Lake Varves

"Our investigations supported de Geer's first contention that sediment-laden flood-waters could generate turbidity underflows to deposit varves, but threw doubt on his second interpretation that varves or varve-like sediment are necessarily annual!" — André Lambert and Kenneth Hsü, Sedimentology, 1979¹

"On average as many as 50% of the varves are indistinguishable." — Gordon Schlolaut and colleagues on the Lake Suigetsu core, 2012²

"a cumulative posterior distribution of counting uncertainties of −13/+7 % indicative of systematic observer undercounting" — Stephanie Arcusa and colleagues, Columbine Lake, Colorado, 2021³

Summary

In a still, deep, cold lake, the sediment that settles in spring is not the same as the sediment that settles in autumn.  Spring brings meltwater silt or a bloom of diatoms;  winter brings fine clay settling through undisturbed water.  If nothing stirs the bottom — no waves, no burrowing worms, which usually means no oxygen down there — the two settle as a couplet, and the couplets stack up like pages.  A couplet that represents one year is a varve.

Counting varves gave the world its first "absolute" chronology, in Sweden, a century before radiometric dating was trusted.  It is also the counting clock with the worst-documented failure modes, and the young-earth critique of it is the strongest young-earth critique of any counting clock.

Eight conclusions, stated first.

1. More than one couplet per year is real and well documented. In Lake Walensee, Switzerland, 300 to 360 laminae accumulated in 165 years — generally two a year, and up to five at some locations — deposited by turbidity currents from snowmelt and rainstorms.¹ In a New Zealand glacier-fed lake, three couplets a year.¹ In Muir Inlet, Alaska, tides and daily meltwater surges produce two couplets a day.¹ These are not fringe results; they are the mainstream sedimentological literature, and the creationist authors who cite them cite them accurately.
2. Every one of those cases is in a setting that varve chronologists exclude. All of them are glacier-fed or tidewater systems with continuous, current-driven sediment delivery. The literature the same creationist authors quote says so directly: in such settings "there is year round delivery of sediments and turbidite activity occurs almost continually resulting in graded laminae that are not true varves."¹ The lakes used for chronology are chosen to be deep, anoxic and non-glacial, where the couplet is a seasonal biological signal rather than a flood pulse.
3. The measured error in real varve chronologies runs in the opposite direction to the young-earth requirement. A review of the field puts most chronologies at ±1–4% counting uncertainty, rising to ±15% for indistinctly varved sequences.³ When the Columbine Lake, Colorado sequence was counted by three independent observers and modelled with a Bayesian framework that allows for both over- and under-counting, the posterior came out at −13/+7%, "indicative of systematic observer undercounting."³ Counters miss layers more often than they invent them.
4. Lake Suigetsu is the load-bearing record, and it is half interpolated. Its counters state that "on average as many as 50% of the varves are indistinguishable" over part of the record, and are filled in by an interpolation program.² This is the single strongest factual point in the young-earth case, it comes from the counters themselves, and it must be printed.
5. Yet the direction of the resulting error is still undercounting. Comparing the Suigetsu varve count against a uranium-thorium-anchored chronology at 593 dated levels gives a mean residual of −32 years at 10–15 ka, growing to +2,744 years at 47,762 varve years — the count falls short of the decay clock by 5–7% in the deep part, not long.⁴ A record that is inflating years does not run short.
6. The decisive test needs no varve count at all. Measure the raw carbon-14 in leaves picked out of individual laminae, in percent modern carbon, and plot it against lamina number. If many couplets formed in a burst, they share almost the same carbon-14, and the plot shows a staircase — long flat treads. There is no staircase. The points descend smoothly from the surface to about 50,000 laminae and lie on top of the independent tree-ring curve through their overlap.⁵
7. The Green River Formation is the one place where the young-earth objection may bite hardest, and it is unresolved here. Bradley's 1929 lamina count gave the Eocene "a little less than 23,000,000 years," within about 4% of the modern radiometric figure, and obtained "entirely independent of" it.⁶ Against that stands a 1988 conference abstract reporting that laminae counts between the same two tuff beds range from 1,160 to 1,568.⁷ The abstract could not be obtained, but the fullest secondary account — by a creationist geologist — adds the detail that decides how to read it: the count rises toward the basin margin, alongside a near-threefold rise in the thickness of sediment between the same two ash beds.⁸ That is a sediment-supply gradient, not a random miscount, and its natural reading is that shoreline flood beds are interleaved with annual ones near the margin. It is a real problem for treating every Green River lamina as a year, and it is not a demonstration that none of them is.
8. The Swedish time scale is the weakest link and everyone knows it. De Geer's chronology was built by matching short sections graphically, has been under revision since the 1940s, and one of its own practitioners wrote that better-looking correlations "can be obtained between limited parts of varve series in a way that is indisputably wrong."⁹ Nobody should rest an argument on it, in either direction.

