1. A strategy for estimating the
components of the carbon and water
budgets for croplands at plot scale over
large areas
Univ. Toulouse III, CNRS, CNES, IRD,
INRA
E. Ceschia, G. Pique, A. Veloso, R. Fieuzal, A. Albitar, JF. Dejoux,
with contributions from A. Brut,T. Tallec, V. Demarez, O. Hagolle, H. Gibrin, B.
Zawilski, C. Marais-Sicre, N. Claverie, F. Granouillac.
3rd
ICOS Science Conference
Prague, September 13th
2018

2. Questioning of the durability of the conventional agriculture (climate change &
other environmental impacts)
Context/societal challenges
Illustrations: Arbre et
Paysage 32
Agro-ecological practices
Lack of large scale multi-criteria diagnostic modelling tools working at plot level to
establish an environmental/agronomical baseline, guide the transition towards agro-
ecological practices and quantify their positive/negative impacts (changes in
ecosystem services)
Strategic challenge for the agricultural profession and the society in general
C storage
= 4/1000
No till
Cover crops
Agroforestery
23rd
ICOS Science Conference, Prague, September 2018

3. • To analyse cropland ecosystem services at plot scale over large
areas (regional to global) : yield, biomass dynamics,
evapotranspiration and net CO2 flux components  annual
carbon & water budgets,
• Test the effect of some management practices (e.g. cover
crops, exporting straw…) on the surface fluxes and C & water
budgets
We developed a multi-temporal high resolution Remote
Sensing (RS) data assimilation scheme in a crop model
(Sentinel…) : avoiding as much as possible the need for ground
data (management)  the Simple Algorithm for Fluxes and
Yield estimates, SAFYE- CO2 (Veloso, 2014 ; Pique G. Fieuzal R.
et al. in prep).
Objectives
3rd
ICOS Science Conference, Prague, September 2018

4. Why is it possible now ?
Clear or cloudy sky conditions
Rugosity & surface water content
Sentinel 1
(10 m, 6j, Radar)
Sentinel 2
(10 m, 5j, Optical)
Clear sky conditions
Reflectances (13 bands)
Land cover LAI/
phenology
Albedo2%
40%
Soil
humidity
Biomass
Land use dynamic mapping
Soil work
Monitored
parameters
Dynamic mapping
fusion
How to use those RS derived products to answer scientific and societal
challenges related to agriculture ? 4
Sentinel
A
Revolution !!!
(free all over
the globe)

5. 5
(since 2002)
(since 2006)
ESU
ESU
The Spatial Regional Observatory (OSR)
Part of the international JECAM & ICOS networks
Sentinel 1&2

6. The Lamasquère ICOS site
More than 250 variables continuoulsy measured since 2005 (same at Auradé)
4 profiles sol (0 à 1m)
Radiation(albedo) et
vertical profiles : wind,
T°C, HR%, CO2
Meteo + soil variables +
fluxes : CO2, H, LE
Soil automatic Chambers :
CO2, N2O
Deported mast Main mast
W
ind direction
Eddy-covariance
method
20 Hz
Humidité T°C
Flux de
chaleur
4 chambers
Inter-row

