Forecasts of Time Averages with a Numerical Weather Prediction Model

View More View Less
  • 1 Scripps Institution of Oceanography, La Jolla, CA 92093
© Get Permissions
Full access

Abstract

Forecasts of time averages of 1–10 days in duration by an operational numerical weather prediction model are documented for the global 500 mb height field in spectral space. The error growth of this model is compared to the error growth in a simple idealized model. Useful forecast intervals for time averages are about twice the useful forecast intervals of instantaneous events. The skill of the forecasts for time averages can be increased still further by ignoring numerical weather prediction model forecasts of instantaneous events past certain stopping point. The window of useful forecast time, for the predictions of ten-day averages considered here, is approximately one week.

The documented transient spectra have the largest values in the planetary scales, along with the largest variability. The error spectra grow from a relatively flat initial spectra to an asymptotic spectral shape similar to the transient spectra. The largest increase in error occurs on the initial day. Temporal variations in the numerical weather forecasts have a large high frequency component. Some notable systematic errors are present; when these errors are removed the time of useful skill for daily forecasts is improved by 6–12 hours and the time of useful skill for forecasts of time averages is improved by 1–2 days. Statistical filters also improve the forecasts although, except for removing systematic errors, they are not likely to prove useful for independent forecasts.

Abstract

Forecasts of time averages of 1–10 days in duration by an operational numerical weather prediction model are documented for the global 500 mb height field in spectral space. The error growth of this model is compared to the error growth in a simple idealized model. Useful forecast intervals for time averages are about twice the useful forecast intervals of instantaneous events. The skill of the forecasts for time averages can be increased still further by ignoring numerical weather prediction model forecasts of instantaneous events past certain stopping point. The window of useful forecast time, for the predictions of ten-day averages considered here, is approximately one week.

The documented transient spectra have the largest values in the planetary scales, along with the largest variability. The error spectra grow from a relatively flat initial spectra to an asymptotic spectral shape similar to the transient spectra. The largest increase in error occurs on the initial day. Temporal variations in the numerical weather forecasts have a large high frequency component. Some notable systematic errors are present; when these errors are removed the time of useful skill for daily forecasts is improved by 6–12 hours and the time of useful skill for forecasts of time averages is improved by 1–2 days. Statistical filters also improve the forecasts although, except for removing systematic errors, they are not likely to prove useful for independent forecasts.

Save