Canadian Centre for Climate Modelling and Analysis (CCCma)
Victoria, BC Canada
Environment Canada's seasonal
forecasts: Current status and
future directionsBill Merryfield
RPN Seminar, 4 Sep 2014
In collaboration with: G. Boer, G. Flato, S. Kharin, W.-S. Lee, J. Scinocca… (CCCma) M. Alarie, B. Archambault, B. Denis, J.-S. Fontecilla, J. Hodgson… (CMC)
Predictability and Prediction
Predictability and Prediction
CanSIPS development and operations
Seasonal forecasting methods
• Earliest standard: empirical/statistical forecasts
• Later standard: two-tier model ensemble forecasts
- model sea surface temperature (SST) prescribed
- used by EC from 1995 until 2011 (anomaly persistence SST)
- forecast range limited to 4 months
• Current standard: coupled climate model ensemble forecasts
- fully interactive atmosphere/ocean/land/(sea ice)
- SSTs predicted as part of forecast
- potentially useful forecast range greatly extended
Motivation for coupled vs
2-tier systemMar 2006
Apr 2006
May 2006
Jun 2006
Jul 2006
Oct 2006
Observed SST anomaly
…
“Forecast” (persisted) SST anomaly
Example: consider 2-tier forecast (persisted SSTA) from 1 April 2006
2-tier system with persisted SSTA cannot predict El Niño or La Niña
Coupled forecast system development
• 2006 Funding from Canadian Foundation for Climate and Atmospheric Sciences (CFCAS) to the Global Ocean-Atmosphere Prediction and Predictability (GOAPP) Network
• 2007-2008 Pilot project using existing AR4 model,
simple SST nudging initialization
• 2008-2009 Model development leading to CanCM3/4,
initialization development
• 2009-2010 Hindcast production
• Dec 2011 Operational implementation
The Canadian Seasonal to Interannual Prediction System (CanSIPS)
• Developed at CCCma
• Operational at CMC since Dec 2011
• 2 models CanCM3/4, 10 ensemble members each
• Hindcast verification period = 1981-2010
• Forecast range = 12 months
• Forecasts initialized at the start of every month
WMO Global Producing Centres for Long Range Forecasts
2-tier (atmosphere + specified ocean temps)
coupled (interactive atmosphere + ocean)
CanSIPS Models
CanAM3 Atmospheric model - T63/L31 (2.8 spectral grid) - Deep convection scheme of Zhang & McFarlane (1995) - No shallow conv scheme - Also called AGCM3
CanAM4 Atmospheric model - T63/L35 (2.8 spectral grid) - Deep conv as in CanCM3 - Shallow conv as per von Salzen & McFarlane (2002) - Improved radiation, aerosols
CanOM4 Ocean model - 1.410.94L40 - GM stirring, aniso visc - KPP+tidal mixing - Subsurface solar heating climatological chlorophyll
SST bias vs obs (OISST 1982-2009)
C C
J0-9J0-8J0-7J0-6J0-5J0-4J0-3J0-2J0-1J0
J0-9J0-8J0-7J0-6J0-5J0-4J0-3J0-2J0-1J0
J0-9J0-8J0-7J0-6J0-5J0-4J0-3J0-2J0-1J0
GEM
GCM2
J0-9J0-8J0-7J0-6J0-5J0-4J0-3J0-2J0-1J0
GCM3
SEF
Month 1 Month 2 Month 3 Month 4
Two-tier initialization (1990s-2011)
atmospheric analyses at 12-hour lags to 120 hours Forecasts
atmospheric models
CanSIPS initialization
assimilation runs
Ensemble member
Atmospheric assimilationSST nudgingSea ice nudging
forecasts
Impacts of AGCM assimilation: Improved land initialization
Correlation of assimilation run vs Guelph offline analysis
SST nudging + AGCM assimSST nudging only
Soil temperature(top layer)
Soil moisture(top layer)
Probabilistic soil moisture forecast Feb 2014 lead 0
1 Feb 2014
9 Feb 2014
28 Feb 2014
25 Feb 2014
Evidence CanSIPS soil moisture initialization is somewhat realistic
21 Jan 2014
Data Sources: Hindcasts vs Operational
(transitioning to daily CMC)
Previous default: Deterministic forecast map
• colours = tercile category of ensemble mean anomaly:
• Issues: - small differences in forecasted anomaly can lead to large differences in in map
- no probabilistic information (climate forecasts are inherently probabilistic)
- no guidance as to magnitude of anomaly, other than tercile category
below normal near normalabove normal
Previous default: Deterministic forecast map
• colours = tercile category of ensemble mean anomaly:
• Issues: - small differences in forecasted anomaly can lead to large differences in in map
- no probabilistic information (climate forecasts are inherently probabilistic)
