r&d plans for water cycle prediction, hydrological modelling … - ec...
Post on 26-Jun-2020
2 Views
Preview:
TRANSCRIPT
R&D Plans for Water Cycle Prediction,
Hydrological Modelling and Forecasting
Vincent Fortin
Recherche en prévision numérique environnementale
Meteorological Research Division
Current team
Research scientists
• Vincent Fortin
• Murray Mackay
Research assistants
• Dominique Bourdin
• Milena Dimitrijevic
• Dorothy Durnford
• Sarah Dyck
Meteorologist
• Guy Roy
Post-docs
• Nasim Alavi (NSERC VF)
• Darwin Brochero (U. Laval)
• Christopher Subich (NSERC VF)
Graduate students
• Mabrouk Abaza (U. Laval)
• Kian Abbasnezhadi (U. Manitoba)
• Haesun Choi (U. Manitoba)
• Marie Courbariaux (AgroParisTech)
• Bruce Davison (McGill)
• Bruce Friesen (U. Manitoba)
Main Acronyms and Model Names
• CaPA: Canadian Precipitation Analysis
• CaLDAS: Canadian Land Data Assimilation System
• WATROUTE: not an acronym, but a routing model
• Land surface schemes:
– CLASS: Canadian Land Surface Scheme
– sloped CLASS: CLASS + horizontal flow parameterization
– ISBA: Interaction Soil-Biosphère-Atmosphère
– SVS: Soil and Vegetation Simulator
• SPS: CMC’s surface prediction system (LSS driver)
• MESH: Modélisation Environnementale – Surface et Hydrologie =
SPS + WATROUTE
• standalone MESH: sloped CLASS + WATROUTE in a standalone
application (no dependencies on CMC, runs on Linux and Windows)
R&D plan for
numerical hydrological prediction
• Continue to improve the Canadian Precipitation Analysis
• Rely on SPS platform for operational applications
• Take advantage of other CMC products:
– CaLDAS: soil moisture and temperature, snow depth, SWE
– High-resolution deterministic surface prediction system
– Meteorological ensemble prediction system
• Introduce more hydrology in land-surface models used
for weather forecasting (coupled modelling)
• Assimilate streamflow observations
• Provide boundary conditions:
– for NEMO ocean model (Great Lakes, Gulf, oceans)
– for H2D2 river hydrodynamic model
Outline
• Canadian Precipitation Analysis (CaPA)
• Hydrological modelling infrastructure
• Hydrological modelling R&D
• Hydrological forecasting R&D
• Water cycle prediction R&D
CaPA precipitation analysis system
• Near real-time quantitative precipitation estimation (QPE)
• Merges different sources of information on precipitation
• Uses background field from the GEM NWP model due to low
network density in most of Canada
• Optimal interpolation technique (aka residual kriging)
• Fully automated quality control algorithm
Surface network
Atmospheric model
Satellite
observations
RADAR
Page 7 – 22 January 2014
24h precipitation for Sept. 18, 2013
Analysis
Page 8 – 22 January 2014
Verification over June-August 2013
• Radar vs no radar, Prairies
CaPA priorities
• Assimilation of radar QPE
• Real-time bias correction of solid precipitation
measurements
• Assimilation of climate stations and cooperative
networks
Hydrological Modelling
Infrastructure
• Based on the coupling of existing numerical models
• Major components:
– land-surface model
– surface and
subsurface flow
parameterization
– simple river and
lake routing model
– bucket model
