Fig. 3: Comparing time series cross-validated predictions of candidate models.

Figure showing time series cross-validated posterior predictions of dengue cases for each model from 2009 to 2022. We used an expanding window cross-validation methodology, where the model is trained on data up to but not including the target week and then posterior predictions are generated for the target week. Coloured lines show the median posterior prediction of weekly dengue cases, shaded areas show the 95% prediction interval and the dark grey lines show the data. From top to bottom the figure shows: predictions for the final selected climate and serotype model with weekly and yearly random effects \({\gamma }_{a[t]}+{\delta }_{w[t]}\) in purple; predictions for a climate only model with weekly and yearly random effects in pink; predictions for a serotype only model with weekly and yearly random effects in green; and predictions from a seasonal baseline model with only weekly random effects \({\delta }_{w[t]}\) in orange.