For a better understanding and explanation of the above results, another issue related to the postprocessing algorithm design is quickly reviewed here—a comparison between the decaying average and equal weight bias estimate approaches. The equal weight approach also makes a first-moment bias calculation over some previous days but with equal weighting for each day.
We applied the equal weight bias estimate method to two seasons of the 2004 operational NCEP/GEFS ensemble and compared the results with the decaying weight OPR_DAV2%. Results from the equal weight and decaying weight are very similar (less than 2% weight for a longer forecast; not shown).
The reasons to choose the decaying weight and not the equal weight include (a) the decay method having a higher weight for the latest information, which is good for a flow-dependent system (short-term forecast), and (b) the application of the decay weight method being operationally cost effective.
There is no need to save extra data on the central computer system, and bias estimates can include more historical information through continuous updates once the latest analysis is available.
In general, the result from the decaying average will be better than any single average (equal weight) method.