for groundwater
– efficient driver for
large grids
Land-surface
sheme
GEM
atmospheric
model
Routing model
Coupling with
GEM can be
one-way (offline)
or two-wayMESH
- 11 -
Quick comparison of
standalone MESH vs SPS+WATROUTE
Characteristic Standalone MESH SPS
LSS Sloped CLASS 3.6, PDMROF, PBSM,
Preliminary infiltration into frozen soils
Sloped CLASS 3.6, ISBA, SVS, VIC
Routing WATROUTE WATROUTE
Platforms Any recent Linux or Windows platform Linux and IBM P7 (only runs at CMC for now,
remote access is possible)
Forcing Can use CMC hourly forcing (GEM and
CaPA) or data from other sources, but
all data needs to be interpolated to the
model grid
Uses CMC operational products at their native
resolution (built-in spatial and temporal
interpolation and downscaling), as long as they are
in RPN standard file format
Land-surface
parameters
All parameters are in config files, can
be calibrated using external software,
e.g. Ostrich
Many parameters are found in look-up tables, can
be changed but code needs to be recompiled
Topology for routing SHED file, typically obtained using
Green Kenue
SHED file, typically obtained by processing
HydroSHEDS data (SRTM based)
Scalability Runs on a single CPU Highly scalable, MPI support for both
computations and I/O
Tiling Built-in support for arbitrary tiles Predefined tiles (vegetated, glacier, open water,
frozen water, urban)
Priorities for Hydrological Modelling
Infrastructure
• Validation of sloped CLASS 3.6 implementation in SPS
against standalone MESH
• Merging standalone MESH and SPS development
streams
• First operational implementation over the Great Lakes
• Coupling of SPS to HYDROTEL, as part of the St.
Lawrence Action Plan
- 13 -
Hydrological Modelling R&D
- 14 -
Overlake evaporation prediction
• Deacu, Fortin et al. (2012), Journal of Hydrometeorology
Average latent heat flux, winter 2011 (W/m²)
GEM 15km GEM 10km OAFlux
Lake Superior supplies
200
150
100
50
0
W/m²
- 15 -
Streamflow prediction, Great Lakes
(sloped CLASS + WATROUTE)
Haghnegahdar, A., B. Tolson, B. Davison, F. Seglenieks, E. Klyszejko, E.D. Soulis, V. Fortin and L.S. Matott (2013). Calibrating
Environment Canada's MESH Modelling System over the Great Lakes Basin. Atmosphere-Ocean, submitted for publication.
- 16 -
Great Lakes Runoff Intercomparison
Project - Lake Michigan (GRIP-M)
- 17 -
Priorities for
Hydrological Modelling R&D
• Imroving the hydrology in SVS, largely based on
CLASS’s GRDRAN and WATROF subroutines
• Stream temperature modelling
• Great Lakes Runoff Intercomparison Project
Hydrological
Forecasting R&D
• Prévision d'ensemble du débit émise le 30-09-2010,
Rivière Chaudière à Saint-Lambert (20 membres)
enveloppe de l'ensemble
membre de l'ensemble
débit obserbé
ensemble forecast
observed total rainfall x
tiré de Abaza et al. (2012)
GEM est en retard mais
fini par avoir la bonne quantité
La prévision de débit montre le même retard mais permet quand même
d'émettre une alerte d'inondation avec une probabilité de 70%
inond. mineure: ~50%
risque d'inondation majeure
le 2 octobre: ~5%
inondation moyenne: ~15%
EnKF performance vs human expert
• Rivière au Saumon
Ass. manuelle par un expert
