-Ensemble forecast -Dealing with model drift obs. climatology Forecast simulations How do we make seasonal forecasts?
obs All Hindcast runs: ~ 12 members 14 years (96-09) Hindcast mean How do we make seasonal forecasts?
obs Hindcast mean Forecast members How do we make seasonal forecasts?
Bias corrected forecast obs How do we make seasonal forecasts?
Skill of seasonal forecasts Between 1990s and late 2000s, skill had barely improved, still minimal outside tropics
Change of strategy Hindcast length Frequency of system upgrades Centre's priorities JMA ~ 30 yr? ? NCEP ~ 40 yr8 yr Link to re-analysis ECWMF ~ 25 yr5 yr Med-range UK Met Office * 14 yr1 yr Link to model development * Arribas et al., 2011: GloSea4 ensemble prediction system for seasonal forecasting. MWR. 139,
Improved GloSea5 Range: 7months Model: HadGEM3 Resolution: N216L85O(0.25)L75 Initialisation: daily, NWP state + NEMOVAR 0.25deg Method: Full field + hindcast bias correction Ensembles: SKEB2 stoch. physics, lagged initialisation Forecasts: Seasonal 2 day -1 : 42 members updated weekly Monthly 2 day -1 : 28 members updated daily Reference: MacLachlan et al., 2013, in prep.
Atmos & land surf: ERA-i Ocean: Seasonal ODA reanalysis Atmos & land surf: NWP anal Ocean/sea-ice : NEMOVAR A coupled system for med-range and ocean forecasting 60 days6 months 6 th Dec th Jan Forecasts Hindcasts 15 days - 24 members - IC perts. from MOGREPS-G - Coupled model as in GloSea5 - Run once a day - Comparison PS32 – PS34 - Operational end13 or 14? New suite:
GloSea5: Skillful NAO predictions! Scaife et al., 2013 Source: MetOffice.gov.uk R=0.60 NAO: November forecast for Dec-Jan-Feb observations forecast
PMSL anomalies from Nov for DJF 05/06 to 09/10 winters
Reasons for skilful NAO forecasts Gulf Stream Bias Wly wind bias => Blocking Deficit No Gulf Stream Bias No Wly wind bias => Good Blocking in N. Atl New Model Scaife et al., Geophys. Res. Lett., Low-res: 1 deg ocean High-res: 0.25 deg ocean
- Courtesy of Warwick Norton -
Not December! As found by Warwick Norton, there is very little skill in December Not a simple persistence of anomalies over the timescale of a few weeks Hints at a possible mechanism in late winter: ENSO/stratosphere => strongest NAO (Ineson and Scaife 2008) From Nick Dunstone
Looking at stratosphere: Data available L38 L85 GloSea4: N96 GloSea5: N
SSWs in L38 and L85 From David Fereday
Obs No Strat Moderate NinoStrong Nino From Anna Maidens
A quick diversion: How good are ENSO forecasts in GloSea5?
Improving ENSO forecasts Obs The westward extension of Nino is a common error in many climate models. It affects remote regions. High-res model has better ENSO pattern and teleconnections Low resolution High resolution
Conclusions - GloSea5: High skill predicting winter NAO - Why? High horizontal res.; sea-ice; stratosphere - Skill highest in late winter, role of stratosphere? - Consistent with ENSO/Europe teleconnection through stratosphere and better SSW
Thanks Any questions?
Tropical storm tracks: impact of resolution 1990s ~240 km horizontal resolution 2000s ~120 km horizontal resolution Observations
Today: GloSea5 ~53 km horizontal resolution Observations Tropical storm tracks: impact of resolution