The thing that is actually surprising

The best modern paper on varve counting error is not a defence of the method.  It is a 2021 study that treats observer subjectivity as a first-class problem, assigns each stretch of core a "varve quality index," has three people count the same thin sections independently, and then reports a systematic bias — and the bias is that human counters miss layers.  The field's own error analysis has arrived at the opposite of what the young-earth argument requires.

Part 1: What a varve is, and when a couplet is not one

The mechanism

Three conditions have to hold for a lake to record years:

1. Seasonal contrast in what settles. Coarse silt in the melt season and clay in winter (a clastic varve); or a diatom bloom in spring and clay in winter (a biogenic varve); or a carbonate precipitation season (a calcareous varve).
2. Still water. Deep enough that waves and currents do not stir the bottom.
3. No burrowers. Which in practice means anoxic bottom water, because worms and clams destroy laminae within weeks.

Lake Suigetsu satisfies all three unusually well.  It is deep, its bottom water is anoxic — probably since a channel cut in 1664 admitted brackish water from the Sea of Japan — and, critically, the coarse sediment load is intercepted before it arrives.  River water enters neighbouring Lake Mikata first, drops its sand there, and passes into Suigetsu carrying only fines.⁵

The Suigetsu couplet is diatom-and-clay:  a white layer of diatom frustules from the spring bloom, a dark grey clay layer from autumn and winter, plus an iron-carbonate (siderite) layer that is "the most frequently and consistently occurring type of seasonal layer" and on which the counting is mainly based.²

When a couplet is not a year

Oard's 1992 survey of the failure modes is the best single collection of them in print on either side, and it is worth reproducing as a list because it is largely correct:

• Turbidity currents. A sediment-laden underflow deposits a graded couplet — coarse below, fine above — that looks exactly like a summer/winter pair. One storm, one couplet. "Sandstone and mud sequences laid down by turbidity currents far removed from a glacial environment may be easily confused with varved sequences."¹
• Winter coarse layers. Glaciers keep discharging in winter, so a coarse layer is not necessarily summer. Misreading them "leads to an over-estimate of the length of time represented by a particular glaciolacustrine sequence. Our review shows such misinterpretation to be common."¹
• Delta slumping. A river delta oversteepened in summer collapses in winter, producing a coarse layer at the wrong time of year.
• Tides and diurnal meltwater. In Muir Inlet, Alaska, two couplets a day, each about half a centimetre, "superficially look[ing] like varves."¹
• Seiches. Standing waves set off by wind or earthquakes.
• Rapid reservoir filling. Wood (1947) recorded three varve-like couplets deposited in a new reservoir in two weeks, by three peak inflows from light showers.¹

And the diagnostic criteria that are supposed to separate real varves from these — a sharp break between coarse and fine, biogenic disturbance in the clay, uniform clay thickness across the basin — are, as Oard correctly says, hard to apply and not decisive on their own.¹

All of this is right.  What follows from it is that varve counting requires a carefully chosen basin and external verification.  What does not follow is that varve counts are worthless, and the reason is in Parts 3 and 4.

