of 17
LETTER • OPEN ACCESS
Tipping point in North American Arctic-Boreal
carbon sink persists in new generation Earth
system models despite reduced uncertainty
To cite this article: Renato K Braghiere
et al
2023
Environ. Res. Lett.
18 025008
View the
article online
for updates and enhancements.
You may also like
Missing pieces to modeling the Arctic-
Boreal puzzle
Joshua B Fisher, Daniel J Hayes,
Christopher R Schwalm et al.
-
Evaluating photosynthetic activity across
Arctic-Boreal land cover types using solar-
induced fluorescence
Rui Cheng, Troy S Magney, Erica L Orcutt
et al.
-
The Arctic-Boreal vulnerability experiment
model benchmarking system
Eric Stofferahn, Joshua B Fisher, Daniel J
Hayes et al.
-
This content was downloaded from IP address 24.24.204.12 on 12/04/2023 at 03:39
Environ. Res. Lett.
18
(2023) 025008
https://doi.org/10.1088/1748-9326/acb226
OPEN ACCESS
RECEIVED
19 May 2022
REVISED
20 December 2022
ACCEPTED FOR PUBLICATION
11 January 2023
PUBLISHED
10 February 2023
Original content from
this work may be used
under the terms of the
Creative Commons
Attribution 4.0 licence
.
Any further distribution
of this work must
maintain attribution to
the author(s) and the title
of the work, journal
citation and DOI.
LETTER
Tipping point in North American Arctic-Boreal carbon sink persists
in new generation Earth system models despite reduced
uncertainty
Renato K Braghiere
1
,
2
,
, Joshua B Fisher
3
, Kimberley R Miner
2
, Charles E Miller
2
,
John R Worden
2
, David S Schimel
2
and Christian Frankenberg
1
,
2
1
Division of Geological and Planetary Sciences, California Institute of Technology, Pasadena, CA 91125, United States of America
2
Jet Propulsion Laboratory, California Institute of Technology, 4800 Oak Grove Drive, Pasadena, CA 91109, United States of America
3
Schmid College of Science and Technology, Chapman University, Orange, CA 92866, United States of America
Author to whom any correspondence should be addressed.
E-mail:
renato.k.braghiere@jpl.nasa.gov
Keywords:
NASA ABoVE
,
CMIP5
,
CMIP6
,
tipping point
,
carbon cycle
,
soil carbon
Supplementary material for this article is available
online
Abstract
Estimating the impacts of climate change on the global carbon cycle relies on projections from
Earth system models (ESMs). While ESMs currently project large warming in the high northern
latitudes, the magnitude and sign of the future carbon balance of Arctic-Boreal ecosystems are
highly uncertain. The new generation of increased complexity ESMs in the Intergovernmental
Panel on Climate Change Sixth Assessment Report (IPCC AR6) is intended to improve future
climate projections. Here, we benchmark the Coupled Model Intercomparison Project (CMIP) 5
and 6 (8 CMIP5 members and 12 CMIP6 members) with the International Land Model
Benchmarking (ILAMB) tool over the region of NASA’s Arctic-Boreal vulnerability experiment
(ABoVE) in North America. We show that the projected average net biome production (NBP) in
2100 from CMIP6 is higher than that from CMIP5 in the ABoVE domain, despite the model
spread being slightly narrower. Overall, CMIP6 shows better agreement with contemporary
observed carbon cycle variables (photosynthesis, respiration, biomass) than CMIP5, except for soil
carbon and turnover time. Although both CMIP ensemble members project the ABoVE domain
will remain a carbon sink by the end of the 21st century, the sink strength in CMIP6 increases with
CO
2
emissions. CMIP5 and CMIP6 ensembles indicate a tipping point defined here as a negative
inflection point in the NBP curve by 2050–2080 independently of the shared socioeconomic
pathway (SSP) for CMIP6 or representative concentration pathway (RCP) for CMIP5. The model
ensembles therefore suggest that, if the carbon sink strength keeps declining throughout the 21st
century, the Arctic-Boreal ecosystems in North America may become a carbon source over the next
century.
