GMDGeoscientific Model DevelopmentGMDGeosci. Model Dev.1991-9603Copernicus PublicationsGöttingen, Germany10.5194/gmd-12-179-2019Representation of fire, land-use change and vegetation dynamics in the Joint
UK Land Environment Simulator vn4.9 (JULES)Representation of fire, land-use change and vegetation dynamics in JULESBurtonChantellechantelle.burton@metoffice.gov.ukhttps://orcid.org/0000-0003-0201-5727BettsRichardCardosoManoelhttps://orcid.org/0000-0003-2447-6882FeldpauschTed R.https://orcid.org/0000-0002-6631-7962HarperAnnahttps://orcid.org/0000-0001-7294-6039JonesChris D.https://orcid.org/0000-0002-7141-9285KelleyDouglas I.https://orcid.org/0000-0003-1413-4969RobertsonEddyWiltshireAndyMet Office Hadley Centre, Exeter, EX1 3PB, UKCollege of Life and Environmental Science, University of Exeter, Exeter, EX4 4SB, UKBrazilian Institute for Space Research (INPE), Earth System Science Center (CCST), São José dos Campos, BrazilCentre for Ecology and Hydrology, Wallingford, OX10 8BB, UKChantelle Burton (chantelle.burton@metoffice.gov.uk)9January201912117919315June201819July20187December201810December2018This work is licensed under the Creative Commons Attribution 4.0 International License. To view a copy of this licence, visit https://creativecommons.org/licenses/by/4.0/This article is available from https://gmd.copernicus.org/articles/12/179/2019/gmd-12-179-2019.htmlThe full text article is available as a PDF file from https://gmd.copernicus.org/articles/12/179/2019/gmd-12-179-2019.pdf
Disturbance of vegetation is a critical component of land cover,
but is generally poorly constrained in land surface and carbon cycle models.
In particular, land-use change and fire can be treated as large-scale
disturbances without full representation of their underlying complexities and
interactions. Here we describe developments to the land surface model JULES
(Joint UK Land Environment Simulator) to represent land-use change and fire
as distinct processes which interact with simulated vegetation dynamics. We
couple the fire model INFERNO (INteractive Fire and Emission algoRithm for
Natural envirOnments) to dynamic vegetation within JULES and use the HYDE
(History Database of the Global Environment) land cover dataset to analyse
the impact of land-use change on the simulation of present day vegetation. We
evaluate the inclusion of land use and fire disturbance against standard
benchmarks. Using the Manhattan metric, results show improved simulation of
vegetation cover across all observed datasets. Overall, disturbance improves
the simulation of vegetation cover by 35 % compared to vegetation
continuous field (VCF) observations from MODIS and 13 % compared to the
Climate Change Initiative (CCI) from the ESA. Biases in grass extent are reduced
from -66 % to 13 %. Total woody cover improves by 55 % compared
to VCF and 20 % compared to CCI from a reduction in forest extent in the
tropics, although simulated tree cover is now too sparse in some areas.
Explicitly modelling fire and land use generally decreases tree and shrub
cover and increases grasses. The results show that the disturbances provide
important contributions to the realistic modelling of vegetation on a global
scale, although in some areas fire and land use together result in too much
disturbance. This work provides a substantial contribution towards
representing the full complexity and interactions between land-use change and
fire that could be used in Earth system models.
JULES (Joint UK Land Environment Simulator) is a land surface model (LSM)
which simulates surface fluxes of water, energy and carbon, along with the
state of terrestrial hydrology, vegetation and carbon stores (Clark et
al., 2011; Best et al., 2011). It forms the land surface component in the
Met Office Unified Model for Numerical Weather Prediction, as well as in the
latest climate and Earth system models of the Hadley Centre family, including
HadGEM3 (Senior et al., 2016) and UKESM1, and can also be used as a
stand-alone LSM to contribute to international scientific studies such
as the Global Carbon Project and TRENDY (Trends in net land atmosphere
carbon exchange model intercomparison project). As documented in Cox (2001)
and Clark et al. (2011), vegetation cover was previously simulated as a
function only of competition between plant species and a large-scale,
spatially constant disturbance term. Here we document updates to the
calculation of vegetation cover, including spatially and temporally varying
changes in land use, and introduce a new disturbance term from fire based on
the fire model INFERNO (Mangeon et al., 2016) as separate from the
large-scale disturbance factor for the first time in JULES. We use these
processes together with dynamic vegetation to address the impact on global
vegetation cover.
JULES can be used in a number of different configurations depending on the
focus of research, and parameters can be switched on or off by the user
accordingly. For example, JULES can be used for studying river routing and
run-off, snow cover and permafrost, or crop modelling. In this context, it
is useful for the community to develop standard configurations that can be
used widely and are thus easily comparable. In this study we use a standard
JULES configuration with dynamic vegetation and focus on the impact of
disturbance from fire and land use on the simulation of vegetation cover.
Land-use change and fire are two of the most important processes which affect
vegetation cover. These disturbances affect vegetation dynamics (e.g. Lasslop
et al., 2016), atmospheric chemistry (Crutzen et al., 1979), the hydrological
cycle (Shakesby and Doerr, 2006) and the carbon cycle (Prentice et al.,
2011), as well as surface albedo (López-Saldaña et al., 2015) and
feedbacks on radiative forcing. Each year around 4 % of vegetation is
burnt (Giglio et al., 2013), releasing approximately 2 PgC, which equates to
around a quarter of emissions from fossil fuel combustion (Hantson et al.,
2016; van der Werf et al., 2017). Land-use and land cover change (LULCC) can
include clearance through fire, as well as other forms of deforestation,
conversion of natural vegetation to agricultural land and abandonment of
agricultural land with subsequent forest regrowth. At least 50 % of the
ice-free land surface has been affected by land-use activities over the last
300 years; 25 % of global forest area has been lost, and agriculture now
accounts for around 30 % of the land surface (Hurtt et al., 2011). LULCC
can result in changes to biogeochemical and biophysical properties of the
Earth system, including changes to surface fluxes of radiation, aerodynamic
roughness, heat and moisture, evaporation patterns, soil moisture, and latent
heat (Betts, 2005). LULCC often represents deliberate conversion from one
land cover type to another, such as forests to cropland, and this can be
long-lasting until the area is subsequently abandoned based on various
socio-economic conditions and decision-making processes (Turner et al.,
1995). Fires may be used in a similar way for land conversion or otherwise
may be unintentional (natural or escaped fire), and thus recovery may be more
temporally variable than with LULCC.
LULCC is known to be one of the most important influencing factors in the
decline of forests in several ways: directly through deforestation and canopy
thinning (cutting as well as use of fire for clearance) and indirectly
through fire leakage, which can extend forest losses into much larger areas
than planned. Fragmentation is also an important contributing factor, causing
increased tree mortality and carbon losses near the forest edges (Laurance et
al., 2000) and increased risk of fire spread into the forest (Soares et al.,
2012; Coe et al., 2013). This can be the result of
land clearance for agriculture and for urban expansion. For example, there is
a clear correlation between distance to roads and increased fire risk in
Amazonia (Cardoso et al., 2003). Even when deforestation itself declines,
fire incidence can remain high due to increased agricultural frontiers where
accidental fires burn out of control (Aragão and Shimabukuro, 2010;
Cano-Crespo et al., 2015) exacerbated by drought conditions (Aragão et
al., 2018). Small-scale forest degradation is sometimes included in the
definition of LULCC and can be an important contributor to carbon and biomass
loss; however, more frequently these contributions are below the level of
detection and are often not accounted for in estimates of LULCC (Watson et
al., 2000; Arneth et al., 2017). Similarly, small fires are difficult to
detect by conventional satellite methods (Randerson et al., 2012), leading to
potential underestimations in LULCC and emission reporting.
The interaction between fire and managed agricultural land is complex. Small-scale
croplands are often burnt to clear land before planting or harvesting
and can also be burnt after harvest to dispose of waste; pasture lands
may be burnt to fertilise the soils between crops (Rabin et al., 2018).
Agricultural land may therefore be an important contributing factor in fire
emissions and fire ignition. Conversely, larger agricultural lands may
provide a fire break, whereby more active fire management takes place to
prevent fires from spreading into crop areas unintentionally, and it has been
shown that burnt area reduces as cropland area increases (Bistinas et al.,
2014). Andela et al. (2017) have shown that fire occurrence has been
decreasing
in many regions because of agricultural expansion and intensification, making
fuel less readily available and decreasing ignitions.
While human ignitions are the main causes of fires in tropical (Cochrane,
2003) and Mediterranean (Mooney, 1977) regions, natural fires from lightning and volcanic activity
are also important for shaping vegetation cover in temperate (Ogden et al.,
1998) and boreal regions (Johnson, 1992; Veraverbeke et al., 2017). In
addition, climate-induced land cover change has been shown to be as important
in the long term as anthropogenic LULCC (Davies-Barnard et al., 2015) and
can continue to fluctuate for decades before a committed state is realised
(Pugh et al., 2018), making it particularly important to incorporate dynamic
vegetation processes in modelling (Seo and Kim, 2018). While previous
modelling studies have considered the impact of each of these processes (e.g.
