Terrestrial biosphere model performance for inter-annual variability of land-atmosphere CO2 exchange.

Abstract

Interannual variability in biosphere-atmosphere exchange of CO2 is driven by a diverse range of biotic and abiotic factors. Replicating this variability thus represents the ‘acid test’ for terrestrial biosphere models. Although such models are commonly used to project responses to both normal and anomalous variability in climate, they are rarely tested explicitly against inter-annual variability in observations. Herein, using standardized data from the North American Carbon Program, we assess the performance of 16 terrestrial biosphere models and 3 remote sensing products against long-term measurements of biosphere-atmosphere CO2 exchange made with eddy-covariance flux towers at 11 forested sites in North America. Instead of focusing on model-data agreement we take a systematic, variability-oriented approach and show that although the models tend to reproduce the mean magnitude of the observed annual flux variability, they fail to reproduce the timing. Large biases in modeled annual means are evident for all models. Observed interannual variability is found to commonly be on the order of magnitude of the mean fluxes. None of the models consistently reproduce observed interannual variability within measurement uncertainty. Underrepresentation of variability in spring phenology, soil thaw and snowpack melting, and difficulties in reproducing the lagged response to extreme climatic events are identified as systematic errors, common to all models included in this study.

Full text

Page 1

Terrestrial biosphere model performance for inter-annual
variability of land-atmosphere CO2exchange
T.F. KEENAN*, IAN BAKER†, ALAN BARR‡, PHILIPPE CIAIS§, KEN DAVIS¶, MICHAEL
DIETZE**, DANILLO DRAGONI††, CHRISTOPHER M GOUGH‡‡, ROBERT GRANT§§,
DAVID HOLLINGER¶¶, KOEN HUFKENS***, BEN POULTER†††, HARRY MCCAUGHEY‡‡‡,
BRETT RACZKA§§§, YOUNGRYEL RYU¶¶¶, KEVIN SCHAEFER****, HANQIN TIAN††††,
HANS VERBEECK‡‡‡‡, MAOSHENG ZHAO§§§§ and ANDREW D. RICHARDSON*
*Department of Organismic and Evolutionary Biology, Harvard University, Cambridge, MA USA, †Atmospheric Science
Department, Colorado State University, Fort Collins, CO USA, ‡Climate Research Division, Atmospheric Science and Technology
Directorate, Saskatoon, Saskatchewan Canada, §Laboratoire des Sciences du Climat et de l’Environnement, CE Orme des Merisiers,
Gif sur Yvette, France, ¶Department of Meteorology, Pennsylvania State University, University Park, PA USA, **Department of
Plant Biology, University of Illinois-Urbana Champaign, Urbana, IL USA, ††Department of Geography, Indiana University,
Bloomington, IN USA, ‡‡Department of Biology, Virginia Commonwealth University, VA USA, §§Department of Renewable
Resources, University of Alberta, Edmonton, Alberta Canada, ¶¶Northern Research Station, USDA Forest Service, Durham, New
Hampshire USA, ***Department of Geography and Environment, Boston University, MA USA, †††Laboratoire des Sciences du
Climat et de l’Environnement, CE Orme des Merisiers, Gif sur Yvette, France, ‡‡‡Department of Geography, Queen’s University,
Kingston, Ontario Canada, §§§Department of Meteorology, Penn State University, Pennsyvannia, NY USA, ¶¶¶Department of
Landscape Architecture and Rural System Engineering, Seoul National University, South Korea, ****National Snow and Ice Data
Center, University of Colorado at Boulder, Boulder, CO USA, ††††School of Forestry and Wildlife Sciences, Auburn University,
Auburn, AL USA, ‡‡‡‡Laboratory of Plant Ecology, Ghent University, Ghent, Belgium, §§§§Department of Ecosystem and
Conservation Sciences, University of Montana, Missoula, MT USA
Abstract
Interannual variability in biosphere-atmosphere exchange of CO2is driven by a diverse range of biotic and abiotic
factors. Replicating this variability thus represents the ‘acid test’ for terrestrial biosphere models. Although such
models are commonly used to project responses to both normal and anomalous variability in climate, they are rarely
tested explicitly against inter-annual variability in observations. Herein, using standardized data from the North
American Carbon Program, we assess the performance of 16 terrestrial biosphere models and 3 remote sensing prod-
ucts against long-term measurements of biosphere-atmosphere CO2exchange made with eddy-covariance flux tow-
ers at 11 forested sites in North America. Instead of focusing on model-data agreement we take a systematic,
variability-oriented approach and show that although the models tend to reproduce the mean magnitude of the
observed annual flux variability, they fail to reproduce the timing. Large biases in modeled annual means are evident
for all models. Observed interannual variability is found to commonly be on the order of magnitude of the mean
fluxes. None of the models consistently reproduce observed interannual variability within measurement uncertainty.
Underrepresentation of variability in spring phenology, soil thaw and snowpack melting, and difficulties in repro-
ducing the lagged response to extreme climatic events are identified as systematic errors, common to all models
included in this study.
Keywords: biosphere-atmosphere interaction, carbon fluxes, carbon sequestration, interannual variability, process-based
models, remote sensing
Received 17 January 2012; revised version received 17 January 2012 and accepted 15 February 2012
Introduction
The terrestrial biosphere acts as a net sink for atmo-
spheric CO2, with global forests absorbing on average
4 Pg C yr?1(Pan et al., 2011), which, excluding defores-
tation, offsets roughly half of all anthropogenic emis-
sions from fossil fuel burning and cement production
(Pan et al., 2011). Interannual variability in this sink is
often on the order of magnitude of the mean (e.g., Zeng
et al., 2005; Reichstein et al., 2007; a; Pan et al., 2011), and
drives interannual variability in the growth rate of atmo-
spheric CO2(Bousquet et al., 2000; Knorr et al., 2007).
Carbon fluxes in forest ecosystems are tightly coupled to
climate (Richardson et al., 2007; Piao et al., 2008; Chen
et al., 2009; Dragoni et al., 2011), and anomalous climatic
Correspondence: T.F. Keenan, tel. + 1 617 496 0825,
fax + 1 617 495 9484, e-mail: tkeenan@oeb.harvard.edu
© 2012 Blackwell Publishing Ltd
1971
Global Change Biology (2012) 18, 1971–1987, doi: 10.1111/j.1365-2486.2012.02678.x

