Upload
others
View
11
Download
1
Embed Size (px)
Citation preview
Coupled Hydrological Atmospheric Modelling for IP3Atmospheric Modelling for IP3
Theme 3 working group
IP3 workshop Nov 9-10, WLU, Waterloo, Ontario
Environmental Prediction FrameworkEnvironmental Prediction Framework
Upper air GEM atmospheric4DVarUpper airobservations
C PA
GEM atmosphericmodel
4DVardata assimilation
“On-line”mode
“Off-line”mode
“On-line”mode
“Off-line”mode
CaPA:Canadian
precipitationanalysis
Surface scheme(EC version of Watflood CLASS or ISBA)
Surfaceobservations
CaLDAS:C di
y
ISBA)and routing model
Canadianland data
assimilation MESHModélisation environnementaleModélisation 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 B C C C ADesigned for a regular grid at a 1-15 km resolution
• Each grid divided into
B C C C A
C B B A A
D C B C C
Sub-gridHetereogeneity(land coverEach grid divided into
grouped response units (GRU or tiles) to deal with subgrid
D C B B C
D D D D B
(land cover,soil type, slope,aspect, altitude)
ghetereogeneity
AA relatively smallnumber of classes
B C
D
number of classesare kept, only the %of coverage foreach class is kept
MESH: A MEC surface/hydrology configuration designed for regional hydrological modelingdesigned for regional hydrological modeling
• The tile connector(1D, scalable) redistributes mass Tile
connectorand energy between tiles in a grid cell
– e.g. snow drift• The grid connector (2D) is
connector
• The grid connector (2D) is responsible for routing runoff
– can still be parallelized by grouping grid cells by
Gridconnector
subwatershed
From Measurements to ModelsFrom Measurements to ModelsResolution 1 m 100 m 100 m - 2 km 2 - 10 km 10 km - 10 km
Landscape type Pattern/tileTile/HRU Tile/HRU Grid/small basin Multi-grid/medium basin Multi-gridPoint Hillslope Sub-basin Basin Mesoscale Regional;
Prediction TerrestrialOpen WaterSnow and Ice
Previous LSS Scaling Methodology
Parametrization TerrestrialOpen WaterSnow and Ice
Process TerrestrialOpen WaterS d I
IP3 Scaling Methodology
Snow and Ice
MESH MESH MESH MESHMODELS CHRM CHRM CHRM CHRM
CEOP Hydrology CEOP HydrologyQuinton CFCAS Study---------------> <----------------------------------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 IP3 scaling methodology involvespoint and then applied directly to regional scales. IP3 scaling methodology involves step-wise transfer of upscaled processes to basin-scale parameterizations and then to regional scales
Scale mattersScale matters
RCM Domain
1Trail Valley Creek
(Arctic Tundra)
Havikpak Creek
GEM Domain North
GEM Domain South1
Mackenzie Basin
1
2Wolf Creek
(Subarctic Tundra Cordillrea)
Havikpak Creek(Taiga Woodland)
243
Peace-Athabasca Sub Basin (Mackenzie)
3 Scotty Creek(Permafrost Wetlands)
Saskatchewan BasinColumbia River
Basin
4
5
Baker Creek(Subarctic Shield
Lakes)
Lake O'Hara(Wet Alpine)
Peyto Creek(Glaciated Alpine)
5
5 (Wet Alpine)
Marmot Creek(Subalpine Forest)
AdvancementsAdvancements
• Establishing MESH domains at the basin scaleg– Partnerships for most research basins have formed.– Single Grid version of CLASS for each basin has been set-up
by U of Wby U of W.▪ Soulis and Seglenieks
– Software Engineering and repository established at HAL labDavison▪ Davison
– DDS working with CLASS and MESH▪ Tolson
Calibration of a Land Surface Hydrology y gyScheme 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, Canada2 Environment Canada, Saskatoon, SK, Canada3 University of Waterloo ON Canada3 University of Waterloo, ON, Canada
Objectivesj
To parameterise a LS-Hydrological model A stepwise procedure isTo parameterise a LS Hydrological model,A stepwise procedure is applied:
1. Calibration of a LSS in a point mode using a single-objective function (snow water equivalent-SWE). Examination of the effects ( q )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 SCA) by keeping thefunction (streamflow and snow cover area-SCA) by keeping the vegetation parameters calibrated in point 1.
