multiple greenhouse-gas feedbacks from the land biosphere under future climate change scenarios
TRANSCRIPT
SUPPLEMENTARY INFORMATIONDOI: 10.1038/NCLIMATE1864
NATURE CLIMATE CHANGE | www.nature.com/natureclimatechange 1
SUPPLEMENTARY INFORMATION1
Multiple greenhouse gas feedbacks from the land bio-2
sphere under future climate change scenarios3
Benjamin D. Stocker1,2∗, Raphael Roth1,2, Fortunat Joos1,2, Renato Spahni1, Marco Steinacher1,2,4
Soenke Zaehle3, Lex Bouwman4,5, Xu-Ri6, Iain Colin Prentice7,85
1Climate and Environmental Physics, Physics Institute, University of Bern, 3012, Bern,6
Switzerland7
2Oeschger Center for Climate Change Research, University of Bern, Bern, Switzerland8
3Max Planck Institute for Biogeochemistry, Department for Biogeochemical Systems, 077459
Jena, Germany10
4Department of Earth Sciences, Geochemistry, Faculty of Geosciences, Utrecht University11
5PBL Netherlands Environmental Assessment Agency, P.O. Box 303, 3720 AH Bilthoven, The12
Netherlands13
6Laboratory of Tibetan Environment Changes and Land Surface Processes, Institute of Tibetan14
Plateau Research, Chinese Academy of Sciences, Beijing 100101, China15
7Department of Biological Sciences, Macquarie University, North Ryde, NSW 2109, Australia16
8Grantham Institute for Climate Change and Division of Ecology and Evolution, Imperial Col-17
lege, Silwood Park, Ascot SL5 7PY, UK.18
S1 Overview19
This document provides supplementary information not provided in the main text of the arti-20
cle Multiple greenhouse gas feedbacks from the land biosphere under future climate change21
scenarios. Setion S2 provides complete references to all input files used to drive simulations.22
Section S3 provides a comprehensive documentation of the model setups used to quantify feed-23
backs and a mathematical framework thereof. Section S4 provides additional figures for results24
referred to in the main text but not given therein. In particular: spatial information of GHG25
emission changes in the 21st century; results from a 2000 yr simulation of the coupled Bern3D-26
LPX model in response to a step increase in radiative forcing by 3.7 Wm−2 ; and a quantification27
1
MULTIPLE GREENHOUSE-GAS FEEDBACKS FROM THE LAND BIOSPHERE UNDER FUTURE CLIMATE CHANGE SCENARIOS
© 2013 Macmillan Publishers Limited. All rights reserved.
2of “traditional” C cycle sensitivities.28
S2 Input data29
Figure S1 illustrates the model setup: which variables are prescribed to individual model com-30
ponents and which are simulated interactively. The naming of variables introduced here is31
followed throughout this document. Components in red represent the model parts used for32
the ’offline’ simulations (see Methods, main article). Inputs prescribed to LPX-Bern 1.0 are N-33
deposition (Ndep)1, mineral N-fertilisation (Nfert)2, distribution of croplands, pastures and urban34
areas (ALU)3, fixed distribution of peatland areas (Apeat)4 and seasonally inundated wetlands35
(Ainund)4. In offline mode, monthly temperature, precipitation and cloud cover are prescribed36
from the CMIP5 outputs (tempCMIP5, precCMIP5, ccovCMIP5, see also Tab.S1). In online mode,37
a spatial pattern per unit temerature change (tempANOM, precANOM, ccovANOM, derived for38
each CMIP5 model used, is scaled by the global mean temperature change simulated online by39
Bern3D (∆T), see Section S2). Simulated terrestrial emissions (eCO2LPX, eN2OLPX, eCH4
LPX)40
are complemented with other sources not simulated by LPX (eCO2EXT, eN2OEXT, eCH4
EXT)41
and an additional flux to close the atmospheric budget in 1900 AD (eN2OADD, eCH4ADD). At-42
mospheric concentrations (cCO2, cN2O, cCH4, cOtropos3 , rOH) are calculated online in Bern3D43
using a simplified atmospheric chemistry model (ATMOS. CHEM.5, 6) to simulate variations in44
the life time of CH4 and using prescribed emissions of reactive gases from the RCP database45
(eVOCRCP, eNOXRCP, eCORCP). cCO2 evolves as a result of the coupled oceanic and ter-46
restrial C cycle and is communicated back to LPX where it affects plant photosynthesis. The47
radiative forcing of all agents affected by variations of terrestrial GHG emissions (fCO2, fN2O,48
fCH4, fOtropos3 , fH2Ostratos) are simulated online in Bern3D after Joos et al. (2001)5 (Bern3D49
RADIATIVE FORCING). Radiative forcing from other agents (fAerosol, fCFCs, fHFCs) are50
prescribed directly from the RCP database. The global mean temperature increase (∆T) is cal-51
culated by Bern3D using a two-dimensional representation of the Earth energy balance7 and52
a three-dimensional physical ocean model8. The top arrow represents the communication of53
land albedo changes (∆α) from LPX to the radiative component of Bern3D. The bottom arrow54
visualizes the feedback from simulated cCO2 and ∆T on LPX. Land surface albedo changes in55
response to vegetation changes and snow cover are simulated based on Otto et al. (2011)9 and56
© 2013 Macmillan Publishers Limited. All rights reserved.
3are described in Steinacher 201110.57
Bern3DATMOS.CHEM.
eCO2EXT
eN2OEXT
eCH4EXT
eVOCRCP
eNOXRCP
eCORCP
eCO2LPX
eN2OLPX
eCH4LPX
eN2OADD
eCH4ADD
emissions
tempCMIP5
precCMIP5
ccovCMIP5
NdepNfertALUApeatAinund
inputs
LPX
cCO2cN2OcCH4cO3
tropos
rOH
concentrations
RCP
fCO2fN2OfCH4fO3
tropos
fH2Ostratos
radiative forcing
fAerosolfCFCsfHFCs
Bern3DRADIATIVEFORCING
RCPtempANOM
precANOM
ccovANOM
Δα
Bern3DCLIMATE
climate change
ΔT
ΔTcCO2
Figure S1: Model setup: Inputs and model components.
Climate58
In offline mode, monthly climate data is prescribed to LPX. CMIP5 climate output for sur-59
face temperature, precipitiation, and cloud cover is applied for all RCP2.6 and RCP8.5 experi-60
ments as given in Tab.S1. To correct CMIP5 model output for bias w.r.t. the present-day CRU61
climatology11, climate fields are anomalized as follows:62
CMIP5∗x,y,t = CMIP5x,y,t − CMIP5x,y + CRUx,y , (S1)
where CMIP5x,y,t is the original and CMIP5∗x,y,t is the offset-corrected CMIP5 climate variable63
field (surface temperature, precipitation, cloud cover) defined for t = 2005-2100 (2300) AD.64
Bars denote the mean over the years 1996-2005 AD.65
In online mode, a spatial pattern per unit temerature change, derived for each CMIP5 model66
and each month is scaled by the global mean temperature change (∆T) simulated online by67
Bern3D. Temperature and precipitation anomaly patterns are illustrated in Figures S2 and S3.68
© 2013 Macmillan Publishers Limited. All rights reserved.
