sensitivity analysis of the land surface model noah-mp for ... · sensitivity analysis of the land...
TRANSCRIPT
UFZ -CENTRE FOR ENVIRONMENTAL RESEARCH LEIPZIG-HALLE
Sensitivity Analysis of the Land Surface Model Noah-MP for Different Output Fluxes in 12 Distinct CatchmentsJ. Mai 1, S. Thober 1, L. Samaniego 1, O. Branch 2, V. Wulfmeyer 2, M. Clark 3, S. Attinger 1, R. Kumar 1, and M. Cuntz 1
1 Helmholtz Centre for Environmental Research - UFZ, Leipzig, Germany, ([email protected]) 2 University of Hohenheim Stuttgart, Germany, 3 National Center for Atmospheric Research NCAR Boulder, CO
1. Motivation
Land Surface Models (LSMs) use a plenitude of process descriptions to rep-resent the carbon, energy and water cycles. They are highly complex andcomputationally expensive. Practitioners, however, are often only interestedin specific parts of the model such as latent heat or surface runoff. In modelapplications like parameter estimation, the most important parameters arethen chosen by experience or expert knowledge. Hydrologists interested insurface runoff therefore choose mostly soil parameters while biogeochemistsfocus on carbon fluxes and hence on vegetation parameters. However, thismight neglect parameters that are important, for example, through stronginteractions with the parameters chosen. Additionally, supposedly unimpor-tant parameters are often fixed (i.e., hard-coded) during model development.However, these might be highly relevant during model calibration but remainnormally undetected.
Noah-MP is applied in 12MOPEX catchments. Metero-logical forcings are obtainedfrom NLDAS-2 and static datafrom LDAS. The spatial reso-lution is 0.125◦ and the simu-lation period is from 1979 to1999 with the first years asspin-up. Additional catchmentcharacteristics are given in thefollowing table.
100 ◦ W 95 ◦ W 90 ◦ W 85 ◦ W 80 ◦ W 75 ◦ W
30◦N
35◦N
40◦N
1
2
3
4
5
6
78
9
10
1112
Mean Mean Runoff Mean Mean Winter DominantNo Name Rainfall Runoff Coefficient Snow Temperature Land Cover
[mm] [mm] [-] [mm] [◦C] [-]1 Tygart Valley River 1198 735 0.61 1901 -0.3 Dec BdLf For2 French Broad 1413 800 0.56 902 3.6 Mixed Forest3 Monocacy River 1050 420 0.40 1737 1.3 Dec BdLf For4 Bluestone River 1036 416 0.40 798 2.0 Dec BdLf For5 Amite River 1563 609 0.38 25 11.2 Evg NdLf For6 East Fork White River 1013 377 0.37 679 -0.5 Grassland7 Rappahannock River 1037 377 0.36 733 3.0 Mixed Forest8 South Branch Pot. 1055 341 0.32 2582 0.3 Dec BdLf For9 English River 900 269 0.29 1776 -2.9 Grassland
10 Spring River 1075 298 0.27 377 3.0 Dec BdLf For11 San Marcos River 825 179 0.21 4 11.2 Grassland12 Guadalupe 763 116 0.15 7 10.1 Veg Mosaic
2. Study Area
The code was scanned for hard-coded values that are termed hidden parameters. 146 such numbers were found in all processes ofNoah-MP but independent of vegetation or soil type. These parameters are in addition to the 93 standard parameters of Noah-MP.38 standard and 80 hidden parameters remained in the analysis when the specific options of Noah-MP were chosen. Informativeparameters were screened for different model outputs[1]: surface runoff, underground runoff, latent heat, and transpiration. Sobol’sensitivity analyses were performed on the parameters identified. Per basin and output flux on average 1,300 model runs for thescreening and 75,000 model runs for the Sobol’ analysis were required. p1
p2
p3
3. Methods
1 2 3 4 5 6 7 8 9101112
RH
OL
01
RH
OS 0
2TA
UL
03
TA
US 0
4X
L 0
5
1 2 3 4 5 6 7 8 9101112
REFS
MC
06
WLT
SM
C 0
7N
RO
OT 0
8C
H2
OP 0
9D
LEA
F 1
0Z
0M
VT 1
1H
VT 1
2H
VB
13
RC
14
CW
PV
T 1
5SA
IM 1
6LA
IM 1
7A
KC
18
AK
O 1
9A
VC
MX
20
VC
MX
25
21
BP 2
2M
P 2
3Q
E2
5 2
4FO
LNM
X 2
5B
B 2
6M
AX
SM
C 2
7SA
TPSI 2
8SA
TD
K 2
9SA
TD
W 3
0FR
ZK
_DA
TA
31
SLO
PE_D
ATA
32
REFD
K_D
ATA
33
REFK
DT_D
ATA
34
ZB
OT_D
ATA
35
FQPR
EC
C 3
6M
QPR
EC
C 3
7FD
IRR
AD
38
