Figure 3
From: Land-surface initialisation improves seasonal climate prediction skill for maize yield forecast
Skill of the seasonal climate forecasting system. (a) Reliability diagrams for low yielding events (CSI low , i.e. below the 25th percentile) for May and June runs, driven by seasonal climate forecasts with climatological and realistic land-surface initialisations. The coloured lines show the linear weighted regression with the associated 75% confidence level (shaded areas). The number of samples for each bin is shown in the lower right sharpness diagram. The reliability diagram has been calculated by grouping the seasonal CSI low forecasts for all countries having significant predictive performance under observed climate conditions. The horizontal and vertical lines indicate the climatological frequency of the events in the observations and forecasts, respectively. The grey area defines a region where seasonal CSI low forecasts contribute positively to the forecast skill with respect to the climatology (the area where the Brier Skill Score is greater than 043). The no skill line separates skilful regions from unskilful ones in the diagram. The deviation from the diagonal provides the conditional bias; the flatter the curve, the less resolution it has (i.e. lower ability of the system to produce reliable forecasts that differ from the naive probability). (b) ROC diagrams for low yielding events (i.e. below the 25th percentile) for May and June forecasts of CSI, driven by seasonal climate forecasts initiated by climatological and realistic land-surface initialisation. The shaded regions indicate the 75% confidence intervals calculated through 1000 bootstrap replications. The hit rate and false alarm rate values consider a set of probability forecasts by stepping a decision threshold with 20% probability through the forecasts. Each ROC diagram displays as well the ROCSS values for both INIT and CLIM forecast experiments.
