Hindcast predictions for malaria outbreaks during the epidemic seasons with the model with rainfall.

The second through fifth columns show the quantiles of simulated epidemics aggregated over September to December in each year, using initial conditions and rainfall covariates based on information available up to August. The column ‘aggregated observed cases’ shows the reported malaria cases accumulated over the same period. We declare a year to be epidemic if cases rose to more than the 75th percentile of aggregated observed cases ( = 2733) (bold black). We forecast an epidemic if the median of the prediction distribution exceeds this threshold. Equivalently, we can measure the forecast probability of exceeding this epidemic threshold (last column), predicting an epidemic if this probability exceeds 0.5. Four out of five epidemics were forecasted correctly (bold black). The false negative (underlined *) and two false positives (underlined @) all contain the actual presence/absence of an epidemic within their central 80% prediction intervals.