ArticlePDF Available

Prior biosphere model impact on global terrestrial CO 2 fluxes estimated from OCO-2 retrievals

Authors:

Abstract and Figures

This study assesses the impact of different state of the art global biospheric CO2 flux models, when applied as prior information, on inverse model “top-down” estimates of terrestrial CO2 fluxes obtained when assimilating Orbiting Carbon Observatory 2 (OCO-2) observations. This is done with a series of observing system simulation experiments (OSSEs) using synthetic CO2 column-average dry air mole fraction (XCO2) retrievals sampled at the OCO-2 satellite spatiotemporal frequency. The OSSEs utilized a 4-D variational (4D-Var) assimilation system with the GEOS-Chem global chemical transport model (CTM) to estimate CO2 net ecosystem exchange (NEE) fluxes using synthetic OCO-2 observations. The impact of biosphere models in inverse model estimates of NEE is quantified by conducting OSSEs using the NASA-CASA, CASA-GFED, SiB-4, and LPJ models as prior estimates and using NEE from the multi-model ensemble mean of the Multiscale Synthesis and Terrestrial Model Intercomparison Project as the “truth”. Results show that the assimilation of simulated XCO2 retrievals at OCO-2 observing modes over land results in posterior NEE estimates which generally reproduce “true” NEE globally and over terrestrial TransCom-3 regions that are well-sampled. However, we find larger spread among posterior NEE estimates, when using different prior NEE fluxes, in regions and seasons that have limited OCO-2 observational coverage and a large range in “bottom-up” NEE fluxes. Seasonally averaged posterior NEE estimates had standard deviations (SD) of ∼10 % to ∼50 % of the multi-model-mean NEE for different TransCom-3 land regions with significant NEE fluxes (regions/seasons with a NEE flux ≥0.5 PgC yr−1). On a global average, the seasonally averaged residual impact of the prior model NEE assumption on the posterior NEE spread is ∼10 %–20 % of the posterior NEE mean. Additional OCO-2 OSSE simulations demonstrate that posterior NEE estimates are also sensitive to the assumed prior NEE flux uncertainty statistics, with spread in posterior NEE estimates similar to those when using variable prior model NEE fluxes. In fact, the sensitivity of posterior NEE estimates to prior error statistics was larger than prior flux values in some regions/times in the tropics and Southern Hemisphere where sufficient OCO-2 data were available and large differences between the prior and truth were evident. Overall, even with the availability of spatiotemporally dense OCO-2 data, noticeable residual differences (up to ∼20 %–30 % globally and 50 % regionally) in posterior NEE flux estimates remain that were caused by the choice of prior model flux values and the specification of prior flux uncertainties.
Content may be subject to copyright.
Atmos. Chem. Phys., 19, 13267–13287, 2019
https://doi.org/10.5194/acp-19-13267-2019
© Author(s) 2019. This work is distributed under
the Creative Commons Attribution 4.0 License.
Prior biosphere model impact on global terrestrial CO2fluxes
estimated from OCO-2 retrievals
Sajeev Philip1,2,a, Matthew S. Johnson1, Christopher Potter1, Vanessa Genovesse3,1, David F. Baker4,5,
Katherine D. Haynes6, Daven K. Henze7, Junjie Liu8, and Benjamin Poulter9
1NASA Ames Research Center, Moffett Field, CA 94035, USA
2NASA Postdoctoral Program administered by Universities Space Research Association, Columbia, MD 21046, USA
3School of Natural Sciences, California State University, Monterey Bay, CA 93955, USA
4NOAA Earth System Research Laboratory, Global Monitoring Division, Boulder, CO 80305-3337, USA
5Cooperative Institute for Research in the Atmosphere, Colorado State University, Fort Collins, CO 80521, USA
6Department of Atmospheric Science, Colorado State University, Fort Collins, CO 80523, USA
7Department of Mechanical Engineering, University of Colorado at Boulder, Boulder, CO 80309, USA
8Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA 91109, USA
9NASA Goddard Space Flight Center, Greenbelt, MD 20771, USA
anow at: NASA Academic Mission Services by Universities Space Research Association, Mountain View, CA 94043, USA
Correspondence: Sajeev Philip (philip.sajeev@gmail.com) and Matthew S. Johnson (matthew.s.johnson@nasa.gov)
Received: 13 October 2018 – Discussion started: 6 February 2019
Revised: 14 September 2019 – Accepted: 25 September 2019 – Published: 28 October 2019
Abstract. This study assesses the impact of different state
of the art global biospheric CO2flux models, when applied
as prior information, on inverse model “top-down” estimates
of terrestrial CO2fluxes obtained when assimilating Orbit-
ing Carbon Observatory 2 (OCO-2) observations. This is
done with a series of observing system simulation exper-
iments (OSSEs) using synthetic CO2column-average dry
air mole fraction (XCO2) retrievals sampled at the OCO-2
satellite spatiotemporal frequency. The OSSEs utilized a 4-
D variational (4D-Var) assimilation system with the GEOS-
Chem global chemical transport model (CTM) to estimate
CO2net ecosystem exchange (NEE) fluxes using synthetic
OCO-2 observations. The impact of biosphere models in
inverse model estimates of NEE is quantified by conduct-
ing OSSEs using the NASA-CASA, CASA-GFED, SiB-4,
and LPJ models as prior estimates and using NEE from
the multi-model ensemble mean of the Multiscale Synthesis
and Terrestrial Model Intercomparison Project as the “truth”.
Results show that the assimilation of simulated XCO2re-
trievals at OCO-2 observing modes over land results in pos-
terior NEE estimates which generally reproduce “true” NEE
globally and over terrestrial TransCom-3 regions that are
well-sampled. However, we find larger spread among pos-
terior NEE estimates, when using different prior NEE fluxes,
in regions and seasons that have limited OCO-2 observa-
tional coverage and a large range in “bottom-up” NEE fluxes.
Seasonally averaged posterior NEE estimates had standard
deviations (SD) of 10 % to 50 % of the multi-model-
mean NEE for different TransCom-3 land regions with sig-
nificant NEE fluxes (regions/seasons with a NEE flux
0.5 PgC yr1). On a global average, the seasonally aver-
aged residual impact of the prior model NEE assumption on
the posterior NEE spread is 10 %–20 % of the posterior
NEE mean. Additional OCO-2 OSSE simulations demon-
strate that posterior NEE estimates are also sensitive to the
assumed prior NEE flux uncertainty statistics, with spread in
posterior NEE estimates similar to those when using variable
prior model NEE fluxes. In fact, the sensitivity of posterior
NEE estimates to prior error statistics was larger than prior
flux values in some regions/times in the tropics and Southern
Hemisphere where sufficient OCO-2 data were available and
large differences between the prior and truth were evident.
Overall, even with the availability of spatiotemporally dense
OCO-2 data, noticeable residual differences (up to 20 %–
30 % globally and 50% regionally) in posterior NEE flux es-
Published by Copernicus Publications on behalf of the European Geosciences Union.
13268 S. Philip et al.: Prior biosphere model impact
timates remain that were caused by the choice of prior model
flux values and the specification of prior flux uncertainties.
1 Introduction
Carbon dioxide (CO2) is the most important greenhouse
gas (GHG) contributing to climate change on a global scale
(IPCC, 2014). The anthropogenic emission of CO2, primar-
ily from fossil fuel usage, has led to average global CO2
mixing ratios reaching historically high levels of > 400 ppm
(parts per million) (Seinfeld and Pandis, 2016). In addition
to fossil fuel emissions, the processes involved in the ex-
change of carbon between the atmosphere and terrestrial bio-
sphere are a major factor controlling atmospheric concen-
trations of CO2(e.g., Schimel et al., 2001) with an esti-
mated global biosphere sink of 3.0 PgC yr1(Le Quéré
et al., 2018). However, current estimates of regional-scale
atmosphere–terrestrial biosphere CO2exchange have large
uncertainties (Schimel et al., 2015). “Bottom-up” (hereafter
quotation marks will be omitted) techniques typically simu-
late the atmosphere–terrestrial biosphere exchange based on
our understanding of these complex exchange processes and
by constraining these estimates with remote-sensing inputs
and limited measurements available for evaluation. Previous
studies intercomparing several of the most commonly used
biospheric flux models (Heimann et al., 1998, Huntzinger et
al., 2012; Sitch et al., 2015; Ott et al., 2015; Ito et al., 2016)
and multi-model ensemble integration projects (Schwalm
et al., 2015) reveal a large spread among global/regional
bottom-up terrestrial biospheric flux estimates and the sub-
components such as ecosystem primary production and res-
piration (Huntzinger et al., 2012).
An alternate approach to estimate biospheric CO2fluxes is
through “top-down” (hereafter quotation marks will be omit-
ted) estimation techniques using inverse models with highly
accurate in situ data (e.g., Baker et al., 2006b) or dense
and globally distributed satellite data (e.g., Chevallier et al.,
2005). The Orbiting Carbon Observatory 2 (OCO-2) satel-
lite, launched in 2014, is the spaceborne sensor with the finest
resolution and highest sensitivity with respect to CO2in the
atmospheric boundary layer to date (Crisp et al., 2017; El-
dering et al., 2017a). Studies applying OCO-2 retrievals re-
vealed the ability to investigate novel aspects of the carbon
cycle (e.g., Eldering et al., 2017b; Liu et al., 2017); however,
the top-down estimates of surface CO2fluxes from numerous
inverse modeling systems, using identical OCO-2 observa-
tions, show differences among optimized/posterior regional
CO2fluxes (Crowell et al., 2019). Previous studies investigat-
ing CO2flux inversions (e.g., Peylin et al., 2013; Chevallier
et al., 2014; Houweling et al., 2015) suggest that this spread
among optimized CO2flux estimates could be due to numer-
ous factors, such as the accuracy and precision of observation
data (Rödenbeck et al., 2006), imperfect observation cover-
age (Liu et al., 2014; Byrne et al., 2017), data density (Law
et al., 2003; Rödenbeck et al., 2003), and poorly character-
ized measurement error covariance (Law et al., 2003; Tak-
agi et al., 2014). Variations in inverse estimation setups be-
tween modeling groups, such as model transport (Chevallier
et al., 2010; Houweling et al., 2010; Basu et al., 2018) and
inversion methods (Chevallier et al., 2014; Houweling et al.,
2015), could also lead to inter-model spread in posterior es-
timates.
In addition to the variables listed above, the assumed prior
fluxes and the associated prior error covariance can also im-
pact top-down global/regional CO2flux estimates (e.g., Gur-
ney et al., 2003). Gurney et al. (2003) assessed the sensitiv-
ity of CO2flux inversions to the specification of prior flux
uncertainty and found that the posterior estimates were sen-
sitive to the prior fluxes over regions with limited in situ ob-
servations. In addition, Wang et al. (2018) found that optimal
CO2flux allocation over land vs. ocean, using satellite and/or
in situ data assimilations, is sensitive to the specification of
prior flux uncertainty. Furthermore, Chevallier et al. (2005)
and Baker et al. (2006a, 2010) highlighted the importance of
accurate assumptions of prior flux uncertainty by conduct-
ing 4-D variational (4D-Var) assimilations of satellite col-
umn retrievals of CO2. However, to date, controlled experi-
mental studies have not been undertaken to isolate and quan-
titatively assess the impact of assumed prior fluxes and prior
uncertainty to inverse estimates of biospheric CO2fluxes us-
ing satellite observations.
Therefore, during this study we conduct a series of con-
trolled experiments to quantitatively assess the impact of as-
sumed prior fluxes and prior uncertainty on global and re-
gional CO2inverse model flux estimates when assimilating
OCO-2 data. In order to achieve this, a series of observing
system simulation experiments (OSSEs) are conducted us-
ing synthetic OCO-2 observations in the GEOS-Chem 4D-
Var assimilation system, with four different prior bottom-up
NEE CO2flux estimates. Section 2 of this study describes the
methods applied during this work including individual mod-
els and model input, synthetic OCO-2 data, and the inver-
sion technique applied in the OSSEs. Section 3 presents the
forward and inverse model results of simulated atmospheric
CO2concentrations and inferred posterior flux estimates. Fi-
nally, our concluding remarks and discussion are presented
in Sect. 4.
2 Methods
To quantify the impact of prior model NEE predictions on
posterior estimates of biospheric CO2fluxes, a series of CO2
forward and inverse model simulations were conducted with
four different state of the art biosphere models. OSSE sim-
ulations were designed to isolate the differences in poste-
rior NEE estimates caused by the selection of prior model
biospheric CO2fluxes and uncertainties when assimilating
Atmos. Chem. Phys., 19, 13267–13287, 2019 www.atmos-chem-phys.net/19/13267/2019/
S. Philip et al.: Prior biosphere model impact 13269
OCO-2 observations. The OSSE framework, input variables,
inversion technique, and analysis method are presented be-
low.
2.1 Prior NEE fluxes
NEE is the net difference of gross primary production (GPP)
and total ecosystem respiration (Re), which itself is the sum
of autotrophic respiration (Ra) and heterotrophic respiration
(Rh). NEE, estimated by terrestrial biospheric CO2flux mod-
els, is commonly applied in CTMs to simulate atmosphere–
terrestrial biosphere carbon exchange. Many biosphere car-
bon models estimate GPP and Re; however, some models
simulate net primary productivity (NPP), which is defined
as the difference between GPP and Ra. In this study, we ap-
ply year-specific NEE fluxes calculated from four state of
the art biosphere models: (1) NASA Carnegie Ames Stanford
Approach (NASA-CASA), (2) CASA-Global Fire Emissions
Database (CASA-GFED), (3) Simple-Biosphere model ver-
sion 4 (SiB-4), and (4) Lund–Potsdam–Jena (LPJ). It should
be noted that the prior biosphere models used in this study
include only NEE, and a single dataset for wild fire and
fuel wood burning CO2emissions was added separately (see
Sect. 2.3). The models applied during this study represent a
range of diagnostic approaches, from models predicting bio-
spheric CO2fluxes using remotely sensed data (e.g., fraction
of absorbed photosynthetically active radiation, leaf area in-
dex, normalized difference vegetation index) to fully prog-
nostic models unconstrained by observations. In addition,
we selected both balanced/neutral (SiB-4) and non-balanced
(NASA-CASA, CASA-GFED, LPJ) biospheric fluxes in our
OSSEs in order to represent the range of prior models cur-
rently being used in CO2inversion modeling studies.
CASA is an ecosystem model predicting NPP based on
light use efficiency and Rhbased on soils/plant production
information (Potter et al., 1993, 2012b). The NASA-CASA
model is a version of the original CASA model (Potter et
al., 1993) currently being developed at the NASA Ames Re-
search Center (Potter et al., 2003, 2007, 2009, 2012a, b).
NASA-CASA specifically utilizes data on global vegetation
cover (enhanced vegetation index, surface solar irradiance
data) and land disturbances retrieved from the NASA Mod-
erate Resolution Imaging Spectroradiometer (MODIS) satel-
lite (Potter et al., 2012b). In addition to Rh, NASA-CASA
includes redistributed crop harvest CO2emissions to the at-
mosphere (Potter et al., 2012b). The CASA-GFED model is
a different version of the original CASA model and is de-
scribed in Randerson et al. (1996) with subsequent versions
being described in recent literature (van der Werf et al., 2004,
2006, 2010). NASA-CASA and CASA-GFED differ in the
use of input parameters and some of the parameterizations
(see Ott et al., 2015 for further description).
The SiB-4 model was developed at Colorado State Uni-
versity (Sellers et al., 1986; Denning et al., 1996), with de-
tails of the newest versions described in Haynes et al. (2013).
This model is a mechanistic, prognostic land surface model
that integrates heterogeneous land cover, environmentally re-
sponsive prognostic phenology, dynamic carbon allocation,
and cascading carbon pools from live biomass to surface lit-
ter and soil organic matter (Haynes et al., 2013; Baker et al.,
2013; Lokupitiya et al., 2009; Schaefer et al., 2008; Sellers et
al., 1996). By combining biogeochemical, biophysical, and
phenological processes, SiB-4 predicts vegetation and soil
moisture states, land surface energy and water budgets, and
the terrestrial carbon cycle. Rather than relying on satellite
input data, SiB-4 fully simulates the terrestrial carbon cycle
by using the carbon fluxes to determine the above- and be-
lowground biomass, which in turn feed back to impact car-
bon assimilation and respiration. Similar to NASA-CASA,
the SiB4 model redistributes crop harvest CO2emission to
the atmosphere. Note that we use a balanced (neutral) bio-
spheric NEE flux for the SiB-4 model.
The LPJ model is a process-based dynamic global veg-
etation model (Sitch et al., 2003; Polter et al., 2014). The
LPJ-wsl dynamic global vegetation model (Sitch et al., 2003)
was used to simulate NEE using meteorological data from
the Climate Research Unit (Harris et al., 2013). LPJ is fully
prognostic, meaning that the establishment, growth, and mor-
tality of vegetation are represented by first-order physio-
logical principles. The model includes nine plant functional
types distinguished by their phenology, photosynthetic path-
way, and physiognomy. Phenology status is determined daily
and photosynthesis is estimated using a modified Farquhar
scheme (Haxeltine and Prentice, 1996). NPP is calculated
from photosynthesis after accounting for Raand reproductive
allocation. The LPJ-wsl model has been evaluated in several
benchmarking activities for stocks and fluxes (Peng et al.,
2015; Sitch et al., 2015).