4
Table S1: CMIP5 ensemble simulations used for offline simulations. ’r1’ refers to the CMIP5
terminology (’r1i1p1’), ’r2’ to ’r2i1p1’, etc. These are different simulation ensemble members
of CMIP5 experiments with equal forcings but slightly different initial conditions.
model RCP2.6 RCP8.5 Modeling Center
HadGEM2-ES r1, r2, r3, r4 r1, r2, r3, r4 Met Office Hadley Centre
MPI-ESM-LR r1, r2, r3 r1, r2, r3 Max-Planck Institute
for Meteorology
IPSL-CM5A-LR r1, r2, r3 r1, r2, r3, r4 Institut Pierre-Simon Laplace
MIROC-ESM r1 r1 Japan Agency for Marine-Earth
Science and Technology
CCSM4 r1, r2, r3, r4, r5 r1, r2, r3, r4, r5 National Center for
Atmospheric Research
N fertiliser input69
Mineral N fertilizer (Nfert) is assumed to be added to croplands only. Nfert inputs on pastures,70
as well as N inputs from manure are not simulated explicitly. Tracking C and N mass flow71
from harvest on agricultural land to soil application of animal manure and recycling of crop72
residues, with denitrification, volatilisation, and N2O emissions along the pathway, is beyond73
the the scope of the present study. N2O emissions from manure are prescribed instead (see74
Section below).75
Four equal doses of mineral N-fertiliser are added during the vegetation period to the soil76
nitrate and ammonium pool with a constant respective split of 1:7. For the historical period77
(1765-2005 AD), Nfert data is from Zaehle et al. (2011)2 (ZAE11), based on country-wise78
ammonium plus nitrate data from the FAO statistical database (1960-2005)12. For years 1910-79
1960, an exponential increase was assumed.80
For the years 2005-2100 AD, spatial Nfert data provided by the the IAM groups (RCP8.5:81
Riahi et al. (2011)13, pers. comm. K. Riahi, January 2012; RCP2.6: VanVuuren et al.82
(2011)14, 15, pers. comm. L. Bouwman, April 2012) is used to scale the 2005 AD-field from83
ZAE11 for each continent separately. Thereby, the relative increase in the total amount of an-84
© 2013 Macmillan Publishers Limited. All rights reserved.
5
Figure S2: Annual mean temperature anomaly patterns per ◦C global temperature change used
for coupled simulations [◦C/(global mean ◦C)]. Note that values above 1 represent locations
where regional temperature change is larger than on global average.
nual Nfert inputs in each continent is conserved from the original data delivered by the IAM85
groups, while the spatial pattern within each continent is conserved from the data of ZAE11 in86
year 2005 AD (see Figures S4, S5). This scaling can be described by87
NRCPt,i = NZAE11
2005,i
∑i=k
NRCP−origt,i∑
i=k
NRCP−orig2005,i
, (S2)
where NRCPt,i is the harmonized RCP Nfert scenario, NZAE11
2005,i=k is the spatialised (index i) field of88
ZAE11 in year 2005 AD. NRCP−origt,i is the original spatialised RCP scenario data for each time89
© 2013 Macmillan Publishers Limited. All rights reserved.
6
Figure S3: Annual mean precipitation anomaly patterns per degree global temperature change
used for coupled simulations [mm/month/(global mean ◦C)].
t and grid cell i. The sum over all grid cells i belonging to continent k is used to scale NZAE112005,i .90
For RCP8.5, the scaling factor is corrected to guarantee that the total Nfert input in 2100 and in91
each continent is identical as in the original data.92
© 2013 Macmillan Publishers Limited. All rights reserved.
7
year AD
TgN
/yr
1900 1950 2000 2050 21000
50
100
150
200
N−fertiliser input
historicalRCP 2.6RCP 8.5
year AD
TgN
/yr
1900 1950 2000 2050 21000
20
40
60
80
N−deposition
historicalRCP 2.6RCP 8.5
year AD
ppm
1900 1950 2000 2050 2100
200
400
600
800
1000
cCO2 for offline simulations
historicalRCP 2.6RCP 8.5
year AD
106 km
2
1900 1950 2000 2050 21000
10
20
30
40
50
60
land use
historicalRCP 2.6RCP 8.5
totalcroppasture
Figure S4: top left: Global mineral nitrogen fertilizer input (Nfert) [TgN/yr] for the historical
period (black), RCP2.6 (blue) and RCP8.5 (red). top right: Global atmospheric N deposition
from Lamarque et al. (2011)1 [TgN/yr] for the historical period (black), RCP2.6 (blue) and
RCP8.5 (red). bottom left: Atmospheric CO2 concentration as prescribed in offline simulations
for the historical period (black), RCP2.6 (blue) and RCP8.5 (red). bottom right: Global land
use area from Hurtt et al. (2006)3 [TgN/yr] for the historical period (black), RCP2.6 (blue) and
RCP8.5 (red).
© 2013 Macmillan Publishers Limited. All rights reserved.
8
Figure S5: Nfert input [gN m−2 yr−1], RCP2.6. (Left) and RCP8.5 (Right), for years 2005,
2030, 2050 and 2100 (top to bottom). Fertiliser is applied to croplands. These are defined by
Hurtt et al. (2006)3.
© 2013 Macmillan Publishers Limited. All rights reserved.
9N deposition93
Annual fields for atmospheric NHx and NOy deposition are from Lamarque et al. (2011)1,94
generated by an atmospheric chemistry/transport model and provided for the historical period95
as well as for RCP scenarios of the 21st century. NHx and NOy are added to the ammonium and96
nitrate pool in LPX along with daily precipitation. For the present study, we treat N deposition97
as an external forcing, meaning that it is not affected by climate or CO2. The assessment of a98
feedback between climate and CO2, emissions of NO, NO2 and NH3 from soils, atmospheric99
transport and chemical reactions, deposition and radiative forcing is beyond the present study.100
We summarize the sum of N deposited and Nfert as “reactive N inputs” (Nr).101
Land use change102
Anthropogenic land use change (LU) is treated as an external forcing (see Figure 1 in the main103
text) and is prescribed also in the ’ctrl’ online and offline simulations. LU is prescribed as104
maps for each year from Hurtt et al. (2006)3. Resulting CO2 emissions from deforestation105
are simulted by the model. A thorough description can be found in previous publications16, 17.106
Note that LU also has indirect effects by changing the C sink capacity under rising cCO216.107
This is reflected in a stronger negative feedback factor rC∆C (Figure S18) when the model is set108
up without accounting for LU (simulation DyNrPt, Table S4, and Figure S10).109
N2O and CH4 emissions not simulated by LPX110
For eN2OEXT we use historical emission data for domestic/industrial sources, fire, and manure111
as described in Zaehle et al. (2011)2 (Figure S6). Domestic/industrial emissions were derived112
from VanAardenne et al. (2001)18 giving a flux of 1.2 TgN2O-N/yr in 2005 AD. The biomass113
burning estimate (0.5 TgN2O-N/yr in 2005 AD) is from Davidson (2009)19. Manure-N2O flux114
is taken as a fraction of global manure-N input yielding 2.2 TgN2O-N/yr in 2005 AD. To extend115
N2O emissions to 2100 AD, we scale the total of domestic/industrial plus fire plus manure116
emissions in year 2005 AD with the relative increase in the sum of respective categories in117
each RCP scenario. RCP emission data are consistent with the economical, demographic, and118
© 2013 Macmillan Publishers Limited. All rights reserved.
