The Community Atmosphere–Biosphere Land Exchange model (CABLE) is a land
surface model (LSM) that can be applied stand-alone and provides the
land surface–atmosphere exchange within the Australian Community Climate
and Earth System Simulator (ACCESS). We describe new developments that extend
the applicability of CABLE for regional and global carbon–climate
simulations, accounting for vegetation responses to biophysical and
anthropogenic forcings. A land use and land cover change module driven by
gross land use transitions and wood harvest area was implemented, tailored to
the needs of the Coupled Model Intercomparison Project 6 (CMIP6). Novel
aspects include the treatment of secondary woody vegetation, which benefits
from a tight coupling between the land use module and the Population Orders
Physiology (POP) module for woody demography and disturbance-mediated
landscape heterogeneity. Land use transitions and harvest associated with
secondary forest tiles modify the annually resolved patch age distribution
within secondary vegetated tiles, in turn affecting biomass accumulation and
turnover rates and hence the magnitude of the secondary forest sink.
Additionally, we implemented a novel approach to constrain modelled GPP
consistent with the coordination hypothesis and predicted by evolutionary
theory, which suggests that electron-transport- and Rubisco-limited rates
adjust seasonally and across biomes to be co-limiting. We show that the
default prior assumption – common to CABLE and other LSMs – of a fixed
ratio of electron transport to carboxylation capacity at standard temperature
(
These new developments enhance CABLE's capability for use within an Earth system model and in stand-alone applications to attribute trends and variability in the terrestrial carbon cycle to regions, processes and drivers. Model evaluation shows that the new model version satisfies several key observational constraints: (i) trend and interannual variations in the global land carbon sink, including sensitivities of interannual variations to global precipitation and temperature anomalies; (ii) centennial trends in global GPP; (iii) coordination of Rubisco-limited and electron-transport-limited photosynthesis; (iv) spatial distributions of global ET, GPP, biomass and soil carbon; and (v) age-dependent rates of biomass accumulation in boreal, temperate and tropical secondary forests.
CABLE simulations agree with recent independent assessments of the global
land–atmosphere flux partition that use a combination of atmospheric
inversions and bottom-up constraints. In particular, there is agreement that
the strong
The Community Atmosphere–Biosphere Land Exchange model (CABLE) is a land surface model (LSM) that can be applied in stand-alone applications and also provides the land surface–atmosphere exchange within the Australian Community Climate and Earth System Simulator (ACCESS; Kowalczyk et al., 2013; Law et al., 2017; Ziehn et al., 2017). In its stand-alone configuration, CABLE was used in the IPCC 5th assessment report (IPCC, 2014) and is one of an ensemble of ecosystem and land surface models contributing to the Global Carbon Project's annual update of the global carbon budget (Le Quéré et al., 2016, 2018). The current paper describes updates to CABLE (Haverd et al., 2017) targeting two key areas that have been identified as limitations in the applicability and utility of the existing generation of LSMs: (i) land use and land cover change (LULCC, hereafter abbreviated to LUC) and (ii) adaptation of photosynthesis to changing environmental conditions.
Additional model updates based on existing parameterisations from the literature include the following: (i) drought and summergreen phenology (Sitch et al., 2003; Sykes et al., 1996); (ii) low-temperature reductions in photosynthetic rates in boreal forests (Bergh et al., 1998); (iii) photoinhibition of leaf day respiration (Clark et al., 2011); and (iv) acclimation of autotrophic respiration (Atkin et al., 2016). These are described in Appendix 2.
The CABLE version that precedes developments described here (hereafter “Prior CABLE”) assumes fixed present-day or pre-industrial vegetation cover in the absence of land management. Capturing the impact of human LUC on the terrestrial carbon and water cycles and on land–atmosphere coupling is a key application of LSMs and associated Earth system models (ESMs) and a prerequisite for the evaluation of the models against observation-based datasets.
For the CMIP6 climate model inter-comparison process, the globally gridded Harmonised Land Use Dataset (LUH2; Hurtt et al., 2016, 2011) specifies a matrix of transitions between land use classes (e.g. primary forest, secondary forest, pasture, cropland) through time (Lawrence et al., 2016). In traditional LSMs, these transitions must be translated into annual land cover maps that specify the fraction of the land surface occupied by each plant functional type (PFT; Lawrence et al., 2012). This approach reduces the transition matrix to a set of net transitions, thereby discarding information about the gross transitions leading to land cover change. Simulations driven by gross land use transitions produce emissions that are 15–40 % higher than the net transitions alone (Hansis et al., 2015; Stocker et al., 2014; Wilkenskjeld et al., 2014).
