Natural Arches, Rock Bridges, and the Age of the Earth

"A simple exponential fit to the measured loss rate in Arches National Park results in a maximum age that is consistent with the biblical account of the flood, but it is at odds with the assumption that the arches are much older." — Danny Faulkner, Answers in Genesis, August 2024¹

"Many of the pinnacles or piles of rock may be the broken remains of former arches, and many of the arches we see may be gone tomorrow, next year, or a few hundreds of years and, certainly, before many thousands of years.  On the other hand, many new arches will form by the processes described above as the geologic clock ticks on." — S. W. Lohman, U.S. Geological Survey, 1975²

"That is, the 43 arch collapses over 49 years may have been an overestimation.  I probably ought to revisit my publication from last year." — Danny Faulkner, April 2025³

Summary

Natural rock arches fall down.  Wall Arch, in Arches National Park, collapsed in the night of 4 August 2008.⁴ Double Arch at Lake Powell — the one tourists called the Toilet Bowl — collapsed on 8 August 2024.⁵ Darwin's Arch in the Galápagos fell in 2021, and Malta's Azure Window fell into the sea in 2017.  These are not slow processes glimpsed through instruments.  They are events with dates, and sometimes with video.

That fact has been turned into an argument for a young earth.  It runs like this: Arches National Park holds about 2,000 arches and has lost roughly one a year in recent decades; run the loss rate backwards and the park would have held absurdly many arches only a few thousand years ago; therefore the landscape cannot be old.  The fullest published version is by Danny Faulkner, an astronomer at Answers in Genesis, prompted by the Double Arch collapse.¹

The observation behind the argument is correct, and it is not even controversial.  Natural arches really are among the youngest landforms on earth.  The Natural Arch and Bridge Society — the people who catalogue them — say most arches are probably 5,000 to 15,000 years old.⁶ The National Park Service says the same in plainer words.  Nobody in mainstream geology thinks Delicate Arch is millions of years old.  A creationist who says "these things are young" is saying what the specialists say.

The inference from that observation to a 4,500-year-old landscape does not work.  It fails for five reasons, and every one of them can be checked by a reader with an internet connection:

1. The model counts deaths and assumes zero births. Faulkner says so himself, in print.⁷
2. The number 43 is not a measured collapse rate. It traces to an anonymous list compiled around 1993, it includes arches that only cracked, and one of the arches on it is still standing.⁸
3. The 2,000 arches did not exist in 1971 in any catalogue. About 90 were known then. The rest were found by people walking around looking for them.⁸
4. The dated collapses are bunched into a single survey window. Read as a rate, the same table gives answers that differ by a factor of thirty depending on which years you pick.
5. The one natural arch anywhere that has been dated radiometrically is about 120,000 years old.⁹

Faulkner has since walked back both of his numeric inputs in his own blog, after a creationist geologist could not find a source for either.^{3 3} That is to his credit, and this article says so at length.  Answers in Genesis has not corrected the original article.

What this article does

It states the argument in its strongest form, traces every number to its origin, and does the arithmetic in the open.  Where a figure is computed rather than quoted, the computation is shown.

It also says where the creationist side is right, because it is right about more than a skeptic expects.  Arches are young.  Nobody can date one.  The mainstream figures floating around — "100,000 years for the largest arches" — turn out to be a geophysicist's off-the-cuff remark to a journalist, which he flagged as a guess.¹⁰ The honest position is that both sides are estimating, and the mainstream estimate is better constrained only because of evidence that comes from outside the arches themselves.

First, what an arch is

A natural arch is a rock exposure with a hole all the way through it, made by the selective removal of rock, with the surrounding frame still intact.⁶ The hole has to go right through: alcoves, caves and overhangs do not count, however arch-shaped they look.

A natural bridge is a kind of natural arch — one where flowing water cut the opening, or still runs through it.⁶ Rainbow Bridge and the three spans at Natural Bridges National Monument are bridges.  Delicate Arch is not.  The distinction matters here because bridges and arches form by different processes and on different timescales, and lumping them together is one way the arithmetic goes wrong.

Size is not part of the definition, so somebody has to pick a cutoff.  The one everybody uses in Utah came from Dale Stevens, a Brigham Young University geographer, in the 1970s: an opening at least three feet across in some direction, with light visible through it.⁸ The Park Service still uses it.¹¹ This is a survey convention, not a fact about rocks, and it is the reason a three-foot hole and the 88-metre span of Landscape Arch end up in the same population count.

The argument, stated fairly

Faulkner's article appeared on 21 August 2024, thirteen days after Double Arch fell.¹ His reasoning:

"An article published in 2000 said that since 1971, at least 43 arches in the park had collapsed."¹

"Taking the loss of 43 arches over 29 years as a good measure of the rate of arch loss, that is an average of 1.48 arches lost per year."¹

He rejects a linear model, correctly, on the grounds that collapse should scale with how many arches there are, and fits an exponential instead:

N(t) = N₀ e^(−λt)

Taking N₁ = 2,043 arches in 1971 and N₂ = 2,000 in 2000, he gets λ = 7.34 × 10⁻⁴ per year.  Running it backwards gives 4,256 arches in AD 971, about 8,900 two thousand years before 1971, and:

"Just a little more than 8,400 years ago, the number of arches would have been a shocking one million!"¹

That is the reductio.  A million arches in a 120-square-mile park is absurd, so the landscape cannot be that old.  At 4,500 years — his date for the Flood — the model gives about 50,000, which he calls "high" but "a reasonable number."¹

His arithmetic is correct.  Recomputing every figure from his stated inputs reproduces his published numbers to within rounding: λ = 7.335 × 10⁻⁴, 4,254 arches at −1,000 years, 8,859 at −2,000, 968,747 at −8,400, 55,438 at −4,500.  Nothing is wrong with the algebra.  Everything that is wrong is in the inputs and in the model's one hidden assumption.