Part 2: The young-earth case, at full strength

Oard 1992: varve chronology is not scientifically sound

Michael Oard's two-part Creation Research Society Quarterly paper is the fullest young-earth treatment.  Its argument runs:

Part I:  the couplets need not be annual.  The list above, plus the observation that ice sheets melt far faster than varve chronologies imply — 10 m/yr or more in the ablation zone of real glaciers, up to 55 m/yr at the Jakobshavn Glacier — so a Scandinavian ice sheet ought to disappear in a few hundred years, not the four thousand the Swedish varve chronology gives.¹

Part II:  the correlations are subjective.  The Swedish chronology is not one long core;  it is hundreds of short sections, "most often represented by less than 200 couplets," strung together by matching thickness curves.⁹  Antevs's own description of the procedure is quoted, and it is uncomfortable reading:

"All individual curves were first matched and corrected for number of varves.  If, for example, out of three measurements two agreed, but one had one varve less or more than the others, the exact location of the mistake was determined and the curve corrected by dividing one varve in two or uniting two varves in one, so that this curve agreed with the two others."⁹

Oard's comment — "it can easily be used to adjust the number of varves and the thickness of each couplet to enhance the correlation with other varve sections" — is fair.

And Lundqvist, a Swedish varve specialist, is quoted saying that a section can be "'indisputably' connected with two or even more different parts of another series," and that "connections which give a completely impossible picture of the deglaciation can be made in this way."⁹

Hebert, Snelling and Clarey 2016: Suigetsu specifically

The Answers Research Journal paper adds four things:

1. A mechanism for many couplets a year at Suigetsu. Diatom blooms need dissolved silica. Wind-blown dust and volcanic ash supply silica. Ice-age Greenland dust concentrations were 3–100 times modern. Dust from eastern Asia crossing Japan "likely occurred in pulses associated with strong, dry cold fronts. Each pulse of dust falling into the lake could have caused a diatom bloom. Consequently, dozens of diatom/clay couplets could have occurred each year."²
2. The interpolation. Quoting Schlolaut et al. 2012 directly: "Usually not all of the sub-layers formed every year and in a considerable portion of years no seasonal layers formed at all, in which case non-seasonal, mixed layers formed… the microscope count of the LGIT is necessarily incomplete." And: "on average as many as 50% of the varves are indistinguishable."²
3. The floating chronology. The older Suigetsu chronology did not reach the surface; its start date of 8,830 cal BP was obtained "from the best overall match between the floating varve chronology and an absolute chronology obtained from German oak and German pine tree-rings. It was not obtained by simple counting."² Hebert's rhetorical question is a good one: "How does one 'simply' count to 8830, if you have 'skipped' the first 8829 integers?"
4. The count itself is smaller than advertised. The widely repeated "100,000 Suigetsu varves" was an estimate from sedimentation rates, not a count; the number actually claimed as counted in the 1998 papers was 29,100.²

Points 2, 3 and 4 are correct and important.  Anyone presenting Suigetsu as a straightforward count of 50,000 years is overstating it, and this article says so.  Point 1 is a real mechanism whose magnitude has never been estimated.

Part 3: What the checks actually show

Suigetsu against uranium-thorium

The 2020 reanalysis of the Suigetsu record publishes, for every one of 593 dated levels, both the varve age and the calendar age obtained by matching the record's carbon-14 to the uranium-thorium-dated Hulu Cave speleothems.  Comparing the two columns directly:⁴

Read that both ways, because it says two things.

Down to 15,000 years the counting clock and a completely different decay clock agree to within a few decades.  At the shallowest levels the residuals are −4, −3, −6, −9 and −16 years.  Two methods with nothing physically in common, landing on the same number.

Below that the count falls behind by 5–7.5%, and the shortfall is real.  Only 16% of the 593 levels have their varve age inside the 2σ band of the uranium-thorium age.  The paper localises the problem to two intervals and offers a mechanism it labels unproven:  thin varves under-represented where sedimentation was slow, which then corrupts the interpolation.⁴

The sign is what matters for this argument.  The count is short.  A record generated by counting storm pulses as years would be long.

Suigetsu against a cave 1,600 km away

The other half of the comparison is whether the atmospheric carbon-14 recorded in Japanese lake leaves matches the atmospheric carbon-14 reconstructed from Chinese cave calcite.  The reanalysis reports offsets of 21 ± 19, 12 ± 17, 5 ± 16, 8 ± 19 and −1 ± 32 years across five model variants, and concludes that all are "within error limits, consistent with a zero systematic offset."⁴

A lake in Japan and a cave in China, in different materials, dated by different physics, agreeing on the carbon-14 content of the atmosphere to within about a dozen years.

