Land use change (LUC) is among the main anthropogenic disturbances in the
global carbon cycle. Here we present the model developments in a global
dynamic vegetation model ORCHIDEE-MICT v8.4.2 for a more realistic
representation of LUC processes. First, we included gross land use change
(primarily shifting cultivation) and forest wood harvest in addition to net
land use change. Second, we included sub-grid evenly aged land cohorts to
represent secondary forests and to keep track of the transient stage of
agricultural lands since LUC. Combination of these two features allows the
simulation of shifting cultivation with a rotation length involving mainly
secondary forests instead of primary ones. Furthermore, a set of decision
rules regarding the land cohorts to be targeted in different LUC processes
have been implemented. Idealized site-scale simulation has been performed for
miombo woodlands in southern Africa assuming an annual land turnover rate of
5
Land use and land use change (LUC) strongly modifies the properties of the
Earth's surface, ecosystem services and the carbon and nutrient fluxes
between the land and the atmosphere. These activities have significant
impacts on the Earth's climate through both biogeochemical and biophysical
effects
(Foley
et al., 2005; Luyssaert et al., 2014; Mahmood et al., 2014). When a forest
is cleared, the majority of carbon stored in the above-ground biomass is lost
as
Globally, LUC activities have contributed significantly to historical
anthropogenic carbon emissions. It is estimated that about 800
In most global studies, only net transitions were accounted for in the LUC
processes simulated by DGVMs (Le Quéré et al.,
2015). Changes in land use over each model grid cell are diagnosed as the
difference in ground fractions of different land cover types between two
consecutive years. At a typical spatial resolution of 0.5
An overview of DGVMs with implemented gross land use change (shifting cultivation) and forest wood harvest.
More and more DGVMs started to include gross transitions and we provide an overview of them in Table 1. All models in Table 1 include shifting cultivation and wood harvest except that shifting cultivation is not included in ISAM, and five of them include sub-grid secondary land tiles when accounting for LUC. A recent review by Arneth et al. (2017) found that including processes that have been previously neglected in DGVMs, including gross transitions and other land management processes such as crop harvest and management, can lead to an upward shift of estimated LUC emissions. Their study thus highlights the importance of including these processes. Furthermore, to more robustly account for shifting cultivation and wood harvest, which often have a certain rotation length and mainly involve secondary forests of different ages, it is critical for DGVMs to include sub-grid differently aged land cohorts. This feature exists in some DGVMs that combine with a forest gap model (e.g. LPJ-GUESS; Bayer et al., 2017) but it would be difficult to represent forest species change because different tree plant functional types (PFTs) are mixed over a model grid cell. The same also applies for LM3V (Shevliakova et al., 2009). Other so-called area-based DGVMs (Smith et al., 2001) such as ISAM (Jain et al., 2013) and LPX-Bern 1.0 (Stocker et al., 2014) included secondary land tiles in the model but their capability to represent different rotation lengths in land use is limited. In the ORCHIDEE model, sub-grid forest cohorts have been recently included in the ORCHIDEE-CAN branch mainly for forest management purposes (Naudts et al., 2015), but a combination of both sub-grid land demography and gross land transition is still missing.
Here we present the new model developments in ORCHIDEE that combine both sub-grid land cohorts and gross LUC. The objectives of this study are (1) to document a new LUC module, including sub-grid vegetation cohorts, forest harvest, and gross LUC in the ORCHIDEE model, that can be run with and without sub-grid age dynamics; (2) to document through an idealized pixel simulation the simulated carbon fluxes from shifting cultivation or land turnover between model set-ups with and without sub-grid age dynamics; and (3) to document the model behaviour and forest age dynamics associated with the historical changes in LUC activities. Whereas the current paper focuses on documenting new model developments and subsequent changes in model behaviour, a companion paper presents a global reanalysis of historical LUC emissions (Yue et al., 2017).