A shorter version had already appeared at Creation Ministries International in 2018, by Don Batten, using a linear model:

"Arches National Park has over 2,000 rock arches.  Forty-three collapsed between 1977 and 2015, according to park rangers.  That gives a rate of collapse of about one per year… Such an attrition rate would mean that all would be gone in about 2,000 years."¹²

Batten cited his source properly — an American Geophysical Union blog post from 2015.¹³ Faulkner did not cite his at all.

The hidden assumption: a model with no births

Everything turns on one line in Faulkner's conclusion:

"While arches in Utah are observed to be collapsing, there is no evidence that arches are forming there today."¹

A month later, after criticism, he stated the assumption explicitly and honestly:

"While it is clear how arches cease to exist, there is much uncertainty about how arches form.  We have some numerical information on the rate at which arches in Utah's Arches National Park are collapsing, but we have no data on the formation rate of arches.  A good model of arch lifetimes must include both processes.  Since we have never observed the formation of an arch, in my initial analysis, I assumed that the current formation rate is zero."⁷

That is a fair and clear statement of what he did, and it is more transparent than most popular science writing.  It is also fatal, because the whole conclusion lives in that assumption rather than in the data.

Write the population as a balance:

dN/dt = R − μN

where R is the formation rate and μ is the per-arch failure rate.  Counting collapses over an interval fixes μ.  It says nothing whatever about R.  With N ≈ 2,000, 43 losses and a 49-year window, μ = 4.39 × 10⁻⁴ per year, so the mean arch lifetime 1/μ is about 2,300 years and the loss rate μN is about 0.88 arches per year.  Now run the same population backwards under different assumptions about R:

Geomorphologists do use exhaustion models of exactly this form — Cruden and Hu applied one to Canadian Rocky Mountain landslides in 1993 — but they state the condition attached to it, which is that the formation rate really is zero.¹⁴

The entire result is controlled by a parameter nobody has measured.  Set it to zero and you get a young landscape; set it to the value that balances the observed losses and you get a landscape of any age at all.  The data cannot tell these apart, so the calculation cannot yield an age.

This is the same structure as arguing that because about 61 million people die each year out of 8.1 billion, and ignoring births, humanity must be young.  Run that honestly: λ = 7.6 × 10⁻³ per year, and 1,000 years ago there would have been 1.5 × 10¹³ people.  Applying Faulkner's own tolerance — he accepts a 25-fold increase as "reasonable" and rejects a 500-fold one as "shocking" — humanity's maximum age comes out at about 420 years.  The method is not wrong about mortality.  It is wrong because people are born.

Where the number 43 actually comes from

This is the part that repays digging, and it ends somewhere nobody expected.

Faulkner's "article published in 2000" does not exist.  The sentence he paraphrases appears in Eos, the American Geophysical Union's magazine, in a feature by Mary Caperton Morton:

"Arches has changed dramatically in the years since it became a national park in 1971: The roads are paved, the campground has flush toilets, and at least 43 arches have collapsed."¹⁰

Its own citation line reads "Published on 13 July 2020."¹⁰ Faulkner appears to have read 2020 as 2000.  That single digit halves his observation window from 49 years to 29 and inflates λ by a factor of 1.69.  Correcting only that, and changing nothing else, his "one million arches" horizon moves from 8,400 years ago to 14,266 years ago, and his figure for 4,500 years ago drops from 55,438 arches to 14,411.

By April 2025 Faulkner was himself using the 49-year window — "the 43 arch collapses over the previous 49 years" — without noting that it changes the published answer.³

The Eos sentence is not a measurement either.  It restates a figure that had been circulating for years.  The AGU blog Batten cited puts it more carefully:

"In all, at least 43 arches have collapsed since 1977, according to rangers at Arches National Park.  A few of these were destroyed by vandals, but most were natural collapses caused by erosion."¹³

So the number is attributed to park rangers, the window is 1977, and some of the events are vandalism rather than erosion.

The list itself exists, and it is worth reading.  The National Park Service commissioned an administrative history of the park, published in 2019, which prints the whole thing as Table 7-2, "Collapsed Arches."⁸ Its provenance is stated plainly:

"In 1993, a list of arches known to have fallen since 1977 was compiled by unknown persons (possibly Dale Stevens).  An undated document with the table calls them 'Fallen Arches,' but notes that their inclusion on the list does not mean that the span has collapsed in every case, but that at a minimum a crack through the span has separated it from the rest of the arch, and it is no longer intact."⁸

Three things follow, and each is independently damaging:

• The source is an anonymous manuscript from about 1993, held in the park archives. It is not a monitoring programme.
• Entries are not all collapses. A cracked but standing span counts.
• The original list of 42 included Landscape Arch, which is still there. The historian removed it and added Wall Arch (2008) and Rainbow Arch (2017–18). 42 − 1 + 2 = 43.