1.Introduction
Global mean surface temperatures have increased
dramatically since the mid-20th century, but have
increased up to four times faster in the Arctic-Boreal
region (Masson-Delmotte
et al
2021
, Rantanen
et al
2022
).Thisphenomenonisreferredtoas‘arcticamp-
lification’ (Scheffer
et al
2012
, Francis
et al
2017
).
Although the exact mechanisms for the arctic amp-
lification are debated, temperature and snow-sea
ice-albedo feedbacks are keys to understanding this
system,andchangesinatmosphericandoceanenergy
transport may play an important role (Previdi
et al
2021
). While Arctic-Boreal ecosystem productivity
may initially benefit from rising atmospheric CO
2
,
higher temperatures, longer growing seasons, and
faster nutrient cycling, these same systems may
increase carbon emissions through permafrost thaw,
plant (autotrophic) respiration and increased micro-
bial (heterotrophic) respiration (Mack
et al
2004
,
©2023TheAuthor(s). PublishedbyIOPPublishingLtd
Environ. Res. Lett.
18
(2023) 025008
R K Braghiere
et al
Natali
et al
2012
,
2019
, Crowther
et al
2015
, Schuur
et al
2015
, Koven
et al
2017
, Huntzinger
et al
2020
,
Miner
et al
2022
).
Projecting the future trajectory of the Arctic-
Boreal system presents a large challenge to Earth
system models (ESMs) (Hinzman
et al
2013
) and
requires critical cryosphere-specific processes to
accurately model its physical, biogeochemical and
ecosystem dynamics (including carbon) (Hawkins
and Sutton
2009
, Knutti and Sedlácˇek
2013
, Slater
and Lawrence
2013
, Koven
et al
2015
, Lawrence
et al
2015
, Schimel
et al
2015
, Ciais
et al
2019
, Braghiere
et al
2021b
,
2022
).TheCoupledModelIntercompar-
isonProjectPhase6(CMIP6;Eyring
et al
2016
)isthe
most recent ESM activity, and builds upon CMIP5
(Taylor
et al
2012
), interpreted in the IPCC Fifth
Assessment Report (Intergovernmental Panel on Cli-
mate Change
2014
). CMIP6 includes the latest gen-
eration of comprehensive ESMs, driven by historical
greenhouse gas concentrations and climate forcing
followed by different future greenhouse gas concen-
trations pathways according to the shared socioeco-
nomic pathways (SSPs) scenarios (Meinshausen
et al
2020
,Tokarska
et al
2020
).TheSSPspicturemultiple
baseline worlds considering underlying factors, such
as population, technological, and economic growth,
andhowthosecouldleadtodifferentfuturescenarios
and global change outcomes. That does not imply a
larger uncertainty in climate change, it rather con-
siders different economic and political choices.
While benchmarking and validation of ESMs has
become increasingly common in recent years (Fisher
JB
et al
2018
), it is still rare to comparatively eval-
uate the performance of a carbon cycle model once
it is updated (Fer
et al
2021
). However, compar-
ing models and observations is required for hypo-
thesis testing and predictive skill evaluation (Fisher
RA
et al
2018
). To this end, the International Land
Model Benchmarking (ILAMB) project (Hoffman
et al
2017
, Collier
et al
2018
) provides the means
to track and compare performance through a com-
prehensive skill score method and to incorporate
multiple observational datasets of the same variable
of interest to account for observational uncertainty.
Moreover, greater agreement between historical runs
and observations may indicate that model compon-
entscanbeupdatedtobettercaptureinaccuratepro-
cesses. This would increase confidence in future pro-
jections, even though forthcoming changes, such as
photosynthesis acclimation or species composition
shifts, may become progressively more important.