Aucune assimilation des débits
EnKF performance vs human expert
• Rivière au Saumon
Ass. auto. des débits (EnKF)
Ass. manuelle par un expert
Aucune assimilation des débits
Pour des prévisions 36h à l’avance ou plus,
le système automatique est aussi efficace qu’un humain
Priorities for Hydrological
Forecasting R&D
• Monthly ensemble forecasting system for the Great
Lakes and St. Lawrence (~1500 km² / 30 days / 20
members)
• High-resolution deterministic forecasting system for the
Pacific Cordillera (~1 km² / 2-3 days)
• Ensemble Kalman Filter for streamflow assimilation
Water Cycle Prediction R&D
H2D2: Suivi des niveaux, des
masses d'eau et des polluants
MESH:
débits en rivière
GEM:
conditions atmosphériques
NEMO: température, courants
niveau et couvert de glace des lacs
Page 23 – 22 January 2014
Hydrological prediction using MESH:
International Upper Great Lakes Study
Monthly water supply to Lake Superior Jun 04 – May 09
Impact of lakes on the weather:
Winter precipitation – DJF 05-09
Models improved
thanks to monitoring
of evaporation
mm
/day
MESH prediction (P-E+Runoff)
Observations Loss through evaporation (-E)
Net precipitation (P-E)
- 24 -
Prediction of monthly NBS
P
E
R
- 25 -
- 26 -
Current areas where CMC’s MESH
system is being implemented
- 27 -
Longer term developments:
the Arctic basin
from Gustafsson et Isberg (2013)
Priorities for
Water Cycle Prediction R&D
• Deterministic two-way coupled atmosphere, surface,
hydrology, lake, and ice forecasts
• Daily analysis of components of the water cycle for the
Great Lakes and St. Lawrence (precipitation,
evaporation, soil moisture, snow water equivalent,
runoff, streamflow)
• Canada-wide assessment of these components
• Runoff prediction for the Arctic basin
Canadian Precipitation Analysis (CaPA)
Vincent Fortin
Recherche en prévision numérique environnementale
Meteorological Research Division
CaPA precipitation analysis system
• Near real-time quantitative precipitation estimation (QPE)
• Merges different sources of information on precipitation
• Uses background field from the GEM NWP model due to low
network density in most of Canada
• Optimal interpolation technique (aka residual kriging)
• Fully automated quality control algorithm
Surface network
Atmospheric model
Satellite
observations
RADAR
Page 31 – 22 January 2014
24h precipitation for Sept. 18, 2013
Observations
Page 32 – 22 January 2014
24h precipitation for Sept. 18, 2013
Observations + radar
Page 33 – 22 January 2014
24h precipitation for Sept. 18, 2013
Analysis
Page 34 – 22 January 2014
24h precipitation for Sept. 18, 2013
Forecast
Page 35 – 22 January 2014
24h precipitation for Sept. 18, 2013
Analysis
Page 36 – 22 January 2014
Verification over June-August 2013
• Radar vs no radar, Canada-wide
Page 37 – 22 January 2014
Verification over June-August 2013
• Radar vs no radar, Prairies
A more quantitative look at bias
Carrera, Marco L., Stéphane Bélair, Vincent Fortin, Bernard Bilodeau, Dorothée Charpentier, Isabelle Doré, 2010: Evaluation of
Snowpack Simulations over the Canadian Rockies with an Experimental Hydrometeorological Modeling System. J. Hydrometeor, 11,
1123–1140.
ANG vs Nipher or RCS obs.