10political development in the respective RCP scenarios as simulated by Integrated Assessment119
Modelling.120
To complete the N2O budget and reproduce the atmospheric concentration for pre-industrial121
conditions, we tuned the oceanic source to 3.3 TgN2O-N/yr (eN2OADD in Figure S1). This is in122
agreement with the broad range of available estimates (1.2-5.8 TgN2O-N/yr)20–22. The oceanic123
source is scaled by 3.3% between 1850 and 2005 AD with the scaling factor following the in-124
crease in atmospheric N deposition. This increase reflects the increase in reactive N in oceans125
due to atmospheric deposition23. After 2005 AD, the oceanic source is held constant in all126
scenarios.127
Non-soil CH4 emissions are taken from the RCP database24. These include emissions128
from biomass burning and wet rice cultivation which are not explicitly simulated by LPX. To129
close the atmospheric CH4 budget and reproduce the atmospheric concentration in 1900 AD,130
we tuned the additional prescribed source (representing geological and small oceanic sources)131
to 38 TgCH4/yr (eCH4ADD in Figure S1). This is based on a data spline of southern-hemisphere132
atmospheric records as provided by the RCP database24.133
year AD
TgN
/yr
1900 1950 2000 2050 21000
2
4
6
8
10Zaehle et al., 2011RCP 2.6RCP 8.5ocean
year AD
TgN
/yr
1900 1920 1940 1960 1980 20000.0
0.5
1.0
1.5
2.0
2.5
3.0domestic/industrialfiremanure
Figure S6: Left: N2O emissions not explicitly simulated by LPX (eN2OEXT) and oceanic emis-
sions (green). The black line (’Zaehle et al., 2011’) is the sum of domestic/industrial, fire, plus
manure as given in right plot. Right: eN2OEXT by sources. Both are given in TgN2O-N/yr.
© 2013 Macmillan Publishers Limited. All rights reserved.
11S3 Simulation protocol and feedback quantification134
Feedback formalism135
Our framework to quantify feedbacks between land and climate follows the formalism applied136
in physical climate science as presented, e.g., in Roe (2009)25 and Gregory et al. (2009)26.137
The latter also address feedbacks between climate and the carbon cycle. Here, we extend138
this concept to other radiative agents (eCH4, eN2O and albedo) mediated by the terrestrial139
biosphere and affected by environmental conditions (climate; and atmospheric CO2 concen-140
trations, cCO2). We start with a brief introduction into the feedback formalism. Consider the141
Earth’s climate to be a system responding to a radiative forcing F with a radiative response H ,142
so that in equilibrium, the net energy flux into the system N is zero and no warming or cooling143
occurs.144
N = F −H , N = 0 ⇒ F = H (S3)
Observations confirm that H can be linearized with respect to the temperature change ∆T 26,145
so that146
F = λ ·∆T (S4)
λ is the climate feedback factor given in Wm−2K−1 and is equal to the inverse of the climate147
sensitivity factor. λ is thus the basic quantity to describe the temperature change of the climate148
system in response to a given radiative forcing. However, λ summarizes all feedbacks operat-149
ing. To quantify an individual feedback, we define a reference system, in which the feedback150
of interest is not operating. The most basic reference system is to consider the Earth as a Black151
Body. Here, we chose the reference system to repesent the ocean-atmosphere climate system152
without any interaction with the land. This is the control simulation (termed ’ctrl’), in which153
the radiative forcing F leads to a temperature change ∆T ctrl (see Figure 4 in main text, bottom154
right, dashed line).155
∆T ctrl =F
λ0
(S5)
λ0 is the sum of all non-land feedbacks operating in the control simulation (the Black Body156
response or Planck feedback (BB), water vapor (WV), ice-albedo (α ), lapse rate (LR), cloud157
(C), etc. λ0 = λBB + λWV + λα + λLR + λC + .... Note, that the radiative forcing F depends158
on the reference system chosen. Note also that in our reference system, the land is still affected159
© 2013 Macmillan Publishers Limited. All rights reserved.
12by external forcings (land use, Nr inputs), which leads to terrestrial GHG emissions and albedo160
change, eventually affecting ∆T ctrl.161
When a feedback is included, the system adjusts to a different temperature ∆T because162
it now “sees” an additional radiative forcing (∆F ) triggered by the feedback. E.g. a warmer163
climate stimulates terrestrial N2O emissions which increase its atmospheric concentration and164
lead to additionally absorbed energy due to its greenhouse effect. Let us look at “land” as a165
feedback element in the climate system interacting via a multitude of feedbacks. We summarize166
these as λland. With the additional radiative forcing from all land feedbacks written as ∆F =167
rland ·∆T we get168
λ0 ·∆T = F + rland ·∆T . (S6)
With r = −λ we get the form presented in the paper (Eq.2)169
F = (λ0 + λland) ∆T . (S7)
This illustrates that the additional radiative forcing per degree temperature change (r) caused170
by the feedback of interest is equal to the the negative of the feedback factor λ. Equations171
(S5) and (S7) are combined to derive λ using a control simulation (’ctrl’) and a fully coupled172
simulation (’CT’, see Table S3). Equation (S6) can be rewritten as173
∆T =F
λ0
+r
λ0
∆T , (S8)
illustrating that the feedback arises because a fraction f = rλ0
of the system output ∆T is174
fed back into the input. We can take a different perspective and characterise the effect of a175
feedback with the gain factor G = ∆T∆T ctrl . By combining Equations (S5) and (S6), the gain176
factor becomes177
G =∆T
∆T ctrl=
Fλ0−λ0f
Fλ0
=λ0
λ0 − λ0f=
1
1− f(S9)
Note that f = rλ0
is often referred to as the “feedback factor”, but not here, where the feedback178
factor is λ = −r. The advantage of the formulation of Equation (S6) and (S7) is that individual179
feedbacks can be added to derive their combined effect.180
λ0 ·∆T = F + ∆T∑i
ri , (S10)
© 2013 Macmillan Publishers Limited. All rights reserved.
