A tree ring is a date. Trees standing hundreds of miles apart put down the same pattern of fat and starved years, and that shared pattern is what lets a piece of wood be dated to the calendar year it stopped growing. This takes 2,474 site chronologies from the International Tree-Ring Data Bank, draws them, and asks the obvious question: in which years did the world's trees all have a bad time at once — and were those the years the history books would predict? Mostly not. Across the 42 largest eruptions with a date somebody wrote down at the time, the share of sites going narrow lifts by 1.3 points, from 8.9% to 10.1%. Even 1816, the year without a summer, is below average narrow worldwide. The reason turns out to be a fact about the archive rather than about trees: half the bank stands in western North America, which in 1816 was having one of its best growing seasons on record.
Every ring is one growing season, and its width is that season written down. This is a real chronology drawn at its real proportions — the disc shows the whole life of the site, the strip below it is the same wood under a hand lens. Drag either one. The dark band at the outer edge of each ring is latewood, laid down as the season closes; it is the line your eye actually counts.
| Year | Index | Pointer | Cores | World that year |
|---|
Counting rings inwards from the bark only works if no ring is missing and none is false, and in a hard year a tree may lay down no ring at all. So dating is done by matching, not counting: you slide an undated sequence against a dated one until the pattern of narrow years locks. Below are two real chronologies as skeleton plots — a spike for every year the site went unusually narrow. Drag the lower one and watch the correlation find its own answer.
Pool every chronology and count, for each year, the share of sites that put down an unusually narrow ring. Above the line is that share; below it, mirrored, is the share that went unusually wide — the same arithmetic with the sign flipped, carried as a control. A spike that appears only above the line is a shock. One that appears in both is thin replication playing tricks, which is why the pale band behind shows how many sites were available at all. Gold ticks mark the eruptions: every volcano of VEI 5 or greater whose date comes from someone writing it down at the time.
Each dot is a site that was alive and measured in the chosen year, coloured by how that year's ring compared with its own neighbourhood. This is where the global average stops being a number and starts being a map — and where you can see that a bad year is almost never bad everywhere.
| Year | Narrow | Wide | Sites |
|---|
The flat global answer is not a fact about trees. It is a fact about where the trees were collected. 1,167 of the 2,474 sites — a little under half — stand in western North America, so a “world” composite is largely a western North American one wearing a bigger name. Cut the same data by region and the years stop cancelling. Note especially what happens to the year without a summer: it appears, sharply, in the one region where the histories say it happened, and is buried by the region that happened to have a wonderful year.
Each site chronology is already standardised: the tree's own age trend has been divided out, leaving an index whose average year is 1000. That still drifts on decadal scales, so an absolute threshold would flag whole stretches of a cool century rather than single shock years. Every ring is therefore rescaled against its own thirteen-year neighbourhood — the pointer-year, or Cropper, transform — giving a value in standard deviations. At or below −1.28, the tenth percentile of a normal distribution, the year is counted narrow for that site; at or above +1.28 it is counted wide.
A site-year needs at least five contributing cores to count, and a year is only reported when at least fifty sites are available. Both numbers are shown throughout rather than hidden, because the honest limit on this page is replication, not method.
The wide-year share runs through the identical pipeline with one sign changed. Anything that inflates the narrow count by accident — a thin year with few sites, an artefact of the window, a bug — should inflate both. Across the whole record the two sit at 8.9% and 8.5%, almost exactly the 10% each threshold implies, which is the main evidence that the arithmetic is doing what it claims. When a year reaches 30% narrow against 2% wide, that asymmetry is the finding.
The eruption test uses only dates established by contemporary written record, from the Smithsonian catalogue. Ice-core volcanic chronologies were themselves synchronised against tree-ring marker years, so testing tree rings against them would be partly circular. Documentary dates are independent of both.
The bank is not a sample of the world's trees. It is overwhelmingly Northern Hemisphere, heavily western North American and European, and collected mostly in the 1980s and 1990s — which is why replication collapses after about 1990 and the most recent decades on this page rest on a couple of hundred sites, not two thousand. Any apparent trend at the right-hand edge is a collecting artefact until proven otherwise.
Site chronologies were built by hundreds of different investigators using different standardisation choices. Pooling them assumes those choices do not bias the sign of a single-year departure. That is a reasonable assumption and not a tested one.
And a narrow ring is not a thermometer. It records that a growing season was poor for that tree, which may mean cold, or drought, or an insect year, or a fire, or a neighbour falling over.
| International Tree-Ring Data Bank — site chronologies, NOAA
Paleoclimatology ncei.noaa.gov/pub/data/paleo/treering/chronologies/ |
public domain |
| NOAA Paleoclimatology study metadata — site coordinates and species ncei.noaa.gov/access/paleo-search/ |
public domain |
| Global Volcanism Program, Smithsonian Institution — Holocene eruptions volcano.si.edu |
cited, not redistributed in bulk |
| Natural Earth 110m land outline naturalearthdata.com |
public domain |