are more distinct, with hardly a fish, shell, or crab in common, than those of the eastern and western shores of South and Central America; yet these great faunas are separated only by the narrow, but impassable, isthmus of Panama."¹ A strip of land a few tens of kilometres wide separates two marine faunas with almost nothing in common, while an unbroken ocean stretching from the eastern Pacific to Africa has no sharp faunal break at all. Distance is not doing the work. Barriers are.

1.2 The oceanic-island argument

Then he narrows to islands, and the argument gets sharp.

On amphibians: "Bory St. Vincent long ago remarked that Batrachians (frogs, toads, newts) have never been found on any of the many islands with which the great oceans are studded. I have taken pains to verify this assertion, and I have found it strictly true."¹ He then closes the loop by showing the absence is not about habitat. Frogs were introduced to Madeira, the Azores and Mauritius, and "have multiplied so as to become a nuisance." The islands suit frogs perfectly. Frogs and their spawn die in sea water. So frogs cannot get there, and that, and not habitat, is why they are missing.

On mammals, the same structure. No terrestrial mammal on any island more than 300 miles out, but bats everywhere: "New Zealand possesses two bats found nowhere else in the world: Norfolk Island, the Viti Archipelago, the Bonin Islands, the Caroline and Marianne Archipelagoes, and Mauritius, all possess their peculiar bats. Why, it may be asked, has the supposed creative force produced bats and no other mammals on remote islands?"¹

1.3 The prediction this article tests

Then comes the sentence that is the reason for this article's central measurement.

Besides the absence of terrestrial mammals in relation to the remoteness of islands from continents, there is also a relation, to a certain extent independent of distance, between the depth of the sea separating an island from the neighbouring mainland, and the presence in both of the same mammiferous species or of allied species in a more or less modified condition.¹

He credits the observation to Windsor Earl and points at the Malay Archipelago "traversed near Celebes by a space of deep ocean," adding that "this space separates two widely distinct mammalian faunas."¹ He is describing the Wallace Line, in print, before Wallace published it, and he ends the paragraph by saying that light will soon be thrown on the archipelago "by the admirable zeal and researches of Mr. Wallace."

Then he gives three worked cases: Britain, separated by a shallow channel, has the same mammals as Europe; islands off Australia behave the same way; and "The West Indian Islands stand on a deeply submerged bank, nearly 1000 fathoms in depth, and here we find American forms, but the species and even the genera are distinct."¹

That is a prediction with a number in it, made in 1859, when bathymetry was a rope with a lead weight on the end. It can now be compared with modern bathymetry, subject to the resolution of the grids.

1.4 And he anticipates the land-bridge answer

Darwin knew the obvious alternative, that all these islands were once connected by land, and he argued against it on a specific ground:

the scarcity of kinds, the richness in endemic forms in particular classes or sections of classes, the absence of whole groups, as of batrachians, and of terrestrial mammals notwithstanding the presence of aërial bats ... seem to me to accord better with the view of occasional means of transport having been largely efficient in the long course of time, than with the view of all our oceanic islands having been formerly connected by continuous land with the nearest continent; for on this latter view the migration would probably have been more complete.¹

A land bridge lets everything walk across. Filtered dispersal lets some things across. What is observed is a filtered sample. The argument is 167 years old and it has not been improved on.

1.5 Wallace turns it into a classification

Alfred Russel Wallace's Island Life, published in 1880, converts Darwin's rule into a three-way scheme with depth thresholds, and it is startling how well the categories still work.

Recent continental islands sit on a submerged bank where "the depth of the intervening sea rarely exceeds 100 fathoms," which is 183 m. Their plants and animals "are either almost identical with those of the continent, or if otherwise, the difference consists in the presence of closely allied species of the same types."⁹

Ancient continental islands are "separated from it by a depth of sea of a thousand fathoms or upwards," which is 1,829 m. They have mammals and amphibians, but "these are highly peculiar, almost all being distinct species," and the fauna is "fragmentary," with "many of the most characteristic continental orders or families being quite unrepresented."⁹ That word, fragmentary, is the disharmony argument.