Page 2

signals generally drive the observed variability in their
sink strength (Dunn et al., 2007; Desai, 2010; le Maire
et al., 2010). Such signals tend to affect photosynthesis
and respiration (the two processes which determine net
ecosystem carbon exchange) to different extents (Rich-
ardson et al., 2007; Luyssaert et al., 2007), and therefore
provide an excellent test-bed to assess the skill of terres-
trial biosphere models.
Terrestrial biosphere models are the primary tools
used for predicting the impact of climate variability on
terrestrial carbon fluxes. Built around the philosophy
that a blend of mechanistic and semi-empirical descrip-
tions can capture functional responses to environmen-
tal drivers, they have been used in conjunction with
remote sensing products (Zhao & Running, 2010) and
data mining tools (Papale & Valentini, 2003) to provide
regional and global estimates of terrestrial carbon
cycling (e.g., Friend, 2010; Beer et al., 2010). They are
also commonly used to quantify terrestrial responses to
climatic variability, including anomalies, and extreme
events (Ciais et al., 2005; Richardson et al., 2007; Vetter
et al., 2008; Zhao & Running, 2010). Future model pro-
jections of the response of terrestrial carbon cycling to
climate change (Heimann & Reichstein, 2008) are neces-
sary to inform policy (IPCC, 2007), although current
models show very divergent sensitivity to long-term
changes in climate (Friedlingstein et al., 2006).
Biogeochemical models are often shown to capture
diel and seasonal dynamics reasonably well (e.g.,
Braswell et al., 2005). This is not surprising, given the
pronounced diurnal and seasonal cycles of climatic
drivers. Over yearly and longer time scales, however,
studies show poor model performance at reproducing
gross fluxes and carbon budgets (e.g., Hanson et al.,
2004; Braswell et al., 2005; Siqueira et al., 2006; Richard-
son et al., 2007; Urbanski et al., 2007). Such comparison
studies are typically restricted to a limited number of
models and sites, and a relatively short time series
length. Nonetheless, the results suggest that although
the response of terrestrial ecosystems to mean climatic
drivers is relatively well captured, sensitivity to the
impact of variability in climatic drivers may not be,
leading to the accumulation of high frequency model
error (e.g., Dietze et al., 2012) over longer time scales
(Schwalm et al., 2010). No study, however, has yet iden-
tified systematic errors in model sensitivity to climatic
variability.
In this analysis, we use 16 terrestrial biosphere mod-
els and 3 remote sensing products, along with eddy-
covariance data from a range of sites included in the
North American Carbon Program interim site synthe-
sis, to assess model ability to reproduce observed vari-
ability in carbon fluxes. We examine the frequency
distribution of temporal anomalies in net ecosystem
exchange (NEE), gross primary productivity (GPP),
and ecosystem respiration (RE), for two plant func-
tional types. We first assess individual model perfor-
mance on an annual/interannual scale. As interannual
variability can be driven by ‘critical’ periods within a
year (le Maire et al., 2010), we examine monthly
systematic model errors (errors consistent across all
models and sites). By using data from sites with a
regionally coherent anomalous year, we then assess
the possible role of extreme within-year climatic
events and lagged effects on model performance for
interannual variability in terrestrial carbon cycling.
Methods
All models and data used were obtained through the North
American Carbon Program (NACP) (http://www.nacarbon.
org/nacp/). To allow for an ensemble approach and reduce
the potential for spurious variability, we selected only sites
with at least 5 years of data, from plant functional types that
were represented by at least three such sites. This resulted in a
total of 11 forested sites distributed through North America
(Table 1). Of those, six were deciduous broadleaf, and five
evergreen needleleaf. This gave a total of 91 site-years for the
analysis.
Eddy-covariance flux data (produced by AmeriFlux and
Fluxnet-Canada investigators) from the 11 selected sites was
processed according to a common protocol from the NACP
site level interim synthesis (http://www.nacarbon.org/nacp/).
The observed NEE were corrected for storage, despiked (i.e.,
outlying values removed), and filtered to remove conditions
of low turbulence (friction velocity filtered). Flux error esti-
mates were calculated (Barr et al., 2009) by combining random
uncertainty (calculated following Richardson & Hollinger
(2007)) and uncertainty due to the selection of the friction
velocity threshold (Barr et al., 2009). Observed monthly and
annual NEE values were then calculated using gap-filled data
from each site (Barr et al., 2009). The gap-filled NEE values
were also partitioned to gross ecosystem photosynthesis
(GPP) and ecosystem respiration (RE). Multiple approaches
were used to quantify additional uncertainty introduced by
the partitioning (Desai, 2010; Barr et al., 2009).
Gaps in the meteorological forcing data occurred due to
instrument failure or quality control. Such gaps were filled
using the nearest available climate station in the National Cli-
matic Data Center’s Global Surface Summary of the Day
(NCDC-GSOD) database. Gaps at sites where no such data
were available were filled using DAYMET (Thornton et al.,
1997). Daily NCDC-GSOD and DAYMET data were temporar-
ily downscaled to hourly or half-hourly values (see Riccuito
et al., 2009fordetails) (http://nacp.ornl.gov/docs/Site_
Synthesis_Protocol_v7.pdf).
Sixteen terrestrial biosphere models (Table 2) were run at
the sites for the period of available measurements (Table 1).
The terrestrial biosphere models simulated carbon cycling
with process-based formulations of varying detail for the
component carbon fluxes of photosynthesis and respiration.
© 2012 Blackwell Publishing Ltd, Global Change Biology, 18, 1971–1987
1972 T.F. KEENAN et al.

Page 3

Table 1
Summary of sites used
Country-site
Name
Lat.
Long.
Elevation
(m.a.s.l.)
Years
Biome
Mean
NEE
d NEE
NEE
error
Mean
GPP
d GPP
GPP
error
Mean
RE
d RE
RE
error
Reference
CA-Ca1
BC, Campbell
river –Mature
forest site
49.87
?125.33
300
1998–2005
EVG
?244
65
32
2310
112
44
2065
130
72
Schwalm
et al., 2007;
CA-Oas
Sask. –SSA Old
aspen
53.63
?106.20
530
1997–2005
DBF
?158
100
15
1090
150
24
932
86
35
Barr et al., 2004;
CA-Obs
Sask. –SSA Old
black spruce
53.99
?105.12
629
2000–2005
EVG
?56
20
8
795
55
14
738
41
19
Griffis et al., 2003;
CA-Ojp
Sask. –SSA Old
jack pine
53.92
?104.69
579
2000–2005
EVG
?30
26
9
612
48
16
582
30
22
Griffis et al., 2003;
CA-TP4
Ontario –
Turkey point
Mature white
pine
42.71
?80.36
219
2001–2005
EVG
?133
72
16
1391
84
20
1258
75
30
Peichl &
Arain, 2007;
US-Ha1
MA –Harvard
forest EMS
tower (HFR1)
42.54
?72.17
303
1992–2005
DBF
?217
117
36
1409
156
87
1192
125
78
Urbanski
et al., 2007;
US-Ho1
ME –Howland
forest (Main
tower)
45.20
?68.74
60
1996–2004
EVG
?223
53
18
1518
90
22
1295
82
33
Richardson
et al., 2009;
US-MMS
IN –Morgan
Monroe state
forest
39.32
?86.41
275
1999–2005
DBF
?348
37
23
1331
56
26
983
68
31
Schmid
et al., 2000;
US-NR1
CO –Niwot
Ridge forest
(LTER NWT1)
40.03
?105.55
3050
1998–2005
EVG
?37
25
15
804
54
25
767
52
38
Bradford
et al., 2008;
US-Pfa
WI –Park falls/
WLEF
45.95
?90.27
485
1995–2005
DBF
45
26
21
1005
49
35
1050
35
45
Davis et al., 2003;
US-UMB
MI –University
of Michigan
biological
station
45.56
?84.71
234
1999–2005
DBF
?132
65
22
1189
43
32
1057
87
42
Schmid et al., 2003
© 2012 Blackwell Publishing Ltd, Global Change Biology, 18, 1971–1987
MODELING INTERANNUAL LAND CO2FLUX
1973