That last detail is the tell.  The famous 43 is an arithmetic accident of one historian's editorial decisions on an anonymous list, not a count of anything.

What the table says when you read it as a rate

Counting the rows of Table 7-2 by the latest year each event could have occurred:

Ninety-five per cent of the events fall in the sixteen years to 1993.  Then nothing for fourteen years, one collapse, nothing for eight more years, one collapse.

Most of those 41 are not dated to a year at all.  They carry ranges — 1977–1986, 1984–1988, 1988–1990 — because they are last-seen-intact to first-seen-fallen brackets generated by a survey.  The administrative history says so: "The exact year of a rock fall event was not routinely noted until 1990," and earlier collapses "could be narrowed only to a range of years."⁸

That is the signature of a discovery process, not a failure rate.  Somebody walked the park in the late 1980s and early 1990s with an older list, found a number of the listed arches missing, and wrote down when each had last been seen.  Arches that fell in 1978 and arches that fell in 1991 both land inside the survey window, because the survey is what found them.

Read as a rate, the same 43-row table gives whatever answer you want:

The last row is the one worth staring at.  Take the period when the park was best observed — twenty-five years of heavy visitation, professional staff, millions of cameras — and the same method, on the same dataset, gives a mean arch lifetime of about 25,000 years and a population 4,500 years ago barely different from today's.

A rate that swings by a factor of thirty depending on which slice of one table you choose is not a rate.

To their credit, creationists noticed.  Gabriel Nulk and Nathan Mogk gave a paper at the 2025 International Conference on Creationism applying Weibull failure analysis to arch populations, explicitly to replace "the constant failure rate model assumed by Faulkner (2024)."¹⁵ Their attempt to check the underlying data failed:

"An unsuccessful attempt to validate the total count of arch collapses in Morton (2020) was made, including scholarly articles, two online arch databases, and inquiries to Arches National Park.  Therefore, four synthetic arch collapse datasets were created to validate the Weibull analysis methodology."¹⁵

Two young-earth researchers went looking for the number their own side's argument rests on, asked the National Park Service directly, could not find it, and built their model on invented data instead.  Note also that they cite the source correctly as Morton (2020) — the misdating is Faulkner's alone.

What is the real collapse rate?

Everything above is negative — this figure is unsound, that window is an artefact.  The obvious next question is what the rate actually is.  The honest answer is that nobody knows within two orders of magnitude, and it is worth showing why, because the reason is instructive rather than evasive.

There are only two bodies of evidence that could give a rate, and they disagree violently.

Approach A — the instrumented arches.  This is the only prospectively designed observation that exists.  Jeff Moore's group began monitoring arches in Arches National Park in 2013 and has measured somewhere between 17 and 25 of them, depending on which paper's sample you take.^{16 17 10} Over thirteen years, one of them failed: Rainbow Arch, in the winter of 2017–18.  That is a proper survival dataset — a defined set of subjects, observed continuously, with failures counted.

Approach B — the modern dated collapses.  Discard the 1993 survey brackets and keep only events with a real date after that survey ended.  From 1994 to the present that is Wall Arch (2008) and Rainbow Arch (2017–18), plus Ring Arch's partial collapse in 2014 if partial failures count.  Set those against the standing population of about 2,000.

Run both, with exact Poisson intervals:

Two caveats on Approach A, both of which make it weaker than the table suggests.  The sources disagree about the sample: the Park Service said in 2018 that the group studies seven sites inside the park and eighteen more across the region, while Eos in 2020 said they had "collected data on 25 arches in the national park."¹⁰ And the arch-years above assume every arch was under observation from 2013, which is certainly too generous — several were measured only once or twice.  Correcting either would shorten the implied lifetime further, not lengthen it.

The point estimates differ by a factor of about a hundred.  And note which way that cuts: Approach A's point estimate of two to three hundred years is far shorter than anything the creationist argument needs.  Taken at face value it would say arches last a few centuries, not a few thousand years.  It is not quoted here as a result because it is not one — a single failure in a sample of twenty produces an enormous confidence interval, and the sample is drawn toward large, famous, accessible arches rather than at random.

The disagreement comes entirely from the denominator.  Approach A has a real denominator — twenty-odd arches genuinely under observation — but almost no numerator.  Approach B has a real numerator but a fictitious denominator, because it treats all 2,000 catalogued arches as though someone were watching them, when in truth nobody checks on a three-foot hole in a fin in the backcountry.  That is exactly why 41 of the 43 historical "collapses" surfaced in one survey: they were not detected as they happened, they were noticed afterwards.

Two things follow, and both matter more than any single number.

The detection bias has a known direction, and it is opposite for large and small arches.  Small arches make up most of the population and are barely observed, so the true overall collapse rate is higher than Approach B suggests.  Large arches are watched constantly by millions of visitors, so their rate is well constrained and is the low one.  These pull in opposite directions and cannot honestly be averaged into a single figure — which is the deepest reason a single decay constant for "arches" is meaningless.