Oceanic islands are the ones where "the total absence of warm-blooded terrestrial animals in an island otherwise well suited to maintain them" shows the island "has been actually produced in mid-ocean."⁹

Wallace is not deriving these from theory. He is sorting islands by their faunas and then noticing that the sorting matches the soundings.

2. Measuring Darwin's rule

2.1 The right quantity

Arguments about island connection usually quote the depth of a named strait. That is the wrong number. An animal takes the easiest route available, not the shortest one, and a shallow detour beats a deep direct crossing.

The quantity that matters is the sill: over every possible route from A to B, the route whose deepest point is shallowest. Lower sea level by that amount and a continuous walkable path exists. Lower it by less and it does not.

That is the minimax path problem, and Dijkstra's algorithm solves it exactly with max() substituted for +. Bathymetry here comes from the GMRT Grid Server, which merges ship soundings with satellite gravity data. Regional runs use about 7.8 km per cell; the Bali-Lombok run uses about 61 m per cell, because coarse grids smooth narrow channels and make them look shallower than they are.

The reference sea level is the Last Glacial Maximum lowstand of 134 m below present, from Lambeck and colleagues' 2014 reconstruction in PNAS, verified verbatim from the paper: "The esl reaches its lowest value of ~134 m at the end of this interval."¹⁰

2.2 Results

Measured here.¹¹

2.3 What the table shows

Sort the rows by sill depth, which is how they are printed, and the fauna column sorts itself. That is the whole argument, and it did not have to come out that way.

The boundary falls exactly where the glacial lowstand puts it. Ten crossings sit under 134 m and every one has a continental fauna. Nine sit above it, at 226 m and deeper, and every one has a filtered fauna. Japan, at 126 m, is the one borderline case, and its fauna is correspondingly intermediate: continental in composition, heavily endemic at the species level, which is what a bridge that opened only at the deepest lowstands should produce.

Distance does not do this. Cuba is roughly ten times farther from Florida than Trinidad is from Venezuela, which is the wrong order of magnitude to explain a fauna difference that stark, and the sills are 718 m against 21 m.

Two independent checks that the method is not tuned. Paul Moody's 1970 textbook gives the continental-shelf criterion as 100 fathoms and names Britain-Europe and Japan-Asia as his examples;¹² measured here, 30 m and 126 m, both under his 183 m line. And Vences and colleagues describe the Comoros as "entirely volcanic and surrounded by sea depths of more than 3500 m";⁶ measured here, the Madagascar-to-Comoros sill is 3,051 m, the same order and independently derived.

Darwin's West Indies estimate deserves its own line. He said the bank was "nearly 1000 fathoms in depth," which is 1,829 m. Measured here, the Cuba-Yucatán sill is 1,661 m. He was working from lead-line soundings and he was within 10%.

3. The vicariance paradigm and its collapse

The next part of the story is one that creationist writers cite and mainstream textbooks tend to skip.

3.1 What vicariance promised

When plate tectonics was accepted in the 1960s, biogeography got a mechanism it had been waiting a century for. If a continent splits, the populations on it split too. No animal has to cross anything. Southern beech in Chile and New Zealand is not a dispersal problem; it is a piece of Gondwana that broke.

Combined with cladistics, this hardened into a methodological rule: vicariance is the default, and dispersal is the explanation of last resort. De Queiroz's summary of the period is that the validation of plate tectonics, "along with cladistic arguments for the primacy of vicariance, helped create a view of oceanic dispersal as a rare phenomenon and an explanation of last resort."²

3.2 What molecular dating did to it

Then divergence dates arrived, and they were repeatedly too young.

Southern beech. Nothofagus had "become iconic as a plant genus whose ancient Gondwanan relationships reach back into the Cretaceous era," in Michael Knapp and colleagues' words, and the seeds are poorly suited to crossing oceans, so vicariance seemed forced.⁴ Their relaxed-clock analysis of 7.2 kb of chloroplast DNA found that some Nothofagus distributions are consistent with vicariance and that "trans-Tasman Sea distributions can only be explained by long-distance dispersal."⁴ The pollen record agrees: Lophozonia and Fuscospora pollen are absent from the New Zealand Cretaceous.