Traditional LSMs are also unable to simulate realistic dynamics resulting
from the accumulation of carbon in forests following harvest and
agricultural abandonment – the so-called secondary forest sink – that is
an important contributor to the extant global terrestrial carbon sink
(Shevliakova et al., 2009) second only to
In contrast to traditional LSMs, demography-enabled dynamic vegetation models (DVMs) can implement gross transitions directly and provide a realistic representation of secondary forest sink by explicitly simulating biomass removal and subsequent recovery following a land use event (e.g. Shevliakova et al., 2009). However, keeping track of a representative distribution of landscape elements (patches) with different times since disturbance can be computationally difficult as repeated land use events can lead to a very high number of such elements in a grid cell.
In this work, we develop a novel LUC scheme for CABLE that is driven by LUH2 gross transitions and represents age effects on biomass dynamics in all tiles with woody vegetation, including those occupied by secondary forest. This is achieved via coupling with the POP module for woody demography and disturbance-mediated heterogeneity (Haverd et al., 2013b). The key simplification in the POP approach, compared with other demography-enabled DVMs, is to compute physiological processes such as photosynthesis at the scale of a land cover tile (“grid scale”), but to partition the grid-scale biomass increment amongst sub-grid-scale patches, each subject to its own dynamics, and distinguished by the time since last disturbance. This makes tracking biomass in a large number of patch ages (as arise through both natural disturbance and human land cover change) easy and circumvents the computational difficulties of tracking land cover classes in DVMs.
Almost all global LSMs use the photosynthesis model of Farquhar et
al. (1980) or a related scheme derived from this model. Different
implementations result in divergent estimates of the response of
photosynthesis to environmental drivers in large-scale models (e.g. Friend et
al., 2014). One reason for this may be that global LSMs have mostly neglected
the constraint imposed by the evolutionary ecological assumption that plants
optimise productivity in their environment through relative investment in
electron-transport- and Rubisco-limited steps in the photosynthesis chain,
which adjust seasonally and across biomes to be co-limiting. This so-called
coordination hypothesis was originally proposed by Chen et al. (1993) and
has been verified experimentally by Maire et al. (2012). Its advantages as an
approach to modelling photosynthetic dynamics using limited data constraints
was pointed out by Wang et al. (2017), while Ali et al. (2016) have
incorporated it into a global mechanistic model of photosynthetic capacity
based on the optimal nitrogen allocation model of Xu et al. (2012). In this
work, we will show that the assumption of a temporally invariant ratio of
Rubisco and electron transport capacities (at standard temperature), adopted
in Prior CABLE and typically in other LSMs, is not only inconsistent with the
coordination hypothesis, but introduces large uncertainty in the simulated
sensitivity of GPP to atmospheric
The paper is structured as follows. In Sect. 2 we review the basic structure
of CABLE. In Sect. 3 we describe the model developments that are the focus of
this work: firstly, updates to the POP module for woody demography and
disturbance; secondly, the new land use and land cover change module;
thirdly, the dynamic optimisation of plant photosynthesis. In Sect. 4, we
describe the modelling protocol that is used to deliver simulations for
evaluating the new model version and assessing terrestrial carbon cycle
implications of changing climate,
Major sub-models of CABLE (revision 4601) showing forcing data, characteristic time steps and information flows between modules, which include fluxes, store updates, and changes to vegetation characteristics and their spatial extent (tile areas) within grid cells. Data from faster modules are aggregated before passing to slower modules. Faster modules are updated with data from slower modules at the rate of the slower time step.
Figure 1 summarises the content of CABLE and how the components interact. Further details are presented in Fig. A1 (Appendix A1) as pseudo-code for each component and Tables A1–A3 (Appendix A3), which document the parameter values and temperature response functions of photosynthesis used in this work. CABLE consists of a biophysics core (Haverd et al., 2016a; Kowalczyk et al., 2013; Wang et al., 2011), the CASA-CNP “biogeochemistry” module (Wang et al., 2010), the POP module for woody demography and disturbance-mediated landscape heterogeneity (Haverd et al., 2013c, 2014), and a completely new module for land use and land management (POPLUC).
The biophysics core (sub-diurnal time step) consists of four components: (1) the radiation module describes radiation transfer and absorption by sunlit and shaded leaves (Goudriaan and van Laar, 1994); (2) the canopy micrometeorology module describes the surface roughness length, zero-plane displacement height and aerodynamic conductance from the reference height to the air within canopy or to the soil surface (Raupach, 1994); (3) the canopy module includes the coupled energy balance, transpiration, stomatal conductance, photosynthesis and respiration of sunlit and shaded leaves (Wang and Leuning, 1998); and (4) the soil module describes heat and water fluxes within soil (six vertical layers) and snow (up to three vertical layers) and at their respective surfaces. The CASA-CNP biogeochemistry module (daily time step) inherits daily net photosynthesis from the biophysical code, calculates autotrophic respiration, allocates the resulting net primary production (NPP) to leaves, stems and fine roots, and transfers carbon, nitrogen and phosphorous between plant, litter and soil pools, accounting for losses of each to the atmosphere and by leaching. POP (annual time step) inherits annual stem NPP from CASA-CNP and simulates woody ecosystem stand dynamics, demography and disturbance-mediated heterogeneity, returning the emergent rate of biomass turnover to CASA-CNP.
