Coupled Hydrological
Atmospheric Modelling for IP3
Theme 3 working group
IP3 workshop
Nov 9-10, WLU, Waterloo, Ontario
Environmental Prediction Framework
Upper air
observations
GEM atmospheric
model
4DVar
data assimilation
CaPA:
C
PA
Canadian
precipitation
analysis
y
Surface
observations
CaLDAS:
C
Canadian
di
land data
assimilation
“ On-line”
mode
“ Off-line”
mode
Surface scheme
(EC version of
Watflood CLASS or
ISBA)
and routing model
MESH
Modélisation environnementale
communautaire (MEC)
de la surface et de l’hydrologie
MESH: A MEC surface/hydrology
configuration designed for regional
hydrological modeling
• Designed for a regular
•
grid at a 1-15 km
resolution
Each grid divided into
grouped response
units (GRU or tiles) to
deal with subgrid
g
hetereogeneity
Sub-grid
Hetereogeneity
(land cover
cover,
soil type, slope,
aspect, altitude)
A relatively small
number of classes
are kept, only the %
of coverage for
each class is kept
B
C
C
C
A
C
B
B
A
A
D
C
B
C
C
D
C
B
B
C
D
D
D
D
B
A
B
C
D
MESH: A MEC surface/hydrology configuration
designed for regional hydrological modeling
• The tile connector
(1D, scalable) redistributes mass
and energy between tiles in a grid
cell
Tile
connector
– e.g. snow drift
• The grid connector (2D) is
responsible for routing runoff
– can still be parallelized by
grouping grid cells by
subwatershed
Grid
connector
From Measurements to Models
Resolution
1m
Tile/HRU
Point
Prediction
Terrestrial
Open Water
Snow and Ice
Parametrization
Terrestrial
Open Water
Snow and Ice
Process
Terrestrial
Open Water
S
Snow
and
d IIce
MODELS
100 m
Landscape type
Tile/HRU
Hillslope
100 m - 2 km
Pattern/tile
Grid/small basin
Sub-basin
2 - 10 km
10 km -
Multi-grid/medium basin Multi-grid
Basin
Mesoscale
10 km
Regional;
Previous LSS Scaling Methodology
IP3 Scaling Methodology
MESH
CHRM
CEOP Hydrology
Quinton CFCAS Study--------------->
CHRM
CHRM
MESH
MESH
MESH
CHRM
CEOP Hydrology
<----------------------------------MAGS
Modelling and parameterization hierarchy. Previous LSS scaling methodology
refers to projects that parameterized and evaluated predictions of processes at a
point and then applied directly to regional scales.
scales IP3 scaling methodology involves
step-wise transfer of upscaled processes to basin-scale parameterizations and then
to regional scales
Scale matters
RCM Domain
GEM Domain North
1
GEM Domain South
1
Trail Valley Creek
(Arctic Tundra)
Havikpak Creek
(Taiga Woodland)
Mackenzie Basin
2
3
4
2
Wolf Creek
(Subarctic Tundra
Cordillrea)
3
Scotty Creek
(Permafrost Wetlands)
4
Baker Creek
(Subarctic Shield
Lakes)
Peace-Athabasca Sub Basin
(Mackenzie)
5
Saskatchewan Basin
Peyto Creek
(Glaciated Alpine)
Columbia River
Basin
5
Lake O'Hara
(Wet Alpine)
Marmot Creek
(Subalpine Forest)
Advancements
• Establishingg MESH domains at the basin scale
– Partnerships for most research basins have formed.
– Single Grid version of CLASS for each basin has been set-up
by U of W
W.