The one thing the two approaches agree on is a floor.  Their 95% intervals overlap only between roughly 7,500 and 12,800 years.  That is not a rigorous combined estimate and should not be treated as one, but it is suggestive, and it lands squarely inside the range the arch specialists give from entirely different reasoning — most arches 5,000 to 15,000 years old.⁶ Both intervals exclude 4,500 years at their lower bound only narrowly, so this is a soft constraint, not a proof.

The only worldwide collapse register, in full

There are two more candidates, both kept by the Natural Arch and Bridge Society, and they are worth showing complete because their size is the argument.

The first is the society's "Fallen Arches" blog category.  Its database reports exactly eight posts, all between 8 June 2015 and 2 May 2018:¹⁸

Eight records.  Worldwide.  Across four years.  Then nothing — the category has had no post since May 2018.

The second is the society's GIS catalogue, where fallen arches are flagged with "(f)" in the coordinate string.¹⁹ Across 31 state and country lists it holds 547 catalogued entries, of which 17 carry the fallen flag.  Twelve have a date or a bracket and five have none; the events span from about 1550–1750 (the Bridge of the Gods) to 2010; and two are not natural collapses at all but deliberate human demolition.

The decisive observation is that these two lists barely overlap.  Wall Arch — the most famous collapse in the park's history — appears in the GIS catalogue but not the blog.  Rainbow Arch appears in the blog but not the catalogue.  Two records kept by the same organisation about the same park disagree on which collapses got written down.  That is what non-systematic capture looks like, and it is the clearest single demonstration that neither is a sample of anything.

Both are fine sets of anecdotes and useless denominators.  Their value is the opposite of statistical: two entries, La Dame Blanche and the Ecuadorian bridge, are cases where the arch's birth was recorded as well as its death.

What would settle it is not subtle.  Someone needs a defined at-risk set — say, every catalogued arch above a stated size in a defined area — resurveyed on a fixed schedule with a stated detection threshold.  That is what the U.S. Geological Survey did for Yosemite rockfalls, where 154 years and more than a thousand events are catalogued with dates, volumes and an explicit completeness discussion.²⁰ Nothing remotely like it exists for arches, anywhere in the world.

The 2,000 arches were not there in 1971

Faulkner's fit needs N₁ = 2,043 arches in 1971.  The catalogue history is documented year by year, and it is nothing like that.

Every one of those steps is documented as discovery.^{8 2} The park administrative history calls the 1989–1995 rise a "Systematic census," carried out by the park's Resource Management and Interpretation staff together with the amateur arch hunters who founded the Natural Arch and Bridge Society.⁸ The 2,000th arch was found in 1995 and named for the milestone.  The figure has been repeated on Park Service materials ever since without revision.

In 1971 the catalogue held about 90 arches.  Faulkner's model puts 2,043 there.  He is off by a factor of roughly twenty-three, and in the wrong direction: over the years he models the population as shrinking, the catalogue grew about twenty-two-fold.

Faulkner reached the same conclusion himself, from a different direction, in April 2025:

"So where did the count of 2,000 arches come from?  With no real documentation of the number of arches, I suspect that someone estimated that figure, and other people merely repeated that number, eventually giving that number an air of authenticity."³

Both of his inputs are now disowned by their own author.

Discovery is not formation — and that cuts both ways

None of this means 1,900 arches formed between 1971 and 1995.  They were found, not built.  But that is exactly the objection to using the 43 as a destruction rate, because both numbers came out of the same counting exercise.  You cannot treat the survey's finds as a catalogue artefact and the survey's misses as physical events.

Faulkner applies the distinction correctly in Kentucky and then drops it in Utah.  Of Red River Gorge he writes that "the number of known arches in Kentucky is increasing with the discoveries of previously unknown or at least undocumented arches" — and attributes it to vegetation hiding them, not to arch formation.¹ That is right.  It is also true of Utah.

Do arches form today?

This is the load-bearing question, and the answer is better documented than Faulkner's sources let him see.

The agencies say yes, in plain language.  The U.S. Geological Survey's official guide to the park, written by S. W. Lohman in 1975:

"On the other hand, many new arches will form by the processes described above as the geologic clock ticks on."²

The National Park Service today:

"Every story of the park's 2,000 arches will end at some point.  However, new arches are forming at a similar pace, which will continue the story of Arches National Park."²¹

A USGS geologist identified a specific newly formed arch and gave his field evidence.  Lohman named it himself:

"Just southwest of Sheep Rock is an unnamed opening in the lower part of the Slick Rock Member which I call 'Baby Arch,' because it is one of the newest ones visible from the park road… Its youthfulness is also indicated by the sharp, angular breaks in the ceiling and by the pile of freshly fallen rocks."²

Fresh angular breaks and unweathered talus are exactly what a recent breakthrough looks like.  A visitor photographed a further rockfall at Baby Arch in May 1989.⁸

Openings are documented growing, with dates.  In November 1940 a large rock fell out of a feature then called Arch-in-the-Making; the opening more than doubled, and the result is the arch now called Skyline Arch.⁸ In June 1992 about 4,000 tons of sandstone fell from inside Kachina Bridge, and the Park Service describes it as "enlarging the opening as it has no doubt been enlarged time and time again."²²

Two complete arch life cycles have been photographed from birth to collapse.  This is the strongest available answer, and both cases come from the Natural Arch and Bridge Society's own records.