Sitch et al., 2015; Betts et al., 2015; Seo and Kim, 2018), considering fire,
LULCC and dynamic vegetation together is still a relatively recent
development.
Future fire activity will depend on a combination of both anthropogenic and
climatic factors. Forest susceptibility to fire is projected to change little
for low-emissions scenarios, but substantially for high-emissions scenarios
(Settele et al., 2014; Burton et al., 2018). Because the frequency of fires
increases with temperature, the IPCC AR5 report concluded that the incidence
of fires is expected to rise over the 21st century (Flato et al., 2013),
although there is low agreement in the models on a regional scale due to the
complexity of interactions and feedbacks and lack of proper representation in
models (Settele et al., 2014). However, while the meteorological conditions
may become more conducive to fire risk in the future, the effects of future
LULCC will also have a direct impact on how fire risk will change. LULCC can
have important impacts on regional climate and has been shown to reduce
evapotranspiration (Cochrane and Laurance, 2008), decrease precipitation and
induce drought (Bagley et al., 2014), which can in turn initiate abrupt
increases in fire-induced tree mortality (Brando et al., 2014; Castello and
Macedo, 2016). The interaction of LULCC, climate change and fire is complex
(Coe et al., 2013) and in order to understand the multiple feedbacks
comprehensively, it is necessary to consider all of these elements together
(Aragão et al., 2008). To do this we need to be able to represent these
processes explicitly within our models.
Currently, the representation of disturbance, in particular fire, drought and
tree mortality, in models is poorly constrained, as identified in the most
recent IPCC report (Ciais et al., 2013; Flato et al., 2013). The purpose of
this paper is to document the developments to JULES to include the explicit
representation of fire and LULCC and their coupling to vegetation dynamics
and to evaluate the impact of these developments on the simulation of
vegetation within the model, with the aim of ultimately being able to
represent these processes within a fully coupled Earth system model. We
begin by describing how dynamic vegetation is simulated in JULES as
documented in Cox (2001) and Clark et al. (2011) before describing the new
processes of fire and land use. We then outline the methods used in this
study for simulating vegetation cover in a number of experiments and
describe the benchmarking approach used to quantify the change. We present
results showing the impact of fire and LULCC on vegetation cover, which
generally decreases woody vegetation cover and increases grass cover,
contributing to an improved simulation of vegetation compared to
observations.
Model description and developments
Within JULES a dynamic global vegetation model (DGVM) called TRIFFID (Top-down Representation of Interactive
Foliage and Flora Including Dynamics) is used to represent the carbon cycle
and the distribution of different plant functional types (PFTs) (Clark et
al., 2011; Cox et al., 2000; Cox, 2001). Here we focus on the simulation of
PFT distribution in a global model run. The area of each grid box covered by
PFT i: vi in the original model is determined by species competition
and large-scale disturbance.
dvidt=λΠv∗Cvi1-∑jcijvj-γvv∗
Equation (1) is used to calculate the evolution of vi. The rate of
increase of vi depends on the carbon available for increasing PFT area
(λΠv∗) and the carbon cost of increasing area given
by the carbon density (Cvi). Two terms balance the constant expansion
of PFTs: a competition term (within the curly brackets) represents the loss
of PFT area due to competition for limited space, and a disturbance term
(γvv∗) represents vegetation loss due to all
mortality processes not related to competition. λ is the fraction
of net primary production (NPP) per PFT area, Π, used for increasing PFT area. v∗ is a
maximum of PFT area, vi, and a minimum of 0.01 grid box fractional area
imposed to ensure PFTs do not get permanently removed from a given grid box.
cij determines which PFT, i or j, is dominant and will out compete
the other. cij is zero for dominant PFTs, meaning the whole grid box
is available for PFT i to expand into; for non-dominant PFTs, cij is
1 and expansion is scaled by the fraction of the grid box in which PFT i is
dominant. The configuration used here has five PFTs; the two tree PFTs
outcompete the shrub PFT, the shrub PFT outcompetes the two grass PFTs, and
the taller of the two tree–grass PFTs outcompetes the shorter tree–grass
PFT. γv is a PFT-dependent disturbance rate. The vegetation is
updated according to these factors on a 10-day time step.
The disturbance rate (γv) and spreading parameter
(λ) implicitly including fire disturbance (top rows) and excluding
fire disturbance (bottom rows).
PFTBroadleaf treeNeedle-leaf treeShrubC3 grassC4 grassγv implicit fire0.0090.00360.050.100.10λ implicit fire3.03.01.01.01.0γv using INFERNO0.00450.00180.150.100.10λ using INFERNO1.01.01.01.01.0
Here we include the effects of land use on vegetation distribution by
modifying the competition term of Eq. (1). Similar to competition, land use
is also represented by a limitation to the space available for a PFT to
expand into. A fraction of each grid box is prescribed as the “disturbed
fraction”, which represents the area covered by agriculture, with no
distinction between cropland and pasture being made. Adding in land use
to Eq. (1), we have
dvidt=λΠv∗Cvi1-αai-∑jcijvj-γvv∗,
where α is the disturbed fraction and ai is 1 for non-woody
PFTs and 0 for woody PFTs. The three woody PFTs (broadleaf trees,
needle-leaf trees and shrubs) are prevented from growing in the disturbed
fraction, while the two grass PFTs (C3 grass and C4 grass) can grow anywhere
in the grid box. Grass PFTs growing in the disturbed fraction are interpreted
as agricultural grasses, although they are physiologically identical to
“natural” grasses. α can increase or decrease over time. As
α increases, first “natural” grasses are relabelled as
“agricultural” grasses, then an area of woody PFTs is replaced by bare
soil, which can be replaced by the non-woody PFTs over time if they are
viable. As α decreases, an area of “agricultural” grasses is
relabelled as “natural” and becomes available for woody PFTs to expand
into.
The effect of fire on vegetation distribution is included by modifying the
disturbance rate, γv. Previously, disturbance due to fire was
implicitly included in γv, along with mortality due to
pests, windthrow and many other processes. With this new development, fire
disturbance, βi, is now included as a PFT-dependent burnt area which
can vary in space and time. βi is calculated within JULES by the
INFERNO (INteractive Fire and Emission algoRithm for Natural envirOnments)
fire model (Mangeon et al., 2016). Now that fire is explicitly represented,
γv must be reduced accordingly, and hence the representation of
fire does not necessarily increase mortality, but makes it spatially and
temporally variable. Table 1 shows the values of γv; in the
top row values implicitly include fire disturbance before the coupling, and
in the bottom row fire is treated separately using INFERNO. Equation (3)
includes fire along with land use.
dvidt=λΠv∗Cvi1-αai-∑jcijvj-γv+βiv∗
The calculation of burnt area depends on fuel availability as documented in
Mangeon et al. (2016) and which now includes the additional feedback of
reduction in fuel from fire (Eq. 3). Also included in fuel availability
is soil carbon density, providing additional mechanisms by which fire and
land use can feed back onto vegetation distribution. The coupling of fire and
the carbon cycle includes a direct impact of fire on soil; some soil
carbon is burnt, resulting in a flux of carbon from the soil to the
atmosphere. The burnt soil carbon flux is diagnosed in INFERNO and we now
allow the flux to effect the evolution of carbon in the soil pools, Ck.
The carbon cycle in JULES does not explicitly represent a litter carbon
store; however, the model includes four soil carbon pools and we use two of
these pools as proxies for flammable litter. The decomposable plant material
soil carbon pool, Cdpm, and the resistant plant material soil carbon
pool, Crpm, both receive the litter carbon flux from vegetation and
have relatively rapid turnover rates, making them reasonable proxies for
the litter carbon store. The calculation of the burnt soil flux is similar
to the INFERNO diagnosis of the burnt vegetation flux (Eq. 8 of Mangeon et
al., 2016).
fs=μmin,k+μmax,k-μmin,k1-θCk∑iβivi
The efficiency of soil burning is inversely proportional to the saturated
soil moisture fraction in the top soil level (0–10 cm), θ, with the
values of the completeness of combustion parameters, μ, for each soil
pool, k, being based on the original values from INFERNO and listed in
Table 2. The burnt soil flux is proportional to the total available fuel,
Ck, and the total burnt area summed over all PFTs.
Fire and land use both affect the soil carbon store by altering the
vegetation-to-soil litter flux. Without fire or land use, the litter flux
comprises a local litter fall rate, Λl, representing
the turnover of leaves, roots and stems, litter due to disturbances, and
litter due to competition. The total litter fall is defined by Clark et al. (2011) as (their Eq. 63)
Λc=∑iviΛli+γviCvi+Πi∑jcijvj.
Including our new disturbance terms produces
ΛCvLoss=∑iviΛli+γvi+βiCvi+Πi∑jαai+cijvj.
