Model results. (a) The empirical distribution of GCV scores for the 104 climate-driven sylvatic plague models, the best 500 of which are marked in red. (b) The effect of increasing the number of models included in the sylvatic plague index Y on its correlation (D) with human plague 1904-1948. D is not sensitive to how many models are included, as long as the best 500 are. (c) The (log
) time series of sylvatic plague abundance (P, black; broken when no plague observed, i.e., P = 0, despite continued sampling of hosts), and the estimated plague abundance (Y, red line). The recorded human plague cases (blue bars) and the predicted human plague from Y (eq. 10, broken red-yellow line) are shown on a linear scale. (d) The black line shows estimated climate forcing on plague (Y) over the past 1500 years, with 95% quantiles in gray and multi-frequency (2-60 years) Gaussian moving average (red). The blue lines mark the long-term (2-400 years) multifrequency mean, maximum (upper broken line), minimum (lower broken line) and sum of minimum and maximum (solid line). The periods leading up to the Justinian plague (1), Black Death (2), Pandemic (3) and the Manchurian epidemics (4) are shaded in blue. The index (W) of conflict between Chinese and nomad societies is shown above the extent of the tree-ring index (T, green), the glacial series (Gann and Gdec, blue), and the decadal coverage in the monsoon proxy (S, brown).