The model version as the starting point for our development is ORCHIDEE-MICT (r3247), a branch of the ORCHIDEE DGVM (the major version is called the trunk version), the land surface component of the French IPSL Earth system model (ESM). ORCHIDEE can simulate the energy, water, and carbon fluxes between the land surface and the atmosphere. The carbon module simulates vegetation carbon cycle processes, including photosynthesis, photosynthate allocation, vegetation mortality and recruitment, phenology, litter fall, and soil carbon decomposition. ORCHIDEE-MICT is a branch initially focusing on improving high-latitude processes (e.g. soil freezing, snow processes, permafrost dynamics, and northern wetlands) but is now under development to include more processes. Of interest for this study is that the grassland management module developed in Chang et al. (2013) is included (r2615). This allows for distinction between natural grassland and pasture that have been mixed together in previous LUC simulations by ORCHIDEE.
In ORCHIDEE, land cover types are represented as PFTs, with each PFT being associated with a set of parameters. A typical
model simulation consists of two stages: a spin-up stage with stable or
constant forcing data until the model reaches an approximately equilibrium
state, to mimic an era with no appreciable human perturbation, and a
transient stage in which the model is forced with temporally varying forcings
(e.g. climate, atmospheric
Schematic illustration of gross versus net land use change, with each
land cover type being represented using a single patch within a model grid
cell. The figure is adapted from Stocker et al. (2014).
The numerical implementation of net transitions is straightforward. However, as explained in the introduction, this scheme omits important sub-grid gross land use transitions. Figure 1 uses an exemplary grid cell to illustrate the difference between the two LUC schemes: one accounting for net transitions only (Fig. 1b), and the other accounting for gross transitions but with no sub-grid cohorts (Fig. 1c, d). Although the areas of forest and cropland after LUC are identical (Fig. 1b, d), carbon stocks for the same vegetation type (e.g. forest) are different between the two schemes. According to the net transition scheme, the carbon stock of the final forest patch shown in Fig. 1b remains intact. But under the gross scheme (Fig. 1d), the post-LUC forest carbon stock is an area-weighted mean between the original forest patch not being impacted by LUC and the newly established forest with a low carbon density that results from cropland abandonment. Consequently the carbon stock of the grid cell is expected to be smaller in Fig. 1d than in 1b and LUC carbon emission in Fig. 1d is conversely larger than in Fig. 1b.
Figure 1c represents the real land cover state after LUC, while the merging shown in Fig. 1d is only a necessary simplification when no sub-grid cohorts are represented in the model. Ideally, the model capability could be expanded to include cohorts to represent the real world case as in Fig. 1c. In addition, inclusion of sub-grid cohorts would allow not only the distinction between original intact forest and newly established forest but also among different forest cohorts (e.g. primary versus secondary forests) regarding which forest patch to be cleared for cropland.
Gross land use change involving forests with different ages under a
model scheme capable of representing sub-grid land cohorts. The figure is
adapted from Stocker et al. (2014). LUC here is similar to in Fig. 1, except
that forest is no longer a single ageless patch but consists of two patches
of primary and secondary forests, i.e. having an age structure.
Figure 2 illustrates a case in which gross LUC is combined with sub-grid cohort representation in the model. Here, multiple patches within a grid cell are used to represent cohorts of a single vegetation type but with different ages since establishment. These cohorts often have different carbon stocks either due to different lengths in carbon accumulation time (e.g. for forest) or due to different extents to which legacy soil carbon is present (e.g. for croplands establishing on former forests). The areas subject to gross LUC transition in Fig. 2a and b remain the same as in Fig. 1a (dashed red rectangles), but primary and secondary forests are cleared in Fig. 2a and b, respectively. Thus, LUC emissions from clearing of primary forest are expected to be higher due to its higher biomass stock. Correspondingly, the legacy soil carbon stocks on the cohort of new cropland are also higher (shown in Fig. 2b and d).
Figures 1 and 2 have shown the example of LUC transitions between forest and cropland, but other types of LUCs, including forest harvest, can be handled in a similar way. In the case of forest harvest, having cohorts avoids the simplification of merging a young re-established forest after harvest with the original forest, which serves as the exact source of harvest. This can effectively simulate forest management practices that induce rotations of different forest cohorts (e.g. see McGrath et al., 2015, for a forest management history in Europe).