13or in the form presented in the paper181
F = (λ0 +∑i
λi) ∆T . (S11)
Note that f = 1λ0
∑i ri and that G 6=
∑iGi.182
Model setups183
So far, we have been looking at feedbacks arising simply “from land”. However, a multitude of184
processes affecting climate are operating in terrestrial ecosystems. One way is to decompose185
the total land feedback into contributions from individual forcing agents (eN2O, eCH4, ∆C,186
and ∆albedo):187
λland = λ∆C + λCH4 + λN2O + λ∆albedo + δ (S12)
δ is a non-linearity term. To isolate individual λs, the model has to be set up, where only the188
respective feedback is operating. In practice, we prescribe the time series of global terrestrial189
emissions from the control run for all non-operating forcing agents. In the case of albedo, we190
prescribe the monthly two-dimensional field from the control run. Table S3 provides a full191
account of all model setups applied.192
A further decomposition of λland can be done by drivers of land feedbacks. Not only193
climate (superscript ’T’) but also atmospheric cCO2 (superscript ’C’) affects terrestrial GHG194
emissions and albedo. We quantify its effects in the same framework.195
λland = λC + λT + δ ≡ λCT (S13)
Simulations where only changes of the respective driver is communicated to the land196
model (see also Figure 1 in the main article) are used to quantify λT and λC. In “climate-197
coupled” simulations, only feedbacks arising from the sensitivity of forcing agents to climate198
are taken into account and lead to a temperature change ∆TT:199
F = (λ0 + λT) ∆TT , (S14)
In “cCO2-coupled” simulation, the land sees only changes in cCO2 and the system attains a200
© 2013 Macmillan Publishers Limited. All rights reserved.
14temperature ∆TC:201
F = (λ0 + λC) ∆TC , (S15)
Additionally, we quantify the modification of feedbacks by the individual effects of C-202
N interactions, peatland C dynamics27, anthropogenic land use change, and Nr inputs (Figure203
S18). Respective feedback factors are quantified identically but with LPX not simulating re-204
spective features (see Tab. S4). This requires the full set of coupled as well as ’ctrl’ simulations205
for each setup. Due to non-linearities in the system, the “expansion” with respect to the full206
setup (by turning one of each feature off in an individual setup) is preferred over an expansion207
w.r.t. the “null-”setup (by turning only one of each feature on). To quantify the modification208
by C-N interactions, results from a carbon-only version of LPX are used.209
To assess the sensitivity of terrestrial GHG emissions to different drivers and to capture210
the effects of the range of climate projections in different CMIP5 models, we use LPX in211
an offline mode (see Table S2 and Figure S1) and only evaluate emissions (Figure 3 in main212
article).213
Table S2: Offline simulations overview. Climate (∆T) and atmospheric CO2 (∆CO2) and Nr
(Ndep and Nfert) are either prescribed from CMIP5 climate models, RCP data, and other sources
or held constant (’0’). Model features anthropogenic land use change (LU), interactive carbon-
nitrogen cycling (DyN), and C-N dynamics/CH4 emissions on peatlands (peat) are turned on in
all simulations. The lines in main article-figure 2 relate as follows: ’standard’ is r1, ’no CO2’
is r3, ’no climate’ is r5, ’no Nr’ is r2.
simulation ∆T ∆CO2 Ndep Nfert LU DyN peat
r1 CMIP5 RCP 1 1 1 1 1
r2 CMIP5 RCP 0 0 1 1 1
r3 CMIP5 0 1 1 1 1 1
r4 CMIP5 0 0 0 1 1 1
r5 0 RCP 1 1 1 1 1
r6 0 RCP 0 0 1 1 1
r7 0 0 1 1 1 1 1
r8 0 0 0 0 1 1 1
© 2013 Macmillan Publishers Limited. All rights reserved.
15
Table S3: Couplings overview. Columns ∆CO2 and ∆T indicate which drivers are communi-
cated to the land model LPX. Columns cCO2, cCH4, cN2O, and ∆α indicate whether variations
in the respective forcing agent affect the climate module in Bern3D (X) or if the climate module
responds to variations in respective agents prescribed from the control run (’ctrl’).
name ∆T ∆CO2 cCO2 cCH4 cN2O ∆α
control
ctrl 0 0 X X X X
fully coupled
CT 1 1 X X X X
cCO2-land coupled
C 0 1 X X X X
climate-land coupled
T 1 0 X X X X
fully coupled - single agent
CT-∆CO2 1 1 X ctrl ctrl ctrl
CT-∆CH4 1 1 ctrl X ctrl ctrl
CT-∆N2O 1 1 ctrl ctrl X ctrl
CT-∆α 1 1 ctrl ctrl ctrl X
fully coupled - CO2/albedo only
CT-∆CO2-∆α 1 1 X ctrl ctrl X
Table S4: Features overview. Model features, variably turned on (’1’) and off (’0’) are: an-
thropogenic land use change (LU), interactive carbon-nitrogen cycling (DyN), N-deposition
(Ndep), N-fertilisation (Nfert), and C-N dynamics/CH4 emissions on peatlands (peat). For the
model setup with DyN turned off, the carbon-only version of LPX was used.
name LU DyN Ndep Nfert peat
LuDyNrPt 1 1 1 1 1
LuDyNr 1 1 1 1 0
LuDyNPt 1 1 0 0 1
LuPt 1 0 0 0 1
DyNrPt 0 1 1 1 1
© 2013 Macmillan Publishers Limited. All rights reserved.
16Equilibrium climate sensitivity214
Climate sensitivity is conventionally defined as the temperature response to a doubling of cCO2,215
thus not involving slowly adjusting biogeochemical feedbacks28. Here, we assess climate sen-216
sitivity to a sustained radiative forcing of 3.7 Wm−2, corresponding to a nominal doubling of217
preindustrial CO2 levels. Note that the climate sensitivity is inversely proportional to the sum218
of all feedbacks 1/(λ0 +λland). Bern3D is tuned to yield a conventionally defined sensitivity of219
∼2.9◦C. We assess results for (i) a simulation with interactive land biosphere and all feedbacks220
operating (setup like ’CT-LuDyNrPt’) (ii) a simulation with interactive land biosphere where221
only feedbacks from albedo and terrestrial C storage are operating (setup like ’CT-∆CO2-∆α’)222
and (iii) a simulation without land-climate interactions (simulation setup like ’ctrl-LuDyNrPt’,223
Tab.S3, see Figure S7). The coupled Bern3D-LPX model is run for 2000 simulation years. All224
boundary conditions (Nr inputs, land use, initial atmospheric CO2, initial climatology) are set225
to preindustrial values. We chose to compare the fully coupled simulation ’CT-LuDyNrPt’ with226
’CT-∆CO2-∆α because the latter represents a setup commonly represented by latest-generation227
Earth system models (e.g., CMIP5 models).228
In our simulations, feedbacks from terrestrial C storage and albedo amplify the equilib-229
rium temperature increase by 0.4-0.5◦C, while the combination of all simulated land-climate230
interactions finally results in 3.4-3.5◦C warming, 0.6-0.7◦C (or 22-27%) above the 2.8◦C warm-231
ing when only non-land climate feedbacks are operating (see Figure S7).232
Values for λland reported here are somewhat higher than derived from the RCP8.5 simula-233
tions as presented in the article. Differences are likely linked to total C in the system. In RCP8.5234
fossil fuel combustion adds C and stimulates C storage on land, acting as a negative feedback.235
Applying present-day boundary conditions would enhance the positive feedback from N2O due236
to higher Nr loads in soils. Assumptions regarding the state of land use used for the equilibrium237
assessment further influence results. This scenario-dependence of any feedback quantification238
may be interpreted as favouring the use of scenarios with consistent future developments in239
land use and emissions of GHGs and other forcing agents.240
© 2013 Macmillan Publishers Limited. All rights reserved.