- no guidance as to magnitude of anomaly, other than tercile category
below normal near normalabove normal
All-in-one probability mapsTemperature probabilities:
individual categories
ucalibrated
White = ‘equal chance’(no category > 40%)
Temperature probabilities: all-in-one
AboveNormal
NearNormal
BelowNormal
Advantages of calibrated probability forecasts
uncalibrated calibrated
• uncalibrated probabilities:
- high probabilities predicted far more frequently than observed
- overconfident, especially for precipitation and near- normal category
- near-normal grossly overpredicted
• calibrated* probabilities:
- much more reliable (forecast probability observed frequency)
- less overconfident
- near-normal less overpredicted
Temperature
perfect forecast
Brier skill score = 0
no resolution
*Kharin et al. , A-O (2009)
Advantages of calibrated probability forecasts
Precipitation
perfect forecast
Brier skill score = 0
no resolution
• uncalibrated probabilities:
- high probabilities predicted far more frequently than observed
- overconfident, especially for precipitation and near- normal category
- near-normal grossly overpredicted
• calibrated* probabilities:
- much more reliable (forecast probability observed frequency)
- less overconfident
- near-normal less overpredicted*Kharin et al. , A-O (2009)
uncalibrated calibrated
Calibrated probabilistic forecasts in the media
Aug 21, 2013 Sep 2, 2014
Current operational configuration
Day of month
Forecastmonths
Official forecast
Backup forecast
1 15 31123456789101112
7
27
Mid-month “preview” forecast(+ lead 0.5 months for BoM ENSO, WMO, APCC)
Fall/Winter/Spring/Summer WPM Briefingsled by Marielle Alarie
…(23 pp., Fr & En)
Daily seasonal forecasts JJA 2014 (unofficial)
Optimal combination = ?
Proposed operational configuration
Day of month
Forecastmonths
Official forecast
Backup forecast
1 15 31123456789101112
7
27
Mid-month “preview” forecast(+ lead 0.5 months for BOM ENSO WMO, APCC)
Benefits of multi-model ensemble (1)
• A desirable property (reliability) of prediction e.g. of ENSO indices is that Ensemble Spread RMSE
• Ensemble Spread << RMSE for each model individually overconfident
• Ensemble Spread RMSE for the two-model combination (except shortest leads)
Benefits of multi-model ensemble (2)Experiment: compare CanSIPS (10xCanCM3 + 10xCanCM4) vs 20xCanCM4 (Jan initialization only): 10xCanCM3 + 10xCanCM4
20xCanCM4
Temperature anomaly correlation:slight advantage for 20xCanCM4 (except lead 0)
Temperature mean-square skill score: big advantage for 10xCanCM3 + 10xCanCM4
Contributions to international forecast
compendia
WMO Global Producing Centres for Long Range Forecasts
2-tier (atmosphere + specified ocean temps)
coupled (interactive atmosphere + ocean)
Asia-Pacific Economic Cooperation (APEC) Climate Center (APCC)
• 7 models: CMCC, MSC_CanCM3, MSC_CanCM4, NASA, NCEP, PMU, POAMA
• month 1-3 and 4-6 probabilistic & deterministic forecasts at ~0.5-1 month lead
CanCM3 CanCM4
• Currently 8 models including CanCM3 and CanCM4
• Temperature forecast for SON 2014 lead 1 shown here
• Besides contributing to combined NMME forecast, enables comparisons between performance of different models
• Temperature anomaly correlation skills for SON lead 1 month shown here
CanCM3 CanCM4
ENSO/Nino Index Forecasts
UK Met Office decadal forecast exchange
http://www.metoffice.gov.uk/research/climate/seasonal-to-decadal/long-range/decadal-multimodel
UK Met Office decadal forecast exchange
http://www.metoffice.gov.uk/research/climate/seasonal-to-decadal/long-range/decadal-multimodel
Annual (12-month average) forecasts
CanSIPS Probabilistic forecast Verification (1981-2010 percentile) + ACC
201
120
12
201
320
14
AC
C skill
Annual T2m forecasts
climatological pdf
forecast pdf
Glo
bal
mea
n f
ore
cas
t v
s c
lim
ato
log
ical
PD
F
Annual Forecast Skills for CanadaDeterministic:
Anomaly correlationProbabilistic:
ROC area/below normal ROC area/above normal
January initialization
Area-averaged score, all initialization months
Climate Indices
CanSIPS ENSO prediction skill
lead 0lead 9
…
0.55 < AC < 0.84 at 9-month lead
Nino3.4 anomaly correlation skill:
Does this translate to long lead skill over Canada?