The biophysics core of CABLE has been benchmarked using prescribed meteorology (e.g. Best et al., 2015; Zhang et al., 2013; Zhou et al., 2012) and its performance evaluated as part of the Australian Community Climate and Earth System Simulator climate model (Kowalczyk et al., 2013). The CASA-CNP module was developed and tested as a stand-alone module (Wang et al., 2010), and its basic performance was demonstrated as part of ACCESS (Law et al., 2017; Ziehn et al., 2017). POP (coupled to CABLE) has been evaluated against savanna data (Haverd et al., 2013b, 2016b) and boreal and temperate forest data (Haverd et al., 2014).
In previous work, POP has been coupled to both the CABLE and HAVANA land surface schemes and demonstrated to successfully replicate the effects of rainfall and fire disturbance gradients on vegetation structure along a rainfall gradient in Australian savanna – the Northern Australian Tropical Transect (Haverd et al., 2013c, 2016b) – and leaf–stem allometric relationships derived from global forest data. For the latter, it may be argued to reflect the simultaneous development of trees in closed forest stands in terms of structural and functional (productivity) attributes (Haverd et al., 2014). The summary below is reproduced from these papers, which describe POP in detail and with full equations. To enable the extension of CABLE to simulate dynamic land use and implications for forest carbon uptake, we used the most recent version of POP's representation of growth partitioning amongst age and size classes (cohorts) of trees established in the same year; it accounts for both cohort-dependent light interception and sapwood respiration. This contrasts with the original growth partitioning, which assumed that individuals capture resources in varying proportion to their size.
POP is designed to be modular, deterministic, computationally efficient and based on defensible ecological principles. POP simulates the allometric growth of cohorts of trees that compete for light and soil resources within a patch. Parameterisations of tree growth, allometry, recruitment and mortality are broadly based on the approach of the LPJ-GUESS Dynamic Vegetation Model (Smith et al., 2001). The time step is 1 year.
Input variables to POP are annual grid-scale stem biomass increment and mean return times for two classes of disturbance: (i) “catastrophic” disturbance, which kills all individuals (cohorts) and removes all biomass in a given patch; and (ii) “partial” disturbances, such as fire, which result in the loss of a size-dependent fraction of individuals and biomass, preferentially affecting smaller (younger) cohorts. For the present study, we adopt a mean catastrophic disturbance return time of 100 years and neglect partial disturbance, such as damage caused by wildfires. Stem biomass increment is provided by the host land surface model (LSM), here CABLE.
State variables are the density of tree stems partitioned among cohorts of trees and representative patches of different age since last disturbance across a simulated landscape or grid cell. Each patch has a number of cohorts. Trees in each cohort are the same age and size because they are established simultaneously and share the same growth rate. Patches are not spatially explicit. Their areal representation in the landscape is given by the patch age distribution.
In the current implementation of POP, the annual stem biomass increment is
partitioned among cohorts and patches in proportion to the current net primary
production of the given cohort (Haverd et al., 2016b). For this purpose,
gross primary production and autotrophic respiration for each woody tile are
passed from CABLE to POP, and each is partitioned amongst patches and
cohorts. Gross resource uptake is partitioned amongst cohorts and patches in
proportion to light interception, which is evaluated for each cohort as the
difference between downward-looking gap probabilities above and below each
cohort. Gap probabilities are calculated using the geometric approach of
Haverd et al. (2012). This requires estimates of cohort-specific crown
cross-sectional area (related allometrically to DBH) and LAI computed using
the CABLE maximum leaf area and distributed amongst patches and cohorts in
proportion to sapwood area. For autotrophic respiration, leaf, fine-root and
sapwood respiration components are also partitioned amongst cohorts and
patches according to the size of each biomass component. Cohort-specific
sapwood is prognosed by assuming sapwood conversion to heartwood at a rate
0.05 year
Cohort stem density is initialised as recruitment density and is episodically reset when the patch experiences disturbance. Mortality, parameterised as the sum of cohort-specific resource limitation and crowding components, reduces the stem density in the intervening period. Resource limitation mortality, a function of growth efficiency (GE; i.e. growth rate relative to biomass), is described by a logistic curve with an inflection point representing a critical GE level at which plants experience a steep increase in mortality risk due to a shortage of resources to deploy in response to stress or biotic damage (Haverd et al., 2013c). The crowding mortality component (Haverd et al., 2014) allows for self-thinning in forest canopies.