17
∆T [o C
]
0
1
2
3
4a ∆T (3.7 Wm−2)
ctrl2 × cCO2
CT−∆C−∆αCT−all agents
simulation year
Wm
−2K
−1
0 500 1000 1500 2000
0.0
0.1
0.2
0.3
0.4
0.5b feedback factor rCT
Figure S7: Upper panel: Global mean temperature increase in response to a radiative forcing
of 3.7 Wm−2. The black curve represents the ’ctrl’ simulation, where no feedbacks from land
are accounted for. The dotted black curve represents the temperature change in response to a
doubling of atmospheric cCO2, the “conventionally defined” climate sensitivity as referred to
in the main article. The blue range represents a setup where only terrestrial feedbacks from ∆C
and albedo are accounted for. The red range represents a setup where also eN2O and eCH4 are
operating (see also Table S3). The difference between the red and blue range is due to effects
from terrestrial eN2O and eCH4. The range of temperature response arises from the sensitiv-
ity to different climate change patterns. Abrupt temperature changes (e.g., ’ctrl’ in simulation
year 1100) are due to abrupt transitions in ocean convection and associated temperature mix-
ing. Lower panel: Total land-climate feedback factor (λland) with colors representing the same
setups as in the upper panel. All external forcings of the land (land use, Nr) and initial state
(cCO2) are preindustrial conditions.
© 2013 Macmillan Publishers Limited. All rights reserved.
18Carbon cycle sensitivities241
Feedback factors rC∆C and rT
∆C are directly linked to the sensitivities of C stocks to temperature242
(γ) and atmospheric cCO2 (β) and linear relationships can be established as an approximation243
26.244
rC∆C ∼ β (S16)
245
rT∆C ∼ γ (S17)
Sensitivities of C storage are commonly quantified, e.g., in model intercomparison projects29,246
and are computationally less expensive to derive as values can be calculated from offline simu-247
lations by regressing C storage changes to prescribed temperature or atmospheric cCO2. Here,248
β and γ are presented for comparison with other studies and are derived from the cCO2-land249
coupled and the climate-land coupled experiments (online, RCP8.5) as follows:250
∆CC = β ·∆cCO2 (S18)251
∆CT = γ ·∆TT (S19)
∆CC is the change in terrestrial C storage in the respective experiment (again, superscript ’C’252
represents the cCO2-land coupled and ’T’ the climate-land coupled experiment). ∆cCO2 is the253
change in atmospheric concentration of CO2 and ∆T the change in global mean temperature.254
Both sensitivities exhibit non-linearity pointing to a stronger positive feedback from land255
under high cCO2 and temperatures (Figure S8). The cCO2 sensitivity (β) is flatening out to-256
wards high cCO2 levels, while the temperature sensitivity (γ) is increasing with the magnitude257
of warming. The single most important model feature reducing β is anthropogenic land use258
change. The replacement of natural vegetation by agricultural land reduces the the ecosystem’s259
sink capacity due to shorter life time of C in grass and crop vegetation as opposed to forests. At260
the same time, land use change implies a reduction of γ due to generally smaller C stocks prone261
to temperature-driven reduction. To derive the net effect of land use change in a scenario for262
future temperature and cCO2 change, one has to turn to the feedback factor as derived above.263
For changes in cCO2 of less than 200 ppm, C-N interactions is the most important feature264
reducing β. This is likely a transient effect of initial N limitation, relieved by higher N reminer-265
© 2013 Macmillan Publishers Limited. All rights reserved.
19alization after the system has adopted to high cCO2 levels and increased the size of total soil266
organic N. Accounting for C-N interactions reduces the value of γ due to higher N availability267
at warmer soil temperatures. Nr inputs have minor impacts on β and γ in our model. Additional268
100 PgC are lost from peatlands under high temperatures and on long time scales leading to an269
increase in γ .
−500 0 500 1000 1500 20000
100
200
300
400
500
600
700
800
∆ Catm
[ppm]
∆ C
terr [
Pg
C]
no DyN
no LU
no peat
no Nr
standard
0 500 1000 15000.3
0.4
0.5
0.6
0.7
0.8
0.9
1
∆ Catm
[ppm]
β [
Pg
C/p
pm
]
no DyN
no LU
no peat
no Nr
standard
0 2 4 6 8 10−1200
−1000
−800
−600
−400
−200
0
200
∆ T [K]
∆ C
terr [P
gC
]
no DyN
no LU
no peat
no Nr
standard
0 2 4 6 8 10−110
−100
−90
−80
−70
−60
−50
−40
−30
−20
−10
0
∆ T [K]
γ [P
gC
/K]
no DyN
no LU
no peat
no Nr
standard
Figure S8: Upper left: Change in terrestrial C storage vs. atmospheric C (CO2). Upper right:
β vs. atmospheric C. Lower left: Change in terrestrial C storage vs. global mean temperature
change. Lower right: γ vs. global mean temperature change.270
© 2013 Macmillan Publishers Limited. All rights reserved.
20S4 Supplementary results271
In the following sections we provide additional documentation of the results from the offline272
simulations. Maps for changes in N2O (Figure S11) and CH4 (Figure S14) emissions and273
terrrestrial C storage (Figures S9 and S10) provide spatial information, given separately for274
each CMIP5 model’s climate input (mean over available ensemble members).275
Terrestrial C balance276
Changes in terrestrial C storage (∆C) as illustrated in Figure S9 are the result of external forc-277
ings (land use change, Nr) and the impact of changes in climate and cCO2. Figure S10 shows278
separated effects of external forcings only (top left) and changes in climate and cCO2 as the279
difference to the total effect (’CT-ctrl’). At high northern latitudes, temperature increase acts280
to relieve the limitation of plant growth by temperature and low nutrient availability and leads281
increased C storage, while at lower latitudes, warmer temperatures generally reduce soil C stor-282
age by enhancing soil C decomposition. The C balance of forest biomes (boreal and tropical) is283
sensibly affected by vegetation dynamics responding to water stress, exceedance of bioclimatic284
limits, etc. and exhibits abrupt transitions (collapse of vegetation) leading to a sharp decline in285
primary productivity and C storage.286
© 2013 Macmillan Publishers Limited. All rights reserved.
21
Figure S9: ∆Ctot, change in total (vegetation, litter, soil) terrestrial C storage [gC/m2], 2100-
2000 AD, in RCP8.5, from offline simulation ’CT’, and based on different CMIP5 climates.
Differences are taken from the means of the years 2006 to 2011 AD and 2095 to 2100 AD.
Brown colors represent C release from the terrestrial biosphere.
© 2013 Macmillan Publishers Limited. All rights reserved.