In order to provide a “true” (hereafter quotation marks
will be omitted) NEE flux for the OSSEs conducted in this
study (Sect. 2.4), we use the multi-model ensemble NEE
mean from the Multiscale Synthesis and Terrestrial Model
Intercomparison Project (MsTMIP) (Huntzinger et al., 2013,
2018; Fisher et al., 2016a, b). The MsTMIP NEE fluxes are
from a weighted ensemble mean of 15 biosphere models
(Schwalm et al., 2015) for the year 2010. Here we apply the
MsTMIP data for the year 2010 as the “truth” (hereafter quo-
tation marks will be omitted) with year-specific prior model
predictions for 2015. This procedure is justified in our case
as within an OSSE framework there needs to be a difference
between true and prior fluxes, as long as the true values are
realistic in nature. The MsTMIP ensemble NEE mean rep-
resents a summary over all 15 models which smooths out
errors particular to any given model. This true NEE flux is
used to produce the synthetic OCO-2 observations applied in
this study (described further in Sect. 2.4.3).
The true and four prior model NEE fluxes were regridded
from their native horizontal resolutions to the grid resolu-
tion of the inverse model simulations (4.0latitude ×5.0
longitude). The MsTMIP NEE fluxes are provided as 3-
www.atmos-chem-phys.net/19/13267/2019/ Atmos. Chem. Phys., 19, 13267–13287, 2019
13270 S. Philip et al.: Prior biosphere model impact
hourly averages and the four year-specific prior models were
provided as monthly mean GPP or NPP and Reor Rh.
Therefore, we imposed diurnal (hourly) and daily variability
to these four prior models following the approach in Car-
bonTracker CT2016 (https://www.esrl.noaa.gov/gmd/ccgg/
carbontracker/CT2016/CT2016_doc.php, last access: 22 Au-
gust 2017), which is based on Olsen and Randerson (2004).
This hourly/daily NEE variability for each prior model was
calculated using the downward solar radiation flux and 2 m
air temperature data from the GEOS-FP (Goddard Earth Ob-
serving System Model, Version 5 “Forward Processing”) me-
teorological product and monthly averaged GPP and Rhfrom
the respective model. We allow the true and prior models to
have different diurnal variability in order to represent a re-
alistic scenario, as prior models will differ some from the
actual diurnal variability of the NEE in nature. In general,
the diurnal variability of NEE is similar between the true and
individual prior models. An example is shown in Fig. S1 in
the Supplement where it can be seen that the diurnal NEE
from the true and prior models for July 2015 at the Park Falls
flux tower site (45.95N, 90.27W) have near-identical tem-
poral diurnal patterns and only differ in NEE magnitude. Ta-
ble 1 shows the global annual NEE flux estimates for the four
prior models and the truth. From this table it can be seen that
the MsTMIP product shows a strong annual global sink of
4.31 PgC yr1. NASA-CASA and CASA-GFED predict a
global sink of 2 PgC yr1and differ by 0.6 PgC yr1.
SiB-4 NEE predicts a source of 1 PgC yr1and the LPJ
model predicts a strong sink of 5.5 PgC yr1. Section 3.1
further describes the spatiotemporal differences of the NEE
fluxes between these four prior models and the truth.
2.2 GEOS-Chem model
The GEOS-Chem chemical transport model (CTM) (Bey
et al., 2001) used in this study has the capability to
run forward CO2simulations (Suntharalingam et al.,
2004; Nassar et al., 2010) and corresponding adjoint
model calculations (Henze et al., 2007; Liu et al., 2014).
In this study, we use the GEOS-Chem adjoint ver-
sion 35 (http://wiki.seas.harvard.edu/geos-chem/index.php/
GEOS-Chem_Adjoint; last access: 27 January 2017), which
is compatible with version 8-02 of the GEOS-Chem forward
model. Liu et al. (2014) tested the accuracy of the GEOS-
Chem CO2adjoint system, which has been used for several
CO2inverse modeling studies (e.g., Liu et al., 2017; Bow-
man et al., 2017; Deng et al., 2014). The model is driven
with assimilated meteorological fields from the GEOS-FP
model of the NASA Global Modeling Assimilation Office
(GMAO). The GEOS-FP meteorology fields have a native
horizontal resolution of 0.25×0.3125and 72 native hy-
brid sigma-pressure vertical levels from the Earth’s surface
to 0.01 hPa. We conduct simulations with a coarser spatial
resolution (4.0×5.0) with 47 reduced vertical levels to at-
tain reasonable computational efficiency.
Table 1. Prior and (posterior) global annual mean NEE fluxes and
CO2emission inventories (PgC yr1) for the year 2015 used in the
OSSE simulations during this study.
NEE Model NEE flux (PgC yr1)
MsTMIP1,24.31
NASA-CASA31.86 (4.14)
CASA-GFED42.42 (4.24)
SiB-450.95 (4.11)
LPJ65.53 (4.36)
Inventory CO2emissions (PgC yr1)
Fossil fuel79.86
Ocean82.41
Biomass burning92.05
Fuel wood burning90.50
1The MsTMIP NEE dataset is representative of the year 2010 and is
an ensemble mean of 15 different NEE models. 2Huntzinger et
al. (2013, 2016); Fisher et al. (2016a, b). 3Potter et al. (1993, 2012a,
b). 4Potter et al. (1993); Randerson et al. (1996). 5Haynes et
al. (2013); Baker et al. (2013). 6Sitch et al. (2003); Poulter et
al. (2014). 7Oda et al. (2018); Nassar et al. (2013). 8CarbonTracker
CT2016; Peters et al. (2007). 9CASA-GFED3; van der Werf et
al. (2004, 2006, 2010).
2.3 Non-NEE CO2fluxes
To simulate concentrations of atmospheric CO2, we used
several land and ocean CO2flux inventories in addition
to the NEE estimates from the prior biosphere models
(global annual budgets listed in Table 1). This study used
the year-specific fossil fuel and cement production inven-
tory from the Open-source Data Inventory for Anthropogenic
CO2(ODIAC-2016) developed by Oda et al. (2018). Fol-
lowing the approach of Nassar et al. (2013), the monthly
ODIAC-2016 inventory is converted from the native tempo-
ral variability into diurnal (hourly) and weekday/weekend
variability (courtesy of Sourish Basu and the OCO-2 Sci-
ence Team). Wild fire emissions and fuel wood burning
emissions were taken from the 3-hourly Global Fire Emis-
sions Database (GFED3). Shipping emissions are from the
International Comprehensive Ocean-Atmosphere Data Set
(ICOADS; Corbett and Koehler, 2003; 2004), and aviation
emissions are from the Aviation Emissions Inventory Code
(AEIC; Olsen et al., 2013). We used 3-D chemical production
of CO2from the oxidation of carbon monoxide, methane,
and non-methane volatile organic compounds (Nassar et
al., 2010). The shipping, aviation and 3-D chemical source
are climatological and are taken from the Bowman (2017)
dataset. To simulate oceanic CO2fluxes, we apply the year-
specific 3-hourly posterior estimates from the CarbonTracker
2016 (CT2016) model constrained with in situ data (Peters et
al., 2007; http://carbontracker.noaa.gov, last access: 22 Au-
gust 2017). All emission inventories, except the NEE fluxes,
are kept constant between the different inverse model simu-
lations.
Atmos. Chem. Phys., 19, 13267–13287, 2019 www.atmos-chem-phys.net/19/13267/2019/
S. Philip et al.: Prior biosphere model impact 13271
Table 2. Summary of the different OSSEs conducted during this work.
Experiment (no. of OSSEs) OCO-2 XCO2Mode Prior NEE model NEE uncertainty
Variable prior NEE (4) LN +LG All1Multi-model SD2
Variable prior NEE (4) OG All1Multi-model SD2
Variable prior uncert. (2) LN +LG CASA-GFED Uniform 10 %/100 %
1NASA-CASA, CASA-GFED, SiB-4 and LPJ. 2SD denotes standard deviation.
2.4 Observing system simulation experiments (OSSEs)
2.4.1 OSSE framework
This study conducted several OSSEs to assess the impact of
prior biospheric CO2models and associated prior uncertainty
specifications on posterior estimates of NEE when assimi-
lating OCO-2 data. To assess the impact of prior fluxes, we
conduct four baseline OSSEs (using the four prior biosphere
models) assimilating synthetic land nadir (LN) and land glint
(LG) observations together, plus an additional four OSSEs
using just ocean glint (OG) observations. These OSSE sim-
ulations were designed in such a way that the differences in
posterior NEE estimates are due solely to the choice of prior
biospheric flux (e.g., identical initial atmospheric CO2condi-
tions, non-NEE fluxes, OCO-2 sampling frequency, observa-
tional data uncertainty, and so on). Furthermore, to assess the
impact of prior uncertainty specifications, we conduct two
additional OSSEs (in addition to our baseline prior uncer-
tainty assumption described in Sect. 2.4.5) with synthetic LN
and LG observations using prior error set uniformly to 10 %
and 100 % of a particular prior NEE flux (CASA-GFED). Ta-
ble 2 shows the summary of OSSEs conducted for this study.
During all OSSE simulations NEE and oceanic CO2fluxes
are optimized, with all other sources kept constant, in order
to be consistent with the methods commonly used by inverse
modeling systems focused on estimating NEE. We optimize
oceanic fluxes along with NEE, although the same ocean
fluxes were used for the truth and the prior in all OSSE sim-
ulations (Sect. 2.3) for simplicity and because the terrestrial
NEE fluxes are the focus of this work. It is noteworthy that
the assimilation of land or ocean data do not in fact produce
substantial deviations from the truth over the TransCom-3
oceanic regions (Fig. S2). For all OSSE simulations, an as-
similation window of 18 months covering the period from
1 August 2014 to 31 January 2016 was applied. NEE/oceanic
fluxes are optimized for every month of the assimilation win-
dow at each surface grid box in the GEOS-Chem model. The
analysis of prior and posterior NEE fluxes is for all months
in 2015, treating the other months as spin-up and spin-down
periods.
2.4.2 Initial CO2concentrations
We use identical initial atmospheric concentrations of CO2
on 1 August 2014 for (1) the GEOS-Chem forward model
simulations generating synthetic XCO2using the true NEE
fluxes (Sect. 2.4.3) and (2) for all of the OSSEs using vari-
able prior biosphere model predictions. The initial CO2con-
centrations were generated by running the GEOS-Chem for-
ward model for 2 years using the true MsTMIP NEE and
other non-NEE CO2sources. The restart file used for this
2-year forward model run was taken from an earlier GEOS-
Chem model simulation constrained with in situ observations
(personal communication from Ray Nassar) in order to rep-
resent a realistic initial condition.
2.4.3 Synthetic OCO-2 retrievals
In this study, we used synthetic satellite data that are directly
representative of version 8 of the OCO-2 product. The OCO-
2 satellite sensor is in sun-synchronous polar orbit with a re-
peat cycle of 16 days and a local overpass time in the early af-
ternoon (Crisp et al., 2017). OCO-2 has three different view-
ing modes: soundings over land from LN and LG and over
oceans from OG. The algorithm from O’Dell et al. (2012) is
used to retrieve column-average dry air mole fraction of CO2
(XCO2) and other retrieval variables. The retrieval of OCO-
2 XCO2is expressed as Eq. (1) using a prior CO2vertical
profile (ca) and prior CO2column (XCO2(a) ) value:
XCO2=XCO2(a) +aT(cca),(1)
where cis the true profile of CO2concentrations and ais the
column averaging kernel. The individual soundings of OCO-
2 are at a fine resolution (24 spectra per second with < 3 km2
spatial resolution per sounding), leading to a very large data
volume (Crisp et al., 2017). This level of detail is lost when
the measurements are used in global inverse models with
much coarser spatial resolution, with numerous individual
OCO-2 soundings occurring in a single model grid box. In
addition, each sounding does not really provide an indepen-
dent piece of information to the inversion system due to spa-
tial and temporal error correlations. Therefore, we use 10 s
averages of the individual XCO2soundings similar to those
developed/described in Basu et al. (2018), although these
soundings are from an updated version with the file name
“OCO2_b80_10sec_WL04_GOOD_v2.nc”. These 10 s data
www.atmos-chem-phys.net/19/13267/2019/ Atmos. Chem. Phys., 19, 13267–13287, 2019
13272 S. Philip et al.: Prior biosphere model impact
contain averages of retrievals with a “good” quality flag and a
“warn level” from 1 to 4. Note that in this study we do not use
actual retrieved 10 s XCO2values for our OSSEs; however,
synthetic XCO2data were generated corresponding to the
spatiotemporal sampling frequency of version 8 OCO-2 10 s
data. These synthetic 10 s XCO2data are calculated with the
CO2concentration profile simulated using the GEOS-Chem
forward model (using MsTMIP as the NEE flux model). In
this manner we produced synthetic XCO2data that are rep-
resentative of the true atmosphere corresponding to the true
NEE fluxes used in this study. We archived these synthetic
XCO2data and applied them for all OSSEs conducted.
2.4.4 Inverse modeling approach
The transport of atmospheric CO2is simulated using GEOS-
Chem along with prescribed surface fluxes as input data
(see Table 1). Subsequently, the GEOS-Chem 4D-Var in-
verse modeling system assimilates synthetic OCO-2 XCO2
data to estimate the posterior/optimized monthly mean NEE
and oceanic fluxes at each surface grid box of the model.
The GEOS-Chem adjoint system applies the L-BFGS numer-
ical optimization algorithm with a no-bound option (Liu and
Nocedal, 1989). Posterior monthly averaged NEE/oceanic
fluxes (x) are inferred for each surface model grid box by
optimizing a vector of scaling factors σjfor the jth model
grid box as follows:
xj=σjxa,j ,(2)
where xarepresents the monthly mean prior NEE/oceanic
fluxes. Scaling factors are assumed to be unity at the first
iteration (that is, the prior NEE flux itself is used). The inver-
sion system, as described below, optimizes the scaling factor
applied to the monthly mean fluxes and the posterior scaling
factors are then used to scale prior fluxes to infer posterior
CO2fluxes.
For each iteration, the inversion system uses the forward
model simulated profiles of CO2concentrations mapped to
OCO-2 retrieval levels (f(x)) in each model grid box in
order to compare them with the synthetic OCO-2 observa-
tions (y). The observation operator (M) represents the model
simulated XCO2corresponding to each synthetic OCO-2 re-
trieval,
M=XCO2(a)+aT(f(x)ca)(3)
using XCO2(a)and cafrom the retrieval (see Eq. 1). The op-
timization approach used in this work defines the 4D-Var cost
function (J) as follows:
J(σ)=1
2Xi(Miyi)TR1
iMiyi
+1
2(σσa)TP1(σσa),(4)
where yiis the vector of synthetic OCO-2 XCO2data across
the assimilation window, with “i” being the number of XCO2
data. Furthermore, Rand Pare the observational error co-
variance matrix and prior error covariance matrix, respec-
tively.
2.4.5 Prior flux uncertainty
For a perfect optimization, the prior error covariance matrix
(P) assumed in the inversion should equal the uncertainty of
the prior model used. However, the estimation of prior error
statistics is a challenging task due to the lack of flux evalu-
ation data. Previous studies have used a range of techniques
to characterize prior error covariance such as developing the
full error covariance matrix with assumed error correlations
(Basu et al., 2013), using a fraction of heterotrophic respi-
ration (Basu et al., 2018), conducting Monte Carlo simula-
tions (Liu et al., 2014), using standard deviations and abso-
lute differences between several different prior flux models
(Baker et al., 2006a, 2010), using continuous in situ mea-
surements of CO2flux compared to model simulations to
inform prior errors (Chevallier et al., 2006), and applying
globally uniform prior flux uncertainty values to satisfy the
posterior χ2(normalized cost function)=1 criteria (Deng et
al., 2014). During this study, a 1σstandard deviation (SD)
of the four prior biosphere models (see Sect. 3.1 for the de-
scription of SD values) is considered to be the measure of
uncertainty in the prior knowledge of bottom-up model pre-
dicted biospheric CO2fluxes. The SD of the four prior NEE
estimates is applied in the prior error covariance matrix and
no spatial or temporal correlations are taken into account.
This assumption is reasonable, as optimized fluxes are at
coarse spatiotemporal scales (monthly mean fluxes at hori-
zontal resolutions of > 400 km2), and is representative of the
majority of inverse modeling studies assimilating CO2satel-
lite data (e.g., Baker et al., 2010; Liu et al., 2014; Deng et
al., 2014). Finally, as the inverse modeling system applied in
this work optimizes scaling factors, we use the square of the
fractional prior error in the Pmatrix, where the fractional er-
ror is calculated for each individual prior model as the SD
of the four prior models divided by the absolute value of the
NEE magnitude. For generating prior error for the oceanic
fluxes, we follow the same method we adopted for generating
prior errors in NEE. The SD of four different state of the art
oceanic CO2flux datasets – NASA-CMS CO2oceanic flux
(from Bowman, 2017), CarbonTracker 2016 prior ocean data
(CT2016; http://carbontracker.noaa.gov, last access: 22 Au-
gust 2017), Takahashi et al. (2009), and Landschützer et al.
(2016, 2017) – was calculated to generate prior error values.
