Using Self-Organizing Maps as a Forecasting Tool

Abstract

Some extreme weather events, such as the early season heavy snow and cold weather outbreak of early November 2014, can be traced back to the influence of tropical or extratropical cyclones on the planetary scale flow. Such planetary scale reorganization also occurs in conjunction with serial extratropical cyclogenesis. Potential temperature on the dynamic tropopause (defined by the 2 PVU surface) allows for a dynamically compact characterization of the flow. NCEP Climate Forecast Systems Reanalysis data spanning 32 years are used to provide this measure, and Self-Organizing Maps (SOM) are then constructed to identify our atmospheric regimes. Key elements of this analysis are the transitions between SOM regimes, which provide means for identifying increased regime predictability at medium and extended ranges. In this study, it was found that 30 regimes defined through the 32 year period were subjectively reasonable for characterizing the variety of hemispheric flow patterns that are observed. The probability of transitions between these regimes over certain time scales (e.g., 10 days, 20 days, and 30 days) was estimated with these same data. This analysis revealed a statistically significant tendency (at the 95% confidence interval) for recurrent patterns at the 30 day lead time, which presents some additional information that may be used in the context of extended range forecasting.