In order to simulate gross LUC combined with sub-grid vegetation cohorts as illustrated in Fig. 2, we expanded the ORCHIDEE-MICT capability to include sub-grid evenly aged cohorts. This necessitates multiple patches within a grid cell for a single PFT, which inherits most of the parameters from its parent PFT (they still belong to the same PFT and thus are largely physically similar). These patches are named cohort functional types (CFTs) here, to be distinguished from the original plant functional types. In this sense, the original PFTs actually become “meta-PFTs” which were named meta-classes (MTCs). As subsequent LUCs generate differently aged CFTs, the computational demand will be greatly increased. Hence, the number of CFTs within an MTC is limited to a user-defined number.
ORCHIDEE-trunk has a feature called “PFT externalization” that allows the creation of a new user-specified PFT by inheriting its parameters from an existing one. A user can then modify specific parameters at their convenience. Based on this feature, the ORCHIDEE-CAN branch (the svn revision number is 2566; Naudts et al., 2015, p. 2037) has developed representation of sub-grid forest age classes (i.e. equivalent to our CFTs here). Each forest age class is an inheritance of a given forest MTC. There, the transitions from one age class to another were defined by tree diameters. When a forest of a certain age class reaches its diameter limit, it moves into the next age class, and is merged with the existing forest patch of that age class if there is one. All associated biophysical and biogeochemical variables are merged as well following an area-weighted mean approach with a few exceptions for discrete variables such as the applied forest management strategy.
ORCHIDEE-MICT also inherits this externalization feature from ORCHIDEE-trunk. Here we ported the codes of forest age class functionality from ORCHIDEE-CAN to develop the CFT functionality needed for LUC simulation with cohorts in ORCHIDEE-MICT. The code base to include sub-grid forest cohorts was migrated from ORCHIDEE-CAN, with substantial adaptions being made in ORCHIDEE-MICT. Except for this, all other LUC developments have been achieved within the current study. Contrary to ORCHIDEE-CAN (see above), ORCHIDEE-MICT uses woody biomass to delimit different forest cohorts, with older cohorts having a higher woody biomass. Forest grows old by moving from the current cohort to the next one when the woody biomass exceeds the cohort upper boundary. Except for the cohort boundaries, no further cohort-specific parameterizations have been performed, so essentially all cohorts are governed by the same set of biophysical and ecological parameter values. However, in ORCHIDEE-MICT there are indeed some simple aging processes to proximate the key changes when a forest grows old: notably, the net primary production (NPP) allocation to below-ground sapwood decreases with the time since establishment.
In addition, we expanded the concept of CFT to croplands, natural grasslands, and pastures. Cohorts are defined with their soil carbon stocks for these herbaceous vegetation types; this is a definition relevant to LUC emission calculation. Because the directional change of soil carbon largely depends on the vegetation types before and after LUC and on climate conditions (Don et al., 2011; Poeplau et al., 2011), ideally agricultural cohorts from different origins should be differentiated. However, to avoid inflating the total number of cohorts and the associated computational demand, as a first attempt, we simply divide each herbaceous MTC into two broad sub-grid cohorts according to their soil carbon stocks and without considering their individual origins. We expect that such a parameterization can accommodate some typical LUC processes, such as the conversion of forest to cropland where soil carbon usually decreases with time, but not all LUC types (for instance, soil carbon stock increases when a forest is converted to a pasture). The biomass or soil carbon thresholds that delineate different CFTs must be properly parameterized in order to have sensible CFT segregation within different contexts of land use change. This will be further detailed in Sect. 2.2.3. In practice, for single-site simulations, the parameterization could be set up via a configuration file enumerating the thresholds for all CFTs. For regional applications, an input file containing spatially explicit thresholds will be used.
Two parallel hierarchies from the model parameterization and land use
change perspective.
The implementation of sub-grid cohort function types as inheritances of
meta-classes and the corresponding hierarchy is exhibited in Fig. 3a.