WhetherArctic-Borealecosystemswillevolveinto
significant carbon sinks (Keenan
et al
2014
, Zhu
et al
2016
, Berner
et al
2020
), net carbon sources (Hayes
et al
2011
, Zhang
et al
2022
), or remain nearly car-
bon neutral (McGuire
et al
2012
) depends on the
trajectory of climate change (McGuire
et al
2018
,
de Vrese and Brovkin
2021
, De Vrese
et al
2021
).
It is also imperative to understand if ESMs accur-
ately describe the major carbon cycle fluxes, stor-
age terms, and processes in the present. The critical
threshold at which a perturbation can qualitatively
alter the system’s state or development is referred to
as a tipping point (Lenton
et al
2008
). An Arctic-
Boreal carbon cycle tipping point would occur when
the rate of release of previously frozen soil carbon to
theatmospherebyecosystemrespirationanddisturb-
ances (DISTs), including wildfires, surpasses photo-
syntheticCO
2
uptake(Ahlström
et al
2015
).Thisper-
mafrost carbon feedback (Schuur
et al
2015
, Miner
et al
2022
) has been identified as a critical Earth sys-
tem tipping point (Lenton
et al
2008
, McKay
et al
2022
).Forexample,therapidwarmingoftheArctic-
Boreal zone has accelerated permafrost degradation
(Mekonnen
et al
2021
) and is remaking the vast,
conifer-dominated boreal forests, lowering species
diversity, increasing ecosystem vulnerability to dis-
ease, decreasing vegetation reproduction rates, mak-
ing fires more frequent and intense, and increasing
mortality rates (Lenton
2012
, Seidl
et al
2017
). In
a stark example of this from the mid-1990s, sum-
merwarmingintheabsenceofsustainedincreasesin
precipitation breached the tipping point in western
central Eurasian boreal forests, sharply shifting eco-
systems into a warmer and drier regime (Buermann
et al
2014
). Ensuring that ESMs reproduce emergent
Arctic-Boreal ecosystem tipping points requires con-
siderable certainty in both ecosystem stability and
carbon cycle drivers. Unfortunately, both of these
complex, interconnected systems are largely uncon-
strained within ESMs.
The goal to reduce carbon cycle uncertainty is
non-trivial (Hausfather
et al
2022
) and it can be
bounded by inherent model uncertainties encom-
passingparametricandstructuraluncertainty,aswell
as forcing data uncertainty (Lovenduski and Bonan
2017
). Yet, interpreting model spread as predictive
failure would not bring as much benefit to science as
muchasacomprehensivediscussionofmodeluncer-
tainty (Bonan
et al
2019
). To this end, we compare
projectionsofthecarboncyclevariablesfromCMIP5
and CMIP6 over Alaska and northwestern Canada,
the domain of NASA’s Arctic-Boreal vulnerability
experiment (ABoVE; figure S1). We benchmark the
historical runs with a suite of state-of-the-art Earth
observations to test if the most recent models con-
vergein theirprojections ofthe Arctic-Borealcarbon
cycle.
The rationale behind benchmarking models in
the historical period before evaluating their future
projections is related to the hypothesis that a model
(or group of models) that better represents the
present has greater predictive skills, and therefore,
should better represent the future. Although mech-
anistic processes represented in these models could
potentially change in the future (i.e. acclimate), the
2
Environ. Res. Lett.
18
(2023) 025008
R K Braghiere
et al
predictiveskilloftwomodelgroupscanonlybeeval-
uated with historical datasets. We compare model
performances and uncertainty, identify improve-
ments and deterioration from older ESMs to newer
ones, and analyze modeled carbon uptake growing
curvestodetermineifthesemodelsprojectapotential
Arctic-Boreal carbon balance tipping point.
