Our cross-validation design is standard [Michaelsen, 1987; Barnston, 1994; WMO, 2002] with correction for temporal autocorrelation. Each of the 51 years 1952-2002 (or 53 years 1950-2002) is forecast by formulating a model on all data excluding a 5 year block centred on the year of interest. Block elimination is employed to minimise potential skill inflation which may arise from the multi-annual persistence of predictor and/or predictand conditions. The 51 (53) hindcasts are compared with the 51 (53) observations to compute the prediction skill. Cross-validation provides a good estimate of the true forecast skill in large samples as here [eg Wilks, 1995].
- Barnston, A. G., Linear statistical short-term climate predictive skill in the northern hemisphere, J. Climate, 7, 1513- 1564, 1994.
- Michaelsen, J., Cross-validation in statistical climate forecast models, J. Climate Appl. Meteor., 26, 1589-1600, 1987.
- Wilks, D. S., Statistical Methods in the Atmospheric Sciences, Academic Press, 467pp, 1995.
- WMO, Standardised Verification System (SVS) for Long-Range Forecasts (LRF) New Attachment II-9 to the Manual on the GDPS (WMO-No. 485), Volume I, WMO, Geneva, 2002.