Wolf Creek – Trail Valley Creek
T il V ll C kTrail Valley Creek
G B iGranger Basin60° 31’N, 135° 07’W Area: 8 km2 TVC Basin
68° 45’N, 133° 30’W Area: 63 km2Area: 63 km2
Landscape Heterogeneity
Modelling strategyLSS
Parameter PLT NF
Open tundra
Shrub tundra
The Canadian Land Surface Scheme (CLASS 3 3)
tundra tundra
Max. LAI (LAMX) 0.53(0.5, 2)
2.81(2, 3)
Min. LAI (LAMN) 0.28(0.5, 3)
0.99(0.4, 1)
LN roughness length (LNZ0) 4 09 2 42(CLASS 3.3)Calibration 2003
Objective function: Snow Water Equivalent (SWE)
LN roughness length (LNZ0)[m]
-4.09(-4.8, -
3.5)
-2.42(-3.7, -
1.8)
Visible albedo (ALVC) 0.183(0.02, 0.2)
0.087(0.03, 0.2)
Near infrared albedo (ALIC) 0 424 0 464Dynamically Dimensioned Search (DDS) global optimisation algorithm (Tolson and Shoemaker WRR 2007) 25 parameters (12 for shrubs, 12 for grass,
Near-infrared albedo (ALIC) 0.424(0.2, 0.4)
0.464(0.3, 0.5)
Biomass Den. (CMAS)[Kg·m-2]
0.11(0.05, 0.35)
6.13(6, 10)
Min. stomatal resist. (RSMN) 251.5(50 300)
51.9(50 300)and 1 for snow-cover depletion, SCD) that
govern snowmelt
Validation 2002 and 2004
(50, 300) (50, 300)
Coef. stomata resp. to light (QA50) [W·m-2]
46.1(20, 60)
21.1(20, 60)
Coef. stomatal resist. to VP deficit (VPDA)
1.31(0.2, 1.5)
1.08(0.2, 1.5)
Effects of initial conditions were analysedfrom extensive field observations whereas forcing data effects were evaluated using the Cold Region Hydrological Model
Coef. stomatal resist. to VP deficit (VPDB)
0.61(0.2, 1.5)
0.93(0.2, 1.5)
Coef. stomatal resist. to soil WS (PSGA)
146.7(50, 150)
93.5(50, 150)
Coef. stomatal resist. to soil WS (PSGB)
4.92(1 10)
1.09(1 10)
g y g(CRHM) as a prepossessing data for CLASS.