The new term, ΛCvLoss, still represents a loss of vegetation
carbon, but now not all of this flux enters the soil carbon pools, and instead
some of the vegetation carbon due to fire is lost to the atmosphere and
some of the loss due to land-use change enters wood-product carbon pools.
All litter fluxes that do enter soil carbon pools are split between
Cdpm and Crpm according to PFT-specific parameters
as described by Clark et al. (2011). To calculate the losses due to the new
processes, the vegetation distribution (Eq. 3) and vegetation loss (Eq. 6)
are calculated with and without the new process, and the difference between
the two values of ΛCvLoss is attributed to the new process.
VariableSymbolUnitSource of variableCombustion completenessμParameterCompetition termcijTRIFFIDCrop indicatoraiParameterDisturbed fractionαFraction of land surfaceInput mapFire disturbanceβiyr-1INFERNOFraction of NPP allocated to PFT area expansionλParameterFractional coveragevFraction of land surfaceTRIFFIDLarge-scale disturbanceγvyr-1ParameterLitter fall rate without fire or land-use changeΛckg C m-2 yr-1TRIFFIDLocal litter fall rateΛlkg C m-2 yr-1TRIFFIDNPP per unit of vegetated areaΠkg C m-2 yr-1JULESPFT indicesi, jSoil carbon in soil pool kCkkg C m-2TRIFFIDSoil fluxfskg C m-2 yr-1INFERNOVegetation carbon densityCvkg C,m-2TRIFFIDVegetation carbon loss due to fireΛFirekg C m-2 yr-1TRIFFIDVegetation carbon loss due to land-use changeΛLUCkg C m-2 yr-1TRIFFIDVegetation carbon loss due to litter, fire and land-use changeΛCvLosskg C m-2 yr-1TRIFFID
The litter due to land-use change, ΛLUC, is calculated by
repeating Eqs. (3) and (6) with the disturbed fraction from the previous time
step, α-1; note that both calculations include some disturbed
fraction and it is the litter due to land-use change that is being
calculated, not the effect of existing land use:
ΛLUC=ΛCvLoss-∑ivLUC,i(Λli+γvi+βiCvi+Πi∑jα-α-1ai+cijvLUC,j),
where vLUC is the PFT area calculated by Eq. (3) with α=α-1. ΛLUC is distributed between the soil
carbon pools and the wood-product pools; the portion that is below-ground
carbon, given by root carbon /Cv, is added to the soil carbon pools
and the remaining above-ground portion is added to the wood-product pools
(Jones et al., 2011).
Carbon loss due to fire, ΛFire, is calculated by repeating Eqs. (3) and (6) with no burnt area (β=0):
ΛFire=ΛCvLoss-∑ivNoFire,iΛli+γviCvi+Πi∑jαai+cijvNoFire,j,
where vNoFire is the PFT area calculated using Eq. (3) with β=0.13, meaning that 13 % of the vegetation carbon killed by fire is
emitted and the remainder enters the soil carbon pools (Li et al.,
2012). All terms expressed above are summarised in Table 3.
Experimental set-up and model evaluation
Here we run JULES vn4.9 from 1860 to present day with re-gridded CRU-NCEP7
forcing data for climate and CO2 and land-use ancillaries from HYDE 3.2
(History Database of the Global Environment) (Klein Goldewijk et al., 2011)
updated to include 2013–2015 as part of the global carbon budget, for which data
were extrapolated based on agricultural trends of the previous 5 years (Le
Quéré et al., 2016). The harmonised HYDE dataset estimates
fractional agricultural land-use patterns and underlying transitions in
land use annually for 1500–2100 and is spatially gridded at half-degree
resolution. It does not include impacts of degradation, climate variability,
forest management, fire management or pollution on land cover (Hurtt et al.,
2006), and it does not specify the nature of the previous land type, whether
forested or not (Le Quéré et al., 2016). This was then re-gridded
for use in JULES at N96 resolution (1.25∘ latitude × 1.875∘ longitude) and implemented from 1860–2015 as annual
land-use change as outlined above. Because the process of land-use change
excludes woody PFT from agricultural areas, it is expected that there will
be a reduction of tree growth and increase in grasses when this term is
included.
For the fire experiments, the model was spun up for 1000 years with fire on
using pre-industrial land use and CO2 at 1860 prescribed as a climatology.
INFERNO was run here with constant natural and anthropogenic ignitions and
with
interactive fire–vegetation on.
The model was tuned with fire towards a PFT distribution from the European
Space Agency Climate Change Initiative (ESA CCI, 2010) observations using
maximum spreading (λ) as LAI_min = 1.0 and the
large-scale disturbance term (γv) modified as per Table 1.
Altering LAI_min is a way of increasing the rate of spread of
vegetation to account for a known deficiency in the model associated with
slow regrowth. The large-scale disturbance of trees has been halved, and
the disturbance of shrub increased by a factor of 3 to be within the error
bars of ESA observations.
JULES was configured to the TRENDY set-up (Sitch et al., 2015) using two
experiments: S2 =CO2 and climate forcing (with land use constant at 1860,
referred to as “no LULCC, no fire”); and S3 =CO2, climate forcing and
land-use change using the standard large-scale constant disturbance rate
for the purposes of comparison (referred to as “LULCC only”). These two
experiment configurations were then repeated including the new explicit
representation of fire for SF2 (referred to as “fire only”) and SF3
(referred to as “LULCC and fire”).
For benchmarking the performance of our model configurations, we use the
protocol used by FireMIP (Rabin et al., 2017) based on the benchmarking
system outlined in Kelley et al. (2013). Annual average burnt area was
assessed using the normalised mean error (NME) metric, which sums the
difference between the model (mod) and observations (obs) over all cells
(i) weighted by cell area (Ai) and normalised by the average distance
from the mean of observations, obs‾.
NME=∑Ai⋅modi-obsi∑Ai⋅obsi-obs‾
NME comparisons are conducted in three steps. Step 1 compares simulated and
observed annual average burnt area. For step 2, modi and
obsi become the difference between modelled or observed and their
respective area-weighted means, i.e. xi→xi-x‾,
thereby removing systematic bias to describe the performance of the model
about the mean. Step 3 additionally removes the mean deviation, i.e.
xi→xi/xi‾, and describes the
model ability to reproduce the spatial pattern in burnt area. Comparisons
are made against fire CCI (Alonso-Canas and Chuvieco, 2015), MCD45
(Archibald et al., 2013), GFED4 (Giglio et al., 2013) and GFED4s (van der Werf
et al., 2017).
Simulated vegetation fractions are compared against vegetation continuous
fields (VCFs) from MODIS (2002–2012), as recommended for FireMIP analysis
(Rabin et al., 2017), and ESA CCI reference observations using the Manhattan
metric (MM):
MM=∑ijAi⋅modij-obsij/∑iAi,
where j is vegetation type. Using the MM, we assess model performance
against different vegetation combinations (see Table S3 in the Supplement for a full list of
comparisons).
Benchmarking results for each experiment by vegetation type using
VCF and CCI reference observations. Vegetation cover: woody, grass and bare
soil cover for VCF and tree, shrub, grass and bare soil cover for CCI. Woody
cover: trees and shrubs vs. other cover. Trees: BL and NL vs. other cover.
Grass cover: grass vs. non-grass cover. Lower results for JULES indicate
closer to observations. Colours indicate how many null models the
configuration exceeds: blue is for all; green is for all but one; yellow
indicates only one exceeded; red indicates none exceeded. Bold font indicates
both fire and land use together.
All benchmark datasets were resampled from their native resolutions to N96
before comparison. Scores for all metrics are directly comparable across
models, e.g. a score of 0.6 is twice as close to observations as 1.2, which
we describe as 100 % improvement from the control as per Kelley et al. (2014).
Three null models are used for further interpretation (Table S4).
The median and mean null model scores compare the median or mean of all
observations with the observation data. Randomly resampled null models
compare resampled observations (without replacement) against observations
using 1000 bootstraps to describe the distribution of the null model.
Individual model quality can be described in terms of the number of null models
exceeded (Table 4).
Present day (2010–2015) vegetation fractions for the TRENDY S2 (no
LULCC, no fire) and S3 experiment (LULCC only) by PFT, without fire, compared
to observations. Left column (a) shows ESA CCI observations (2010),
second column (b) shows vegetation without LULCC (S2), third
column (c) shows vegetation with LULCC (S3), fourth
column (d) shows the change resulting from LULCC (difference between
column 2 and 3), and right column (e) shows bias of S3 compared to
observations (difference between column 1 and 3). BL: broadleaf,
NL: needle-leaf, C3: C3 grasses, C4: C4 grasses,
Sb: shrub, BS: bare soil.
Results
Here we present results showing the effect of LULCC and fire on the
simulation of vegetation in JULES. First, we present global vegetation by
PFT to assess the present day spatial distribution of vegetation as a result
of LULCC disturbance compared to observations. We then move on to fire
disturbance, first reviewing how the new fire disturbance term modelled by
the coupled INFERNO model compares to GFED observations of burnt area as
validation for the fire model. We then present global vegetation by PFT for
fire disturbance and show how this compares to observations. Finally, we show
the global distribution of vegetation in the context of observations
considering uncertainty bounds.