7. Maïs
Bois
Bâti/Surf. minérale
Blé-Tournesol
Blé- Colza - Tournesol
Eau
Surf. enherbées - Blé
Blé-Colza
Maïs- Blé
Maïs - Tournesol
Surf. enherbées
Surf. enherbées - Tournesol
Maïs
Bois
Bâti/Surf. minérale
Blé-Tournesol
Blé- Colza - Tournesol
Eau
Surf. enherbées - Blé
Blé-Colza
Maïs- Blé
Maïs - Tournesol
Surf. enherbées - Tournesol- Blé
Blé - Tournesol - Sorgho
Blé - Colza - Orge
Maïs - Soja
Surf. enherbées
Surf. enherbées - Tournesol
Bare soil
Early regrowth
Late regrowth
Late cover crops
Early cover crops
D. Ducros, J. Inglada, C. Marais-Sicre , S. Valero , J.F Dejoux,, E. Ceschia
Crop rotations
Autres rotations
Maïs
Bois
Bâti/Surf. minérale
Blé-Tournesol
Blé- Colza - Tournesol
Eau
Surf. enherbées - Blé
Blé-Colza
Maïs- Blé
Maïs - Tournesol
Surf. enherbées - Tournesol- Blé
Blé - Tournesol - Sorgho
Blé - Colza - Orge
Maïs - Soja
Blé - Maïs - Tournesol
Surf. enherbées
Surf. enherbées - Tournesol
CICC & Bag’ages projects
Cover crops
H2020 Sensagri project
Crop mapping based on remote sensing
7
 operational algorithms for annual crop mapping at regional/national scale
(ESA Sentinel 2 Agri & H2020 Sensagri projects),
 mapping of some management practices : cover crops, crop rotations,
irrigation, soil work.
Legend
objectivize the cover crop
development and associated
C storage
3rd
ICOS Science Conference, Prague, September 2018
Sen2agri
Sensagri

8. 29/06/006
26/07/2006
09/09/2006
Cartes de Green Area Index
Biomass,
Yield,
irrigation
CO2 & water fluxes
+
C & water budgets
Validation par stations de
mesures des flux du
Radiative transfer model, SVM…
Crop
mapps
Soil maps
(e.g. global
soil map)
Climatic
data (e.g.
SAFRAN)
29 Juin 2006
Spot, Landsat,
Sentinel 2
Yield
SAFYE-CO2
Monteith approach+FAO56
(Veloso, 2014) Validation with regional stat.
(yield/irrigation)
Validation with regional stat.
(yield/irrigation)
8 km
Modelling approach with SAFYE-CO2
ASW
C
Crop param.
Calibration of
phenology/LUE
Leaf area index
m2
leaves/m2
soil
C budgets
for winter
wheat fields
in 2011
(gC.m-2
)
8
Biomass,SWC
H2020
Sensagri
Dynamic land
use mapp
(crop &
management)
Crop & soilparam.
National observatoriesNational observatories
JECAM Network
Ecosystem component
flux stations
H2020
Sensagri
3rd
ICOS Science Conference, Prague, September 2018

9. 9
Regional estimates for winter wheat
3rd
ICOS Science Conference, Prague, September 2018

10. 10
Net CO2 fluxes (NEP) & C budget (NECB)
Regional estimates for winter wheat
3rd
ICOS Science Conference, Prague, September 2018

11. 11
WUEagronomical = yield or biomass exported/ETR
WUEenvironnemental = C budget /ETR
SAFYE-
CO2
Usefull approach to find compromises between productive and
environmental ecosystem services.
Grains exportésGrain + straw exported
Agronomical vs environmental WUE
Tallec et al (2013)
in AFM

12. Validation des flux nets de CO2 cumulés sur blé
SAFYE-
CO2
Performances/originality of our approach
3rd
ICOS Science Conference, Prague, September 2018
GPP Reco NEE
R² RMSE* Slope R² RMSE* Slope R² RMSE* Slope
AUR2006 0.91 1.42 1.06 0.77 0.80 0.96 0.85 1.21 1.02
LAM2007 0.94 1.4 0.81 0.80 1.23 0.69 0.87 1.06 0.81
AUR2008 0.94 1.26 0.95 0.74 0.89 0.84 0.89 1.05 0.87
LAM2009 0.93 1.13 1.04 0.71 0.90 0.87 0.79 1.19 0.88
AUR2010 0.94 1.27 0.89 0.82 0.85 0.85 0.88 1.28 0.79
1
GPP Reco
R² RMSE* Slope R² RMSE* Slope R
AUR2006 0.91 1.42 1.06 0.77 0.80 0.96 0.8
LAM2007 0.94 1.4 0.81 0.80 1.23 0.69 0.8
AUR2008 0.94 1.26 0.95 0.74 0.89 0.84 0.8
LAM2009 0.93 1.13 1.04 0.71 0.90 0.87 0.7
AUR2010 0.94 1.27 0.89 0.82 0.85 0.85 0.8
1

