CIP-EIP - CEA

piloturuguayanAI and Robotics

Oct 15, 2013 (4 years and 23 days ago)

237 views




Deliverable D610.1

Ref

CARBONES
-
D610.1
-
REP
-
LSCE
-
023
-
01
-
00

Date

05/06/2012

Page

1
/
36



1



Information content on the carbon cycle

brought by

the

CARBONES project



Table of content

1

Introduction

________________________________
____________________________

2

2

Information about fossil fuel emissions

________________________________
______

3

2.1

Compared products

________________________________
________________________

3

2.2

Horizontal & vertical spatial distributions of CARBONES IER data

_______________

4

2.3

Temporal distributions

________________________________
_____________________

5

3

Information on ocean fluxes from OCVR system

_______________________________

7

3.1

Information about pCO2 spatial and temporal distributions

______________________

7

3.2

Evaluation of the
OCVR ocean flux

________________________________
___________

9

4

Evaluation of the net surface fluxes from the CCDAS

__________________________

13

4.1

Approach: product used for the evaluation of CARBONES

______________________

13

4.2

Global annual totals

________________________________
_______________________

14

4.3

Long term means

________________________________
_________________________

15

4.4

Inter
-
annual variability

________________________________
____________________

18

4.5

Seasonal flux variations

________________________________
____________________

22

5

Evaluation of land gross carbon fluxes

________________________________
______

23

5.1

Evaluation at the site level

________________________________
__________________

24

5.2

Evaluation at global scal
e from MTE estimates

________________________________

25

6

Evaluation of land carbon stocks
________________________________
___________

29

6.1

Product used for the evaluation

________________________________
_____________

29

6.2

Results of the comparison

________________________________
__________________

29

7

Conclusions and perspectives
________________________________
______________

34

8

References

________________________________
_____________________________

35






Deliverable D610.1

Ref

CARBONES
-
D610.1
-
REP
-
LSCE
-
023
-
01
-
00

Date

05/06/2012

Page

2
/
36



2


1

Introduction

The objectives of this

deliverable are to analys
e the information brought by
CARBONES

on
the
carbon cycle
in terms of
carbon fluxes and carbon stocks at several spatial and

temporal
scales. Th
ese concern the

study

of
:



The global net annual carbon balance by apportioning between key regions of the
globe like North America, Europe, North Eurasia, the Tropics and
key
ocean
basins



The inter
-
annual variability of the sub
-
continen
tal regions and ocean basins



The trend in the net carbon uptake of the lands in comparison with that of the oceans



The spatial distribution of the forest carbon stocks compared with “observations”
mapped globally (produced by CARBONES)

The analys
es
were

pl
anned
initially
to take into account the uncertainties
of

th
e inverted
fluxes from CARBONES
-
CCDAS
(error propagation)
in order to discriminate between robust
signals and other ones.


However, t
his repo
rt will not cover all the above
-
mentioned objectives
,
given some delays in
producing the final 20 years carbon flux and stock re
-
analysis
. Moreover, we stress
on

the

fact
that the report
mainly
evaluate
s the last

version of the CARBONES product
(Version V2
.0

but also in some cases version V1.0
)
against other

independent flux/stock products
.
It is
indeed very difficult to present the information content of a given carbon product
that
combine observation and models
(CARBONES) as “true” information without comparing the
estimated quantities w
ith other independent

estimates. It is indeed impossible to establish the
true fluxes
at regional to global scales
, given that there is no direct measurement of such
quantity
.
Note also that we consider fossil fuel emissions (WP300) and ocean flux
estimates
from ocean pCO2 dat
a (further used as prior in the Carbon Cycle Data Assimilation System)
as CARBONES products. Given that

s
ignificant progress has been made for
the
se two fluxes
,
we discuss the

information
content o
f these products with respect to
the

carbon cycle
(
especially
the

temporal variations)
.


When making the evaluations of the
above
-
mentioned CARBONES

products, the
uncertainties in both C
A
RBONES and other

products

are
not
considered

yet
, because of lack
of time. However, a specific report will analyse
independently the estimated errors on the
carbon fluxes derived from the CCDAS
.
Finally, we stress
again that

th
is

report:




shows

the potential of th
e

evaluation exercise conducted within CARBONES



should not be considered as the most advanced answer to the

above scientific
questions but rather a new “contribution” to the overall knowledge
.

The outline of th
e

report is as follows:

We describe in section 2) the information
content on carbon cycle
about fossil fuel emissions.
The
n, the

ocean products are evaluated in section 3.
The
net surface fluxes of CARBONES



Deliverable D610.1

Ref

CARBONES
-
D610.1
-
REP
-
LSCE
-
023
-
01
-
00

Date

05/06/2012

Page

3
/
36



3


are confronted to those derived from direct inversion systems
and from global dynamic
ecosystem models (DGVMs)
in section 4.
In section 5, t
he CARBONES land carbon gross
fluxes
are compared to
estimates
from machine learning algorithm

that uses observations from
water, energy, and carbon fluxes
.

The CARBONES land carbon stocks are evaluated in
section 6. Finally, conclusions and perspectives are presented in section
7
.



2

Information about fossil fuel emissions

2.1

Compared products

CARBONES CCDAS considers new global spatial and temporal resolved CO2 emissions
based on EDGAR v4.2 and time profiles developed by USTUTT (see
CARBONES
deliverable D300 for details). The spatial res
olution of
the CARBONES fossil fuel emissions
product
(called hereafter IER data)
is
at
1°x 1°

with an hourly temporal resolution. Th
is
product is
derive
d from
annual EDGAR
v4.2

CO
2

emissions
by
using country, sector, year,
month, day and time zone specific monthly, weekly and daily time profiles.
The temporal
variations of
IER products are compared to the existing other
ones
, which are defined as
f
ol
lows:



The emissions of dioxide from fossil
-
fue
l combustion and cement production reported
in Andres et al. (2012). The spatial resolution of the product is
1°x 1° with a monthly
temporal resolution. T
his product is called hereafter
Andres
.



The emissions of fossil fuel CO
2

emission inventory at global scale
from a
combination of
a worldwide point source database and satellite observations of the
global nightlight distribution (Oda and
Maksyutov, 2011)
.
The product used for this
exercise
is 1°x 1° with a

monthly

temporal res
olution

and for
only
the year 200
8
.

The
product is called ODA.




The emissions from the University of Beijing (hereafter
PKU
)

at global scale are also
considered
. PKU uses

a 16 sub
-
national disaggregation method (SDM) applied to
establish a global 0.1°×0.1°

geo
-
referenced inventory of fuel combustion (PKU
-
FUEL) and a corresponding CO2 emission inventory (PKU
-
CO2) based

on

18 upon
64 fuel sub
-
types for the year 2007 (
Rong et al., 2012)
.



The emissions used in the CarbonTracker

model
-
data fusion

system
:

(see
http://www.esrl.noaa.gov/gmd/ccgg/carbontracker/documentation_ff.html#ct_doc
)
.
The product is based on
CDIAC country total using

EDGv4
.0 spatial distribution
and
standard temporal

profiles
.






Deliverable D610.1

Ref

CARBONES
-
D610.1
-
REP
-
LSCE
-
023
-
01
-
00

Date

05/06/2012

Page

4
/
36



4


In what follows, we first describe the n
ovel

IER products
derived

for three altitudes,
and
then
monthly variations of IER data are confronted to Andres, ODA
, and PKU fossil fuel emission
products

for few regions
.

2.2

Horizont
al & vertical spatial distribution
s of CARBONES IER data


Besides spatial and temporal resolved emissions, it is also important to consider the effective
emission height
,

which significantly influences model
l
ed concentration values (Pregger &

Friedrich 2008).
In particular it is crucial to separate:

1.

