Asseng et al. 2014 Nature CC

Report
Uncertainties of climate change impacts in agriculture
Senthold Asseng
F. Ewert, P. Martre, R.P. Rötter, D.B. Lobell, D. Cammarano, B.A. Kimball, M.J. Ottman, G.W. Wall,
J.W. White, M.P. Reynolds, P.D. Alderman, P.V.V. Prasad, P.K. Aggarwal, J. Anothai, B. Basso,
C. Biernath, A.J. Challinor, G. De Sanctis, J. Doltra, E. Fereres, M. Garcia-Vila, S. Gayler,
G. Hoogenboom, L.A. Hunt, R.C. Izaurralde, M. Jabloun, C.D. Jones, K.C. Kersebaum,
A.-K. Koehler, C. Müller, S. Naresh Kumar, C. Nendel, G. O’Leary, J. E. Olesen, T. Palosuo,
E. Priesack, E. Eyshi Rezaei, A.C. Ruane, M.A. Semenov, I. Shcherbak, C. Stöckle, P. Stratonovitch,
T. Streck, I. Supit, F. Tao, P. Thorburn, K. Waha, E. Wang, D. Wallach, J. Wolf, Z. Zhao and Y. Zhu
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
Overview
1.
AgMIP
2.
Crop models – modeling CO2
3.
Model uncertainty
a) What is it?
b) Quantification
c)
Comparison with other sources
d) Can it be reduced?
4.
Conclusions
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
AgMIP
Agricultural Model Intercomparison and Improvement Project
Led by Cynthia Rosenzweig, James W. Jones, Jerry Hatfield & John Antle
Goals
 To improve the characterization of risk of hunger and world food security
due to climate change,
 To enhance adaptation capacity in both developing and developed countries.
www.agmip.org
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
AgMIP Wheat
AgMIP
AgMIP
Wheat
30 wheat models
AgMIP :
• combines climate – crop – economic models in a multi-model approach
• started in 2010, open, > 600 members from around the world, >30 projects
Rosenzweig et al. 2013 AFM
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
Wheat yield and climate
CO2
Light
Temperature
H2O
Management
Genotype
Carter 2013
Soil
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
Scale
ExM
P
G
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
Crop
models
APSIM - NWheat
max & min
temperature
rainfall
Es
Ep
solar
radiation
CO2
C
TUE
Harvest
C
Assimilation
C,N
C
Grain
Shoot + Leaf
N
N
Root
N
Nwheat
C,N
runoff
Denitrification
Residues
(surface)
Fertiliser
CO2
SoilN
1
Residues
(roots)
2
FOM
3
Mineralisation
carbohydartes
cellulose
Immobilisation
lignin
n
LL
DUL
SoilWAT
Mineral-N
NH4
urea
NO3
NH4
CO2
SAT
CO2
Drainage
BIOM
C:N
CO2
HUM
C:N
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
Leaching
Model output
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
Models vs observations
input
Model
output
output
1. wrong input
observation
 observation
2. wrong/poor estimate for input
3. wrong observation
4. wrong model/routine
a) wrong number
b) wrong unit
c) value with large variability
d) outside model design
c) ‘not a measurement’ - just another ‘model’
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
Modeling CO2
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
Photosynthesis
leaf
cell
CO2 H2O
CO2 + H2O + light energy ---> C6H12O6 + O2 + H2O
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
Simple approaches to compute Photosynthesis:
RUE - model
RUE
Monteith 1977 PTRSL
Sinclair and Weiss (2010) In: Principles
of Ecology in Plant Production
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
Climate change - Photosynthesis
leaf
CO2 H2O
CO2 + H2O + light energy ---> C6H12O6 + O2 + H2O
 Radiation use effciency (RUE) and transpiration effciency (TE)
both increases with increased CO2
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
RUE model
RUE = Radiation use efficiency in g[crop] MJ-1[intercepted light]
incoming light
interception
dW/dt = RUE x FCO2 x I0 x [1-exp(-k . LAI)]
1.30
o
1.25
25 C
o
FFCO
CO
2
20 C
1.20
o
15 C
2
10 oC
1.15
1.10
1.05
Reyanga et al. 1999 EMS
1.00
0.95
300
400
500
600
Atmospheric CO2 (ppm)
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
700
Observed grain yield – CO2 effect
Grain Yield (t/ha)
12
10
Observed data after
Kimball et al. 1995 GCB
+CO2 = 550ppm
(by 2050)
8
6
4
2
0
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
Observed grain yield – CO2 effect
Grain Yield (t/ha)
12
10
Observed data after
Kimball et al. 1995 GCB
+CO2 = 550ppm
(by 2050)
8
6
4
2
0
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
Observed & simulated grain yield – CO2 effect
Grain Yield (t/ha)
12
observed & simulated
10
Asseng et al. 2004 FCR
8
6
4
2
0
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
Uncertainty
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
Modeling climate impact
Climate models
Climate models
Impact model
Impact model
Climate models
(+scenarios)
e.g. Crop model
(or model for:
- hydrology,
- biodiversity,
- health…)
Impact model
e.g. Economic model
(or model for:
- land-use…)
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
A meta-analysis of crop yields (wheat)
Challinor et al. 2014 Nature CC
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
Distinguishing variability from uncertainty
Due to model, process,
measurements errors
Variability = Natural
variability in space & time
e.g. impact simulation
Lehmann & Rillig 2014 Nature CC
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
Distinguishing variability from uncertainty
Natural variability
Uncertainty
Time
After Lehmann & Rillig 2014 Nature CC
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
AgMIP Wheat - Background
1. Crop model = main tool to assess climate change impact
2. But, simulated effect due to chosen crop model ?
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
AgMIP Wheat Pilot