The first is La Dame Blanche, near Dieulefit in the Drôme, France, catalogued as NABSQNO 31T-663186-4933509 in a crumbly clay-rich rock.  Three photographs by three different people track it: a small young opening in 2009, a mature span of 20 feet in June 2014, and a collapse by April 2015.²³ The society's own headline for it is "Rapid Evolution of an Arch," and its opening line is the point exactly: "Arch evolution can occur over thousands of years, but La Dame Blanche… evolved from youth to maturity to collapse in the course of just a few years."

The second is the San Rafael Waterfall natural bridge in Ecuador, NABSQNO 18M-212655-9988525, "suddenly created on February 2, 2020, by the collapse of a sinkhole behind the San Rafael Waterfall."²⁴ There is a NASA satellite image from before and an Ecuadorian government image from after.  NABS thought it might be the longest natural bridge in the world, and wrote at the time: "It will be interesting also to see if this bridge lasts more than a few years."  It lasted one.

A third case is reported in the peer-reviewed literature: a sea arch on Kukuihoʻolua Island at Laie Point, Oʻahu, which local witnesses say did not exist before the tsunami of 1 April 1946.²⁵ That one rests on testimony plus wave modelling rather than photographs, and its author hedges accordingly — but it is a geologist publishing in a refereed journal.

So the claim that arch formation has never been observed is not merely unproven.  It is false, with photographs.

And one famous arch demonstrably formed and died inside two centuries.  Malta's Azure Window collapsed on 8 March 2017.  Working back through paintings and photographs, its documentary record establishes that it had formed, or nearly formed, only about 193 years earlier.²⁶ That is the strongest single datum for the "arches fail fast" premise — and it is self-undermining, because an arch that both appeared and vanished within 200 years is direct evidence that the population is being replenished, not drawn down.

All three are fast, water-driven or weak-rock cases, and a fair critic will say they do not show that a desert arch forms today.  That is right, and it is the honest limit of the point.  What they establish is narrower and still sufficient: arch populations demonstrably gain members as well as lose them, so a model that sets the birth rate to zero is not a simplification of reality but a contradiction of it.  The reason no Utah desert case is on the list is that the process there is slow and silent, not that it has stopped.

Arches exist at every stage of development, which is what an ongoing process looks like.  The Park Service's own account runs alcove → hole → widening → collapse, and Paul Geimer, quoted in the very article Faulkner mined for the number 43, says: "If you look around the park, you see so many huge fins and alcoves that are on their way to becoming the next North Window Arch."¹⁰ At Natural Bridges National Monument the Park Service presents its three bridges explicitly as a sequence from youngest and growing to oldest and near failure.

The honest qualification.  Nobody has photographed a Utah desert arch breaking through for the first time.  The Natural Arch and Bridge Society says only "a handful" of arches anywhere have had their formation observed and recorded.⁶ And Jeff Moore's group, who have instrumented more arches than anyone, concede the point directly:

"Rates of arch formation and collapse are unfortunately poorly known, so at this time it is not possible to quantitatively relate arch geometry and longevity."¹⁷

But an absence of eyewitnesses proves nothing here.  A collapse is a loud, dated, sometimes filmed event.  A formation is the quiet loss of the last grains from a widening alcove, with no moment to witness.  Even collapses are almost never seen: the Park Service's chief of resource stewardship told Eos that "Rockfall is pretty rare, and it's even rarer that it's actually witnessed by people."¹⁰ Expecting a photographed arch birth, and treating its absence as evidence, sets a standard the physics cannot meet regardless of the earth's age.

How fast does this rock actually erode?

The model has a second problem that is independent of everything above.  It needs about 50,000 arches to exist 4,500 years ago.  Those arches have to be carved out of solid rock in less time than that, and the measured erosion rates do not permit it.

The most direct measurement in the region comes from Inscription Rock at El Morro National Monument, where the Park Service set reference pins into the sandstone and measured the face against them.  The rock retreated about 0.8 mm in five years, 2000–2005 — 0.16 mm per year.²⁷ That is a ceiling, not a typical rate: the pins sit in friable, clay-cemented Zuni Sandstone at spots chosen because they were actively degrading.

The archaeology gives much lower numbers on ordinary surfaces.  Juan de Oñate cut his name into that same rock in 1605, and it is still legible 420 years later.  Petroglyphs pecked one to five millimetres deep through desert varnish remain crisp after 1,500 years.  Painted panels in Colorado Plateau alcoves have survived exposure for well over a thousand years.

The broadest measured baseline comes from cosmogenic ¹⁰Be.  Portenga and Bierman compiled every published measurement then available and reported that bare rock outcrops erode at a mean of 12 metres per million years, median 5.4 — against 218 m/Myr for whole drainage basins.²⁸ The units are worth keeping straight, because one metre per million years is exactly one millimetre per thousand years.  So a bare outcrop loses on the order of 12 millimetres per thousand years.