Emission at the surface mainly from transport and residential sectors. These emission
will be emitted in the lowest level of the transport model

2.

Emission from power plant that are still injected in

the Planetary Boundary Layer
(PBL) but directly mixed within the PBL because the injection is made though “high
chimney”. This correspond
s

to part of the industrial sector.

3.

Emission from aviation above the PBL in mid troposphere.

Most
existing global sp
atially and temporally resolved fossil fuel emission models do not
consider effective emission heights at all or not sufficient
ly

enough. As a consequence,
the
application of effective emission heights is a major improvement of the fossil fuel
emission mod
el
l
ing on the global scale.

The report describing the CCDAS evolution (D420)
displays

the considered effective emission heights based on air quality model from EMEP and
specific
assumptions.

JRC, responsible for the EDGAR emissions, delivered within
the framework of CARBONES
also the emissions from aviation distinct into “Take
-
Off and Landing”, “Climb and Descent”
and “Cruise”. Therefore it was also possible to apply emission heights to the subsecto
rs of
aviation,
which is also an innovation for the f
ossil fuel emission mode
l
ling.

This last
distinction will be crucial for the Carbon Data Assimilation System with respect to the
atmospheric CO2 data.


The results based on the current time profiles for the three emission heights cla
sses are shown
in the

Figure 2.1
.

According to the actual results the highest absolute values occur at the
middle altitude. This is the case in particular because the emissions from the energy industry
as well as the emissions from the industrial combustion are effectively emi
tted between 92m


781m. Besides, the hourly shift between the different regions in the current version of the
fossil fuel emissions

is studied and corrected (not shown).


A direct evaluation of this vertical splitting of fossil fuel emissions in terms of

impact at
atmospheric CO2
stations

is only undergoing
. Especially we will evaluate the “improvement”
of the simulated concentrations.





Deliverable D610.1

Ref

CARBONES
-
D610.1
-
REP
-
LSCE
-
023
-
01
-
00

Date

05/06/2012

Page

5
/
36



5



Figure 2.
1
: Spatial distributions of
fossil fuel hourly emissions) for the 16th of January 2008
at 14:00 hours in a 1°x1° resolution in [Mg/h] and at
3 altitude levels: < 92 m (
right
), between
92 m and781 m (middle), and > 781 m (
left
)
.


2.3

Temporal distribution
s


Figure
s

2.2

and 2.3 display

th
e
temporal variations of IER fossil fluxes
together with the

estimates from Andres, ODA, PKU
, and Ctracker

at global
scale
and for
two continental
regions
.
This analysis of the temporal variations indicates
:



At global scale, IER data compared well with And
res data, with
however
a
larger

seasonal
amplitude obtained
in the
IER
-
CARBONES
product than ODA and Andres
but
similar to that of Ctracker.

ODA
product

for 2008
is

in goo
d agreement with
Andres but the phase of the seasonal cycle for both product

is slightly different than
IER
-
CARBONES. The
PKU yearly data
(evenly distributed over the year) for

2007 is
smaller

than the other

flux estimates

given that it does not include
“bunker” fuel and
air planes
.
Note that Andres total flux is
also
slightly sma
ller than IER total, given that
it does not include “bunker” fuel.

The main differences after correction for differences
in fuel categories included in each product concern the temporal variations. Given the
significant effort brought by IER to construct v
arying temporal profiles, we believe
that the CARBONES product brings new information to the carbon cycle through the
temporal disaggregation of EDGARv4.2 spatial product.



Over Europe, IER data give
s

the strongest seasonal amplitude
, similar to Ctracker
pr
oduct and significantly larger that both Andres and ODA product
.
In this case, given
the detailed analysis and the large collection of temporal profile data made by IER, we
can be confident that such large seasonality is probably more realistic and that th
e
CARBONES product brings new information. Note that

Ctracker provide a slightly
different phase for the seasonal cycle than IER
.




Deliverable D610.1

Ref

CARBONES
-
D610.1
-
REP
-
LSCE
-
023
-
01
-
00

Date

05/06/2012

Page

6
/
36



6




Over Eurasia, IER
presents a larger
amplitude

than ODA and Andres but smaller than
Ctracker

that uses similar temporal profile as for Europe. Whether one product is more
realistic that the other one still need to be

evaluated against local proxy data
.

Overall the IER
-
CARBONES product provides new information with seasonal amplitude
around 30% t
o 50% depending on the considered regions and that these variations will be
crucial for the assimilation of atmospheric observations
.






Figure
2.2
:
Temporal variations of the fossil emissions from ODA, Andres, IER

(CARBONES)
, and
PKU products for
the
globe.











Deliverable D610.1

Ref

CARBONES
-
D610.1
-
REP
-
LSCE
-
023
-
01
-
00

Date

05/06/2012

Page

7
/
36



7



Figure 2.
4
: Temporal variations of the
fossil emissions from
ODA, Andres, IER, and PKU products
for
Europe (left), and Eurasia (right)
.




3

Information
on

ocean
fluxes

from OCVR system



The Ocean Carbon Variation
al Reanalyzer (OCVR) is used to produce a twenty years
global ocean

carbon
flux

reanalys
es. OCVR

is a neural network framework developed by
CLIMMOD. As input variables, it uses observations from satellites and/or model outputs (as
for instance sea surface temperature, mixed layer depth, wind speed, etc), which control at the
first
-
or
der the surface ocean pCO2. Furthermore, a variational data assimilation scheme
incorporates efficiently
recent

sets of
raw
pCO2 observations
to adjust for the
trend and to
take into account extreme events like
El Niño
. The spatial and temporal resolutions

of the
system are adjustable. The system then uses supplied atmospheric CO2 (Globalview product)
concentration to calculate air
-
sea flux according to a selectable exchange parameterization
(e.g. Liss et Merlivat 1986, Wanninkhof 1992, Takahashi 2009 etc).

The system is
described
in more details in previous report
s

(D410

and D420
)
.

The fluxes that are produced by OCVR
are further used in the CCDAS as prior ocean fluxes.


3.1

Information about pCO2 spatial
and

temporal distribution
s


Figure 3.1.1 displays the sp
atial variations of pCO2 for three years for
the month of
January
,
from the OCVR system
. Results show the
spatial
variations of pCO2
for
the selected months
-
years.

The classical spatial pattern is obtained with
low p
CO2

at high latitudes and high
values

in the tropics. However, the new information brought by CARBONES concern the



Deliverable D610.1

Ref

CARBONES
-
D610.1
-
REP
-
LSCE
-
023
-
01
-
00

Date

05/06/2012

Page

8
/
36



8


inter
-
annual variability (IAV). Indeed most optimization system
s

are using
so far a

climatology field with no year
-
to
-
year flux variations

(such at Takahashi
et al. products)
. In

the
example below, we see that 2009 has a larger high pCO2 over the tropic
al Pacific than the
other years
.



Figure 3.1.1: Global pCO2sw
(in micro
-
atmosphere)
maps from OCVR are shown

and for simulations relevant for

January
1990, 2000
,

and 2009
.







Deliverable D610.1

Ref

CARBONES
-
D610.1
-
REP
-
LSCE
-
023
-
01
-
00

Date

05/06/2012

Page

9
/
36



9


As already discussed in the D420 reports, t
he temporal variations of the OCVR pCO2 data
over the 1989
-
2009 periods
at

selected ocean locations
outperform the

Takahashi standard
climatology
estimates
(i
.e. using a fixed growth rate of pCO2).
O
ver the Equato
rial Pacific,
we clearly see an improvement in the OCVR product compared to the “scaled climatology”
wit
h

a drawdown of pCO2 during the 1998
El Niño

period and with a seasonal cycle
much
less pronounced than Takahashi and more in line with the observations
.