Page 4

Table 2
Summary of model characteristics
Attribute
Model
BEPS
BIOME-BGC
Can-IBIS
CNCLASS
DLEM
ECOSYS
ED2
EDCM
ISAM
LoTEC-DA
Temporal
resolution
Daily
Daily
Half-hourly
Half-hourly
Daily
Hourly
Half-hourly
Monthly
Weekly
Half-hourly
Vegetation
pools
4
7
3
4
6
9
6
8
5
4
Soil Pools
9
4
7
3
3
9
4
5
8
5
Canopy
phenology
Semi-prognostic
Prognostic
Prognostic
Prognostic
Semi-
prognostic
Prognostic
Prognostic
Prognostic
Prognostic
Gross primary
photosynthesis
(GPP)
Enzyme kinetic
model
Stomatal
conductance
model
Enzyme kinetic
model
Enzyme kinetic
model
Stomatal
conductance
model
Enzyme
kinetic
model
Enzyme
kinetic
model
Light use
efficiency
Enzyme
kinetic
model
Enzyme
kinetic
model
Heterotrophic
respiration
(HR)
Temperature
(air + soil)
precipitation
soil
evaporation
soil carbon +
nitrogen
Soil
temperature
soil moisturesoil carbon
First or
greater
order model
First or greater
order model
Air
temperature
soil
temperature
litter and soil
carbon soil
nitrogen soil
moisture
Dissolved
carbon loss
soil
temperature
soil moisture shortwave &
long wave
radiation soil
carbon
vegetation
carbon soil
nitrogen leafnitrogen
Soil
temperature
soil moisture soil carbonsoil
nitrogen
Soil
temperature
soil moisture soil carbon
dissolved
carbon loss
vegetation
carbon soil
nitrogen
First or
greater
order
model
Soil
temperature
soil
moisture
soil carbon
Autotrophic
respiration
(AR)
Air
temperature
GPP
Air
temperature
vegetation
carbon leaf
nitrogen
Air
temperature
soil
temperature
precipitation
soil moisture
incident
shortwave +
long wave
radiation
vegetation
carbon
Fraction of
instantaneous
GPP
Air
temperature
vegetation
carbon leaf
nitrogen GPP
Air
temperature
soil
temperature
vegetation
carbon leaf
nitrogen
Air
temperature
soil
temperature
vegetation
carbon leaf
nitrogen GPP
Proportional
to growth
Proportional
to growth
Air
temperature
soil
temperature
soil
moisture
vegetation
carbon GPP
Ecosystem
respiration
AR+HR
Air
temperature
soil
temperature
AR + HR
AR + HR
AR + HR
AR + HR
AR + HR
AR + HR
AR + HR
AR + HR
© 2012 Blackwell Publishing Ltd, Global Change Biology, 18, 1971–1987
1974 T.F. KEENAN et al.

Page 5

Table 2 (continued)
Attribute
Model
BEPS
BIOME-BGC
Can-IBIS
CNCLASS
DLEM
ECOSYS
ED2
EDCM
ISAM
LoTEC-DA
soil moisture soil carbon
vegetation
carbon
Net primary
productivity
(NPP)
GPP–AR
Shortwave
radiation
vapor
pressure
deficit CO2
vegetation
carbon leaf
nitrogen
GPP–AR
Fraction of
instantaneous
GPP
GPP–AR
GPP–AR
GPP–AR
Air
temperature
precipitation
soil carbonsoil nitrogensoil moisture
vegetation
carbon leaf
nitrogen
GPP–AR
GPP–AR
Net
ecosystemexchange
NPP–HR
Soil
temperature
soil moistureshortwave
radiation
vapor
pressure
deficit
NPP–HR
GPP–R
NPP–HR
GPP–R
NPP–HR
NPP–HR
NPP–HR
Reference
Liu et al.,
1999;
Thornton
et al., 2005;
Williamson
et al., 2008;
Arain
et al., 2006;
Tian
et al., 2010;
Grant et al.,
2005;
Medvigy
et al.,
2009;
Liu
et al., 2003;
Yang et al.,
2009;
Hanson
et al., 2004;
Attribute
Model
Remote Sensing
LPJml
ORCHIDEE
SiB
SiB-CASA
SSiB2
TECO
BESS
MODISc5
MODISc5.1
Temporal
resolution
Daily
Half-hourly
Half-hourly
10 min
Half-hourly
Hourly
Daily
Daily
Daily
Vegetation
pools
3
8
0
8
0
3
n/a
n/a
n/a
Soil pools
2
8
0
5
0
5
n/a
n/a
n/a
Canopy
phenology
Prognostic
Prognostic
Prescribed
Prescribed
Prescribed
Prognostic
Prescribed
Prescribed
Prescribed
Gross primary
photosynthesis
(GPP)
Stomatal
conductance
model
Enzyme
kinetic
model
Enzyme
kinetic
model
Enzyme
kinetic
model
Stomatal
conductance
model
Stomatal
conductance
model
Enzyme
kinetic
model
Light-use-
efficiency
model
Light-use-
efficiency
model
© 2012 Blackwell Publishing Ltd, Global Change Biology, 18, 1971–1987
MODELING INTERANNUAL LAND CO2FLUX
1975

Page 6

Table 2 (continued)
Attribute
Model
Remote Sensing
LPJml
ORCHIDEE
SiB
SiB-CASA
SSiB2
TECO
BESS
MODISc5
MODISc5.1
Heterotrophic
respiration (HR)
Soil
temperature
soil moisturesoil carbon
Soil temperature
soil
moisture
soil carbon
Zero-order
model
Soil
temperature
soil moisture soil carbon
Zero-order
model
First or greater
order model
n/a
n/a
n/a
Autotrophic
respiration (AR)
Air temperature
soil moisture
vegetation
carbon
Air temperature
vegetation
carbon
Fraction of
instantaneous
GPP
Air temperature
soil moisture
vegetation
carbon
Air temperature
soil moisturesurface
incident
shortwave
radiation relative
humidity LAI
fPAR CO2
forced annual
balance
Air temperature
vegetation
carbon
n/a
n/a
n/a
Ecosystem
respiration
AR + HR
AR + HR
Forced Annual
Balance
AR + HR
GPP–AR
AR + HR
n/a
n/a
n/a
Net primary
productivity
(NPP)
GPP–AR
GPP–AR
GPP–AR
Air temperature
soil moisture
CO2relative
humidity
GPP–R
GPP–AR
GPP
GPP
GPP
Net ecosystem
exchange
NPP–HR
GPP–R
GPP–R
GPP–R
Zero-order
model
GPP - R
n/a
n/a
n/a
Reference
Sitch
et al., 2003;
Krinner
et al., 2005;
Baker
et al., 2008;
Schaefer
et al., 2009;
Zhan
et al., 2003;
Weng & Luo,
2008;
Ryu et al.,
2011;
Running
et al., 2004;
Zhao
et al., 2005
© 2012 Blackwell Publishing Ltd, Global Change Biology, 18, 1971–1987
1976 T.F. KEENAN et al.

