On experiments using an auto-regressive moving-average model for predicting the monthly 50–100 kPa thickness anomalies

A series of 75 auto-regressive integrated moving-average (ARIMA) forecasts of monthly mean 100–50 kPa thickness anomalies with lead time of one month are produced for 455 grid points over the northern hemisphere. The results are indistinguishable from persistence forecasts when the scores are averaged over the domain. A more detailed view indicates some differences which may merit closer investigation. Differences in the behaviour of positive and negative anomaly forecasts are also described.Some aspects of the methodology are discussed briefly in the introductory sections. The model used in the experiment is described in detail. A shortcoming of the method is that it does not lend itself to seasonal stratification.