One layer of Greenland ice, two counts of its years, and the trees that settled it

Counted Twice

In 774 a storm of particles, most likely from the Sun, put a burst of carbon-14 into the air and of beryllium-10 into the snow. Trees around the world record it as a jump in carbon-14 across their rings of 774 and 775. In Greenland it is a layer of ice, and GICC05, the count of annual layers published in 2006, put that layer in 768, with a maximum counting error of one year. The same ice was counted again for a chronology published in 2022, tied only to eruptions in 1783 and 1362, and the layer came out in 774. Lay both counts on one core, and the gap opens where the old count had trusted two eruptions it had misnamed.

A year in the wood

Carbon-14, like beryllium-10 and chlorine-36, comes from what Florian Mekhaldi and colleagues call the nuclear cascade triggered when cosmic rays reach the atmosphere, and a tree takes it in with the rest of its carbon, ring by ring: much of the carbon a tree absorbs, write A. Scifo and colleagues, is immediately locked into the cellulosic structure of the growth-rings. In 2012 Fusa Miyake and colleagues measured it in the annual rings of Japanese cedar, one or two rings at a time, and reported a rapid increase of about 12‰ in the (14)C content from ad 774 to 775, which is about 20 times larger than the change attributed to ordinary solar modulation. In 2018 U. Büntgen and colleagues measured 484 individual tree rings formed in the periods 770–780 and 990–1000 CE, and the two spikes, they wrote, ensure the precise dating of 44 tree-ring records from five continents. The likeliest cause, from beryllium and chlorine in polar ice, is the Sun: Mekhaldi’s group found that the spikes were most likely produced by extreme solar events.

A ring’s year is not found by counting in from the bark. You cannot date a piece of wood by counting its rings, as another layer here begins; What you do instead is find the year the weather went wrong and look for it in everything else that was standing (Every Tree Wrote the Same Bad Year). So the jump has a calendar year that does not depend on anyone counting layers of ice.

Ring 774 reads like the rings before it in both trees, and ring 775 does not. The oak keeps climbing for another year; the juniper’s next measured ring, 777, sits no higher than 775. Spikes like this are often drawn out: Mekhaldi’s group found the rises in beryllium and carbon imprinted over a time span of 2–3 years despite the probable ephemeral aspect of the cosmic-ray events. In both trees the first raised ring is 775, and Büntgen’s network puts the start of the event in the boreal summer of 774.

The same year in the ice

The burst that made carbon-14 also made beryllium-10, which is deposited on the ice sheets, so the same event is a layer in Greenland’s ice: Mekhaldi’s group measured it there and at the other pole, a symmetrical production and deposition of ¹⁰Be at both poles. Greenland’s ice is dated by counting. Some of what falls with the snow follows the seasons (sodium, for one, has a strong winter peak), so each year leaves a band, and you count the bands down from the top. GICC05, the Greenland Ice Core Chronology 2005, was made on three cores, DYE-3, GRIP and NorthGRIP, and B. M. Vinther and colleagues described it in 2006: The main annual layer counting is carried out on the most suited records only, and After the verification, the main dating is transferred to all records using the volcanic reference horizons as tie points.

GICC05’s table in the NOAA archive gives, for every twentieth year, the depth at which that year begins in each core and the count’s maximum counting error there: half the sum of the uncertain layers accumulated until the corresponding age, an uncertain layer being one of the features in the ice core that could neither be dismissed nor confirmed. At the depth of the beryllium layer that error is 1 year. Giulia Sinnl and colleagues, looking back at it in 2022, put the published figure plainly: the published uncertainty until about 2.7 kyr b2k is only 2 years. On that count the layer is 768.

Sinnl’s group counted the same ice again: six deep ice cores and three shallow ice cores, with automated counting of annual layers on multiple parallel proxies and manual finetuning, and almost no historical dates. Hence, we use Laki as a reference datum to calculate relative ages for the rest of the timescale. Laki erupted in 1783; one core, DYE-3, whose measurements start below Laki, was tied to Öræfajökull in 1362 instead. Of the dated events in their table, Only two of the events are used to anchor our timescale. The new count, GICC21, puts the beryllium layer in 774.

Where the gap opens

Sinnl’s group published the difference between the two counts for every year of the new one, a transfer function for moving anything dated on GICC05 onto GICC21. It is the thick line below, with GICC21’s own one-sigma uncertainty around it. The band around zero is GICC05’s maximum counting error: if the old count had been off by no more than it said, and the new one were right, the line would stay inside it.

From Laki down to Samalas, the eruption of 1257, the two counts are never more than 3 years apart. Then the line climbs: on average about 3 years in the second half of the eleventh century, 6 in the tenth, 7 in the sixth, and 10 at the layer GICC05 had taken for Vesuvius. In the nineteen centuries below that it never falls under 9 again. GICC05’s error band is 2 years at its widest anywhere in the Common Era.

Four years for Hekla, and a match with Vesuvius

Sinnl’s group traced the gap to two of GICC05’s pins: beyond the Samalas eruption (1258 CE, 742 b2k), the offset increases rapidly because of the mismatch of the Hekla (1104 CE) and the Vesuvius (79 CE) eruptions. GICC05 had matched its cores to each other by volcanic reference horizons in electrical conductivity measurements, and tied some of those layers to eruptions whose dates were known from history. Two of those matches were wrong.