Page 7

Simulated NEE was based on model specific runs using gap
filled observed weather at each site and locally observed val-
ues of soil texture according to a standard protocol (Riccuito
et al., 2009). Each model used species or plant functional type
specific parameterizations as defined by the individual model
teams, with the exception of LoTEC where parameters were
optimized using data assimilation (Riccuito et al., 2008). Three
remote sensing products of terrestrial gross primary produc-
tivity (MODISc5 (Running et al., 2004), MODISc5.1 (Zhao
et al., 2005), BESS (Ryu et al., 2011)), not included in the North
American Carbon Program, were also used to provide annual
estimates of GPP for each site.
To assess interannual variability, we normalized the mea-
sured/modeled values of NEE, GPP, and RE by subtracting
the long-term calendar year measured/modeled mean for
each site from individual site-year flux values, giving Fobsand
Fsimfor each flux and year. By comparing the long-term calen-
dar year mean of measured and modeled fluxes, we also iden-
tified biases in model estimates of NEE, GPP and RE. Model-
data agreement for interannual variability in annual flux sums
was assessed in terms of the normalized root mean squared
error (NRMSE) and the v2statistic.
The NRMSE is the root mean square error of model-data
mismatch normalized by the magnitude of observed variabil-
ity at each site:
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
NRMSEijk¼
1
n
PlðFobs? FsimÞ2
rðFobsÞ
q
ð1Þ
where F represents the observed (obs) or modeled (sim) value
of a particular flux, i (NEE, GPP, or RE), for a particular year,
l. Note that each flux here is represented as the interannual
variability (Fobsand Fsim), not the mean flux. r(Fobs) is the
standard deviation of observed interannual variability at site
k. NRMSE values are calculated for each model j at site k. The
NRMSE thus reports the mean difference between the simu-
lated and observed flux, relative to the variability in the
observed flux.
The v2statistic complements the NRMSE by incorporating
measurement error. Here it is calculated for each model and
PFT as the squared residual between paired model and data
points for each flux (after normalization to the long-term mean
as described above), relative to the observational error:
?
v2
ijk¼1
n
X
ijkl
Fobs? Fsim
2dðFobsÞ
?2
ð2Þ
where d(Fobs) is uncertainty related to the annual observed
value of that flux, 2 normalizes the uncertainty in the observed
flux to correspond to a 95% confidence interval. A v2value of
less than one indicates that the model agrees with the data rel-
ative to data uncertainty. That is interannual variability for a
model with a v2value of less than one will always fall within
one standard deviation of data error. Above one, the v2scales
model error relative to observation uncertainty.
Interannual variability in observed fluxes commonly stems
from specific, short-lived, periods of anomalous fluxes within
the year (Krishnan et al., 2008, 2009; Chen et al., 2009; le Maire
et al., 2010). We therefore also assessed model performance for
variability on a monthly scale. The variability of monthly
fluxes between years was calculated in the same way as
annual variability, as the difference between the observed or
modeled monthly value and the associated long-term mean
for the month in question.
By differencing the observed and predicted monthly
variability (herein termed variance residuals) specific periods
during the year at which the models under- or over-repre-
sent the observed monthly variability can be identified. We
define periodsofsystematic
common to all models) as times when all models show the
same-signed bias in variance residuals with 95% confidence.
We also assess persistent biases, which are mean biases of
more than 1 month in duration that are not necessarily
systematic.
Extreme climatic events, detectable as regionally coherent
deviations outside the normal range of variability, provide a
strong test of model performance. We identified one such
event in our database. At three sites in mid-western Canada,
mean spring monthly temperatures in 2002 were between 8
and 10 °C below the long-term mean. We used this event to
assess model skill and to identify systematic model error.
modelerror(statistically
Results
Biases and the magnitude of variability
In order to quantify interannual variability, we normal-
ized all models and data by subtracting respective
mean annual totals from individual annual totals. This
process identified considerable biases in model esti-
mates of all total annual fluxes at all sites (Fig. 1). In
particular, biases in annual NEE were commonly of
similar magnitude to mean observed annual NEE
fluxes. The majority of models were biased toward
underestimating ecosystem carbon uptake for both
deciduous and evergreen sites (Fig. 1). Note that biases
herein are reported relative to the observed mean NEE
for each site, and therefore have the potential to be par-
ticularly larger for sites with low mean annual NEE.
See Table 1 for per-site mean annual NEE values.
The magnitude of modeled interannual variability in
each annual flux was on average of the same order of
that observed (Fig. 2). A large range in model perfor-
mance was evident (Table 3), but in general, the models
proved ‘flexible’ enough to reproduce the observed
range of variability. Observed interannual variability in
NEE for deciduous broadleaved forests was twice that
of evergreen needle-leaved forests, a distinction only
reproduced by six of the included models. The magni-
tude of interannual variability in both GPP and RE was
greater (55%, 23%) in deciduous broadleaved forests
than in evergreen needleleaved forests. The remote
sensing products, however, consistently predicted
higher GPP variability in evergreen than in deciduous
forests.
© 2012 Blackwell Publishing Ltd, Global Change Biology, 18, 1971–1987
MODELING INTERANNUAL LAND CO2FLUX
1977

Page 8

Fig. 1 Mean model bias (Modeled-Measured, gC m?2yr?1) over all years when compared to annual gap-filled observations of net eco-
system exchange (NEE), gross primary productivity (GPP), and ecosystem respiration (RE). Results are grouped by two plant func-
tional types (EVG: Evergreen needleleaf forest; DBF: Deciduous broadleaf forest). The right panel is the normalized frequency
distribution of model biases grouped by plant functional type, showing the distribution of values in the bar charts on the left, reported
as a total bias for NEE, and a percentage of the annual total ([Modeled-Measured]/Measured) for GPP and RE. X-axis ranges are trun-
cated to represent only observed range of biases.
© 2012 Blackwell Publishing Ltd, Global Change Biology, 18, 1971–1987
1978 T.F. KEENAN et al.