(PSGB) (1-10) (1-10)
Lower snow depth limit for 100% SCA (D100) [m]
0.42(0.05-0.5)
0.81(0.05-1)
Modelling strategyCRHM – Short wave correction
Modelling strategyCLASS – Point mode
SWE - Calibration period - 2003m
m)
120
160
200
obs SWE UBSim SWE UB
mm
) 120
160obs SWE PLTsim SWE PLT
SWE
(
0
40
80
SWE
(0
40
80
Time (days)
Apr 20 Apr 28 May 06 May 14 May 22 May 30 Jun 07 0
Time (days)
Apr 20 Apr 28 May 06 May 14 May 22 0
NS= 0.93 NS= 0.76
SWE - Calibration period - 2003m
m)
160
200
240obs SWE NFsim SWE NF
mm
)
200240280320
obs SWE SFsim SWE SF
SWE
(m
0
40
80
120
SWE
(m0
4080
120160
Time (days)
Apr 20 Apr 28 May 06 May 14 May 22 May 30 Jun 07 0
Time (days)
Apr 20 Apr 28 May 06 May 14 May 22 May 30 Jun 07 0
NS= 0.87 NS= 0.70
SWE - Calibration period - 2003
mm
)
120
160
200obs SWE VBsim SWE VB
SWE
(m
40
80
120
Time (days)
Apr 20 Apr 28 May 06 May 14 May 22 May 30 0
NS= 0.98
SWE - Validation period - 2002W
E (m
m)
120160200240280320
obs SWE NFsim SWE NF
WE
(mm
)
6080
100120140
obs SWE SFsim SWE SF
NF=0.92 NS= 0.89
Time (days)
Apr 19 Apr 28 May 07 May 16 May 25 Jun 03 Jun 12
SW
04080
120
Time (days)
Apr 19 Apr 27 May 05 May 13 May 21
SW
0204060
Time (days) Time (days)
m) 120
160obs SWE VBsim SWE VBNS= 0.78
SWE
(m
0
40
80
Time (days)
Apr 19 Apr 27 May 05 May 13 May 21 May 29 0
SWE - Validation period - 2004
280 240b SWE SF
SWE
(mm
)
80120160200240 obs SWE NF
sim SWE NF
SWE
(mm
)
80
120
160
200 obs SWE SFsim SWE SF
NS= 0.72NS= 0.73
Time (days)
Apr 20 Apr 27 May 04 May 11 May 18 May 25
S
04080
Time (days)
Apr 20 Apr 27 May 04 May 11 May 18 May 25
S
0
40
mm
)
120
160
200obs SWE VBsim SWE VB
mm
)
60
80
100obs SWE PLTsim SWE PLT
NS= 0 94
A 20 A 27 M 04 M 11 M 18 M 25
SWE
(
0
40
80
Apr 16 Apr 23 Apr 30 Ma 07 Ma 14
SWE
(
0
20
40
NS= 0.94NS= 0.86
Time (days)
Apr 20 Apr 27 May 04 May 11 May 18 May 25
Time (days)
Apr 16 Apr 23 Apr 30 May 07 May 14
Avg Initial ConditionsAvg. Initial Conditions
20022002
WE
(mm
)
120160200240280320
obs SWE NFsim SWE NF
WE
(mm
)
80
120
160
200obs SWE SFsim SWE SF
NS= -0.36 R2= 0 75
20022002
Ti (d )
Apr 19 Apr 28 May 07 May 16 May 25 Jun 03 Jun 12
SW
04080
120
Ti (d )
Apr 19 Apr 27 May 05 May 13 May 21
SW
0
40
80 R 0.75
m)
160
200
240obs SWE PLTsim SWE PLT
Time (days) Time (days)
m)
160
200
240obs SWE PLTsim SWE PLT
2003 2004
SWE
(mm
0
40
80
120
160
NS= -4.87SW
E (m
m
0
40
80
120
160
NS= -10.69
Time (days)Apr 20 Apr 28 May 06 May 14 May 22
0
Time (days)
Apr 16 Apr 23 Apr 30 May 07 May 14 0
Aggregated Forcing data - 2003gg g g
240b SWE NF 280
320obs SWE SF
SWE
(mm
)
80
120
160
200 obs SWE NFsim SWE NF
SWE
(mm
)
80120160200240280 obs SWE SF
sim SWE SFNS= -0.44 NS= -0.54
Time (days)
Apr 20 Apr 28 May 06 May 14 May 22 May 30 Jun 07 0
40
Time (days)
Apr 20 Apr 28 May 06 May 14 May 22 May 30 Jun 07 0
4080
mm
)
120
160
200
obs SWE UBsim SWE UB
NS= -0.38
Apr 20 Apr 28 May 06 May 14 May 22 May 30 Jun 07
SWE
(
0
40
80
Time (days)
Apr 20 Apr 28 May 06 May 14 May 22 May 30 Jun 07
Aggregated Forcing data – 2002-2004gg g g20022002