Winter 2010-2011, Prairies Equitable threat score Partial mean
Assimilation of
bias-corrected
observations
(Nipher+Geonor):
Assessment of
bias and skill
largely depends
on what is chosen
as “truth”
GEM
CaPA
Obs (Nipher)
ANG
vs Nipher
ANG
vs RCS
New C-SPICE site at Forêt
Montmorency Observations émises le 21-01-14 à 7h
Précipitations reçues Neige (cm) Pluie (mm)
Depuis le dernier relevé 0 0
Totales* 252 118
Présente au sol 90 --
* Depuis le 30/10/13
Hydrological modelling infrastructure
Vincent Fortin
Recherche en prévision numérique environnementale
Meteorological Research Division
Hydrological modelling
infrastructure
• Based on the coupling of existing numerical models
• Major components:
– land-surface model
– surface and
subsurface flow
parameterization
– simple river and
lake routing model
– bucket model
for groundwater
– efficient driver for
large grids
Land-surface
sheme
GEM
atmospheric
model
Routing model
Coupling with
GEM can be
one-way (offline)
or two-wayMESH
- 44 -
MESH
- 45 -
MESH
SPS
WATROUTE
- 46 -
SPS: Surface Prediction System
- 47 -
Quick comparison of
standalone MESH vs SPS+WATROUTE
Characteristic Standalone MESH SPS
LSS Sloped CLASS 3.6, PDMROF, PBSM,
Preliminary infiltration into frozen soils
Sloped CLASS 3.6, ISBA, SVS, VIC
Routing WATROUTE WATROUTE
Platforms Any recent Linux or Windows platform Linux and IBM P7 (only runs at CMC for now,
remote access is possible)
Forcing Can use CMC hourly forcing (GEM and
CaPA) or data from other sources, but
all data needs to be interpolated to the
model grid
Uses CMC operational products at their native
resolution (built-in spatial and temporal
interpolation and downscaling), as long as they are
in RPN standard file format
Land-surface
parameters
All parameters are in config files, can
be calibrated using external software,
e.g. Ostrich
Many parameters are found in look-up tables, can
be changed but code needs to be recompiled
Topology for routing SHED file, typically obtained using
Green Kenue
SHED file, typically obtained by processing
HydroSHEDS data (SRTM based)
Scalability Runs on a single CPU Highly scalable, MPI support for both
computations and I/O
Tiling Built-in support for arbitrary tiles Predefined tiles (vegetated, glacier, open water,
frozen water, urban)
- 48 -
A new community modelling
paradigm for MESH
• Now • Vision
MSC MSC
Hydrological Modelling R&D
Vincent Fortin
Recherche en prévision numérique environnementale
Meteorological Research Division
- 50 -
Improving the modelling of
hydrological processes in SVS
- 51 -
Parameterization for field capacity
and lateral flow
• Soulis, Craig, Fortin et Liu (2011), Hydrol. Proc.
WATDRAIN
Field capacity at 4 feet, estimated or measured
Fie
ld c
ap
acity p
red
icte
d
- 52 -
Overlake evaporation prediction
• Deacu, Fortin et al. (2012), Journal of Hydrometeorology
Average latent heat flux, winter 2011 (W/m²)
GEM 15km GEM 10km OAFlux
Lake Superior supplies
200
150
100
50
0
W/m²
- 53 -
Carrera et al., 2010: Evaluation of
Snowpack Simulations over the Canadian
Rockies with an Experimental
Hydrometeorological Modeling System. J.
Hydrometeor, 11, 1123–1140.
Impact of horizontal resolution
SWE evaluation vs
snow pillow data
• Grid size matters
• Simple downscaling
helps (e.g. constant
lapse-rate)
- 54 -
Use of hydrological observations
for model verification
Bow River at Calgary
mean annual flow (▀)
confirms results
obtained using
precipitation gauges
and snow pillows
Carrera et al., 2010: Evaluation of
Snowpack Simulations over the Canadian
Rockies with an Experimental
Hydrometeorological Modeling System. J.
Hydrometeor, 11, 1123–1140.
- 55 -
Streamflow prediction, Great Lakes
(sloped CLASS + WATROUTE)
Haghnegahdar, A., B. Tolson, B. Davison, F. Seglenieks, E. Klyszejko, E.D. Soulis, V. Fortin and L.S. Matott (2013). Calibrating
Environment Canada's MESH Modelling System over the Great Lakes Basin. Atmosphere-Ocean, submitted for publication.
- 56 -
Great Lakes Runoff Intercomparison
Project - Lake Michigan (GRIP-M)
Hydrological Forecasting R&D
Vincent Fortin
Recherche en prévision numérique environnementale
Meteorological Research Division
Prévision d'ensemble
• Prévision d'ensemble du débit émise le 30-09-2010,
Rivière Chaudière à Saint-Lambert (20 membres)
enveloppe de l'ensemble
membre de l'ensemble
débit observé
ensemble forecast
observed total rainfall x
tiré de Abaza et al. (2012)
GEM est en retard mais
fini par avoir la bonne quantité
La prévision de débit montre le même retard mais permet quand même
d'émettre une alerte d'inondation avec une probabilité de 70%
inond. mineure: ~50%
risque d'inondation majeure
le 2 octobre: ~5%
inondation moyenne: ~15%
Ensemble vs
deterministic forecasting
• CRPS vs MAE
Velázquez, J.A., Th. Petit, A. Lavoie, M.-A.
