Spatial and temporal coherence of quasi-periodic components of meteorological fields as a basis for long-term weather forecasts

Abstract

A new method of teleconnections studding is proposed which is based on the identification of conjugate regions in the global meteorological fields of temperature and pressure by their characteristic coherent quasi-periodic oscillation. This method was implemented in order to select predictors of winter air temperature in Belarus with an advance of 2 months. The degree of coherence of sea level pressure and winter temperature in Belarus on a quasi-8-year cycle was considered as a criterion for the selection of predictors. The forecast was implemented using the advanced deep machine learning model TimesNet and showed rather high metrics of quality for seasonal meteorological forecasting: the correlation coefficient between actual and predicted temperature values was 0.66, and the weighted macro-average values of precision and recall of the forecast in the gradations “normal”, “above normal” and “below normal” were 0.61 and 0.56, respectively.