Fig. 4.

Performance of weekly case forecasts from component forecasters and selected ensembles, along with component forecaster weights. Component forecasters that were given high weight at key times are highlighted. The top row shows the relative WIS of forecasts made each week. The second row shows the relative WIS over the 12 weeks before the forecast date, for forecasts of quantities that were observed by the forecast date. These scores, which are used to compute the component weights in the relative WIS weighted median ensemble, are calculated using data available as of the forecast date. The third row shows component forecaster weights for the post hoc weighted mean ensemble, and the bottom row shows the component model weights for the relative WIS weighted median ensemble; each component forecaster is represented with a different color. Over the time frame considered, 31 distinct component forecasters were included in this top-10 ensemble.