Journal cover Journal topic
Nonlinear Processes in Geophysics An interactive open-access journal of the European Geosciences Union
Journal topic

Journal metrics

Journal metrics

  • IF value: 1.699 IF 1.699
  • IF 5-year value: 1.559 IF 5-year
    1.559
  • CiteScore value: 1.61 CiteScore
    1.61
  • SNIP value: 0.884 SNIP 0.884
  • IPP value: 1.49 IPP 1.49
  • SJR value: 0.648 SJR 0.648
  • Scimago H <br class='hide-on-tablet hide-on-mobile'>index value: 52 Scimago H
    index 52
  • h5-index value: 21 h5-index 21
Volume 12, issue 6
Nonlin. Processes Geophys., 12, 1021–1032, 2005
https://doi.org/10.5194/npg-12-1021-2005
© Author(s) 2005. This work is licensed under
the Creative Commons Attribution-NonCommercial-ShareAlike 2.5 License.
Nonlin. Processes Geophys., 12, 1021–1032, 2005
https://doi.org/10.5194/npg-12-1021-2005
© Author(s) 2005. This work is licensed under
the Creative Commons Attribution-NonCommercial-ShareAlike 2.5 License.

  20 Dec 2005

20 Dec 2005

A comparison of predictors of the error of weather forecasts

M. S. Roulston M. S. Roulston
  • Department of Meteorology, Pennsylvania State University, University Park, USA

Abstract. Three different potential predictors of forecast error - ensemble spread, mean errors of recent forecasts and the local gradient of the predicted field - were compared. The comparison was performed using the forecasts of 500hPa geopotential and 2-m temperature of the ECMWF ensemble prediction system at lead times of 96, 168 and 240h, over North America for each day in 2004. Ensemble spread was found to be the best overall predictor of absolute forecast error. The mean absolute error of recent forecasts (past 30 days) was found to contain some information, however, and the local gradient of the geopotential also provided some information about the error in the prediction of this variable.

Ensemble spatial error covariance and the mean spatial error covariance of recent forecasts (past 30 days) were also compared as predictors of actual spatial error covariance. Both were found to provide some predictive information, although the ensemble error covariance was found to provide substantially more information for both variables tested at all three lead times.

The results of the study suggest that past errors and local field gradients should not be ignored as predictors of forecast error as they can be computed cheaply from single forecasts when an ensemble is not available. Alternatively, in some cases, they could be used to supplement the information about forecast error provided by an ensemble to provide a better prediction of forecast skill.

Publications Copernicus
Download
Citation