27 wheat models

4 contrasting field experiments (natural variability)

Standardized protocols
• “Blind test”
• Full calibration
• Sensitivity analysis
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
AgMIP Wheat Pilot
27 wheat models
4 contrasting field
experiments
Wheat area after Monfreda et al. (2008)
Above-ground biomass (t/ha)
20
ME 11, High rainfall; cold temperature, winter wheat
ME 2, High rainfall; temperate temperature, spring wheat
ME 1, Irrigated; temperate temperature, spring wheat
ME 4, Low rainfall; temperate temperature, spring wheat
15
10
5
CIMMYT’s mega-environments (ME) for wheat
0
50
100
150
200
250
300
Days after sowing
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
Observations versus simulations
NL
O b s e rv e d
Line = median
Box = 50%
Bars = 80%
"B lin d "
C a lib ra te d
AR
O b s e rv e d
"B lin d "
C a lib ra te d
IN
O b s e rv e d
"B lin d "
C a lib ra te d
AU
O b s e rv e d
"B lin d "
Asseng et al. 2014 Nature CC
C a lib ra te d
0
2
4
6
8
10
G ra in y ie ld (t/h a )
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
Observations versus simulations
L o c a tio n
N L + A R + IN + A U
NL+AR+AU
N L + IN + A U
A R + IN + A U
N L + A R + IN
IN + A U
AR +AU
A R + IN
NL+AU
N L + IN
NL+AR
AU
IN
AR
NL
N .a .N .
Fully calibrated
25
20
15
10
5
0
“Blind”
N u m b e r o f m o d e ls w ith in
1 3 .5 % o f o b s e rv a tio n
13.5% = coefficient of variation for field experimental observation (Taylor et al. 1999)
Asseng et al. 2014 Nature CC
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
Observations versus simulations
G ra in Y ie ld
G ra in N u m b e r
G ra in P ro te in
Above-ground biomass (t/ha)
H a rv e s t In d e x (H I)
20
B io m a s s @ A n th e s is
15
B io m a s s @ M a tu rity
“Blind”
10
M a x im u m L A I
5
C u m u la tiv e E T
0
C ro p N @ A n th e s is
50
100
150
200
250
Days after sowing
300
C ro p N @ M a tu rity
G ra in N
Asseng et al. 2014 Nature CC
0
10
20
RMSE %
Fully calibrated
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
30
40
Model detail
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
Model detail
Relative
RMSE
(%)
90
80
70
60
50
40
30
20
10
0
R² = 0.21
0
5
10
15
20
Number of cultivar parameter (#)
Challinor et al. 2014 Nature CC
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
25
Model response to changes in T, rainfall and CO2
CV% of model response to climate change scenario (A2 2100)
“Blind”
fully calibrated
50% of models with the closest simulations to the
observed yields across all location
50% of models with closest simulation per location
NL
30
20
10
AU 30
20
AR
10
10
10
20
30
IN
20
30
 e.g. the best models (i.e. smallest RMSE with
observations) have smallest CV at 3 locations,
but not at AU;
i.e. performance of models with historical data
is no guidance for future impact studies
Asseng et al. 2014 Nature CC
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
Model response to rainfall
Argentina
Australia
Line = median
Box = 50%
Bars = 80%
R a in fa ll c h a n g e (% )
25
10
0
-1 0
-2 5
-1 2 0 -8 0 -4 0
0
40
80 120
-1 2 0 -8 0 -4 0
0
40
S im u la te d % y ie ld c h a n g e
Asseng et al. 2014 Nature CC
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
80 120
Model response to CO2 and T
Argentina
Australia
0
550 ppm
+3
Temperature change (oC)
Atmospheric CO2 concentration (ppm)
720
540
360
observed
+6
+3oC
00
Plot 1
+3
+3
+6oC
+6
+6
0
-80 -40
+3
0
40
80