Additional mortality occurs as a result of disturbances. Patches representing stands of differing age since last disturbance are simulated for each grid cell. It is assumed that each grid cell is large enough to accommodate a landscape in which the frequency of patches of different ages follows a negative exponential distribution with an expectation related to the current disturbance interval. This assumption is valid if grid cells are large relative to the average area affected by a single disturbance event and disturbances are a Poisson process occurring randomly with the same expectation at any point across the landscape independent of previous disturbance events. To account for disturbances and the resulting landscape structure, state variables of patches of different ages are linearly interpolated between ages and weighted by probability intervals from the negative exponential distribution. The resultant weighted average of, for example, total stem biomass or annual stem biomass turnover is taken to be representative for the grid cell as a whole.
In earlier applications, CABLE–POP coupling consisted of just two exchanges: (i) stem NPP passed from the host LSM to POP and (ii) woody biomass turnover returned from POP to the host LSM. To convert between stem biomass (POP) and tree biomass (CABLE), we assume a ratio of 0.7, a representative average for forest and woodland ecosystems globally (Poorter et al., 2012). The POP biomass lost by mortality is applied as an annual decrease in the CASA-CNP tree biomass pool and replaces the default fixed biomass turnover rate. In the current work, the coupling also includes the return of sapwood area and sapwood biomass to the CASA-CNP biogeochemical module of CABLE, in which these variables respectively influence C allocation to leaves and autotrophic respiration. Combined allocation to leaves and wood is partitioned following the pipe model (Shinozaki et al., 1964) such that a target ratio of leaf area to sapwood area (a global value of 5000 is assumed) is maintained. Sapwood replaces stem-wood biomass in the CASA-CNP calculation of stem respiration. These feedbacks of POP structural variables on leaf area and autotrophic respiration result in net primary production (NPP) that reflects the area-averaged sapwood area and mass of each woody tile.
POP is not a replacement for a full-featured dynamic vegetation model (DVM), but does overcome key limitations of Prior CABLE and many DVMs adopted by most Earth system models (Arora et al., 2013), for which biomass turnover is often represented as a first-order decay process expressed as the product of grid cell biomass and a bulk rate parameter. This “big wood” approximation does not resolve underlying population and community processes such as recruitment, mortality and competition between individuals for limiting resources (e.g. Sitch et al., 2003) and has been demonstrated to lead to an inaccurate trajectory of biomass accumulation with stand age (Wolf et al., 2011; Haverd et al., 2014). Big wood models are additionally unable to directly exploit the wealth of information on forest stand structure and dynamics available from forest inventories. By discriminating individual and population growth and explicitly representing asymmetric competition among age and size classes of trees co-occurring within forest stands, POP overcomes the limitation of the big wood approach and has proved able to reproduce allometric relationships reflecting linkages between productivity, biomass and density in widely distributed forests (Haverd et al., 2014). This is achieved without a marked increase in model complexity or computational demand thanks to a modular design that separates the role of the parent land surface model (prognosing whole-ecosystem production) and the population dynamics model (partitioning the production among cohorts, computing mortality for each and returning the stand-level integral as whole-ecosystem biomass turnover to the parent model; Fig. 1).
Spatial distribution of BIOME1 biomes (Table 1) that determines the type of primary vegetation cover
CABLE primary vegetation: mapping of BIOME1 biomes to CABLE plant functional types.
A drawback of this modular approach is that age effects on leaf area and NPP are not accounted for explicitly at the scale of the individual because these variables are computed for each woody tile and in turn distributed amongst POP patches and cohorts. Feedbacks of stand structure on leaf area and NPP thus reflect the area-averaged structural properties (sapwood area and sapwood mass) of each woody tile.
POP does not represent competitive interactions among PFTs that provide an
important explanation for global biome distributions and may modulate the
responses of vegetation to future climate and
This development enables the simulation of the effect of LUC on land cover fractions and associated carbon flows into and out of soil, litter, vegetation and product pools.
Three land use tile types are considered: primary woody vegetation (p),
secondary woody vegetation (s) and open grassy vegetation (g), the latter
encompassing natural grassland, rangeland, pasture and cropland. Forcing data
comprising four possible annual gross transition rates are used to drive the
annual LUC-induced changes to land use area fractions. These transition rates
are the following: (i) primary clearing (p
Potential vegetation cover is prescribed using BIOME1 (Prentice et al., 1992), a semi-mechanistic climate-envelope approach, to construct global spatial distribution of biomes according to CABLE's own climate drivers, which are accumulated from 30 years (1901–1930) of meteorological inputs (Fig. 2).
