Predictability of the Terrestrial Carbon Cycle
Yiqi Luo University of Oklahoma, USATsinghua University, China
Trevor Keenan Harvard University, USAMatthew Smith Microsoft Research, Cambridge, UKYingPing Wang CSIROJianyang Xia University of Oklahoma, USASasha Hararuk University of Oklahoma, USAEnsheng Weng Princeton University, USAYaner Yan Fudan University, China
http://[email protected]
Soil carbon modeled in CMIP5 vs. HWSD
Yan et al. submitted
IPCC assessment report
Friedlingsterin et al. 2006
1. Great uncertainty among models
2. Does the uncertainty reflect variability in the nature or result from artifacts?
Current efforts to improve the
predictive understanding
Satellite measurement of CO2
中科院碳循环研究专项
Observation and experiment
Can they make IPCC assessment report better?
Modeling
Method Merit Issue
Adding components
Simulating more processes realistically
Complexity Intractability
Model intercomparison
Illustrating uncertainty
Attribution to its sources
Benchmark analysis
Performance skills Objective benchmarks
Data assimilation Improving models Various challenges
Theoretical analysis1. The terrestrial carbon cycle is a relatively simple
system
2. It is intrinsically predictable.
3. The uncertainty shown in model intercomparison studies can be substantially reduced with relatively easy ways
4. We should sharpen our research focus on key issues
1. Photosynthesis as the primary C influx pathway
2. Compartmentalization,
3. Partitioning among pools
4. Donor-pool dominated carbon transfers
5. 1st-order transfers from the donor pools
Properties of terrestrial carbon cycle
Luo and Weng 2011
A: Basic processesB: Shared model structure
C: Similar algorithmD: General model
Model development
Enc
odin
g
The
oret
ical
an
alys
is
Generalization
Leaf (X1) Wood (X3)
Metabolic litter (X4)
Microbes (X6)
Structure litter (X5)
Slow SOM (X7)
Passive SOM (X8)
CO2
CO2CO2
CO2
CO2
CO2
CO2
CO2
Photosynthesis
Root (X2)
Luo et al. 2003 GBCLuo and Weng 2011 TREELuo et al. 2012Luo et al. submitted
Theoretical analysis
General equations
Empirical evidence First-order decay of litter decomposition (Zhang et al. 2008) Carbon release from soil incubation data (Schaedel et al. 2013) Ecosystem recovery after disturbance (Yang et al. 2011)
Model structure analysis 11 models in CMIP5 (Todd-Brown et al. 2013)
1. Internal C processes to equilibrate efflux with influx as in an example of forest succession
2. C sink strength becomes smaller as efflux is equalized with influx
3. When initial values of C pools differ, the magnitude of disequilibrium varies without change in the equilibrium C storage capacity.
Focusing research on dynamic disequilibrium
Convergence
An ultimate goal of carbon research is to quantify
Carbon-climate feedbackWhich occurs only when carbon cycle is at
disequilibriumLuo and Weng 2011
Periodic climate(e.g., seasonal)
Periodicity
Disturbance event(e.g. fire and land use)
Pulse-recovery
Climate change(e.g., rising CO2)
Gradual change
Disturbance regime disequilibrium
Ecosystem state change(e.g., tipping point)
Abrupt change
External forcing Response
Given one class of forcing, we likely see a highly predictable pattern of response
Predictability of the terrestrial carbon cycle
Luo, Smith, and Keenan, submitted
Terrestrial carbon cycle
Nonautomatous system
working groupNonautonamous system
The 3-D parameter space is expected to
bound results of all global land models and
to analyze their uncertainty and traceability.
Computational efficiency of spin-up
Xia et al. 2012 GMD
Computational efficiency of spin-up
Xia et al. 2012 GMD
Semi-analytical spin-up with CABLE
Initial step: 200 yearsFinal step: 201 years
Initial step: 200 yearsFinal step: 483 years
Traditional: 2780 years
Traditional: 5099 years
-92.4%
-86.6%
Xia et al. 2012 GMD
http://ecolab.ou.eduXia et al. 2013 GCB
'1EE
WT
http://ecolab.ou.eduXia et al. 2013 GCB
Traceability of carbon cycle in land models
'E
E
NPP ( )ssU
ssX
Preset Residence times
Soil textureLitter lignin fraction
Climate forcing
Precipitation Temperature
Tw
http://ecolab.ou.eduXia et al. 2013 GCB
Traceability of carbon cycle in land models
Traceability for differences among biomes
Based on spin-up results from CABLE with 1990 forcings.
http://ecolab.ou.edu
0
100
200
300
400
500
0 500 1000 1500 2000
Res
iden
ce ti
me
(Yea
r)
(a)
NPP (g C m-2 year-1)
0
40
80
120
160
0 500 1000 1500
ENF EBF
DNF DBF
Shrub C3G
C4G Tundra
Barren
4030
20101
(b)
NPP (g C m-2 year-1)
Res
iden
ce ti
me
(Yea
r)
(τE)
(τE)
Long τE but low NPP.