FVIS
RA
D 3
9C
SN
OW
PR
EC
IP1
40
CSN
OW
PR
EC
IP2
41
CSN
OW
PR
EC
IP3
42
CTSN
OW
PR
EC
IP1
43
CTSN
OW
PR
EC
IP2
44
CTSN
OW
PR
EC
IP3
45
C2
STR
EA
MK
OPEN
46
C2
STR
EA
MC
HIL
1 4
7C
2STR
EA
MC
HIL
2 4
8C
2STR
EA
MC
HIL
3 4
9C
2STR
EA
MC
HIL
4 5
0Z
0SN
O 5
1K
RSU
RF1
52
CR
SU
RF2
53
CM
AX
LAIE
54
CSTA
BLE
ATM
2 5
5C
STA
BLE
ATM
3 5
6C
RB
OU
ND
LEA
F 5
7C
TIN
HIB
1 5
8C
TIN
HIB
2 5
9M
EX
P 6
0SN
OW
_IW
C 6
1C
BU
LKSN
OW
1 6
2C
BU
LKSN
OW
2 6
3C
BU
LKSN
OW
3 6
4C
BU
LKSN
OW
4 6
5C
WETC
AN
OPY 6
6C
SN
OW
CO
MPA
CTC
2 6
7C
SN
OW
CO
MPA
CTD
M 6
8C
SN
OW
CO
MPA
CTETA
0 6
9C
SN
OW
CO
MPA
CTC
6 7
0C
MIN
SN
OW
L1 7
1C
MIN
SN
OW
L2 7
2D
CM
AX
SN
OW
L1 7
3C
MA
XSN
OW
L2 7
4C
FIM
PER
MFR
OZ
EN
2 7
5SW
EM
X 7
6C
TK
SN
O1
77
CTK
SN
O2
78
CSN
OW
CLA
SSA
LB0
79
KSN
OW
CLA
SS 8
0C
SN
OW
CLA
SSFR
ESH
81
CR
AD
ALB
SA
T 8
2C
FSN
O 8
3C
ZPD
84
CG
RO
UN
DM
AX
INC
85
KG
RO
UN
DM
OIS
T 8
6
Radia
tion
Soil
Physi
o
Vegeta
tion
Str
uct
ure
Physi
olo
gy
Soil
Wate
r
Runoff
Input
Radia
tion
Tra
nsf
er
Physi
olo
gy
Snow
Wate
r
Snow
Energ
y
Soil
Energ
y
Late
nt
Heat
Catc
hm
ents
Surf
ace
Runoff
Catc
hm
ents
Standard Parameters Hidden Parameters
• Informative parameters depend more on model output thanon catchment characteristics
• On average 29% of in total 118 parameters are informative
• On average 38% of informative parameters are hidden
• Snow parameters are highlighted in surface runoff
• Parameters for vegetation, i.e. plant-soil interface, structureand physiology, are highlighted in latent heat
4. Parameter ScreeningRadiation
Physio.
Soil
StructureVegetation
Physiology
Soil
Water
Runoff
Input*
Radiation*
Transfer*
Physio.*
Water*
Snow
Energ
y*Snow
Energy*
Soil
0
0.8
2 6
101418222630343
8424
650
545862667074
7882
86
8S
outhB
ranchP
otomac
RadiationPhysio.
Soil
StructureVegetation
Physiology
Soil
Water
Runoff
Input*
Radiation*
Transfer*
Physio.*
Water*
Snow
Energ
y*Snow
Energy*
Soil
0
0.8
2 6
101418222630343
8424
650
545862667074
7882
86
12G
uadalupe
9E
nglishR
iver
RadiationPhysio.
Soil
StructureVegetation
Physiology
Soil
Water
Runoff
Input*
Radiation*
Transfer*
Physio.*
Water*
Snow
Energ
y*Snow
Energy*
Soil
0
0.8
2 6
101418222630343
8424
650
545862667074
7882
86
Underground Runoff
RadiationPhysio.
Soil
StructureVegetation
Physiology
Soil
Water
Runoff
Input*
Radiation*
Transfer*
Physio.*
Water*
Snow
Energ
y*Snow
Energy*
Soil
0
0.8
2 6
101418222630343
8424
650
545862667074
7882
86
Surface Runoff
RadiationPhysio.
Soil
StructureVegetation
Physiology
Soil
Water
Runoff
Input*
Radiation*
Transfer*
Physio.*
Water*
Snow
Energ
y*Snow
Energy*
Soil
0
0.4
2 6
101418222630343
8424
650
545862667074
7882
86
Latent Heat
RadiationPhysio.
Soil
StructureVegetation
Physiology
Soil
Water
Runoff
Input*
Radiation*
Transfer*
Physio.*
Water*
Snow
Energ
y*Snow
Energy*
Soil
0
0.1
2 6
101418222630343
8424
650
545862667074
7882
86
Transpiration
• Few informative parameters show highsensitivity
• Large number of snow parameters impor-tant in snowy catchments and these areall hidden
• Runoff in snowy catchments sensitive tohidden separation of rain and snow
• Saturated hydraulic conductivity very im-portant for runoff processes
• Influence of vegetation parameters de-pends on land cover type
• Sensitivities of underground runoff aremore similar to those of latent heat thanto those of surface runoff
• Sensitivities of latent heat dominated byevaporation processes
• Hidden parameter in formulation of sur-face resistance for evaporation is over-sensitive, similar to findings for CLM v3.5[2]
5. Sensitivity Analysis
[1] M. Cuntz, J. Mai et al., “Computationally inexpensive identificationof non-informative model parameters by sequential screening,” Jan2015, submitted to WRR.
[2] M. Gohler, J. Mai, and M. Cuntz, “Use of eigendecomposition in aparameter sensitivity analysis of the Community Land Model,” JGR,vol. 118, pp. 904–921, 2013.