Developing the next generation ECCC seasonal forecasting system Hai Lin RPN, ECCC FORCAII, Beijing Climate Center April 24-26, 2017 Acknowledgements: contributions of many RPN/CMC and CCCma colleagues 

Outlines Early generations of the ECCC seasonal system Current operational system: CanSIPS Contribution to NMME The GEM coupled model Hindcast verifications Summary

Seasonal forecasts in CMC Sept. 1995 – 2003: based on 12-member ensemble of 2 atmospheric models (GCM2, SEF) (2-tier system). Multi-model approach: 1 climate models and 1 NWP models; Forecast range limited to 3 months 2003 – Nov. 2011: based on 40-member ensemble of 4 atmospheric models (GCM2, GCM3, SEF, GEM) (2-tier system). Multi-model approach: 2 climate models and 2 NWP models; Forecast range limited to 4 months Since Dec. 2011: based on 20-member ensemble of 2 coupled atmosphere-ocean-land physical climate models (CanSIPS); Multi-model approach: 2 climate models; Forecast range up to 12 months. Pioneered in MME 3

Current Seasonal System (CanSIPS) 2-coupled models (CCCma models) AGCM3 (T63L31) -> CanCM3 AGCM4 (T63L35) -> CanCM4 coupled with CanOM4 (1.41°× 0.94° / L40) ocean model 1-tier system 10 members for each model 1-12 month forecast issued every month Hindcast period: 1980-2010 4

CanSIPS contribution to the NMME

NMME Model History Forecast release months  Aug 2011 Aug 2012 Year 1 NCEP CFSv1 NCEP CFSv2 IRI ECHAM4f IRI ECHAM4i GFDL CM2.1 NASA GEOS5 NCAR CCSM3 CMC1/CanCM3 CMC2/CanCM4 NCEP CFSv1 Year 1 Year 2 Aug 2013 Aug 2014 GFDL FLOR NCAR CCSM4 Year 3 Year 4

User/applications Community NMME Operational Centers Research Centers NCEP EC GFDL NASA NCAR 8th of each month Hindcasts Real-time Forecasts Research Community User/applications Community

Niño3.4 Plumes (Nov-Jun 2014-15) Ensemble means for each model

Real time verifications – summary statistics init Aug 2011 – Jul 2012 init Aug 2012 – Jul 2013

Plan for next system Multi-model (currently already CanCM3+CanCM4) Climate model + NWP Development of GEM coupled model

Coupled GEM model Coupled system of Atmosphere + Land Ocean Sea ice

Atmospheric Model Global Environmental Multiscale (GEM) Horizontal resolution: 256 x 128 (~1.4° × 1.4° ) 79 levels, top at 0.075 hPa Time step: 1 hour Land surface scheme: ISBA Deep convection scheme: Kain-Fritsch (KFC)

Ocean + Sea ice NEMO Horizontal resolution: 1° × 1° , 1/3 degree meridionally near the equator 50 levels Time step: 30 minutes coupled with sea ice --- CICE

GEM-NEMO Seasonal System Two components: 1) Hindcast (model statistics, verification) 2) Real time forecast 14

Hindcast: initial conditions Atmosphere: ERA-interim pefield of GEPS to generate 10 members (random isotropic perturbations) Ocean: ORAP5 from ECMWF: ocean T, S, H, U, V Land: off-line SPS forced by ERA-interim atmosphere Sea ice concentration: ORAP5 Sea ice thickness: ORAP5

Hindcast 12 month integrations 10 members 31 years of 1980-2010

Real-time Start at beginning of each month, 10 members 12 month integrations Atmosphere: 10 members from ENKF of GEPS Land: SPS forced by CMC analysis Ocean: CMC GIOPS Sea ice concentration: CMC GIOPS Sea ice thickness: CMC GIOPS

May 1 Start

DJF T2m lead=1 month

JJA T2m lead=1 month

Nov 1 Start

DJF PR lead=0

JJA PR lead=0

Lead months Lead months Lead months Lead months

NAO skill 1 for JFM, 2 for FMA, etc

PNA skill

MJO skill

Seasonal forecast system Current operational system (CanSIPS): CM3+CM4 Possible next system option 1: GEM+CM4 option 2: GEM+CM3+CM4

TAS, Land, seasonal mean, vs. ERA-Interim GEM+CM4 CanSIPS GEM GEM+CM3+CM4 CM4 CM3

TAS, N.America, seasonal mean, vs. ERA-Interim

PR, Globe, seasonal mean, vs. GPCP2.3

PR, N.America, monthly mean, vs. GPCP2.3

Summary ECCC continues to develop MME in seasonal forecasting GEM in general has a higher skill for T2m and PR than CM3 and CM4. CM3 has the least skill. Better MJO and NAO skill in GEM Combining GEM with CM4 gives better skill than CanSIPS. Adding CM3 does not improve the skill.