OISST obs
Oceanic Indices (http://ioc-goos-oopc.org/state_of_the_ocean/sur/)Pacific :1.Niño1+2 : SST Anomalies in the box 90°W - 80°W, 10°S - 0°.2.Niño3 : SST Anomalies in the box 150°W - 90°W, 5°S - 5°N.3.Niño4 : SST Anomalies in the box 160°E - 150°W, 5°S - 5°N4.Niño3.4 : SST Anomalies in the box 170°W - 120°W, 5°S - 5°N5.SOI : difference of SLP anomalies between Tahiti and Dawin6.El Niño Modoki Index (EMI) EMI = SSTA(165E-140W, 10S-10N)-0.5*SSTA (110W-70W, 15S-5N)-0.5*SSTA (125E-145E, 10S-20N Ashok, K., S. K. Behera, S. A. Rao, H. Weng, and T. Yamagata, 2007 : El Niño Modoki and its possible teleconnection. J. Geophys. Res., 112, C11007, doi:10.1029/2006JC003798.
Atlantic :1. North Atlantic Tropical SST index(NAT) ; SST anomalies in the box 40°W - 20°W, 5°N - 20°N.2. South Atlantic Tropical SST index(SAT) SST anomalies in the box 15°W - 5°E, 5°S - 5°N.3. TASI = NAT – SAT4. Tropical Northern Atlantic index(TNA) SST anomalies in the box 55°W - 15°W, 5°N -25°N.5. Tropical Southern Atlantic index(TSA) SST anomalies in the box 30°W - 10°E, 20°S - EQ.
Indian Ocean :1. Western Tropical Indian Ocean SST index (WTIO) : SST anomalies in the box 50°E - 70°E, 10°S - 10°N2. Southeastern Tropical Indian Ocean SST index(SETIO) : SST anomalies in the box 90°E - 110°E, 10°S - 0°3. South Western Indian Ocean SST index(SWIO) : SST anomalies in the box 31°E - 45°E, 32°S - 25°S4. Indian Ocean Dipole Mode Index (IOD) : WTIO - SETIO
Monsoon Indices
Pacific :
1. Western North Pacific Monsoon Index WNPMI = U850 (5ºN -15ºN, 90ºE-130ºE) – U850 (22.5ºN - 32.5ºN, 110ºE-140ºE) Wang, B., and Z. Fan, 1999: Choice of South Asian summer monsoon indices. Bull. Amer. Meteor. Soc., 80, 629–638.2. Australian Summer Monsoon Index AUSMI = U850 averaged over 5ºS-15ºS, 110ºE-130ºE Kajikawa, Y., B. Wang and J. Yang, 2010: A multi-time scale Australian monsoon index, Int. J. Climatol, 30, 1114-11203. South Asia Monsoon Index SAMI= V850-V200 averaged over 10ºN -30ºN, 70ºE-110ºE Goswami, B. N., B. Krishnamurthy, and H. Annama lai, 1999: A broad-scale circulation index for interannual variability of the Indian summer monsoon. Quart. J. Roy.. Meteorol. Soc., 125, 611- 633.4. East Asian Monsoon Index EASMI= U850(22.5°–32.5°N, 110°–140°E) - U850 (5°–15°N, 90°–130°E) Wang, Bin, Zhiwei Wu, Jianping Li, Jian Liu, Chih-Pei Chang, Yihui Ding, Guoxiong Wu, 2008: How to Measure the Strength of the East Asian Summer Monsoon. J. Climate, 21, 4449–4463. doi: http://dx.doi.org/10.1175/2008JCLI2183.1
Indian :
1. Indian Monsoon Index IMI=U850(5ºN -15ºN, 40ºE-80ºE) – U850(20ºN -30ºN, 70ºE-90ºE) Wang, B., R. Wu, and K-M. Lau, 2001: Interannual variability of Asian summer monsoon: Contrast between the Indian and western North Pacific–East Asian monsoons. J. Climate, 14, 4073–4090.2. Webster-Yang Monsoon Index WYMI=U850-U200 averaged over 0-20ºN, 40ºE-110ºE Webster, P. J., and S. Yang, 1992: Monsoon and ENSO: Selectively interactive systems. Quart. J. Roy. Meteor. Soc., 118, 877-926.3. All Indian Rainfall Index4. Indian Summer Monsoon Circulation Index
22.0%
26.1%
44.8%
• PDO index of PC of 1st EOF of North Pacific SST
• Comparison of obs and CanSIPS EOF patterns:
Pacific Decadal Oscillation (PDO)
Woo-Sung Lee plots
Obs
CanSIPS lead 0
CanSIPS lead 5
(a) Monthly Mean
AC
C
0.0
0.2
0.4
0.6
0.8
1.0
CanCM3 CanCM4 CanCM34
MEAN
0.6630.6930.703
(b) Seasonal Mean
AC
C
0.0
0.2
0.4
0.6
0.8
1.0
CanCM3 CanCM4 CanCM34
MEAN
0.6950.7210.731
Averaged PDO anomaly correlation skill for all initial months (1979-2010)
Woo-Sung Lee plots
Snow Prediction
Evidence CanSIPS snow initialization is somewhat realistic
Example: BERMS Old Jack Pine Site (Saskatchewan, Canada)
2002
-200
3CanCM3 assimilation runs CanCM4 assimilation runs
1997-2007 climatology vs in situ obs
Sospedra-Alfonso et al. , in preparation
3-category probabilistic forecast (left)
MERRA verification(right)
JFM 2012 (lead 0)
SWE (left)
2m temperature (right)
Anomaly correlation
JFM (lead 0)
Higher than for T2m in snowy regions!