Chameleons. Raxworthy, Forstner and Nussbaum tested the Gondwanan-vicariance hypothesis on 52 chameleon taxa with mitochondrial, morphological and life-history characters. The tree put the two most basal lineages in Madagascar and required dispersal outward, not fragmentation. Their title states the result: chameleon radiation by oceanic dispersal.⁵

Frogs, which are the hardest case. Amphibians die in sea water. That is why Darwin used them, and why their distributions were treated as pure vicariance. Miguel Vences and colleagues then found two endemic mantellid frog species on Mayotte in the Comoros, an island chain that is "entirely volcanic and surrounded by sea depths of more than 3500 m," and which "did not have any past physical contact to other land masses."⁶ The frogs are genuinely distinct species, not introductions. Using them to recalibrate, the same paper identified two further Cenozoic crossings previously read as vicariance. Their conclusion: "overseas dispersal of amphibians exists and is no rare exception, although vicariance certainly retains much of its importance."⁶

Baobabs. Africa and Australia separated about 120 million years ago. The molecular split between their baobabs dates to 5 to 23 million years. Even the oldest estimate is off by a factor of five.²

Madagascar's mammals. If Madagascar's four endemic mammal radiations were Gondwanan leftovers, they would share one arrival date. Poux and colleagues dated them from three nuclear genes and got four different windows: lemurs 60 to 50 million years ago, tenrecs 42 to 25, carnivorans 26 to 19, rodents 24 to 20.¹³ Four separate arrivals from Africa. Jason Ali and Matthew Huber then modelled the currents and argued in Nature that they rafted, stating plainly that the evidence is "inconsistent with the land-bridge hypothesis."¹⁴

Monkeys and rodents in South America. South America was an island continent for most of the Cenozoic, and its monkeys and guinea-pig relatives are immigrants. Pierre-Olivier Antoine and colleagues described South America's oldest rodents from about 41 million years ago in Peru and concluded that the phylogeny supports "a trans-Atlantic dispersal of these mammals during Middle Eocene times."¹⁵ Their 2012 paper noted that this widened the gap between the rodent and primate arrival dates to about 15 million years, which was awkward. Three years later Bond and colleagues described Eocene primates from the same region of Peru that resemble Eocene African anthropoids, and the gap closed.¹⁶ Two independent mammal groups crossing the Atlantic, dated by fossils rather than clocks, arriving together.

3.2b The Atlantic crossings, counted

Those two are not the whole list, and the length of the list is the point. The Atlantic between Africa and South America has to be crossed repeatedly, by unrelated groups, at different dates, for the modern South American fauna to exist. This is the single most-attacked inference in the field and it deserves to be laid out with a number on it.

Nicolas Vidal and colleagues gave the count in 2008 while adding to it: "Until now, only four or five transatlantic dispersal events were known in terrestrial vertebrates."¹⁷ Their own paper made it five or six. The roll-call, as Pereira and Schrago summarise it a decade later, runs to hystricognath rodents, platyrrhine primates, amphisbaenians, testudinid tortoises, emballonurid bats and mabuyine skinks, the last of these on "two independent transoceanic dispersal events from the Old World."¹⁸

The addition Vidal's team made is the awkward one. Amphisbaenians are burrowing, legless, worm-like reptiles. They crossed "in the Eocene (40 Myr ago) resulting in a tropical American radiation representing one-half of all known amphisbaenian species," and the authors flag exactly what is odd about it: "this is the first such dispersal event to involve a group that burrows, an unexpected lifestyle for an oceanic disperser."¹⁷

How wide was that ocean? Narrower than it is now, and this matters enough to get right. Vidal's paper says the continents were then "between 3500 and 6000 km" apart,¹⁷ and that cannot be the crossing distance: the Atlantic has widened continuously since the Cretaceous, the narrowest modern gap between Brazil and West Africa is about 2,900 km, and equatorial spreading accounts for another 800 to 1,600 km of that over 40 million years. The figure usually attached to the monkey crossing is about 1,450 km, at the Eocene-Oligocene sea-level lowstand, and Seiffert and colleagues place the dispersal window at 35.1 to 31.7 Ma precisely because it "includes the major eustatic drop in sea level that occurred in the earliest Oligocene."¹⁹ The mechanism proposed is a vegetation raft "aided by favourable marine currents (North Equatorial Current) and palaeowinds."¹⁷

Four to six crossings of roughly fifteen hundred kilometres of open ocean, by rodents, monkeys, tortoises, worm-lizards and skinks. Section 6.1 gives the objection to this the space it deserves, because the objection is a good one and the best version of it comes from inside the field.

