Table 1.

Retrospective and prospective predictive accuracy of climate-driven Lassa fever incidence model.

Model	RMSE (2012–2019)	RMSE (2016–2019)	RMSE (2020)	DIC (WS)
Baseline	3.287	4.328	3.932	5030.5
Climate-driven	2.915	3.483	4.518	4793.9
Proportion change in prediction error	−0.113	−0.195	0.149

The table shows the differences in the predictive performance of weekly LF cases by the baseline (random and reporting effects only) and best climate-driven model. We measured predictive performance as OOS RMSE, calculated for retrospective predictions across the entire study period (2012–2019) and following the widespread rollout of surveillance (2016–2019), and for prospective predictions for 2020. The climate-driven model also substantially improved within-sample model fit, measured using Deviance Information Criteria (DIC).