3.2

Evaluation of the
OCVR

ocean flux



We now try to evaluate the estimated air
-
sea fluxes from OCVR.
We

present here a
first
comparison of

OCVR product with
few other

“independent” ones. These are:

1.

The result from the Takahashi (2009) flux climatology

2.

The results from an ensemble of “Ocean Interior Inversion” from Gruber et al. 2009.
These estimates combine information on DIC measurement in the ocean and ocean
circulation model (ocean inversion)

3.

Th
e results from Steinkamp (2012) that

were produced at ET
H, one partner of
CARBONES. These fluxes correspond to the estimates from an atmospheric inversion
combining the “ocean interior inversion” product and the information c
ontent of
atmospheric CO2 data using a
classical
atmospheric
inversion
with
the transpo
rt
models
from

the TRANSCOM int
er
-
comparison exercise.

4.

The results from five Ocean General Circulation Models (OGCM) that were used and
compared within the recent RECCAP synthesis, as part of the Global Carbon Project.
These models comprise
different ocean physical models and biogeochemical models
(see table
3.1
below).



Table
3.1
: List of OGCM used to compare with the CARBONES ocean product.

Model

Ocean model

BGC model

Forcing

Bergen

MICOM (isopycnic)

HAMOCC

NCEP

CSIRO

OGCM

P
-
based

NCEP

LSCE

NEMO

PISCES

NCEP




Deliverable D610.1

Ref

CARBONES
-
D610.1
-
REP
-
LSCE
-
023
-
01
-
00

Date

05/06/2012

Page

10
/
36



10


UEA

NEMO

PlankTOM5

NCEP

WHOI

CCSM

BEC

NCEP



The figure
3.2

compares the mean seasonal cycle

over the 1990
-
1999 period from the
estimates
1, 2, and 3

(OCVR results are in red) for and ensemble of 11 ocean basin
(TRANSCOM regions). Major features from this analysis are:



As expected, for most basins the seasonal cycle of OCVR fluxes is close to that of the
Takahashi climatology. The southern ocean and th
e north Pacific present the largest
deviations from the climatology. Further analysis need to be done to evaluate the level
of improvement brought by OCVR
.



As for the mean fluxes, the OCVR results are generally in good agreement with the
results from the o
cean interior inversion
, with a mean global sink slightly larger than
Takahashi 2009.



The large seasonal variation
s

obtain
ed

in the Steinkamp 2012 product, from the
atmospheric inversion
, are

partly incompatible with our OCVR product, especially in
the mid

to high latitude basins.
Such results, would indicate that the atmospheric CO2
data tend to impact
the

seasonal ocean fluxes in a way that may be incompatible with
the raw ocean pCO2 surface data
. These

results will be investigated in
a paper under
prepar
ation
.






Deliverable D610.1

Ref

CARBONES
-
D610.1
-
REP
-
LSCE
-
023
-
01
-
00

Date

05/06/2012

Page

11
/
36



11



Figure

3.2:

Mean seasonal air
-
sea flux estimates from OCVAR for the period 1990
-
1999, for 11 regions, compared with
independent estimates (see legend in the figure).


We now focus on the year
-
to
-
year air
-
sea flux variability.
Figure 3.3 compares the annual air
-
sea flux
for several “latitudinal scale” regions and
one specific basin

obtained by the OCVR system and the
five OGCM models presented above. The major features of OCVR are:



A smaller global uptake during the period 1998
to 2002
compared to the early 90s and late
2000s,
not present in most OGCM (except one)
.



A pronounced increased of the global ocean uptake after 2002
up to 2009,
that is not
present in the OGCM models.



The increased uptake after 2002 is mostly explained by

the northern ocean (north of 30°N)
and to a lower extend by the southern ocean (south of 30°S).



The tropical ocean present nearly no trend during the 20 years periods, with a mean release
of carbon around 0.6 PgC/year
.



The tropical ocean shows a decrease
of the carbon source in 1998 linked to the El
-
nino
conditions, a feature captured by some OGCM but not all. Such tropical variation in 1998 is



Deliverable D610.1

Ref

CARBONES
-
D610.1
-
REP
-
LSCE
-
023
-
01
-
00

Date

05/06/2012

Page

12
/
36



12


dominated by the contribution of the tropical
Pacific Ocean, mostly affected by El
-
Nino
conditions
.

Overall the m
ain feature brought by the assimilation of surface pCO2 observations into a statistical
ocean model (OCVR) concerns the trend in the ocean uptake after 2002. The mean carbon sink for
the different ocean basin
s

across the 20 years remains similar to what ha
s been estimated by most
approaches. Whether the increased global trend from 1.4 PgC/year in 2002 to 2.6 PgC/year in 2009 is
real or an artefact/bias of the OCVR model is the main
question. Possible sources of biases are: i) the
non uniform coverage of the

raw pCO2 data over time which could thus bias the neural network
performance towards fitting preferentially a given period and thus introducing a temporal bias; ii)
the
non uniform coverage of the raw data over

space with more observations over the coasta
l area at
the end of the period. These crucial points are under investigations and will be discussed in a paper
presenting OCVR system (Kane et al.).


Figure
3.3: Comparison of the air
-
sea fluxes for different ocean regions estimated by CARBONES OCVR re
-
a
nalyses (prior
flux used in the CCDAS) with the fluxes from
5 Ocean General Circulation Models used in the RECCAP synthesis.




Deliverable D610.1

Ref

CARBONES
-
D610.1
-
REP
-
LSCE
-
023
-
01
-
00

Date

05/06/2012

Page

13
/
36



13


4

Evaluation of
the
optimized

surface fluxes from the
CCDA
S


4.1

Approach: product used for the evaluation of CARBONES

The CARBONES products are compared to
two different types of approaches:



T
he

results from
classical recent atmospheric
inversions
following the synthesis
performed
for

the RECCAP exercise (Peylin et al.,
submitted to Biogeocsciences
).
We
use t
welve

participating
standard atmospheric

inversion systems and associated key
attributes are listed in Table 1.
F
u
rther details on these systems can be found in Peylin
(2012)
with a general description given on

the TransCom website
(
http://transcom.lsce.ipsl.fr
).



The results
of 8 Dynamic global vegetation models (DGVMs) also compared within
the RECCAP exercise (sitch et al., submitted to Biogeosciences). These data are not
published yet and the figures including their

results should not be distributed yet. We
used the results of DGVMs to evaluate the inter
-
annual variations of CARBONES flux
product only. We considered that these model fluxes do not provide better information
than atmospheric inversions for the mean car
bon uptake.



In what follows, we discuss CARBONES land and ocean fluxes in terms of annual totals, long
term means, and inter
-
annual and seasonal variations with regard to those derived from
the

ensemble of
standard
inversions

(and from DGVM for the IAV)
.

The results discussed here
correspond to version V1 of CARBONES and the ocean fluxes are thus those from
OCVR further corrected after the assimilation of atmospheric data.

G
lobal patterns of the
results derived from the ensemble of the
standard
inversions

are first given and those from
CARBONES are

then

confronted.
We

will only give the main characteristics of this ensemble
of results, with
an
emphasis
on the
CARBONES results
when they
significant
ly

differ from
the
envelope of the direct inversion inferenc
es.

More detailed discussions
on the following
rich information on carbon budget
from the standard atmospheric inversions
are
given in
Peylin et al. (2012).
Indeed, the differences due to the different set ups (e.g., fossil emissions,
meteorological forcin
g used, biomass burning,
the inversion systems themselves,
etc…) of the
direct systems are not discussed here.