Page 9

Statistical performance of models on an interannual scale
Although the mean magnitude of model variability on
the interannualscale was
observed magnitude of variability (Fig. 2), all models
fell outside the data error of the observed for individ-
ual site-years (>1 v2, Fig. 3, S1). This means that the
general magnitude of interannual variability was well
reproduced, but not the timing. Interannual variability
similar tothemean
in the annual net ecosystem exchange of evergreen
forests was better simulated on average than decidu-
ous forests (Fig. 3). A larger range of model perfor-
mance was observed for variability in annual GPP
than that of RE. Our results suggest that on average
the inability of models to match the timing of
observed variability in GPP is the main cause of errors
in the simulation of interannual variability in NEE,
although this is very model-dependent (Fig. 3). The
Fig. 2 The distribution of the magnitude of interannual variability (IAV) in annual totals of net ecosystem exchange (NEE), gross pri-
mary production (GPP), and ecosystem respiration (RE), over all sites for each model (gray lines) and the data (black line). See Table 3
for individual model values.
Table 3
mary productivity (GPP) and ecosystem respiration (RE). Sites are grouped as deciduous broadleaved forest or evergreen needle-
leaved forest. Mean error of observed annual sums is also given for reference (in brackets). IAV: Interannual variability. Model
codes, given in brackets, are those used in Fig. 3
Mean standard deviation of observed and modeled interannual variability for net ecosystem exchange (NEE), gross pri-
NEEGPP RE
DBF EVGDBFEVGDBFEVG
Observed IAV
Annual error
Models
BEPS (A)
Biome-BGC (B)
CanIBIS (C)
CNCLASS (D)
DLEM (E)
EDLUEEDCM (F)
ECOSYS (G)
ED2 (H)
ISAM (I)
LoTEC-DA (J)
LPJml (K)
ORCHIDEE (L)
SiB (M)
SiBCASA (N)
SSIB2 (O)
TECO (P)
BESS (Q)
MODISc5 (R)
MODISc5.1 (S)
85.53
(24.73)
44.49
(17.12)
115.05
(47.73)
74.30
(24.54)
91.56
(50.89)
74.32
(37.50)
213.54
59.66
57.70
76.98
81.57
169.20
47.79
134.66
40.36
77.99
71.36
31.34
39.28
65.57
59.43
17.69
110.54
115.54
91.42
55.87
202.73
240.33
110.59
156.14
178.31
121.13
127.76
175.23
33.23
54.86
74.59
94.41
104.32
82.17
48.92
69.16
129.05
78.94
55.84
81.54
93.72
64.85
23.85
98.10
56.05
145.35
63.33
32.87
40.53
66.23
57.15
178.26
136.14
52.06
234.33
103.42
66.11
68.36
224.81
103.86
86.59
54.08
85.04
78.51
87.81
39.41
59.75
43.23
64.58
14.79
88.26
72.97
93.58
0.28
37.93
93.55
64.32
92.03
104.96
26.33
0.54
38.81
53.50
38.87
57.77
86.87
113.73
33.14
65.39
124.01
121.13
103.90
93.58
47.96
32.90
56.86
95.89
54.89
© 2012 Blackwell Publishing Ltd, Global Change Biology, 18, 1971–1987
MODELING INTERANNUAL LAND CO2FLUX
1979

Page 10

MODISc5 remote sensing product performed worse
than the average model (Fig. 3, Fig S2). The MO-
DISc5.1 data product proved to be a large improve-
ment over the MODISc5 GPP product. The BESS
remote sensing product, a process-based model inter-
pretation of remotely sensed data (Ryu et al., 2011),
performed better than either MODIS product for
deciduous forests, although that was not the case for
evergreens. Although process-based models of differ-
ent types were represented (e.g., light use efficiency
vs. enzyme kinetic model for GPP, Table 2) no model
characteristic performed statistically better than any
other (data not shown). This could be due to the lim-
ited number of models with particular characteristics.
Variability within the year
The models showed persistent systematic biases (see
definition in Methods section) for monthly flux variabil-
ity. In deciduous forests, models consistently underesti-
mated monthly variability in NEE throughout spring
(May and June) (Fig. 4). Model underestimation of vari-
ability in deciduous spring NEE fluxes was mostly due
to underrepresentation of variability in spring GPP
(Fig. 4). A smaller peak in the deciduous GPP variance
residuals (predicted monthly variability – observed
monthly variability) was also evident in September and
October. Variability in deciduous RE showed no bias
that was consistent across all models.
Systematic underrepresentation of monthly variabil-
ity during May was also evident for evergreen forests
(Fig. 4). Here, however, model error for NEE was dom-
inated by the lack of variability in modeled RE during
spring. Although evergreen forests do not exhibit the
marked phenological transitions observed in deciduous
forests, all evergreen forests included in this study
maintain a large snowpack throughout winter. Persis-
tent, non-systematic biases were evidenced throughout
the year, in particular an overestimation of winter
variability in evergreen NEE and GPP, and a persistent
Fig. 3 Statistical comparison (on a log-log scale) of model performance (normalized root mean square error vs. v2statistic) for interan-
nual variability (IAV) in annual totals of net ecosystem exchange (NEE), gross primary productivity (GPP) and ecosystem respiration
(RE) for the two plant functional types (DBF (green): Deciduous broadleaved forests; EVG (blue): Evergreen needleleaf forests). See
supplementary material for graphs with error bars (Fig. S1, S2).
© 2012 Blackwell Publishing Ltd, Global Change Biology, 18, 1971–1987
1980 T.F. KEENAN et al.

Page 11

underestimation of variability in evergreen RE during
the summer.
Response to anomalous climate forcing
Three sites (CA-Ojp, CA-Obs, CA-Oas; see site descrip-
tion Table 1) experienced a regionally coherent extreme
climatic event during 2002, where monthly mean tem-
peratures were between 8 (CA-Ojp, CA-Obs) and 10 °C
(CA-Oas) below the long-term mean. The anomaly lar-
gely affected canopy GPP at all three sites, and to a les-
ser extent RE (Fig. 5). At CA-Ojp and CA-Obs,
anomalously low temperatures during the month of
April lowered observed GPP by more than twice the
normal range of variability for that month (Fig. 5). The
models accurately captured the drop in productivity,
with the mean of all model projections capturing both
the sign and the magnitude of the April GPP anomaly
Fig. 4 Residuals (predicted-observed) of monthly variability in net ecosystem exchange (NEE), gross primary productivity (GPP), and
ecosystem respiration (RE). Positive values indicate a higher variability in the observations than in a model. The mean model-data
residuals are presented as a dashed line. The gray area represents the standard deviation about the mean. Values represent averages
over all sites for each plant functional type, and all sites taken together. Model codes – A: BEPS, B: Biome-BGC, C: CanIBIS, D:
CNCLASS, E: DLEM, F: EDLUEEDCM, G: ECOSYS, H: ED2, I: ISAM, J: LoTEC-DA, K: LPJml, L: ORCHIDEE, M: SiB, N: SiBCASA, O:
SSiB2, P: TECO, Q: BESS, R: MODISc5, S: MODISc5.1, X: Mean
© 2012 Blackwell Publishing Ltd, Global Change Biology, 18, 1971–1987
MODELING INTERANNUAL LAND CO2FLUX
1981

Page 12

at both sites. The temperature anomaly was observed at
CA-Oas 1 month later, and again the models accurately
reproduced the observed magnitude in anomalous
GPP. During the following May, June, and July,
observed temperature remained colder than normal but
returned to within the normal range of variability for
the three sites. Observed GPP, however, remained
anomalously low during those months and did not
return to within the normal range of variability until
July at each site. This lagged effect between anomalous
climate forcing and resulting fluxes was not well
reproduced by the models. At CA-Obs, all models
returned to within the normal range of GPP variability
in the month directly following the temperature
Fig. 5 Monthly modelled and observed anomalies (mean model: green dashed line; standard deviation of models: green shaded area;
observed: solid black line) in gross primary production (GPP), ecosystem respiration (RE), and net ecosystem exchange (NEE) for the
year 2002 at three sites (Ca-Ojp, Ca-Obs, Ca-Oas) which exhibit a regionally coherent anomaly in that year. The standard deviation of
normal observed monthly variability is presented as the gray area with dark gray columns. The duration of the anomalous temperature
event is shown in the crosshatched gray area for each site.
© 2012 Blackwell Publishing Ltd, Global Change Biology, 18, 1971–1987
1982 T.F. KEENAN et al.