22
Figure S10: ∆Ctot, change in total (vegetation, litter, soil) terrestrial C storage [gC/m2], 2100-
2000 AD, in RCP8.5 and based on different CMIP5 climates. upper left: from offline sim-
ulation ’ctrl’ where the land model “sees” no changes in climate or CO2 and is affected only
by external forcings (land use, N-deposition, N-fertiliser). rest: difference between ’CT’ and
’ctrl’ simulation; represents effects due to changes in climate and CO2. Differences are taken
from the means of the years 2006 to 2011 AD and 2095 to 2100 AD. Brown colors represent C
release from the terrestrial biosphere.
© 2013 Macmillan Publishers Limited. All rights reserved.
23N2O emissions287
Simulated present-day N2O emissions from terrestrial ecosystems of 9.1 TgN2O-N/yr are within288
the range of other studies2, 20, 21, 30, 31. The spatial pattern of the N2O increase in the 21st century289
is similar for all prescribed CMIP5 climates (Figure S11). CCSM4 shows the smallest increase290
across different regions. Most of the increase in N2O emissions is from agricultural land (crop-291
land and pastures, Figure S12). The amplification of the N2O source from agricultural soils292
(from 1.4 in 1900 AD to 4.9 TgN2O-N/yr in 2005 AD) is a combination of expansion of areas293
under anthropogenic land use and an increase in fertiliser-N inputs and N-deposition per unit294
area. Due to the continuous increase of Nr inputs in RCP8.5 throughout the 21st century, N2O295
emissions from agricultural soils reach 9-11 TgN2O-N/yr by 2100 AD.296
Interestingly, the relative increase in agricultural N2O emissions is larger than the relative297
increase in anthropogenic Nr inputs to agricultural land. We define here a dimensionless N2O298
emission factor as the ratio of N2O emissions from agricultural land divided by anthropogenic299
Nr inputs on agricultural land (N-fertilisation plus N-deposition). Thus the values of the emis-300
sion factor quantified here cannot be compared directly to results of Davidson (2009)19 and301
Crutzen et al. (2008)32. Note, that other inputs of fixed N which did not see a magnitude in the302
relative increase as for N-deposition and fertiliser-N (biological N fixation or manure), are not303
accounted for here. The temporal evolution of this emission factor for simulations with climate304
change (r1, red) and a simulation without climate change (r5, red) is illustrated by Figure S13.305
Two important features of this evolution emerge: (i) A drop of the emission factor from 0.24 to306
0.04 from pre-industrial times to present. This is due to the drastic increase in anthropogenic Nr307
inputs. Nr inputs on agricultural land (N-deposition plus N-fertiliser) increase from 3 TgN/yr308
in 1850 AD to 131 at present and to 239 TgN/yr in 2100 AD in RCP8.5. Its relative increase309
is much stronger than the relative increase in N2O emissions. (ii) The divergence of emission310
factors in the 21th century for the RCP8.5 scenario, as climate change is responsible for an311
increase in the share of reactive N input lost as N2O.312
© 2013 Macmillan Publishers Limited. All rights reserved.
24
Figure S11: ∆eN2O[gN2O-N/m2/yr], 2100-2000 AD, in RCP8.5, from offline simulation ’CT’,
and based on different CMIP5 climates. Differences are taken from the means of the years 2006
to 2011 AD and 2095 to 2100 AD.
Figure S12: Global total N2O emissions on natural (left) and agricultural (right) land.
© 2013 Macmillan Publishers Limited. All rights reserved.
25
Figure S13: N2O emission factor. Defined as the ratio of N2O emissions from agricultural land
divided by Nr inputs on agricultural land. For a simulation with climate change (r1, black) and
a simulation without climate change and cCO2 changes (r7, red). RCP8.5 climates change from
all CMIP5 models is prescribed for the 21st century.
© 2013 Macmillan Publishers Limited. All rights reserved.
26CH4 emissions313
Modelled CH4 emissions from natural ecosystems increase from 195 at preindustrial to 219314
TgCH4/yr at present and further to 228-241 in RCP2.6 and 304-343 TgCH4/yr in RCP8.5 (Fig-315
ure 2b). Increased substrate availability for methanogenesis due to a strong stimulation of NPP,316
and faster soil turnover leads to an amplification of CH4 emissions with the sharpest increase317
in peatlands (plus 120-200%). Other CH4-related analyses with the same model are presented318
in Spahni et al. (2011)33 and Zurcher et al. (2012)34. Changes in tropical inundated wetland319
emissions are less pronounced, and perhaps underestimated in our model that does not account320
for wetland expansion under future climate change35, 36. The additional CH4 release from nat-321
ural land ecosystems is not accounted for in the climate projections in preparation of the Fifth322
Assessment Report of the Intergovernmental Panel on Climate Change24, 37. In our simulations323
cCH4 rises up to 4500 ppb in RCP8.5 by 2100 AD (see online simulation, below), about 800324
ppb more than in the RCP data.325
© 2013 Macmillan Publishers Limited. All rights reserved.
27
Figure S14: ∆eCH4LPX [gCH4/m2/yr], 2100-2000 AD, in RCP8.5, from offline simulation
’CT’, and based on different CMIP5 climates. Based on different CMIP5 climates. Differences
are taken from the means of the years 2006 to 2011 AD and 2095 to 2100 AD.
© 2013 Macmillan Publishers Limited. All rights reserved.
28Albedo326
Figure S15: ∆albedo, 2100-2000 AD, in RCP8.5. Combined effect of climate and cCO2 (upper
left) , effect of climate only (upper right), effect of cCO2 (lower left); on natural land only (no
land use). Map in lower right illustrates effect of external forcings (land use, Nr) for RCP8.5.
Differences are taken from the means of the years 2000 to 2010 AD and 2090 to 2100 AD.
Negative values (red, decreasing albedo) imply more absorbtion of shortwav radiation at the
surface, which leads to warming.
© 2013 Macmillan Publishers Limited. All rights reserved.