Boucher, R. Turcotte, V. Fortin et F. Anctil (2009).
An evaluation of the Canadian global
meteorological ensemble prediction system for
short-term hydrological forecasting, Hydrology and
Earth System Sciences, 13: 2221-2231.
EnKF performance vs human expert
• Rivière au Saumon
Ass. manuelle par un expert
Aucune assimilation des débits
EnKF performance vs human expert
• Rivière au Saumon
Ass. auto. des débits (EnKF)
Ass. manuelle par un expert
Aucune assimilation des débits
Pour des prévisions 36h à l’avance ou plus,
le système automatique est aussi efficace qu’un humain
Water Cycle Prediction R&D
Vincent Fortin
Recherche en prévision numérique environnementale
Meteorological Research Division
Great Lakes Environmental Prediction System Deacu, D., V. Fortin, E. Klysejko, C. Spence et P.D. Blanken (2012). Predicting the Net Basin Supply of the Great Lakes
with a Hydrometeorological Model. Journal of Hydrometeorology, 13(6): 1739-1759.
Dupont, F., P. Chittibabu, V. Fortin, Y.R. Rao et Y. Lu (2012). Assessment of a NEMO-based hydrodynamic modelling
system for the Great Lakes. Water Quality Research Journal of Canada, 47(3-4): 198-214.
H2D2: Suivi des niveaux, des
masses d'eau et des polluants
MESH:
débits en rivière
GEM:
conditions atmosphériques
NEMO: température, courants
niveau et couvert de glace des lacs
- 64 -
Page 65 – 22 January 2014
Hydrological prediction using MESH:
International Upper Great Lakes Study
Monthly water supply to Lake Superior Jun 04 – May 09
Impact of lakes on the weather:
Winter precipitation – DJF 05-09
Models improved
thanks to monitoring
of evaporation
mm
/day
MESH prediction (P-E+Runoff)
Observations Loss through evaporation (-E)
Net precipitation (P-E)
- 66 -
Prediction of monthly NBS
P
E
R
- 67 -
- 68 -
Current areas where CMC’s MESH
system is being implemented
- 69 -
Longer term developments:
the Arctic basin
from Gustafsson et Isberg (2013)
- 70 -
Arctic-HYPE
Model performance (NSE, monthly average)
WATCH Forcing Data (ERA 40 corrected with CRU+GPCC ) 1961-1988
NSE = tradeoff between variance,
correlation and bias
- 71 -
Arctic-HYPE
Arctic and Cold Regions challenges
Many lakes in some areas
Permafrost landscape
Palsa bogs
Inland sinks
Meandring and
human activity
Greenland still no routing
Ice plugs
Glaciers
Winter
evaporation
Generating Geophysical files for
SPS+hydrology
CCRN meeting
Montreal
January 21, 2014
Dorothy Durnford, Vincent Fortin
Page 73 – 22 January 2014
Process overview
1. Generation of regular fields
– SPS
2. Generation of extra fields
– hydrology
• Single script calculates 1) and/or 2)
Page 74 – 22 January 2014
1. Generation of SPS fields
• A. Preparation of domain’s grid
• B. Generation of SPS geophysical fields
Page 75 – 22 January 2014
1A. Preparation of domain’s grid
• From gem_settings.nml
• Requires grille
– specified version of GEM
• Few minutes
Page 76 – 22 January 2014
1A. gem_settings.nml:
Great Lakes/St Lawrence at 2 arcmin
&grid
Grd_typ_S = 'LU’,
Grd_ni = 795,
Grd_nj = 420,
Grd_iref = 1,
Grd_jref = 1,
Grd_latr = 39.0,
Grd_lonr = -94.5,
Grd_dx = 0.0333332,
Grd_dy = 0.0333332,
/
start of grid description
lat-lon grid