▪ Soulis and Seglenieks
– Software Engineering and repository established at HAL lab
▪ Davison
– DDS working with CLASS and MESH
▪ Tolson
Calibration of a Land Surface Hydrology y
gy
Scheme in Arctic Environments
Pablo Dornes1, Bruce Davison2, Alain Pietroniro2
Bryan Tolson3, Ric Soulis3, Philip Marsh2, and John Pomeroy1
1 Centre for Hydrology, University of Saskatchewan, Saskatoon, SK, Canada
2 Environment Canada, Saskatoon, SK, Canada
3 University of Waterloo
Waterloo, ON
ON, Canada
Objectives
j
To parameterise a LS-Hydrological
LS Hydrological model
model,A
A stepwise procedure is
applied:
1. Calibration of a LSS in a point mode using a single-objective
function ((snow water equivalent-SWE).
q
) Examination of the effects
of including an explicit representation in a LSS of:
a) Fully distributed (calibrated)
b) Initial conditions,
c) Forcing data.
d) all
2. Calibration of a LS-Hydrological model using a multi-objective
function (streamflow and snow cover area
area-SCA)
SCA) by keeping the
vegetation parameters calibrated in point 1.
Wolf Creek – Trail Valley Creek
T il V
Trail
Valley
ll C
Creek
k
Granger Basin
G
B i
60° 31’N, 135° 07’W
Area: 8 km2
TVC Basin
68° 45’N, 133° 30’W
Area: 63 km2
Landscape Heterogeneity
Modelling strategy
Parameter
LSS
Max. LAI (LAMX)
The Canadian Land Surface Scheme
(CLASS 3
3.3)
3)
ƒ Calibration 2003
Objective function: Snow Water Equivalent
(SWE)
Dynamically Dimensioned Search (DDS)
global optimisation algorithm (Tolson and
Shoemaker WRR 2007)
25 parameters (12 for shrubs, 12 for grass,
and 1 for snow-cover depletion, SCD) that
govern snowmelt
Min. LAI (LAMN)
LN roughness length (LNZ0)
[m]
Visible albedo (ALVC)
Near infrared albedo (ALIC)
Near-infrared
Biomass Den. (CMAS)
[Kg·m-2]
Min. stomatal resist. (RSMN)
PLT
NF
Open
tundra
Shrub
tundra
0.53
2.81
(0.5, 2)
0.28
(0.5, 3)
4 09
-4.09
-2.42
2 42
(-3.7, 1.8)
0.183
0.087
(0.02,
0.2)
0 424
0.424
(0.2, 0.4)
0.11
(0.05,
0.35)
251.5
(50 300)
(50,
(20, 60)
ƒ Validation 2002 and 2004
Coef. stomatal resist. to VP deficit
(VPDA)
(0.2, 1.5)
ƒ Effects of initial conditions were analysed
Coef. stomatal resist. to VP deficit
(VPDB)
(0.2, 1.5)
Coef. stomatal resist. to soil WS
(PSGA)
(50, 150)
Coef. stomatal resist. to soil WS
(PSGB)
(1 10)
(1-10)
Lower snow depth limit for 100% SCA
(D100) [m]
0.99
(0.4, 1)
(-4.8, 3.5)
Coef. stomata resp. to light (QA50)
[W·m-2]
from extensive field observations whereas
forcing data effects were evaluated using
the Cold Region
g
Hydrological
y
g
Model
(CRHM) as a prepossessing data for
CLASS.
