The Influence of Sampling Variability upon Model Output-based
Predictability Statistics
James A. Renwick and John M. Wallace
Submitted to Monthly Weather Review, 16 June 1995.
Abstract
The robustness of relationships between forecast model output and forecast skill is tested using a fourteen winter set of Northern Hemisphere 500-mb geopotential height analyses and forecasts produced by the European Centre for Medium-Range Weather Forecasts operational model. During the period of record, model skill improved substantially at the medium-range, as the model itself went through many upgrade cycles. It is found, through independent trials using sub-samples of the full 14 winter record, that apparently useful forecast correction and skill prediction relationships are sensitive to sampling variability and are of limited use in an operational setting. Interannual variability and non-stationarity in the time series of forecasts both appear to contribute to the lack of robustness. The occurrence of non-occurrence of El Nino conditions appears to have a large effect on the form of many of the results.
The amplitude of mean errors associated with the day-10 forecast show no dependence on the initial analysis polarity of the Pacific/North American (PNA) pattern, and the day-10 forecast amplitude of the PNA pattern appears to be of little use in the prediction of forecast skill. One relatively stable result is found: errors in predicting upper-level ridges or blocks over the Alaskan region are influential in determining the hemispherically-integrated rms erros at the medium range, throughout the entire period of record. Some implications of these results for the re-analysis and possible re-prediction of meteorological fields at global forecasting centers are discussed.
Compressed tar file (1.4 Mbytes) containing PostScript files of the manuscript and figures.
Corresponding author: James A. Renwick (renwick@wcd1.kelburn.cri.nz)
[Home][Up]