The first was a prominent peak attributed to the great eruption of Hekla, in Iceland, in 1104. When Coulter and colleagues looked in 2012 for Hekla’s ash near that peak, neither DYE-3 nor GRIP or NorthGRIP supported the identification with the Hekla eruption. Sinnl’s group found the mark it had left on the count: The erroneous attribution of Hekla (1104 CE) explains the steep 4-year offset which was introduced between the Samalas eruption and 1104 CE. In hindsight, the four annual layers appear poorly supported by the DYE-3, GRIP, and NorthGRIP1 data.

The second was the deepest pin, the eruption of Vesuvius in 79. In the construction of GICC05, an acidity spike attributed to the Vesuvius eruption (79 CE, Italy) was considered an exact time marker, carrying no age uncertainty. And the count agreed with it: The authors also observed that the count between 1362 CE (Öræfajökull, Iceland) and 79 CE (Vesuvius) was correct within 1 year, which is 1,283 years of history. The same two layers are in the new count’s table, so you can measure the interval on both. In each of the three cores GICC05 counts about 1,283 years between them and GICC21 about 1,273.

The agreement was real. The layer was not Vesuvius. In 2022 Gill Plunkett and colleagues analysed ash from the same acid layer: Major and trace element composition of the particles indicates that the tephra does not derive from Vesuvius but most likely originates from an unidentified eruption in the Aleutian arc. GICC21’s table calls the layer UE 88, gives its age on another ice chronology as about 88, and dates it on its own count to 89.

Neither mistake could show in the error bar. Sinnl’s group again, on what the maximum counting error left out: Hence, the MCE does not consider the bias one can introduce, for example because of misleading assumptions on the tie points or abruptly changing layer shapes. In summary, GICC05 was considered exact for the part younger than the eruption peak previously assigned to Vesuvius because many well-known historical eruptions tied the chronology together.

A witness that does not count layers

The trees check the count only where the spikes are. A check along the whole count came six years before the recount, from Florian Adolphi and Raimund Muscheler, who matched the slow wander of beryllium-10 in Greenland ice to the slow wander of carbon-14 in tree rings, decade by decade, with no annual spikes at all. Their abstract: Our results imply that the GICC05 counting error is likely underestimated during the most recent 2000 years leading to a dating bias that propagates throughout large parts of the Holocene. Their estimate is the dashed line on the chart; at 775 it says GICC05 is 6 years too old, with a 95% range of 1 to 11. And at the old pin: our results indicate a timescale offset of −11 ± 6 (2σ ) years at AD 79, which would imply an age of AD 90 ±6 for the GRIP tephra layer – incompatible with an attribution to the age of the Vesuvius eruption within 2σ.

They checked themselves against the two spikes, which their decadal data could not see: Consistent with these findings, we obtain IntCal13-GICC05 differences of −4 ± 4 and −6 ± 5 years (2σ ) for the 994 and 775 AD event, respectively. The second spike, in 993 by the trees, is in GICC21’s table too, at 992 ± 3.6, and it was Not used as tie point across ice cores.

What moved with it: the year the sun went dim

A date in the ice is rarely wanted for itself. It is wanted to set an eruption beside a harvest, a famine, a cold summer in the trees, and a count that is six years out sets them beside the wrong ones. Procopius, writing of the tenth year of Justinian’s reign: For the sun gave forth its light without brightness, like the moon, during this whole year, and it seemed exceedingly like the sun in eclipse. The edition dates it 536 to 537.

In 2008 L. B. Larsen and colleagues, re-reading the acid in GICC05’s three cores, found large eruptions at 529 and at 533 to 534, each give or take two years, and took the second for the cause: a substantial and extensive atmospheric acidic dust veil at A.D. 533–534 ± 2 years. The same year M. G. L. Baillie argued from the tree rings that the ice was out of step: Moving the ice-core chronology forward 7 years places two large volcanoes in 536 and 541. Sinnl’s group built their transfer function for this: The transfer function can be used to translate any age previously matched to GICC05 to the new revised GICC21 ages. Run the old dates through it and they move 7 years: the eruption the old count put in 529 lands in 536, and the one it put in 533 to 534 lands in 540 to 541.

What this page does not show

• The beryllium itself. The layer is placed where GICC21’s table of tie points puts it, the tie point it calls “775 CE SPE”; the beryllium measurements are not drawn here.
• An exact old year at every depth. GICC05’s table gives every twentieth year, so between its ticks the old count’s year is interpolated in a straight line. At the beryllium layer the three cores give 768, 768 and 769, and the transfer function, which Sinnl’s group built from GICC05’s own year-by-year layers, gives an offset of 6 there; published estimates of the gap at 775 run from six to seven years.
• A smooth difference. The transfer function is, for each year, a weighted mean of several cores’ offsets, each read from two sets of layers at matched depths, and it steps by as much as 1.4 years between neighbouring years. The steps are in the data, not in the drawing.
• The paper’s own rounding. GICC21’s paper prints the beryllium layer at 774 ± 4.1; its supplement’s table of tie points gives the uncertainty of the same layer as 4.199, which this page shows as 4.2.
• That the trees are beyond question. The rings’ years rest on crossdating, tree against tree. The spike is itself a test of it, the one Büntgen’s network passed when the spikes, in its words, ensure the precise dating of 44 tree-ring records from five continents.