Page 13

anomaly. The same behavior was apparent at CA-Ojp
and CA-Oas, although the average model GPP esti-
mates remained just outside the normal range of vari-
abilitydueto persistent
extended period of low productivity in CA-Oas may be
in part also due to consistently low precipitation during
the year.
A similar, although smaller anomaly pattern was
observable for RE (Fig. 5). Low spring temperatures
caused a prolonged anomaly of low ecosystem respira-
tion. The models tended to overestimate the reduction
in RE as a result of the colder temperatures. After the
initial anomaly, RE as estimated by eddy-covariance
measurements took a few months to return to within
the normal range of variability. Modeled RE quickly
returned to ‘normal’ at CA-Ojp and CA-Obs. Temporal
dynamics at CA-Oas differed from those of the other
two sites due to the additional pressure of persistently
low temperatures and precipitation during the year.
lowtemperatures.The
Discussion
This analysis has shown that, although capable of
reproducing the magnitude of interannual variability,
terrestrial biosphere models are not consistent with the
timing of observations of interannual variability in sur-
face-atmosphere exchanges of CO2at mid-latitude for-
ests over North America. By examining interannual
variability in measured and modeled monthly fluxes,
we show that all the models used for the NACP interim
site synthesis systematically fail to reproduce observed
variability during spring. Underestimation of spring
variability is largest for GPP in deciduous forests, and
RE for evergreens, suggesting different processes may
be responsible for plant functional type specific model
error.
It has been shown that terrestrial biosphere models
are typically unable to adequately explain the observed
interannual variability in deciduous canopy phenology
(Richardson et al., 2012), and that variability in spring
GPP often drives observed interannual variability in
net ecosystem exchange (Krishnan et al., 2008, 2009).
Here we show that this is a systematic cause of the low
agreement between modeled and observed interannual
variability in terrestrial carbon fluxes.
In a similar fashion, it has been shown that the cur-
rent available models of snow pack dynamics perform
poorly for both spatial and interannual variability. Rut-
ter et al. (2009) tested 33 models of snowpack dynamics
across a range of sites, and found that although a model
could perform well when tuned to a particular site-
year, this did not transfer to good performance for
other years at the same site, or other sites. Interactions
between snowmelt, soil thaw and water table depth are
known to directly affect interannual variability in NEE
(Goulden et al., 1998; Dunn et al., 2007; Hu et al., 2010).
Results here suggest that this may be a direct system-
atic contributor to the low agreement between observed
and modeled interannual variability in net ecosystem
carbon exchange, in particular for evergreen sites.
These results do not imply, however, that a lack of phe-
nological variability in canopy or soil dynamics are nec-
essarily the main culprits for the lack of agreement
between the observations and output from any one
model, as individual models showed large persistent
biases at other times of the year (Fig. 3).
The remote sensing products performed comparably
to the average process-based model when assessed
against interannual variability in GPP. The MODISc5.1
data set is a post-processed version of the MODISc5 data
set where corrections are made for poor quality driver
data (Zhao et al., 2005). The remote sensing products,
which are themselves models, are driven by a global
daily meteorological reanalysis dataset not site-specific
meteorology and the uncertainties in the meteorological
reanalysis can introduce biases in GPP estimates (Zhao
et al., 2006). Although estimates of GPP based on remote
sensing have been used to evaluate process-based
models (e.g., Poulter et al., 2009), results herein suggest
that estimates of interannual variability from both
approaches are subject to similar magnitudes of error.
Although there was a general tendency for the mod-
els to persistently underestimate flux variability in
summer, it should be noted that the flux data are sub-
ject to random error roughly in proportion to the size of
the flux (Richardson & Hollinger, 2007; Richardson
et al., 2008). Even if the model were perfect, modeled
variability should be smaller than that observed. Car-
bon fluxes are typically higher in the summer, and sub-
jectto larger uncertainty.
variability in the data during summer could therefore
be due to random errors in the flux measurements
generating larger variability in monthly totals.
The analysis of model responses to the regionally
coherent climatic anomaly of spring 2002 suggests that
models have the potential to correctly reproduce the
magnitude of instantaneous biological response to
climate anomalies (Desai, 2010). Although the models
accurately captured the direct effect of an isolated
climate extreme, the models included here failed to
accurately reproduce lagged effects of climate anoma-
lies on both gross primary production and ecosystem
respiration. Lagged effects of climate variability on eco-
system function have previously been reported (Gough
et al., 2009), and our results suggest that such lagged
effects are not well incorporated into models. The nat-
ure of such lagged effects depends on the type of cli-
matic anomaly. Spring frosts (Gu et al., 2008; Marino
Theapparenthigher
© 2012 Blackwell Publishing Ltd, Global Change Biology, 18, 1971–1987
MODELING INTERANNUAL LAND CO2FLUX
1983