3.2a What replaced it at Wallace's Line

The vicariance paradigm's collapse did not leave a vacuum. The current mainstream account of the Indo-Australian break is Skeels and colleagues' 2023 paper in Science, and it is worth stating because it is the thing a creationist model has to beat.

They took spatial and phylogenetic information on 20,433 species belonging to all 227 families of terrestrial vertebrates in the archipelago and ran it through a model of geoclimate and diversification.²⁰ Three results matter here.

The line is not a wall, it is a filter with a measurable leak rate. They estimated "a minimum of 381 independent colonization events across Wallace's Line."²⁰ This is a reconstruction from phylogenies, conditional on their trees and method, rather than 381 directly observed crossings.

The traffic is lopsided, and the asymmetry is the thing to explain.  In their introduction, Skeels and colleagues report earlier work estimating exchange "at least twice as high from Sunda to Sahul than in the opposite direction."²⁰  That background estimate is distinct from their own reconstruction of 381 colonization events.

The explanation is climate, not distance. Southeast Asian lineages evolved in wet conditions like the humid "stepping stones" of Wallacea, which let them island-hop east. Australian lineages evolved in dry conditions and could not establish in wet Sunda. In their words: "Sundanian (Southeast Asian) lineages evolved in a climate similar to the humid 'stepping stones' of Wallacea, facilitating colonization of the Sahulian (Australian) continental shelf. By contrast, Sahulian lineages predominantly evolved in drier conditions, hampering establishment in Sunda and shaping faunal distinctiveness."²⁰

This is a real explanation with a real weakness, and Michael Oard names the weakness accurately in his 2025 critique: the model needs southern Wallacea to have been drier in the past than it is now, and that is an assumption of the reconstruction rather than an observation.²¹ He is right that it is an assumption. He does not show it is a wrong one.

3.3 Why this is a good outcome and not an embarrassment

A field that had made vicariance a methodological rule went looking for confirmation, found the opposite, and published it. The paper that names the shift, de Queiroz's, calls for the assumption itself to be "re-evaluated and perhaps discarded."²

Creationist writers use this history as evidence that biogeography is unreliable. It is worth noticing what it actually shows. The vicariance paradigm made a prediction, the prediction was tested by a method it did not control, and it lost. That is the mechanism by which a field is supposed to work, and it is the mechanism that the alternative accounts of biogeography have not yet subjected themselves to.

4. The Hawaiian test, which is the cleanest one available

The Pacific Plate moves north-west over a fixed hotspot, so the Hawaiian islands were built one at a time in a line. Potassium-argon dating gives Kauai 5.1 million years and Hawaii 0.43 million.⁷ The ages come from radioactive decay in the lavas and have nothing to do with biology.

That sets up a prediction that nothing else in biogeography matches for cleanness. If a lineage colonised the chain as it was built, then its family tree should branch in island order, and the genetic distance between sister forms on adjacent islands should scale with the age difference between those islands.

Robert Fleischer, Carl McIntosh and Cheryl Tarr tested it in three unrelated groups: honeycreepers, Drosophila and Laupala crickets. They got rate calibrations of 0.016 sequence divergence per million years for honeycreeper cytochrome b, 0.019 for the Yp1 gene in Hawaiian Drosophila, and 0.024 to 0.102 for crickets.⁷ Three groups, three genes, one plate.

Their paper is unusually honest about what the test assumes, and lists seven conditions by name, including that lineages did not colonise long after an island emerged and that the coalescent point for sister taxa does not much predate the younger island.⁷ Those assumptions can fail, and where they fail the rates come out too slow.

The reason this test matters is that the prediction is not adjustable. The island order was fixed by a hotspot before anyone looked at a fly. Any account in which Hawaiian Drosophila did not descend from earlier Hawaiian Drosophila has to explain why their tree matches a geological conveyor belt.