All different “inversion estimates” have used different fossil fuel emissions so that a direct
comparison of the natural fluxes, which should be

considered as a residual flux, are biased.
The systems that have used more fossil fuel emissions should have a larger land/ocean carbon
sink to match the atmospheric growth rate. In order to cope with that problem we applied a
correction to all products:
we took the total estimated flux (natural + fossil) and then subtract
a common fossil fuel emission (EDGAR v4.2). In the remaining we thus compare the so
-



Deliverable D610.1

Ref

CARBONES
-
D610.1
-
REP
-
LSCE
-
023
-
01
-
00

Date

05/06/2012

Page

14
/
36



14


called “fossil corrected fluxes”.

Note that the EDGARv4.2 correspond to the emissions used
by IER to d
erived the CARBONES fossil fuel product.


Table 4.1
:

I
nversion systems
used to compare with CARBONES product with their

key attributes.
“MM” denotes
monthly mean.

IAV indicates inter
-
annual variations used for meteorological forcing.


4.2

Global annual totals

Figure

4.1

display
s
for each inversion the posterior estimate of the natural global total fluxes
(land plus ocean)
,

and the global fossil fuel fluxes.
The
year
-
to
-
year

variations of the global
total flux (land pl
us ocean)
reflect the variations in global atmospheric CO
2

growth rate. As
expected, they are robust across the different inversions, with large fluctuations assoc
iated
with the occurrence of El
Niño and La Ni
ñ
a conditions. For instance, in 1998, and to a lesser
extent in 1994, the strong
El Niño
condition led to a small carbon uptake by the land and
ocean ecosystems
.
Significant differences in prescribed fossil fuel emissions are noteworthy.
The JENA
fossil flu
xes are larger than other inversions (~0.45 PgC/yr), and consequently that
inversion requires more uptake to match the atmospheric CO
2

growth

(note that this is taken
care

with

the “fossil correction” we described above)
.


CARBONES product compare
s

well with these
independent
estimates

and provides a similar
global picture
.
First, we should notice that the assimilation of atmospheric data (step 4 of the
sequential approach) does not change too much the ocean fluxes compared to results of the
OCVR mo
del (used as prior in step4). Thus,

the

monotonous
increase of ocean uptake
obtained from CARBONES
after

2002

is a new feature, not present in all other inversions. As
noticed above such feature is crucial and need to be confirmed by independent proxy
.
The

land ocean partition of the global carbon
fluxes is directly affected by the results of the OCVR

system and we thus obtain a lower land carbon sink in the early 90s and late 2000s compared
to the other atmospheric inversion systems.



Inverse System

No

regions

Contact

Time
Period

Obs

# of observing
stations

IAV
trans
port

A

LSCE1

Lsce_an_v2.1

Grid
-
cell
(96x72)

Philippe Peylin

1996
-
2004

MM

76

Yes

B

LSCE2

Lsce_var_v1.0

Grid
-
cell
(96x72)

Frederic Chevallier

1988
-
2008

Raw

128

Yes

C

CCAM

C13_CCAM_LAW

146

Rachel Law

1992
-
2008

MM

73 CO
2
, 7 C13

No

D

MATCH

C13_MATCH_Rayner

116

Peter Rayner

1992
-
2008

MM

73 CO
2
, 7 C13

No

E

CTRUS

Carbontracker_US

156

Andy Jacobson

Wouter Peters

2000
-
2008

Raw


Yes

F

CTREU

Carbontracker_EU

156

Wouter Peters

2000
-
2008

Raw

117

Yes

G


Jena_s96_v3.3

Grid
-
cell

(72x48)

Christian Roedenbeck

1996
-
2008

Raw

53

Yes

H

RIGC

Rigc_Patra

64

Prabir Patra

1993
-
2007

MM

74

Yes

I

JMA

JMA_2010

22

Kazutaka Yamada

1985
-
2008

MM


Yes

J

TRC

TRCOM_mean

22

Kevin Gurney

1995
-
2008

MM


No

K

NICAM

Nicam_Niwa

40

Yosuke Niwa

1988
-
2007

MM


Yes




Deliverable D610.1

Ref

CARBONES
-
D610.1
-
REP
-
LSCE
-
023
-
01
-
00

Date

05/06/2012

Page

15
/
36



15




Figure
4.1
:

Annual
mean posterior flux estimate of the individual inversion
s and CARBONES (in
black)
. Shown here are a) natural “fossil corrected” global total carbon exchange, b) fossil fuel
emission, c) natural “fossil corrected” total land, and d) natural total ocean
fluxes
.


4.3

Long term means


Since the
ensemble of the
standard
inversions ha
s

been run for different time periods
,
a
common time period
that allows
reduc
ing

the inter
-
comparison timespan when calculating
multi
-
year means
,

was selected
.
Thus,
the
2001
-
2006
period

was chosen
.

Figure 4.2

displays the total fluxes for the globe and three latitudinal bands as well as the
partition between the land and ocean
. From the perspective of the long
-
term mean, the land
(fossil corrected) and ocean have similar values for

total uptake, around
-
1.5 PgC/yr.




Deliverable D610.1

Ref

CARBONES
-
D610.1
-
REP
-
LSCE
-
023
-
01
-
00

Date

05/06/2012

Page

16
/
36



16



Figure 4.2
: Mean natural fluxes for the period 2001
-
2006 of the individual inversion posterior fluxes
(exception for LSCE_ana system which is averaged over 2001
-
2004)

and CARBONES (in black)
. Shown
here are total (first

column), natural “fossil corrected” land (second column)
and natural ocean (third
column) carbon exchange aggregated over the Globe, the Northern
hemisphere (rough
ly > 25N), the
tropic (roughly 25S
-
25N) and the southern hemisphere (roughly < 25S). Numbers

in parenthesis
represent the mean flux and the standard deviation across all inversions.





Deliverable D610.1

Ref

CARBONES
-
D610.1
-
REP
-
LSCE
-
023
-
01
-
00

Date

05/06/2012

Page

17
/
36



17


Except for JENA inversion, which has much larger uptake on land and smaller uptake
by the ocean (both compensating), and NICAM which gives the smallest land uptake (
~
-
0.5
PgC/yr) compensated by the largest ocean sink (~
-
2.5 PgC/yr)

CARBONES is

at the lower
end of the results range. In details, global total flux from CARBONES is slightly lower than
results obtained from most
of the direct inversions,
which

is explain
ed by the lower
carbon

uptake of the land either
at
North
ern

latitude or in the Tropics. This lower uptake from the
land is compensated by the relatively large uptake of
the
ocean at global scale. The North and
South oceans contribute to this large uptake,

while a lower uptake is found
over
the tropics.
The large
carbon
uptake of the ocean
,

derived from CARBONES
CCDAS,
can partly be
explained by the large uptake observed from 2002 as shown in Figure 4.3.1b.

Overall CARBONES confirms the results from the
inversions that were solving the fluxes at
high resolution (JENA, LSCE, CTRACKER systems) with a nearly neutral tropical land
budget and a very small southern land source.


We briefly

investigate the long term mean natural fluxes within
continental/basin
-
scale
subdivisions for a breakdown of the northern hemisphere into three selected continental/basin
-
scale regions: North America, Europe, North Asia (
Figure 4.3
). These three land regions show
a significant carbon sink, from nearly
-
0.5 P
gGtC/yr over Europe to

-
1.0 GtCPgC/yr over
North Asia. A large spread among the
results from the different
inversions
is obtained. For
the
three
selected
regions, the standard deviation reaches around 0.5 GtCPgC/yr.
CARBONES
results are one of

the lowest
carbon
sink
for these regions and especially for North Asia
.