2.Materialandmethods
2.1.Studydomainandmodels
This study focuses on the ABoVE domain, includ-
ing the Arctic and Boreal regions of Alaska, and the
western provinces of Canada (Fisher JB
et al
2018
,
Stofferahn
et al
2019
, Huntzinger
et al
2020
). We
benchmark six ecosystem and carbon cycle variables:
above-ground biomass, gross primary productivity
(GPP),ecosystemrespiration(RECO),leafareaindex
(LAI), net ecosystem exchange (NEE), and topsoil
carbon. We incorporate 3 meteorological forcing
datasets (i.e. surface air temperature, precipitation,
and surface downward shortwave radiation), output
from a total of 20 CMIP models, with 8 models par-
ticipating in the CMIP5 and the remaining 12 latest
model versions participating in the CMIP6 (table
1
).
Thesemodelswerechosenbasedontheavailabilityof
all the required variables included in the analysis. We
also evaluated monthly historical simulations (1850–
2005 for CMIP5 and 1850–2015 for CMIP6) driven
by observation-based forcing data, including green-
housegasconcentrations,griddedland-usedata,vol-
canic aerosols, and other meteorological variables
(Eyring
et al
2016
).
Previously, in preparation for CMIP5, the land-
use harmonization v1 (LUH1) project provided har-
monized land use data for the years 1500–2100 at
0.5 deg
×
0.5 deg resolution (Hurtt
et al
2011
).
These data served as required land use forcing for
CMIP5 climate model experiments and have been
used in a number of related studies to assess the
effects of land use change on carbon cycle and cli-
mate. More recently, as part of CMIP6 (Eyring
et al
2016
), the international research community has
developedthenextgenerationofadvancedESMsable
to estimate the combined effects of human activit-
ies (e.g. land use and fossil fuel emissions) on the
carbon–climatesystem.Thestrategy describedin the
updatedLUH2buildsontheapproachforharmoniz-
ing land use patterns and transitions in CMIP5. The
new version is completely updated with new inputs
and includes higher spatial resolution (0.25 deg vs.
0.5 deg), increased detail (12 states vs. 5 and all asso-
ciated transitions), added management layers, new
future scenarios (8 vs. 4), and a longer time domain
(850–2100 vs. 1500–2100)—in all more than a 50-
fold increase in data from its predecessor (Hurtt
et al
2020
). Despite these differences, carbon fluxes
associatedwithDISTsareseveralordersofmagnitude
smaller than photosynthesis and respiratory terms
(see figure S2 in supporting information).
CMIPmodelsprovidemultiplesimulationsbased
on the different experimental configurations for
ensemblememberanalysestocapturetheclimatesys-
tem’s natural variability. However, some of the parti-
cipant members have only released their first realiza-
tion,denominatedasr1i1f1;therefore,weutilizedthe
firstrealizationonly,followingpreviousrecommend-
ations (Anav
et al
2013
, Park and Jeong
2021
).
2.2.Modelbenchmarking
We used the ILAMB v2.6 package for model bench-
marking (Collier
et al
2018
) focusing on global
patterns of ecosystem and carbon cycle variables,
including datasets: (a) aboveground living biomass
based on inventory plots upscaled using remote
sensing imagery from GlobalCarbon (Saatchi
et al
2011
), USForest (Blackard
et al
2008
), and Thurner
(Thurner
et al
2014
); (b) GPP and RECO from
FLUXCOM (Jung
et al
2019
,
2020
) and from
FLUXNET2015 (Pastorello
et al
2020
); (c) LAI
from the moderate resolution imaging spectrora-
diometer (MODIS) (De Kauwe
et al
2011
), AVHRR
and AVH15C1 (Claverie
et al
2016
); (d) NEE from
global bio-atmosphere flux (GBAF) (Jung
et al
2010
) and FLUXNET2015 (Pastorello
et al
2020
);
and (e) soil carbon stocks from the Harmonized
World Soil v1.2 Database (HWSD; Nachtergaele
et al
2012
), the Northern Circumpolar Soil Carbon
Database version 2.2 (NCSCDv2.2) (Hugelius
et al
2013a
,
2013b
),andsoilcarbonturnovertime(Koven
et al
2017
). Results from the global models were
masked out to focus benchmarking on the ABoVE
domain.