Without explicit fire or LULCC disturbance, the model produces too much
broadleaf vegetation compared to observations, especially over South America
and SE Asia (Fig. 1, second column). Both broadleaf and needle-leaf trees are
not simulated well in the high-latitude boreal regions in JULES and do not
extend far enough across this region, which is not improved by adding
disturbance. Overall the model performs poorly at simulating tree cover, as
indicated by an MM score of 0.78 for vegetation cover comparison and 0.64 for
wood cover (Table 4) when comparing against VCF (generally worse than our
null models – Table S4) and 0.72 and 0.45, respectively, compared to CCI. The
introduction of LULCC generally results in a reduction in broadleaf,
needle-leaf and shrub vegetation and an increase in C3 and C4
grasses (Fig. 1, fourth column), improving the simulation of vegetation cover
by 23 % compared to VCF and 17 % against CCI (Table 4). This is as
expected, with the purpose of this disturbance term being to represent crop
area with C3 and C4 grasses. With LULCC, the broadleaf fraction is
much improved over South America compared to observations, but is not
improved in the high-latitude regions. C3 grass is improved with LULCC,
but the fraction is still too low, whereas shrub fraction remains too high
(also shown in Fig. 5). The bare soil fraction is too high in the model, but
the inclusion of LULCC has little effect on this.
Now considering fire, compared to observations of burnt area from GFED 4.1s
(including small fires) INFERNO captures the spatial extent and level of fire
relatively well (Fig. 2 and Table S4), and it is clear that the integrity of
the model to accurately simulate global burnt area (as presented in Mangeon
et al., 2016) is preserved through the coupling of fire and vegetation, both
with and without land use. INFERNO accurately simulates the areas of high
fire occurrence found in GFED4.1s, especially over Africa, northern
Australia, South America and SE Asia, although the model also shows high fire
occurrence over India, which is not seen in the observations. This is likely
due to the current lack of representation of fire suppression in agricultural
and urbanised areas. An NME score of 0.79–0.95 (Table S4) outperforms all
but one null model and is better than published assessments of other global
fire–vegetation models using the same metrics (Lasslop et al., 2014; Kelley
et al., 2013; Kloster and Lasslop, 2017; Hantson et al., 2016). NME step 2
and step 3 scores also fall in a similar range (Table S4), demonstrating a
strong performance in overall fire magnitude, variance and spatial pattern.
Similarly to LULCC, fire disturbance also improves the representation of
vegetation cover, this time by 31 % compared to VFC and 11 % against CCI
(Table 4). The balance of tree to grass cover over South America, for
example,
shows particular improvement (Fig. 3, third column), although in other areas
fire creates too much disturbance and results in the tree fraction being too
sparse, notably across Africa (although still within the range of
uncertainty; see Fig. 5). C3 grass fractions are generally too low without
fire compared to observations, and this is improved with fire. C4 grasses
are well modelled both with and without fire (Table S4). The shrub
fraction is too high in the model compared to observations, but this is also
improved when fire is included (28 %; Table S4, also shown in Fig. 5). There is too much bare soil in the model without disturbance, and this
increases further with fire. The overall change as a result of fire is
generally a reduction in the larger PFTs (broadleaf and needle-leaf trees)
and an increase in grasses and bare soil (Fig. 3, fourth column). Broadleaf
trees show a loss in all regions, including the Cerrado region to the south
of the Amazon and across the arid regions in Africa, SE Asia and northern high
latitudes. The changes in shrub and C4 grasses are more variable and are
region dependent. The increase in grasses and bare soil reflects the burnt
area as modelled by INFERNO (Fig. 2b), indicating a shift away from woody
vegetation (broadleaf trees, needle-leaf trees and shrubs) towards faster-growing vegetation and bare ground as a result of fire.
Average 2010–2015 burnt area from GFED 4.1s
observations (a) and as modelled by JULES-INFERNO (b).
Present day (2010–2015) vegetation fractions for the TRENDY S2
experiment (no LULCC, no fire) and SF2 (fire only) by PFT compared to
observations. Left column (a) shows ESA CCI observations (2010),
second column (b) shows vegetation without fire or LULCC (S2), third
column (c) shows vegetation with fire only (SF2), fourth
column (d) shows the change resulting from fire (difference between
column 2 and 3), and right column (e) shows the bias of SF2 compared
to observations (difference between column 1 and 3). BL: broadleaf,
NL: needle-leaf, C3: C3 grasses, C4: C4 grasses,
Sb: shrub, BS: bare soil.
Present day (2010–2015) vegetation fraction PFT, as modelled by
JULES with LULCC and fire (SF3, central column), compared to the range of
uncertainty from ESA CCI observations (V1, 2010) (minimum fractions left
column, maximum fractions right column). (a–c) Broadleaf,
(d–f) needle-leaf, (g–i) C3 grass,
(j–l) C4 grass, (m–o) shrub, (p–r) bare
soil.
Present day (2010–2015) total vegetation cover (percentage)
globally (a) and by WWF biome (five out of eight shown here:
tropical forest, temperate forest, boreal forest, tropical savanna and
temperate grasses. Tundra, Mediterranean wood and desert not shown). Trees:
total broadleaf and needle-leaf trees, grasses: total C3 and C4
grasses. Panel (a) includes results prior to tuning, plus
uncertainty bars for the ESA observations of vegetation cover shown in blue.
As with all observational datasets, there are uncertainties associated with
retrieving observations of land cover and the classification of these into a
small number of plant functional types. The observations used here are from
ESA CCI, which have been processed into the five PFTs used by JULES so as to be
comparable with the model output (Hartley et al., 2017), introducing a range
of possible values for each vegetation type. The representation of
vegetation distribution is further complicated by seasonal variation,
whereby the peak growing season will have a higher fraction of vegetation than the low
season, and high-fire-risk areas will show burnt area as high bare soil in
peak fire season. These uncertainties give a range of potential vegetation
cover, and the developments to the representation of disturbance in JULES
described here have been tuned to give a reasonable distribution within this
range of uncertainty as far as possible (Figs. 4 and 5, top left panel).
The “best estimate” of vegetation cover from the ESA, known as the reference
case, is otherwise used for comparison, and VCF is used to provide additional
comparison in the benchmarking assessment.
Considering the distribution by vegetation type (trees, grasses, shrubs and
soil), in all cases adding disturbance to the model brings the global total
vegetation closer to reference observations, although bare soil increases in
the opposite trend (Fig. 5, top left panel). In the case of trees and
shrubs, fire plus LULCC creates too much disturbance (42 % and 47 % less
coverage than observations, respectively), but grasses increase (13 % more
coverage than observations) (Table S1). This is reflected in only slight
improvements in MM scores for vegetation cover and wood cover comparisons
against VCF and a slight degradation when compared against CCI (Table 4).
Trees are reduced by 43 % when both disturbances are included (S2 compared
to SF3), shrubs are reduced by 71 %, and grasses increase by 127 % (Table S1),
taking into account the updated terms for γv (Table 1). There is an increase of 20 %
in bare soil with disturbance included.
Overall, adding disturbance into JULES reduces the bias of shrubs from
72 % to -47 % and grasses from -66 % to 13 % compared to
observations (Table S1). All results show a statistically significant
difference with disturbance compared to no disturbance (Table S5).
However, there is more variation by biome. In all cases tree fraction is
simulated as too low with both fire and LULCC, although the extent of this
varies (Fig. 5). In some cases shrubs improve (in the temperate and boreal
forests), but in others the results show too much disturbance (tropics,
savanna and temperate grasses). Grasses are generally higher than
observations, except for the temperate grasses biome. Both disturbance terms
reduce the tree and shrub fractions and increase grasses and bare soil
fractions. In most biomes the bare soil fraction is too high compared to
observations, except in the tropics and boreal regions where the fraction is
well represented compared to observations.
Overall, the inclusion of these disturbance terms within JULES leads to a
shift towards grass cover and a reduction in woody PFTs. This is as expected
for land use, which replaces trees with grasses as a representation of
crops. The regrowth rates for trees is much slower than for grasses, which
spread fast and recover quickly (see Sect. 2); this may be an important
factor in the response to fire. With continuous disturbance which varies
spatially and temporally now included in the model, the vegetation seems
unable to recover trees in some areas, notably around the Cerrado and Congo
regions, instead encouraging the growth of grasses in their place (Fig. 4).
DiscussionImpact of fire and land-use changes
Fire and land use are important global disturbances, and the results
presented here have shown that when considered, they have a significant
impact on modelled vegetation as represented by JULES. In all cases,
including disturbance brings the vegetation fractions closer to the
observations compared to no disturbance, although in some cases there is a
tendency towards too much disturbance when both fire and LULCC are included,
and bare soil increases too much compared to observations (Fig. 5).