2.4.6 XCO2uncertainty
As described in Sect. 2.4.3, the synthetic XCO2used in this
study is calculated at the spatiotemporal sampling frequency
of the OCO-2 10 s average dataset. Although we use syn-
thetic XCO2, we apply the same observation error statistics
generated with the actual OCO-2 XCO210 s dataset in or-
Atmos. Chem. Phys., 19, 13267–13287, 2019 www.atmos-chem-phys.net/19/13267/2019/
S. Philip et al.: Prior biosphere model impact 13273
der to develop the observational error covariance matrix (R).
The final observation error for the 10 s average data is gener-
ated as a quadratic sum of the retrieval error from individual
OCO-2 soundings, 10 s averaging error, and a “model repre-
sentation error” as described in Basu et al. (2018). Similar
to the treatment of prior error statistics, we neglect observa-
tion error correlations and assume a diagonal observational
error covariance. No random perturbations were added to the
synthetic XCO2used in this study, as the goal of this work
was not to quantify the analytical posterior flux uncertainty
but instead to analyze the spread among posterior NEE esti-
mates. We note that other inverse modeling groups assimilat-
ing OCO-2 data also do not add random perturbations to the
data and use the same error statistics generated along with
the OCO-2 10 s product that are applied in this study (e.g.,
Basu et al., 2018; Crowell et al., 2019). During this study
the square of the observation error for the 10 s average data
was applied as the diagonal in the Rmatrix. From our ini-
tial OSSE tests it was determined that the use of 10 s error
statistics led to posterior χ2(normalized cost function) val-
ues that were much lower than unity. Therefore, we divided
the 10 s error values uniformly by a factor of 5 to approxi-
mately satisfy the χ2=1 criteria for all of the OSSEs. This
deflation procedure reduced the average 10 s observational
error values for LN+LG (OG) data from 1.5 (0.9) to
0.3 (0.2) ppm. These procedures give more confidence
to the observational data and lead to results in this study
which can be assumed as the lower limit of the impact of
prior model flux and uncertainty statistics in inverse model
estimates.
2.4.7 Evaluation of OSSE results
During this study, the posterior NEE values from the OSSEs
are compared to the true fluxes to assess accuracy and are
also intercompared to assess the spread in posterior estimates
due to the assumed prior NEE and prior error statistics. The
primary statistical parameters used to evaluate the spread in
posterior NEE fluxes are the SD (hereafter the term “spread”
will be used to represent SD) and range (difference between
maximum and minimum in NEE). The SD/spread and range
of posterior NEE estimates, when using the different prior
models, will provide an understanding of the spatiotemporal
residual impact of the prior models in top-down estimates of
global/regional NEE fluxes when assimilating OCO-2 data.
In order to evaluate the spatiotemporal variability of
prior and posterior regional NEE fluxes, we aggregate in-
dividual model grid boxes to the TransCom-3 land regions
(TransCom-3 regions illustrated in Fig. 1). To further inter-
pret the OSSE results, we produce additional classifications
of three broad hemisphere-scale TransCom-3 land regions:
northern land (NL), tropical land (TL), and southern land
(SL). TL includes tropical South America, North Africa, and
tropical Asia; SL includes South American temperate, South
Africa, and Australia; and NL includes the other five land re-
gions. The evaluation of the SD and range of prior model and
posterior/optimized NEE fluxes were calculated for the 11
individual TransCom-3 regions, for the 3 hemisphere-scale
TransCom-3 regions, and globally. Throughout the paper,
seasonally averaged prior and posterior NEE fluxes will be
discussed and these seasons are presented with respect to the
Northern Hemisphere.
2.4.8 Pseudo data assimilations
In order to test our OSSE framework, we first run four
“pseudo” (hereafter quotation marks will be omitted) exper-
iments by conducting inverse modeling studies using pseudo
surface observations. These test OSSE simulations were con-
ducted for 5-month assimilation windows for two separate
seasons, November 2014 to March 2015 (analysis for win-
ter, DJF) and May to September 2015 (analysis for summer,
JJA), using all four prior model NEE values separately. Simu-
lated hourly concentrations of CO2for all surface grid boxes
of GEOS-Chem are taken as pseudo surface observations. In
order to check whether the model framework can converge to
the truth, a simple controlled experiment was performed as-
suming a very small observational data uncertainty (0.001 %)
and with the prior flux uncertainty set equal to the absolute
magnitude of the truth – prior NEE (divided by the absolute
value of the NEE magnitude for that respective prior model).
The robustness of the flux inversions conducted in the sub-
sequent sections is validated by the results of these pseudo
tests. Figure S3 shows the results of the four pseudo tests us-
ing the four different NEE flux model predictions as the prior
information. From this figure it is apparent that, regardless of
the prior NEE assumed, posterior NEEs were able to repro-
duce the truth with near perfect accuracy for all TransCom-3
regions, with the range between the posterior NEEs typically
approaching 0 PgC yr1. This test also demonstrates that a
“perfect” assimilation (using uniform and dense surface data
coverage, highly accurate data, and known/loose prior un-
certainty) is almost insensitive to the prior assumed. Having
tested the robustness of our inversion setup, we feel confi-
dent in presenting the output from our OSSE framework, us-
ing synthetic OCO-2 remote-sensing data, in the following
sections.
3 Results and discussion
3.1 Prior NEE fluxes
Figure 2 shows the seasonally averaged multi-model-mean
and SD of the NEE fluxes from the four prior biosphere mod-
els used in the OSSE simulations (individual prior model
and true seasonally averaged NEE fluxes are displayed in
Fig. S4). This figure shows the main features of NEE that
are expected, such as the Northern Hemispheric fall/winter
maximum in Reand summer maximum in GPP due to the
seasonality of photosynthesis and respiration. Figure 2 also
www.atmos-chem-phys.net/19/13267/2019/ Atmos. Chem. Phys., 19, 13267–13287, 2019
13274 S. Philip et al.: Prior biosphere model impact
Figure 1. The TransCom-3 land region boundaries used to aggregate CO2fluxes for evaluation.
shows the spread of the four prior model fluxes (used as prior
uncertainty), which is typically highest over the temperate re-
gions of the Northern Hemisphere in the spring and summer
and high in the tropical regions during all seasons (the left
panels of Fig. 3 shows a spatial map of the corresponding
range of the four prior model fluxes). Furthermore, Fig. S5
shows the time series of monthly mean prior NEE fluxes, and
the corresponding prior error uncertainty (error bars), for the
individual prior models. From this figure it can be seen that
the 1σprior uncertainties for the global land are 50 %–
70 % of the total NEE for the different prior models, with
variability among other TransCom-3 land regions.
Table 3 displays the statistics of the prior NEE multi-
model-mean and SD and range for the 11 individual
TransCom-3 land regions. The SD values for prior NEE
fluxes range from 20 % to frequently >100 % of the multi-
model NEE mean for different regions/seasons with signifi-
cant NEE fluxes (hereafter this refers to regions/seasons with
NEE flux 0.5 PgC yr1). When comparing the magnitude
of NEE between the four prior models, it can be seen (from
Table 3) that the range of the NEE values is large in some
regions (up to 6 PgC yr1). In general, all regions/seasons
tend to have at least an 1 PgC yr1range among the four
prior models, indicating the large diversity in NEE predicted
by current bottom-up biosphere models (Table 3). Figure S5
shows the time series of monthly mean NEE for individual
prior models averaged over the globe, hemispheric-scale land
regions (NL, TL, and SL), and the 11 individual TransCom-3
land regions. It can be seen from this figure that the major-
ity of the seasonality in the global NEE flux is controlled by
the NL regions. Figure S5 also shows that the spread in prior
NEE fluxes is larger in general for TL and SL regions com-
pared with the NL, except for the North American temper-
ate region. Furthermore, when focusing on individual mod-
els, differences in NEE seasonality are evident. The impact
of these differences among the four prior biospheric CO2
flux models on simulated XCO2and posterior estimates of
global/regional NEE fluxes is evaluated in the following sec-
tions.
3.2 Simulated XCO2
Figure 4 shows the number of observations sampled in the
OCO-2 LN and LG modes during the different seasons of
2015 summed in each model grid box. Large spatiotempo-
ral variability can be seen in the OCO-2 observation den-
sity, with the largest values over regions with minimal cloud
coverage (e.g., desert regions of North/South Africa, Middle
East, Australia, and so on.). The opposite is true for many
tropical regions (e.g., Amazon, central Africa, tropical Asia,
and so on.) where cloud occurrence is prominent and the
number of OCO-2 observations is lowest. From Fig. 4 it can
also be seen that the OCO-2 observation density has notice-
able seasonality. For example, during the winter months low
numbers of OCO-2 observations are made in the northern
boreal regions and the largest amounts are observed during
the summer. Furthermore, larger numbers of OCO-2 obser-
vations are made in the SL during the summer (JJA) com-
pared with other seasons.
The seasonally averaged multi-model-mean GEOS-Chem
simulated XCO2using the four prior model NEE fluxes is
shown in the right column of Fig. 4. The most notable fea-
ture in this figure is the Northern Hemisphere seasonality,
with higher XCO2concentrations in the winter months and
lowest XCO2values in the growing seasons of the summer.
Atmos. Chem. Phys., 19, 13267–13287, 2019 www.atmos-chem-phys.net/19/13267/2019/
S. Philip et al.: Prior biosphere model impact 13275
Table 3. Data corresponding to Fig. 6. Seasonally averaged NEE (PgC yr1) averaged over the 11 TransCom-3 land regions (refer to Fig. 1) for the MsTMIP (truth), multi-model prior
mean, and multi-model posterior mean (PgC yr1). The differences between the prior and posterior model NEE values are presented as SD (1σ) and range. Prior model values are
presented in standard font and posterior estimates are in bold. Seasons are represented as Winter (W), December–February; Spring (Sp), March–May; Summer (Su), June–August; and
Fall (F), September–November. The synthetic observations in these OSSE simulations correspond to the OCO-2 LN+LG observing modes.
NEE: truth NEE: mean NEE: standard deviation NEE: range
Region* W Sp Su F W Sp Su F W Sp Su F W Sp Su F
1 1.1 0.22.8 0.7 1.1 1.0 0.4 0.3 3.42.7 1.3 0.9 0.5 0.4 0.5 0.1 0.8 0.2 0.5 0.1 1.2 0.9 1.0 0.3 1.9 0.5 1.2 0.2
2 1.5 1.52.3 0.0 1.9 1.6 1.01.6 3.22.1 0.6 0.2 1.1 0.3 1.3 0.1 2.0 0.2 0.8 0.1 2.3 0.6 3.2 0.2 4.3 0.4 1.6 0.2
30.60.21.41.3 0.4 0.8 0.6 0.2 0.71.2 0.1 1.3 0.1 0.1 0.8 0.2 1.4 0.0 2.6 0.1 0.2 0.1 1.8 0.3 3.1 0.1 5.9 0.2
41.50.4 0.7 0.41.11.3 0.10.4 0.6 0.7 0.0 0.1 0.9 0.1 0.4 0.2 0.8 0.1 0.9 0.1 2.0 0.1 0.9 0.5 1.9 0.2 2.0 0.3
5 1.3 1.0 1.92.0 0.7 0.9 0.6 0.4 0.81.7 1.42.0 1.0 0.2 0.8 0.1 0.9 0.1 1.2 0.2 2.3 0.5 1.6 0.3 2.0 0.2 3.0 0.5
62.41.6 1.2 1.2 0.91.9 1.11.8 0.3 0.8 0.7 1.2 0.5 0.2 1.0 0.1 1.2 0.2 0.7 0.1 1.2 0.3 2.2 0.2 2.6 0.5 1.7 0.2
7 1.2 1.0 4.6 1.2 1.5 1.2 0.9 0.9 5.64.8 2.0 1.3 0.9 0.7 1.0 0.1 1.6 0.1 0.6 0.2 1.8 1.4 2.2 0.3 3.3 0.2 1.3 0.6
8 1.4 0.0 1.90.4 1.4 1.3 1.00.2 2.32.3 0.1 0.7 1.3 0.3 0.9 0.1 2.5 0.2 1.7 0.1 3.0 0.6 2.1 0.3 5.7 0.5 3.7 0.2
90.1 0.4 0.30.5 0.1 0.3 0.0 0.6 0.40.1 0.50.4 0.4 0.2 0.4 0.1 0.4 0.3 0.8 0.0 0.8 0.5 0.7 0.2 0.8 0.5 1.9 0.1
10 0.60.60.20.70.10.6 0.0 0.7 0.1 0.4 0.10.8 0.4 0.1 0.3 0.1 0.3 0.1 0.6 0.1 0.9 0.2 0.6 0.2 0.7 0.3 1.2 0.3
11 2.8 1.73.2 1.6 2.3 2.6 0.91.2 3.73.1 1.6 1.3 0.4 0.6 1.2 0.2 1.5 0.1 0.6 0.1 0.9 1.3 2.5 0.5 3.5 0.3 1.5 0.3
TransCom-3 region name and location displayed in Fig. 1.
www.atmos-chem-phys.net/19/13267/2019/ Atmos. Chem. Phys., 19, 13267–13287, 2019
13276 S. Philip et al.: Prior biosphere model impact
Figure 2. Prior multi-model (NASA-CASA, CASA-GFED, SiB-4,
and LPJ biosphere models) seasonally averaged NEE (gC m2d1;
left column) and NEE SD (standard deviation) (gC m2d1; right
column) for the year 2015.
Seasonality in model-predicted XCO2values is also evident
in the TL and SL, with largest values in the autumn and low-
est values in the spring. Figure 5 (left column) shows the
range of XCO2values simulated using the four prior model
NEE fluxes. The differences between individual model sim-
ulations of XCO2values deviated among themselves by up
to 10 ppm. These large differences in XCO2values across
the four different prior NEE flux models show that the choice
of prior NEE has a large impact on simulated XCO2values.
3.3 Optimized global NEE fluxes
From Table 1 it can be seen that annual global mean pos-
terior NEE flux, when using the different prior models and
assimilating synthetic LN+LG OCO-2 XCO2, ranges from
4.11 to 4.36 PgC yr1, which is generally close in mag-
nitude to the true flux of 4.31 PgC yr1. Although these
posterior NEEs generally converged to the truth, there are
Figure 3. Seasonally averaged NEE range (gC m2d1) of the four
prior biosphere models (NASA-CASA, CASA-GFED, SiB-4, and
LPJ) (left column) and posterior estimates (right column) from the
OSSE simulations. The synthetic observations in these OSSE sim-
ulations correspond to the OCO-2 LN+LG observing modes.
some remaining differences, with an annual global mean
posterior NEE range of 0.25 PgC yr1(6 % of the multi-
model-mean posterior NEE; Table 1). From the results of the
OSSE simulations, it was found that the spread and range in
XCO2simulated using the optimized posterior NEE fluxes
was greatly reduced compared with the spread in XCO2sim-
ulated using prior NEE fluxes. This is evident from the right
column of Fig. 5 where XCO2values simulated using pos-
terior NEE fluxes differ among themselves by <0.5 ppm,
which is greater than an order of magnitude lower, on av-
erage, than the spread among XCO2values simulated using
prior NEEs.
Figure 3 shows the spatial distribution of the range of prior
and posterior NEEs. As expected, the range in optimized pos-
terior NEE flux estimates starting from the four separate prior
models was substantially reduced compared with the spread
in prior NEE fluxes. However, the posterior NEE fluxes for
Atmos. Chem. Phys., 19, 13267–13287, 2019 www.atmos-chem-phys.net/19/13267/2019/
S. Philip et al.: Prior biosphere model impact 13277
Figure 4. Total number of OCO-2 LN and LG XCO2observations
(left column) and the corresponding seasonally averaged multi-
model (NASA-CASA, CASA-GFED, SiB-4, and LPJ biosphere
models) mean GEOS-Chem-simulated prior XCO2(right column)
in 2015.
individual surface grid boxes of the model still depict some
residual range among the posteriors, with the largest residu-
als being found across South America and South Africa in all
seasons and in temperate regions of the Northern Hemisphere
in the spring months. As shown in Fig. 3, the geographical
pattern of the range of prior and posterior NEEs does not in-
dicate any noticeable correlations. From comparing Figs. 3
and 5, it is apparent that the spread in posterior XCO2is
significantly reduced in all regions of the globe compared
with prior model simulations; however, while posterior NEE
values are reduced compared to the prior, noticeable resid-
ual spread remains in some regions. This emphasizes the fact
that the OSSEs successfully converge to match the synthetic
OCO-2 XCO2values by optimizing NEE in different ways
depending on the prior NEE model used. The following sec-
tions investigate the regional differences in posterior NEE
Figure 5. Seasonally averaged XCO2range from GEOS-Chem
forward model simulations using the four prior biosphere mod-
els (NASA-CASA, CASA-GFED, SiB-4, and LPJ) (left column)
and the corresponding posterior estimates (right column) from the
OSSE simulations. The synthetic observations in these OSSE sim-
ulations correspond to the OCO-2 LN+LG observing modes.
estimates due to the residual impact of prior biospheric CO2
flux predictions.
3.4 Optimized regional NEE fluxes
Figure 6 shows the seasonally averaged true, prior, and pos-
terior NEE flux values for the 11 individual TransCom-3 land
regions (with detailed statistics in Table 3 and monthly mean
time series in Fig. S6). The first thing noticed from this fig-
ure is that all posterior NEE values, using variable priors,
tend to reproduce the truth in most TransCom-3 land regions.
