Application of Markov Chain Model to Long-Range Temperature Prediction

Despite progress in numerical weather prediction (NWP) over the last half century, the upper limit of deterministic forecast of day-to-day weather is at most 10 days for current NWP models. Recent studies indicate that general circulation models (GCMs), though capable of providing useful information to extended climate outlook over a broad area, generally underestimate the seasonal forecastability of the atmosphere as a result of their inability to reproduce adequately the atmospheric response to sea-surface temperature anomalies and their exaggeration of the effects of the chaotic behavior of the atmosphere. Transforming predicted large-scale atmospheric features to regional scales by GCMs has become even more difficult. Because of these limitations, statistical forecast models continue to be valuable supplements to the dynamic models. An exploratory study is under way to evaluate the application of Markov Chain model to TVA long-range temperature prediction over the Tennessee Valley on a seasonal-to-annual time scale. Monthly temperature records starting from 1871 at specified locations are used as the primary data base, supplemented by a variety of other time series data sets. The purpose of the study is to determine if the Markoc Chain model, given its customized flexibility in both time and space, can provide competitive skills in long-range temperature prediction.