NERC Centre for Terrestrial Carbon Dynamics & University of Sheffield
Assimilating EO Data into Terrestrial Carbon Cycle Models
Shaun Quegan (+ CTCD, CESBIO, JRC et al.)
The current challenge in C cycle research
Objective To produce estimates & predictions of ecosystem carbon exchange with quantifiable uncertainty.
Complications Observations have gaps & instrumental weaknesses. Models tend to oversimplify and may miss key processes and linkages.
Solution Data assimilation provides a method to combine models and data to produce a more accurate description of ecosystem dynamics.
Global Carbon Data Assimilation System
Geo-referenced emissions inventories
Geo-referenced emissions inventories
Atmospheric measurements
Atmospheric measurements
Remote sensing of atmospheric CO2
Remote sensing of Remote sensing of atmospheric COatmospheric CO22
Atmospheric Transport Model
Atmospheric Transport Model
Ocean Carbon Model
Ocean Carbon Model Terrestrial
Carbon ModelTerrestrial
Carbon Model
Remote sensing of vegetation properties
Growth cycleFires
BiomassRadiation
Land cover/use
Ocean remote sensingOcean colour
AltimetryWinds
SSTSSS
Water column inventories
Ocean time seriesBiogeochemical
pCO2
Surface observation
pCO2nutrients
Optimised model
parameters
Optimised model
parameters
Optimised fluxes
Optimised fluxes
Ecological studies
Biomass soil carbon
inventories
Eddy-covariance flux towers
Coastal studiesCoastal studies
rivers
Lateral fluxes
Data assimilation
link
Climate and weather fields
Geo-referenced emissions inventories
Geo-referenced emissions inventories
Atmospheric measurements
Atmospheric measurements
Remote sensing of atmospheric CO2
Remote sensing of Remote sensing of atmospheric COatmospheric CO22
Atmospheric Transport Model
Atmospheric Transport Model
Ocean Carbon Model
Ocean Carbon Model Terrestrial
Carbon ModelTerrestrial
Carbon Model
Remote sensing of vegetation properties
Growth cycleFires
BiomassRadiation
Land cover/use
Ocean remote sensingOcean colour
AltimetryWinds
SSTSSS
Water column inventories
Ocean time seriesBiogeochemical
pCO2
Surface observation
pCO2nutrients
Optimised model
parameters
Optimised model
parameters
Optimised fluxes
Optimised fluxes
Ecological studies
Biomass soil carbon
inventories
Eddy-covariance flux towers
Coastal studiesCoastal studies
rivers
Lateral fluxes
Data assimilation
link
Climate and weather fields
Source: Ciais et al. 2003 IGOS-P Integrated Global Carbon Observing Strategy
Terrestrial component
Remote sensing of atmospheric CO2
Remote sensing of Remote sensing of atmospheric COatmospheric CO22
Atmospheric Transport Model
Atmospheric Transport Model
Terrestrial Carbon Model
Terrestrial Carbon Model
Remote sensing of vegetation properties
Growth cycleFires
BiomassRadiation
Land cover/use
Optimised model
parameters
Optimised model
parameters
Optimised fluxes
Optimised fluxes
Ecological studies
Biomass soil carbon
inventories
Eddy-covariance flux towers
rivers
Lateral fluxes
Climate and weather fields
Remote sensing of atmospheric CO2
Remote sensing of Remote sensing of atmospheric COatmospheric CO22
Atmospheric Transport Model
Atmospheric Transport Model
Terrestrial Carbon Model
Terrestrial Carbon Model
Remote sensing of vegetation properties
Growth cycleFires
BiomassRadiation
Land cover/use
Optimised model
parameters
Optimised model
parameters
Optimised fluxes
Optimised fluxes
Ecological studies
Biomass soil carbon
inventories
Eddy-covariance flux towers
rivers
Lateral fluxes
Climate and weather fields
Carbon Cycle – Earth Observation Interfaces
VEGETATION MODELLINGCLIMATE &
WEATHER
VEGETATION-SOILCARBON
ALLOCATION
SOIL TYPESOIL CARBON
EARTH OBSERVATION
LAND COVER
INITIALISATION
DISTURBANCE,PHENOLOGY
CALIBRATION
BIOMASS
TESTING
fAPAR, SNOW COVER
ASSIMILATION
This lecture
Issues
u Monitoring Consistency of models and data:
- are model and measurement quantities compatible?
- comparability of valuesTimescales (re-analysis)
u PredictionAre model internal processes and parameters testable and credible?
S is the state vector describing the vegetation-soil system.
Soil
1S
Climate
DVM 2S
Parameters
The Functioning of a DVM
Calibrating the SDGVM phenology module with EO data
SIBERIA-II: Multi-Sensor Concepts for Greenhouse Gas Accounting of Northern Eurasia
5th Framework Project , 2002-2005
Land cover (IIASA) 1o x 1o forest map
Lake Baikal
80E 120E 80E 120E
50N
75N
The Central Siberia dataset: ~ 2 M km2
The CESBIO budburst algorithm
u Data set: SPOT-VEG 1999-2001u Based on minimum in time-series of NDWI datau Uncertainties in recovered budburst date ~ 7
days
The Date of budburst derived from minimum NDWI (VGT sensor, 2000) N. Delbart, CESBIO
Day of year
Start of budburst
T0
∑days
min(0, T – T0) > Threshold, budburst occurs.
The sum is the red area. Optimise over the 2 parameters, Threshold and T0 (minimum effective temperature).