(2, 3)
46.1
1.31
0.61
146.7
4.92
0.42
(0.05-0.5)
(0.03,
0.2)
0 464
0.464
(0.3, 0.5)
6.13
(6, 10)
51.9
(50 300)
(50,
21.1
(20, 60)
1.08
(0.2, 1.5)
0.93
(0.2, 1.5)
93.5
(50, 150)
1.09
(1 10)
(1-10)
0.81
(0.05-1)
Modelling strategy
CRHM – Short wave correction
Modelling strategy
CLASS – Point mode
SWE - Calibration period - 2003
160
obs SWE UB
Sim SWE UB
160
120
80
40
0
Apr 20
Apr 28 May 06 May 14 May 22 May 30 Jun 07
Time (days)
NS= 0.93
SWE (mm)
SWE (mm)
200
obs SWE PLT
sim SWE PLT
120
80
40
0
Apr 20
Apr 28
May 06
Time (days)
NS= 0.76
May 14
May 22
SWE - Calibration period - 2003
obs SWE NF
sim SWE NF
SWE (m
mm)
200
160
120
80
40
0
Apr 20
Apr 28 May 06 May 14 May 22 May 30 Jun 07
Time (days)
NS= 0.87
SWE (m
mm)
240
320
280
240
200
160
120
80
40
0
Apr 20
obs SWE SF
sim SWE SF
Apr 28 May 06 May 14 May 22 May 30 Jun 07
Time (days)
NS= 0.70
SWE - Calibration period - 2003
SWE (m
mm)
200
obs SWE VB
sim SWE VB
160
120
80
40
0
Apr 20
Apr 28
May 06
May 14
Time (days)
NS= 0.98
May 22
May 30
320
280
240
200
160
120
80
40
0
Apr 19
NF=0.92
obs SWE NF
sim SWE NF
Apr 28 May 07 May 16 May 25 Jun 03 Jun 12
obs SWE SF
sim SWE SF
100
80
60
40
obs SWE VB
sim SWE VB
120
NS= 0.78
80
40
Apr 27
May 05
May 13
Time (days)
0
Apr 19
Apr 27
May 05
Time (days)
160
SWE (mm)
120
20
Time (days)
0
Apr 19
NS= 0.89
140
SW
WE (mm)
SW
WE (mm)
SWE - Validation period - 2002
May 21
May 29
May 13
May 21
SWE - Validation period - 2004
280
240
obs SWE NF
sim SWE NF
SWE (mm)
S
NS=200
0.72
160
120
80
sim SWE
NS=
0.73SF
160
120
80
40
40
0
Apr 20
obs
b SWE SF
200
SWE (mm)
S
240
Apr 27
May 04
May 11
May 18
0
Apr 20
May 25
Apr 27
Time (days)
May 18
May 25
100
obs SWE VB
sim SWE VB
160
120
NS= 0.94
0 94
80
SWE (mm)
SWE (mm)
May 11
Time (days)
200
obs SWE PLT
sim SWE PLT
80
60
40NS=
0.86
20
40
0
Apr 20
A
May 04
A 27
Apr
M 04
May
M 11
May
Time (days)
M 18
May
M 25
May
0
Apr 16
Apr 23
Apr 30
Time (days)
Ma 07
May
Ma 14
May
320
280
240
200
160
120
80
40
0
Apr 19
2002
2002
200
obs SWE NF
sim SWE NF
NS= -0.36
SW
WE (mm)
SW
WE (mm)
Avg Initial Conditions
Avg.
120
R2= 0.75
0 75
80
40
0
Apr 19
Apr 28 May 07 May 16 May 25 Jun 03 Jun 12
obs SWE SF
sim SWE SF
160
Apr 27
Ti
Time
(days)
(d
)
NS= -4.87
80
40
0
Apr 20
obs SWE PLT
sim SWE PLT
200
SWE (mm
m)
SWE (mm
m)
160
120
May 21
2004
240
obs SWE PLT
sim SWE PLT
200
May 13
Ti
Time
(days)
(d
)
2003
240
May 05
160
NS= -10.69
120
80
40
Apr 28
May 06
Time (days)
May 14
May 22
0
Apr 16
Apr 23
Apr 30
Time (days)
May 07
May 14
Aggregated
gg g
Forcing
g data - 2003
SWE (mm)
200
160
NS=
obs
b SWE NF
sim SWE NF
-0.44
120
80
40
0
Apr 20
Apr 28 May 06 May 14 May 22 May 30 Jun 07
Time (days)
SWE (mm)
120
obs SWE UB
sim SWE UB
NS= -0.38
80
40
0
Apr 20
Apr 28 May 06 May 14 May 22 May 30 Jun 07
Time (days)
320
280
240
200
160
120
80
40
0
Apr 20
obs SWE SF
sim SWE
SF
NS=
-0.54
Apr 28 May 06 May 14 May 22 May 30 Jun 07
Time (days)
200
160
SWE (mm)
240
Aggregated
gg g
Forcing
g data – 2002-2004
320
280
240
200
160
120
80
40
0
Apr 19
2002
NS= 0.44
obs SWE NF
sim SWE NF
120
80
60
40
0
Apr 19
Apr 28 May 07 May 16 May 25 Jun 03 Jun 12
2004
Apr 27
Time (days)
May 05
2004
May 13
May 21
Time (days)
240
obs SWE NF
sim SWE NF
200
160
120
NS= -3.44