The staircase that is not there

This is the argument that does not require trusting the varve count at all, and it is the strongest thing in this article.

Suppose the young-earth model is right and, in the early post-Flood centuries, hundreds of couplets formed each year.  Then hundreds of consecutive laminae contain leaves that died within months of each other, and therefore hold almost identical carbon-14.  Plot measured carbon-14 against lamina number and you get a staircase:  long horizontal treads where a year's worth of couplets sits at one carbon-14 value, then a drop to the next.

Davidson and Wolgemuth ran exactly this test on the Suigetsu measurements — over 800 of them, from the surface to about 41 m depth.  There is no staircase.  The carbon-14 declines smoothly and continuously to background.  Nor is there the sharp change in sediment character that a Flood/post-Flood boundary would produce:  "Not only do the varves appear the same above and below, but the frequency and thickness of interspersed ash, flood, and earthquake-induced turbidite deposits also vary little above and below."⁵

And the required rate is not modest.  Davidson and Wolgemuth compute it:  to fit the observed carbon-14 profile into a few thousand years would require, in the early post-Flood period, "over 1,000 couplets per year—on the order of 3 per day," each one settling out as a distinct, unmixed couplet hours after the last.⁵

Oard's "dozens per year" mechanism, taken at its own valuation, is short of that by a factor of about thirty.

Turbidites are identified and removed

One reasonable worry is that flood and earthquake layers get counted as varves.  In the modern Suigetsu work they do not:  event layers are identified — 369 of them, of which 362 are attributed to floods — and the depth scale used for comparison is an "event free depth" with the thickness of ash, flood and turbidite deposits subtracted.⁵  The young-earth critique is right that most of those events are inferred rather than historically documented;  it is the physical character of the deposit that identifies them, not a written record.

Part 4: How the field measures its own counting error

The best available answer to "counting is subjective" is not a denial.  It is the 2021 Columbine Lake study, which treats subjectivity as the thing to be measured.³

Their procedure:  three observers count the same thin sections independently;  each stretch of core gets a quantitative varve quality index;  indistinct stretches are coded and assigned their own prior distributions for over- and under-counting;  multiple cores are cross-correlated on marker layers;  and the whole thing is combined with lead-210, caesium-137 and radiocarbon measurements in a Bayesian model.

Their catalogue of error sources is the honest one, and it includes over-counting explicitly:

"(1) inter-site differences in varve counts (missing varves), (2) subjectivity in identifying varves due to varve quality, (3) expert judgement in identifying marker layers, (4) compound single varves that are misinterpreted as representing multiple years (over counting), (5) indistinct varves that are combined with adjacent varves (under counting), (6) intermittent (floating) varves, (7) technical issues (missing varves), and (8) counting strategies."³

Their result for a deliberately difficult, indistinctly varved alpine lake:  3,137 varve years, with a posterior counting uncertainty of −13/+7%, "indicative of systematic observer undercounting."³

And their summary of the field:  "most are associated with ±1–4 % counting uncertainty with some indistinctly varved sequences having counting errors up to ±15 %."³

Set that against the requirement.  Fitting 50,000 Suigetsu laminae into 4,300 years needs an over-count of about 1,060%.  The worst-behaved varve sequences in the literature, deliberately chosen for their difficulty, come in at ±15%, and the bias runs the other way.

Part 5: The Green River Formation

The Green River Formation of Wyoming, Utah and Colorado is the classic ancient varve deposit:  roughly 2,000 feet of laminated oil shale over more than 25,000 square miles, laid down in a system of Eocene lakes.  Wilmot Bradley measured the laminae in 1929 and found them averaging 0.18 mm thick;  dividing thickness into the section gave the Green River epoch 5 to 8 million years.⁶

What makes Bradley's paper worth reading is that he then did something a modern reader would not expect.  He extended the estimate to the whole Eocene by fluviatile accumulation rates, got "a little less than 23,000,000 years," and observed that this "agrees rather closely with estimates of the duration of the Eocene epoch based on the age determinations of radioactive minerals, but it is entirely independent of them."⁶  The Eocene on the current international scale runs 56.0 to 33.9 Ma — 22.1 Myr.  Bradley, counting mud in 1929, was within about 4%.