Disturbance generally improves the simulation of shrubs and grasses, but
tree fractions are often simulated as too sparse. LULCC mainly decreases
trees and shrubs and replaces them with C3 and C4 grasses (representing crop
and pasture). Fire creates a more mixed response, decreasing vegetation in
the boreal regions and high-fire-risk areas and showing an increase in
grasses. Both fire and LULCC reduce the larger vegetation types when added
to the model (Figs. 1 and 3). Without the inclusion of fire, this could
result in an overestimation in the amount of carbon released due solely to
LULCC, which may have a significant impact on carbon budgets.
Previous work has shown that fire may be an important contributor to the
existence of savannas (Cardoso et al., 2008; Bond et al., 2005; Staver et
al., 2011). The results shown here seem to support this conclusion, showing
that when fire is included in the model there is a shift towards open
savanna-like states in areas that climatologically could support trees
without the incidence of fire, including the Cerrado area of southern Brazil
and savanna areas in Africa. Here we have shown that a large savanna region
in South America is completely forested in the model without the addition of
fire or anthropogenic LULCC.
Uncertainty
Here we have used the ESA CCI land cover product as our observational data
for comparison with the model output. The CCI product has been translated
into the five PFTs that are used in JULES (Poulter et al., 2015), and through
the process of data collection and classification, a number of uncertainties
are introduced which result in a range of possible outcomes for land cover
distribution (Hartley et al., 2017). These uncertainties can include
variation in classifying the surface reflectance products into the 22 land
cover classes and aggregating these by dominant vegetation type into just five PFTs for JULES using a consultative cross-walking technique. This
classification also takes into account seasonal variation in normalised difference vegetation index
(NDVI; greenness), burnt area, cloud cover and snow occurrence that can all vary
throughout the year, giving a large range between the minimum and maximum
possible vegetation cover for any one PFT, as shown in Fig. 4 and the blue
bars in Fig. 5. For this reason we also use the MODIS VCF for benchmarking
comparison. The VCF product is a characterisation of the land surface into
just three components of ground cover using satellite data: tree cover,
non-tree vegetation cover and bare ground. The model performs well compared
to a simple classification of tree and non-tree vegetation cover, showing
that the spatial coverage of vegetation is simulated well when both disturbances
are added to JULES. The benchmarking results compared to CCI still show an
improvement compared to the control, but on a global scale this is better
when each disturbance is considered separately, suggesting further
parameterisation may be beneficial for each PFT. However, it is important to
consider regional improvements or degradation as well, which can be masked in
global scale analyses (Figs. 1, 3, 5). It also suggests that there may be some
overlap in the disturbances, which reflects the complicated nature of how
fire and LULCC are often used together for land clearance, and future
development would benefit from reducing burnt area in cropland areas
(Bistinas et al., 2014). The HYDE LULCC dataset in this study has been
developed from a combination of model, satellite and historical
reconstructions of agricultural and population data, and the biomass
quantities are noted to contain uncertainties due to a lack of direct
observations from the historical period (Hurtt et al., 2011). Some of what
has been attributed to LULCC may include fire clearance, which is a key
point for consideration for other DGVMs including fire and land use
together.
Limitations and future developments
When interactive fire was initially added to JULES, there was a tendency
towards complete dominance by shrubs and significant tree reduction (see
Fig. 5, top left panel). This was tuned to the observations by increasing
the large-scale disturbance term (γ) and increasing spreading
(λ) (Table 1) to account for the fact that fire was previously
included in the total mortality rate. Grasses spread and recover quickly
with TRIFFID, whereas larger PFTs take longer to re-establish. On this
timescale the tree cover is not able to recover fast enough with constant
disturbance from fire, and the results indicate that fire restricts tree
growth and encourages a shift towards the more responsive vegetation types.
There are a number of ways this could be addressed in future developments.
Grasses can be given a higher mortality rate to prevent overgrowth, but
this has been tested and results in too much bare soil because trees are
unable to recover. The fractions were low from the start of the run (1860)
as fire was included in the spin-up, and the vegetation does not recover
through the transient simulation due to continual disturbance, leading to
present day levels being low. This perhaps points to a need to develop
faster regrowth of trees within TRIFFID to cope with disturbance, for
example by representing age or mass classes within each PFT to enable a
range of successional stages to be represented. It is also worth noting that
the fire disturbance is high in some areas in the model compared to
observations (Fig. 2), which may lead to too much disturbance in these
regions, whereas in other areas the burnt area maximum is underestimated. In
addition, there remains significant underlying complexity around the
interaction of LULCC and fire as discussed in Sect. 2. For example,
agricultural land in some regions may be a cause of fire ignition, whereas
in other areas it may act as a fire break or generate anthropogenic fire
suppression, and future development would benefit from reducing burnt area
in cropland areas (Bistinas et al., 2014). One way forward for this could
be to identify the average field size based on surrounding vegetation and
mask fire in larger agricultural regions, but allow smaller fields to
include the probability of burning or include fragmentation effects such as
described in Pfeiffer et al. (2013). There will also be
additional complexity around the PFTs themselves, and some species will be
more fire resilient than other species; for example,
vegetation in high-fire-risk areas often develops thicker bark for protection from fire,
whereas other species may adapt to the fire and use it as a method of
reproduction and resprouting (Kelley et al., 2014; Pellegrini et al., 2017).
The representation of just five PFTs is a considerable simplification of the
real world. Finally, we have just considered two of the main disturbances
here. We have not considered windthrow, pests or diseases, for example, which for
now are still aggregated into the generic large-scale disturbance term in
JULES.
There are still a number of regions that require improvement in the
simulation of vegetation. In all of the JULES simulations there are too few
needle-leaf trees across the boreal regions compared to observations. With
fire, the trees across the extratropics and savanna regions, such as
the Congo region in Africa, are notably reduced. Further work could be to develop
these configurations into the nine PFTs set up by Harper et al. (2016). In
particular, recent work has shown that the distinction between evergreen and
deciduous needle-leaf trees has led to an improved representation of boreal
forests within JULES, which could improve these simulations (Harper et al.,
2018).
Conclusion
This work has described the first steps in developing the land surface model
JULES to represent fire and land use as separate disturbances. The results
have shown the significant contributions of these disturbances to changes in
vegetation on a global scale. Without disturbance JULES simulates too much
vegetation in most PFTs compared to observations, which is generally
improved with the addition of fire and LULCC, although there is still
regional variation. Disturbance generally has the effect of decreasing tree
cover (43 %) and shrubs (71 %) and increasing grasses (127 %). In
places the disturbance is too high with both fire and LULCC and leads to
vegetation being reduced too much. Slow regrowth rates in TRIFFID also mean
that with constant disturbance from fire, there is a shift towards faster-growing PFTs that can recover and spread quickly. Overall, representing
disturbance in JULES improves the simulation of total vegetation cover
compared to both VCF and CCI datasets by 35 % and 13 %, respectively,
with woody vegetation improving by 55 % and 20 %, respectively. The
simulation of shrubs and grasses is much improved, with the bias reduced
from 72 % to -47 % and from -66 % to 13 %, respectively. It is
expected that fire risk will increase in the future with climate change as a
result of hotter, drier conditions, but fire occurrence depends heavily on
the interaction with LULCC. The developments to the model that have been
outlined in this paper now give the capability to model future interactions
between fire and LULCC and the impact that this could have on future
vegetation density, spread and carbon storage. Overall we have presented
results for an improved representation of mechanistic processes of
disturbance in JULES using a non-optimised approach, with positive results
for vegetation cover. This is a significant first step in the representation
of highly complex factors surrounding anthropogenic and natural disturbances
in the model, and it lays the foundation for future developments into Earth
system models.
The JULES code used in these experiments is freely
available on the JULES trunk from version 4.8 (revision 6925) onwards. The
rose suite used for these experiments is u-ap845 at vn4.9 r9986 (located in
the repository at trac/roses-u/log/a/p/8/4/5 r69824). Both the suite and the
JULES code are available on the JULES FCM repository:
https://code.metoffice.gov.uk/trac/jules (registration required; last
access: 4 January 2019).
The supplement related to this article is available online at: https://doi.org/10.5194/gmd-12-179-2019-supplement.
CB updated the code in the JULES trunk to include fire mortality, with help
and advice from ER, AW, RB, CDJ and AH. CB drafted the text and made the
figures. ER and CB co-wrote Sect. 2 and the equations. DIK performed the
benchmarking. All authors (CB, RB, MC, TRF, AH, CDJ, DIK, ER, AW) have
contributed to the analysis methods and to the text.
The authors declare that they have no conflict of
interest.
Acknowledgements
This work and its contributors (Chantelle Burton, Richard Betts, Chris Jones, Eddy Robertson, Andy Wiltshire)
were supported by the
Newton Fund through the Met Office Climate Science for Service Partnership
Brazil (CSSP Brazil).
Manoel Cardoso acknowledges support from the Brazilian BNDES/Amazon Fund Project
MSA/BNDES/BIOMASSA-SUB 7.