One objection to it deserves more room than it usually gets. Michael Heads, a panbiogeographer and a persistent critic of molecular dating, argues that island-age calibration has a structural flaw: "Studies correlating the age of nodes with age of volcanic islands often overlook the fact that these islands have been produced at subduction zones or hot spots where small, individually ephemeral islands are constantly being produced and disappearing, and a metapopulation can survive indefinitely."²² The point is that the current islands have known ages, but the hotspot has been making and drowning islands for far longer, and a lineage can hop from a sinking island to a rising one without ever being as young as the rock it now sits on. If that happened, the calibration is a minimum age read as an absolute one, and every rate derived from it is too fast.

This cannot be dismissed. It is also testable in principle, because the drowned Emperor seamounts are dated, and it does not touch the branching order, which is the part of the Hawaiian argument that carries the evidential weight. A lineage that hopped along older, now-drowned islands would still branch in island order. It would just be older than the calibration says.

5. Marsupials: living relationships and the fossil record

5.1 What the living tree resolves

The relationships among marsupial orders resisted resolution for decades. Sequence data kept giving conflicting answers. Maria Nilsson and colleagues attacked it with retroposon insertions instead, screening about 217,000 loci from opossum and kangaroo and recovering 53 informative markers. Retroposons land essentially at random and are almost never removed cleanly, which is why the authors call them "virtually homoplasy-free."⁸

The answer: the four Australasian orders share a single origin, with the South American Dromiciops group as their closest relative, and the South American opossums branch first. Put in geography, the pattern "indicates a single marsupial migration from South America to Australia."⁸

One migration, on a route through Antarctica when the three southern continents were joined. That is a strong, specific, falsifiable claim.

And the fossils do not agree with it. The current review of the field is blunt about this: "It is also clear there was not simply a single dispersal of marsupials from South America to Australia (the scenario implied by phylogenies of extant marsupials only); instead, the fossil evidence from Tingamarra, South America and Antarctica suggests a much more complex biogeographical relationship between these landmasses, characterised either by multiple dispersals, differential extinctions, or a combination of these."²³ Beck and colleagues push further, and argue from Djarthia's primitive anatomy that the South American microbiotheres "may be the result of back-dispersal from eastern Gondwana, which is the reverse of prevailing hypotheses."²⁴

So the honest statement is narrower than the headline. Retroposons resolve the tree of living marsupials to a degree sequence data could not. They do not settle how many crossings there were or which way they ran, because the lineages that would answer that are extinct. This is worth stating plainly because the single-migration line is repeated in popular accounts as though it were settled, and the review literature says it is not.

5.2 The unexplained fact

Now the census. Every metatherian occurrence in the Paleobiology Database was pulled for this article and binned by epoch and continent.²⁵

Measured here.²⁵

Read the Cretaceous row. The entire early record of the group is in North America, with a scatter in Asia and Europe. South America, Australia and Africa have nothing. Then the centre of gravity moves south and the northern record dies out.

Statham quotes Richard Cifelli and Brian Davis's 2003 description of a "geographical switch" that "remains unexplained."²⁶  That is a historical assessment, rather than a new test of the present census. There are partial answers. Most of the Cretaceous northern forms are stem metatherians, not crown marsupials, so the switch is less abrupt than the raw row suggests. North American Cretaceous mammal beds are far better sampled than South American ones, which inflates the top-left cell. Both are real, and neither turns 773 against zero into a wash.

5.3 The Australian record has a 30-million-year hole in it

Australia's own record is thin in exactly the interval where the radiation is supposed to have begun. In this census the Eocene gives one named Australian genus, Djarthia, from the Tingamarra fauna at Murgon. The Oligocene gives 38. Robin Beck and colleagues, describing the new Djarthia material, call it "the oldest known Australian marsupial by 30 million years."²⁴

Beck's team also read the anatomy in a direction that cuts against Nilsson's: Djarthia is the most primitive australidelphian known, which "suggests that the South American microbiotheres may be the result of back-dispersal from eastern Gondwana, which is the reverse of prevailing hypotheses."²⁴ So the fossils and the retroposons currently point in opposite directions about which way the single migration ran. Nobody should pretend that is settled.

6. Where the evolutionary account is weak

An article that only lists the successes is advocacy. Here is the other column.

6.1 Dispersal explanations are cheap

This is the serious objection, and it is best stated by the field's own dispersal advocate. Summarising why the improbable cases should be believed, de Queiroz writes: "given enough time, many things that seem unlikely can happen."²

That is true, and it is also a licence to explain anything. A frog on a volcanic island 3,500 m from the nearest land is explained by a raft. A monkey across the Atlantic is explained by a raft. When vicariance fails, dispersal is available, and dispersal has no upper bound on improbability as long as the clock has enough time in it. A model that can absorb any observation is not being tested by that observation.

The critics put it more sharply. Casey Luskin, a geologist at the Discovery Institute, asks the obvious questions of the Atlantic crossings: how did the monkeys not starve, how did they not die of thirst on a raft they could not drink from, and what did the founding population's genetic diversity look like afterwards.²⁷ He compares the accumulating dispersal explanations to Ptolemaic epicycles, patches added to save a theory rather than predictions made by it. Günter Bechly gives the version that stings most: "It looks like there was some very busy ocean travel going on in those times, which suddenly stopped as soon as humans could have observed and recorded it."²⁸

Bechly's objection is half right, and the half that is right is the important half.

It is wrong as stated. The traffic did not stop, and it has been observed. Fishermen on Anguilla watched green iguanas come ashore on a raft of hurricane debris in 1995.²⁶ Carlton and colleagues recovered 289 living Japanese coastal species off tsunami debris that crossed the Pacific between 2012 and 2017.²⁹ Those are records, not reconstructions.

It is right about the thing that matters. No mammal has ever been observed crossing an ocean on a raft. Every documented case involves invertebrates, or reptiles, or plants. The four to six transatlantic crossings in section 3.2b include rodents, primates and bats, and for those the mechanism rests entirely on inference. A monkey needs fresh water. The observed record of oceanic rafting is a record of animals that do not.

The honest scorecard, then. Rafting as a mechanism: observed, quantified, not in doubt. Rafting by small ectotherms and invertebrates: observed directly. Rafting by mammals across roughly 1,450 km: inferred from phylogeny and never witnessed, and the inference is doing real work that nothing else currently does.

Two further things blunt the general objection without removing it. Dispersal explanations do make checkable predictions, about direction, about currents, about which groups cross and which do not, and those predictions can fail. And the frog case was not a rescue after the fact: Vences and colleagues found new endemic species on a volcanic island, which is a discovery, not an interpretation.

The sharpest version of this objection is not from the Discovery Institute. It is in Science, from two primatologists.

Alfred Rosenberger and Mark Hubbe replied to the Peruvian monkey paper in May 2020. They accept the phylogeny and challenge the dispersal story: transatlantic rafting is "a scenario that cannot be empirically tested," and its difficulties "tend to be glossed over or minimized by the scenario's advocates, which results in overrating its viability and contributes to its widespread acceptance."³⁰

They then do the arithmetic nobody else does. Houle's rafting model lists twelve conditions a successful colonisation must clear, from the raft detaching with a viable breeding group aboard, through currents and storms, through food and fresh water against salt spray, to establishing and reproducing on arrival. Treat those as a chain: "if we assume that each step had a probability ranging between 0.01 (unlikely) to 0.001 (very unlikely), the overall chance of success would range between to 1x10^-24 (if all steps are assumed to have a probability of 0.01) to 1x10^-36," which they gloss as "roughly of the same order as selecting one specific cell of the body (~100 trillion cells) of a random human being on the planet (~7 billion people)."³⁰

Those input probabilities are stipulated, not measured, and the authors say so. The structural point stands regardless: "the odds of survival across voids of oceanic scale do not decrease linearly with distance, but rather exponentially."³⁰

Which is why the distance comparison decides this, and why it does not decide it the way an earlier draft of this article claimed. It is tempting to answer the critics by noting that creationists invoke rafting too. That answer fails, because the two sides are not asking for the same trip.

