sea ice, land, … Model improvements

CanSIPS development plans
Bill Merryfield
CCCma
CanSISE Workshop - 30 Oct 2013
Avenues for CanSIPS development
• Initialization improvements: sea ice, land, …
• Model improvements : all physical components + ESM
• System improvements: larger ensemble size, new
products, …
Initialization improvements
CanSIPS sea ice thickness initialization
• Based on relaxation to (not very realistic) model seasonal thickness
climatology
1 Mar 1981
1 Sep 1981
1 Mar 2010
1 Sep 2010
• Unlikely to accurately
capture thinning
trend
meters
Sea ice thickness on
first day of forecasts
(~initial values)
CanSIPS sea ice thickness initialization
• Based on relaxation to (not very realistic) model seasonal thickness
climatology
1 Mar 1981
1 Sep 1981
1 Mar 2010
1 Sep 2010
• Unlikely to accurately
capture thinning
trend
meters
Sea ice thickness on
first day of forecasts
(~initial values)
Ice extent trends:
HadISST1.1
= 0.55
NASA Bootstrap
CanSIPS fcsts = 0.36 !
NASA Bootstrap
CanSISE sub-project A2.3: Improved
CanSIPS sea ice initialization
Example of empirical relationships:
lagged correlations between volume
(as predictor) and extent (as predictand)
• Approach 2: allow thickness to set itself
in assimilating models, possibly with
corrections to reduce bias
Predictor: Sea Ice Volume (PIOMASS)
• Approach 1: find empirical relationships between ice thickness and
observables (e.g. September and current-month ice concentration),
based on models that validate best with observations
Predictor: Sea Ice Extent (NSIDC)
correlation
M. Chevallier, G. Smith
CanSIPS Land initialization (current)
Direct atmospheric
initialization through
assimilation of
6-hourly T, q, u, v
Indirect land
initialization through
response to model
atmosphere
www.eoearth.org/view/article/152990
Data Sources: Hindcasts vs Operational
*
*pending availability of CMC
NEMOVAR analysis
Change in atmospheric data source:
Effect on soil moisture
• Plots below compare soil moisture in first forecast month for ERA vs
CMC-based initialization
• VFSM = volume fraction of soil moisture (%)
• Anomalies are relative to 1981-2010 hindcast climatology
CanCM3
CanCM4
Global mean VFSM anomaly
Canada mean VFSM anomaly
ERA assimilation
CMC assimilation
CMC assimilation began 1 Jan 2010
Effects of soil moisture biases on forecasts
Mean differences in JJA forecasts for 2010-12 (lead 0)
Dots indicate statistical significance of CMC – ERA diffs according to t test
precipitation
2m temperature
C
plots by Slava Kharin
mm day-1
Problem solved using bias correction methodology of Kharin
& Scinocca (GRL 2012), but
 data constraint on land variables would eliminate such drifts
New approach: Constrain land variables with CaLDAS?
• CaLDAS = Canadian Land Data Assimilation System
• Will be global, available in real time

• Variable sets, ranges differ from CLASS  will need to map CaLDAS
variables into CLASS variables
• Would need to extend CaLDAS to cover hindcast period, ideally back to
1981 (proposed under CanSISE)
Model improvements
Atmospheric/land/earth-system model
development
Next CanSIPS target model: CanESM4.2?
• CLASS2.7  3.6: improved snow physics with liquid water component
• Additional snow model improvements (K. von Salzen talk)
• Interactive vegetation (CTEM = Canadian Terrestrial Ecosystem Model)
Drought  reduced leaf area index  reduced evapotranspiration
 persisted drought (positive feedback)
• Ocean ecosystem model (CMOC = Canadian Model of Ocean Carbon)
• Atmospheric and ocean physics improvements
Model resolution
CanSIPS
AGCM
CanSIPS
OGCM
Model resolution
CanSIPS
AGCM
NCEP
UK Met
CanSIPS
OGCM
NCEP
UK Met
Model resolution
CanSIPS
AGCM
NCEP
UK Met
CanSIPS
OGCM
NCEP
UK Met
CanSIPSv3?
or downscale using
Canadian Regional Climate Model?
CanCM3/4 ice model
resolution
• Sea ice and coastlines represented at AGCM resolution (128x64)
• “Pole problem” due to convergence of meridians
CanCM3/4 ice model
resolution
OPA/NEMO
ORCA1
resolution
CanCM3/4 ice model
resolution
OPA/NEMO
ORCA1
resolution
OPA/NEMO
ORCA025
resolution
CanCM3/4 ice model
resolution
OPA/NEMO
ORCA1
resolution
OPA/NEMO
ORCA025
resolution
Summary
• Near-term CanSIPS initialization improvement will focus on
land and sea ice thickness
• Near-term model development will focus on snow + ecosystem
components (+ atmosphere/ocean physics improvements)
• Longer-term model development will include coupling to
OPA/NEMO, which will vastly better resolve Arctic coastlines &
sea ice and eliminate pole problem
• Coupled GEM to be applied to seasonal prediction
 CanSIPSv2 = CanESM4.2 + coupled GEM?
• RCM downscaling for Canadian regions a possibility
Solution: Modify CMC-based assimilation runs using
bias correction method of Kharin & Scinocca (GRL 2012)
1. Extend ERA-based assimilation runs to mid-2012
2. From these runs make 6-hourly soil moisture time series from 1 Jan
2010
3. Repeat CMC-based assimilation runs, assimilating soil moisture from
assimilated ERA-based
model soil
ERA-based runs from step 2 using:
moisture
usual model equations
soil moisture
assimilation terms
4. Construct cyclostationary bias correcting forcing (“G”) from soil moisture
assimilation term:
mean annual cycle
The bias correcting term “G” is not a relaxation term. For a given grid point, it
only depends on the day of the year.