13. Comparaison des différents modèles: flux net de CO2
Wattenbach et al. (2010)
Very good performance of this simple modelling approach that does not requires data
on management (fertilisation, sowing date..) compared to other models.
Shows the power of remote sensing for constraining this crop model
CumulatednetCO2fluxes(gC.m-2
)
Accounting for
regrowth, weeds,
cover crops from
remote sensing
Likely improvement by accounting for the dynamic mapping of some of the
management practices (H2020 Sensagri & Bag’ages) : soil work, cover crops,
irrigation...
ObservationsObservations
Simulations
vs.
+ 5%
Fluxes
CO2
Cimp
- 0.854%
+ 7.8%
No accounting for regrowth (or
cover crops)
No accounting for
regrowth (or cover
crops)
Fluxes
CO2
Cimp
+ 7.8%
- 24%
- 84.1%
C
budget
Charvest
C
budget
Charvest
Performances/originality of our approach
ComponentsoftheCBudget(gC.m-2
.an-1
)
3rd
ICOS Science Conference, Prague, September 2018

14. • According to literature cover crop (CC) allow C storage of 313 ± 313
kg C.ha-1
.yr-1
considering trial > 5 years i.e. in optimum conditions
(Justes et al. 2013 ; Poeplau & Don, 2015) but…
• In real conditions CC development is very
heterogeneous (in time and space) and therefore
the C storage effect too  can be quantified by our approach
• And the C storage effect of CC will be compared to the albedo
effect (Carrer et al, 2018 in ERL) : essential for optimal mitigation
strategy
Effect of cover crops
Radiative forcing (W.m-2
) Mitigation effect (kt CO2-eq/yr)
Albedo effect of CC is
approx. 40% of the C
storage effect on the short
term but represents 170%
of it on a 100 years basis
(Tribouillois et al. 2018 ;
Carrer et al. 2018)
Albedo increase with CC

15. Limits of the approach
• Diagnostic approach : but some scenarii can be tested,
• Optical RS data must be combined with radar data
(Sentinel 1) in cloudy aeras & for strong crop
development (optical RS saturates for high LAI values);
ongoing research  H2020 Sensagri,
• Not suited for areas with animal farming : impossible to
quantify organic fertilisation from RS and very difficult
to locate fields where straw is exported  main causes
of uncertainties on the C budget,
Still this approach could be combined with a smart use of
inventory data enlightened by remote sensing data (or
the next generation of LPIS).
3rd
ICOS Science Conference, Prague, September 2018

16. Conclusions
• This approaches was developed in the perspective of
generalising it by using Sentinel data/products and global soil
maps (for model input) and the JECAM & ICOS networks (for
validation); with some limits… and some challenges (huge
amount of RS data… DIAS),
• Well suited for assessing the effects of straw management and
cover crops on cropland C budgets (and other benefits for
climate mitigation like the albedo effects; Carrer et al. 2018),
• The transposability of this modelling approach as been verified
(Morocco, Mexico, India…) for the SAFY-WB version, next is to
test SAFY-CO2 at other ICOS crop sites,
• Research tool that needs improvements before it can be used
in operational mode for mapping ecosystem services.
3rd
ICOS Science Conference, Prague, September 2018

17. Thanks for your attention and
thanks to our financers
If you want to have more details concerning our work please contact me at :
eric.ceschia@cesbio.cnes.fr
For complete description of the model see : http://www.cesbio.ups-tlse.fr/data_all/theses/Th_Veloso_2014.pdf
3rd
ICOS Science Conference, Prague, September 2018

18. Example of C and GHG Budgets
Auradé 2006 Lamasquère 2006
Blé
FluxdeCarboneetGES[geqCm-2
y-1
]
-500
0
500
1000
Maïs
-500
0
500
1000
NEP FO Exp NBP GES
Machines
Intrants
N2
O
Irrigation
NEP FO Exp NBP GES
Machines
Intrants
N2
O
Irrigation
(Béziat et al. 2009, Ceschia et al. 2010 AGEE)
Winter wheat
(Auradé 2006)
Maize for silage
(Lamasquère 2006)
??
EFO EFO
CarbonandGHGfluxes[gC-eqm-2
y-1
]
+ 35 gC.m-2
.yr-1
with cover crops
3rd
ICOS Science Conference, Prague, September 2018

