ICOS Science Conference 2020, Session 14

1. Reconciling the carbon balance of northern Sweden
using measurements and modelling.
Anusha Sathyanadh1, Marko Scholze,Guillaume Monteil,Hjalmar Laudon, Per Marklund, Mikaell Ottosson
Löfvenius,Anne Klosterhalfen,Zhendong Wu, Christoph Gerbig,Erik van Schaik, Vladislav Bastrikov,Mats B.
Nilsson, Matthias Peichl
1Department of Forest Ecology and Management, Swedish University of Agricultural Sciences, Umeå, Sweden

2. Background & motivation forthe project
2
▪ Boreal forests, sinks for atmospheric CO2 and key component in climate mitigation strategies.
▪ Not yet known as to what extent the boreal region contribute to the global carbon sink
▪ Hence it is necessary to provide the accurate accounting of C budgets.
▪ Common measurement techniques plot-scale inventories to large-scale modeling approaches
Plot
Ecosystem
Landscape
RegionAtmospheric concentrations
Tall tower EC
Eddy covariance (EC)
Flux chamber & Inventory
Svartberget
Figure 1: Schematic diagram showing a GHG emission assessment system from local to regional scale

3. Studyaim and objectives
3
In this study, the aim is to connect different carbon cycle observations to reconcile carbon fluxes
and to estimate the carbon balance of northern Sweden with specific objectives
1. Compare the fluxes and concentration at different spatial scales
2. Determine how well the atmospheric observation network can constrain the fluxes in the
Nordic region.
3. Investigate how C fluxes and concentration compares during 2018 drought episode at an
Experimental site Svartberget (SVB), northern Sweden.

4. Materialsand Methods
4
▪ The measured regional fluxes are collected from a tall tower EC system installed at 60m on the 150m
ICOS tower in Svartberget measuring CO2, water vapour, and energy exchange between the boreal
landscape and the atmosphere.
▪ It consists of a CSAT3 3-D sonic anemometer and an EC155 closed-path gas analyzer (Campbell
Scientific, Inc., USA).
▪ The measured NEE was processed following ICOS protocol, gap-filled and source partitioned into
the two flux componentsGPP and Reco.
1. Local Fluxes - Observations at SVB
2. Regional Fluxes -Ecosystemand Diagnostic vegetation models
▪ Process based ecosystem models
▪ LPJG
▪ ORCHIDEE
▪ Diagnostic models
▪ VPRM
▪ SiBCASA

5. Period: 2016 January to 2018 December
Forward transport of the surface fluxes to the observation sites
Transport model:
FLEXPART footprints
(0.5x0.5)
+ TM5 boundary
condition
Model
concentration
Land C flux:
LPJ-GUESS
ORCHIDEE
VPRM
SiBCASA
Anthropogenic
emissions
(EDGAR)
Nordic study domain
Materialsand Methods
5
1. Atmospheric CO2 Concentration Measurements
2. Transport model (LUMIA system)
▪ Continuous in situ observations of CO2 concentration -GLOBALVIEWplus observation package
▪ For 2016-2018, observations from eight sites within our regional study domain are available.
▪ High frequency CO2 concentration measurements at several vertical levels up to 150m using tall towers
part of the European ICOS atmospheric network. Selected the observations at highest level
Figure 2: Map showing the Scandinavian study
domain with ICOS atmospheric observation sites
and different geographical regions.

6. Results
6
1. Comparison of fluxes from vegetation models with EC in situ measurements at the
SVB landscape scale (LPJG, VPRM, SiBCASA, ORCHIDEE vs EC_60m)
2. Comparison of fluxes at regional scale (Reconcile site and regional scale flux)
3. Comparison of concentration simulated by LUMIA transport model using these
vegetation model fluxes with observed atmospheric concentration to verify whether
the conclusions at site scale hold at the regional scale. (LUMIA conc from LPJG,
VPRM, SiBCASA, ORCHIDEE, EC_60m vs in situ concentration)
4. Comparison of flux and concentration during 2018 drought episode on Northern
Sweden.