Three objections, of increasing force.

The fossils.  Green River fish, birds and bats are preserved articulated and undecayed.  Hebert and colleagues argue that "because these fish were preserved, the thin layers must have formed quickly around them, before the fish could decay or be eaten."²  This is a real observation and it is right about the local burial event — a fish is buried by a smothering pulse, not by 0.18 mm a year.  It does not follow that the intervening 2,000 feet is also event beds;  exceptional preservation is exceptional, which is why articulated Green River fish are collectable and most of the formation is barren shale.

Laminae can form fast.  Guy Berthault's flume experiments produced spontaneous lamination in graded sediment mixtures, and comparable results were published in Nature in 1997.²  Later flume work by Juergen Schieber showed that mudstones — long assumed to require still water — can be deposited by currents at 0.3 m/s.²  Both results are genuine and both are correctly cited.  They establish that lamination does not require slow settling.  They do not establish that any particular laminated sequence formed fast, which is a question about that sequence.

The lateral count problem.  This is the serious one.  If Green River laminae are annual, then between any two ash beds — which fell across the whole lake in a day — the number of laminae must be the same everywhere.  A 1988 Geological Society of America abstract by Paul Buchheim and R. Biaggi is reported to have found counts between the same two tuff beds ranging from 1,160 to 1,568 — roughly 35% variation.⁷

This article could not obtain that abstract.  What it could obtain is the fullest secondary account of it, by the creationist geologist Paul Garner, and it carries physical detail the bare number does not:⁸

• The site is near Kemmerer, Wyoming.
• The two tuff layers are each about 2–3 cm thick.
• The shale between them ranges from 8.3 to 22.6 cm thick across the basin.
• The lamina count between them runs from 1,160 to 1,568.
• And the count increases toward the basin margin, by up to 35%.

That last point is the one that matters, and it cuts less cleanly than the bare 35% figure suggests.  A random counting error would be random.  A systematic centre-to-margin increase, accompanied by a near-threefold centre-to-margin increase in thickness, is a sediment-supply gradient:  more material is arriving at the margin, close to the shore and the river mouths, than in the middle of the lake.  The natural reading of extra laminae accompanying extra sediment near a shoreline is that flood and storm beds are being interleaved with the annual ones — which is why varve chronologies are taken from basin centres, and why the minimum count is the candidate for the annual one.

That reading is not established here;  nobody has published it, so far as this article can find.  What can be said is that the observation is real, that it is a genuine problem for anyone who treats every Green River lamina as a year, and that it does not by itself show that no Green River laminae are annual.  The abstract should still be obtained, because the direction of the gradient is the whole argument and it is currently known only at second hand.

Two related sources also remain unobtained and would settle the matter:  the argon-argon geochronology of the Green River tuffs, which would give an independent duration to compare against Bradley's count, and Meyers's 2008 spectral analysis, which addresses whether an orbital signal is present.

And the young-earth side is divided about what Green River even is

Worth recording, because it bears on how much weight the formation can carry for either side.  Creationist geologists do not agree on whether the Green River Formation is a Flood deposit or a post-Flood one.  Michael Oard and Peter Klevberg argue it "very likely did not form in a postdiluvial lake";  John Whitmore and Paul Garner argue the post-Flood lake position;  and a multi-round forum on the question ran in the creationist literature.¹⁰  Hebert, Snelling and Clarey note the disagreement in passing and set it aside.

A formation whose depositional setting is disputed within the model cannot simultaneously be strong evidence for the model.

Objections and responses

"Lambert and Hsü found five laminae in a single year.  Varve counting is dead." They found up to five at some locations in Lake Walensee and about two on average, produced by turbidity underflows.  That is a real and important finding, and it is why nobody builds a chronology in a lake like Walensee.  It is also a factor of two to five, in a setting the same literature classifies as not producing true varves — against a requirement of about eleven, sustained, in a lake with none of Walensee's characteristics.