However, if we compare only the 5 fives estimates from the left (JENA, LSCE
_var and the
two CTRACKER syste
ms)
, CARBONES results do not appear as outliers. These estimates
come from standard inver
sions that solve for fluxes either at the resolution of the transport
model or for a large number of regions, which avoid the so
-
called “aggregation error”
associated to the other estimates that solve for a restricted number of flux
-
regions. These
technica
l details are discussed in Peylin et al. 2012.


Overall, CARBONES with the use of FluxNet data and MODIS
-
NDVI thus decrease the
carbon uptake over North Asia compared to LSCE_var inversion that uses the same
atmospheric transport model. The main changes

of the sequential optimization (step 1 and 2)
,
related to
a decrease of the growing season length in ORCHIDEE (through MODIS
-
DVI data)
and the decrease of the amplitude of the seasonal cycle of NEE for north ecosystems (through
FluxNet data)
,

lead to a st
ronger reduction of the carbon sink in North Asia compared to
Europe and North America. These results are under
investigation and will be published with
the 20 year CARBONES flux reanalysis (Peylin et al., in preparation).






Deliverable D610.1

Ref

CARBONES
-
D610.1
-
REP
-
LSCE
-
023
-
01
-
00

Date

05/06/2012

Page

18
/
36



18




Figure 4.3:

As for Figure
4.2
, but for three continental/basin
-
scale regions: North America,
Europe, and North Asia.


4.4

Inter
-
annual variability



Figure
4.3

show
s

the inter
-
annual variability
of land
and ocean
fluxes
for the northern,
tropical and southern aggregate
s
. The results
represent annual means with the individual
model long
-
term means removed (a long
-
term mean defined over the entirety of the submitted
model timespan). We refer to these as inter
-
annual carbon exchange anomalies

(hereafter
IAV)
.




Deliverable D610.1

Ref

CARBONES
-
D610.1
-
REP
-
LSCE
-
023
-
01
-
00

Date

05/06/2012

Page

19
/
36



19




Figure 4.4
: Annual mean
smoothed average (smoothing window of 3 years) of the individual
participating inversion posterior flux estimates. Shown here are land fluxes for northern, tropics and
south regions a
s well as for the global total, for land and ocean.







Deliverable D610.1

Ref

CARBONES
-
D610.1
-
REP
-
LSCE
-
023
-
01
-
00

Date

05/06/2012

Page

20
/
36



20


All
the
standard
i
nversion systems

tend to exhibit greater IAV on land
versus ocean
(Figure 4.4
)
, particularly in the tropical latitude band. Within the land aggregates, the tropical
land exhibits the greatest amount of inter
-
annual

variability while in the ocean

similar
inter
-
annual variability is seen in the
different
ocean
basins (note that for the southern ocean the
scale is different than for the two other latitude bands).

Note that
in the southern ocean
a
large part of the IAV
is associated with the 1997/1998 time p
eriod
,

in which several model
inversions show large anomalies, though of differing sign.

Overall, CARB
ONES results are in agreement with the IAV produced by the ensemble
of
inversions. In details, CARBONES exhibit a large negative IAV in
the
South
ern land
. This
feature mainly arise
s

over South

America during the 1997/1998 period, which seems to be
compensated by the large positive IAV obtained in South Asia

(not shown)
. The large
biomass burn
ing fluxes observed during the 1997
-
1998 period and considered in

the
CAR
BONES optimization may explain

these results.
Further analysis will be conducted to
evaluate whether other observational evidences support the different flux IAV in
CARBONES for the southern land.

As a direct consequence, CARBONES produces a smalle
r
flux IAV over the tropical land than the other inversions. Such feature is a
direct
consequence
of the flux optimization with larger IAV in the southern land. The results from version V3
(under analysis) with a correction of the parameters of ORCHIDEE in

step 4 (assimilation of
atmospheric CO2 data) will confirm or not if such reduce
d

tropical land IAV can be supported
by the current ORCHIDEE processes.


A second aspect of the evaluation of CARBONES product was performed against the results
of DGVMs
that
comparable to the

one used in our CCDAS. In the set of DGVMs
a standard
version of ORCHIDEE was also used
following the

specific protocol applied to all DGVMs.
Figure 4.5 provides for “big latitudinal bands” and the northern hemisphere split into the three

continental
regions
the annual flux variations, after subtraction of the mean flux over the 20
years period

(i.e., the flux IAV). Major features are:



First the CARBONES product significantly differs from the ORCHIDEE DGVM
version, indicating that the MODIS
-
NDVI, the FluxNet data, and the atmospheric
data led to significant changes in the model IAV signal.



For the latitudinal breakdown, we obtain IAV sig
nals that are compatible with the
DGVM ones but with slightly lower IAV over the tropics, especially from 1999 to
2004. The tropical positive anomaly en 2005 is significantly larger than the DGVMs.
Over the northern land, CARBONES IAV is slightly smaller t
hat most DGVMs with
a significant increase of the land carbon sink in 2004, remaining the following years.



For the northern breakdown,
North America, Europe, and North Asia all show
relatively similar flux IAV. CARBONES provides similar results than the DG
VMs
for Europe and North America in terms of phase and amplitude, while over North



Deliverable D610.1

Ref

CARBONES
-
D610.1
-
REP
-
LSCE
-
023
-
01
-
00

Date

05/06/2012

Page

21
/
36



21


Asia the phase of the IAV varies between all estimates. More in depth analysis are
needed to attribute these IAV to underlying processes.

Overall, if CARBONES does not chan
ge radically our knowledge on the land carbon flux
IAV, it will provide new insight on the processes than underline these flux variations and
these processes are likely to differ from the standard DGVM, given that ORCHIDEE
-
CARBONES parameters have signific
antly changed from their original values. This analysis
is however only underway and will be finalized after the duration of the project.



Figure
4.5
:
Annual mean smoothed average

of the
CARBONES flux estimates (red) compared to the results from 8
land d
ynamic global vegetation models (DGVMs)
. Shown here are land fluxes for northern, tropics and south regions
as well as for the
North America, Europe, and North Asia.





Deliverable D610.1

Ref

CARBONES
-
D610.1
-
REP
-
LSCE
-
023
-
01
-
00

Date

05/06/2012

Page

22
/
36



22


4.5

Seasonal flux variations


Figure
4.6

show
s

the mean seasonal cycle on land and ocean for the latitudinal aggregate
regions
.

Note that the land and ocean panels use different numerical scales. For this diagnostic
we consider the raw natural fluxes and not the “fossil corrected” fluxes, to avoid any

spurious
monthly flux corrections. The global land seasonality is driven by the northern land with close
agreement regarding both the magnitude and phasing of the growing season and dormant
season fluxes.
Regarding ocean fluxes, the largest differences be
tween CARBONES and the
other systems are more pronounced over the South hemisphere. Only d
e
tails

relevant for land
fluxes
, which largely contribute to the total global fluxes
and for the latitudinal aggregate
regions are given hereafter
:



T
he amplitude of
the seasonal cycle
of land fluxes over the Northern hemisphere
is
close to 3 PgC/yr (range needed) and the peak of the growing season is located in July
for all
inversion systems
,

including CARBONES
.




Seasonality for the tropical land is quite low and
larger differences can occur across
the different

systems
,

including CARBONES
. We notice that CARBONES is closer to
the
LSCE inversions
, which reflect the fact that they use both the ORCHIDEE model,
and that the atmospheric constraint over the tropic is re
latively weak
.
CARBONES
show two peaks: one in March and the other in July, which are not shown by almost
all the other systems.

This feature is less prominent in the recent version V2 of
CARBONES and will be analysed at the final meeting
.