From Fig. 6 it can also be seen that the assimilation of syn-
thetic OCO-2 LN+LG XCO2retrievals resulted in a large
reduction in the range among the four modeled NEE val-
ues (Table 3 shows the corresponding SD values). The re-
duction in the SD of NEE (also known as uncertainty reduc-
www.atmos-chem-phys.net/19/13267/2019/ Atmos. Chem. Phys., 19, 13267–13287, 2019
13278 S. Philip et al.: Prior biosphere model impact
tion) in most regions/seasons, calculated as 100×(1(pos-
terior NEE SD)/(prior NEE SD)), is generally > 70 % and
up to 98 %. However, the range of seasonal mean posterior
NEEs over individual TransCom-3 regions is still as large as
1.4 PgC yr1when applying different prior NEEs, with the
largest ranges occurring in northern boreal regions (North
America boreal, Eurasian boreal, and Europe) in the winter
months. During the spring and summer months, regions in
the TL (e.g., tropical Asia) and SL (e.g., South American
temperate and South Africa) have ranges in posterior NEEs
up to 0.5 PgC yr1. The larger residual range among pos-
terior NEE estimates for winter months in northern boreal
regions is likely due to the insufficient OCO-2 observations
during this time (see Fig. 4), while the larger range in the TL
and SL regions is due to differences in the priors (see Fig. 2).
This demonstrates that the impact from the prior model has
regional and seasonal variability depending on (1) the spa-
tiotemporal flux variabilities inherent in prior NEEs and
(2) the observation density and coverage of synthetic OCO-
2 data. Figure 6 and Table 3 show that the seasonally aver-
aged posterior NEE spread varies from 10 % to 50 % of
the multi-model-mean for different TransCom-3 land regions
with significant NEE fluxes. When evaluating this residual
spread in posterior NEEs on a global average, seasonally av-
eraged values ranged from 10 % (JJA) to 20 % (DJF) of
the posterior NEE mean. These statistics reveal that the im-
pact of prior models lead to a much larger spread/range for
regional/seasonal posterior fluxes (up to 50 %) compared
with annual global averaged values (6%). This emphasizes
that while OCO-2 observations on average constrain global,
hemispheric, and regional biospheric fluxes, noticeable resid-
ual differences in posterior NEE flux estimates remain due to
the choice of prior model values. Overall, the results of this
evaluation suggest that when intercomparing inverse model
results assimilating similar OCO-2 observational data, dif-
ferences in posterior NEE in regions with significant NEE
fluxes could vary by up to 50 % when using different prior
flux assumptions.
3.5 Impact of prior uncertainty
Results of this study have demonstrated the sensitivity of
posterior NEE estimates to prior NEE flux assumptions. In
this section, the sensitivity of posterior NEE estimates to the
assumed prior uncertainty is tested, when assimilating syn-
thetic OCO-2 LN+LG XCO2observations. The general im-
portance of prior uncertainty values is highlighted in the TL
regions. In these regions the largest differences in prior mod-
els are calculated, thus the largest prior uncertainty is as-
signed, resulting in larger deviations from the prior and pos-
terior NEE spread similar to other TransCom-3 land regions
(see Fig. 6). In order to quantify the sensitivity of posterior
NEE to prior uncertainty statistics, a single prior NEE flux
model (CASA-GFED) is applied in the OSSE framework,
with variable prior flux uncertainty assumptions. Two addi-
tional OSSE simulations (in addition to the baseline simu-
lations using the SD of the four prior models as the prior
uncertainty; see right column of Fig. 2 for SD maps) are per-
formed using prior NEE magnitudes from CASA-GFED and
setting the prior uncertainty uniformly as 10 % and 100 %
of the CASA-GFED NEE values. Figure 7 shows the results
of these additional OSSE simulations over the TransCom-3
land regions. From this figure it can be seen that the range of
seasonal mean posterior NEEs over individual TransCom-3
regions vary from 0.1 to > 1 PgC yr1when applying vari-
able prior error assumptions. Seasonally averaged posterior
NEE SD varies from 10 % to 50 % of the multi-model-
mean for different TransCom-3 land regions with significant
NEE fluxes. On a global average, the seasonal-average SD
values range from 15 % (JJA) to 30 % (DJF) of the pos-
terior NEE mean. Note that these posterior NEE SD/range
values are similar here to the baseline OSSEs conducted by
changing prior NEE flux magnitudes (see Fig. 6). However,
when comparing Figs. 7 and 6 (and Fig. S7), it is noticed
that posterior NEE estimates are more sensitive to prior error
assumptions compared with prior flux values in some sea-
sons/regions of TL and SL (e.g., northern Africa, southern
Africa, and South America temperate). It appears that NEE
estimates during this study are more sensitive to prior er-
ror assumptions when sufficient observations are available
and large differences between the prior and truth are present.
Also, from Fig. 7 it can be seen that prior uncertainty as-
sumptions in the baseline runs (using SD of prior models)
and the assumption of 100 % prior uncertainty tend to re-
produce the truth more accurately than NEE estimates us-
ing 10 % prior error. Overall, the results demonstrate that the
posterior NEE fluxes over TransCom-3 land regions are in
general similarly sensitive (up to 50 %) to the specifica-
tion of prior flux uncertainties and the choice of bottom-up
prior biospheric NEE model estimates.
3.6 OCO-2 ocean data
This portion of the study investigates the impact of assim-
ilating OCO-2 OG XCO2data on posterior NEE flux esti-
mates in our OSSE framework. To do this, four additional
OSSE simulations were conducted with the four prior model
NEEs when only assimilating synthetic OG retrievals (in-
stead of LN+LG) in the inversions (everything else remains
the same as in the baseline simulations). Figure 8 shows the
results of these four additional OSSE simulations averaged
over the TransCom-3 land regions. From these simulations
it can be seen that OCO-2 OG indeed reduces the range in
posterior NEE flux estimates, when applying different pri-
ors, compared with prior model predictions, and can gener-
ally reproduce the truth. On average, the spread in posterior
NEE fluxes is 20 % to 50 % of the multi-model-mean
for different TransCom-3 land regions with significant NEE
fluxes. As expected, the comparison of Figs. 6 and 8 suggests
that LN+LG data are better able to constrain biospheric CO2
Atmos. Chem. Phys., 19, 13267–13287, 2019 www.atmos-chem-phys.net/19/13267/2019/
S. Philip et al.: Prior biosphere model impact 13279
Figure 6. Seasonally averaged NEE averaged over the 11 TransCom-3 land regions from MsTMIP (truth) vs. the prior biosphere models
(NASA-CASA, CASA-GFED, SiB-4, and LPJ) (left column), posterior estimates (middle column) from the OSSE simulations, and the
corresponding range of prior and posterior NEE estimates (right column). The synthetic observations in these OSSE simulations correspond
to the OCO-2 LN+LG observing modes. Detailed statistics of the truth, multi-model means of prior and posterior NEE estimates, standard
deviations, and ranges displayed in this figure are listed in Table 3.
fluxes compared with OG data, as the spread among the pos-
teriors is generally lower in assimilations using only LN+LG
data (70 % lower on a global average) compared with as-
similations using only OG data. However, there were some
cases where OSSE simulations using OCO-2 OG data alone
did in fact result in slightly lower posterior NEE spreads in
some TransCom-3 land regions compared with LN+LG as-
similation runs (e.g., northern boreal regions during summer
months and Australia during winter months). Overall, our
OSSE simulations using the OCO-2 OG data demonstrate
the importance of these oceanic retrievals to constrain land
NEE fluxes, as the posterior NEE range is much lower than
prior NEE estimates (see Fig. 8). This generally agrees with
previous studies that demonstrated the importance of satel-
lite data over the ocean in constraining NEE fluxes over land
regions (e.g., Deng et al., 2016).
4 Conclusions
To the best of our understanding, this is the first study directly
quantifying the impact of different prior global land bio-
sphere models on the estimate of terrestrial CO2fluxes when
assimilating OCO-2 satellite observations. We conducted a
series of OSSEs that assimilated synthetic OCO-2 observa-
tions applying four state of the art biospheric CO2flux mod-
els as the prior information. These controlled experiments
were designed to systematically assess the impact of prior
NEE fluxes and the impact of prior error assumptions on top-
down NEE estimates using OCO-2 data. The OSSEs incor-
porated NEE fluxes from the NASA-CASA, CASA-GFED,
SiB-4, and LPJ biosphere models as prior estimates and vari-
able prior flux error assumptions.
We found that the assimilation of synthetic OCO-2 XCO2
retrievals resulted in posterior monthly/seasonal NEE esti-
mates that generally reproduced the assumed true NEE glob-
www.atmos-chem-phys.net/19/13267/2019/ Atmos. Chem. Phys., 19, 13267–13287, 2019
13280 S. Philip et al.: Prior biosphere model impact
Figure 7. Seasonally averaged NEE averaged over the 11 TransCom-3 land regions from MsTMIP (truth) vs. CASA-GFED prior biosphere
model (left column), posterior estimates with the three different prior uncertainties (middle column), and the corresponding range of posterior
NEE (right column). The synthetic observations in OSSE simulations correspond to the OCO-2 LN+LG observing modes. Detailed statistics
of the truth, prior, multi-model mean of posterior NEE estimates, standard deviations (SD), and ranges displayed in this figure are listed in
Table S1 in the Supplement.
ally and regionally. However, spread in posterior NEE ex-
ists in regions during seasons with poor data coverage, such
as the northern boreal regions and some of the tropical and
southern hemispheric regions (e.g., South American temper-
ate, South Africa, tropical Asia). This spread among poste-
rior NEEs is likely due to the insufficient OCO-2 observa-
tions during winter over northern boreal regions and the large
range among the priors in some of the northern boreal, tropi-
cal, and southern hemispheric regions. Residual spread from
10 % to 50 % in seasonally averaged posterior NEEs in
TransCom-3 land regions with significant NEE flux were cal-
culated due to using different prior models in inverse model
simulations. We also found similar spreads in the magnitudes
of posterior NEEs by conducting additional OSSEs using a
single prior NEE flux model with variable prior flux uncer-
tainty assumptions. While the spread in posterior NEE esti-
mates, when using variable prior error statistics, was simi-
lar to when applying variable NEE flux models, the impact
was larger in some seasons in the TL and SL regions. We
determined that while OCO-2 observations constrain global,
hemispheric, and regional biospheric fluxes on average, no-
ticeable residual differences (up to 20 %–30 % globally
and 50 % regionally) in seasonally averaged posterior NEE
flux estimates remain that were caused by the choice of prior
model values and the specification of prior flux uncertainties.
There have been previous studies that have investigated
similar scientific objectives, such as the impact of prior un-
certainties on inverse model estimates of NEE (Gurney et al.,
2003; Chevalier et al., 2005; Baker et al., 2006a, 2010). The
sensitivity of CO2flux inversions to the specification of prior
flux information was first assessed by Gurney et al. (2003)
using ground-based in situ data. One main conclusion from
Gurney et al. (2003) was that CO2flux estimates were sen-
sitive to the prior flux uncertainty over regions with limited
Atmos. Chem. Phys., 19, 13267–13287, 2019 www.atmos-chem-phys.net/19/13267/2019/
S. Philip et al.: Prior biosphere model impact 13281
Figure 8. Seasonally averaged NEE averaged over the 11 TransCom-3 land regions from MsTMIP (truth) vs. the prior biosphere models
(NASA-CASA, CASA-GFED, SiB-4, and LPJ) (left column), posterior estimates (middle column) from the OSSE simulations, and the
corresponding range of prior and posterior NEE (right column). The synthetic observations in these OSSE simulations correspond to the
OCO-2 OG observing mode. Detailed statistics of the truth, multi-model means of prior and posterior NEE estimates, standard deviations,
and ranges displayed in this figure are listed in Table S2.
observations and insensitive over data-rich regions. Cheval-
lier et al. (2005) suggested the importance of an accurate
formulation of prior flux uncertainty by conducting 4D-Var
assimilation of satellite column retrievals of CO2. Baker et
al. (2010) investigated the importance of assumed prior flux
uncertainties by conducting sensitivity tests that mistuned
the assimilations by using incorrect prior flux errors. Finally,
Baker et al. (2006a, 2010) suggested the need for realistic
prior models in the 4D-Var assimilations using OCO syn-
thetic satellite CO2data. The results of this research are gen-
erally consistent with the findings of these past studies. How-
ever, in comparison with these previous efforts, our study is a
step forward, because we quantify the specific impact of prior
model NEE spatiotemporal magnitude and prior uncertain-
ties in optimizing regional and seasonal NEEs using satel-
lite data in a more controlled manner by applying an OSSE
framework.
As explained earlier in this study, estimates of surface CO2
fluxes from numerous inversion systems in the OCO-2 model
intercomparison project (MIP) ensemble model framework,
using identical OCO-2 observations, result in different opti-
mized/posterior regional NEE fluxes (Crowell et al., 2019).
This inverse model variance can be due to numerous fac-
tors (e.g., model transport, inversion methods, observation
errors, and so on.) including prior model mean and uncer-
tainty estimates. In order to estimate the amount of vari-
ance in the results of posterior NEE values from the OCO-2
MIP which could be due to prior flux estimates, we com-
pare our OSSE derived residual posterior NEE range (using
LN+LG) to the range in the posterior Level-4 OCO-2 flux
data (using both LN and LG) (https://www.esrl.noaa.gov/
gmd/ccgg/OCO2/index.php, last access: 31 October 2018)
in each TransCom-3 region. This comparison suggests that
prior NEE and uncertainty statistics could contribute 10 %–
www.atmos-chem-phys.net/19/13267/2019/ Atmos. Chem. Phys., 19, 13267–13287, 2019
13282 S. Philip et al.: Prior biosphere model impact
30 % (average 20 %) of the annually averaged NEE vari-
ance calculated for each TransCom-3 region in the OCO-2
Level-4 MIP flux data. Comparing this contribution of prior
model impact to the OSSE study by Basu et al. (2018), which
calculated the impact of atmospheric transport on posterior
NEE estimates when assimilating OCO-2 observations, this
contribution is 50 % less than the impact of atmospheric
transport. From our study and Basu et al. (2018) it is esti-
mated that the combination of prior flux/uncertainty assump-
tions and atmospheric transport could contribute on average
50 % of the annually averaged posterior NEE variance of
the OCO-2 MIP study.
The results of this study suggest the need to be aware of the
residual impact from prior assumptions for CO2global flux
inversions, especially for regions and times (1) where current
bottom-up biosphere models diverge greatly and (2) without
sufficient observational coverage from spaceborne platforms.
For example, larger spread in posterior NEE estimates were
calculated in portions of the northern boreal regions that tend
to have insufficient satellite data coverage and moderate dif-
ferences among prior biosphere models. In addition to these
northern boreal regions, tropical and southern hemispheric
regions with large spread among prior biosphere models,
which are assigned higher prior uncertainty values resulting
in largely reduced spreads in posterior NEE estimates (large
deviation from the prior), still have residual impact from the
prior NEE predictions regardless of the fact that OCO-2 data
coverage is dense in these regions. Results of this study also
indicate that in some regions/seasons of the TL and SL, in-
verse model estimates of NEE can be more sensitive to prior
error statistics compared with prior flux values. Overall, in
data-poor regions/times, posterior estimates from inversion
techniques relying on Bayesian statistics can result in similar
estimates to the prior flux, although with some improvements
over broader regions. Additionally, in regions/seasons where
uncertainty in NEE fluxes are large (e.g., in the TL where
prior model NEE differences are large), inverse model esti-
mates, applying large prior uncertainty values, will still have
some residual impact from the choice of prior NEE flux. Fi-
nally, care should be given when interpreting flux estimates
constrained with real OCO-2 satellite data over some of the
regions identified in this study as it is suggested here that
residual differences (up to 20 %–30 % globally and 50 %
regionally) in seasonal posterior NEE flux estimates can be
produced by the choice of prior model values and the speci-
fication of prior flux uncertainties. Finally, the results of this
study suggest that multi-inverse model intercomparison stud-
ies should consider the differences in posterior NEE flux esti-
mates caused by variable prior fluxes and error statistics used
in different models.
Code and data availability. The forward and inverse model simu-
lations for this work were performed using the GEOS-Chem model
which is publicly available for free download at http://wiki.seas.
harvard.edu/geos-chem/index.php/GEOS-Chem_Adjoint (Henze et
al., 2007; Liu et al., 2014; last access: 27 January 2017). The 10 s
OCO-2 data used to produce synthetic observations during this
study are stored on the authors data storage system and are available
upon request. All OCO-2 10 s files are also available upon request
from the OCO-2 Science Team, and individual OCO-2 sounding
data can be downloaded from https://co2.jpl.nasa.gov/\#mission=
OCO-2 (Eldering et al., 2017a).
Supplement. The supplement related to this article is available on-
line at: https://doi.org/10.5194/acp-19-13267-2019-supplement.
Author contributions. SP and MJ designed the methods and exper-
iments presented in the study and analyzed the results. CP, VG, DB,
KH, and BP were instrumental in providing biosphere model and
OCO-2 data and guidance when applying these products. DH, JL,
and DB provided components implemented in the modeling frame-
work applied during this study. Finally, SP prepared the paper with
contributions from all coauthors.
Competing interests. The authors declare that they have no conflict
of interest.