Put the three rate families side by side over Faulkner's 4,500 years:

Now look at what has to be removed.  Delicate Arch's opening is roughly 10 by 14 metres.  Landscape Arch's is 88 metres across.  Excavating those openings requires metres to tens of metres of rock removal, and that is after the fins themselves have been carved out from between the joints, which takes tens of metres more.

There is no rate consistent with the measurements at which 50,000 arches can be cut in 4,500 years.  A young-earth geologist can answer this by invoking catastrophic post-Flood runoff, and Michael Oard's flood-geomorphology programme does exactly that.²⁹ That is a coherent position.  But it is a different argument, resting on different evidence, and the collapse-rate data cannot support it — indeed the collapse-rate argument quietly assumes the opposite, that present rates are a reliable guide to the past.

What can actually be dated

One natural arch in the world has a radiometric formation age.  The Pont d'Arc, a limestone natural bridge over the Ardèche in France, sits at the neck of an abandoned meander near the Chauvet cave.  A 2021 study in Scientific Reports dated the meander's abandonment surface by uranium-series and electron spin resonance:

"Results show that the meander became totally cutoff between 108 and 138 ka (95%).  Hence, the natural arch formed before the Upper Paleolithic and the first known human presence in the Chauvet Cave, dated to 37 ka cal BP."⁹

The authors add that the opening itself began earlier, when "the channel through the arch would have been no more than a few meters wide."⁹

So the only natural arch anyone has dated is roughly 120,000 years old, with a small opening older still.  That is about twenty-five times the age Faulkner's framework allows for any arch — and, notably, four times older than the 30,000-year ceiling the Natural Arch and Bridge Society proposes for arches generally.⁶ It cuts against both estimates.

The caveat has to be stated: Pont d'Arc is limestone, cut by a river through a meander neck.  Utah's arches are eolian sandstone fins weathered from the inside out.  The dating method does not transfer, and neither does the age.

For the Utah arches, nothing has been dated at all.  No cosmogenic exposure age, no luminescence age, no varnish age exists for any arch, any fallen block, or any talus pile in the park.  The Utah Geological Survey put it plainly when Wall Arch fell:

"We do not have a good geologic tool for dating arch formation, but we are sure that Wall Arch had stood nearly unchanged for hundreds, and probably thousands of years."⁴

The frequently quoted figure that the largest arches "may be in the range of 100,000 years old" comes from a single interview.  Paul Geimer said it to Eos, prefaced with "Our best guess is," in a passage where he explained that erosional landforms are notoriously hard to date "because most of the diagnostic evidence is transported away by wind and water."¹⁰ It appears nowhere in his published work, and his group's own paper says arch formation rates are "poorly known."¹⁷

This is a genuine, symmetric weakness, and it should be conceded.  A creationist saying "you have not dated these arches" is correct.  The mainstream's arch ages are informed estimates, not measurements.  What tilts the balance is not the arches but everything around them: measured erosion rates on the same rock, dated archaeology on the same surfaces, and dated fragile rock features elsewhere.  Those are measurements, and they all point the same way.

The fragile features that have been dated

Nobody dates arches, but there is a whole seismological subfield devoted to dating precariously balanced rocks — boulders perched on pedestals so delicately that strong ground shaking would knock them over.  Their survival puts a ceiling on how hard the ground can have shaken at a site, so engineers pay to date them properly, by cosmogenic ¹⁰Be exposure dating of the rock and its pedestal.

A balanced rock is, by any reasonable measure, more fragile than an arch.  It has no load path holding it together and no compressive stress locking its grains; it simply sits there.  So its survival time is a conservative lower bound on what a rock landform in a desert can do.

Anna Rood's 2022 doctoral thesis measured twenty of them in southern California:

"The fragility ages of the 20 studied PBRs range from approximately 1.5 ka to 50 ka…  The oldest PBR is LJ5, which has a fragility age of 49.8 ka."³⁰

Half of the twenty are younger than 10,000 years, which is worth saying because it again shows these are genuinely young features.³⁰ But the oldest has stood in its precarious state for about 49,800 years — eleven times the age Faulkner's framework permits for anything on the post-Flood landscape, and measured rather than estimated.

The point is not that arches are 50,000 years old.  It is that the objection "such delicate rock features cannot last that long" is testable, has been tested on the most delicate rock features anyone could find, and failed.

What the instruments have found

Since 2013 a University of Utah group has instrumented arches with seismometers, reading their resonant frequencies the way an engineer reads a bridge.  Nineteen arches have been modelled in three dimensions from photogrammetry, spanning 4 to 88 metres.¹⁷ Seventeen have measured resonant modes between 1 and 40 Hz.¹⁶

Three results matter here.

Arches are nowhere near crushing.  Compressive stresses in every arch modelled are "at least an order of magnitude lower than the reported compressive strengths for these materials."¹⁷ Arches do not fail by being squashed.  They fail in tension, at cracks, and the models predict where those cracks appear well enough that the predicted tension zones match the observed cracks in the field.