80
obs SWE SF
sim SWE SF
200
SWE (mm
m)
240
SWE (mm
m)
obs SWE SF
sim SWE SF
100
20
280
NS= 0.54
160
120
80
40
40
0
Apr 20
NS= 0.87
140
SWE (mm)
SWE (mm)
2002
Apr 27
May 04
May 11
Time (days)
May 18
May 25
0
Apr 20
Apr 27
May 04
May 11
Time (days)
May 18
May 25
Aggregated vs Distributed
2003
2002
280
200
SW
WE (mm)
240
Obs SWE
Aggregated
Distributed
200
160
120
120
80
80
40
40
0
Apr 20
0
Apr 28 May 06 May 14 May 22 May 30 Jun 07 Apr 19
240
2004
SWE ((mm)
200
160
Obs SWE
Aggregated
Distributed
120
80
40
0
Apr 20
Obs SWE
Aggregated
Distributed
160
Apr 27
May 04
May 11
Time (days)
May 18
May 25
Apr 28 May 07 May 16 May 25 Jun 03
Time (days)
Jun 12
Modelling strategy
Transference of parameters
Modelling
g strategy
gy
LS-Hydrological model
The MESH modelling
g system
y
Calibration 1996
Objective functions:
St
Streamflow
fl
and
d basin
b i average S
Snow C
Cover A
Area (SCA)
Dynamically Dimensioned Search (DDS) global
optimisation algorithm
15 parameters (7 for shrubs
shrubs, 7 for grass
grass, and 1 for snowsnow
cover depletion, SCD)
ƒ Validation 1999
Trail Valley Creek - SCA
1.0
1.0
Obs SCA
Sim regional.
Sim default
0.8
Fractional SCA
A
Fractional SCA
A
0.8
0.6
0.4
0.6
0.4
0.2
0.2
0.0
0.0
May-11
y
May-18
y
May-25
y
Jun-01
Date (1996)
(a)
calibration
Jun-08
Jun-15
Obs SCA
Sim regional.
Sim default
Mayy 11
May
y 18
May
y 25
Jun 01
Date (1999)
(b)
validation
Jun 08
Jun 15
Trail Valley Creek - Streamflow
8
10
Obs
Sim regional.
Sim default
8
-1
1
Q [m . s ]
4
6
3
3
-1
1
Q [m . s ]
6
Obs
Sim regional.
Sim default
2
4
2
0
0
May-01
May
01 May
May-11
11 May
May-21
21 May
May-31
31 Jun
Jun-10
10 Jun
Jun-20
20 Jun
Jun-30
30
Date (1996)
calibration
May 01 May 11 May 21 May 31 Jun 10 Jun 20 Jun 30
Date (1999)
validation
Conclusions
• A regionalization approach for transferring parameters of a physically based
•
LSH model in sub
sub-arctic
arctic and arctic environments has been presented
presented.
This approach was based on a landscape similarity criterion and focused on
two aspects.
– First, model parameters are landcover-based rather than basin-based, and
– second a step
step-wise
wise calibration procedure was used to estimate the effective
parameters.
• The landcover-based parameters offer an interesting alternative for PUB
•
•
due to the difficulties in finding basin-based criteria for transferring
parameters.
Distributed and physically based models, landscape-based parameters
appear to be a more feasible framework for transferring information
between catchments than regionalisation schemes using regression
methods based on basin characteristics.
A special case however, was the inclusion of the SDC parameter in the
calibration process at TVC. The main reasons were its poor physical basis
and the resulting difficulty in deriving a landscape-base value from
observations.
What Next
• Extend Wolf Creek analysis
y
to entire basin
• Examine basin segmentation approaches and
combinations of topographic and land
land-cover
cover GRU’s
GRU s.
• Look at continuous simulation to assess impacts on IC.
– Tile connectors for redistribution of blowing snow
• Extend analysis to other basins
• Pay attention to IP-1 and IP-2 findings