Seasonality in t
he Southern Land also shows consistency in terms of phasing though
somewhat less than the northern land. Maximum carbon uptake across the models
spans the February to April time period. The peak of the dormant season carbon
emission varies from June to Sep
tember depending upon the model. CARBONES
follows the ensemble of the inversion systems, but the maximum of carbon uptake
occur relatively earlier, i.e., around August, while most of the other systems show this
maximum during September to October period.
T
his feature arise from the correction
of the phenology parameters with MODIS
-
NDVI.





Deliverable D610.1

Ref

CARBONES
-
D610.1
-
REP
-
LSCE
-
023
-
01
-
00

Date

05/06/2012

Page

23
/
36



23



Figure 4.
6
:
Mean seasonal cycle of the posterior carbon exchange for the individual participating
inversion submissions. Shown here are the natural land (first column)
and natural ocean (second
column) carbon exchange aggregated over the Northern hemisphere (> 30N), the tropic (30S
-
30N) and
the southern hemisphere (< 30S).


5

Evaluation of land gross
carbon
fluxes



CARBONES gross carbon fluxes

are compared to
:



GPP and
Reco estimates directly
derived
from
the net flux
o
bservations at site level
(FLUXNET data)
as performed in e.g., Reichstein et al. (2005) by using
a flux
partitioning method

(e.g., Baldocchi, 2003 and Papale et al., 2006; see the dedicated
website:
http:/
/www.fluxnet.ornl.gov
)




Global
GPP estimates
from
data

oriented approach. The
data
-
oriented
method
used
a

Model
T
ree
E
nsemble

(hereafter MTE)
which
is a machine learning system builds on
an empirical model. Such a model has been applied to the upscaling

of eddy



Deliverable D610.1

Ref

CARBONES
-
D610.1
-
REP
-
LSCE
-
023
-
01
-
00

Date

05/06/2012

Page

24
/
36



24


covariance measurements
(i.e., water, energy and carbon fluxes)
from local to
continental
scales
(Jung et al., 2011)
.


5.1

Evaluation at the site level


The
CARBONES
GPP and Reco fluxes
are compared to
estimates derived directly from
the
NEE
observations
(
as performed in Reichstein et al. (2005)
)
.
Note that the NEE data were
assimilated in the CCDAS in a sequential approach (see details in the report D420).
Although
not independent, these “data
-
oriented” estimates provide valuable insights on

the
ORCHIDEE
model performances.
Hence, here we evaluate the performances of the model with regard to
the optimization of its process based parameters
(step 2 of our sequential assimilation
approach)
through different
metrics
as
described

in Kuppel et al.

(2012):
In this comparison
we use the results of a so
-
called
single site
optimization
to constrain the ORCHIDEE
parameters
(
SS optimization
)

and a second approach
used in CARBONES
that considers
the
observations from
all
sites
of a given Plant Functional
Type
to optimize these parameters (MS
method).

Figure
5.1

shows the seasonal cycle of GPP and Reco at two sites
for
a 2
-
year time
period
.

We observe that in general the optimizations decrease the seasonal amplitude of GPP
and Reco at these sites

and as ex
pected are closer to the observations
, with a shortening of the
period where GPP is significant. Besides, we observe that the model
-
data fit is generally
improved both for Reco and GPP, although more significantly for Reco.

More details of these
comparison
s are given in Kuppel et al. (2012).





Deliverable D610.1

Ref

CARBONES
-
D610.1
-
REP
-
LSCE
-
023
-
01
-
00

Date

05/06/2012

Page

25
/
36



25



Figure 5.1
:

Seasonal cycle of GPP and R
eco

at a) Hainich and b) Harvard Forest sites, smoothed with a
15
-
day moving average window. The estimations derived from flux
-
partitioning

of NEE (black) are
compared with the prior model (green), the MS optimization (blue)
corresponding to CARBONES
and
the SS optimization (red
, see text
). The averaged annual fluxes in gC/m² are given between brackets.


5.2

Evaluation at global scale
from MTE es
timates


The global GPP estimates from the MTE model are compared to those obtained from
CARBONES project. The comparison is performed over the 1990
-
2008 period. MTE uses
different datasets to upscale GPP from
F
L
UXNET
station
data
: FAPAR satellite observations,
the SYNMAP land cover map, temperature observations from CRU and precipitation from
GPCC (Jung et al. 2011)
.

The global spatial patterns of GPP from MTE and CARBONES agree reasonable well.
Figure
5.2.1 displays the spatial

distributions of both the yearly mean GPPs derived from MTE and
CARBONES together with their differences over
the
1990
-
2008 period at global scale.
Although

the two GPP estimates agree reasonably well, differences can be significant in some
areas. We obta
in a good agreement in boreal regions, while CARBONES gives higher GPPs
in
mainly agricultural used areas
. The largest differences between CARBONES and MTE do
occur in South East of Asia in the Tropics, where differences can reach up to 4000 gC/m
2
/y at
pix
el level. The large differences are partly explained by the fact that MTE does not permit
extreme GPPs values above around 3500 gC/m
2
/y as can be seen from Figure 5.2.2.




Deliverable D610.1

Ref

CARBONES
-
D610.1
-
REP
-
LSCE
-
023
-
01
-
00

Date

05/06/2012

Page

26
/
36



26


Differences in mean annual GPP are associated to different climate regions.
Mean ann
ual
global GPP over 1990
-
2008 period from CARBONES are plotted against MTE
stratified by

the Koeppen
-
Geiger climate classes (Figure 5.2.2).
The largest differences in mean annual
GPP occur in the Aw climate region (equatorial savannah with dry winter). Al
so, the
seasonalities

of both GPP estimates are well correlated (not shown)
, except in some arid and
tropical regions where GPP does not show a pronounced seasonal cycle
. However, we found
large differences over few areas and during particular seasons, e.g
., dry period over savannah
(not shown).



Figure 5.2.1
: Yearly mean GPP (gC/m
2
/y) over 1990
-
2008 estimated from MTE and CARBONES are
shown. The differences between CARBONES and MTE (CARBONES

MTE) are also given.






Deliverable D610.1

Ref

CARBONES
-
D610.1
-
REP
-
LSCE
-
023
-
01
-
00

Date

05/06/2012

Page

27
/
36



27



Figure 5.2.2:

Mean annual GPP from
CARBONES against MTE (1990
-
2008). The different colo
u
rs
represent the Koeppen
-
Geiger climate classes used in the MTE estimates.


GPP from CARBONES has different sensitivities to temperature and precipitation than GPP
from MTE (Figure 5.2.3). The maximum G
PP under a specific temperature condition
increases in both approaches with higher mean annual temperatures in case of fully humid
climate regions (Df


purple, Cf


darkgreen, Af


red; upper panel in Figure 5.2.3). In case of
water limited climate region
s (ET


blue, BS and BW


yellow, Aw


light red) GPP is clearly
below
the GPP of fully humid climate regions

and does not show this relationship with
temperature. Nevertheless, CARBONES reaches under all temperature conditions higher
maximum GPP values th
an MTE. The increase in GPP with increasing temperatures under
non
-
water limited conditions in boreal climate regions is much stronger for CARBONES than
for MTE

(upper limit at purple colors

in Figure 5.2.3
)
. GPP increases also with increasing
mean annual
precipitation in both approaches (lower panel in Figure 5.2.3). Again, for
CARBONES the increase in GPP with increasing mean annual precipitation is much stronger.
At a certain high amount of annual precipitation, increasing precipitation does not increase

GPP further, i.e. water availability is no longer a limiting factor for GPP. In MTE this point is
reached at ca. 1800 mm of mean annual precipitation whereas in CARBONES this point is
reached already at a lower amount of mean annual precipitation (
ca.