Acknowledgements. Resources supporting this work were provided
by the NASA High-End Computing Program through the NASA
Advanced Supercomputing Division at the NASA Ames Research
Center. We thank the OCO-2 Science Team for providing the ver-
sion 8 OCO-2 product. We also thank the OCO-2 Flux Inversion
Team, the GEOS-Chem model developers, the CASA-GFED team,
and the NASA Carbon Monitoring System program for the free
availability of their products. CarbonTracker CT2016 prior and
posterior ocean fluxes were provided by National Oceanographic
and Atmospheric Administration’s Earth System Research Labora-
tory, Boulder, Colorado, USA, from http://carbontracker.noaa.gov
(last access: 22 August 2017). We are thankful to Sourish Basu,
Feng Deng, Ray Nassar, and Tom Oda for sharing data. We are
also grateful for the support from the Earth Science Division of the
NASA Ames Research Center. The views, opinions, and findings
contained in this report are those of the authors and should not be
construed as an official NASA or United States Government posi-
tion, policy, or decision.
Financial support. Sajeev Philip’s research was supported by an
appointment to the NASA Postdoctoral Program at the NASA
Ames Research Center, administered by the Universities Space Re-
search Association under contract with NASA. Sajeev Philip ac-
knowledges partial support from the NASA Academic Mission
Services by the Universities Space Research Association at the
NASA Ames Research Center. Daven K. Henze recognizes sup-
port from National Oceanic and Atmospheric Administration (grant
no. NA14OAR4310136).
Atmos. Chem. Phys., 19, 13267–13287, 2019 www.atmos-chem-phys.net/19/13267/2019/
S. Philip et al.: Prior biosphere model impact 13283
Review statement. This paper was edited by Joshua Fu and re-
viewed by four anonymous referees.
References
Baker, D. F., Doney, S. C., and Schimel, D. S.: Variational
data assimilation for atmospheric CO2, Tellus B, 58, 359–365,
https://doi.org/10.1111/j.1600-0889.2006.00218.x, 2006a.
Baker, D. F., Law, R. M., Gurney, K. R., Rayner, P., Peylin, P.,
Denning, A. S., Bousquet, P., Bruhwiler, L., Chen, Y. H., Ciais,
P., Fung, I. Y., Heimann, M., John, J., Maki, T., Maksyutov,
S., Masarie, K., Prather, M., Pak, B., Taguchi, S., and Zhu,
Z.: TransCom 3 inversion intercomparison: Impact of trans-
port model errors on the interannual variability of regional
CO2fluxes, 1988–2003, Global Biogeochem. Cy., 20, GB1002,
https://doi.org/10.1029/2004gb002439, 2006b.
Baker, D. F., Bösch, H., Doney, S. C., O’Brien, D., and Schimel,
D. S.: Carbon source/sink information provided by column CO2
measurements from the Orbiting Carbon Observatory, Atmos.
Chem. Phys., 10, 4145–4165, https://doi.org/10.5194/acp-10-
4145-2010, 2010.
Baker, I. T., Harper, A. B., da Rocha, H. R., Denning, A. S., Araújo,
A. C., Borma, L. S., Freitas, H. C., Goulden, M. L., Manzi, A.
O., Miller, S. D., Nobre, A. D., Restrepo-Coupe, N., Saleska, S.
R., Stöckli, R., von Randow, C., and Wofsy, S. C.: Surface eco-
physiological behavior across vegetation and moisture gradients
in tropical South America, Agr. Forest Meteorol., 182–183, 177–
188, https://doi.org/10.1016/j.agrformet.2012.11.015, 2013.
Basu, S., Guerlet, S., Butz, A., Houweling, S., Hasekamp, O., Aben,
I., Krummel, P., Steele, P., Langenfelds, R., Torn, M., Biraud, S.,
Stephens, B., Andrews, A., and Worthy, D.: Global CO2fluxes
estimated from GOSAT retrievals of total column CO2, At-
mos. Chem. Phys., 13, 8695–8717, https://doi.org/10.5194/acp-
13-8695-2013, 2013.
Basu, S., Baker, D. F., Chevallier, F., Patra, P. K., Liu, J., and
Miller, J. B.: The impact of transport model differences on
CO2surface flux estimates from OCO-2 retrievals of col-
umn average CO2, Atmos. Chem. Phys., 18, 7189–7215,
https://doi.org/10.5194/acp-18-7189-2018, 2018.
Bey, I., Jacob, D. J., Yantosca, R. M., Logan, J. A., Field,
B. D., Fiore, A. M., Li, Q. B., Liu, H. G. Y., Mickley,
L. J., and Schultz, M. G.: Global modeling of tropospheric
chemistry with assimilated meteorology: Model description
and evaluation, J. Geophys. Res.-Atmos., 106, 23073–23095,
https://doi.org/10.1029/2001JD000807, 2001.
Bowman, K. W.: Carbon Monitoring System Flux for Shipping,
Aviation, and Chemical Sources L4 V1, Greenbelt, MD, USA,
Goddard Earth Sciences Data and Information Services Center
(GES DISC), https://doi.org/10.5067/RLT7JTCRJ11M, 2017.
Bowman, K. W., Liu, J., Bloom, A. A., Parazoo, N. C., Lee, M.,
Jiang, Z., Menemenlis, D., Gierach, M. M., Collatz, G. J., Gur-
ney, K. R., and Wunch D.: Global and Brazilian carbon response
to El Niño Modoki 2011–2010, Earth Space Sci., 4, 637–660,
https://doi.org/10.1002/2016EA000204, 2017.
Byrne, B., Jones, D. B. A., Strong, K., Zeng, Z.-C., Deng, F.,
and Liu, J.: Sensitivity of CO2surface flux constraints to ob-
servational coverage, J. Geophys. Res.-Atmos., 122, 6672–6694,
https://doi.org/10.1002/2016JD026164, 2017.
Chevallier, F., Fisher, M., Peylin, P., Serrar, S., Bousquet, P.,
Bréon, F.-M., Chédin, A., and Ciais, P.: Inferring CO2sources
and sinks from satellite observations: Method and applica-
tion to TOVS data, J. Geophys. Res.-Atmos., 110, D24309,
https://doi.org/10.1029/2005JD006390, 2005.
Chevallier, F., Viovy, N., Reichstein, M., and Ciais, P.: On
the assignment of prior errors in Bayesian inversions of
CO2surface fluxes, Geophys. Res. Lett., 33, L13802,
https://doi.org/10.1029/2006GL026496, 2006.
Chevallier, F., Feng, L., Bösch, H., Palmer, P. I., and Rayner, P. J.:
On the impact of transport model errors for the estimation of CO2
surface fluxes from GOSAT observations, Geophys. Res. Lett.,
37, L21803, https://doi.org/10.1029/2010GL044652, 2010.
Chevallier, F., Palmer, P. I., Feng, L., Bösch, H., O’Dell, C.
W., and Bousquet, P.: Toward robust and consistent regional
CO2flux estimates from in situ and spaceborne measure-
ments of atmospheric CO2, Geophys. Res. Lett., 41, 1065–1070,
https://doi.org/10.1002/2013GL058772, 2014.
Corbett, J. J. and Koehler, H. W.: Updated emissions from
ocean shipping, J. Geophys. Res.-Atmos., 108, 4650,
https://doi.org/10.1029/2003JD003751, 2003.
Corbett, J. J. and Koehler, H. W.: Considering alternative input pa-
rameters in an activity-based ship fuel consumption and emis-
sions model: Reply to comment by Øyvind Endresen et al. on
“Updated emissions from ocean shipping”, J. Geophys. Res.-
Atmos., 109, D23303, https://doi.org/10.1029/2004JD005030,
2004.
Crisp, D., Pollock, H. R., Rosenberg, R., Chapsky, L., Lee, R. A.
M., Oyafuso, F. A., Frankenberg, C., O’Dell, C. W., Bruegge, C.
J., Doran, G. B., Eldering, A., Fisher, B. M., Fu, D., Gunson, M.
R., Mandrake, L., Osterman, G. B., Schwandner, F. M., Sun, K.,
Taylor, T. E., Wennberg, P. O., and Wunch, D.: The on-orbit per-
formance of the Orbiting Carbon Observatory-2 (OCO-2) instru-
ment and its radiometrically calibrated products, Atmos. Meas.
Tech., 10, 59–81, https://doi.org/10.5194/amt-10-59-2017, 2017.
Crowell, S., Baker, D., Schuh, A., Basu, S., Jacobson, A. R., Cheval-
lier, F., Liu, J., Deng, F., Feng, L., McKain, K., Chatterjee, A.,
Miller, J. B., Stephens, B. B., Eldering, A., Crisp, D., Schimel,
D., Nassar, R., O’Dell, C. W., Oda, T., Sweeney, C., Palmer, P. I.,
and Jones, D. B. A.: The 2015–2016 carbon cycle as seen from
OCO-2 and the global in situ network, Atmos. Chem. Phys., 19,
9797–9831, https://doi.org/10.5194/acp-19-9797-2019, 2019.
Deng, F., Jones, D. B. A., Henze, D. K., Bousserez, N., Bowman, K.
W., Fisher, J. B., Nassar, R., O’Dell, C., Wunch, D., Wennberg, P.
O., Kort, E. A., Wofsy, S. C., Blumenstock, T., Deutscher, N. M.,
Griffith, D. W. T., Hase, F., Heikkinen, P., Sherlock, V., Strong,
K., Sussmann, R., and Warneke, T.: Inferring regional sources
and sinks of atmospheric CO2from GOSAT XCO2data, At-
mos. Chem. Phys., 14, 3703–3727, https://doi.org/10.5194/acp-
14-3703-2014, 2014.
Deng, F., Jones, D. B. A. O’Dell, C. W., Nassar, R.,
and Parazoo N. C.: Combining GOSAT XCO2obser-
vations over land and ocean to improve regional CO2
flux estimates, J. Geophys. Res.-Atmos., 121, 1896–1913,
https://doi.org/10.1002/2015JD024157, 2016.
Denning, A. S., Collatz, G. J., Zhang, C., Randall, D. A., Berry,
J. A., Sellers, P. J., Colello, G. D., and Dazlich, D. A.:
Simulations of terrestrial carbon metabolism and atmospheric
www.atmos-chem-phys.net/19/13267/2019/ Atmos. Chem. Phys., 19, 13267–13287, 2019
13284 S. Philip et al.: Prior biosphere model impact
CO2in a general circulation model, Tellus B, 48B, 521–542,
https://doi.org/10.1034/j.1600-0889.1996.t01-2-00009.x, 1996.
Eldering, A., O’Dell, C. W., Wennberg, P. O., Crisp, D., Gunson, M.
R., Viatte, C., Avis, C., Braverman, A., Castano, R., Chang, A.,
Chapsky, L., Cheng, C., Connor, B., Dang, L., Doran, G., Fisher,
B., Frankenberg, C., Fu, D., Granat, R., Hobbs, J., Lee, R. A. M.,
Mandrake, L., McDuffie, J., Miller, C. E., Myers, V., Natraj, V.,
O’Brien, D., Osterman, G. B., Oyafuso, F., Payne, V. H., Pol-
lock, H. R., Polonsky, I., Roehl, C. M., Rosenberg, R., Schwand-
ner, F., Smyth, M., Tang, V., Taylor, T. E., To, C., Wunch, D.,
and Yoshimizu, J.: The Orbiting Carbon Observatory-2: first
18 months of science data products, Atmos. Meas. Tech., 10,
549–563, https://doi.org/10.5194/amt-10-549-2017, 2017a (data
available at: https://co2.jpl.nasa.gov/\#mission=OCO-2, last ac-
cess: 24 January 2018).
Eldering, A., Wennberg, P. O., Crisp, D., Schimel, D. S., Gun-
son, M. R., Chatterjee, A., Liu, J., Schwandner, F. M., Sun, Y.,
O’Dell, C. W., Frankenberg, C., Taylor, T., Fisher, B., Oster-
man, G. B., Wunch, D., Hakkarainen, J., Tamminen, J., and Weir,
B.: The Orbiting Carbon Observatory-2 early science investiga-
tions of regional carbon dioxide fluxes, Science, 358, eaam5745,
https://doi.org/10.1126/science.aam5745, 2017b.
Fisher, J. B., Sikka, M., Huntzinger, D. N., Schwalm, C., and Liu,
J.: Technical note: 3-hourly temporal downscaling of monthly
global terrestrial biosphere model net ecosystem exchange,
Biogeosciences, 13, 4271–4277, https://doi.org/10.5194/bg-13-
4271-2016, 2016a.
Fisher, J. B., Sikka, M., Huntzinger, D. N., Schwalm, C. R., Liu, J.,
Wei, Y., Cook, R. B., Michalak, A. M., Schaefer, K., Jacobson,
A. R., Arain, M. A., Ciais, P., El-masri, B., Hayes, D. J., Huang,
M., Huang, S., Ito, A., Jain, A. K., Lei, H., Lu, C., Maignan, F.,
Mao, J., Parazoo, N. C., Peng, C., Peng, S., Poulter, B., Ricciuto,
D. M., Tian, H., Shi, X., Wang, W., Zeng, N., Zhao, F., and Zhu,
Q.: CMS: Modeled Net Ecosystem Exchange at 3-hourly Time
Steps, 2004–2010, ORNL DAAC, Oak Ridge, Tennessee, USA,
https://doi.org/10.3334/ORNLDAAC/1315, 2016b.
Gurney, K. R., Law, R. M., Denning, A. S., Rayner, P. J., Baker, D.,
Bousquet, P., Bruhwiler, L., Chen, Y.-H., Ciais, P., Fan, S., Fung,
I. Y., Gloor, M., Heimann, M., Higuchi, K., John, J., Kowal-
czyk, E., Maki, T., Maksyutov, S., Peylin, P., Prather, M., Pak,
B. C., Sarmiento, J., Taguchi, S., Takahashi, T., and Yuen, C.-W.:
TransCom 3 CO2inversion intercomparison: 1. Annual mean
control results and sensitivity to transport and prior flux infor-
mation, Tellus B, 55B, 555–579, https://doi.org/10.1034/j.1600-
0889.2003.00049.x, 2003.
Harris, I., Jones, P. D., Osborn, T. J., and Lister, D. H.: Up-
dated high-resolution grids of monthly climatic observations
– the CRU TS3.10 Dataset, Int. J. Climatol., 34, 623–642,
https://doi.org/10.1002/joc.3711, 2013.
Haxeltine, A. and Prentice, I. C.: A general model for the light-
use efficiency of primary production, Funct. Ecol., 10, 551–561,
https://doi.org/10.2307/2390165, 1996.
Haynes, K. D., Baker, I. T., Denning, A. S., Stöckli, R., Schaefer,
K., and Lokupitiya, E.: Global Self-Consistent Carbon Flux and
Pool Estimates Utilizing the Simple Biosphere Model (SiB4),
Abstract B31F-01 presented at 2013 AGU Fall Meeting, AGU,
San Fransisco, CA, 9–13 December, 2013.
Heimann, M., Esser, G., Haxeltine, A., Kaduk, J., Kicklighter, D.
W., Knorr, W., Kohlmaier, G. H., McGuire, A. D., Melillo,
J., Moore III, B., Otto, R. D., Prentice, I. C., Sauf, W.,
Schloss, A., Sitch, S., Wittenberg, U., and Würth, G.: Evalu-
ation of terrestrial Carbon Cycle models through simulations
of the seasonal cycle of atmospheric CO2: First results of a
model intercomparison study, Global Biogeochem. Cy., 12, 1–
24, https://doi.org/10.1029/97GB01936, 1998.
Henze, D. K., Hakami, A., and Seinfeld, J. H.: Development of
the adjoint of GEOS-Chem, Atmos. Chem. Phys., 7, 2413–2433,
https://doi.org/10.5194/acp-7-2413-2007, 2007.
Houweling, S., Aben, I., Breon, F.-M., Chevallier, F., Deutscher,
N., Engelen, R., Gerbig, C., Griffith, D., Hungershoefer,
K., Macatangay, R., Marshall, J., Notholt, J., Peters, W.,
and Serrar, S.: The importance of transport model uncer-
tainties for the estimation of CO2sources and sinks using
satellite measurements, Atmos. Chem. Phys., 10, 9981–9992,
https://doi.org/10.5194/acp-10-9981-2010, 2010.
Houweling, S., Baker, D., Basu, S., Boesch, H., Butz, A., Cheval-
lier, F., Deng, F., Dlugokencky, E. J., Feng, L., Ganshin, A.,
Hasekamp, O., Jones, D., Maksyutov, S., Marshall, J., Oda, T.,
O’Dell, C. W., Oshchepkov, S., Palmer, P. I., Peylin, P., Poussi,
Z., Reum, F., Takagi, H., Yoshida, Y., and Zhuravlev, R.: An in-
tercomparison of inverse models for estimating sources and sinks
of CO2using GOSAT measurements, J. Geophys. Res.-Atmos.,
120, 5253–5266, https://doi.org/10.1002/2014JD022962, 2015.
Huntzinger, D. N., Post, W. M., Wei, Y., Michalak, A. M., West
T. O., Jacobson, A. R., Baker, I. T., Chen, J. M., Davis, K. J.,
Hayes, D. J., Hoffman, F. M., Jain, A. K., Liu, S., McGuire, A.
D., Neilson, R. P., Potter, C., Poulter, B., Price, D., Raczka, B.
