Predictability of Stratospheric Sudden Warmings in the Beijing Climate Center Forecast System with Statistical Error Corrections

Abstract

Previous studies have reported that the predictive limit of stratospheric sudden warming (SSW) events in the Beijing Climate Center forecast system (BCC_CSM) is shorter than 2 weeks. This study continues to analyze the general characteristics of this model in forecasting SSWs and carries out a trial of error corrections. The ratio of the ensemble members that forecast the zonal wind reversal with a 5-day delay allowed (hit ratio) is higher for SSW events with a small decrease in the zonal mean zonal winds (moderate SSWs) than for events with a large decrease in the zonal mean zonal winds (radical SSWs) in hindcasts initialized around 1 (D-7) and 2 (D-14) weeks in advance. The underestimated cumulative eddy heat flux associated with weak wave activities accounts for the weaker-than-observed deceleration of westerlies. The preexisting extratropical wave patterns are satisfactorily forecast in D-14 for most (9/12) cases, and the wave phase bias is reasonably small for those cases. After the climatology bias is deducted from the hindcasts, an increase in the hit ratio can be identified for moderate SSW events as the evolutions of zonal winds are improved. Following the error correction by remapping the zonal wind probability distribution function in the forecast system to the reanalysis, the SSW hit ratios increase in the D-7 (43% to 57%) and/or D-14 (11% to 21%) initializations. Based on the composite result, this error correction method robustly improves the hit ratio not only for both radical SSWs and moderate SSWs but also for both polar vortex split SSWs and polar vortex displacement SSWs. The best error correction method might propel the prediction limit of SSW events to around 2 weeks.

The 2019 New Year Stratospheric Sudden Warming and Its Real‐Time Predictions in Multiple S2S Models

“…For example, the SSW hit ratio for the 6 December initialization ensemble is higher (~50%) in those models than that in other models (<10%, below the 15–20% threshold in Rao, Ren, Chen, Liu, Yu, Hu, and Zhou ()). However, most models show a high hit ratio (>50%, exceeding the 15–20% threshold in Rao, Ren, Chen, Liu, Yu, Hu, and Zhou ()) for the 13 December initialization ensemble, which is common to nearly all models. Therefore, it is suggested that the average predictive limit for this SSW onset is at least 18 days.…”

“…The possible reasons for the longer predictability of this SSW event include the polar vortex displacement SSW type (for the initial SSW), the easterly QBO phase, and the MJO. The predictability of SSW events depends on the dominant wave number before the SSW onset (longer for wave‐1 type vortex displacement events than wave‐2 type vortex split events; see Taguchi, , ), and the boundary conditions (Butler & Polvani, ; Rao, Ren, Chen, Liu, Yu, Hu, & Zhou, ; Taguchi & Hartmann, ), and is longer by a few days when the MJO is strong (Garfinkel & Schwartz, ). While the initial wave‐1 disturbance is successfully forecasted at 18‐day leads, the splitting and persistence after onset are much more difficult to forecast in four different prediction MMEs initialized on 6, 13, 20, and 27 December. The predictability of this SSW onset depends on the model, but most models can forecast it at a lead time of 18 days, and some (e.g., NCEP, UKMO, and ECMWF) forecast it at 25‐day leads.…”

“…However, those results are obtained from monthly or even seasonal mean data, and SSW can also happen during the westerly QBO winter such as in February 2018. Based on the SSW events identified in the NCEP/NCAR reanalysis by Charlton and Polvani () for 1958–2001 and by Rao, Ren, Chen, Liu, Yu, Hu, and Zhou () Rao, Ren, Chen, Liu, Yu, and Yang () for 2002–2018, there are 37 SSW events during 1958–2019 in total (Table and Figure S1). The SSW frequency is 0.6 per year, consistent with the recent studies using other SSW definitions (Butler et al, ).…”

“…Based on a particular model, the average predictability of different SSW cases has also been studied (e.g., Karpechko, ; Marshall & Scaife, ; Rao, Ren, Chen, Liu, Yu, Hu, & Zhou, ; Rao, Ren, Chen, Liu, Yu, & Yang, ; Taguchi, , ). Among all the forecast systems, the JMA (Hirooka et al, ; Mukougawa et al, ; Mukougawa & Hirooka, ; Taguchi, , ) and ECMWF (Dörnbrack et al, ; Karpechko, ; Simmons et al, ) are two of the most widely used models for the prediction of SSW events, although the predictability of SSW events in other models such as CMA‐BCC (Rao et al, ; Rao, Ren, Chen, Liu, Yu, Hu, & Zhou, ; Rao, Ren, Chen, Liu, Yu, & Yang, ), NOGAPS (Kim et al, ; Kim & Flatau, ), HadGEM (Marshall & Scaife, ), ACCESS (Roff et al, ), NCEP‐CFS (Stan & Straus, ), and CMAM (Sigmond et al, ) have also been explored. For example, five SSW events were analyzed and their average predictability was 9–15 days in the HadGEM model (Marshall & Scaife, ); nine SSW event were analyzed and the average ratio of ensemble members that predict the SSW occurrence is 20%, 30%, and 70% at lead times of 15, 10, and 5 days, respectively, in the JMA model (Taguchi, ); the probabilistic prediction limit for a few SSW events is around 13 days ( p > 0.9), but this probability decreases to 0.3 for hindcasts initialized 13 days before the SSW central date in the ECMWF model (Karpechko, ).…”

Comparison of stratospheric evolution during the major sudden stratospheric warming events in 2018 and 2019

“…During major SSWs, the polar vortex moves southward from the North Pole or splits into two daughter vortices, which are always classified as “displacement‐type SSW” events and “split‐type SSW” events. The displacement‐type SSWs are generally believed to be caused by strong planetary waves with a zonal wavenumber of 1 (wave 1), whereas the split‐type SSWs are mainly due to the enhancement of planetary waves with a zonal wavenumber of 2 (wave 2; e.g., Charlton and Polvani, 2007; Karpechko et al, 2018; Rao J et al, 2018, 2019a, 2019b). The statistical characteristics of the planetary waves during major SSWs and the two vortex types have also been verified by modeling analyses (e.g., Cao C et al, 2019; Liu SM et al, 2019).…”