The relationships between these variables and
precipitation, temperature, and incident shortwave
radiationwereanalyzedusingdataproductsfromthe
Global Precipitation Climatology Project Monthly
Analysis version (Adler
et al
2018
), the Climatic
Research Unit monthly temperature version 4.02
(Harris
et al
2014
), and the Clouds and the Earth’s
RadiantEnergySystemsurfaceirradiancesedition4.1
(Kato
et al
2018
, Loeb
et al
2018
).
WeusetheresultsfortheILAMBoverallscorefor
the absolute values (
S
overall
) defined as:
S
overall
=
S
bias
+
2
S
RMSE
+
S
phase
+
S
dist
1
+
2
+
1
+
1
(1.1)
where
S
bias
isthespatiallyintegratedbiasscore,
S
RMSE
istheroot-mean-squarederror(RMSE)scoredoubly
weighted to emphasize its importance,
S
phase
is the
phase shift score, and
S
dist
is the spatial distribu-
tion score. For the whole set of equations of each
term in equation (
1
) refer to Collier
et al
(
2018
). All
the ILAMB results and plots are available in
https://
braghiere.github.com
(accessedon:April05th2022).
3
Environ. Res. Lett.
18
(2023) 025008
R K Braghiere
et al
Table1.
General specifications of ESMs used in this study.
CMIP5
Modeling
group
ESM
Land model
Number
of live
carbon
pools
Number
of dead
carbon
pools
No of
plant
functional
types
(PFTs) Fire
Dynamic
vegetation
cover
Nitrogen
cycle
Phosphorus
cycle
No
soil
layers
Soil
depth
(m) References
CSIRO
ACCESS1-3 CABLE1.8
3
6
13 No No
Yes
Yes
4 2.0 Kowalczyk
et al
(
2013
)
CCCMA CanESM2 CLASS2.7-
CTEM
3
2
9 No No
No
No
3 2.2 Arora and Boer
(
2005
); Arora
et al
(
2011
)
GFDL
GFDL-
ESM2M
LM3
6
4
5 Yes Yes
No
No
20 8.8 Shevliakova
et al
(
2009
);
Shao
et al
(
2013
)
UK
HadGEM2-
CC
JULES
3
4
5 No Yes
Yes
No
4 2.0 (Martin
et al
2011
, Best
et al
2011
, Clark
et al
2011
)
IPSL
IPSL-CM5A-
LR
ORCHIDEE 8
3
15 No No
No
No
7 3.9 (Dufresne
et al
2013
, Krinner
et al
2005
)
JAMSTEC MIROC-ESM MATSIRO
(physics)
VISIT-e
(BGC)
3
6
13 No No
Yes
No
6 9.0 (Watanabe
et al
2011
, Sato
et al
2007
)
MPI
MPI-ESM-LR JSBACH
3
18
13 Yes Yes
Yes
No
5 7.0 (Giorgetta
et al
2013
, Raddatz
et al
2007
,
Knorr
2000
)
NCC
NorESM1-M CLM4
22
7
22 Yes No
Yes
No
15 35.2 (Iversen
et al
2013
, Bentsen
et al
2013
)
CMIP6
Modeling
group
ESM
Land model
Number
of live
carbon
pools
Number
of dead
carbon
pools
Number
of plant
functional
types
(PFTs) Fire
Dynamic
vegetation
cover
Nitrogen
cycle
Phosphorus
cycle
No.