"Half the Suigetsu varves were never seen.  They were interpolated." Correct, and it comes from the counters.  The consequence is measurable and it is a 5–7% shortfall against uranium-thorium, not a surplus.  A record inflated by counting storms would run long.  Suigetsu runs short.

"Dust storms in the ice age could cause dozens of diatom blooms a year." The mechanism is real:  silica limits diatom growth, and dust supplies silica.  The magnitude has never been estimated.  And the requirement is not dozens.  The carbon-14 profile requires on the order of a thousand couplets a year in the early post-Flood period — about three a day, each settling as a distinct unmixed couplet.

"Mudstones can be deposited by flowing water, so lamination proves nothing about time." Correct, and Schieber's flume work is genuinely important.  It removes lamination as such from the argument.  What it does not remove is the seasonal chemistry:  a Suigetsu couplet is a diatom bloom over a clay layer over an iron-carbonate layer, in a fixed order, which is a biological and geochemical cycle rather than a hydrodynamic one.

"Antevs adjusted varve counts to make the sections match." He did, and described doing it.  That is a real weakness of the classical Swedish chronology and of early New England work, and no argument in this article rests on either.  Modern practice — multiple independent counters, quantified quality indices, cross-correlated cores, external radiometric anchors — exists precisely because of that history.

"The Green River laminae counts vary by 35% between the same two ash beds." Possibly.  The source is a 1988 conference abstract that could not be obtained.  If it is accurate it is a decisive objection to Green River specifically, and this article will say so.  It has no bearing on Suigetsu, which is checked against uranium-thorium and against raw carbon-14.

"Fossil fish in the Green River shale prove rapid deposition." They prove rapid burial, at that spot, at that moment.  Every argument for slow accumulation is an argument about the average, and averages tolerate rare fast events.  The reverse inference — that because some beds are event beds, all are — is the same mistake as inferring from one quiet year that no storms happen.

Issues and unanswered questions

• Buchheim & Biaggi 1988 is unobtained, and it is the most important missing item in this article.
• The argon-argon duration of the Green River Formation has not been checked against Bradley's count for this article. Smith, Singer & Carroll's 2003 GSA Bulletin geochronology and Meyers's 2008 Geology spectral analysis are both unobtained.
• Lambert & Hsü 1979 has been read only through Oard's quotation of it. The original is behind a Wiley paywall. Oard's quotations are internally consistent and there is no reason to doubt them, but they have not been verified against the original.
• The two Suigetsu intervals that cannot be reconciled — about 300 years at 15–18 kvyr and about 700 years at 29.3–32.6 kvyr — have a proposed mechanism that its own authors call "theoretical."
• The Swedish time scale has been revised repeatedly and by hundreds of years in places; no argument here depends on it, but no argument against it should be built on its early history either.

Verdict

Varves are the counting clock most vulnerable to the young-earth critique, and the critique is largely accurate about the mechanisms.  Multiple couplets per year happen.  Turbidites mimic varves.  Correlations between short sections are subjective and have been demonstrably wrong.  Half of the Suigetsu record is interpolated rather than seen.

None of that gets to the conclusion, for three reasons that are independent of each other.

The direction is wrong:  where varve counts have been checked, against uranium-thorium at Suigetsu and against multi-observer Bayesian modelling at Columbine Lake, the dominant error is undercounting.

The magnitude is wrong:  documented over-counting reaches a factor of two to five in the worst settings, and the requirement is about eleven in the best.

And the shape is wrong:  the raw carbon-14 in leaves picked from individual Suigetsu laminae declines smoothly to background with no staircase, which is what a record of many couplets deposited in bursts would have to produce.

The one place where the young-earth argument may still have a live empirical case is the Green River Formation, and it rests on a conference abstract nobody has been able to read.  It should be read.

Further reading

• Non-Radiometric Clocks — the overview of this series
• Ice-Core Layer Counting
• Tree-Ring Dating
• Radiocarbon Dating
• Chalk and Limestone — how fast carbonate can accumulate, treated separately

Sources