Page 14

et al., 2011), for example, are known to directly effect
canopy structure, an aspect not currently accounted for
in models. The affects of other disturbances, such as ice
storms, strong winds and insect outbreaks are known to
be poorly represented by models (Liu et al., 2011) and
affect long-term carbon dynamics. Lagged effects unre-
lated to disturbances can be caused by changes in nutri-
ent cycling (Richardson et al., 2009) or changes in the
size of carbon pools such as litter (Rocha et al., 2008), or
non-structural carbohydrates (Gough et al., 2009) due to
climatic conditions in previous years. Model aspects
related to lagged and cumulative effects can be
improved through direct comparisons with observa-
tions (e.g., Keenan et al., 2009), although many related
issues remain (Liu et al., 2011). Although lagged effects
are apparent at the three sites showing a coherent regio-
nal extreme event, we did not detect similar lagged
events for other climatic anomalies in the database. This
is likely due to two confounding effects: that smaller
anomalous climate signals do not produce lagged (on
monthly scales) ecosystem effects, and that biotic effects
could play a role in driving some of the interannual
variability in observed fluxes (Richardson et al., 2007).
Open questions remain as to the proportion of interan-
nual variability in land-atmosphere carbon exchange that
is directly explainable by variability in climate (Hui et al.,
2003; Polley et al., 2010; Richardson et al., 2007). Controls
on interannual variability can also be manifest in the
form of functional changes in the ecosystem, or lagged
effects on pool sizes and dynamics. By contrasting the in-
terannual performance of a simple empirical model with
fixed parameters against the same model with interan-
nually varying parameters, Richardson et al. (2007)
reported that forest functional change at a spruce forest
was responsible for 55% of interannual variations in
land-atmosphere CO2exchange. i.e., 45% of the observed
variability was not explainable by the direct impacts of
climate. Polley et al. (2010) used a similar approach to
determineasignificantcontribution
functional change to interannual variability in grass-
lands. Using an optimized process-based model, how-
ever, Desai (2010) found that 81% of interannual
variability in annual CO2exchange could be explained
by variability in climate for five mature hardwood
forests, a value that likely underestimates model perfor-
mance given that it does not account for observational
error. This result supports multi-site synthesis efforts
that show that ~79% of interannual variability for mid-
altitude deciduous broadleaved forests can be explained
by variability in temperature (Yuan et al., 2009). Clearly
a detailed assessment of the relative roles of climate and
functional change on the interannual variability of CO2
flux across a wide range of sites and climate zones is
needed.
ofecosystem
We could not distinguish any model structure or
characteristic (see Table 2) that tended to give a better
model performance. All models are subject to errors
resulting from both parameter choice (parameter mis-
specification) and model structure (process mis-repre-
sentation) (Keenan et al., 2011). The fact that no model
structure proved consistently better suggests that
parameter error was excessively large. In future efforts,
model-data fusiontechniques
Keenan et al., 2011) could aid in reducing the relative
magnitude of parameter errors, thus allowing for a more
rigorous assessment of model structural differences.
Our estimates of the magnitude of observed interan-
nual variability in land-atmosphere CO2 exchange
(DBF: ~85 gC m?2; EVG: 44 gC m?2, Table 3) are
roughly 50% and 33% of the mean flux, respectively.
Given that this represents one standard deviation about
the mean, variability in ecosystem carbon uptake is
commonly on the order of magnitude of the mean. This
supports previous results from single sites (Richardson
et al., 2007; a), and modeling studies (Zeng et al., 2005),
across the range of sites included herein. Variability in
GPP has been found to be the main contributor to vari-
ability in NEE for a variety of terrestrial ecosystems
(Luyssaert et al., 2007). Here, we show that for decidu-
ous forests, the interannual variability in GPP is on
average 26% greater than that of RE (Table 3).
Although on average both GPP and RE show a similar
magnitude of variability at the evergreen needleleaf for-
est sites, four of the six evergreen sites had higher vari-
ability in GPP. This suggests that variability in GPP
dominates variability in NEE in mid-latitude forests,
though this rule is not applicable to all sites included
herein.
Using 91 site-years at 11 long-term eddy-covariance
forest sites, we show that terrestrial biosphere models
have difficulty in simulating land-atmosphere CO2
exchange at annual and interannual time scales, with
the potential for large biases on the interannual scale,
and incorrect simulation of the timing of interannual
variability. Instead of focusing on model-data agree-
ment, we present a variability-oriented approach of
diagnosing systematic and persistent model-data dis-
agreement. Given that studies of the impact of climate
variability on terrestrial fluxes are likely to reveal a
more informative picture of biosphere-atmosphere
interactions (le Maire et al., 2010), such a variability ori-
entated approach should greatly aid modeling teams in
future model assessment and development. Our results
highlight three potential mechanisms - spring canopy
phenology, soil thaw and the melting of the snow pack,
and lagged effects - common to all models included in
the study, which contribute to the low agreement
between the models and the observed interannual vari-
(Wang
et al., 2009;
© 2012 Blackwell Publishing Ltd, Global Change Biology, 18, 1971–1987
1984 T.F. KEENAN et al.

