< Back to Article

Anticipating epidemic transitions with imperfect data

Fig 5

Heat maps showing impact of reporting process parameters and infectious period on performance of EWS.

AUC values further from 0.5 imply better performance. For each reporting scenario, 1000 20 year long replicates of both the test (emerging) and null (stationary) SIR model are simulated using the Gillespie algorithm, for fixed model parameters see Methods. All EWS are then calculated for each replicate. To quantify ability to identify emergence, AUC values are calculated from the distributions of the rank correlation coefficient for each EWS, see Methods. The scales for both the overdispersion and reporting probability are logarithmic.

Fig 5