# points: x-direction
# points: y-direction
ref. point: x-direction index
ref. point: y-direction index
ref. point: latitude
ref. point: longitude
grid spacing (deg): x-direction
grid spacing (deg): y-direction
end of grid description
Page 77 – 22 January 2014
1B. Generation of SPS geophysical
fields
• 4 categories:
– topography
– masks
– vegetation
– soils
• Multiple databases
– resolutions from 1 km
– covering Canada to global
– Can combine datasets with specified priority
• Requires GenPhysX, ARMNLIB and CMC databases
• Several hours
Page 78 – 22 January 2014
1B. SPS geophysical fields:
Great Lakes/St Lawrence at 2 arcmin Topography: elevation Mask: ocean/continent
Veg: evergreen needle-leaf trees Soil: fraction of sand, level 1
Page 79 – 22 January 2014
2. Generation of hydrological fields
1. Average subgrid-scale topographical slope (SLOP)
2. Drainage density (DRND)
• Multiple databases
– resolutions from 25 m
– covering Canada and U.S.
– Can combine datasets with specified priority
• Requires GenPhysX, ARMNLIB and CMC databases
• Several hours/days
Page 80 – 22 January 2014
2.1. Generation of hydrological field:
SLOP
• Definitions:
The slope is the magnitude of the gradient
– gradient: perpendicular to the topography contours
– slope: maximum change in elevation
– slope: > 0
• Calculation:
For each database grid cell within the domain grid cell
– calculate the slope (degrees)
– calculate their average (degrees)
– convert from degrees to m/m
Page 81 – 22 January 2014
2.1. Hydrological geophysical field, SLOP:
Great Lakes/St Lawrence at 2 arcmin
Average subgrid-scale topographical slope (SLOP): m/m
Page 82 – 22 January 2014
2.1. Generation of hydrological field:
DRND
• Definition wrt a grid cell:
– m/m2
– from zero to infinity (zero)
total length of rivers/streams + total perimeters of bodies of water
grid cell area - total area of bodies of water
Page 83 – 22 January 2014
2.1. Hydrological geophysical field, DRND:
Great Lakes/St Lawrence at 2 arcmin
Drainage density (DRND): mm/m2
From HydroSHEDS data
to MESH shed fields
CCRN meeting
Montreal
January 21, 2014
Dorothy Durnford, Milena Dimitrijevic, Vincent Fortin
- 85 -
Process overview
1. Preparation of input fields
2. Determination of essential coordinates
3. Generation of MESH shed fields
- 86 -
1. Preparation of input fields
• Script
• Format conversion of HydroSHEDS data:
– .bil to .fst
– flow directions (DIR), elevations (DEM)
– Requires ARMNLIB and SPI
• Generation of ocean-continent mask (MG)
– same domain and grid
– Requires ARMNLIB and CMC databases
• Few hours
• Done for Canada (from 35N) at 30 arc seconds
- 87 -
Nechako Res., BC: input fields
DIR DEM
MG
- 88 -
2. Determination of essential coordinates
• Manual
• Of outlet and all reaches (any point)
• (x,y) values
• Domain relative
• Uses SPI
• Few minutes
- 89 -
Nechako Res., BC: outlet
MG DIR
SPI
- 90 -
3. Generation of MESH shed fields
• Fortran program
• Input:
– DIR, DEM, MG
– (x,y) values of outlet, each reach
– names, values of constant parameter fields
• Output:
– shed fields: rank, next, reach, drainage area …
– constant fields: flz, pwr, r1n, aa2, aa3, aa4 …