7. Results
7
1. Comparison of fluxes at SVB
Figure 3: Comparison of the weekly net CO2 exchange (NEE)
estimated from 2016 to 2018 from four different models and
landscape level eddy covariance measurements at 60m height at the
ICOS-Svartberget (SVB) station.
➢ Larger difference between model and
observation during winters (for NEE) and
during summers (GPP and Reco)
➢ VPRM and LPJG zero winter fluxes.
➢ In 2018, strikingly increased net CO2 uptake
during June in the EC data and an anomalously
high net emission in August in LPJG model.
➢ The ecosystem models LPJG and ORCHIDEE
showed an overestimated GPP and Reco during
summers.
➢ The diagnostic model SiBCASA and VPRM
simulated fluxes fits well with EC fluxes in
general.

8. Results
8
2. Comparison of regional fluxes
Figure 4: Spatial map of NEE Annual Budget for 2018 from all different
models.
To examine how much widespread the features seen at the local landscape level is and how much
SVB can represent the northern Sweden region
All the models showed the site as well as
the region is a C sink for all the years.
The year 2017 is a lesser sink (and even
a small source in SiBCASA) for both site
and region in all models, which is
contradicted only for LPJG in 2018.
-300
-250
-200
-150
-100
-50
0
50
Site Region Site Region Site Region
2016 2017 2018
NEE Annual Budget (gC m-2year-1)
LPJ VPRM ORCHIDEE SiBCASA
Annual Budget
(gC m-2
year-1
)

9. Figure 5: Time series comparison of LUMIA simulated concentrations with observed concentration at SVB. (b) Concentration is taken
only when the observations are sensitivite to fluxes of northern Sweden region is more than 90%.
LUMIA simulated concentration using fluxes
from vegetation models is compared with
observed concentration at SVB. Concentration
can be sensitive to fluxes from any region. In
second figure concentration is plotted only when
the site is more than 90 % sensitive to fluxes
within north Sweden region. The background
uncertaintyis the same for all simulations.
Results
9
3. Comparison of CO2 concentration
a)
b)

10. ❑ During summer (JJAS) 2018, the
observed fluxes (site scale) and
concentration (regional scale)
compares well with all models
except LPJG.
❑ LPJG model simulated a higher
NEE but similar concentration at
SVB site during drought => no
effect of NEE on concentration, as
SVB site was not sensititve enough
to the fluxes in the region.
Figure 4: Time series of NEE and concentrations simulated and observed at SVB during 2018 drought
Results
10
4. Comparison of NEE and CO2 concentration during 2018 summer drought

11. Conclusions
11
➢ The Swedish boreal region is a C sink for all the years (2016-2018) but the magnitude differs
among different models. The year 2017 is a lesser sink (and even a small source in SiBCASA)
compared to 2018 and 2016 in all models and observation.
➢ Models and observations of fluxes and concentrations generally agreed well during the 2018 at
site and regional scale. Results indicate similar reductions in the net CO2 uptake during
drought for ecosystem models and flux observations, except LPJG estimates high emissions.
➢ The difference among modelled concentration when the observations are sensitive to N
Sweden proves the potential of observations to be used for refining the flux estimates.
➢ Our study overall highlights the need to further improve vegetation models through model-
data inter-comparisons and using inverse transport models which is the next step.

12. Conclusions
12
➢ The Swedish boreal region is a C sink for all the years (2016-2018) but the magnitude differs
among different models. The year 2017 is a lesser sink (and even a small source in SiBCASA)
compared to 2018 and 2016 in all models and observation.
➢ Models and observations of fluxes and concentrations generally agreed well during the 2018 at
site and regional scale. Results indicate similar reductions in the net CO2 uptake during
drought for ecosystem models and flux observations, except LPJG estimates high emissions.
➢ The difference among modelled concentration using four modelled vegetation fluxes when the
observations are sensitive to N Sweden proves the potential of observations to be used for
refining the flux estimates.
➢ Our study overall highlights the need to further improve vegetation models through model-
data inter-comparisons and using inverse transport models which is the next step.