Tier 1 of the model parameterization hierarchy corresponds to the four basic vegetation types
(forest, natural grassland, pasture, and croplands, abbreviated as f, g, p,
and c respectively). Tier 2 corresponds to meta-classes in ORCHIDEE-MICT,
which contain one bare soil MTC and 14 vegetative MTCs, with each
vegetative MTC belonging to one of the four basic vegetation types. Tier 3
corresponds to CFTs. A CFT is noted
as CFT
With sub-grid cohorts, the model spin-up run is initiated with an input MTC map, essentially the same as in the case without sub-grid cohorts (recall that in Sect. 2.1.1 this MTC map is called a PFT map). But the difference is that the initial prescribed areas (as fractions of grid cell area) of different MTCs are all assigned to their youngest cohorts. During model spin-up forest woody mass will grow to exceed the thresholds of the first cohort, so that forests will move to the second cohort, and so on. At the end of spin-up, all forests thus end up in the oldest cohort of each MTC. The same case applies to herbaceous MTCs, given that cohort thresholds are properly defined (see more details in Sect. 2.2.3).
A set of implemented rules regarding cohort selection for different land use change processes.
Natural forest mortality in ORCHIDEE could be either prescribed as a constant rate or dynamically simulated, but in the case of prescribed vegetation cover, mortality takes effect by reducing the amount of existing biomass only, with the coverage of the concerned forest patch being unchanged. Likewise, recruitment increases forest individual density and updates leaf age and other relevant variables, but again, forest coverage remains unchanged. These features are necessary, as the original ORCHIDEE model does not take into account forest demography. As explained in Krinner et al. (2005, p. 8), recruitment sapling biomass is only incorporated when the existing biomass is virtually zero while a larger-than-zero ground coverage is prescribed. These features remain the same when sub-grid cohorts are used, i.e. forest mortality or recruitment does not modify forest cohort ground coverage. In addition, forest mortality and subsequent regeneration due to forest fires are handled in a similar manner. ORCHIDEE-MICT has integrated a prognostic fire module to simulate open grassland and forest fires arising from both natural and anthropogenic ignitions (Yue et al., 2014). Other forest disturbances, such as wind-throw, diseases, and insect outbreaks, are not explicitly considered in ORCHIDEE-MICT. Because of these reasons, after the spin-up, the only way to create secondary cohorts in the model is through LUC.
When entering transient simulations with LUC, younger cohorts will begin to be created. From a modelling perspective, the oldest cohorts in ORCHIDEE-MICT are somewhat equivalent to the primary lands (especially, the oldest forest cohorts are equivalent to primary forests), and other younger cohorts are analogue to secondary lands.
This section describes the implementation of gross LUC and forest harvest with sub-grid CFTs. We focus on the implementation with sub-grid cohorts because the same LUC process without cohorts could be simply treated as a particular case in which all MTCs have only one single cohort. The module interface is designed to receive forcing information on land area fluxes among four basic land cover types of forest (f), natural grassland (g), pasture (p), and cropland (c), taking into account the current LUC modelling landscape in DGVMs (as briefly reviewed in the introduction) and the availability of LUC reconstructions (e.g. Hurtt et al., 2011). The present developments are intended for the case in which changes in vegetation coverage are only driven by historical LUC activities and so there is no need to use the dynamic vegetation module of ORCHIDEE. This is different from the LUC implementation in JSBACH DGVM in Reick et al. (2013) in which a lot of effort has been devoted to reconciling the vegetation types in the forcing data (primary and secondary natural lands in the Land-Use Harmonization data set version 1 or LUH1 data) and the vegetation distributions simulated by the dynamic vegetation module of JSBACH. We focus on including sub-grid land cohorts in the model and implementing a set of hierarchical rules for which land cohorts are subjected to different LUC processes (Table 2). The allocation of natural lands into forest versus grasslands in the model, and the reconciliation of LUH1 land cover distribution and model PFT map, are instead handled by independent preparations of reconstructed historical land cover map time series.
In order to compare the simulation results from the gross LUC module with
the original net-transition-only LUC module, we separate the gross LUC areas
into two additive terms: net change equivalent to the original net
transition (prescribed by the matrix
Schematic representation of the new LUC scheme in ORCHIDEE-MICT v8.4.2 accounting for net land use change, land turnover, and forest harvest in combination with sub-grid cohort representation.