Biomes (combinations of dominant plant types; Prentice et al., 1992) are
mapped to a single CABLE plant functional type (PFT) or in some cases to two
CABLE PFTs (one woody and one herbaceous) with fixed relative areal
proportions (Table 1). We make use of five woody vegetation types (evergreen
needle-leaf, evergreen broadleaf, deciduous needle-leaf, deciduous broadleaf,
shrub) and six non-woody types (C
Each land use tile has an associated areal fraction representing its
fractional area cover of the grid cell. Land transition area rates augment
and deplete land use area fractions, subject to land availability. In
secondary forest tiles, the areal fraction of each integral age class
(0–400 years) is also tracked: a transition to secondary forest (p
The POPLUC code provides the secondary forest patch age distribution to POP. POP tracks biomass in each of a set of patches with different ages based on patch-dependent growth and turnover. It then computes biomass for each integral age class represented by the secondary forest tile patch age distribution by interpolating biomass in the simulated patches.
POPLUC represents integral secondary forest ages classes from 0 to 1000 years old, although many ages may have a weight of zero. The frequency distribution is fully dynamic. In contrast, POP represents 60 patches in each woody tile spanning a distribution of ages from 0 to 1000.
Changes in pool sizes of biomass, soil and litter carbon in the
biogeochemical module are updated to reflect the areal changes from gross
land use transitions. Analogous updates occur for nitrogen pools. The mass
balance equation for the carbon density
The carbon gained by receiver transitions is generally
Carbon losses by secondary forest harvest and clearing need to be resolved from net biomass loss in secondary forest tiles, which also includes components from natural disturbance and areal expansion. POP diagnoses a change in biomass resulting from the aggregate shift in age distribution contributed by natural disturbance, forest expansion, harvest and clearing. The proportional contributions of each of these processes to total biomass change is recorded. The carbon flux implied by this total biomass change is subsequently disaggregated according to the previously recorded proportional contributions of each process.
Carbon removal from the landscape by crop harvest and grazing are treated simply. Crops and pasture are not treated in separate land use tiles, but are simulated as grass in the open “grassy” tile of each grid cell. The areal fractions of cropland and pasture in each open tile are tracked via the gross transitions to and from these land use types. These fractions, combined with assumed respective removals of 90 % and 50 % of above-ground NPP by crop harvest and grazing (Lindeskog et al., 2013), are used to prescribe leaf-litter transfer to an agricultural product pool with a turnover time of 1 year. Following Lindeskog et al. (2013), soil carbon loss by tillage is simulated by increasing the turnover of soil carbon by 50 % in croplands. Where crops and pasture occupy more than 10 % of a grass tile, it is assumed that there is no nutrient limitation to growth.
Photosynthesis, as represented by the Farquhar et al. (1980) model, may be
limited by the Rubisco-catalysed maximum rate of carboxylation
(
Here we review the equations of the C
Net photosynthesis (
Net photosynthesis is also equated with biochemical demand for
Stomatal conductance is expressed as a linear function of
Equations (3), (4) and (
The approach to the optimisation of Leaf nitrogen resources may be dynamically redistributed at a 5-day
timescale at no cost; i.e. Leaf nitrogen resources available for partitioning between Rubisco and
electron transport capacity are proportional to effective nitrogen content
( The prior values of The emerging contributions of electron-transport- and Rubisco-limited
rates contribute approximately equally to total net photosynthesis (Chen et
al., 1993). In practice, this requires a relative cost factor
The method for implementing these assumptions in CABLE is as follows.
Maintain a 5-day history of sub-diurnal leaf-level meteorology (absorbed
PAR; leaf–air VPD difference; leaf temperature, Construct a
function that calculates leaf nitrogen cost per unit net photosynthesis
( Implement a search algorithm to find Insert a call to the optimisation algorithm at the end of each day at the
point in the code at which
Global simulations were performed at
Simulation scenarios.