M., Tian, H., Q., Thornton, P., Tomelleri, E., Viovy, N., Xiao, J.,
Yuan, W., Zeng, N., Zhao, M., and Cook, R.: North American
Carbon Program (NACP) regional interim synthesis: Terrestrial
biospheric model intercomparison, Ecol. Model., 232, 144–157,
https://doi.org/10.1016/j.ecolmodel.2012.02.004, 2012.
Huntzinger, D. N., Schwalm, C., Michalak, A. M., Schaefer, K.,
King, A. W., Wei, Y., Jacobson, A., Liu, S., Cook, R. B., Post,
W. M., Berthier, G., Hayes, D., Huang, M., Ito, A., Lei, H., Lu,
C., Mao, J., Peng, C. H., Peng, S., Poulter, B., Riccuito, D.,
Shi, X., Tian, H., Wang, W., Zeng, N., Zhao, F., and Zhu, Q.:
The North American Carbon Program Multi-Scale Synthesis and
Terrestrial Model Intercomparison Project – Part 1: Overview
and experimental design, Geosci. Model Dev., 6, 2121–2133,
https://doi.org/10.5194/gmd-6-2121-2013, 2013.
Huntzinger, D. N., Schwalm, C. R., Wei, Y., Cook, R. B., Micha-
lak, A. M., Schaefer, K., Jacobson, A. R., Arain, M. A., Ciais,
P., Fisher, J. B., Hayes, D. J., Huang, M., Huang, S., Ito, A.,
Jain, A. K., Lei, H., Lu, C., Maignan, F., Mao, J., Parazoo, N.
C., Peng, C., Peng, S., Poulter, B., Ricciuto, D. M., Tian, H., Shi,
X., Wang, W., Zeng, N., Zhao, F., Zhu, Q., Yang, J., and Tao,
B.: NACP MsTMIP: Global 0.5-degree Model Outputs in Stan-
dard Format, Version 1.0. ORNL DAAC, Oak Ridge, Tennessee,
USA, https://doi.org/10.3334/ORNLDAAC/1225, 2018.
IPCC: Climate Change 2014: Synthesis Report. Contribution of
Working Groups I, II and III to the Fifth Assessment Report
of the Intergovernmental Panel on Climate Change, edited by:
Pachauri, R. K. and Meyer, L. A., IPCC, Geneva, Switzerland,
151 pp., 2014.
Ito, A., Inatomi, M., Huntzinger, D. N., Schwalm, C., Michalak, A.
M., Cook, R., King, A. W., Mao, J., Wei, Y., Post, W. M., Wang,
W., Arain, M. A., Huang, S., Hayes, D. J., Ricciuto, D. M., Shi,
Atmos. Chem. Phys., 19, 13267–13287, 2019 www.atmos-chem-phys.net/19/13267/2019/
S. Philip et al.: Prior biosphere model impact 13285
X., Huang, M., Lei, H., Tian, H., Lu, C., Yang, J., Tao, B., Jain,
A., Poulter, B., Peng, S., Ciais, P., Fisher, J. B., Parazoo, N. C.,
Schaefer, K., Peng, C., Zeng, N., and Zhao, F.: Decadal trends in
the seasonal-cycle amplitude of terrestrial CO2exchange result-
ing from the ensemble of terrestrial biosphere models, Tellus B,
68, 28968, https://doi.org/10.3402/tellusb.v68.28968, 2016.
Landschützer, P., Gruber, N., and Bakker, D. C. E.:
Decadal variations and trends of the global ocean car-
bon sink, Global Biogeochem. Cy., 30, 1396–1417,
https://doi.org/10.1002/2015GB005359, 2016.
Landschützer, P., Gruber, N., and Bakker, D. C. E.: An updated
observation-based global monthly gridded sea surface pCO2
and air-sea CO2flux product from 1982 through 2015 and
its monthly climatology (NCEI Accession 0160558), Version
2.2, NOAA National Centers for Environmental Information,
Dataset, [2017-07-11], 2017.
Law, R. M., Chen, Y-H., Gurney, K. R., and TransCom 3 model-
ers: TransCom 3 CO2inversion intercomparison: 2. Sensitivity
of annual mean results to data choices, Tellus B, 55B, 580–595,
https://doi.org/10.1034/j.1600-0889.2003.00053.x, 2003.
Le Quéré, C., Andrew, R. M., Friedlingstein, P., Sitch, S., Pongratz,
J., Manning, A. C., Korsbakken, J. I., Peters, G. P., Canadell,
J. G., Jackson, R. B., Boden, T. A., Tans, P. P., Andrews, O.
D., Arora, V. K., Bakker, D. C. E., Barbero, L., Becker, M.,
Betts, R. A., Bopp, L., Chevallier, F., Chini, L. P., Ciais, P.,
Cosca, C. E., Cross, J., Currie, K., Gasser, T., Harris, I., Hauck,
J., Haverd, V., Houghton, R. A., Hunt, C. W., Hurtt, G., Ily-
ina, T., Jain, A. K., Kato, E., Kautz, M., Keeling, R. F., Klein
Goldewijk, K., Körtzinger, A., Landschützer, P., Lefèvre, N.,
Lenton, A., Lienert, S., Lima, I., Lombardozzi, D., Metzl, N.,
Millero, F., Monteiro, P. M. S., Munro, D. R., Nabel, J. E. M.
S., Nakaoka, S., Nojiri, Y., Padin, X. A., Peregon, A., Pfeil, B.,
Pierrot, D., Poulter, B., Rehder, G., Reimer, J., Rödenbeck, C.,
Schwinger, J., Séférian, R., Skjelvan, I., Stocker, B. D., Tian, H.,
Tilbrook, B., Tubiello, F. N., van der Laan-Luijkx, I. T., van der
Werf, G. R., van Heuven, S., Viovy, N., Vuichard, N., Walker,
A. P., Watson, A. J., Wiltshire, A. J., Zaehle, S., and Zhu, D.:
Global Carbon Budget 2017, Earth Syst. Sci. Data, 10, 405–448,
https://doi.org/10.5194/essd-10-405-2018, 2018.
Liu, D. C. and Nocedal J.: On the limited memory BFGS
method for large scale optimization, Math. Prog., 45, 503–528,
https://doi.org/10.1007/bf01589116, 1989.
Liu, J., Bowman, K. W., Lee, M., Henze, D. K., Bousserez, N., Brix,
H., Collatz, G. J., Menemenlis, D., Ott, L., Pawson, S., Jones, D.,
and Ray Nassar, R.: Carbon monitoring system flux estimation
and attribution: impact of ACOS-GOSAT XCO2sampling on the
inference of terrestrial biospheric sources and sinks, Tellus B, 66,
22486, https://doi.org/10.3402/tellusb.v66.22486, 2014.
Liu, J., Bowman, K. W., Schimel, D. S., Parazoo, N. C., Jiang,
Z., Lee, M., Bloom, A. A., Wunch, D., Frankenberg, C., Sun,
Y., O’Dell, C. W., Gurney, K. R., Menemenlis, D., Gierach, M.,
Crisp, D., and Eldering A.: Contrasting carbon cycle responses of
the tropical continents to the 2015–2016 El Niño, Science, 358,
eaam5690, https://doi.org/10.1126/science.aam5690, 2017.
Lokupitiya, E., Denning, S., Paustian, K., Baker, I., Schae-
fer, K., Verma, S., Meyers, T., Bernacchi, C. J., Suyker, A.,
and Fischer, M.: Incorporation of crop phenology in Simple
Biosphere Model (SiBcrop) to improve land-atmosphere car-
bon exchanges from croplands, Biogeosciences, 6, 969–986,
https://doi.org/10.5194/bg-6-969-2009, 2009.
Nassar, R., Jones, D. B. A., Suntharalingam, P., Chen, J. M., Andres,
R. J., Wecht, K. J., Yantosca, R. M., Kulawik, S. S., Bowman,
K. W., Worden, J. R., Machida, T., and Matsueda, H.: Model-
ing global atmospheric CO2with improved emission inventories
and CO2production from the oxidation of other carbon species,
Geosci. Model Dev., 3, 689–716, https://doi.org/10.5194/gmd-3-
689-2010, 2010.
Nassar, R., Napier-Linton, L., Gurney, K. R., Andres, R. J., Oda,
T., Vogel, F. R., and Deng, F.: Improving the temporal and
spatial distribution of CO2emissions from global fossil fuel
emission data sets, J. Geophys. Res.-Atmos., 118, 917–933,
https://doi.org/10.1029/2012JD018196, 2013.
Oda, T., Maksyutov, S., and Andres, R. J.: The Open-source Data
Inventory for Anthropogenic CO2, version 2016 (ODIAC2016):
a global monthly fossil fuel CO2 gridded emissions data product
for tracer transport simulations and surface flux inversions, Earth
Syst. Sci. Data, 10, 87–107, https://doi.org/10.5194/essd-10-87-
2018, 2018.
O’Dell, C. W., Connor, B., Bösch, H., O’Brien, D., Frankenberg,
C., Castano, R., Christi, M., Eldering, D., Fisher, B., Gunson, M.,
McDuffie, J., Miller, C. E., Natraj, V., Oyafuso, F., Polonsky, I.,
Smyth, M., Taylor, T., Toon, G. C., Wennberg, P. O., and Wunch,
D.: The ACOS CO2retrieval algorithm – Part 1: Description and
validation against synthetic observations, Atmos. Meas. Tech., 5,
99–121, https://doi.org/10.5194/amt-5-99-2012, 2012.
Olsen, S. C. and Randerson, J. T.: Differences between sur-
face and column atmospheric CO2and implications for car-
bon cycle research, J. Geophys. Res.-Atmos., 109, D02301,
https://doi.org/10.1029/2003JD003968, 2004.
Olsen, S. C., Wuebbles, D. J., and Owen, B.: Comparison of global
3-D aviation emissions datasets, Atmos. Chem. Phys., 13, 429–
441, https://doi.org/10.5194/acp-13-429-2013, 2013.
Ott, L. E., Pawson, S., Collatz, G. J., Gregg, W. W., Mene-
menlis, D., Brix, H., Rousseaux, C. S., Bowman, K. W., Liu,
J., Eldering, A., Gunson, M. R., and Kawa, S. R.: Assess-
ing the magnitude of CO2flux uncertainty in atmospheric
CO2records using products from NASA’s Carbon Monitor-
ing Flux Pilot Project, J. Geophys. Res.-Atmos., 120,734–765,
https://doi.org/10.1002/2014JD022411, 2015.
Peng, S., Ciais, P., Chevallier, F., Peylin, P., Cadule, P., Sitch,
S., Piao, S., Ahlström, A., Huntingford, C., Levy, P., Li, X.,
Liu, Y., Lomas, M., Poulter, B., Viovy, N., Wang, T., Wang,
X., Zaehle, S., Zeng, N., Zhao, F., and Zhao, H.: Bench-
marking the seasonal cycle of CO2fluxes simulated by terres-
trial ecosystem models, Global Biogeochem. Cy., 29, 46–64,
https://doi.org/10.1002/2014GB004931, 2015.
Peylin, P., Law, R. M., Gurney, K. R., Chevallier, F., Jacobson,
A. R., Maki, T., Niwa, Y., Patra, P. K., Peters, W., Rayner, P.
J., Rödenbeck, C., van der Laan-Luijkx, I. T., and Zhang, X.:
Global atmospheric carbon budget: results from an ensemble of
atmospheric CO2inversions, Biogeosciences, 10, 6699–6720,
https://doi.org/10.5194/bg-10-6699-2013, 2013.
Peters, W., Jacobson, A. R., Sweeney, C., Andrews, A. E., Con-
way, T. J., Masarie, K., Miller, J. B., Bruhwiler, L. M. P., Pétron,
G., Hirsch, A. I., Worthy, D. E. J., van der Werf, G. R., Ran-
derson, J. T., Wennberg, P. O., Krol, M. C., and Tans, P. P.: An
atmospheric perspective on North American carbon dioxide ex-
www.atmos-chem-phys.net/19/13267/2019/ Atmos. Chem. Phys., 19, 13267–13287, 2019
13286 S. Philip et al.: Prior biosphere model impact
change: CarbonTracker, P. Natl. Acad. Sci. USA, 104, 18925–
18930, https://doi.org/10.1073/pnas.0708986104, 2007.
Potter, C., Randerson, J. T., Field, C. B., Matson, P. A., Vi-
tousek, P. M., Mooney, H. A., and Klooster, S A.: Terrestrial
ecosystem production: A process model based on global satel-
lite and surface data, Global Biogeochem. Cy., 7, 811–841,
https://doi.org/10.1029/93GB02725, 1993.
Potter, C., Klooster, S., Myneni, R., Genovese, V., Tan, P.-
N., and Kumar, V.: Continental scale comparisons of terres-
trial carbon sinks estimated from satellite data and ecosys-
tem modeling 1982–98, Global Planet. Change, 39, 201–213,
https://doi.org/10.1016/j.gloplacha.2003.07.001, 2003.
Potter, C., Kumar, V., Klooster, S., and Nemani, R.: Recent
history of trends in vegetation greenness and large-scale
ecosystem disturbances in Eurasia, Tellus B, 59, 260–272,
https://doi.org/10.1111/j.1600-0889.2006.00245.x, 2007.
Potter, C., Klooster, S., Huete, A., Genovese, V., Bustamante, M.,
Guimaraes Ferreira, L., R. C. de Oliveira Jr., and Zepp, R.: Ter-
restrial carbon sinks in the Brazilian Amazon and Cerrado re-
gion predicted from MODIS satellite data and ecosystem mod-
eling, Biogeosciences, 6, 937–945, https://doi.org/10.5194/bg-6-
937-2009, 2009.
Potter, C., Klooster, S., and Genovese, V.: Net primary pro-
duction of terrestrial ecosystems from 2000 to 2009, Cli-
matic Change, 115, 365–378, https://doi.org/10.1007/s10584-
012-0460-2, 2012a.
Potter, C., Klooster, S., Genovese, V., Hiatt, C., Boriah, S.,
Kumar, V., Mithal, V., and Garg, A.: Terrestrial ecosys-
tem carbon fluxes predicted from MODIS satellite data and
large-scale disturbance modeling, Int. J. Geosci., 3, 469–479,
https://doi.org/10.4236/ijg.2012.33050, 2012b.
Poulter, B., Frank, D., Ciais, P., Myneni, R. B., Andela, N., Bi, J.,
Broquet, G., Canadell, J. G., Chevallier, F., Liu, Y. Y., Running,
S. W., Sitch, S., and van der Werf, G. R.: Contribution of semi-
arid ecosystems to interannual variability of the global carbon cy-
cle, Nature, 509, 600–603, https://doi.org/10.1038/nature13376,
2014.
Randerson, J. T., Thompson, M. V., Malmstrom, C. M., Field,
C. B., and Fung, I. Y.: Substrate limitations for heterotrophs:
Implications for models that estimate the seasonal cycle of
atmospheric CO2, Global Biogeochem. Cy., 10, 585–602,
https://doi.org/10.1029/96GB01981, 1996.
Rödenbeck, C., Houweling, S., Gloor, M., and Heimann, M.: CO2
flux history 1982–2001 inferred from atmospheric data using a
global inversion of atmospheric transport, Atmos. Chem. Phys.,
3, 1919–1964, https://doi.org/10.5194/acp-3-1919-2003, 2003.
Rödenbeck, C., Conway, T. J., and Langenfelds, R. L.: The effect
of systematic measurement errors on atmospheric CO2inver-
sions: a quantitative assessment, Atmos. Chem. Phys., 6, 149–
161, https://doi.org/10.5194/acp-6-149-2006, 2006.
Schaefer, K., Collatz, G. J., Tans, P., Denning, A. S., Baker, I.,
Berry, J., Prihodko, L., Suits, N., and Philpott, A.: Combined
Simple Biosphere/Carnegie-Ames-Stanford Approach terrestrial
carbon cycle model, J. Geophys. Res.-Biogeo., 113, G03034,
https://doi.org/10.1029/2007JG000603, 2008.
Schimel, D., Stephens, B. B., and Fisher, J. B.: Effect of increasing
CO2on the terrestrial carbon cycle, P. Natl. Acad. Sci. USA, 112,
436–441, https://doi.org/10.1073/pnas.1407302112, 2015.
Schimel, D. S., House, J. I., Hibbard, K. A., Bousquet, P., Ciais, P.,
Peylin, P., Braswell, B. H., Apps, M. J., Baker, D., Bondeau, A.,
Canadell, J., Churkina, G., Cramer, W., Denning, A. S., Field,
C. B., Friedlingstein, P., Goodale, C., Heimann, M., Houghton,
R. A., Melillo, J. M., Moore III, B., Murdiyarso, D., Noble, I.,
Pacala, S. W., Prentice, I. C., Raupach, M. R., Rayner, P. J., Sc-
holes, R. J., Steffen, W. L., and Wirth, C.: Recent patterns and
mechanisms of carbon exchange by terrestrial ecosystems, Na-
ture, 414, 169–172, https://doi.org/10.1038/35102500, 2001.
Schwalm, C. R., Huntzinger, D. N., Fisher, J. B., Michalak, A. M.,
Bowman, K., Ciais, P., Cook, R., El-Masri, B., Hayes, D., Huang,
M., Ito, A., Jain, A., King, A. W., Lei, H., Liu, J., Lu, C., Mao, J.,
Peng, S., Poulter, B., Ricciuto, D., Schaefer, K., Shi, X., Tao, B.,
Tian, H., Wang, W., Wei, Y., Yang, J., and Zeng, N.: Toward “op-
timal” integration of terrestrial biosphere models, Geophys. Res.
