Innsbruck talk, May 2009 - School of GeoSciences

Improving understanding and forecasts of
the terrestrial carbon cycle
Mathew Williams
School of GeoSciences, University of Edinburgh
With input from: BE Law, A Fox, RF Fisher, J Grace, J
Moncrieff, T Hill, P Meir, REFLEX team
Motivation
 How is the Earth changing?
 What are the consequences of these
changes for life on Earth?
The Global Carbon Cycle – a simple model
The Carbon Cycle
Atmosphere
Fossil
Fuels (7 per yr) &
volcanoes
Climate
Litterfall/
sedimentation
Vegetation
Ocean
Respiration
Photosynthesis
Combustion
Soils
Sediments
Understanding, prediction and control of the Carbon cycle
Research Vision
 To use EO data to test, constrain, modify and
evolve models of the terrestrial biosphere
 To focus on uncertainty throughout the process
of linking observations to models
 To guide experimental and observational
science towards critical areas of uncertainty
 To generate global bottom-up estimates of the
terrestrial C cycle with quantified uncertainty
Outline
 The problems
 Progress so far
 Challenges for the future
Intercomparison of 11 coupled carbon climate
models
Friedlingstein et al 2006: C4MIP
Matrix of R2 for simulations of mean annual GPP for 36 major watersheds in
Europe from different process- and data oriented models
Williams et al. 2009, BGD
Time and space scales in ecological processes
time
dec
Nutrient
cycling
yr
Succession
Climate
change
Adaptation
Disturbance
month
Climate variability
Growth and phenology
day
Photosynthesis and
respiration
hr
Flask
Site
Physiology
s
Space (km)
0.1
1.0
10
100
1000
10000
Time and space scales in ecological observations
time
dec
Flask
Site
MODIS
yr
GOSAT
month
Flux
Tower
Tall
tower
day
Field
Studies
Aircraft
hr
Flask
Site
s
Space (km)
0.1
1.0
10
100
1000
10000
Williams et al. 2009, BGD
Progress so far in MDF
 Model-data fusion with multiple constraints to
improve analyses of C dynamics (Williams et
al. 2005, GCB)
 Assimilating EO data to improve C model state
estimation (Quaife et al. 2008, RSE)
 REFLEX: Intercomparison experiment on
parameter estimation using synthetic and
observed flux data (Fox et al, in press, AFM)
 “Improving land surface models with FLUXNET
data” (Williams et al 2009, BGD)
C cycling in Ponderosa Pine, OR
Flux tower (2000-2)
Sap flow
Soil/stem/leaf respiration
LAI, stem, root biomass
Litter fall measurements
Chambers
Sap-flow
A/Ci
Chambers
EC
Williams et al GCB (2005)
Time (days since 1 Jan 2000)
Time (days since 1 Jan 2000)
Photosynthesis &
plant respiration
Senescence
&
disturbance
Phenology &
allocation
Af
Cfoliage
Microbial &
soil processes
Lf
Rh
Ra
GPP
Ar
Croot
Lr
Clitter
D
Climate
drivers
Feedback from Cf
Aw
Cwood
Lw
Simple linear functions
C
SOM/CWD
Non linear f(T)
The Kalman Filter
Initial state
Drivers
At
MODEL
P
Forecast
Ft+1
Predictions
OPERATOR
F´t+1
Assimilation
At+1
Analysis
Observations
Dt+1
= observation
— = mean analysis
| = SD of the analysis
Time (days since 1 Jan 2000)
Williams et al GCB (2005)
= observation
— = mean analysis
| = SD of the analysis
Time (days since 1 Jan 2000)
Williams et al GCB (2005)
Data bring confidence & test the model
-2
a) Model only: -251 ±197 g c m
2
2
-2
b) GPP data + model: -413±107 gC m
1
0
0
-1
-2
-2
-4
-4
0
-2
-1
NEE (g C m d )
-3
365
730
1095
0
365
730
1095
= observation
— = mean analysis
| = SD of the analysis
-2
c) GPP & respiration data + model: -472 ±56 gC m
2
-2
d) All data: -419 ±29 g C m
2
1
0
0
-1
-2
-2
-3
-4
0
365
730
1095
-4
0
365
Time (days, 1= 1 Jan 2000)
Williams et al, GCB (2005)
730
1095
REFLEX experiment
 Objectives: To compare the strengths and weaknesses
of various MDF techniques for estimating C model
parameters and predicting C fluxes.
 Evergreen and deciduous models and data
 Real and synthetic observations
 Multiple MDF techniques
 Links between stocks and fluxes are explicit
www.carbonfusion.org
Parameter constraint
“truth”
Fox et al. in press
Consistency among methods
Confidence intervals constrained by the data
Consistent with known “truth”
DALEC Model
Ra
Clab
Atolab
Afromlab
Cf
Lf
Cr
Lr
Rh1
Af
GPP
Ar
Clit
D
Aw
Cw
Fox et al. in press
Lw
CSOM
Rh2
Fox et al. in press
Problems with SOM and wood
Fox et al. in press
Problems so far




Varied estimation of confidence intervals
Equifinality
Problems in defining priors
Multiple time scales of response
Challenges for the future
FLUXNET
Quantifying model skill across biomes
Williams et al. 2009, BGD
WP6 Earth
observation
WP4 Towers
WP5 Airborne
Arctic Biosphere-Atmosphere
Coupling across multiple Scales
ABACUS
WP1 Plants
WP2 Soils
WP Moss
WP York
WP3 Fluxes
Other data constraints?








Tree rings
FPAR, NDVI, EVI time series
Stem inventories
chronosequences
Phenology observations
Soil moisture, LE, stream-flow
Surface temperature
Soil chambers
Manipulation Experiments
SPA model output vs. data
Control : R2=0.81
Rp
 lmin
vK
v
Drought : R2=0.75
LAI
Root
5
Fisher et al. 2007
Soil-Root Resistance
(modelled)
Met.
Links to atmospheric CO2 observations…
Workflow for interpretation of GOSAT, flask, aircraft and tall tower data
Land
surface
model
Atmos.
transport
Model XCO2
Aircraft/
ground XCO2
Satellite XCO2
Calibration/
Validation
Satellite XCO2
vs
Global C fluxes
Science
questions
MODIS
Assimilation
Flux analysis
Fire
Models
Flasks/aircraft
Ground XCO2
Model
intercomparison
Error/bias
characterisation
Science
questions
Funding support:
NERC
NASA
DOE
Thank you
Information content of data
(——) aircraft soundings + flux data
(- - - -) flux data only; (— — —) aircraft soundings only
Hill et al. in prep.
Quantifying
driver
uncertainty in
carbon flux
predictions
Spadavecchia et al. in prep.
Parameter retrieval from a synthetic experiment
using the DALEC model using EnKF
Williams et al. 2009, BGD