Simulations were driven by (i) daily CRU-NCEP V7 (1901–2016; Viovy, 2016)
downscaled to 3-hourly resolution using a weather generator (Haverd et al.,
2013a), (ii)
Simulations (Table 2) were performed to quantify the net land–atmosphere
carbon flux and attribute it to three components: (i) the land–atmosphere
exchange that would occur in response to changing climate,
This allows the net flux
Scenario (iv) is included so that the net ecosystem production (NPP minus
heterotrophic respiration) on secondary forest tiles can be partitioned
between secondary forest regrowth and legacy emissions from post-harvest and
post-clearing residues, which are zero in Scenario (iv). Note here that
Scenario (v) is included to resolve the net LUC emissions associated with
grazing and cropland management as the difference
The loss of additional sink capacity (1860 reference year)
The initialisation phase of each scenario was designed to establish the
dynamic equilibrium between model state (biomass and soil carbon pools) and
the forcing data. All scenarios were initialised from zero biomass (to
ensure biomass variables in POP and CASA-CNP start from the same value) and
arbitrary soil carbon and nutrient stocks and brought to equilibrium with
1901–1920 climate by five repetitions of a pair of model runs. This pair
comprised a full model run (1901–1920 climate, 1860 land cover,
In addition to the above scenarios, we also explored the impact on global GPP
of dynamically optimising
Observation-based
Model–data comparisons of the spatial distributions of key fluxes and stocks are presented in Fig. 3. We choose to evaluate the model against GPP, biomass and soil carbon because these are key quantities that are critical constraints on the global terrestrial carbon cycle and for which global distributions are available. We include evapotranspiration (ET) here as it is a key constraint on GPP because both ET and GPP are regulated by stomatal conductance.
The mean of evapotranspiration (ET) was obtained from the LandFlux
Evaluation of CABLE (1990) above-ground biomass predictions against
biomass compartments data (Teobaldelli, 2008; Teobaldelli et al., 2009) separated into deciduous
broadleaf and evergreen needle-leaf classes:
Observation-based global gross primary production (GPP) was obtained from
upscaled FLUXNET eddy covariance tower measurements (1982–2011; Jung et
al., 2010). CABLE and FLUXNET estimates of the latitudinal distribution of
GPP differ by a mean absolute error of 147 g C m
Observation-based above-ground forest biomass at
Soil carbon density in the top 1 m of soil for the year 2000 was obtained
from the Harmonized World Soil Database (HWSDA; version 1.2).
(FAO/IIASA/ISRIC/ISSCAS/JRC, 2009). Latitudinal profiles of soil carbon from
CABLE (total soil carbon and litter) differ from the HWSDA product by a mean
absolute error of 1.8 Pg C deg
Forest inventory data for above-ground biomass and age were sourced from the
Biomass Compartments Database (Teobaldelli, 2008; Teobaldelli et al., 2009). This database contains
data from around 5790 plots and represents a harmonised collection of
the Cannell (1982) and Usoltsev (2001) datasets covering the temperate and
boreal forest region globally. In earlier work we used the database to
construct biomass density plots for the purpose of calibrating the crowding
mortality component of POP and to evaluate CABLE leaf-stem allometry plots
relating foliage and stem biomass per tree (Haverd et al., 2014). Here we
directly evaluate CABLE predictions of above-ground stem biomass for 1990
(approximate median year for the observational data; Fig. 4) for a wide
range of stand ages (2–200 years). Despite significant scatter, predictions
show low bias (Fig. 4i and ii) and biomass–age relationships that accord with
the data (Fig. 4iii and iv): (DBL,
CABLE and observation-based estimates (Poorter et al., 2016) of Neotropical secondary forest biomass after 20 years of regrowth versus mean annual precipitation. CABLE estimates are extracted from secondary forest tiles in tropical rainforest, tropical seasonal forest and tropical dry forest–savanna biomes (Fig. 2) in South America. The lower distinct cloud of CABLE-simulated values corresponds to the tropical dry forest–savanna biomes.
CABLE regrowth rates of secondary forests in the tropical rainforest, tropical seasonal forest and tropical dry forest–savanna biomes (Fig. 2) in South America compare well with observation-based estimates by Poorter et al. (2016). This database has 1500 forest plots at 45 sites spanning the major environmental gradients across the Neotropics (Fig. 5), where mean annual rainfall is the strongest environmental predictor of biomass accumulation after 20 years (Poorter et al., 2016).