High NPP but short τE.
Xia et al. 2013 GCB
Act
ual r
esid
ence
tim
e (y
ear)
Baseline residence time (year)
1 3 10 32 100 3161
3
10
32
100
316
Traceability for model intercomparison
http://ecolab.ou.eduModel-model Intercomparison
CABLECLM3.5
ENF EBFDNF DBFShrub C3GC4G Tundra
Traceability for impact of additional model component
http://ecolab.ou.edu
0
60
120
180
0 500 1000 1500
Res
iden
ce ti
me
(yea
r)
NPP (g C m-2 year-1)
ENF EBF
DNF DBF
Shrub C3G
C4G Tundra
Barrens
4030
20101
(a)
0
10
20
30
N in
duce
d re
duct
ions
(%)
(b)
EN
F
EB
F
DN
F
DB
F
Shr
ub
C3G
C4G
Tund
ra
Bar
ren
NPPτE
'
Xia et al. 2013 GCB
Carbon influx
Initial values of carbon pools
Environmental scalar
Transfer coefficient
Partitioning coefficient
Residence time
Sources of model uncertainty
Sources of model uncertainty
Carbon influx
Initial values of carbon pools
Residence time
Soil carbon modeled in CMIP5 vs. HWSD
Todd-Brown et al. 2013 BG
Initial values of carbon pools
Carbon influx
Residence time
Soil carbon modeled in CMIP5 vs. HWSD
Yan et al. submitted
Data assimilation to reduce uncertainty
A: Basic processesB: Shared model structure
C: Similar algorithmD: General model
Model development
Enc
odin
g
The
oret
ical
an
alys
is
Generalization
Leaf (X1) Wood (X3)
Metabolic litter (X4)
Microbes (X6)
Structure litter (X5)
Slow SOM (X7)
Passive SOM (X8)
CO2
CO2CO2
CO2
CO2
CO2
CO2
CO2
Photosynthesis
Root (X2)
Luo et al. 2003 GBCLuo and Weng 2011 TREELuo et al. 2012Luo et al. submitted
Summary
A: Basic processes
D: General model The
oret
ical
an
alys
is
Luo et al. 2003 GBCLuo and Weng 2011 TREELuo et al. 2012Luo et al. submitted
Applications
a. Research focus on dynamic disequilibrium (Luo and Weng 2011)
b. Computational efficiency of spin-up (Xia et al. 2012)
c. Traceability for structural analysis (Xia et al. 2013)
d. Predictability of the terrestrial carbon cycle (Luo et al. submitted)
e. Sources of uncertaintyf. Data assimilation to improve
models (Hararuk et al. submitted)
g. Parameter space (work in progress)
Periodic climate(e.g., seasonal) Periodicity
Disturbance event(e.g. fire and land use)
Pulse-recovery
Climate change(e.g., rising CO2)
Gradual change
Disturbance regime disequilibriumEcosystem state change(e.g., tipping point) Abrupt change
External forcing Response
Relevance to empirical research
Terrestrial carbon cycle
1. Find examples to refute this equation2. Disequilibrium vs. equilibrium states3. Disturbance regimes have not been quantified4. Mechanisms underlying state change have not been understood5. Response functions to link carbon cycle processes to forcing are not well
characterized.6. Disturbance-recovery trajectory, especially to different states, is not
quantified
Relevance to modeling studiesMethod Merit IssueAdding components Simulating more processes realistically Complexity Intractability
Model intercomparison Illustrating uncertainty Attribution to its sources
Benchmark analysis Performance skills Objective benchmarks
Data assimilation Improving models Various challenges
1. Benchmarks to be developed from data and used to evaluate models2. Traceability of modeled processes3. Pinpointing model uncertainty to its sources 4. Data assimilation to improve models5. Standardize model structure for those processes we well understood but
allow variations among models for processes we have alternative hypotheses
6. Parameter ensemble analysis