Measured for the companion article, on the same bathymetry as section 2: the longest single open-water leg on the easiest island-hopping route from the Sunda shelf to the Sahul shelf is 85 km at a 50 m drawdown and 110 km today.³¹ The Atlantic crossing is about 1,450 km. That is a factor of about seventeen, a bit over one order of magnitude. An earlier version of this article put it at forty to seventy by using Vidal's ocean-width figure as though it were a crossing distance. It is not, and the corrected ratio is still large enough to carry the argument.

The observed record sits on the short side of the line. The Anguilla iguanas covered 320 km. Carlton's tsunami species crossed the Pacific, but on plastic, which is that paper's central finding and does not help a Palaeogene raft. So a critic who accepts a hundred-kilometre crossing and doubts a fifteen-hundred-kilometre one is not being inconsistent. He is drawing the line where the evidence draws it.

What the mainstream account has, and the objection does not remove, is time. Four to six successful crossings spread over tens of millions of years is roughly one per ten million years, and a once-per-ten-million-year event is not required to be observable in a human lifetime. That is a real answer, and it is exactly the answer de Queiroz gives, which is why his sentence about "given enough time" is doing so much load-bearing work.

Rosenberger and Hubbe's own alternative is worth recording, because it shows the debate is live rather than settled: they argue that dispersal from North America deserves equal standing, its advantage being that "the entire enterprise is not premised on a long interlocking chain of unverifiable assumptions."³⁰ That hypothesis has its own problem, an absent fossil record, which they concede and argue is not fatal.

6.2 Clock circularity, in the words of the field's own critics

Most of section 3 rests on molecular dates. Those dates need calibration, and calibration usually comes from fossils or from geological events. Use a continental breakup to calibrate a clock, then use the clock to show that a distribution postdates that breakup, and the argument has eaten its own tail.

This is not a creationist complaint. It was made inside the field, by biogeographers, in the field's own journals, and it has never been fully answered.

Jonathan Waters and Dave Craw put it directly in Systematic Biology in 2006, in a paper about New Zealand titled "Goodbye Gondwana?": "When such distinct fields of scientific research intersect, however, there is potential for associated 'multidisciplinary' inferences to be clouded by circular reasoning."³² They name the failure mode. Geological and biological evidence should be assessed independently, and often is not: "it is more common to merely invoke a geological explanation a posteriori on the basis of biological data," and they list nine studies that do it.³²

Michael Heads goes further, and his 2005 review in Cladistics is the strongest secular statement of the problem. He takes apart all three standard calibration methods. On fossils: "Studies often, correctly, introduce oldest known fossils as providing minimum ages for divergences. However, in the actual analyses these ages, and ages derived from them, are often treated as absolute ages and earlier geological events are deemed irrelevant to the phylogeny. In fact, only younger geological events can be irrelevant."²² On palaeogeographic calibration, which he rates the best of the three: it "has often been used in a simplistic way, for example in assuming that all divergence across the Isthmus of Panama dates to its final rise."²² And on the relaxed clocks that made the dispersal revival possible: "New methods of estimating branch lengths do not assume a strict clock, but the number of models for molecular evolution is then effectively infinite."²²

His conclusion is the sharpest sentence available on this: "Problems with calibrating the nodes, as well as with substitution models, mean that phylogeography's claim to be able to test between vicariance and dispersal is not justified."²²

How much of section 3 does that take down? Less than it first appears, and more than nothing.

The results that do not depend on a clock survive untouched. Bond's Eocene primates and Antoine's Eocene rodents are fossils in dated rock. Vences's frogs are on an island that is volcanic all the way down, so no clock is needed to know they crossed water. The Comoros are 3,051 m from Madagascar by the measurement in section 2, and frogs die in seawater. Heads's objection cannot reach any of that.

The results that do depend on a clock are exactly the ones he is aiming at: the Nothofagus trans-Tasman date, the baobab date, the four Madagascar arrival windows. Those should be held more loosely than the papers announcing them usually suggest, and Poux's own abstract already does so, conceding that a land bridge "cannot be ruled out if credibility intervals are taken into account."¹³

Heads is not a neutral party. He is a panbiogeographer defending a research programme that molecular dating has been dismantling, and his alternative is less widely held than his criticism. His criticism is still correct.