Lett., 42, 4418–4428, https://doi.org/10.1002/2015GL064002,
2015.
Seinfeld, J. H. and Pandis, S. N.: Atmospheric chemistry and
physics: from air pollution to climate change, 3rd Edn., Wiley,
2016.
Sellers, P. J., Mintz, Y., Sud, Y. C., and Dalcher, A.: A simple
biosphere model (SiB) for use within general circulation mod-
els, J. Atmos. Sci., 43, 505–531, https://doi.org/10.1175/1520-
0469(1986)043<0505:ASBMFU>2.0.CO;2, 1986.
Sellers, P. J., Randall, D. A., Collatz, G. J., Berry, J. A.,
Field, C. B., Dazlich, D. A., Zhang, C., Collelo, G. D.,
and Buonoua, L.: A revised Land Surface Parameteriza-
tion (SiB2) for atmospheric GCMs. Part I: Model formu-
lation, J. Climate, 9, 676–705, https://doi.org/10.1175/1520-
0442(1996)009<0676:ARLSPF>2.0.CO;2, 1996.
Sitch, S., Smith, B., Prentice, I. C., Arneth, A., Bondeau, A.,
Cramer, W., Kaplan, J. O., Levis, S., Lucht, W., Sykes, M. T.,
Thonicke, K., and Venevsky, S.: Evaluation of ecosystem dynam-
ics, plant geography and terrestrial carbon cycling in the LPJ dy-
namic global vegetation model, Glob. Change Biol., 9, 161–185,
https://doi.org/10.1046/j.1365-2486.2003.00569.x, 2003.
Sitch, S., Friedlingstein, P., Gruber, N., Jones, S. D., Murray-
Tortarolo, G., Ahlström, A., Doney, S. C., Graven, H., Heinze,
C., Huntingford, C., Levis, S., Levy, P. E., Lomas, M., Poul-
ter, B., Viovy, N., Zaehle, S., Zeng, N., Arneth, A., Bonan,
G., Bopp, L., Canadell, J. G., Chevallier, F., Ciais, P., Ellis,
R., Gloor, M., Peylin, P., Piao, S. L., Le Quéré, C., Smith, B.,
Zhu, Z., and Myneni, R.: Recent trends and drivers of regional
sources and sinks of carbon dioxide, Biogeosciences, 12, 653–
679, https://doi.org/10.5194/bg-12-653-2015, 2015.
Suntharalingam, P., Jacob, D. J., Palmer, P. I., Logan, J. A., Yan-
tosca, R. M., Xiao, Y., Evans, M. J., Streets, D. G., Vay, S. L., and
Sachse, G. W.: Improved quantification of Chinese carbon fluxes
using CO2/CO correlations in Asian outflow, J. Geophys. Res.-
Atmos., 109, D18S18, https://doi.org/10.1029/2003JD004362,
2004.
Takagi H., Houweling, S., Andres, R. J., Belikov, D., Bril, A.,
Boesch H., Butz, A., Guerlet, S., Hasekamp, O., Maksyutov,
S., Morino, I., Oda T., O’Dell, C. W., Oshchepkov, S., Parker,
R., Saito, M., Uchino, O., Yokota, T., Yoshida, Y., and Valsala,
V.: Influence of differences in current GOSAT XCO2retrievals
on surface flux estimation, Geophys. Res. Lett., 41, 2598–2605,
https://doi.org/10.1002/2013GL059174, 2014.
Atmos. Chem. Phys., 19, 13267–13287, 2019 www.atmos-chem-phys.net/19/13267/2019/
S. Philip et al.: Prior biosphere model impact 13287
Takahashi, T., Sutherland, S. C., Wanninkhof, R., Sweeney, C.,
Feely, R. A., Chipman, D. W., Hales, B., Friederich, G., Chavez,
F., Sabine, C., Watson, A., Bakker, D. C. E., Schuster, U., Metzl,
N., Yoshikawa-Inoue, H., Ishii, M., Midorikawa, T., Nojiri, Y.,
Körtzinger, A., Steinhoff, T., Hoppema, M., Olafsson, J., Arnar-
son, T. S., Tilbrook, B., Johannessen, T., Olsen, A., Bellerby, R.,
Wong, C. S., Delille, B., Bates, N. R., and de Baar, H. J. W.: Cli-
matological mean and decadal change in surface ocean pCO2,
and net sea-air CO2flux over the global oceans, Deep-Sea Res.
Pt. II, 56, 554–577, https://doi.org/10.1016/j.dsr2.2008.12.009,
2009.
van der Werf, G. R., Randerson, J. T., Collatz, G. J., Giglio, L.,
Kasibhatla, P. S., Arellano, A. F., Olsen, S. C., and Kasischke,
E. S.: Continental-scale partitioning of fire emissions during the
1997 to 2001 El Nino/La Nina period, Science, 303, 73–76,
https://doi.org/10.1126/science.1090753, 2004.
van der Werf, G. R., Randerson, J. T., Giglio, L., Collatz, G. J.,
Kasibhatla, P. S., and Arellano Jr., A. F.: Interannual variability
in global biomass burning emissions from 1997 to 2004, Atmos.
Chem. Phys., 6, 3423–3441, https://doi.org/10.5194/acp-6-3423-
2006, 2006.
van der Werf, G. R., Randerson, J. T., Giglio, L., Collatz, G.
J., Mu, M., Kasibhatla, P. S., Morton, D. C., DeFries, R. S.,
Jin, Y., and van Leeuwen, T. T.: Global fire emissions and the
contribution of deforestation, savanna, forest, agricultural, and
peat fires (1997–2009), Atmos. Chem. Phys., 10, 11707–11735,
https://doi.org/10.5194/acp-10-11707-2010, 2010.
Wang, J. S., Kawa, S. R., Collatz, G. J., Sasakawa, M., Gatti, L. V.,
Machida, T., Liu, Y., and Manyin, M. E.: A global synthesis in-
version analysis of recent variability in CO2fluxes using GOSAT
and in situ observations, Atmos. Chem. Phys., 18, 11097–11124,
https://doi.org/10.5194/acp-18-11097-2018, 2018.
www.atmos-chem-phys.net/19/13267/2019/ Atmos. Chem. Phys., 19, 13267–13287, 2019
... The four-dimensional variational (4D-Var) assimilation system with the GEOS-Chem global chemical transport model (CTM; Philip et al., 2019) was used to estimate global NEE and oceanic CO 2 fluxes on a monthly scale at the spatial resolution of the model from 2015 to 2018 (see Section 3). Regionally aggregated NEE and Net Biome Exchange (NBE = NEE + unoptimized biomass burning emissions [BBE]) fluxes over the South Asia region were assessed. ...
... For posterior flux calculations, we use the GE-OS-Chem adjoint, version 35a (Henze et al., 2007;Liu et al., 2014). The GEOS-Chem forward model (e.g., Krishnapriya et al., 2020;Nassar et al., 2010) and GEOS-Chem-based 4D-Var data assimilation system have been extensively validated and used for CO 2 flux inversions in previous studies (e.g., Bowman et al., 2017;Deng et al., 2014;Liu et al., 2017;Philip et al., 2019;Wang et al., 2019). In this study, MERRA-2 assimilated meteorology was used to drive the forward model. ...
... The 4D-Var data assimilation system with the GEOS-Chem CTM was used to conduct the CO 2 surface flux estimations following Philip et al. (2019). The inversion system minimizes the cost function (Equation 1) iteratively to optimize the state vector, σ representing scaling factors to be multiplied with monthly NEE and oceanic fluxes in surface grid boxes of the model (x), ...
Article
Full-text available
The spatiotemporal variability of terrestrial biospheric carbon dioxide (CO2) fluxes over South Asia has large uncertainty. The Orbiting Carbon Observatory 2 (OCO‐2) satellite provides much‐needed retrievals of column‐average CO2 on a global‐scale, with the highest sensitivity to surface CO2 fluxes and spatiotemporal resolution available to‐date. This study conducted global inverse model simulations, assimilating in situ (IS) data and OCO‐2 retrievals, to assess optimized CO2 net ecosystem exchange (NEE) fluxes for South Asia. Annual Net Biome Exchange (NBE = NEE + biomass burning) fluxes over South Asia were estimated to be near neutral (0.04 ± 0.14 PgC yr⁻¹) using both IS and OCO‐2 observations. The most robust result found by assimilating OCO‐2 observations was the constraint imposed on the seasonal cycle of NBE fluxes. The amplitude of the seasonal cycle of NEE was found to be larger than previously assumed. The OCO‐2 inversion led to an NBE seasonal amplitude of 0.34 PgC month⁻¹, which was larger compared to IS constrained NBE (0.19 PgC month⁻¹) and MsTMIP ensemble mean NEE (0.16 PgC month⁻¹). Moreover, OCO‐2 data imposed a phase shift in the NBE seasonal cycle predicted by the prior model. The larger magnitude of NEE seasonality, and phase shift, simulated when assimilating OCO‐2 observations are in general agreement with previous studies assimilating regional aircraft observations in addition to global IS observations. This result suggests that OCO‐2 provides valuable data that allows for the estimate of NBE on a regional scale in a similar manner as regional in situ aircraft networks.
... One possible explanation for this seasonal bias is that the prior NEE estimates used in the inversions have biased the posterior NEE estimates. Previous studies (e.g., Philip et al., 2019) have demonstrated the large impact of prior information on the current global inversion systems. Feng et al. (2021) showed that one model commonly used as a flux prior appears to underestimate the magnitude of summer NEE. ...
Article
Full-text available
Atmospheric inversion estimates of net ecosystem exchange (NEE) of CO2 are increasingly relevant to climate policy. We evaluated sub‐continental, seasonal estimates of CO2 NEE from nine global inversion systems that participated in the Orbiting Carbon Observatory‐2 model intercomparison project (OCO‐2 v9 MIP), using 98 research flights conducted over the central and eastern United States from 2016 to 2018 as part of the Atmospheric Carbon and Transport ‐ America mission. We found that the seasonal amplitude of NEE in the central and eastern United States is underestimated in these models and model‐data biases are largest for those inversions with the smallest seasonal flux amplitudes. These results were independent of whether the inversions used satellite or in situ data. The largest NEE biases were observed in the Midwest croplands and eastern forests. Future experiments are needed to determine the causes of the persistent biases and if they are associated with biases in annual flux estimates.
... The uncertainties in the regional fluxes mainly arise from prior flux distribution and seasonality, selection of observational network and data uncertainty, transport model resolution leading to site-representation error, and uncertainties arising from parameterization of transport processes (Basu et al., 2018;Patra et al., 2005a;Philip et al., 2019;Qu et al., 2021;Wang et al., 2018). The uncertainties associated with the subcontinental-scale CO 2 fluxes are often much greater than the interannual and interdecadal flux changes in non-FFC sectors, which allows us to make a better assessment of the changes in regional CO 2 fluxes compared to knowledge gained in regional flux magnitudes (Baker et al., 2006;Gurney et al., 2008;Patra et al., 2005a, b;Peylin et al., 2013;Rayner et al., 2008;. ...
Article
Full-text available
Global and regional sources and sinks of carbon across the earth's surface have been studied extensively using atmospheric carbon dioxide (CO2) observations and atmospheric chemistry-transport model (ACTM) simulations (top-down/inversion method). However, the uncertainties in the regional flux distributions remain unconstrained due to the lack of high-quality measurements, uncertainties in model simulations, and representation of data and flux errors in the inversion systems. Here, we assess the representation of data and flux errors using a suite of 16 inversion cases derived from a single transport model (MIROC4-ACTM) but different sets of a priori (bottom-up) terrestrial biosphere and oceanic fluxes, as well as prior flux and observational data uncertainties (50 sites) to estimate CO2 fluxes for 84 regions over the period 2000–2020. The inversion ensembles provide a mean flux field that is consistent with the global CO2 growth rate, land and ocean sink partitioning of −2.9 ± 0.3 (± 1σ uncertainty on the ensemble mean) and −1.6 ± 0.2 PgC yr−1, respectively, for the period 2011–2020 (without riverine export correction), offsetting about 22 %–33 % and 16 %–18 % of global fossil fuel CO2 emissions. The rivers carry about 0.6 PgC yr−1 of land sink into the deep ocean, and thus the effective land and ocean partitioning is −2.3 ± 0.3 and −2.2 ± 0.3, respectively. Aggregated fluxes for 15 land regions compare reasonably well with the best estimations for the 2000s (∼ 2000–2009), given by the REgional Carbon Cycle Assessment and Processes (RECCAP), and all regions appeared as a carbon sink over 2011–2020. Interannual variability and seasonal cycle in CO2 fluxes are more consistently derived for two distinct prior fluxes when a greater degree of freedom (increased prior flux uncertainty) is given to the inversion system. We have further evaluated the inversion fluxes using meridional CO2 distributions from independent (not used in the inversions) aircraft and surface measurements, suggesting that the ensemble mean flux (model–observation mean ± 1σ standard deviation = −0.3 ± 3 ppm) is best suited for global and regional CO2 flux budgets than an individual inversion (model–observation 1σ standard deviation = −0.35 ± 3.3 ppm). Using the ensemble mean fluxes and uncertainties for 15 land and 11 ocean regions at 5-year intervals, we show promise in the capability to track flux changes toward supporting the ongoing and future CO2 emission mitigation policies.
... As shown in Fig. S6, except for southern Africa, the difference between the two is significantly smaller in 2010-2014 than in 2015-2018, especially in temperate S. America, northern Africa, and Australia, con-firming that the significant differences are mainly from the different XCO 2 products used in these two studies. In addition to XCO 2 data, the prior carbon flux can also have a significant impact on the inversion results (Philip et al., 2019). We further examine the prior and posterior NBE over southern Africa in these two studies and find that the prior NBEs used in these two systems are quite different (a strong sink in CMS-Flux and a source in this study). ...
Article
Full-text available
A global gridded net ecosystem exchange (NEE) of CO2 dataset is vital in global and regional carbon cycle studies. Top-down atmospheric inversion is one of the major methods to estimate the global NEE; however, the existing global NEE datasets generated through inversion from conventional CO2 observations have large uncertainties in places where observational data are sparse. Here, by assimilating the GOSAT ACOS v9 XCO2 product, we generate a 10-year (2010–2019) global monthly terrestrial NEE dataset using the Global Carbon Assimilation System, version 2 (GCASv2), which is named GCAS2021. It includes gridded (1∘×1∘), globally, latitudinally, and regionally aggregated prior and posterior NEE and ocean (OCN) fluxes and prescribed wildfire (FIRE) and fossil fuel and cement (FFC) carbon emissions. Globally, the decadal mean NEE is -3.73±0.52 PgC yr−1, with an interannual amplitude of 2.73 PgC yr−1. Combining the OCN flux and FIRE and FFC emissions, the net biosphere flux (NBE) and atmospheric growth rate (AGR) as well as their inter-annual variabilities (IAVs) agree well with the estimates of the Global Carbon Budget 2020. Regionally, our dataset shows that eastern North America, the Amazon, the Congo Basin, Europe, boreal forests, southern China, and Southeast Asia are carbon sinks, while the western United States, African grasslands, Brazilian plateaus, and parts of South Asia are carbon sources. In the TRANSCOM land regions, the NBEs of temperate N. America, northern Africa, and boreal Asia are between the estimates of CMS-Flux NBE 2020 and CT2019B, and those in temperate Asia, Europe, and Southeast Asia are consistent with CMS-Flux NBE 2020 but significantly different from CT2019B. In the RECCAP2 regions, except for Africa and South Asia, the NBEs are comparable with the latest bottom-up estimate of Ciais et al. (2021). Compared with previous studies, the IAVs and seasonal cycles of NEE of this dataset could clearly reflect the impacts of extreme climates and large-scale climate anomalies on the carbon flux. The evaluations also show that the posterior CO2 concentrations at remote sites and on a regional scale, as well as on vertical CO2 profiles in the Asia-Pacific region, are all consistent with independent CO2 measurements from surface flask and aircraft CO2 observations, indicating that this dataset captures surface carbon fluxes well. We believe that this dataset can contribute to regional- or national-scale carbon cycle and carbon neutrality assessment and carbon dynamics research. The dataset can be accessed at https://doi.org/10.5281/zenodo.5829774 (Jiang, 2022).
... On a final note, errors stemming from biases in the transport model must be quantified alongside other sources of systematic errors-such as biases in the prior fluxes and assimilated data [11][12][13] -before the Chinese carbon budget can be considered to be well understood. Ensembles of inverse models that differ in CTM and inversion approach probably provide the best option for quantifying systematic sources of error. ...
... On a final note, errors stemming from biases in the transport model must be quantified alongside other sources of systematic errors-such as biases in the prior fluxes and assimilated data [11][12][13] -before the Chinese carbon budget can be considered to be well understood. Ensembles of inverse models that differ in CTM and inversion approach probably provide the best option for quantifying systematic sources of error. ...
... The Ames inversion system used the transport model GEOS-Chem (Goddard Earth Observing System -Chemistry; Bey et al., 2001), driven by meteorological parameters from the MERRA-2 reanalysis (Bosilovich et al., 2017) and run at a 4 • × 5 • resolution (further description provided in Philip et al. (2019)). Surface fluxes were optimized monthly. ...