The contribution by Douglas I. Kelley was supported by the UK Natural Environment Research
Council through The UK Earth System Modelling Project (UKESM, grant no. NE/N017951/1)
We would like to thank Nicolas Viovy and Philippe Ciais for making available
their CRU-NCEP forcing data and for their kind permission for its use in
these model runs.
Edited by: Christoph Müller
Reviewed by: four anonymous referees
ReferencesAlonso-Canas, I. and Chuvieco, E.: Global burned area mapping from
ENVISAT-MERIS and MODIS active fire data, Remote Sens. Environ., 163,
140–152, 10.1016/j.rse.2015.03.011, 2015.
Andela, N., Morton, D., Giglio, L., Chen, Y., van der Werf, G., Kasibhatla,
P., DeFries, R., Collatz, G., Hantson, S., Kloster, S., Bachelet, D.,
Forrest, M., Lasslop, G., Li, F., Mangeon, S., Melton, J., Yue, C., and
Randerson, J.: A human-driven decline in global burned area, Science, 356,
1356–1361, 2017.
Aragão, L. and Shimabukuro, Y.: The Incidence of Fire in Amazonian
Forests with Implications for REDD, Science, 328, 1275–1278, 2010.
Aragão, L., Malhi, Y., Barbier, N., Lima, A., Shimabukuro, Y., Anderson,
L., and Saatchi, S.: Interactions between rainfall, deforestation and fires
during recent years in the Brazilian Amazonia, Philos. T. Roy. Soc. B, 363,
1779–1785, 2008.Aragão, L., Anderson, L. O., Fonseca, M. G., Rosan, T. M., Vedovato, L.
B., Wagner, F. H., Silva, C. V. J., Silva, C. H. L., Arai, E., Aguiar, A. P.,
Barlow, J., Berenguer, E., Deeter, M. N., Domingues, L. G., Gatti, L., Gloor,
M., Malhi, Y., Marengo, J. A., Miller, J. B., Phillips O. L., and Saatchi,
S.: 21st Century drought-related fires counteract the decline of Amazon
deforestation carbon emissions, Nat. Commun., 9, 536,
10.1038/s41467-017-02771-y, 2018.Archibald, S., Lehmann, C. E. R., Gomez-Dans, J. L., and Bradstock, R. A.:
Defining pyromes and global syndromes of fire regimes, P. Natl. Acad. Sci.
USA, 110, 6442–6447, 10.1073/pnas.1211466110, 2013.Arneth, A., Sitch, S., Pongratz, J., Stocker, B. D., Ciais, P., Poulter, B.,
Bayer, A. D., Bondeau, A., Calle, L., Chini, L. P., Gasser, T., Fader, M.,
Friedlingstein, P., Kato, E., Li, W., Lindeskog, M., Nabel, J. E. M. S.,
Pugh, T. A. M., Robertson, E., Viovy, N., Yue, C., and Zaehle, S.: Historical
carbon dioxide emissions caused by land-use changes are possibly larger than
assumed, Nat. Geosci.,10, 79–84, 10.1038/ngeo2882, 2017.
Bagley, J., Desai, A., Harding, K., Snyder, P., and Foley, J.: Drought and
Deforestation: Has Land Cover Change Influenced Recent Precipitation Extremes
in the Amazon?, J. Climate, 27, 345–361, 2014.Best, M. J., Pryor, M., Clark, D. B., Rooney, G. G., Essery, R. L. H.,
Ménard, C. B., Edwards, J. M., Hendry, M. A., Porson, A., Gedney, N.,
Mercado, L. M., Sitch, S., Blyth, E., Boucher, O., Cox, P. M., Grimmond, C.
S. B., and Harding, R. J.: The Joint UK Land Environment Simulator (JULES),
model description – Part 1: Energy and water fluxes, Geosci. Model Dev., 4,
677–699, 10.5194/gmd-4-677-2011, 2011.
Betts, R.: Integrated approaches to climate-crop modelling: needs and
challenges, Philos. T. Roy. Soc. B, 360, 2049–2065, 2005.Betts, R. A., Golding, N., Gonzalez, P., Gornall, J., Kahana, R., Kay, G.,
Mitchell, L., and Wiltshire, A.: Climate and land use change impacts on
global terrestrial ecosystems and river flows in the HadGEM2-ES Earth system
model using the representative concentration pathways, Biogeosciences, 12,
1317–1338, 10.5194/bg-12-1317-2015, 2015.Bistinas, I., Harrison, S. P., Prentice, I. C., and Pereira, J. M. C.: Causal
relationships versus emergent patterns in the global controls of fire
frequency, Biogeosciences, 11, 5087–5101,
10.5194/bg-11-5087-2014, 2014.Bond, W. J., Woodward, F. I., and Midgley, G. F.: The global distribution of
ecosystems in a world without fire, New Phytol., 165, 525–537,
10.1111/j.1469-8137.2004.01252.x, 2005.
Brando, P., Balch, J., Nepstad, D., Morton, D., Putz, F., Coe, M., Silverio,
D., Macedo, M., Davidson, E., Nobrega, C., Alencar, A., and Soares, B.:
Abrupt increases in Amazonian tree mortality due to drought-fire
interactions, P. Natl. Acad. Sci. USA, 111, 6347–6352, 2014.Burton, C., Betts, R. A., Jones, C. D., and Williams, K.: Will fire danger be
reduced by using Solar Radiation Management to limit global warming to
1.5 ∘C compared to 2.0 ∘C?, Geophys. Res. Lett., 45,
3644–3652, 2018.
Cano-Crespo, A., Oliveira, P., Boit, A., Cardoso, M., and Thonicke, K.:
Forest edge burning in the Brazilian Amazon promoted by escaping fires from
managed pastures, J. Geophys. Res.-Biogeo., 120, 2095–2107, 2015.
Cardoso, M., Hurtt, G., Moore, B., Nobrem C., and Prins, E.: Projecting
future fire activity in Amazonia, Glob. Change Biol., 9, 656–669, 2003.Cardoso, M., Nobre, C., Lapola, D., Oyama, M., and Sampaio, G.: Long-term
potential for fires in estimates of the occurrence of savannas in the
tropics, Global Ecol. Biogeogr., 17, 222–235,
10.1111/j.1466-8238.2007.00356.x, 2008.
Castello, L. and Macedo, M. N.: Large-scale degradation of Amazonian
freshwater ecosystems, Glob. Change Biol., 22, 990–1007, 2016.
Ciais, P., Sabine, C., Bala, G., Bopp, L., Brovkin, V., Canadell, J.,
Chhabra, A., DeFries, R., Galloway, J., Heimann, M., Jones, C., Le
Quéré, C., Myneni, R. B., Piao, S., and Thornton, P.: Carbon and
Other Biogeochemical Cycles, in: Climate Change 2013: The Physical Science
Basis. Contribution of Working Group I to the Fifth Assessment Report of the
Intergovernmental Panel on Climate Change, edited by: Stocker, T. F., Qin,
D., Plattner, G.-K., Tignor, M., Allen, S. K., Boschung, J., Nauels, A., Xia,
Y., Bex, V., and Midgley, P. M., Cambridge University Press, Cambridge, UK
and New York, NY, USA, 2013.Clark, D. B., Mercado, L. M., Sitch, S., Jones, C. D., Gedney, N., Best, M.
J., Pryor, M., Rooney, G. G., Essery, R. L. H., Blyth, E., Boucher, O.,
Harding, R. J., Huntingford, C., and Cox, P. M.: The Joint UK Land
Environment Simulator (JULES), model description – Part 2: Carbon fluxes and
vegetation dynamics, Geosci. Model Dev., 4, 701–722,
10.5194/gmd-4-701-2011, 2011.Cochrane, M. A.: Fire Science for Rainforests, Nature, 421, 913–919,
10.1038/nature01437, 2003.
Cochrane, M. A. and Laurance, W. F.: Synergisms among Fire, Land Use, and
Climate Change in the Amazon, Ambio, 37, 522–527, 2008.Coe, M., Marthews, T., Costa, M., Galbraith, D., Greenglass, N., Imbuzeiro,
H., Levine, N., Malhi, Y., Moorcroft, P., Muza, M., Powell, T., Saleska, S.,
Solorzano, L., and Wang, J.: Deforestation and climate feedbacks threaten the
ecological integrity of south-southeastern Amazonia, Philos. T. Roy. Soc. B,
368, 1619, 10.1098/rstb.2012.0155, 2013.
Cox, P. M.: Description of the “TRIFFID” Dynamic Global Vegetation Model,
Tech. Note 24, Hadley Centre, Met Office, Bracknell, UK, 16 pp. 2001.
Cox, P. M., Betts, R. A., Jones, C. D., Spall, S. A., and Totterdell, I. J.:
Acceleration of global warming due to carbon-cycle feedbacks in a coupled
climate model, Nature, 408, 184–187, 2000.Cruzten, P. J., Heidt, L. E., Krasnec, J. P., Pollock, W. H., and Seiler, W.