Observed
(% yield change)
+6
-120 -80 -40
0
40
80 120 -120 -80 -40
0
40
80 120
Simulated % yield change
Line = median
Box = 50%
Bars = 80%
Asseng et al. 2013 Nature CC
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
CO2 response: Amthor 2001, Ewert
et al. 2002, Hogy et al. 2010,
Kimball 2011, Ko et al., 2010, Li et
al. 2007
T response (extrapolated): Amthor
2001, Singh et al. 2008, Xiao et al.
2005
Model response to heat stress
NL
AR
7 x 35 oC after
anthesis
IN
AU
-40
-20
0
20
Simulated relative heat impact (%)
Asseng et al. 2013 Nature CC
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
Line = median
Box = 50%
Bars = 80%
Models with heat stress routine
What about other crops?
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
Maize model response
23 models
Bassu et al. 2014 GCB
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
Crop models vs GCMs
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
Modelling climate impact
Climate models
Climate models
(+scenarios)
Impact model
e.g. Crop model
(or model for:
- hydrology,
- biodiversity,
- health…)
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
Impact uncertainties
C o e ffic ie n t o f v a ria tio n (% )
A2 scenario for Mid-Century
25
Model uncertainty in simulating
climate change yield impact
20
15
Mean exp CV% (Taylor et al. 1999)
10
Uncertainty due to 16 GCM’s scenarios
5
0
N e th e rla n d s
A rg e n tin a
In d ia
A u s tra lia
Asseng et al. 2014 Nature CC
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
Reducing uncertainty
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
Multi-model ensembles to reduce uncertainty
C o e ffic ie n t o f V a ria tio n (% )
35
o
In d ia : + 3 C & 4 5 0 p p m
30
25
20
13.5% = Mean exp CV%
(Taylor et al. 1999)
15
10
5
0
0
2
4
6
8
10
12
14
16
18
20
22
24
26
N u m b e r o f M o d e ls (# )
Asseng et al. 2014 Nature CC
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
Multi-model ensembles to reduce uncertainty
C o e ffic ie n t o f V a ria tio n (% )
35
30
25
Required number of crop models to achieve
20
15
<13.5% simulated impact variability (-)
10
5
0
0
2
4
6
8
10
12
14
16
18
20
22
24
26
N u m b e r o f M o d e ls (# )
15
12
Colors represent
different CO2 levels
9
6
Mean (+/- STD) of
all locations
3
0
-6
-3
0
3
6
9
Changes in temperature (oC)
Asseng et al. 2014 Nature CC
(13.5% = Mean exp CV% (Taylor et al. 1999))
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
12
Reducing uncertainty
via model improvements
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
Model improvements to reduce uncertainty
PD Alderman, E Quilligan, S Asseng, F Ewert
and MP Reynolds (Editors)
CIMMYT, El Batan, Texcoco, Mexico
June 1921, 2013
Bruce Kimball
Wall et al. 2011 GCB; Ottman et al. 2012 AJ
 Improve high temperature impacts in models
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014
Conclusions
1.
Many of the crop models can reproduce observed experiments
2.
However, there is an uncertainty in climate change impact assessments
due to crop models
3.
This uncertainty is similar to experimental error, but larger than from
GCM’s
4.
Uncertainty in modeling T and T x CO2 interactions
>>> model improvements
5.
Multi-model ensembles can reduce simulated impact uncertainties.
Contact: Senthold Asseng, [email protected]
Senthold Asseng, Seminar for NCAR ASP Summer Colloquium on Uncertainty, Bolder, CO, July 22, 2014

similar documents