Tracing the origin and spread of agriculture in Europe

Abstract

The origins of early farming and its spread to Europe have been the subject of major interest for some time. The main controversy today is over the nature of the Neolithic transition in Europe: the extent to which the spread was, for the most part, indigenous and animated by imitation (cultural diffusion) or else was driven by an influx of dispersing populations (demic diffusion). We analyze the spatiotemporal dynamics of the transition using radiocarbon dates from 735 early Neolithic sites in Europe, the Near East, and Anatolia. We compute great-circle and shortest-path distances from each site to 35 possible agricultural centers of origin--ten are based on early sites in the Middle East and 25 are hypothetical locations set at 5 degrees latitude/longitude intervals. We perform a linear fit of distance versus age (and vice versa) for each center. For certain centers, high correlation coefficients (R > 0.8) are obtained. This implies that a steady rate or speed is a good overall approximation for this historical development. The average rate of the Neolithic spread over Europe is 0.6-1.3 km/y (95% confidence interval). This is consistent with the prediction of demic diffusion (0.6-1.1 km/y). An interpolative map of correlation coefficients, obtained by using shortest-path distances, shows that the origins of agriculture were most likely to have occurred in the northern Levantine/Mesopotamian area.

Figure 1. Location of the 735 Archaeological Sites Used in the Analysis as well as the Ten POAs Listed in Table 1

Figure 2. Linear Regression Fits to the Data (n = 735 Sites) for Uncalibrated Dates in Years BP and Distances of Sites Computed from the POA with the Highest R-Value in Table 1

(A) Based on great-circle distances. The speed implied by the distance-versus-time regression is the slope of the dashed line, namely 0.71 ± 0.04 km/y (in agreement with statistical theory, the error range of 0.04 km/y has been computed as twice the standard error of the slope and corresponds to a 95% confidence interval). The speed implied by the time-versus-distance regression (full line) is the inverse of the corresponding regression slope, namely 1.04 ± 0.05 km/y (95% confidence interval). Therefore, we estimate the overall speed range as 0.7–1.1 km/y. If calibrated dates are used in the analysis (top axis), the result is 0.6–1.0 km/y (see the first figure in Protocol S1). (B) Based on shortest-path distances. The distance-versus-time regression yields 0.85 ± 0.04 km/y, whereas the time-versus-distance regression yields 1.22 ± 0.06 km/y. The overall estimated speed range is thus 0.8–1.3 km/y. If calibrated dates are used (top axis), the result is 0.7–1.1 km/y (see the second figure in Protocol S1).

Figure 3. Interpolation Map of R-Values of HOAs (n = 765 Sites)

Using great-circle distances (A) and shortest-path distances (B), these maps are based on uncalibrated dates and a slightly larger number of sites than those used in Figures 1 and 2 and Table 1. As a consistency test, the dataset now includes 30 Arabian sites. However, the results for the speed range are very similar to those obtained for the set of 735 sites. In addition, the use of calibrated dates does not lead to substantial changes in the maps (see the third and fourth figures in Protocol S1).