Investigation of the trigger perturbation for a simulated case of arctic oscillation transition using initial‐condition perturbation denial experiments

Abstract

This case study seeks to identify the initial-condition perturbation structure that triggers a prominent example of the positive-to negative-phase arctic oscillation (+AO to −AO) transition in a numerical forecast ensemble. Experiments with spectral filtering and perturbation-denial sensitivity analysis are used to test the hypothesis that the transition originated from a localized perturbation feature.The denial experiments are guided by binary search, anomaly detection, and adjoint sensitivity methods. The spectral filtering experiments indicate that perturbation scales in the wavenumber range v ∊ [10, 30] are critical to initiating the transition. Conversely, neither the largest perturbation scales (v < 10) nor the smallest scales (v > 30) have significant influence on the transition. The perturbation denial experiments likewise find no evidence to support the hypothesis that a localized perturbation feature initiates the transition. Rather, the transition is only substantially curtailed when the perturbation denial is applied across a broad portion of the domain. In effect, the trigger perturbation behaves like a redundant system, such that the initiation of transition does not critically depend on any singular perturbation feature. Practically, the redundant nature of the trigger means that if one wants to severely constrain the AO index forecast uncertainty such that its envelope falls within only one phase of the AO, then it will require very broad and unfeasible reduction of initial condition errors. However, the denial experiments also indicate that the triggering of transition and the subsequent amplification within the −AO phase are somewhat independent processes.