G.: Biomass burning as a source of atmospheric gases CO, H2,
N2O, NO, CH3, Cl and COS, Nature, 282, 253–256, 1979.Davies-Barnard, T., Valdes, P. J., Singarayer, J. S., Wiltshire, A. J., and
Jones, C. D.: Quantifying the relative importance of land cover change from
climate and land use in the representative concentration pathways, Global
Biogeochem. Cy., 29, 842–853, 10.1002/2014GB004949, 2015.ESA CCI: https://maps.elie.ucl.ac.be/CCI/viewer/ (last access:
4 January 2019), 2010.
Flato, G., Marotzke, J., Abiodun, B., Braconnot, P., Chou, S. C., Collins,
W., Cox, P., Driouech, F., Emori, S., Eyring, V., Forest, C., Gleckler, P.,
Guilyardi, E., Jakob, C., Kattsov, V., Reason, C., and Rummukainen, M.:
Evaluation of Climate Models, in: Climate Change 2013: The Physical Science
Basis. Contribution of Working Group I to the Fifth Assessment Report of the
Intergovernmental Panel on Climate Change, edited by: Stocker, T. F., Qin,
D., Plattner, G.-K., Tignor, M., Allen, S. K., Boschung, J., Nauels, A., Xia,
Y., Bex, V., and Midgley, P. M., Cambridge University Press, Cambridge, UK
and New York, NY, USA, 2013.
Giglio, L., Randerson, J. T., and Werf, G. R.: Analysis of daily, monthly,
and annual burned area using the fourth generation global fire emissions
database (GFED4), J. Geophys. Res.-Biogeo., 118, 317–328, 2013.Hantson, S., Arneth, A., Harrison, S. P., Kelley, D. I., Prentice, I. C.,
Rabin, S. S., Archibald, S., Mouillot, F., Arnold, S. R., Artaxo, P.,
Bachelet, D., Ciais, P., Forrest, M., Friedlingstein, P., Hickler, T.,
Kaplan, J. O., Kloster, S., Knorr, W., Lasslop, G., Li, F., Mangeon, S.,
Melton, J. R., Meyn, A., Sitch, S., Spessa, A., van der Werf, G. R.,
Voulgarakis, A., and Yue, C.: The status and challenge of global fire
modelling, Biogeosciences, 13, 3359–3375,
10.5194/bg-13-3359-2016, 2016.Harper, A. B., Cox, P. M., Friedlingstein, P., Wiltshire, A. J., Jones, C.
D., Sitch, S., Mercado, L. M., Groenendijk, M., Robertson, E., Kattge, J.,
Bönisch, G., Atkin, O. K., Bahn, M., Cornelissen, J., Niinemets, Ü.,
Onipchenko, V., Peñuelas, J., Poorter, L., Reich, P. B., Soudzilovskaia,
N. A., and Bodegom, P. V.: Improved representation of plant functional types
and physiology in the Joint UK Land Environment Simulator (JULES v4.2) using
plant trait information, Geosci. Model Dev., 9, 2415–2440,
10.5194/gmd-9-2415-2016, 2016.Harper, A. B., Wiltshire, A. J., Cox, P. M., Friedlingstein, P., Jones, C.
D., Mercado, L. M., Sitch, S., Williams, K., and Duran-Rojas, C.: Vegetation
distribution and terrestrial carbon cycle in a carbon cycle configuration of
JULES4.6 with new plant functional types, Geosci. Model Dev., 11, 2857–2873,
10.5194/gmd-11-2857-2018, 2018.Hartley, A., MacBean, N., Georgievski, G., and Bontempts, S.: Uncertainty in
plant functional type distributions and its impact on land surface models,
Remote Sens. Environ., 203, 71–89, 10.1016/j.rse.2017.07.037, 2017.
Hurtt, G. C., Frolking, S., Fearon, M. G., Moore, B., Shevliakova, E.,
Malyshev, S., Pacala, S. W., and Houghton, R. A.: The underpinnings of
land-use history: three centuries of global gridded land-use transitions,
wood harvest activity, and resulting secondary lands, Glob. Change Biol., 12,
1208–1229, 2006.
Hurtt, G. C., Chini, L., Frolking, S., Betts, R., Feddema, J., Fischer, G.,
Fisk, J., Hibbard, K., Houghton, R., Janetos, A., Jones, C., Kindermann, G.,
Kinoshita, T., Goldewijk, K., Riahi, K., Shevliakova, E., Smith, S.,
Stehfest, E., Thomson, A., Thornton, P., van Vuuren, D., and Wang, Y.:
Harmonization of land-use scenarios for the period 1500–2100: 600 years of
global gridded annual land-use transitions, wood harvest, and resulting
secondary lands, Climatic Change, 109, 117–161, 2011.
Johnson, E. A.: Fire and the vegetation dynamics: studies from the North
American boreal forest. Cambridge studies in ecology, Cambridge University
Press, Cambridge, UK, 1992.Jones, C. D., Hughes, J. K., Bellouin, N., Hardiman, S. C., Jones, G. S.,
Knight, J., Liddicoat, S., O'Connor, F. M., Andres, R. J., Bell, C., Boo,
K.-O., Bozzo, A., Butchart, N., Cadule, P., Corbin, K. D., Doutriaux-Boucher,
M., Friedlingstein, P., Gornall, J., Gray, L., Halloran, P. R., Hurtt, G.,
Ingram, W. J., Lamarque, J.-F., Law, R. M., Meinshausen, M., Osprey, S.,
Palin, E. J., Parsons Chini, L., Raddatz, T., Sanderson, M. G., Sellar, A.
A., Schurer, A., Valdes, P., Wood, N., Woodward, S., Yoshioka, M., and
Zerroukat, M.: The HadGEM2-ES implementation of CMIP5 centennial simulations,
Geosci. Model Dev., 4, 543–570, 10.5194/gmd-4-543-2011,
2011.Kelley, D. I., Prentice, I. C., Harrison, S. P., Wang, H., Simard, M.,
Fisher, J. B., and Willis, K. O.: A comprehensive benchmarking system for
evaluating global vegetation models, Biogeosciences, 10, 3313–3340,
10.5194/bg-10-3313-2013, 2013.Kelley, D. I., Harrison, S. P., and Prentice, I. C.: Improved simulation of
fire–vegetation interactions in the Land surface Processes and eXchanges
dynamic global vegetation model (LPX-Mv1), Geosci. Model Dev., 7, 2411–2433,
10.5194/gmd-7-2411-2014, 2014.Klein Goldewijk, K., Beusen, A., de Vos, M., and van Drecht, G.: The HYDE 3.1
spatially explicit database of human induced land use change over the past
12,000 years, Global Ecol. Biogeogr., 20, 73–86,
10.1111/j.1466-8238.2010.00587.x, 2011.
Kloster, S. and Lasslop, G.: Historical and future fire occurrence (1850 to
2100) simulated in CMIP5 Earth System Models, Global Planet. Change, 150,
58–69, 2017.
Lasslop, G., Kirsten T., and Kloster, S.: SPITFIRE within the MPI Earth
system model: Model development and evaluation, J. Adv. Model. Earth Sy.,
6.3, 740–755, 2014.
Lasslop, G., Brovkin. V., Reick, C. H., Bathiany, S., and Kloster, S.:
Multiple stable states of tree cover in a global land surface model due to a
fire-vegetation feedback, Geophys. Res. Lett., 43, 6324–6331, 2016.Laurance, W. F., Vasconcelos, H. L., and Lovejoy, T. E.: Forest loss and
fragmentation in the Amazon: implications for wildlife conservation, Oryx,
34, 39–45, 10.1046/j.1365-3008.2000.00094.x, 2000.Le Quéré, C., Andrew, R. M., Canadell, J. G., Sitch, S., Korsbakken,
J. I., Peters, G. P., Manning, A. C., Boden, T. A., Tans, P. P., Houghton, R.
A., Keeling, R. F., Alin, S., Andrews, O. D., Anthoni, P., Barbero, L., Bopp,
L., Chevallier, F., Chini, L. P., Ciais, P., Currie, K., Delire, C., Doney,
S. C., Friedlingstein, P., Gkritzalis, T., Harris, I., Hauck, J., Haverd, V.,
Hoppema, M., Klein Goldewijk, K., Jain, A. K., Kato, E., Körtzinger, A.,
Landschützer, P., Lefèvre, N., Lenton, A., Lienert, S., Lombardozzi,
D., Melton, J. R., Metzl, N., Millero, F., Monteiro, P. M. S., Munro, D. R.,
Nabel, J. E. M. S., Nakaoka, S.-I., O'Brien, K., Olsen, A., Omar, A. M., Ono,
T., Pierrot, D., Poulter, B., Rödenbeck, C., Salisbury, J., Schuster, U.,
Schwinger, J., Séférian, R., Skjelvan, I., Stocker, B. D., Sutton, A.
