Figure 2

Accuracy of the outbreak detection, depending on the number of individuals N in the sample. In these and further plots, results are averaged over 1000 measurements for each dataset and sample size N. Error bars indicate the standard error. A The distance error Δx between true and estimated location decreases quickly with growing sample size N. B The method is able to infer the correct outbreak location with high accuracy \(P_{\mathrm{loc}}\), even for a small number of input trajectories. C The temporal error Δt decreases fast for the synthetic datasets, but less so for the empirical datasets, which we attribute to repeated behavioral patterns in the datasets’ populations (see main text). The dashed line at \(\Delta t=42\mathrm{h}\) marks the expected error for a random guess of the outbreak time within the sample interval of 7 days. D The accuracy \(P_{\mathrm{time}}\) of the time inference increases more slowly, and the accuracy is generally lower for the empirical datasets, which we again attribute to repeated behavioral patterns