Page 15

ability in land-atmosphere CO2exchange. Addressing
these issues in future model efforts will be the first step
toward improving the sensitivity of models to climatic
variability on interannual time scales.
Acknowledgments
We thank all those involved in the NACP Site Synthesis, in par-
ticular the modeling teams, and the AmeriFlux and Fluxnet-
Canada Research Network/Canadian Carbon Program PIs who
provided the data on which this analysis is based. We also thank
the funding agencies that have supported model development
and long-term flux measurements. TFK and ADR acknowledge
support from Office of Science (BER), US Department of Energy,
through the Northeastern Regional Center of the National Insti-
tute for Climatic Change Research. TFK acknowledges Philippe
Peylin for useful comments on an earlier draft.
References
Arain MA, Yuan F, Andrew Black T (2006) Soil–plant nitrogen cycling modulated car-
bon exchanges in a western temperate conifer forest in Canada. Agricultural and
Forest Meteorology, 140, 171–192, doi: 10.1016/j.agrformet.2006.03.021.
Baker IT, Prihodko L, Denning AS, Goulden M, Miller S, da Rocha HR (2008) Sea-
sonal drought stress in the Amazon: reconciling models and observations. Journal
of Geophysical Research, 113, G00B01, doi: 10.1029/2007JG000644.
Barr AG, Black TA, Hogg EH, Kljun N, Morgenstern K, Nesic Z (2004) Inter-annual
variability in the leaf area index of a boreal aspen-hazelnut forest in relation to net
ecosystem production. Agricultural and Forest Meteorology, 126, 237–255, doi:
10.1016/j.agrformet.2004.06.011.
Barr A, Hollinger D, Richardson AD (2009) CO2flux measurement uncertainty esti-
mates for NACP. EOS Transactions, American Geophysical Union, (Fall Meeting Sup-
plement) 90, Abstract B54A-04.
Beer C, Reichstein M, Tomelleri E et al. (2010) Terrestrial gross carbon dioxide uptake:
global distribution and covariation with climate. Science, 329, 834–838, doi:
10.1126/science.1184984.
BousquetP,PeylinP,CiaisP,LeQue ´rC,FriedlingsteinP,TansPP(2000)Regionalchanges
incarbondioxidefluxesoflandandoceanssince1980.Science,290,1342–1347.
Bradford JB, Birdsey RA, Joyce LA, Ryan MG (2008) Tree age, disturbance history,
and carbon stocks and fluxes in subalpine Rocky Mountain forests. Global Change
Biology, 14, 2882–2897, doi: 10.1111/j.1365-2486.2008.01686.x.
Braswell BH, Sacks WJ, Linder E, Schimel DS (2005) Estimating diurnal to annual eco-
system parameters by synthesis of a carbon flux model with eddy covariance net
ecosystem exchange observations. Global Change Biology, 11, 335–355, doi: 10.1111/
j.1365-2486.2005.00897.x
Chen B, Black TA, Coops NC, Krishnan P, Jassal R, Bru ¨mmer C, Nesic Z (2009) Sea-
sonal controls on interannual variability in carbon dioxide exchange of a near-end-
of rotation Douglas-fir stand in the Pacific Northwest, 1997-2006. Global Change
Biology, 15, 1962–1981, doi: 10.1111/j.1365-2486.2008.01832.x.
Ciais P, Reichstein M, Viovy N et al. (2005) Europe-wide reduction in primary pro-
ductivity caused by the heat and drought in 2003. Nature, 437, 529–533, doi:
10.1038/nature03972.
Davis KJ, Bakwin PS, Yi C, Berger BW, Zhao C, Teclaw RM, Isebrands JG (2003) The
annual cycles of CO2and H2O exchange over a northern mixed forest as observed
from a very tall tower. Global Change Biology, 9, 1278–1293, doi: 10.1046/j.1365-
2486.2003.00672.x.
Desai AR (2010) Climatic and phenological controls on coherent regional interannual
variability of carbon dioxide flux in a heterogeneous landscape. Journal of Geophysi-
cal Research, 115, 1–13, doi: 10.1029/2010JG001423.
Dietze M, Rodrigo V, Andrew DR et al. (2011) Characterizing the performance of eco-
system models across time scales: a spectral analysis of the North American Car-
bon Program site-level synthesis. Journal of Geophysical Research, 116, G04029, doi:
10.1029/2011JG001661.
Dragoni D, Schmid HP, Wayson CA, Potter H, Grimmond CSB, Randolph JC (2011)
Evidence of increased net ecosystem productivity associated with a longer vege-
tated season in a deciduous forest in south-central Indiana, USA. Global Change
Biology, 17, 886–897, doi: 10.1111/j.1365-2486.2010.02281.x.
Dunn AL, Barford CC, Wofsy SC, Goulden ML, Daube BC (2007) A long-term record
of carbon exchange in a boreal black spruce forest: means, responses to interannu-
al variability, and decadal trends. Global Change Biology, 13, 577–590, doi: 10.1111/
j.1365-2486.2006.01221.x.
Friedlingstein P, Cox P, Betts R et al. (2006) Climate-carbon cycle feedback analysis:
results from the C4MIP model intercomparison. Journal of Climate, 19, 3337–3353.
Friend AD (2010) Terrestrial plant production and climate change. Journal of experi-
mental botany, 61, 1293–1309, doi: 10.1093/jxb/erq019.
Gough CM, Flower CE, Vogel CS, Dragoni D, Curtis PS (2009) Whole-ecosystem
labile carbon production in a north temperate deciduous forest. Agricultural and
Forest Meteorology, 149, 1531–1540, doi: 10.1016/j.agrformet.2009.04.006.
Goulden ML, Wofsy SC, Harden JW et al. (1998) Sensitivity of Boreal Forest Carbon
Balance to Soil Thaw. Science, 279, 214–217, doi: 10.1126/science.279.5348.214.
Grant RF, Arain A, Arora V et al. (2005) Intercomparison of techniques to model
high temperature effects on CO2and energy exchange in temperate and boreal
coniferous forests. Ecological Modelling, 188, 217–252, doi: 10.1016/j.ecolmodel.
2005.01.060.
Griffis T, Black TA, Morgenstern K et al. (2003) Ecophysiological controls on the car-
bon balances of three southern boreal forests. Agricultural and Forest Meteorology,
117, 53–71, doi: 10.1016/S0168-1923(03)00023-6.
Gu L, Hanson PJ, Post WM et al. (2008) The 2007 eastern US spring freeze: increased
cold damage in a warming world. BioScience, 58, 253, doi: 10.1641/B580311.
Hanson PJ, Amthor JS, Wullschleger SD et al. (2004) Oak forest carbon and water sim-
ulations: model intercomparisons and evaluations against independent data. Eco-
logical Monographs, 74, 443–489, doi: 10.1890/03-4049.
Heimann M, Reichstein M (2008) Terrestrial ecosystem carbon dynamics and climate
feedbacks. Nature, 451, 289–292, doi: 10.1038/nature06591.
Hu J, Moore DJP, Burns SP, Monson RK (2010) Longer growing seasons lead to less
carbon sequestration by a subalpine forest. Global Change Biology, 16, 771–783, doi:
10.1111/j.1365-2486.2009.01967.x.
Hui D, Luo Y, Katul G (2003) Partitioning interannual variability in net ecosystem
exchange between climatic variability and functional change. Tree physiology, 23,
433–442.
IPCC (2007) Climate Change 2007: The Physical Science Basis (eds Solomon S, Qin D,
Manning M, Marquis M, Averyt K, Tignor MMB, Miller HLJ, Chen Z), pp. 501–
586. Cambridge University Press, Cambridge, UK. ISBN-13: 9780521705967.
Keenan T, Garcia R, Friend AD, Zaehle S, Gracia C, Sabate S (2009) Improved under-
standing of drought controls on seasonal variation in Mediterranean forest canopy
CO2and water fluxes through combined in situ measurements and ecosystem
modelling. Biogeosciences, 6, 1423–1444, doi: 10.5194/bg-6-1423-2009.
Keenan TF, Carbone MS, Reichstein M, Richardson AD (2011) The model-data fusion
pitfall: assuming certainty in an uncertain world. Oecologia, 167, 587–597, doi:
10.1007/s00442-011-2106-x.
Knorr W, Gobron N, Scholze M, Kaminski T, Schnur R, Pinty B (2007) Impact of ter-
restrial biosphere carbon exchanges on the anomalous CO2increase in 2002–2003.
Geophysical Research Letters, 34, 1–6, doi: 10.1029/2006GL029019.
Krinner G, Viovy N, de Noblet-Ducoudre N et al. (2005) A dynamic global vegetation
model for studies of the coupled atmosphere-biosphere system. Global Biogeochemi-
cal Cycles, 19, GB1015, doi: 10.1029/2003GB002199.
Krishnan P, Black TA, Barr AG, Grant NJ, Gaumont-Guay D, Nesic Z (2008) Factors
controlling the interannual variability in the carbon balance of a southern boreal
black spruce forest. Journal of Geophysical Research, 113, 1–16, doi: 10.1029/
2007JD008965.
Krishnan P, Black TA, Jassal RS, Chen B, Nesic Z (2009) Interannual variability of the
carbon balance of three different-aged Douglas-fir stands in the Pacific Northwest.
Journal of Geophysical Research, 114, 1–18, doi: 10.1029/2008JG000912.
Liu J, Chen J, Cihlar J, Chen W (1999) Net primary productivity distribution in the
BOREAS region from a process model using satellite and surface data. Journal of
Geophysical Research, 104, 27375–27754.
Liu S, Bliss N, Sundquist E, Hungtington TG (2003) Modeling carbon dynamics in
vegetation and soil under the impact of soil erosion and deposition. Global Biogeo-
chemical Cycles, 17, 1074, doi: 10.1029/2002GB002010.
Liu S, Bond-Lamberty B, Hicke JA et al. (2011) Simulating the impacts of disturbances
on forest carbon cycling in North America: processes, data, models, and chal-
lenges. Journal of Geophysical Research, 116, 1–22, doi: 10.1029/2010JG001585.
Luyssaert S, Janssens IA, Sulkava M et al. (2007) Photosynthesis drives anomalies in
net carbon-exchange of pine forests at different latitudes. Global Change Biology, 13,
2110–2127, doi: 10.1111/j.1365-2486.2007.01432.x.
le Maire G, Delpierre N, Jung M et al. (2010) Detecting the critical periods that under-
pin interannual fluctuations in the carbon balance of European forests. Journal of
Geophysical Research, 115, 1–16, doi: 10.1029/2009JG001244
© 2012 Blackwell Publishing Ltd, Global Change Biology, 18, 1971–1987
MODELING INTERANNUAL LAND CO2FLUX
1985