196
Chapter 10: The Evaluation of Forecasts
tation of climatologies frQm the subsampie and recalculation of anomalies
(i.e. departures from climatology). Otherwise, inflation of skill or certain
degeneracies will occur in the scoring (see Van den Dool, 1987, and Barnston
and Van den Dool, 1993, for examples).
Finally, forecast year data in each iteration must be independent of the
developmental sampie. To ensure this a forecast year predictor or predictand
should not be serially correlated with itself in the years reserved for model
building. Suppose, for example, that for a particular predictand adjacent
years are serially correlated but every other year is effectively not. In this
instance three years must be set aside in each iteration and a forecast made
only for the middle withheld year. To determine the effective time between
independent sampies, refer to the procedures outlined in the previous chapter.
Chapter 10: The Evaluation of Forecasts
tation of climatologies frQm the subsampie and recalculation of anomalies
(i.e. departures from climatology). Otherwise, inflation of skill or certain
degeneracies will occur in the scoring (see Van den Dool, 1987, and Barnston
and Van den Dool, 1993, for examples).
Finally, forecast year data in each iteration must be independent of the
developmental sampie. To ensure this a forecast year predictor or predictand
should not be serially correlated with itself in the years reserved for model
building. Suppose, for example, that for a particular predictand adjacent
years are serially correlated but every other year is effectively not. In this
instance three years must be set aside in each iteration and a forecast made
only for the middle withheld year. To determine the effective time between
independent sampies, refer to the procedures outlined in the previous chapter.