Systematic and Random Error in an Extended-Range Forecasting Experiment

View More View Less
  • 1 Canadian Climate Centre, Downsview, Ontario, Canada
© Get Permissions
Full access

Abstract

A dynamical extended-range forecast consisting of a series of six monthly forecasts for each of the eight Januarys from 1979 to 1986 is carried out with the Canadian Climate Centre low-resolution general circulation model. Results, in terms of the 500-rnb height field, are presented for the mean January systematic and random error and for the skill of the January mean forecast. The evolution of error as a function of time and spatial scale is investigated and the equations governing the growth of systematic and random error are derived and evaluated.

The mean January systematic error in the 500-mb height field is modest and is largely in other than the zonal structures. Systematic-error variance is about an order of magnitude smaller than random-error variance and is concentrated in the larger scales of the flow. The January mean anomaly correlation indicates marginal forecast skill, as well as some connection between the spread of the forecasts and the skill of the average forecast.

The budget equation for random error indicates that the interaction between it and the systematic error is small so that, for this model at least, the removal of systematic error by subtraction is plausible. The nonbarotropic source-sink term dominates random-error growth early in the forecast, while nonlinear barotropic generation does so at later times. The growth of the (smaller) systematic-error variance is importantly affected by its interaction with random error at early forecast times and also by nonlinear barotropic generation at later times. The relative sizes of the nonlinear barotropic generation and baroclinic source-sink terms, together with the interactions between the two forms of error, may reveal differences between model behavior and suggest areas of improvement.

Abstract

A dynamical extended-range forecast consisting of a series of six monthly forecasts for each of the eight Januarys from 1979 to 1986 is carried out with the Canadian Climate Centre low-resolution general circulation model. Results, in terms of the 500-rnb height field, are presented for the mean January systematic and random error and for the skill of the January mean forecast. The evolution of error as a function of time and spatial scale is investigated and the equations governing the growth of systematic and random error are derived and evaluated.

The mean January systematic error in the 500-mb height field is modest and is largely in other than the zonal structures. Systematic-error variance is about an order of magnitude smaller than random-error variance and is concentrated in the larger scales of the flow. The January mean anomaly correlation indicates marginal forecast skill, as well as some connection between the spread of the forecasts and the skill of the average forecast.

The budget equation for random error indicates that the interaction between it and the systematic error is small so that, for this model at least, the removal of systematic error by subtraction is plausible. The nonbarotropic source-sink term dominates random-error growth early in the forecast, while nonlinear barotropic generation does so at later times. The growth of the (smaller) systematic-error variance is importantly affected by its interaction with random error at early forecast times and also by nonlinear barotropic generation at later times. The relative sizes of the nonlinear barotropic generation and baroclinic source-sink terms, together with the interactions between the two forms of error, may reveal differences between model behavior and suggest areas of improvement.

Save