In this region, CABLE predicts that secondary forest biomass recovers to
Four examples of contrasting regional land use histories (
Each column in Fig. 6 corresponds to one site, and the four rows show the following:
(1) land use transition rates for clearing (p
Contrasting land use and land management for sample
The global terrestrial carbon balance (1860–2016) and its
partitioning, as influenced by LUC, land management,
The land use history for this grid cell is dominated by the clearing of primary
forest with peak clearing events in 1940 and 1960 corresponding to respective
conversion to rangeland and cropland. The 1860 carbon stocks are partitioned
approximately equally between soil and vegetation. Cumulative carbon loss of
30 kg C m
The land use history is dominated by shifting cultivation (s
There is no primary forest. Land use activity is dominated by secondary
forest harvest pre-1920 and the abandonment of pasture. The cessation of harvest
leads to significant carbon accumulation in secondary forest vegetation
post-1920. Of the total carbon accumulation since 1860 (7 kg C m
This is a landscape dominated by agricultural activity. All secondary forest
is cleared by 1900; however, the abandonment of cropland post-1945 leads to an
expansion of secondary forest land. Carbon stocks in vegetation are very low
because of secondary forest harvest. Soil carbon in open land is depleted
because of cropland management (tillage and removal of biomass). The
cumulative carbon loss from 1860 is 4 kg C m
Figure 7 shows the combined impacts of changing climate,
The net effect of clearing and abandonment has been a decline in forest area
of
Forest area has declined by
This region has been subject to particularly aggressive deforestation, with
Global primary forest area has decreased by
LUC emissions have been declining steadily since 1960 (albeit with a slight
upturn since 2005), while the
The net land carbon sink (Pg C year
While the
Illustrative simulations of net photosynthesis, fractional
Rubisco limitation, elasticity of net photosynthesis with respect to surface
(
Table 3 shows that CABLE's partitioning of the net land–carbon sink between
the tropics and NH extratropics accords well with a recent synthesis by
Schimel et al. (2015), which utilised atmospheric inversion data (selected
according to assessment against aircraft vertical profile observations),
biomass inventory data and an ensemble of model estimates of global land
carbon uptake in response to rising
The effect of dynamically optimising the ratio of
In the tropics, the dynamic values of
Latitudinal profiles of
Global land carbon sink, as predicted by CABLE and five terrestrial biosphere models contributing to TRENDY-v5 (Le Quéré et al., 2016), and the Global Carbon Project (GCP) estimate, as the sum of atmosphere and ocean sinks minus fossil fuel emissions (Le Quéré et al., 2016).
The impacts of optimising
Simulated annual time series of the global land carbon sink: correlation with GCP, linear trend and mean sink.
Interannual global carbon–climate sensitivities, as defined by Eq. (18).
Key functions of global terrestrial biosphere models such as CABLE are the
attribution and projection of the global net land carbon sink. Therefore we
assess CABLE predictions against observation-based estimates of this
important quantity. Figure 10 depicts simulated annual times series of the
global land carbon sink from CABLE and the corresponding Global Carbon
Project (GCP) estimate, diagnosed as the sum of atmosphere and ocean sinks
minus fossil fuel emissions (Le Quéré et al., 2016). Of the 14 land
models represented in the GCP's 2016 assessment of the global carbon budget
(Le Quéré et al., 2016), the five contributing simulations of the net
land carbon sink (as opposed to the residual land sink, equivalent to the net
land sink plus net LUC emissions, represented by all land models) are also
shown in Fig. 10. For each model, the correlation of annual values with GCP
estimates (1959–2015), trend (1980–2015) and magnitude (2006–2015) is
quantified in Table 4. Uncertainty on the GCP estimates is
0.4 Pg C year
CABLE captures a high proportion of the variance in the GCP estimate relative to the other models in Table 4. This is in part attributable to its relatively good representation of the 1973–1974 and 1975–1976 positive anomalies corresponding to very strong La Niña events. Moisture sensitivities of both productivity and decomposition are important for capturing the response of the net flux to such events: in particular the high temporal correlation of heterotrophic respiration with NPP in water-limited environments reduces the response of the net flux compared with the response of NPP (Haverd et al., 2016c).
In contrast, CABLE under-predicts large negative anomalies corresponding to 1987–1988 and 1997–1998 El Niño events. Possible explanations are that wildfire is not represented, and the simulated drought response of tropical forests may be too weak.
We evaluate the global land carbon–climate sensitivity, following the
analysis by Piao et al. (2013), of 10 terrestrial biosphere models. A linear
model relates anomalies in the annual detrended land carbon sink
(
We have presented CABLE model developments that improve its applicability as a terrestrial biosphere model for use within an Earth system model and in stand-alone applications to attribute trends and variability in the terrestrial carbon cycle to regions, processes and drivers. Model evaluation has shown that the new model version satisfies several key observational constraints: (i) trend and interannual variations in the global land carbon sink, including sensitivities of interannual variations to global precipitation and temperature anomalies; (ii) centennial trends in global GPP; (iii) coordination of Rubisco-limited and electron-transport-limited photosynthesis; (iv) spatial distributions of global ET, GPP, biomass and soil carbon; and (v) secondary forest rates of biomass accumulation in boreal, temperate and tropical forests.
Model evaluation highlighted a few discrepancies that warrant further investigation: (i) under-prediction of ET in tropical forests in Amazonia; (ii) over-prediction of GPP in SH temperate evergreen broadleaf forests; and (iii) under-prediction of large negative anomalies in the global land carbon sink corresponding to 1987–1988 and 1997–1998 El Niño events.