19. Contexte
Growth
Senescence
D0
BiomassGAI
FT(Ta)
End of
growing
Time (Day of Year)
Senescence
starts
The SAFYE-CO2 model
Stt
Rs
Pla / Plb
19
),()(1 21 HHFwTaFfELUEeRgGPP T
GAIk
c
ext ×××−××=∆ ×−
ε
( ) elueeafELUE
bRR gdf
××=
×/
Correction of the elue
according to diffuse/total
radiation
(x GAI/GAImax)
On each pixel/plot
Emergence
Only during the senescence phase
ELUE(GPP/Rg)
Rgdiffuse/Rgtotal)
FAPAR
Optimised by comparison with
RS derived LAI
The SAFYE-CO2 model

20. growing
Plateau
Sénescence
D0
Biomass
Soil module = 3 layers
GreenAI
Plant module = 3 phases
Evaporation
= f(ET0)
Transpiration = f(GAI,ET0)
Diffusivité
Drainage
Rain +
irrigation
End of foliage
development
Time (Day of Year)
H1
H2
H3
Beginning of
sénescence
Simplified Monteith
equation
GAI
Coupling the vegetation module with the soil
water module
),()(1 21 HHFwTaFfELUEeRgGPP T
GAIk
c
ext ×××−××=∆ ×−
ε
Couplage avec un module hydrique simple de type FAO56 (et activation possible d’un module
d’irrigation)
3rd
ICOS Science Conference, Prague, September 2018

21. Details concerning SAFYE-CO2
GPP
NPP Biomasse /
Rendement
RtSv
NPP
a
NPP
+
=
1
RaGPPNPPd −=
RhNPPNEE −=
STb
eaRh
×
×=
( ) ( )RmGPPGYgR
aT
QRRm
RmNPPdmR
gRmRRa
−×−=
−
×=
×−=
+=






1
10
10
1010
1
Empirical relationship
(PhD E. Delogu)
NEE
Baret et al. (1992)
Ra
Rh
See former slide
Yield = f(ΣTa) cf STICS
or a simple harvest index
3 PhDs P. Béziat, A. Veloso & E. Delogu
GAI
• Partition to leaf
• SLA
3rd
ICOS Science Conference, Prague, September 2018

22. Validating biomass & yield simulations
Measured biomass (Kg m-2
)
Simulatedbiomass(Kgm-2
)
SAFY SAFY-CO2 SAFYE-CO2
RMSE (g.m-2
) 180.3 171.0 186.8
RRMSE (%) 25.0 23.7 25.9
Bias (g.m-2
) -20.1 -22.6 -30.7
R² 0.88 0.91 0.89
Comparison with destructive biomass samplings
Claverie et al (2012) in RSE Veloso (2014)
Summer crops (ESU 2006, 2008) Winter wheat
Underestimation for the highest values (saturation effect of
the RS optical derived LAI) : could be improved by
assimilating biomass data derived from radar satellites (ex.
Sentinel 1)

23. Contexte
0 20 40 60 80 100
0
20
40
60
80
100
Field campaign
Measured Yield (q/ha)
SimulatedYield(q/ha)
RMSE (q.ha-1
) = 12.22
RRMSE (%) = 25.73
Bias = 4.031
R2
= 0.83
0 20 40 60 80 100
0
20
40
60
80
100
RMSE (q.ha-1
) = 19.49
RRMSE (%) = 36.96
Bias = 6.715
R2
= 0.04
Farmers surveys
Measured Yield(q/ha)
EstimatedYield(q/ha)
0 20 40 60 80 100
0
20
40
60
80
100
RMSE (g.m-2
) = 26.81
RRMSE (%) = 39.64
Bias = -24.98
R2
= 0.95
Yield monitors
Measured Yield(q/ha)
EstimatedYield(q/ha)
y(x) = a x
a = 1.11
y(x) = a x
a = 1.12
y(x) = a x
a = 0.64
Validating yield estimates
Yield monitor on combined
harvester may have to be
calibrated against mean plot
values (weigthed) and data
require post processing 
very interesting for assessing
plot level spatial variablility
Yield
2011 campaign
(Veloso, 2014)
3rd
ICOS Science Conference, Prague, September 2018

