Abstract
An analogue forecast scheme is described for multifield prediction of monthly and seasonal New Zealand climate anomalies on the basis of the methodology of Livezey and Barnston (1988) for US seasonal temperatures. The method is applied to predicting terciles of temperature and precipitation for six regions of New Zealand. Empirical orthogonal function analysis is used to reduce sea surface temperature and sea‐level pressure predictors down to a set of five independent indices, which incorporate variations due to El Niño‐Southern Oscillation, Indian Ocean sea temperatures and a wave 3 pattern in the Southern Hemisphere westerlies. A full bootstrap cross‐validation procedure is carried out, along with Monte Carlo tests, to assess the skill of the method on independent data and to determine the significance of the results. Significant skill is found for seasonal temperature forecasts for the summer and winter seasons; there is less success in predicting monthly temperatures or rainfall at either timescale. Considerable care is required to constrain the climate state vector, from which analogues are defined, and to constrain the search procedure itself, in order to produce results that are stable with respect to small parameter changes in the model. For the New Zealand region, 5 to 7 is found to be the optimum number of ‘closest analogues’, and the inclusion of anti‐analogues improves the predictions, at least in the seasonal case. Skill in predicting regional temperature and rainfall is shown to be related to a combination of skill in predicting sea‐level pressure patterns and to how strongly these patterns project onto temperature and rainfall anomalies. Copyright © 2005 Royal Meteorological Society.