Further work on the model configuration presented here should include formal benchmarking in the International Land Model Benchmarking Project framework (Hoffman et al., 2017) and model–data fusion (Trudinger et al., 2016). The latter would aim to quantify data constraints on the regional and process attribution of the global land carbon sink using multiple parameter sets that are consistent with the observations, in the same way that Trudinger et al. (2016) did for the Australian region. Data for this task would comprise observation-based constraints presented in this work extended, for example, to include remotely sensed vegetation cover.
Priorities for further process enhancement are (i) wildfire impacts on vegetation and related emissions, (ii) explicit cropland management, (iii) dynamic biogeography and PFT interactions, and (iv) dynamic allocation of carbon that optimises plant fitness.
The source code for CABLE, including revision number 4601
(Haverd et al., 2017; Kowalczyk et al., 2006), can be accessed after
registration at
Part 1: CABLE biophysics. Part 2: CASA-CNP biogeochemistry. Part 3: POP and POPLUC components of CABLE.
Additional model updates include the following: (i) drought and summergreen phenology (Sitch et al., 2003; Sykes et al., 1996); (ii) low-temperature reductions in photosynthetic rates in boreal forests (Bergh et al., 1998); (iii) photoinhibition of leaf day respiration (Clark et al., 2011); and (iv) acclimation of autotrophic respiration (Atkin et al., 2016). These are described below.
Prior CABLE predicts phenology based on an annual climatology of remotely sensed vegetation cover. This precludes simulating the effects of interannual variations and trends in phenology on the terrestrial carbon and water cycles and land–atmosphere exchange. We addressed this deficiency by implementing drought and summergreen phenology following the LPJ model (Sitch et al., 2003), with extensions to account for chilling requirements of bud burst (Sykes et al., 1996).
Summergreen phenology applies to deciduous forest types (deciduous
needle-leaf and deciduous broadleaf; Table 1) and C
Raingreen phenology applies to C
For both summergreen and raingreen phenology, green-up translates to high allocation of NPP to leaves. Leaf turnover rate is set to zero outside of the senescence period, when turnover time is set to 4 weeks.
Three processes that contribute to the low-temperature reduction of
photosynthesis in boreal conifer forests are the following: (i) reduction caused by frozen
soils; and (ii) incomplete recovery of photosynthetic capacity during spring;
(iii) frost-induced autumn decline. The first effect is largely accounted for
in Prior CABLE, because soil moisture limitation on stomatal conductance
(Eq. 9) depends on liquid water content, meaning that soil freezing induces
soil moisture limitation. Our treatment of the other two processes follows
that of Bergh et al. (1998). The rate of post-winter recovery of
In Prior CABLE, the rate of leaf respiration at standard temperature is
assumed the same day and night. However, many studies have shown that at a
given temperature the rate of leaf respiration in daylight is less than that
in darkness (Brooks and Farquhar, 1985; Hoefnagel et al., 1998; Atkin et al.,
1998, 2000). To account for this, we implement the inhibition of leaf
respiration by light, as demonstrated by Brooks and Farquhar (1985),
implemented by Lloyd et al. (1995) and successfully tested in the JULES land
surface model for an Amazonian rainforest site by Mercado et al. (2007) and
globally by Clark et al. (2011). The light-dependent non-photorespiratory
leaf respiration (
Prior CABLE assumes a fixed PFT-dependent value of leaf respiration at
standard temperature (25
For consistency with Atkin et al. (2016), we adopt the “variable Q10”
instantaneous temperature response of
CABLE biophysics and CASA-CNP biogeochemistry parameters for
evergreen needle-leaf (ENL), evergreen broadleaf (EBL), deciduous needle-leaf
(DNL), deciduous broadleaf (DBL), shrub, C
Continued.
POP parameters.
Temperature response functions for photosynthesis.
VH and BS conceived, designed and performed the analysis. VH and LN coded the developments. LN, WW, CMT, MC and JGC contributed to the analysis. MC formulated the elasticity diagnostic. PRB collected and prepared the model inputs. VH drafted the paper, which all co-authors commented on and edited.
The authors declare that they have no conflict of interest.
Vanessa Haverd, Cathy M. Trudinger, Peter R. Briggs and Josep G. Canadell acknowledge support from the Earth Systems and Climate Change Hub funded by the Australian Government's National Environmental Science Program. The contributions of Benjamin Smith to this work were supported by the Strategic Research Area MERGE and by a CSIRO Distinguished Visiting Science grant. We thank Graham Farquhar for helpful discussions about the optimisation-based approach to simulating plant coordination of photosynthesis. Edited by: Philippe Peylin Reviewed by: two anonymous referees