SWE T2m
CanSIPS snow water equivalent (SWE)
forecasts & skill
Sea Ice Prediction
WMO Global Producing Centres for Long Range Forecasts
2-tier (atmosphere + specified ocean temps)
coupled (interactive atmosphere + ocean) interactive sea ice climatological sea ice
CanSIPS predictions (hindcasts)Predictions of Arctic sea ice area: Anomaly correlation skill
Trend included Trend removed
Skill of anomaly persistence “forecast” Value added by CanSIPS
Sigmond et al. GRL (2013), Merryfield et al. GRL (2013)
Regional verification of CanSIPS sea ice forecastsWoo-Sung Lee, CCCma/UVic
Subregions of the Arctic Oceanas defined by the Navy/NOAA Joint Ice Center
Example: Beaufort Sea
Monthly Climatology
Forecast time series (lead 0)raw values
anomalies
CanSIPS
persistence
Correlation skill
0
1
CanSIPS predictions (forecasts)Prediction of monthly Arctic sea ice extent from 1 June 2012
Aug 2012 ice concentrationsNASA Team
CMC - NASA TeamCMC - NASA Bootstrap
NASA BootstrapCMC
CanSIPS predictions (forecasts)What of we adjust for higher CMC ice cover?
Original prediction
Original predictionminus
mean(CMC-NSIDC)
sea ice forecasts aligned with North American Ice Service products
• Initially, attempt to develop probabilistic forecasts for freeze-up and breakup dates, e.g.
3%12%
20%32%
25%
8%
1-5Jun
6-10Jun
11-16Jun
16-20Jun
21-25Jun
26-30Jun• Will require
New bias correction methods, e.g. seasonal cycle mapping Historical verification data back to ~1981
Towards CanSIPSv2
CanSIPS Development Efforts
• Improved ocean initialization
• Improved sea ice initialization
• Improved land initialization based on EC’s Canadian Land Data Assimilation System (CaLDAS)
• Improved climate model components (atmosphere, ocean, land, sea ice)
• New coupled model based on MSC’s GEM weather prediction model
• Regional downscaling of global model forecasts?
Current CanCM3/4 ice model grid
OPA/NEMOORCA1 grid
OPA/NEMOORCA025grid
Planned CanSIPS ice/ocean model improvements
1 M
ar 1
981
1 M
ar 2
010
1 S
ep 1
981
1 S
ep 2
010
• Based on relaxation to (not very realistic) model seasonal thickness climatology
• Unlikely to accurately capture thinning trend
Sea ice thickness on first day of forecasts (~initial values)
meters
Current CanSIPS sea ice thickness initialization
Real-time sea ice thickness estimation through statistical relationships to observables
Arlan Dirkson, UVic grad student
Thickness reconstructions based on 3 SVD modes
Sep 1996
2012Sep
Experimental downscaling of CanSIPS forecasts
• CanRCM4 = Canadian Regional Climate Model version 4• CORDEX North America grid – 0.22 ~ 25 km resolution• RCM runs will be initialized from downscaled assimilation runs• Atmospheric scales > T21 spectrally nudged in interior domain• Global model output files = RCM input global, downscaled forecasts run concurrently
Soil moisture probabilistic forecast on CanSIPS global grid
Surface temperature on CanRCM4 0.22 CORDEX North America grid
Global vs regional model topography
Global model: x 300 km Regional model: x 25 km
Summary• CanSIPS has reliably produced EC’s seasonal forecasts to a range
of 12 months since December 2011
• Multi-model approach appears to have been justified
• CanSIPS contributes to many international forecast compendia
• Many new products are under development
• CanSIPS R & D includes development of improved and new models (including GEM/NEMO), improvements in initialization (e.g. sea ice thickness), and downscaling to 25 km resolution using CanRCM4
Research supported by: