The Canadian Numerical Weather Prediction
System: Present and Future
Gilbert Brunet
Recherche en Prévision Numérique (RPN)
Meteorological Research Branch
Meteorological Service of Canada
Environment Canada
Québec, Canada
Thursday, November 1, 2001, CAS 2001 meeting, Annecy, France
Introduction

The main mandate of the Meteorological Research Branch (MRB) is to
improve the operational Numerical Weather Prediction (NWP) and data
assimilation system at the Canadian Meteorological Center (CMC) of the
Meteorological Service of Canada (MSC)

We have to integrate in the NWP system observation instrument systems that
has a maximum impact on improving prediction skill, like satellites, and radar
for very-short-range forecasts

Environmental Prediction: We have to couple our NWP system with ocean,
hydrological and chemistry models
RPN - Recherche en Prévision Numérique
Many major world-wide innovations













During 1960’s and 1970’s
Semi-implicit method, Robert-Kwizak
First integration of spectral model, Robert
Optimum Interpolation, Rutherford
Operational spectral model, Daley-Girard
Finite Element model, Staniforth-Daley
During 1980’s and 1990’s
Semi-Lagrangian technique, Robert et al.
First operational TKE PBL, Benoit et al.
Ultra-fast FFT’s (Cray,CDC), Temperton
First SI-SL fully non-hydrostatic model Tanguay, Laprise , Robert (later MC2)
Unified GEM (global, uniform or variable resolution), Staniforth, Côté,
Gravel et al.
MC2 is internationally recognized for mesoscale modeling, Benoit et al.
Increasing computer power

Increasing computer power





1960 ’s - Bendix G20, IBM370.
1970’s: Control Data 7600, Control Data 176
1980’s: Cray 1S, Cray XMP-2/8, Cray XMP-4/16
1990’s: NEC SX-3/44, SX-3/44R, SX-4/64M2 and SX-5/32M2
2000’s: Requirement for a new contract & new HPC systems
Supercomputing Cluster
Front Ends
SGI
O2000's
28 PEs
(MIPS R10K)
1.2TB
10x FC
dual
attach
0.7 TB RAID
NEC
SX-4/64M2
FC switched 32 ports
IXS (16 GB/s)
IXS (8GB/s)
NEC
SX-5/32M2
2x ea
HIPPI
RAIDs
(0.8 TB)
24x
7x ea.
Climate Archive
O2000 4 x
R10K PEs
1.2 TB
14x SCSI
from OSS
Central File
Server
145 TB
4 DST drives
20 MB/s ea.
SGI O2000: 4 PEs
Max Strat
Gen5
RAID
0.7 TB
4xFC
ADIC AML-E (Tape Robot)
HIPPI 1000 Mb/s switched
128 ports. Each host has
links to ops & dev networks
Trend in skill 1958-2000
Different problems need different
modeling approaches and physics
ONGOING RESEARCH at RPN (in collaboration with Canadian
and International Institutions)


Global predictions need an uniform space grid model with a good climatic balance
between dynamics and physics (Collaboration with CCMA/Victoria)
Regional predictions need a variable space grid model with improved implicit
physics, like Kain-Fritsch deep convection scheme
(Collaboration with McGill U. and Cloud Physics/SMC,Toronto)

Montainous and high resolution predictions of precipitation need a non-hydrostatic
and limited area model with explicit physics, like Kong-Yau. (Collaboration with
Mesoscale Alpine Project (MAP), McGill U. )



Environmental prediction needs to couple the NWP system with hydrology , and
ocean, and chemistry and wave models. (Collaboration with York U., Dalhousie U.,
Waterloo U., McGill U., Atlantic and Ontario SMC/Regions and Canadian Space
Agency)
Middle atmosphere capability for integrating Canadian space and ground-based
measurements and chemical modelling activities, like ozone (Collaboration with York
U. and Canadian Space Agency)
Multi-seasonal forecasting needs a low resolution model with a particular attention to
surface forcing, like sea temperature and soil moisture (Collaboration with
CCMA/Victoria and McGill U.)
Different problems need different
modeling approaches and physics
Variational data assimilation, 3Dvar and
4Dvar, needs the development of highly
sophisticated numerical tools that are tied
closely to the model dynamics and
physics
 Adaptation to new computer architecture
is highly model dependant.
 Global Environmental Multiscale (GEM):
An integrating concept, one tool for
different problems.
 Leader: J. Côté and S. Gravel

S
P
A
C
E
S
C
A
L
E
Global Environmental Multiscale Forecasting & Modeling System
2001-2006
Middle Atmosphere Model
&
Data assimilation
Multi- Seasonal Forecast
Monthly Forecast
Medium-range Forecast
( 240 h 35-100 km )
&
Data assimilation
Ensemble Forecast
Regional and
Mesoscale Forecast
( 24-48 h 10-24 km )
&
Data assimilation
Regional Climate Model
Limited-area Environmental Multiscale Model
0-24h 1-4km
TIME SCALE
Uniform resolution
Variable resolution
Hydrostatic
Nonhydrostatic
Global
Limited-area
Distributed memory
--------------------------------3D Var Data Assimilation
4D Var Data Assimilation
Operational forecast
Emergency response
Volcanic ashes
Air quality
Stratospheric ozone
Wave model
Coupling to Hydrology
Coupling to Oceanology
Simulations
etc
Physical Processes
Leader: S. Bélair and J. Mailhot
Solar
Radiation
Stratiform Precipitation
Deep
Convection
M
H
E
Infrared
Radiation
Boundary-Layer Turbulence
Surface Processes
G
S. Bélair
Physics (2000-2005)
Global model with a 35km resolution with
an optimized Kain-Fritsch, Tremblay
(Cloud Physics/SMC) phase mixed scheme
and ISBA surface scheme that drives a
continuous data assimilation cycle
 Regional model with an optimized KainFritsch and Tremblay (Cloud Physics/SMC)
phase mixed scheme with a 10km
resolution that drives a continuous data
assimilation cycle

Non-hydrostatic modeling of a severe weather
squall line in Oklahoma
Radar reflectivity
of the squall line event
Cloud water/ice
Cloud water/ice
Hydrostatic
Non-Hydrostatic
Non-hydrostatic effect: important
for timing?
Regional GEM Model
(rough estimates for research)
Application:
Configuration:
Time to complete application (hrs)
Runs per week
Concurrent (simultaneous) runs
Number of ensemble members
Number of FLOPS (PetaFlops)
Memory needed (GB)
Volume of files archived (GB)
Regional GEM Model
2000/2001
20Km-L35-48Hr
0.75
30
1
1
2002/2003
10Km-L60-48Hr
1
30
1
1
2005/2006
10Km-L60-48Hr
1
30
1
1
0.046
7.3
1.1
1.48
50
7.5
1.6
50
7.5
GLOBAL GEM GRID AT 35km
An 35km global uniform resolution GEM with an improved physics and
cloud package for the Meteorological Service of Canada
Model
Results
Outgoing longwave
radiation for a 1-day
forecast valid at 0000
UTC 2 November 2000
using the proposed
high-resolution model
configuration
A SAT(OBS)
A
C
B
SAT(OBS)
OP
NEW
B
OP
NEW
SAT(OBS)
A) Mid-latitude synoptic-scale
systems over North America:
Better representation of cold
frontal convection and of
occluded cyclonic circulations
B) Convective activity over South OP
America:
More widespread continental
convective activity and
better representation of lowlevel clouds
C) InterTropical Convergence
NEW
Zone:
Better representation of
convective activity and of a
Typhoon over southeast Asia
C
Global GEM Model
(rough estimates for research)
4DVAR GEM Global Model
(rough estimates for research)
Application:
Configuration:
Time to complete application (hrs)
Runs per week
Concurrent (simultaneous) runs
Number of ensemble members
Number of FLOPS (PetaFlops)
Memory needed (GB)
Volume of files archived (GB)
4DVAR Global GEM Model
2000/2001
2002/2003
2005/2006
50Km-L35-120Hr
24
0.5
1
1
35Km-L60-240Hr
24
2
1
1
18Km-L70-240Hr
24
5
1
1
0.85
27
15
37.5
45
230
350
175
930
Community Of Mesoscale
Modeling(COMM) group
Leader: R. Benoit



Community model (MC2) support is essential in
order to partner effectively with universities and
benefit from funds provided through Canadian
Foundation Climate and Atmospheric Science
($10 000 000 per year)
Community model would be configured to focus
on region with potential active or extreme
weather events at 1-3km horizontal resolution.
MC2 is worldwide recognized as one of the most
computer efficient non-hydrostatic model
MC2 run at 2km horizontal resolution over Vancouver Island
17H forecast valid 26 June 1997 2000 UTC. Near surface flow (arrows with scale in knots in
lower left corner). Superimposed over topography (gray shades every 500m). Only one
arrow every other grid point for each directions is displayed.
(M. Desgagné)
Configuration: 1500 x 1300 x 31; 2880 time steps of 30 sec (24H); 14 PEs SX5; total memory: 46 Gb, wall
clock : 16 H, Flops rate : 29 Gflops
Limited-Area GEM Model for high resolution
meteorology (2000km X 2000km)
(rough estimates for research)
Application:
Configuration:
Time to complete application (hrs)
Runs per week
Concurrent (simultaneous) runs
Number of ensemble members
Number of FLOPS (PetaFlops)
Memory needed (GB)
Volume of files archived (GB)
Peak GFLOPS/S
High Res. Meteorology
2000/2001
2002/2003
2005/2006
2km L80 24H
48
0.5
1
1
1km L80 12H
24
1
1
1
1km L80 12H
12
2
1
1
7.52
52
8
28
19.5
70
11
463
19.5
70
11
463
Coupled Modeling for
Environmental Prediction
Leader: H. Ritchie
RPN Environmental Prediction and Coupled Modeling
group is supporting/conducting R&D based on coupling
a variety of numerical prediction models

Now feasible due to advances in numerical modeling
in various domains, together with advances in
computer power.


Very high level of collaboration with international and
Canadian university and institutions. We are the provider
of the numerical modeling and computer expertise

High potential for Canadian Foundation for Climate and
Atmospheric Science collaborations
Key projects








Atmosphere-hydrology Model (Waterloo U., IML, MAP,
Ontario/MSC Region, …)
Regional Ocean Modeling and Prediction (Dalhousie U., BIO,IML,...)
3-D ocean circulation models being coupled with MSC models for
atmosphere-ocean prediction (Dr. Greatbatch, Dalhousie U.,...)
Coastal Modelling Systems for Storm surge forecasts (Atlantic/MSC
region, Dr. Thompson, Dalhousie U.,...)
Atmosphere-wave Modeling (Atlantic/MSC Region,...)
St. Lawrence Estuary Models (IML)
Marine Environmental Prediction System: Coupled
atmosphere/ocean/biology/chemistry ecosystem model to be
developed for demonstration site for Lunenburg Bay, NS (Dalhousie
U., Bedford Institute of Oceanography,...)
Extra-tropical hurricane transition (Dalhousie U., McGill U., ...)
Marine Environmental Prediction
System (MEPS)

To establish demonstration site for
Lunenburg Bay, NS.
Goal: interdisciplinary marine
environmental prediction guided and tested
using advanced observing systems.
 Coupled
atmosphere/ocean/biology/chemistry
ecosystem model being developed.

26
The Dream
The Reality
Extratropical Transition
- Examining mid latitude transition of hurricanes and typhoons.
- Eventually to use two-way interactive atmosphere-ocean data assimilation and prediction
system for direct modeling.
10km/60 levels Kuosym/Sundqvist
E
D
(MC2, M. Desgagné) 24-72 H Forecast of relative vorticity at 25m (frame every hour) COMPARE
FLO
2km/40 levels Kong&Yau (MC2, M. Desgagné)
16-30 H Forecast of Relative Vorticity at 20m (frame every hour)
COMPARE
Limited-Area GEM Model for high resolution
meteorology (2000km X 2000km)
(rough estimates for research)
Configuration:
Time to complete application (hrs)
Runs per week
Concurrent (simultaneous) runs
Number of ensemble members
Number of FLOPS (PetaFlops)
Memory needed (GB)
Volume of files archived (GB)
Peak GFLOPS/S
2000/2001
2003/2004
2006/2007
4km L35 48H
6
5
1
1
2km L80 24H
6
2
1
1
2km L80 24H
6
4
1
1
0.6
8
2
28
10
70
8
463
10
70
8
463
Conclusion



Meteorological Research Branch used in general 40% of the computer
resources
Climate Research Branch used in general 40% of the computer
resources
Canadian Meteorological Center used in general 20% of the computer
resources
For (2001-2006) the R&D strategy




Global NWP with a MESOGLOBAL GEM (35km) with a lid at the
stratopause (1mb) with the Regional GEM physics package
Global NWP ensemble forecast with GEM-100km with an improved
physics package
Collaborating with CFCAS to improved our Regional and Local NWP
Collaborating with CFCAS and other partners for Environmental
Prediction (coupling with chemistry, hydrology and ocean)