soil
layers
Soil
depth
(m) Reference
CSIRO
ACCESS-
ESM1-5
CABLE2.4
with
CASA-CNP
3
6
13 No No
Yes
Yes
6 2.9 (Bi
et al
2020
)
BCC
BCC-CSM2-
MR
BCC-
AVIM2
8
16
16 No No
No
No
10 2.9 (Wu
et al
2019
,
Li
et al
2019
)
CCCMA CanESM5 CLASS3.6-
CTEM
3
2
9 No No
No
No
3 4.1 (Swart
et al
2019
, Arora
et al
2020
)
CESM
CESM2
CLM5
22
7
22 Yes No
Yes
No
25 42.0 (Danabasoglu
et al
2020
,
Lawrence
et al
2019
)
CNRM
CNRM-
ESM2-1
ISBA-
CTRIP
6
7
16 Yes No
No
(implicit,
derived
from Yin,
2002)
No
14 10.0 (Delire
et al
2020
)
GFDL
GFDL-ESM4 LM4.1
6
4
5 Yes Yes
No
No
20 8.8 (Dunne
et al
2020
, Zhao
et al
2018
)
NASA
GISS-E2-1-G Ent TBM
5
2
17 Yes Yes
Yes
No
6 2.7 (Ito
et al
2020
)
IPSL
IPSL-CM6A-
LR
ORCHIDEE,
branch 2.0
8
3
15 No No
No
No
18 65.6 (Boucher
et al
2020
)
JAMSTEC MIROC-
ESM1-2-LR
MATSIRO
(physics)
VISIT-e
(BGC)
3
6
13 No No
Yes
No
6 9.0 (Ito and
Oikawa
2002
)
MPI
MPI-ESM1-
2-LR
JSBACH3.2 3
18
13 Yes Yes
Yes
No
5 7.0 (Mauritsen
et al
2019
)
NCC
NorESM2-
LM
CLM5
22
7
22 Yes No
Yes
No
25 42.0 (Seland
et al
2020
)
UK
UKESM1-0-
LL
JULES-ES-
1.0
3
4
13 No Yes
Yes
No
4 2.0 (Sellar
et al
2019
, Walters
et al
2019
)
4
Environ. Res. Lett.
18
(2023) 025008
R K Braghiere
et al
3.Results
3.1.BenchmarkingCMIPmodels
Modeling if an Arctic-Boreal carbon cycle tipping
point will occur is rather complex and dependent on
a number of feedback loop interactions, but results
indicatenewerESMsaregenerallybetteratcapturing
the present-day carbon cycle picture over the Arctic-
Borealecosystemsthanwerepreviousmodelversions.
The exception to the general improvement in more
recentESMsisrelatedtotherepresentationofcarbon
in soils.
Figure
1
shows the overall scores for ecosystem
andcarboncyclevariablesfortheCMIP5andCMIP6
ensemblemembers.Resultsforindividualmodelscan
be found in supporting information (figure S3). The
ensemble member is per definition fluxes and states
averaged across individual model members within a
modelinggroup.Theseusemultipledatasetsincluded
in the ILAMBv2.6. Triangles corresponding to dif-
ferent ensemble members (CMIP5 and CMIP6) and
spatial domains (ABoVE and global) are represented
byaschematicinblue.ComparisonsbetweenCMIPs
globally versus the ABoVE domain is performed to
determine if the expected improvements in newer
model versions are comparable when only consider-
ing the Arctic-Boreal North America.
For aboveground biomass, the CMIP6 ensemble
member presents a higher overall score than the
CMIP5 ensemble member across all the evaluated
data products for the ABoVE domain and the globe.
However, the overall score for the ABoVE domain is
always smaller than the global overall score for both
ensemble members. This suggests that most process-
based models included in this study better capture
carbon allocation to living stock for other ecosys-
tems and regions of the globe on average than to
the Arctic-Boreal North America. The highest over-
all score between models and biomass datasets is
given for Thurner, followed by GlobalCarbon, and
USForest.
