Enhanced seasonal forecast skill following SSWs Michael Sigmond

Enhanced seasonal forecast skill following SSWs Michael Sigmond (CCCma) John Scinocca, Slava Kharin (CCCma), Ted Shepherd (Reading) Part I: Scinocca et al. poster (today 4: 30 pm) Are SSWs associated with enhanced forecast skill in dynamical forecast systems, and if yes, how can it be quantified? Dyn. Var/SNAP Workshop, Reading, UK, 22 -26 April 2013

Introduction • Skillful seasonal forecasts rely on predictability of slowly varying components of climate system (SSTs, soil moisture) • Due to the limited influence of ENSO, forecast skill in NH extratropics is relatively small Forecast skill for February mean ST (issued Jan 1) Courtesy: B. Merryfield • Additional predictability may be realized by exploiting long time scales variations that are introduced by SSWs

Predictability introduced by SSWs Weak, warm vortex Mean surface conditions after SSW (day 15 -60) Surf. Temperature Precipitation Sigmond et al. (2013), Thompson et al. (2002) • Averaged surface conditions after SSWs characterized by more blocking and equatorward shift of storm tracks (negative NAM) Composites NAM around SSWs Baldwin and Dunkerton, 2001 • Long timescale disturbances in lower stratosphere influences troposphere for up to 2 months

Can potential predictability associated with SSWs be realized in dynamical forecast models ? • Sigmond et al. (2013): yes, but important to realize that: – SSWs are only predictable up to 1 -2 weeks in advance – Potential predictability is highly conditional (i. e. , only after SSW) – Seasonal forecasts will only benefit from SSWs when they are initialized close to an observed SSW

Method: • Tool: Dynamical seasonal forecast system that includes a wellresolved stratosphere (CMAM, T 63 L 71) • Experiments: Retrospective ensemble forecasts initialized at the onset date of all 20 observed SSWs (1970 -2009) • Initial model states: - Taken from state at the date of the SSWs in 10 assimilation runs, which are nudged towards time-evolving ERA reanalyses - Provides a consistent way (balanced fields) to initialize the land atmosphere above ERA • Forecast skill metric: Anomaly correlation skill score (linear dependency between observational and model anomalies, day 16 -60)

Model captures observed surface response Surf. Temperature Observations Forecast Precipitation

Forecast skill following SSWs Forecast NAM 1000 h. Pa • Significant forecast skill of the surface circulation in SSW runs • What part of the skill can be attributed to SSWs? • Perform ‘control’ forecasts that are not initialized during SSWs (40 forecasts, same calendar dates as SSWs, in year prior and following SSW) • Skill difference between SSW and control runs is due to SSWs

Forecast skill enhancement following SSWs NAM 1000 h. Pa No forecast skill of surface NAM in control runs Skill in SSW-runs comes entirely from SSWs are associated with significant skill enhancement of surface circulation

Forecast skill enhancement following SSWs Forecast skill SLP • SSWs associated with significant skill enhancement of SLP

Forecast skill enhancement following SSWs Forecast skill SLP • SSWs associated with significant skill enhancement of SLP • What about other more socioeconomically relevant variables?

Forecast skill enhancement following SSWs Forecast skill ST • Significant skill enhancement of ST northern Russia and eastern Canada Forecast skill PCP • Significant skill enhancement of north Atlantic PCP

Conclusions: • Potential predictability associated with SSWs can be realized in dynamical seasonal forecast systems • Following SSWs we find enhanced forecast skill of SLP, ST and PCP • Follow up: How far in advance can SSWs be predicted and usefully add skill to tropospheric forecasts? • Practical suggestion: issue special forecasts (at nonstandard times) once a SSW has been identified in observations • Implication: Operational seasonal forecasts which happen to be initialized close to the onset of a SSW will yield enhanced forecast skill (e. g. , Jan 2013 SSW)

January 2013 SSW • January 2013 happened close to beginning of the month • Was the forecast for February more skillful than average?

Operational forecast for Feb. 2013 (issued January 1) Mean ST response after SSW EC Forecast

Operational forecast for Feb. 2013 (issued January 1) Mean ST response after SSW EC Forecast Observed anomaly 2013

Conclusions: • Potential predictability associated with SSWs can be realized in dynamical seasonal forecast systems • Following SSWs we find enhanced forecast skill of SLP, ST and PCP • Follow up: How far in advance can SSWs be predicted and usefully add skill to tropospheric forecasts? • Practical suggestion: issue special forecasts (at nonstandard times) once a SSW has been identified in observations • Implication: Operational seasonal forecasts which happen to be initialized close to the onset of a SSW will yield enhanced forecast skill (e. g. , Jan 2013 SSW)

EXTRA SLIDES

Part I: Methods that don’t work (poster, Sigmond et al. , in prep) • Strat. HFP runs (hindcasts initialized on Nov 1, high and low top CMAM): – SSW climatology is more realistic in the high top model, but specific SSW events are not captured (in high and low top models) Standard set of Strat. HFP runs do not benefit from SSWs, and can not provide quantitative estimate of forecast skill enhancement associated with SSWs • Nudged stratosphere runs: – Enhanced forecast skill scores in DJF NH – But enhanced skill scores not limited to region and season with SSWs Synchronization of SSWs with observations by stratospheric nudging can not isolate the influence of SSWs on seasonal forecast skill

Ensemble spread (NAM)

Forecast skill metric • • • After initialization, models tend to drift from observations to their mean behavior/climatology (which is often biased) Solution: statistical bias correction: from many simulations started from the same calendar date, calculate the average bias/drift (function of forecast lag) Problem with simulations started from non-standard calendar dates (such as hindcasts initialized during SSWs): bias correction is usually not known Statistical bias corrected MSE can not be determined Sigmond et al. (2013) focussed on anomaly correlation score, which measures the linear dependence between anomalies (deviations from climatology) in observations and the forecast model Model anomaly is calculated relative to the climatology of the freely running model In the first 15 days, the model drifts from observations to the mean behavior of the freely running (AMIP) runs following SSWs discard the first 15 days and focus on days 16 -60