Shape controls stability, and it varies enormously.  A flat beam-like lintel carries high tension; an inverted-catenary lintel carries almost none.  The measured ratio of mean principal stresses runs from 0.7 at Two Bridge to 32.9 at Rainbow Bridge — a factor of nearly fifty across a single population of nineteen.¹⁷ Treating such a population with one decay constant is not a simplification; it is a category error.

Repeated monitoring has not detected progressive damage.  Mesa Arch was checked nine times over eighteen months: "We observed no permanent resonant frequency shifts related to irreversible damage of Mesa Arch during our study period."³¹ The crackmeter on North Window, installed in November 2015, found the crack opening and closing by more than 2 cm a year and more than half a millimetre a day — with no net trend.  Jeff Moore: "To the best of our current analysis, there is no opening trend… However, the annual opening and closing is larger than we'd thought it would be."²¹

That last is the closest thing to a direct measurement of arch ageing anyone has, and over the observation period it is zero.

Three things cut the other way, and they should be stated plainly.

First, Rainbow Arch — a small span, 11.7 feet, catalogued as SA-137 — was measured four times during 2017 by Moore and Geimer, who "observed no changes in the crack or vibration characteristics over 12 months."³² It collapsed that winter; a ranger found it gone in February.  The monitoring programme's most direct test of prediction, on the one instrumented arch that failed, was a miss.  Its own literature does not even agree on when: Moore et al. 2020 and Finnegan et al. 2022 say the winter of 2017–18, while Geimer et al. 2020 says 2018–19.

Second, the instruments have a detection floor, and it is high.  Reversible weather-driven change swamps any damage signal — resonant frequencies move 0.5 to 6 per cent per 10 °C, with jumps of about 17 per cent when pore water freezes.³³ At a slab in Courthouse Mesa that was actively toppling, with crack aperture opening a measured 2–4 mm a year, three continuous years of monitoring still found "no clearly detectable irreversible changes in modal parameters," and the authors say why: detection "may be limited due to the magnitude of seasonal changes," which reached 29 per cent annually.³⁴ A null result from an instrument that cannot see a landform visibly coming apart is a weak null.

Third — the honest counterweight — these methods can detect impending failure when it is close.  At the Chamousset limestone column in the French Alps, continuous ambient-noise monitoring caught the first resonance frequency dropping roughly 20 per cent in under two weeks before a 21,000 m³ column fell in November 2007.³⁵ The technique works.  What it detects is the terminal phase, not the long middle of a landform's life — which is the part that bears on age.

The cement argument, and why it fails inside its own footnotes

Faulkner offers a mechanism for why Utah loses arches and Kentucky does not.  Utah's sandstones are cemented with calcite, which dissolves in rainwater; Kentucky's Corbin Sandstone is cemented with silica, which does not.¹

The chemistry is real.  The USGS guide to the park describes the Entrada as quartz sand "cemented together by calcium carbonate," dissolved by carbonic acid in rain and groundwater — so Faulkner's mechanism has a genuine USGS pedigree.²

But his rule is inverted by the geologic map he cites for it.  Doelling and Kuehne describe the Slick Rock Member of the Entrada — the unit holding almost all the park's arches — as:

"Well-indurated, red-orange or brown… all cemented with calcite and iron-oxide cement… most arches formed in this and in the immediately adjoining units."³⁶

And the Navajo Sandstone as:

"generally friable and weakly cemented with silica or calcite."³⁶

So on his own source, the calcite-cemented unit is the well-indurated one and the silica-bearing unit is the friable one.  Worse, Double Arch — the collapse that prompted the whole article — is in the Navajo.

The same map attributes arch formation not to cement at all, but to structure:

"Arch formation is a product of the Pleistocene and Holocene.  Massive, moderately hard sandstones jointed by folding, resting on or containing soft layers or partings, and located near salt-cored anticlines undergoing dissolution, favor the formation of arches in this region."³⁶

Three conditions — jointing from folding, a soft layer underneath, and salt dissolution below — which "rarely… occur in one place, but they do in Arches National Park."³⁶ That is why the park has the world's densest arch concentration, and it has nothing to do with what glues the sand together.

There is also a much simpler confound.  Moab gets about 9 inches of rain a year; Red River Gorge gets about 47, under a forest canopy, with fewer freeze-thaw cycles and no salt-cored anticline underneath.  Attributing the entire durability difference to cement mineralogy skips every other variable.

Kentucky, and a retraction worth reading

Faulkner used Red River Gorge as a control: over a thousand arches, "There has been no recorded arch collapse in Red River Gorge."¹

On 18 September 2025 he retracted it:

"I have often said that there has not been a documented natural arch collapse in Kentucky.  I now stand corrected."³⁷

Moose Head Arch, newly documented by arch photographer Bill Patrick, collapsed between its discovery and its confirmation — apparently struck by falling tree debris during Hurricane Helene.  Faulkner maintains a distinction that is fair: Utah arches fall under their own weight, whereas Moose Head was knocked down.³⁷ In the same post he reports Patrick's Kentucky count as 1,175 arches with solid leads on a couple of hundred more.³⁷

The control case is worth taking seriously in its own right.  If Kentucky's arches genuinely do not fall on human timescales while Utah's do, then arch lifetime varies by orders of magnitude with rock and climate — and a single decay constant read off one park cannot be a clock for anything.  The Kentucky observation is one of the better things in the creationist treatment, and it argues against the conclusion it was offered to support.