For GPP and RECO, the CMIP6 ensemble mem-
ber also presents a higher overall score than the
CMIP5 ensemble member across all the evaluated
data products for the ABoVE domain and the globe,
with the overall score for the ABoVE domain being
higherthantheglobaloverallscoreforbothensemble
members. The higher score is mainly due to lar-
ger bias and RMSE in more productive areas of the
globe, including the tropics and temperate forests.
For GPP, the overall score related to the FLUXCOM
data product is systematically larger than the those
associated with FLUXNET2015, while for RECO,
CMIP6 over ABoVE presents better agreement for
FLUXNET2015 than to FLUXCOM.
Spatiotemporal GPP biases between CMIP5 and
CMIP6 with FLUXCOM for the period from 1980
to 2014 are shown in supporting information (figure
S4). The positive bias in GPP presented in CMIP5 in
thenortheasternandeastoftheABoVEdomainisnot
observed in CMIP6, as well as a negative bias in GPP
inthepacific.Aslight(
1gm
2
d
1
)positivebiasin
CMIP6 GPP remains in central parts of the ABoVE
domain, as well as southern croplands in Canada.
For GPP annual cycle, CMIP6 also presents higher
agreement with FLUXCOM (bias
=
0.56 g m
2
d
1
,
RMSE
=
0.84 g m
2
d
1
) than CMIP5
(bias
=
0.79 g m
2
d
1
, RMSE
=
1.23 g m
2
d
1
),
approximately30%reductioninbiasandRMSE,with
especially accurate performance during spring.
To assess the representation of mechanistic pro-
cesses in the models, we also evaluate variable-
to-variable relationships of GPP with precipita-
tion, surface downward shortwave radiation, and
temperature. The response curves are then scored by
computing a relative error based on the RMSE of
reference datasets to the relationship diagnosed in
models. Acrossall the evaluated relationshipsofGPP
with meteorological variables, the scores of CMIP6
were higher than those of CMIP5 (refer to support-
ing information; figure S5).
ForLAI,theCMIP6ensemblememberspresenta
larger overall score than the CMIP5 ensemble mem-
bers across both spatial domains and data products.
Still, the CMIP6 ensemble member over the ABoVE
domainpresentsalargeroverallscorethantheCMIP6
ensemble member over the globe, which is not
observedincomparisonwithAVHRRandAVH15C1.
Spatiotemporal LAI biases between CMIP5 and
CMIP6 with MODIS for the period from 2000 to
2006 are shown in supporting information (figure
S6). A strong positive bias (
>
2 m
2
m
2
) in LAI
presented in CMIP5 in most of the ABoVE domain
including Alaska, the northeastern and eastern parts
ofthedomain,aswellastheborealcordillera(higher
elevation terrain of the Rocky Mountains and the
Coast Mountains) are corrected in CMIP6, as well
as a negative bias in LAI in the Pacific coast, boreal
plain and taiga plain. A positive bias in CMIP6 LAI
remains in high elevation areas, as well as south-
ern croplands in Canada. For LAI annual cycle,
CMIP6 also presents higher agreement with MODIS
(bias
=
0.22 m
2
m
2
, RMSE
=
0.53 m
2
m
2
) than
CMIP5(bias
=
0.67m
2
m
2
,RMSE
=
0.91m
2
m
2
),
approximately40%reductioninbiasand70%reduc-
tioninRMSE,withhighperformanceinthefirsthalf
of the year and slight overestimation in the second
half of the year. Across all the evaluated relationships
of LAI with meteorological variables, the scores of
CMIP6 were higher than those of CMIP5 (refer to
supporting information; figure S7).
Finally, soil carbon overall scores highlight a
downgrade in model performance from CMIP5 to
CMIP6 for both spatial domains in comparison to
the HWSD data product and inferred turnover rates
fromKoven
et al
(
2017
)overtheglobe.Incomparison
to NCSDV22 soil carbon data, CMIP6 presents a lar-
ger overall score than CMIP5 models for both spatial
5