When
The SDGVM budburst algorithm
The calibration procedure
Data - model comparison 1999
Budburst from NDWI dataModel budburst:
optimised parameters
Calibration parameters (forest only)
7.0294.42000
6.0117-2.91999
MMSE(days)
Threshold(degree-
days)
T0
(degrees)Year
Green-up relation to N Pacific Index
Effect of uncertainty in green-up day
‘True’ Assimilation
Vegetation model Scattering or reflectance model
Compare prediction with measurement
Modify model state to improve consistency between data and prediction
Observation model(forward model)
Basis of radiation models (optical)
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
400 600 800 1000 1200 1400 1600 1800 2000 2200 2400
wavelength / nm
no
rmal
ised
ab
sorp
tio
n
chlorophyll
water
dry matter
Leaf scattering model
pigments
‘structure’
dry matter
water
Leaf reflectance and transmittance
Basis of radiation models (optical)
u Model of canopy scattering:– Leaf properties– Scattering object density
(LAI), orientation, and spatial distribution
– Soil / understoreyproperties for low density canopies
u solutions by analytical or numerical methods
Link to the C models
u C models include concept of radiation model– For calculation of intercepted radiation
u Observation model– Provides link from subset of C-model state
variables to EO observation– Main linkages:
u LAI, Density (for limited conditions)u leaf properties (hyperspectral data)
– leaf dry matter, chlorophyll (nitrogen), water– xanthophyll cycle (light use efficiency)
Exploiting quantities derived from radiance
CLIMATECO2
SOIL
VEGETATION PHYSIOLOGY AND
BIOPHYSICS
DISTURBANCEGENERATOR
VEGETATION DYNAMICS
VEGETATION PHENOLOGY
NUTRIENT CYCLING
time
LAI
The propagation of the light follows the Beer-Lambert Law:
).exp(.)( lKPARlI −=One gets:
).exp(1 LAIKfAPAR −−=
1=l2=l
LAIl =
PARLight propagation
-> In SDGVMd, fAPAR and LAI are linked in a deterministic way.
Sheffield Dynamic Vegetation Global Model
Application of the models
u Testing C models (SDGVM)– Confront predictions with observations
Time update“predict”
Measurement update
“correct”
A prediction-correction system
Earth Observation
-Phenology,-LAI,-Land cover…
PAR (weekly)
LAI/FAPAR(weekly)
APAR (weekly)
Pot. GPP(weekly)
ΕLand cover(yearly)
*
Light
GPP (daily)
Gs (daily)Temp(daily)Temp
Rainfall-Runoff=ET + ε
Simple SVAT model at the catchmentscale:
TranspirationEvaporation
(daily)
Rainfall,temp, humidty, LAI(daily)
Water
Run Off(daily)
Assimil. Obs. Run Off(monthly)
Pixelscale
Williams et al., 2004
GPP Croot
Cwood
Cfoliage
Clitter
CSOM/CWD
Ra
Af
Ar
Aw
Lf
Lr
Lw
Rh
D
Temperature controlled
6 model pools10 model fluxes9 rate constants10 data time series
Rtotal & Net Ecosystem Exchange of CO2
C = carbon poolsA = allocationL = litter fallR = respiration (auto- & heterotrophic)
0 365 730 1095-4
-3
-2
-1
0
1
2
0 365 730 1095-4
-2
0
2
Time (days, 1= 1 Jan 2000)
b) GPP data + model: SD=123
0 365 730 1095-4
-3
-2
-1
0
1
2 c) GPP & respiration data + model: SD=53
NE
E (g
C m
-2 d
-1)
0 365 730 1095-4
-2
0
2
a) Model only: SD=364
d) All data: SD=26
0
50
100
150
200
0
100
200
300
400
0 365 730 1095
600
800
1000
1200
Foliage
Cf (
gC m
-2)
Fine root
Cr (
gC m
-2)
Wood
Cw (g
C m
-2)
Cf
Cr
Cw
Cl
Cs
GPP
Wc
WS1
WS2
WS3
E T
Ppt
Flux emulator State variable
Flux
Q
Rh
Predictedflux
Carbon modelHydrology
Pools and fluxes
Ra
Cf
Cr
Cw
Cl
Cs
GPPRadiation, VPDtemperature
Wc
WS1
WS2
WS3
E T
Ppt
Radiation, VPDtemperature
Soilsdata
Driver data Flux emulator
State variable
Flux
Input to emulator
Q
Rh
Predictedflux
Ra
Linking C and water fluxes
Cf
Cr
Cw
Cl
Cs
GPP
LAI data
Biomass
Radiation, VPDtemperature
Wc
WS1
WS2
WS3
E T
Ppt
Radiation, VPDtemperature
Soilsdata
Assimilateddata Driver data Flux emulator
State variable
Observationoperator
Flux
Data assimilation
Input to emulator
Q
Local tower
Rh
Predictedflux
Ra
Local assimilation
Cf
Cr
Cw
Cl
Cs
GPP
Reflectance
Radar
Radiation, VPDtemperature
Wc
WS1
WS2
WS3
E T
Ppt
Radiation, VPDtemperature
Soilsdata
Assimilateddata Driver data Flux emulator
State variableObservation operator
Flux
Data assimilation
Input to emulator
QRiver gauges
Tall tower &Aircraft
Rh
Predictedflux
Ra
Regional assimilation
Conclusions
u Calibration of DVM parameters with EO data provides a means to improve the predictive power of the models, e.g., phenology, fire.
u Well-developed forward models for scattering and reflectance exist; a current challenge is to interface them to biosphericmodels for monitoring and assimilation.
u Because of possible problems in derived products, such asfAPAR, assimilation of radiances sems preferable. However, this is dependent on how radiation absorption is represented in thebiospheric model.
u Successful assimilation schemes have just been developed forbiospheric models. By using existing forward models, these provide a framework for assimilating EO data.
u Watch this space!