Where are the ruins?

Faulkner's model requires roughly 2,000 arch collapses in the park since AD 1000 and about 48,000 since the Flood, all inside 310 square kilometres.  Every one of them leaves something behind.

We know what that looks like, because Wall Arch is photographed.  The Utah Geological Survey published an image two days after it fell: two orange fin stubs still standing with a pile of bright, angular, unvarnished slabs between them.⁴ Eighteen years later the Park Service still routes visitors past the debris.

Old collapse sites are recognisable too — the Park Service's geologic report for Arches notes that "former arches that have collapsed" are present in the Fiery Furnace, and Lohman wrote in 1975 that "many of the pinnacles or piles of rock may be the broken remains of former arches."²

But nobody has ever counted them.  There is no inventory of former arches, at Arches or anywhere.  Talus in the park is mapped only as undifferentiated Quaternary cover.  No fallen block has been dated by any method, though cosmogenic and weathering-rind dating are routinely applied to rockfall debris elsewhere.

So this is a real, testable prediction that has not been tested.  Faulkner's model says the park should contain tens of thousands of collapse sites in various states of decay; a steady-state model says it should contain a few thousand.  Someone could go and count.  That nobody has is a gap in the mainstream literature, not a point for either side.

Faulkner's own strongest objection, turned around

The best quantitative argument against the young-earth model was made by Faulkner himself, though he was aiming it elsewhere.  In April 2025 he questioned whether 2,000 arches can physically fit in the park:

"Arches National Park contains nearly 77,000 acres of land.  If there are 2,000 arches in the park, that works out to an average of 0.026 arches per acre.  But the arches in Arches National Park are found only within large sandstone outcroppings, particularly among the fins.  Much of the park is rolling scrubland where there are no arches.  Examining detailed maps of the park, I estimate that, at best, 10% of the land in the park is arch-dense territory.  Therefore, for there to be 2,000 arches in the park, the portions of the park rich in arches would have to have 0.26 arches per acre, or about one arch for every four acres… Having now hiked in five parts of the park where some of the more memorable arches are found, is the arch density that great?  I am doubtful."³

Apply that reasoning to his own model.  In the arch-bearing tenth of the park:

If one arch per four acres strains belief, then seven arches per acre — his own requirement at the Flood — is 25 times harder to believe, and his reductio case is 484 times harder.  The density objection he raises against the standing count is far more damaging to his model than to the count.

What survives

Setting the argument aside, several things the creationist writers say are true, and a skeptical reader should know that a Christian source will say so.

Arches are young.  Not "young" in the creationist sense, but genuinely among the most transient landforms on the planet.  The Natural Arch and Bridge Society: "it is safe to say that no natural arch is older than about 30 thousand years.  Most are probably between 5 and 15 thousand years old, i.e., not incomparable to the span of recorded history."⁶ Sea arches they put at "a few centuries, or even decades."  Batten's instinct that these are not million-year-old features is simply correct.

Nobody has dated one.  For the Utah arches this is flatly true, and the Utah Geological Survey says so.

The mainstream's own numbers are soft.  The "100,000 years" figure is an interview remark labelled a guess.  The "over 2,000 arches" figure dates from 1995 and has never been revised.  The "43 collapses" figure comes from an anonymous list.  A creationist pointing out that geologists repeat unsourced numbers is, in this instance, right.

Faulkner's cement chemistry has a real basis, even though his durability rule is inverted by his own citation.  Calcite dissolution in carbonic acid is a genuine mechanism in the Entrada, and the USGS says so.

He corrected himself in public, twice, without being made to.  He conceded the zero-formation assumption a month after publishing; he disowned both numeric inputs eight months later after a fellow creationist checked them; he retracted the Kentucky claim a year after that.  That is how it is supposed to work.  The failure here belongs to Answers in Genesis, which has left the 2024 article standing unchanged and uncorrected while its author has walked away from its inputs on the same website.

Verdict

The collapse of natural arches is not evidence for a young earth.

It is not evidence for an old earth either.  A loss rate measured over decades constrains one parameter of a two-parameter system, and the parameter it does not constrain is the one that determines the answer.  No amount of better counting fixes this; the calculation is not underdetermined by accident but by structure.

What does bear on the question is everything the arches sit in: measured erosion rates on the same sandstone, dated inscriptions and rock art on the same surfaces, dated fragile rock landforms elsewhere, and the single dated arch in France at 120,000 years.  Those are measurements, and they are consistent with each other and inconsistent with 4,500 years.

The most useful thing about this episode is not the geology.  It is that a number with no source travelled from an anonymous 1993 archive box into an encyclopedia article, a magazine, a ministry publication, and a mathematical model, growing more precise-sounding at every step, until a creationist geologist asked where it came from and nobody could say.  That happens on every side of every argument.  It is worth knowing that it happened here.

Companion pages

• The literature — every source consulted, catalogued with what it measured and which way it cuts.
• Erosion Rates and the Age of the Earth — the same style of argument at continental scale. Rivers really would level the continents in about twenty million years, mainstream textbooks say so, and the conclusion still does not follow.

Sources