29Supplementary References327
3281. Lamarque, J.-F. et al. Global and regional evolution of short-lived radiatively-active gases329
and aerosols in the Representative Concentration Pathways. Clim. Change 109, 191–212330
(2011).331
2. Zaehle, S., Ciais, P., Friend, A. D. & Prieur, V. Carbon benefits of anthropogenic reactive332
nitrogen offset by nitrous oxide emissions. Nature Geoscience 4, 601–605 (2011).333
3. Hurtt, G. C. et al. The underpinnings of land-use history: three centuries of global gridded334
land-use transitions, wood-harvest activity, and resulting secondary lands. Global Change335
Biol. 12, 1208–1229 (2006).336
4. Tarnocai, C. et al. Soil organic carbon pools in the northern circumpolar permafrost region.337
Global Biogeochem. Cycles 23, GB2023+ (2009).338
5. Joos, F. et al. Global warming feedbacks on terrestrial carbon uptake under the Inter-339
governmental Panel on Climate Change (IPCC) emission scenarios. Global Biogeochem.340
Cycles 15, 891–907 (2001).341
6. Strassmann, K. M., Plattner, G.-K. & Joos, F. CO2 and non-CO2 radiative forcings in342
climate projections for twenty-first century mitigation scenarios. Climate Dynamics 33,343
737–749 (2009).344
7. Ritz, S. P., Stocker, T. F. & Joos, F. A coupled dynamical oceanenergy balance atmosphere345
model for paleoclimate studies. J. Climate 24, 349–75 (2011).346
8. Muller, S. A., Joos, F., Edwards, N. R. & Stocker, T. F. Water mass distribution and347
ventilation time scales in a cost-efficient, three-dimensional ocean model. J. Clim. 19,348
5479–5499 (2006).349
9. Otto, J., Raddatz, T. & Claussen, M. Strength of forest-albedo feedback in mid-holocene350
climate simulations. Climate of the Past 7, 1027–1039 (2011). URL http://www.351
clim-past.net/7/1027/2011/.352
10. Steinacher, M. Modeling changes in the global carbon cycle-climate system. Ph.D. thesis,353
University of Bern (2011).354
© 2013 Macmillan Publishers Limited. All rights reserved.
30
cCO2-land coupling
cN2O
cCH4
albedo
ΔC
eN2O
eCH4
Δalbedo
RF ΔTC
ΔC
eN2O eCH4
Δalbedo
cCO2
cN2O
cCH4
albedo
RF
land use N-deposition N-fertiliser
climate-land coupling
LAND
cCO2
eGHGEXT
eGHGEXT
ΔTT
Figure S16: Schematic illustration of feedbacks loops in a climate-land coupled (red back-
ground, superscript ’T’) and a cCO2-land coupled (blue background, superscript ’C’) model
setup. The land model simulates terrestrial emissions of GHGs (∆C , eN2O, eCH4) and land
surface albedo changes. Here, GHGs and albedo are referred to as forcing agents. In combi-
nation with other non-soil GHG emissions not simulated by the land model (eGHGEXT), these
determine the atmospheric GHG concentrations (cGHG) and the surface albedo, cause a radia-
tive forcing (RF) and affect climate, here represented by the global mean temperature (∆T).
In all setups, external forcings are prescribed to the land model (land use, N deposition+N fer-
tiliser=Nr) or are added directly into the atmosphere (eGHGEXT). Note that even in the ’ctrl’
setup, land use, N-deposition, and N-fertiliser affect terrestrial GHG emissions, and eGHGEXT
is added and ∆Tctrl is not zero. In cCO2-land coupled (’C’) and climate-land coupled (’T’)
experiments only changes in one variable (cCO2 in the case of ’C’ and climate/temperature in
the case of ’T’) are communicated to the land model (not shown).
© 2013 Macmillan Publishers Limited. All rights reserved.
31
observed/RCP cCO2
CRU/CMIP5 climate
ΔC
eN2O
eCH4
land use N-deposition N-fertiliser
LAND
eEXT
T C
cCH4
cN2O
eEXT
Figure S17: Schematic illustration of the model setup used for offline simulations. Prescribed
inputs to the land model are the evolution of land use areas, reactive nitrogen addition (N
deposition and N fertilizer), historical observed cCO224 and climate11. The land response in
forcing agents (∆C, eN2O, eCH4) are presented in Figure 3 in the main article. Letters ’C’ and
’T’ indicate which changes in drivers (cCO2 and climate) are communicated to the land model.
Simulated emissions are complemented with other sources not simulated by LPX (eEXT) to
derive cN2O and cCH4 as presented in Figure 2 in the main article. Note that the prescribed
emissions of reactive gases (CO, VOC,NOX) affect cCH4. cCO2 cannot be derived from offline
simulations as no ocean carbon cycle model is operating in this setup. Note also that in offline
simulations, the ’ctrl’ simulation sees no changes in cCO2 and climate with respect to their
preindustrial state.
© 2013 Macmillan Publishers Limited. All rights reserved.
32−0.4 −0.3 −0.2 −0.1 0.0 0.1 0.2 0.3 0.4
present2100
2300
a
clim
ate
coup
led
Tfull featuresno Nrno C−N interactionsno peatlandno landuse
present2100
2300
b
cCO
2 −
land
cou
pled
C
feedback parameter r [Wm−2K−1]−0.4 −0.3 −0.2 −0.1 0.0 0.1 0.2 0.3 0.4
present2100
2300
c
fully
cou
pled
CT
Figure S18: Feedback factor rT (a), rC (b), rCT (c) evaluated by model features. Feedback
factors represent the combined feedbacks from all forcing agents (eCO2, eN2O, eCH4, and
albedo). Quantified at three time periods: present (mean of 2000-2010 AD), 2100 (mean of
2095-2105), and 2300 (mean of 2290-2300). Rectangles represents minimum (left edge), max-
imum (right edge), and mean (middle line) of values derived from simulations with different
climate change anomaly patterns from the five CMIP5 models applied. Results are from online
simulations. © 2013 Macmillan Publishers Limited. All rights reserved.
33R
F [W
m−2
]
−0.4
−0.2
0.0
0.2
0.4
0.6
0.8
1.0
a ∆RF(N2O)
−0.4
−0.2
0.0
0.2
0.4
0.6
0.8
1.0
RF
[Wm
−2]
b ∆RF(CH4)
RF
[Wm
−2]
1900 2000 2100 2200 2300
−0.4
−0.2
0.0
0.2
0.4
0.6
0.8
1.0
c ∆RF(CO2)
1900 2000 2100 2200 2300
−0.4
−0.2
0.0
0.2
0.4
0.6
0.8
1.0
RF
[Wm
−2]
d ∆RF(albedo)
RCP 2.6RCP 8.5
Figure S19: Additional radiative forcing due the higher N2O (a), CH4 (b), CO2 (c), and albedo
(d) in the fully coupled simulation compared to the ctrl simulation. ∆RF is caused by terres-
trial feedbacks. Note that changes in cCH4concentrations affect stratospheric cH2O and cO3
and hence invoke an indirect radiative forcing. Resulting radiative forcings are included in
RF(CH4). The sum ∆RF(N2O+CH4) is presented in Figure 4 in the main article (bottom left).
© 2013 Macmillan Publishers Limited. All rights reserved.
34
year AD
cCO
2 [p
pmv]
1900 2000 2100 2200 2300
0
500
1000
1500
2000
2500
cCO2
RCP 2.6RCP 8.5
Figure S20: Simulated CO2 concentrations in the fully coupled (range) and the ctrl (dashed
lines) simulations for RCP2.6 (blue) and RCP8.5 (red). Higher (lower) concentrations in the
fully coupled simulation RCP8.5 (RCP2.6) are due to enhanced C sources/reduced C sinks in
response to changes in both climate and CO2.
© 2013 Macmillan Publishers Limited. All rights reserved.