120
0 mm). In summary,
CARBONES GPP has a higher sensitivity to temperature in boreal regions and a higher
sensitivity to precipitation globally than GPP estimates from
MTE
.





Deliverable D610.1

Ref

CARBONES
-
D610.1
-
REP
-
LSCE
-
023
-
01
-
00

Date

05/06/2012

Page

28
/
36



28



Figure 5.2.
3
:

Mean annual GPP
from MTE and CARBONES against mean annual
temperature (top)
and mean annual total precipitation (bottom)
. The different colo
u
rs represent the Koeppen
-
Geiger
climate classes
as in Figure 5.2.2
.
The black line is a spline fitted to the upper 95% percentile of the
distribution and represents the uppe
r boundary of GPP under different temperature or and precipiation
conditions, respectively.


This
comparison of flux estimates from a data
-
driven approach (MTE) with a model
-
data
integration approach (CARBONES)
should be considered as a first step that

wi
ll be further



Deliverable D610.1

Ref

CARBONES
-
D610.1
-
REP
-
LSCE
-
023
-
01
-
00

Date

05/06/2012

Page

29
/
36



29


improved
after the end of the project within an article on the Evaluation of CARBONES 20
-
year flux reanalysis. This will

provide a more comprehensive and detailed analysis of
CARBONES strengths and weaknesses, although the MTE should not cons
idered as the truth!

A non
-
exhaustive list of the actions planned can be sum up as follows:



Perform the analysis by land cover or plant functional types



Include other data
-
oriented upscaled estimates (TER, LE, H)



Perform the estimates from MTE with d
ifferent forcing data sets



Compare CARBONES to other biogeochemical model gross carbon flux estimates
(e.g. LPJ, JSBACH)


6

Evaluation of land carbon stocks

6.1

Product used for the evaluation



We compared
CARBONES
biomass estimates to
the
DLO
-
EFI product
s
,

which are
based
on
FAO data further compiled by
ALTERRA
,

and
are
available in the

database

(http://www.alterra.wur.nl/UK/research/Specialisation+Geo
-
information/LGN/)
. This
database includes

data
on forest area, growing stock, increment, harvest levels
(scattered info
only) and age classes (Europe only), based on historical and recent international assessments
and national forest inventory statistics.
For detailed description of these data, consult the
deliverable D300.1 of CARBONES project. Note that DL
O
-
EFI
estimates
are yearly
data
at
global scale and cover

the

1950
-
2010

period.

In what follows, inferences on forest above biomass of CARBONES are compared to DLO
-
EFI estimates. For CARBONES, two simulations from ORCHIDEE are considered: i)
model
l
ed bio
mass by using the default values of the process based parameters of ORCHIDEE
(hereafter ORCHIDEE REFERENCE)
and ii) mode
l
led biomass by using optimized
parameters
(ORCHIDEE OPTIMIZED)
that are constrained by the satellite MODIS NDVI
data

and then with sate
llite and FluxNet data
(the first
and second
step
s

of our sequential data
assimilation system)
.

Note
that we were not able to

compare
with
the version of ORCHIDEE
that has been optimised
by

all three data streams

(V3), given the delay in realisation of the last
step of the sequential approach (step 4). However, given that only few parameters
that control
the above biomass
were further adjusted in step 4, the comparison below can be considered as
the most relevant
one
.



6.2

Results of the comparison






Deliverable D610.1

Ref

CARBONES
-
D610.1
-
REP
-
LSCE
-
023
-
01
-
00

Date

05/06/2012

Page

30
/
36



30


Figure 6.2.1 displays t
he spatial distrib
u
tions of
the
mean annual
above ground
forest
biomass
derived from
CARBONES and
the estimates from
DLO
-
EFI for year 2005.
Overall,
the two CARBONES estimates agree
reasonably

well with DLO
-
EFI data

in term of global
spatial pattern
, but the following differences can be highlighted:



CARBONES tends to produce more biomass over North of Europe, Russia, and over
tropical Asian regions
.



CARBONES tends to produce lower biomass in th
e south
part
of central Africa
.




Qualitatively, differences between the
two CARBONES estimates

(Figure 6.2.1)

appear to be
relatively small
.

However, as shown in Figure 6.2.2, the
estimates from the
optimized
ORCHIDEE
model
re
present a
better

agreement with the
DLO
-
EFI product based on FAO
data

compared to the reference simulations
. Indeed,
a

lowe
r

bias is obtained
between
the
optimized
ORCHIDEE
estimates and
DLO
-
EFI

data.

This is primarily the result of a reduced
growing season length followi
ng the optimisation of the phenology parameters.
Figure 6.2.3
shows that the difference in the biomass between the optimized and reference versions of
ORCHIDEE is greatest in northe
rn temperate and boreal regions, though there is a general
reduction in
biomass across all forest PFTs.


The
re is a similar

reduction in
the positive
bias
of ORCHIDEE
after optimization with
both
satellite NDVI and FluxNet data
, though
it has increased slightly compared to the
optimization using onl satellite NDVI data
.
This i
s due to the fact that the
value of
Vcmax
increases
as a result of

the optimization with the NEE data, and therefore
there is also an
increase in biomass
.

T
he
root mean square error
s

(RMSE)
derived from the t
hree

CARBONES estimates
against
DLO
-
EFI data
r
emain

rather large
,

and nearly
unchanged

after optimisation.

This
demonstrates that there is

potentially a structural error, or lack of process representation,
in
the model that cannot be accounted for by parameter optimization. It is likely that disturban
ce
and forest age need to be
properly considered

in order to achieve more accurate biomass
estimates.

Note
finally that the comparison is done for one year

only,

as the change in biomass is rather
small across the 20 y
ea
r

period but that further analysis will consider the change between
1990 and 2009.







Deliverable D610.1

Ref

CARBONES
-
D610.1
-
REP
-
LSCE
-
023
-
01
-
00

Date

05/06/2012

Page

31
/
36



31




Figure 6.2.1
:
Spatial distributions of mean annual forest above biomass derived from DLO
-
EFI and
CARBONES

for year 2005
. For CARBONES, two products are shown:
simulations of
ORCHIDEE
by
using i) the default parameters of the model (REFERENCE) and ii) optimized ph
e
nological
parameters
of the
model
when constraining them with
the

satellite NDVI data (OPTIMIZED).

DLO

-

EFI

ORCHIDEE

REFERENCE

ORCHIDEE

OPTIMIZED




Deliverable D610.1

Ref

CARBONES
-
D610.1
-
REP
-
LSCE
-
023
-
01
-
00

Date

05/06/2012

Page

32
/
36



32





Figure 6.2.2:
The differences given by the mean bias (top) and the RMS
E

(bottom) between
DLO
-
EFI and the t
hree

CARBONES estimates are shown.

The t
hree

CARBONES estimates
are simulations of ORCHIDEE by using i
) the default parameters of the model
(REFERENCE) and ii) optimized phenological parameters of the model when constraining
them with the satellite NDVI data (
optim. with MODIS
)

and satellite NDVI and FluxNet NEE
and LE daa (optim. with MODIS + FluxNet)
.





Deliverable D610.1

Ref

CARBONES
-
D610.1
-
REP
-
LSCE
-
023
-
01
-
00

Date

05/06/2012

Page

33
/
36



33



Figure 6.2.3:

The difference in biomass
between the optimized and reference
(default
parameters)

version
s

of ORCHIDEE.