– extras: bver, tour
• Requires ARMNLIB environment
• Minutes: inland watershed
• Days: tidal regions (manual intervention)
- 91 -
Nechako Res., BC: shed fields
RANK
REAC
outlet
reach 1
reach 2
reach 3 reach 4
- 92 -
Nechako Res., BC: extras
outlet
BVER
TOUR
reach 3
reach 4
reach 2
reach 3
reach 4 reach 2
reach 1
- 93 -
Page 94 – 22 January 2014
Verification over June-August 2013
• Radar vs no radar, Canada-wide
Page 95 – 22 January 2014
Verification over June-August 2013
• Radar vs no radar, Prairies
Page 96 – 22 January 2014
Verification over June-August 2013
• Impact of radar QPE post-processing, Canada-wide
Page 97 – 22 January 2014
Verification over June-August 2013
• Impact of radar QPE post-processing, Prairies
• Known negative bias in the gauge measurement of solid precipitation
• Standard precipitation gauge used in the Canadian Reference Climate
Stations (RCS) only measures 50% of total snowfall at wind speeds of 5 m/s
• CaPA throws out observed precipitation from SYNO stations when 2m wind
speed > 3 m/s and Ta < 0°C
• No automated (RCS) data used when Ta < 0°C and Ws > 0
Solid Precipitation Issues Work done in collaboration with Craig Smith, CRD
Catch Efficiency of Geonor-SA vs. Wind Speed
0.00
0.20
0.40
0.60
0.80
1.00
1.20
0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0 10.0 11.0
Wind Speed (m/s)
Ca
tch
Eff
icie
nc
y
CE = e - 0.15Ws
r2 = 0.68
Geonor
Objective and Methods
• Objective: use bias adjusted RCS Geonor data (DJF 2010/2011) for selected
locations on the Canadian Prairies for CaPA verification
• Daily RCS Geonor precipitation adjusted using: CE=e-0.15Ws when maximum
daily Ta < -2°C (cold snow)
• 11 climate stations as a start
Alberta Saskatchewan
Manitoba
• AO: Assimilation of all observations
• OP: AO + strict quality control
(operational configuration)
• AN: OP + assimilation of bias-adjusted manual
Nipher obs
• ANG: AN + assimilation of bias-adjusted Geonor obs
CaPA Configurations
Catch Efficiency vs Wind Speed: Nipher
0
0.2
0.4
0.6
0.8
1
1.2
0 2 4 6 8 10
Wind Speed (2m) m/s
CE
(N
iph
er/
DF
IR)
(Goodison et al 1998)
Assimilation
of unbiased
observations only:
Slightly more
skill than forecast,
little bias
Equitable threat score Partial mean
All stations
assimilated:
Negative bias
Less skill than
forecast
GEM
CaPA
Obs (Nipher)
AO and OP vs Nipher obs.
Winter 2009-2010, Canada
AO
vs Nipher
OP
vs Nipher
Assimilation
of bias-corrected
Nipher obs:
Makes almost no
difference
Equitable threat score Partial mean
GEM
CaPA
Obs (RCS w/BC)
Assimilation
of unbiased
observations only:
Slightly more
skill than forecast,
large bias
OP and AN vs RCS obs.
Winter 2010-2011, Prairies
OP
vs RCS
AN
vs RCS
ANG vs Nipher or RCS obs.
Winter 2010-2011, Prairies Equitable threat score Partial mean
Assimilation of
bias-corrected
observations
(Nipher+Geonor):
Assessment of
bias and skill
largely depends
on what is chosen
as “truth”
GEM
CaPA
Obs (Nipher)
ANG
vs Nipher
ANG
vs RCS
ANG vs Nipher + RCS obs.
Winter 2010-2011, Prairies
GEM
CaPA
Obs (RCS w/BC + Nipher)
Verification
against both
bias-corrected
Nipher and Geonor
observations
ANG
vs Nipher
+ RCS
top related