Utilizing long memory and circulation patterns for stochastic forecasts of temperature extremes

TL;DR

This paper develops nonlinear stochastic models incorporating long memory and circulation patterns, notably the Arctic Oscillation, to improve subseasonal-to-seasonal temperature extreme forecasts, demonstrating significant predictive improvements over traditional models.

Contribution

It introduces a data-driven method combining fractional calculus and stochastic difference equations to model temperature with long memory and external circulation influences.

Findings

Long memory models outperform traditional models in temperature forecasting.

Forecasts achieve up to 20 days lead time for maximum temperature.

AO index significantly influences winter temperatures in southern Scandinavia.

Abstract

Long memory and circulation patterns are potential sources of subseasonal-to-seasonal predictions. Here, we infer one-dimensional nonlinear stochastic models of daily temperature which capture both long memory and external driving by the Arctic Oscillation (AO) index. To this end, we employ a data-driven method which combines fractional calculus and stochastic difference equations. A causal analysis of AO and North-Atlantic Oscillation indices and European daily extreme temperatures reveals the largest influence of the AO index on winter temperature in southern Scandinavia. Stochastic temperature forecasts for Visby Flygplats, Sweden, show significantly improved performance for long memory models. Binary temperature forecasts show predictive power for up to 20 (11) days lead time for maximum (minimum) daily temperature (66% CI) while an AR(1) model possesses predictive power for 8 (3) days lead time for daily maximum (minimum) temperature (66% CI). Our results show the potential of long memory and circulation patterns for extreme temperature forecasts.