Article
Full-text available
The Orbiting Carbon Observatory 2 (OCO-2) satellite has been providing information to estimate carbon dioxide (CO2) fluxes at global and regional scales since 2014 through the combination of CO2 retrievals with top–down atmospheric inversion methods. Column average CO2 dry-air mole fraction retrievals have been constantly improved. A bias correction has been applied in the OCO-2 version 9 retrievals compared to the previous OCO-2 version 7r improving data accuracy and coverage. We study an ensemble of 10 atmospheric inversions all characterized by different transport models, data assimilation algorithms, and prior fluxes using first OCO-2 v7 in 2015–2016 and then OCO-2 version 9 land observations for the longer period 2015–2018. Inversions assimilating in situ (IS) measurements have also been used to provide a baseline against which the satellite-driven results are compared. The time series at different scales (going from global to regional scales) of the models emissions are analyzed and compared to each experiment using either OCO-2 or IS data. We then evaluate the inversion ensemble based on the dataset from the Total Carbon Column Observing Network (TCCON), aircraft, and in situ observations, all independent from assimilated data. While we find a similar constraint of global total carbon emissions between the ensemble spread using IS and both OCO-2 retrievals, differences between the two retrieval versions appear over regional scales and particularly in tropical Africa. A difference in the carbon budget between v7 and v9 is found over this region, which seems to show the impact of corrections applied in retrievals. However, the lack of data in the tropics limits our conclusions, and the estimation of carbon emissions over tropical Africa require further analysis.
Article
Land carbon sink is a vital component for the achievement of China’s ambitious carbon neutrality goal, but its magnitude is poorly known. Atmospheric observations and inverse models are valuable tools to constrain the China’s land carbon sink. Space‐based CO2 measurements from satellites form an emerging data stream for application of such atmospheric inversions. Here, we reviewed the satellite missions that is dedicated to the monitoring of CO2, and the recent progresses on the inversion of China’s land carbon sink using satellite CO2 measurements. We summarized the limitations and challenges in current space platforms, retrieval algorithms, and the inverse modelling. It is shown that there are large uncertainties of contemporary satellite‐based estimates of China’s land carbon sink. We discussed future opportunities of continuous improvements in three aspects to better constrain China’s land carbon sink with space‐based CO2 measurements.
Article
Full-text available
The ability of current global models to simulate the transport of CO2 by mid‐latitude, synoptic‐scale weather systems (i.e., CO2 weather) is important for inverse estimates of regional and global carbon budgets but remains unclear without comparisons to targeted measurements. Here, we evaluate ten models that participated in the Orbiting Carbon Observatory‐2 model intercomparison project (OCO‐2 MIP version 9) with intensive aircraft measurements collected from the Atmospheric Carbon Transport (ACT)‐America mission. We quantify model‐data differences in the spatial variability of CO2 mole fractions, mean winds, and boundary layer depths in 27 mid‐latitude cyclones spanning four seasons over the central and eastern United States. We find that the OCO‐2 MIP models are able to simulate observed CO2 frontal differences with varying degrees of success in summer and spring, and most underestimate frontal differences in winter and autumn. The models may underestimate the observed boundary layer‐to‐free troposphere CO2 differences in spring and autumn due to model errors in boundary layer height. Attribution of the causes of model biases in other seasons remains elusive. Transport errors, prior fluxes, and/or inversion algorithms appear to be the primary cause of these biases since model performance is not highly sensitive to the CO2 data used in the inversion. The metrics presented here provide new benchmarks regarding the ability of atmospheric inversion systems to reproduce the CO2 structure of mid‐latitude weather systems. Controlled experiments are needed to link these metrics more directly to the accuracy of regional or global flux estimates.
Article
Full-text available
WOMBAT (the WOllongong Methodology for Bayesian Assimilation of Trace-gases) is a fully Bayesian hierarchical statistical framework for flux inversion of trace gases from flask, in situ, and remotely sensed data. WOMBAT extends the conventional Bayesian synthesis framework through the consideration of a correlated error term, the capacity for online bias correction, and the provision of uncertainty quantification on all unknowns that appear in the Bayesian statistical model. We show, in an observing system simulation experiment (OSSE), that these extensions are crucial when the data are indeed biased and have errors that are spatio-temporally correlated. Using the GEOS-Chem atmospheric transport model, we show that WOMBAT is able to obtain posterior means and variances on non-fossil-fuel CO2 fluxes from Orbiting Carbon Observatory-2 (OCO-2) data that are comparable to those from the Model Intercomparison Project (MIP) reported in Crowell et al. (2019). We also find that WOMBAT's predictions of out-of-sample retrievals obtained from the Total Column Carbon Observing Network (TCCON) are, for the most part, more accurate than those made by the MIP participants.
Article
Full-text available
The Orbiting Carbon Observatory-2 has been on orbit since 2014, and its global coverage holds the potential to reveal new information about the carbon cycle through the use of top-down atmospheric inversion methods combined with column average CO2 retrievals. We employ a large ensemble of atmospheric inversions utilizing different transport models, data assimilation techniques, and prior flux distributions in order to quantify the satellite-informed fluxes from OCO-2 Version 7r land observations and their uncertainties at continental scales. Additionally, we use in situ measurements to provide a baseline against which to compare the satellite-constrained results. We find that within the ensemble spread, in situ observations, and satellite retrievals constrain a similar global total carbon sink of 3.7±0.5 PgC yr⁻¹, and 1.5±0.6 PgC yr⁻¹ for global land, for the 2015–2016 annual mean. This agreement breaks down in smaller regions, and we discuss the differences between the experiments. Of particular interest is the difference between the different assimilation constraints in the tropics, with the largest differences occurring in tropical Africa, which could be an indication of the global perturbation from the 2015–2016 El Niño. Evaluation of posterior concentrations using TCCON and aircraft observations gives some limited insight into the quality of the different assimilation constraints, but the lack of such data in the tropics inhibits our ability to make strong conclusions there.
Article
Full-text available
The precise contribution of the two major sinks for anthropogenic CO2 emissions, terrestrial vegetation and the ocean, and their location and year-to-year variability are not well understood. Top-down estimates of the spatiotemporal variations in emissions and uptake of CO2 are expected to benefit from the increasing measurement density brought by recent in situ and remote CO2 observations. We uniquely apply a batch Bayesian synthesis inversion at relatively high resolution to in situ surface observations and bias-corrected GOSAT satellite column CO2 retrievals to deduce the global distributions of natural CO2 fluxes during 2009–2010. The GOSAT inversion is generally better constrained than the in situ inversion, with smaller posterior regional flux uncertainties and correlations, because of greater spatial coverage, except over North America and northern and southern high-latitude oceans. Complementarity of the in situ and GOSAT data enhances uncertainty reductions in a joint inversion; however, remaining coverage gaps, including those associated with spatial and temporal sampling biases in the passive satellite measurements, still limit the ability to accurately resolve fluxes down to the sub-continental or sub-ocean basin scale. The GOSAT inversion produces a shift in the global CO2 sink from the tropics to the north and south relative to the prior, and an increased source in the tropics of ∼ 2 Pg C yr−1 relative to the in situ inversion, similar to what is seen in studies using other inversion approaches. This result may be driven by sampling and residual retrieval biases in the GOSAT data, as suggested by significant discrepancies between posterior CO2 distributions and surface in situ and HIPPO mission aircraft data. While the shift in the global sink appears to be a robust feature of the inversions, the partitioning of the sink between land and ocean in the inversions using either in situ or GOSAT data is found to be sensitive to prior uncertainties because of negative correlations in the flux errors. The GOSAT inversion indicates significantly less CO2 uptake in the summer of 2010 than in 2009 across northern regions, consistent with the impact of observed severe heat waves and drought. However, observations from an in situ network in Siberia imply that the GOSAT inversion exaggerates the 2010–2009 difference in uptake in that region, while the prior CASA-GFED model of net ecosystem production and fire emissions reasonably estimates that quantity. The prior, in situ posterior, and GOSAT posterior all indicate greater uptake over North America in spring to early summer of 2010 than in 2009, consistent with wetter conditions. The GOSAT inversion does not show the expected impact on fluxes of a 2010 drought in the Amazon; evaluation of posterior mole fractions against local aircraft profiles suggests that time-varying GOSAT coverage can bias the estimation of interannual flux variability in this region.
Article
Full-text available
We estimate the uncertainty of CO2 flux estimates in atmospheric inversions stemming from differences between different global transport models. Using a set of observing system simulation experiments (OSSEs), we estimate this uncertainty as represented by the spread between five different state-of-the-art global transport models (ACTM, LMDZ, GEOS-Chem, PCTM and TM5), for both traditional in situ CO2 inversions and inversions of XCO2 estimates from the Orbiting Carbon Observatory 2 (OCO-2). We find that, in the absence of relative biases between in situ CO2 and OCO-2 XCO2, OCO-2 estimates of terrestrial flux for TRANSCOM-scale land regions can be more robust to transport model differences than corresponding in situ CO2 inversions. This is due to a combination of the increased spatial coverage of OCO-2 samples and the total column nature of OCO-2 estimates. We separate the two effects by constructing hypothetical in situ networks with the coverage of OCO-2 but with only near-surface samples. We also find that the transport-driven uncertainty in fluxes is comparable between well-sampled northern temperate regions and poorly sampled tropical regions. Furthermore, we find that spatiotemporal differences in sampling, such as between OCO-2 land and ocean soundings, coupled with imperfect transport, can produce differences in flux estimates that are larger than flux uncertainties due to transport model differences. This highlights the need for sampling with as complete a spatial and temporal coverage as possible (e.g., using both land and ocean retrievals together for OCO-2) to minimize the impact of selective sampling. Finally, our annual and monthly estimates of transport-driven uncertainties can be used to evaluate the robustness of conclusions drawn from real OCO-2 and in situ CO2 inversions.
Article
Full-text available
Accurate assessment of anthropogenic carbon dioxide (CO2) emissions and their redistribution among the atmosphere, ocean, and terrestrial biosphere – the “global carbon budget” – is important to better understand the global carbon cycle, support the development of climate policies, and project future climate change. Here we describe data sets and methodology to quantify the five major components of the global carbon budget and their uncertainties. CO2 emissions from fossil fuels and industry (EFF) are based on energy statistics and cement production data, respectively, while emissions from land-use change (ELUC), mainly deforestation, are based on land-cover change data and bookkeeping models. The global atmospheric CO2 concentration is measured directly and its rate of growth (GATM) is computed from the annual changes in concentration
Article
Full-text available
The Open-source Data Inventory for Anthropogenic CO2 (ODIAC) is a global high-spatial-resolution gridded emissions data product that distributes carbon dioxide (CO2) emissions from fossil fuel combustion. The emissions spatial distributions are estimated at a 1 × 1 km spatial resolution over land using power plant profiles (emissions intensity and geographical location) and satellite-observed nighttime lights. This paper describes the year 2016 version of the ODIAC emissions data product (ODIAC2016) and presents analyses that help guide data users, especially for atmospheric CO2 tracer transport simulations and flux inversion analysis. Since the original publication in 2011, we have made modifications to our emissions modeling framework in order to deliver a comprehensive global gridded emissions data product. Major changes from the 2011 publication are (1) the use of emissions estimates made by the Carbon Dioxide Information Analysis Center (CDIAC) at the Oak Ridge National Laboratory (ORNL) by fuel type (solid, liquid, gas, cement manufacturing, gas flaring, and international aviation and marine bunkers); (2) the use of multiple spatial emissions proxies by fuel type such as (a) nighttime light data specific to gas flaring and (b) ship/aircraft fleet tracks; and (3) the inclusion of emissions temporal variations. Using global fuel consumption data, we extrapolated the CDIAC emissions estimates for the recent years and produced the ODIAC2016 emissions data product that covers 2000–2015. Our emissions data can be viewed as an extended version of CDIAC gridded emissions data product, which should allow data users to impose global fossil fuel emissions in a more comprehensive manner than the original CDIAC product. Our new emissions modeling framework allows us to produce future versions of the ODIAC emissions data product with a timely update. Such capability has become more significant given the CDIAC/ORNL's shutdown. The ODIAC data product could play an important role in supporting carbon cycle science, especially modeling studies with space-based CO2 data collected in near real time by ongoing carbon observing missions such as the Japanese Greenhouse gases Observing SATellite (GOSAT), NASA's Orbiting Carbon Observatory-2 (OCO-2), and upcoming future missions. The ODIAC emissions data product including the latest version of the ODIAC emissions data (ODIAC2017, 2000–2016) is distributed from http://db.cger.nies.go.jp/dataset/ODIAC/ with a DOI (https://doi.org/10.17595/20170411.001).
Article
Full-text available
The 2015–2016 El Niño led to historically high temperatures and low precipitation over the tropics, while the growth rate of atmospheric carbon dioxide (CO2) was the largest on record. Here we quantify the response of tropical net biosphere exchange, gross primary production, biomass burning, and respiration to these climate anomalies by assimilating column CO2, solar-induced chlorophyll fluorescence, and carbon monoxide observations from multiple satellites. Relative to the 2011 La Niña, the pantropical biosphere released 2.5 ± 0.34 gigatons more carbon into the atmosphere in 2015, consisting of approximately even contributions from three tropical continents but dominated by diverse carbon exchange processes. The heterogeneity of the carbon-exchange processes indicated here challenges previous studies that suggested that a single dominant process determines carbon cycle interannual variability.
Article
Full-text available
The El Niño Modoki in 2010 led to historic droughts in Brazil. In order to understand its impact on carbon cycle variability, we derive the 2011–2010 annual carbon flux change (δF↑) globally and specifically to Brazil using the NASA Carbon Monitoring System Flux (CMS-Flux) framework. Satellite observations of CO2, CO, and solar-induced fluorescence (SIF) are ingested into a 4D-variational assimilation system driven by carbon cycle models to infer spatially resolved carbon fluxes including net ecosystem production, biomass burning, and gross primary productivity (GPP). The global 2011–2010 net carbon flux change was estimated to be δF↑=−1.60 PgC, while the Brazilian carbon flux change was −0.24 ± 0.11 PgC. This estimate is broadly within the uncertainty of previous aircraft-based estimates restricted to the Amazon basin. The 2011–2010 biomass burning change in Brazil was −0.24 ± 0.036 PgC, which implies a near-zero 2011–2010 change of the net ecosystem production (NEP): The near-zero NEP change is the result of quantitatively comparable increases GPP (0.31 ± 0.20 PgC) and respiration in 2011. Comparisons between Brazilian and global component carbon flux changes reveal complex interactions between the processes controlling annual land-atmosphere CO2 exchanges. These results show the potential of multiple satellite observations to help quantify and spatially resolve the response of productivity and respiration fluxes to climate variability.
Article
NASA’s Orbiting Carbon Observatory-2 (OCO-2) mission was motivated by the need to diagnose how the increasing concentration of atmospheric carbon dioxide (CO2) is altering the productivity of the biosphere and the uptake of CO2 by the oceans. Launched on 2 July 2014, OCO-2 provides retrievals of the column-averaged CO2 dry-air mole fraction () as well as the fluorescence from chlorophyll in terrestrial plants. The seasonal pattern of uptake by the terrestrial biosphere is recorded in fluorescence and the drawdown of during summer. Launched just before one of the most intense El Niños of the past century, OCO-2 measurements of and fluorescence record the impact of the large change in ocean temperature and rainfall on uptake and release of CO2 by the oceans and biosphere.
Data
The observation-based pCO2 fields were created using a 2-step neural network method extensively described and validated in Landschützer et al. 2013, 2014, 2016. The method first clusters the global ocean into biogeochemical provinces and in a second step reconstructs the non-liner relationship between CO2 driver variables and observations from the 4th release of the Surface Ocean CO2 Atlas (SOCATv4, Bakker et al. 2016). This file contains the resulting monthly pCO2 fields at 1°x1° resolution covering the global ocean with the exception of the Arctic Ocean and few marginal seas. The air-sea CO2 fluxes are computed from the air-sea CO2 partial pressure difference and a bulk gas transfer formulation following Landschützer et al. 2013, 2014, 2016. Furthermore, the monthly climatology is created from the monthly average of the period 1985-2015. (Website: https://www.nodc.noaa.gov/ocads/oceans/SPCO2_1982_2015_ETH_SOM_FFN.html)
Article
Inverse modeling of regional CO2 fluxes using atmospheric CO2 data is sensitive to the observational coverage of the observing network. Here we use the GEOS-Chem adjoint model to examine the sensitivity to CO2 fluxes of observations from the in situ surface network, the Total Carbon Column Observing Network (TCCON), the Greenhouse Gases Observing Satellite (GOSAT), and the Orbiting Carbon Observatory (OCO-2). We find that OCO-2 has high sensitivity to fluxes throughout the tropics and Southern Hemisphere, while surface observations have high sensitivity to fluxes in the northern extratropics throughout the year. For GOSAT viewing modes, ocean glint data provide the strongest constraints on fluxes in the tropics and Southern Hemisphere during Northern Hemisphere fall and winter relative to other viewing modes. In contrast, GOSAT nadir land data offer the greater sensitivity to fluxes in these regions during Northern Hemisphere spring and summer. For OCO-2 viewing modes, ocean glint data provided the dominant sensitivity to the surface fluxes in the northern subtropics, tropics, and Southern Hemisphere. We performed a series of inversion analyses using pseudodata and found that the varying sensitivities can result in large differences in regional flux estimates. However, combining measurements from different observing systems to exploit their complementarity may lead to a posteriori flux estimates with improved accuracy.