Deliverable D610.1

Ref

CARBONES
-
D610.1
-
REP
-
LSCE
-
023
-
01
-
00

Date

05/06/2012

Page

34
/
36



34


7

Conclusion
s


We presented in this report
a

first evaluation of CARBONES products against other

independent products
. This comparison

was performed

to highlight the information content
brought by CARBONES on the carbon
budgets

at continental to hemispheric scales
. The
se
results
should still be considered as preliminary as they will be updated with version V2 of
CARBONES for the final me
eting and possibly version V3 (parameter optimizations in step
4) before the end of the project. Note that a possible revision of this report will be considered
before final submission
.
Overall, CARBONES performs as well as the other products for this
curr
ent version (V1.0
)

and bring new feature
s

on the carbon cycle
,

that need further
investigations:



Fossil fuel emission
s

with hourly temporal variations
appear to be

significantly larger
that those from other products



Ocean carbon fluxes
show new IAV with a
pronounced increased ocean uptake after
2002 that still needs to

be validated

or confirmed with other proxies.



Net land carbon fluxes follow most standard atmospheric inversion

results

but with
slight diffe
rences: the IAV over the tropic is slightly lower
in CARBONES than the
other products; the net land carbon uptake in the northern continental regions is
slightly lower than most inversions, especially for North America; the southern land
IAV tend to be larger than in the other products. All these features

will be summarized
in paper focussing on the analysis of the 20
-
year CARBONES reanalysis.



A favourable comparison against biomass
forest data from FAO statistics appears with
lower biases after the assimilation of MODIS
-
NDVI and FluxNet data.


As reported

in the introduction, th
is report

aims to present
the potential of CARBONES
products and should be considered as a framework to analyse the results of a Carbon Cycle
Multi
-
Data Assimilation System.

The development of the various tools
for
the

evaluation exercise presented in this report
represent a

continuing effort

and these tools

will be
used

after the duration of the
CARBONES project to further valorised CARBONES products
.







Deliverable D610.1

Ref

CARBONES
-
D610.1
-
REP
-
LSCE
-
023
-
01
-
00

Date

05/06/2012

Page

35
/
36



35


8

References

Andres, R. J., T. A. Boden, F.
-
M. Breon, P. Ciais
, S. Davis, D. Erickson, J. S. Gregg, A.
Jacobson, G. Marland, J. Miller, T. Oda, J. G. J. Olivier, M. R. Raupach, P. Rayner, and K.
Treanton, 2012: A synthesis of carbon dioxide emissions from fossil
-
fuel combustion
,
Biogeosciences, 9, 1845

1871, doi:10.5
194/bg
-
9
-
1845
-
2012


Baldocchi, D. D. (2003), Assessing the eddy covariance technique for evaluating carbon
dioxide exchange rates of ecosystems: past, present and future. Global Change Biology, 9,
479

492.


Bush et al., 2010: Bush T., Tsagatakis

I., Passant N., Griffin A., Pearson B. (2010): NAEI UK
Emission Mapping Methodology 2007 (URL: http://www.naei.org.uk/reports.php)


Chevallier, F., P. Ciais, T. J. Conway, T. Aalto, B. E. Anderson, P. Bousquet, E. G. Brunke,
L. Ciattaglia, Y. Esaki, M. Fr
öhlich, A.J. Gomez, A.J. Gomez
-
Pelaez, L. Haszpra, P.
Krummel, R. Langenfelds, M. Leuenberger, T. Machida, F. Maignan, H. Matsueda, J. A.
Morguí, H. Mukai, T. Nakazawa, P. Peylin, M. Ramonet, L. Rivier, Y. Sawa, M. Schmidt, P.
Steele, S. A. Vay, A. T. Verm
eulen, S. Wofsy, D. Worthy, (2010), CO2 surface fluxes at grid
point scale estimated from a global 21
-
year reanalysis of atmospheric measurements, J.
Geophys. Res., 115, D21307, doi:10.1029/2010JD013887.


Chevallier, F., et al., CO2 surface fluxes at grid
point scale estimated from a global 21 year
reanalysis of atmospheric measurements, JOURNAL OF GEOPHYSICAL RESEARCH, Vol.
115, D21307, doi:10.1029/2010JD013887, 2010



EDGAR, 2010: Emissions Database for Global Atmospheric Research
(URL:http://edgar.jrc.e
c.europa.eu/methodology.php)


GLOBALVIEW
-
CO2: Cooperative Atmospheric Data Integration Project
-

Carbon Dioxide.
CD
-
ROM, NOAA ESRL, Boulder, Colorado [Also available on Internet via anonymous FTP
to ftp.cmdl.noaa.gov, Path: ccg/co2/GLOBALVIEW], 2011.


Jung
, M., et al. (2011), Global patterns of land

atmosphere fluxes of carbon dioxide, latent
heat, and sensible heat derived from eddy covariance, satellite, and meteorological
observations, J. Geophys. Res., 116, G00J07, doi:10.1029/2010JG001566.


Kuppel S.,
P. Peylin,

F. Chevallier, C. Bacour,
and
F. Maignan
,

2012:


Constraining a
global ecosystem model with multi
-
site eddy
-
covariance data,
Submitted to Biogeosciences





Deliverable D610.1

Ref

CARBONES
-
D610.1
-
REP
-
LSCE
-
023
-
01
-
00

Date

05/06/2012

Page

36
/
36



36


Oda, T. and Maksyutov, S.: A very high
-
resolution (1 km×1 km) global fossil fuel CO2
em
ission inventory derived using a point source database and satellite observations of
nighttime lights, Atmos. Chem. Phys., 11, 543
-
556, doi:10.5194/acp
-
11
-
543
-
2011, 2011.


Peylin Philippe, Gurney Kevin, Law Rachel, Chevallier Frederic, Jacobson Andy, Niwa

Yosuke, Patra Prabir, Rayner Peter, Roedenbeck Christian, Wouter Peters, Yamada Kazutaka
,

2012: Global Atmospheric Carbon Budget: results from an ensemble of atmospheric CO2
inversions,
In preparation


Pregger, T., Friedrich, R. 2009: Effective pollutant
emission heights for atmospheric transport
modelling based on real

world information, Environmental Pollution 157 (2009), S. 552

560.


Rong Wang, Shu Tao, Philippe Ciais, Huizhong Shen, Ye Huang, Han Chen, Guofeng Shen, Bin
Wang, Wei Li, Yanyan Zhang, Yan
Lu, Dan Zhu, Yuanchen Chen, Xiaopeng Liu, Wentao Wang,
Xilong Wang, Bengang Li, and Shilong Piao; 2012: High resolution mapping of combustion
processes and implications for CO2 emissions, Submitted to ACP


Takahashi

T.
, Stewart C. Sutherland, Rik
Wanninkhof, Colm Sweeney, Richard A. Feely, David W.
Chipman, Burke Hales, Gernot Friederich, Francisco Chavez, Christopher Sabine, Andrew Watson,
Dorothee C.E. Bakker, Ute Schuster, Nicolas Metzl, Hisayuki Yoshikawa
-
Inoue, Masao Ishii,
Takashi Midorikawa,

Yukihiro Nojiri, Arne Körtzinger, Tobias Steinhoff, et al. Corrigendum to
“Climatological mean and decadal change in surface ocean pCO2, and net sea

air CO2 flux over the
global oceans” [Deep Sea Res. II 56 (2009) 554

577].


Tiedtke, M. (1989), A Comprehe
nsive Mass Flux Scheme for Cumulus Parameterization in

Large
-
Scale Models, Monthly Weather Review, 117, 1779
-
1800.


Wanninkhof, R., 1992. Relationship between wind speed and gas exchange. J. Geophys. Res.
97, 7373

7382


Zhu C., Byrd R.H., Lu P., Nocedal J
. (1995), A Limited Memory Algorithm for Bound
Constrained Optimization, (1995), SIAM Journal on Scientific and Statistical Computing , 16,
5, pp. 1190
-
1208.