An archaeological site can compress several different worlds into a few metres of earth. A Mesolithic camp may sit below the remains of Neolithic farmers; a Bronze Age cemetery may cut into both. Add an ancient climate reconstruction and the sequence becomes hard to resist. Did warmer or colder conditions help move societies along that historical path?
In a previous post, I asked a version of that question, but the first models left an obvious objection. Ancient people were not scattered randomly across Europe. They arrived in related populations, often in waves of migration, and many genomes came from the same handful of sites. A climate result could therefore be an ancestry result, a site result, or a mixture of both.
I have now rerun the analysis with a stricter model of genetic similarity and a separate effect for archaeological site. The main warming pattern remains. People sampled after greater warming over the preceding thousand years were more likely to fall into earlier archaeological-stage categories. The pattern appears in two differently constructed datasets and is clearest for annual and winter temperature.
This does not mean that one degree of warming knocked a society back by one historical age nor does it prove that climate caused a transition. It is an association in an uneven ancient record. Even so, it survived the most serious statistical challenge to the earlier result.
A two thousand year dating error
An earlier version of this project had a dating error in my extraction of CHELSA-TraCE21k, not in the climate dataset itself. I treated time code zero as the present, although it denotes 2,000 years before present (Karger, 2021, Table 4). That shifted both ends of each 1,000-year temperature comparison 2,000 years too far back. For a person dated to 5,000 years before present, the old code compared 7,000 with 8,000 years before present instead of 5,000 with 6,000.
I corrected the time-code mapping, rebuilt the climate values, and reran the archaeological-stage analyses before the ancestry-and-site models described here. The correction changed some coefficients and uncertainty estimates, but the broad warming association persisted. All figures and estimates in this post use the corrected dates. The original, misdated estimates should not be read as evidence about the intended millennium.
The historical ladder is only an ordering
The outcome in this analysis is not a numerical score for civilisation. It is an ordering of broad archaeological periods: Mesolithic, Early or Middle Neolithic, Late Neolithic or Chalcolithic, Bronze Age, and Iron Age or later. These labels describe archaeological packages that differ across regions.
The statistical model respects that limitation. It estimates whether an exposure shifts the probability toward an earlier or later category without pretending that the step from the Mesolithic to the Neolithic is equal to the step from the Bronze Age to the Iron Age. A negative climate coefficient points toward earlier categories; a positive one points toward later categories.
The climate exposure is also specific. For every ancient individual, I compared reconstructed temperature at the person’s place and date with temperature at the same location about one thousand years earlier. Warming and cooling were allowed to have separate slopes. This avoids a subtle trap: if one straight line is forced through both directions, evidence about warming automatically creates a mirror-image claim about cooling.
Figure 1 makes the comparison concrete. For one Bronze Age person at Cliffs End Farm, reconstructed annual temperature is 10.98°C about 1,000 years before the sample and 11.50°C at the sample date: a change of +0.52°C. Later Iron Age samples there follow cooling. The four panels also include weaker contrasts and a counterexample at Hasanlu, where later Iron Age samples follow warming. These selected records explain the exposure; they do not show a society moving backward or estimate the adjusted model effect.
Figure 1. Warming and archaeological labels at four multi-period sites.
