Headwater Box Plot: 126 Hour Lead

Once we go out to a lead time forecast of 126 hours we are solidly in phase 3 where climatologically-based QPF input has a dominant influence. When climatological QPF dominates we see that the ensemble forecast members are indicating conditional bias�too high on the low end and too low on the high end. There is also tendency for underspread on the high end as seen by the small vertical extent of the boxes. But bias is the more obvious characteristic.

The upper 20% of flow observations are often not captured by any of the ensemble members as seen in the box plots located mostly or entirely below the zero error line. This indicates underforecasting.

At the same time the low end forecasts have a tendency to be skewed above the zero line, indicating an overforecasting bias there. However, the low end values are not as badly overforecast as they were at lead hour 30. Let's zoom in to the area left of the blue line. As with the zoomed in box plot we examined for lead hour 30, this zoomed in plot will allow closer examination of the ensemble forecasts associated with the lower 80% of observed flows.

Box Plot of 126-hour Ensemble Forecast Errors Against Observed Value for Headwater Point at Williamsburg