(mm
)
160200240280320
obs SWE NFsim SWE NF
(mm
)
80100120140
obs SWE SFsim SWE SF
NS= 0.44 NS= 0.87
Apr 19 Apr 28 May 07 May 16 May 25 Jun 03 Jun 12
SWE
04080
120160
Apr 19 Apr 27 May 05 May 13 May 21
SWE
0204060
20042004Time (days) Time (days)
m) 200
240280
obs SWE NFsim SWE NF
m)
160
200
240
obs SWE SFsim SWE SFNS= 0.54
20042004
SWE
(mm
04080
120160200
SWE
(mm
0
40
80
120
160
NS= -3.44
Time (days)
Apr 20 Apr 27 May 04 May 11 May 18 May 25 0
Time (days)
Apr 20 Apr 27 May 04 May 11 May 18 May 25 0
Aggregated vs Distributed
2002802003 2002
80
120
160Obs SWE AggregatedDistributed
WE
(mm
)
120160200240 Obs SWE
AggregatedDistributed
Time (days)Apr 19 Apr 28 May 07 May 16 May 25 Jun 03 Jun 12
0
40
Apr 20 Apr 28 May 06 May 14 May 22 May 30 Jun 07
SW
04080
Time (days)
(mm
)
160
200
240Obs SWE AggregatedDistributed
2004
SW
E (
0
40
80
120
Time (days)Apr 20 Apr 27 May 04 May 11 May 18 May 25
0
Modelling strategyTransference of parameters
Modelling strategyg gyLS-Hydrological model
The MESH modelling system g yCalibration 1996Objective functions: St fl d b i S C A (SCA)Streamflow and basin average Snow Cover Area (SCA)
Dynamically Dimensioned Search (DDS) global optimisation algorithm
15 parameters (7 for shrubs 7 for grass and 1 for snow-15 parameters (7 for shrubs, 7 for grass, and 1 for snowcover depletion, SCD)
Validation 1999Validation 1999
Trail Valley Creek - SCAA
0.8
1.0Obs SCASim regional.Sim default
A
0.8
1.0Obs SCASim regional.Sim default
Frac
tiona
l SC
A
0.4
0.6
Frac
tiona
l SC
A
0.4
0.6
May-11 May-18 May-25 Jun-01 Jun-08 Jun-15 0.0
0.2
May 11 May 18 May 25 Jun 01 Jun 08 Jun 15 0.0
0.2
calibration validation
Date (1996)
y y y
Date (1999)
y y y
(a) (b)
Trail Valley Creek - Streamflow
8 10
1 ]
6
ObsSim regional.Sim default
1 ]
6
8ObsSim regional.Sim default
Q [m
3 . s-1
2
4
Q [m
3 . s-1
4
6
May-01 May-11 May-21 May-31 Jun-10 Jun-20 Jun-300
2
May 01 May 11 May 21 May 31 Jun 10 Jun 20 Jun 300
2
calibration validation
Date (1996)
May 01 May 11 May 21 May 31 Jun 10 Jun 20 Jun 30
Date (1999)
May 01 May 11 May 21 May 31 Jun 10 Jun 20 Jun 30
ConclusionsConclusions
• A regionalization approach for transferring parameters of a physically based LSH model in sub arctic and arctic environments has been presentedLSH model in sub-arctic and arctic environments has been 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 wise calibration procedure was used to estimate the effective– second a step-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.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 NextWhat Next
• Extend Wolf Creek analysis to entire basiny
• Examine basin segmentation approaches and combinations of topographic and land cover GRU’scombinations of topographic and land-cover GRU s.
• Look at continuous simulation to assess impacts on IC.– Tile connectors for redistribution of blowing snow
• Extend analysis to other basins• Extend analysis to other basins
• Pay attention to IP-1 and IP-2 findings