6.3 New Zealand

The worst open problem in the field. Landis and colleagues examined the Waipounamu Erosion Surface and concluded that they "can find no geological evidence to indicate that land areas were continuously present throughout the middle Cenozoic," and that New Zealand's biota therefore "evolved largely from fortuitous arrivals during the past 22 million years."³

Take that seriously and the tuatara, the last survivor of an order otherwise extinct since the Cretaceous, crossed 4,621 m of ocean. So did Leiopelma, a frog family with no close living relatives. So did the flightless ancestors of the moa. Each is individually improbable and the conjunction is worse.

The alternative is that the geologists are wrong and some land persisted. That is the more popular reading now, and Waters and Craw's paper is the case for reading the evidence carefully in both directions rather than letting either discipline's default settle it.³² The point stands that the field cannot presently decide, and that it is arguing about whether an entire continental biota is 80 million years old or 22.

6.4 The Cretaceous marsupial switch

Covered in section 5.2. It is a real anomaly, it is thirty years old, and it has no accepted explanation.

6.5 Areas of endemism

De Candolle's twenty regions of concentrated plant endemism, identified in 1820 and later found to match animal endemism, are not an obvious prediction of any dispersal-plus-diversification model. Croizat spent a career on the pattern and got nowhere with the mainstream. Wise and Croxton's 2003 creationist paper puts this at the centre of its case, and the complaint is not answered simply by pointing out that their own explanation fails.

7. Objections and responses

"Biogeography does not prove evolution. Nelson and Platnick said so." They said something narrower. Their 1981 sentence is about whether cladograms can be shown to be genealogical trees, the evidence for which they said was scarce in 1981 and "worthy of further investigation." The full passage and the audit are in the companion article. It is also true in a broader sense: biogeography on its own shows filtered dispersal and local descent. Universal common ancestry needs the rest of the evidence.

"A designer could have placed each species where it best fits." Darwin met this one and the reply is the same. Frogs thrive on oceanic islands once introduced, to the point of becoming pests. Their absence is not about fit. And a placement model has to explain why the placements track sill depth, a quantity nothing in an organism can sense.

"Convergence between marsupials and placentals shows the similarity is not from ancestry." Convergence is real and the thylacine skull is a striking case. It is also the wrong lever. Convergence explains resemblances between separately-evolving lineages under similar selection, which is what the theory predicts on two isolated continents. What needs explaining is not that a marsupial and a placental both became wolf-shaped. It is that all the wolf-shaped marsupials are on one continent.

"The vicariance failures show the whole field is guesswork." The failures were found by biogeographers, using biogeography's own methods, and published in the field's own journals. A discipline that publicly discards its dominant paradigm on the evidence is behaving better, not worse.

"Sill depths today do not tell you about sill depths in the past." Correct, and it limits what the measurement can do. Over hundreds of millions of years, sills move. This is why the sill table is presented here as a test of Darwin's rule about present faunas and present bathymetry, which is what he claimed, and not as a reconstruction of deep history. The measurement is far more constraining for a model that needs everything to have happened in the last five thousand years, which is the subject of the companion article.

8. Verdict

Biogeography is the strongest evidence in biology that is visible without a microscope, and the reason is that it makes a prediction nothing else does. Life is distributed as though it spread from where it already was, filtered by whatever the barriers happened to be.

The measurement in section 2 is the compact form of that. Darwin wrote in 1859 that the depth of the sea between an island and the mainland would predict its mammals. In 2026 the seafloor is mapped and the claim can be checked against bathymetry, subject to grid resolution. Across nineteen crossings, it holds every time, with no exception and no overlap between the shallow group and the deep group. The dividing line falls at the glacial lowstand, which is where the theory says it should fall and nowhere an author could have put it.

The twentieth-century elaboration of the theory did much worse. Vicariance was oversold, and the field said so.

What remains uncomfortable is real. New Zealand's biota may be 22 million years old or 80 and the geology cannot presently decide. The oldest marsupials are on the wrong continent. Dispersal explanations are cheap enough that a careful reader should always ask what would have counted as a failure.

None of that touches the central pattern, and the central pattern is not subtle. Frogs are missing from oceanic islands and thrive there when carried. Bats are on every island and nothing else is. The Wallace Line falls at 226 m and not at a distance. Sri Lanka has elephants and Cuba does not.

Sources