J., Takahashi, T., Tian, H., Tilbrook, B., van der Laan-Luijkx, I. T., van
der Werf, G. R., Viovy, N., Walker, A. P., Wiltshire, A. J., and Zaehle, S.:
Global Carbon Budget 2016, Earth Syst. Sci. Data, 8, 605-649,
10.5194/essd-8-605-2016, 2016.Li, F., Zeng, X. D., and Levis, S.: A process-based fire parameterization of
intermediate complexity in a Dynamic Global Vegetation Model, Biogeosciences,
9, 2761–2780, 10.5194/bg-9-2761-2012, 2012.López-Saldaña, G., Bistinas, I., and Pereira, J. M. C.: Global
analysis of radiative forcing from fire-induced shortwave albedo change,
Biogeosciences, 12, 557–565, 10.5194/bg-12-557-2015, 2015.Mangeon, S., Voulgarakis, A., Gilham, R., Harper, A., Sitch, S., and
Folberth, G.: INFERNO: a fire and emissions scheme for the UK Met Office's
Unified Model, Geosci. Model Dev., 9, 2685–2700,
10.5194/gmd-9-2685-2016, 2016.
Mooney, H. A. (Ed.): Convergent in Chile and California Mediterranean Climate
Ecosystems, Dowden, Hutchinson and Ross, Inc. Stroudsberg, Penn., USA, 1977.Ogden, J., Basher, L., and McGlone, M.: Botanical Briefing: Fire, Forest
Regeneration and Links with Early Human Habitation: Evidence from New
Zealand, Ann. Bot.-London, 81, 687–696, 10.1006/anbo.1998.0637, 1998.Pellegrini, A. F. A., Anderegg, W. R. L., Paine, C. E. T., Hoffmann, W. A.,
Kartzinel, T., Rabin, S. S., Sheil, D., Franco, A. C., and Pacala, S. W.:
Convergence of bark investment according to fire and climate structures
ecosystem vulnerability to future change, Ecol. Lett., 20, 307–316,
10.1111/ele.12725, 2017.Pfeiffer, M., Spessa, A., and Kaplan, J. O.: A model for global biomass
burning in preindustrial time: LPJ-LMfire (v1.0), Geosci. Model Dev., 6,
643–685, 10.5194/gmd-6-643-2013, 2013.Poulter, B., MacBean, N., Hartley, A., Khlystova, I., Arino, O., Betts, R.,
Bontemps, S., Boettcher, M., Brockmann, C., Defourny, P., Hagemann, S.,
Herold, M., Kirches, G., Lamarche, C., Lederer, D., Ottlé, C., Peters,
M., and Peylin, P.: Plant functional type classification for earth system
models: results from the European Space Agency's Land Cover Climate Change
Initiative, Geosci. Model Dev., 8, 2315–2328,
10.5194/gmd-8-2315-2015, 2015.Prentice, I. C., Kelley, D., Foster, P., Friedlingstein, P., Harrison, S.,
and Bartlein, P.: Modeling fire and the terrestrial carbon balance, Global
Biogeochem. Cy., 25, GB3005, 10.1029/2010GB003906, 2011.Pugh, T., Jones, C., Huntingford, C., Burton, C., Arneth, A., Brovkin, V.,
Ciais, P., Lomas, M., Robertson, E., Piao, S., and Sitch, S.: A large
committed long-term sink of carbon due to vegetation dynamics, Earth's
Future, 6, 1413–1432, 10.1029/2018EF000935, 2018.Rabin, S. S., Melton, J. R., Lasslop, G., Bachelet, D., Forrest, M., Hantson,
S., Kaplan, J. O., Li, F., Mangeon, S., Ward, D. S., Yue, C., Arora, V. K.,
Hickler, T., Kloster, S., Knorr, W., Nieradzik, L., Spessa, A., Folberth, G.
A., Sheehan, T., Voulgarakis, A., Kelley, D. I., Prentice, I. C., Sitch, S.,
Harrison, S., and Arneth, A.: The Fire Modeling Intercomparison Project
(FireMIP), phase 1: experimental and analytical protocols with detailed model
descriptions, Geosci. Model Dev., 10, 1175–1197,
10.5194/gmd-10-1175-2017, 2017.Rabin, S. S., Ward, D. S., Malyshev, S. L., Magi, B. I., Shevliakova, E., and
Pacala, S. W.: A fire model with distinct crop, pasture, and non-agricultural
burning: use of new data and a model-fitting algorithm for FINAL.1, Geosci.
Model Dev., 11, 815–842, 10.5194/gmd-11-815-2018, 2018.Randerson, J., Chen, Y., van der Werf, G., Rogers, B., and Morton, D.: Global
burned area and biomass burning emissions from small fires, J. Geophys. Res.,
117, G04012, 10.1029/2012JG002128, 2012.Senior, C. A., Andrews, T., Burton, C., Chadwick, R., Copsey, D., Graham, T.,
Hyder, P., Jackson, L., McDonald, R., Ridley, J., Ringer, J., and Tsushima,
Y.: Idealized climate change simulations with a high-resolution physical
model: HadGEM3-GC2, J. Adv. Model. Earth Sy., 8, 813–830,
10.1002/2015MS000614, 2016.Seo, H. and Kim, Y.: Interactive Impacts of Fire and Vegetation Dynamics on
Global Carbon and Water Budgets using Community Land Model version 4.5,
Geosci. Model Dev. Discuss., 10.5194/gmd-2018-231, in review,
2018.
Settele, J., Scholes, R., Betts, R., Bunn, S., Leadley, P., Nepstad, D.,
Overpeck, J. T., and Taboada, M. A.: Terrestrial and inland water systems,
in: Climate Change 2014: Impacts, Adaptation, and Vulnerability. Part A:
Global and Sectoral Aspects. Contribution of Working Group II to the Fifth
Assessment Report of the Intergovernmental Panel on Climate Change, edited
by: Field, C. B., Barros, V. R., Dokken, D. J., Mach, K. J., Mastrandrea, M.
D., Bilir, T. E., Hatterjee, M., Ebi, K. L., Estrada, Y. O., Genova, R. C.,
Girma, B., Kissel, E. S., Levy, A. N., MacCracken, S., Mastrandrea, P. R.,
and White, L. L., Cambridge University Press, Cambridge, UK and New York, NY,
USA, 271–359, 2014.
Shakesby, R. A. and Doerr, S. H.: Wildfire as a hydrological and
geomorphological agent, Earth-Sci. Rev., 74, 269–307, 2006.Sitch, S., Friedlingstein, P., Gruber, N., Jones, S. D., Murray-Tortarolo,
G., Ahlström, A., Doney, S. C., Graven, H., Heinze, C., Huntingford, C.,
Levis, S., Levy, P. E., Lomas, M., Poulter, B., Viovy, N., Zaehle, S., Zeng,
N., Arneth, A., Bonan, G., Bopp, L., Canadell, J. G., Chevallier, F., Ciais,
P., Ellis, R., Gloor, M., Peylin, P., Piao, S. L., Le Quéré, C.,
Smith, B., Zhu, Z., and Myneni, R.: Recent trends and drivers of regional
sources and sinks of carbon dioxide, Biogeosciences, 12, 653–679,
10.5194/bg-12-653-2015, 2015.
Soares, B., Silvestrini, R., Nepstad, D., Brando, P., Rodrigues, H., Alencar,
A., Coe, M., Locks, C., Lima, L., Hissa, L., and Stickler, C.: Forest
fragmentation, climate change and understory fire regimes on the Amazonian
landscapes of the Xingu headwaters, Landscape Ecol., 27, 585–598, 2012.Staver, A. C., Archibald, S., and Levin, S. A.: The global extent and
determinants of savanna and forest as alternative biome states, Science,
334, 230–232, 10.1126/science.1210465, 2011.
Turner, B. L., Skole, D. L., Sanderson, S., Fischer, G., Fresco, L. O., and
Leemans, R.: Land-use and land-cover change. Science/Research Plan, IGBP
Report No. 35 and HDP Report No. 7, 132 pp., Stockholm, Sweden and Geneva,
Switzerland, 1995.van der Werf, G. R., Randerson, J. T., Giglio, L., van Leeuwen, T. T., Chen,
Y., Rogers, B. M., Mu, M., van Marle, M. J. E., Morton, D. C., Collatz, G.
J., Yokelson, R. J., and Kasibhatla, P. S.: Global fire emissions estimates
during 1997–2016, Earth Syst. Sci. Data, 9, 697–720,
10.5194/essd-9-697-2017, 2017.Veraverbeke, S., Rogers, B. M., Goulden, M. L., Jandt, R. R., Miller, C. E.,
Wiggins, E. B., and Randerson, J. T.: Lightning as a major driver of recent
large fire years in North American boreal forests, Nat. Clim. Change, 7,
529–534, 10.1038/nclimate3329, 2017.
Watson, R. T., Noble, I. R., Bolin, B., Ravindranath, N. H., Verardo, D. J.,
and Dokkenm D. J. (Eds.): IPCC, Cambridge University Press, Cambridge, UK,
375 pp., 2000.