35−0.3 −0.2 −0.1 0.0 0.1 0.2 0.3
a
fully
cou
pled
CT
present2100
2300
b
clim
ate
− la
nd c
oupl
edT
present2100
2300
albedo∆CN2OCH4
feedback factor r [Wm−2K−1]−0.3 −0.2 −0.1 0.0 0.1 0.2 0.3
c
cCO
2 −
land
cou
pled
Cpresent
21002300
Figure S21: Feedbacks factors evaluated from RCP2.6. (a) Feedback factors in fully coupled
simulations rCTi , with i = CO2,N2O,CH4, albedo. (b) Feedback factors in climate-land cou-
pled simulations rTi . (c) Feedback factors in cCO2-land coupled simulations rC
i . Quantified at
three time periods: present (mean of 2000-2010 AD), 2100 (mean of 2095-2105), and 2300
(mean of 2290-2300). Rectangles represents minimum (left edge), maximum (right edge), and
mean (middle line) of values derived from simulations with different climate change anomaly
patterns from the five CMIP5 models applied. Results are from online simulations.© 2013 Macmillan Publishers Limited. All rights reserved.
3611. Mitchell, T. D. & Jones, P. D. An improved method of constructing a database of monthly355
climate observations and associated high-resolution grids. Int. J. Climatol. 25, 693–712356
(2005).357
12. FAOSTAT. Resourcesstat: Fertilisers. Website (2009). Available from http://358
faostat.fao.org accessed 08/12/2009.359
13. Riahi, K. et al. RCP 8.5-A scenario of comparatively high greenhouse gas emissions. Clim.360
Change 109, 33–57 (2011).361
14. van Vuuren, D. P. et al. RCP2.6: exploring the possibility to keep global mean temperature362
increase below 2 degrees C. Clim. Change 109, 95–116 (2011).363
15. Bouwman, A. F., Beusen, A. H. W. & Billen, G. Human alteration of the global nitrogen364
and phosphorus soil balances for the period 1970-2050. GLOBAL BIOGEOCHEMICAL365
CYCLES 23 (2009).366
16. Strassmann, K. M., Joos, F. & Fischer, G. Simulating effects of land use changes on367
carbon fluxes: past contributions to atmospheric CO2 increases and future commitments368
due to losses of terrestrial sink capacity. Tellus B 60, 583–603 (2008).369
17. Stocker, B. D., Strassmann, K. & Joos, F. Sensitivity of Holocene atmospheric CO(2) and370
the modern carbon budget to early human land use: analyses with a process-based model.371
Biogeosciences 8, 69–88 (2011).372
18. van Aardenne, J., Dentener, F., Olivier, J., Goldewijk, C. & Lelieveld, J. A 1 degrees x 1373
degrees resolution data set of historical anthropogenic trace gas emissions for the period374
1890-1990. Global Biogeochem. Cycles 15, 909–928 (2001).375
19. Davidson, E. A. The contribution of manure and fertilizer nitrogen to atmospheric nitrous376
oxide since 1860. Nature Geoscience 2, 659–662 (2009).377
20. Hirsch, A. et al. Inverse modeling estimates of the global nitrous oxide surface flux from378
1998-2001. Global Biogeochem. Cycles 20 (2006).379
21. Denman, K. L. et al. Chapter 7: Couplings between changes in the climate system and380
biogeochemistry. In Solomon, S. et al. (eds.) Climate Change 2007: The Physical Science381
© 2013 Macmillan Publishers Limited. All rights reserved.
37Basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergov-382
ernmental Panel on Climate Change (Cambridge Univ. Press, United Kingdom and New383
York, NY, USA, 2007).384
22. Rhee, T. S., Kettle, A. J. & Andreae, M. O. Methane and nitrous oxide emissions from the385
ocean: A reassessment using basin-wide observations in the Atlantic. Journal of Geohpys-386
ical Research - Atmospheres 114 (2009).387
23. Suntharalingam, P. et al. Quantifying the impact of anthropogenic nitrogen deposition on388
oceanic nitrous oxide. Geophys. Res. Lett. 39 (2012).389
24. RCP database. RCP database, version 2.0.5. Website (2009). Available from http:390
//www.iiasa.ac.at/web-apps/tnt/RcpDb/ accessed 27/10/2011.391
25. Roe, G. Feedbacks, Timescales, and Seeing Red. ANNUAL REVIEW OF EARTH AND392
PLANETARY SCIENCES 37, 93–115 (2009).393
26. Gregory, J. M., Jones, C. D., Cadule, P. & Friedlingstein, P. Quantifying Carbon Cycle394
Feedbacks. J. Climate 22, 5232–5250 (2009).395
27. Spahni, R., Joos, F., Stocker, B. D., Steinacher, M. & Yu, Z. C. Transient simulations of396
the carbon and nitrogen dynamics in northern peatlands: from the last glacial maximum397
to the 21st century. Climate of the Past Discussions 8, 5633–5685 (2012). URL http:398
//www.clim-past-discuss.net/8/5633/2012/.399
28. Knutti, R. & Hegerl, G. C. The equilibrium sensitivity of the Earth’s temperature to radia-400
tion changes. Nature Geoscience 1, 735–743 (2008).401
29. Friedlingstein, P. et al. Climate-carbon cycle feedback analysis: Results from the C(4)MIP402
model intercomparison. J. Climate 19, 3337–3353 (2006).403
30. Syakila, A. & Kroeze, C. The global nitrous oxide budget revisited. Greenhouse Gas404
Measurement and Management 1, 17–26 (2011).405
31. Xu-ri, Prentice, I. C., Spahni, R. & Niu, H. S. Modelling terrestrial nitrous oxide emissions406
and implications for climate feedback. New Phytologist 2, 472–88 (2012).407
© 2013 Macmillan Publishers Limited. All rights reserved.
3832. Crutzen, P. J., Mosier, A. R., Smith, K. A. & Winiwarter, W. N2o release from agro-biofuel408
production negates global warming reduction by replacing fossil fuels. Atmospheric Chem-409
istry and Physics 8, 389–395 (2008).410
33. Spahni, R. et al. Constraining global methane emissions and uptake by ecosystems. Bio-411
geosciences 8, 1643–1665 (2011).412
34. Zurcher, S., Spahni, R., Joos, F., Steinacher, M. & Fischer, H. Impact of an 8.2-kyr-like413
event on methane emissions in northern peatlands. Biogeosciences Discussions 9, 13243–414
13286 (2012). URL http://www.biogeosciences-discuss.net/9/13243/415
2012/.416
35. Shindell, D., Walter, B. & Faluvegi, G. Impacts of climate change on methane emissions417
from wetlands. Geophys. Res. Lett. 31 (2004).418
36. Melton, J. R. et al. Present state of global wetland extent and wetland methane modelling:419
conclusions from a model intercomparison project (wetchimp). Biogeosciences Discus-420
sions 9, 11577–11654 (2012). URL http://www.biogeosciences-discuss.421
net/9/11577/2012/.422
37. CMIP5. CMIP5 Coupled Model Intercomparison Project. Website (2009). Available from423
http://cmip-pcmdi.llnl.gov/index.html accessed 27/10/2011.424
© 2013 Macmillan Publishers Limited. All rights reserved.