24. 24
SAFYE-
CO2
Evapotranspiration (ETR)
Années de
calibration
(données
sites flux)
Years of
validation
(flux data
at both
sites)
ETR
Performances of the model : ETR

25. 25
SAFYE-
CO2
cumulated ETR
calibration
validation
validation
validation
calibration
Performances of the model : ETR
3rd
ICOS Science Conference, Prague, September 2018

26. 26
Auradé 2006
Regrowth
Performances of the model : GPP
Hydric stress

27. 27
Lamasquère 2007
Low
temperatures
Under-
estimation
Performances of the model : GPP
3rd
ICOS Science Conference, Prague, September 2018

28. Lamasquère 2007Lamasquère 2009
RhRaR +=eco
Under-estimation
28
Performances of the model : Reco
3rd
ICOS Science Conference, Prague, September 2018

29. 29
NEE= GPP-Reco
GPPRecoNEE
Lamasquère 2007
Performances of the model : NEE

30. Biomass
Residues
Soil orga. C
Photosynthesis
(GPP)
Ra Rh
Eco. resp. (RE)
Net CO2 flux (NEE)
Orga. fertil +
seeds
C budgetC budget GHG budget (LCA)GHG budget (LCA)
Fieldoperations
N2O(IPCC/measures)
Albedo effectAlbedo effect
CH4
DOC ?
See Ceschia et al. 2010 inAGEE Adapted from Muñoz et al. 2010
Other GHG
emissions
Time or space
Bare
soil
Growing period
t0 t1 t2
αs LO1
αs LO2
Surfacealbedo(αs)
RFLMC 1
< 0
C sink : C-CO2 eq1 < 0
RFLMC 2
> 0
C source : C-CO2 eq2 > 0
harvestseeding
1 yr
Bare
soil
Betts (2000), Bird et al. (2008)
Radiative forcing (in C-CO2 eq) = (C budget + N2O + Field operations ) + albedo effect
LMC1 LMC2
Harvest
System of study : the agricultural plot & interfaces
seeBéziat et al. 2009 inAFM

31. For France
MitigationpotentialMtCO2-eq.yr-1
Years0 25 50 75 100
~ 40% of the C
storage effect
C storage effect
N2O
Albedo effect
Tribouillois et al.
(2018)
~ 80%
of the C storage
effect
~ 170%
of the C storage
effect
Tribouillois et al.
(2018)
Carrer et al.(2018)
3
0
1.2

32. RFCC
Détermination des
zone et périodes
d’interculture
Données de
rayonnement &
transmittance
atmo.
(ERA-INTERIM)
Land use and desagregated vegetation index,
bare soil albedo & vegetation albedo derived
desagregated MODIS data at 5*5 km (Kalman
filter ;Carrer et al., 2012)
Radiative
Forcing of
Cover Crop
Ta*SWi
n
Albedo mitigation effect of cover crops estimated from RS
Carrer et al. (2018)
Gain d’albédo avec CIMS
Forçage radiatif (W.m-2
)
32

33. (Carrer et al. 2018)
• Les pays qui ont le plus fort potentiel d’effet albédo lié à l’introduction des CIMS sont la
France, la Roumanie, la Bulgarie et l’Allemagne,
• En moy. pour les 28 pays de l’UE, l’effet albédo des CIMS représente 10% de l’effet
stockage de C mais sans présenter d’effet saturation à long terme (au contraire !!)
Effet albédo des CIMS par pays en équivalent CO2
33