The key processes of the gross LUC module with CFTs are shown in Fig. 4,
comprising in total six steps. The LUC module is called at the first day of
each year. Input data are the three matrices.
As explained in Sect. 2.1.3, the construction of CFTs within the model follows the model parameterization hierarchy shown in Fig. 3a. The cohort age subjected to LUC is one of the most important considerations in LUC decisions, especially in the context of land turnover and forest harvest. This necessitates a re-organization of the CFTs to derive the LUC hierarchy shown in Fig. 3b, in which Tier 2 information is about areas of different cohorts of the same land cover type, and Tier 3 remains on the level of CFTs. Thus, Step 1 in the LUC module (Fig. 4) is to construct the LUC hierarchy, i.e. to calculate within the model the areas of each cohort for each vegetation type.
When implementing LUC matrices, all information of land transitions between
the four basic land cover types must first be downscaled on the cohort tier
(i.e. decision on which cohort is subjected to LUC) and then on the CFT
tier (i.e. how LUC-affected area is distributed among different comprising
meta-classes within each cohort; refer also to Fig. 3b). This is achieved in
Step 2 as shown in Fig. 4. Because all the newly established lands,
regardless of their originating LUC process, must belong to the youngest CFT
of the MTCs that comprise the target land cover type, the ultimate outcome
of Step 2 is a single (large) matrix
Step 3 handles forest wood collection (here “collection” rather than
“harvest” is used, to avoid the confusion with forest wood harvest, which is
a means of forest management), from forest being converted to other land
cover types, and forestry harvest (forest remaining forest). We assume that
a certain fraction of above-ground woody biomass (i.e. sapwood and
heartwood) is lost as instant
We now return to Step 2, explaining the different rules used to build the
Rules of selection of forest cohorts in secondary wood harvest to
account for the dynamics in harvest area over time.
Implementation of primary forest harvest is straightforward: we always start with the oldest cohort and move sequentially downwards to younger ones if older cohorts are exhausted until the prescribed harvest demand is fulfilled (Table 2). For secondary forest harvest, we start with intermediately aged cohorts. But if the existing area of intermediately aged cohorts is not sufficient to fulfill the prescribed harvest area, we are left with two options to either search upwards for older cohorts or downwards for younger ones. We decide to first search upward and then search downward, if all cohorts older than the intermediate age still cannot fulfill the prescribed harvest demand (Table 2). This rule allows potential temporal changes in harvested area to be accommodated, as explained in Fig. 5. Under such a scheme, (1) at the very beginning (after spin-up) and before the existence of any secondary forests, harvest will start with the oldest cohort, i.e. corresponding to harvest of primary forests (sometimes, because of the inconsistency between the input harvest information and existing forest cohort structure in the model, secondary forest harvest could be prescribed for pixels in which only primary forests exist in the model). (2) If harvest area of secondary forests remains stable, then as soon as sufficient intermediately aged cohorts are created via conversion of primary forest to regrowing younger cohorts, a corresponding stable rotation cycle would be maintained in the model as well. (3) If the harvest area increases, the upward searching would allow additional harvest of primary forests (i.e. area subject to the stable rotation is expanded). (4) If the harvest area decreases, moving cohorts from younger to older ones independent of any LUC activities would allow the restoration of older cohorts – e.g. a consequence of abandonment of forest management. (5) Finally, the downward searching for younger cohorts after exhausting all other older cohorts is solely to ensure the consistency between prescribed input harvest area and that actually realized in the model. Hence, this scheme is designed in order to faithfully implement the prescribed harvest areas in the model with an explicit consideration of forest successional states (i.e. primary or secondary). But when this is not possible because of inevitable mismatch between the model and forcing data, harvest areas of primary and secondary forests could mutually compensate for each other in the model to ensure that their prescribed total harvest area remains realized.
A number of studies reported that fallow lengths for shifting cultivation could range from a few years to more than 50 years depending on different regions, with the majority being 10–40 years (Bruun et al., 2006; Mertz et al., 2008; Thrupp et al., 1997; van Vliet et al., 2012), and there is a tendency in reduction of fallow lengths possibly because of increased population pressure (van Vliet et al., 2012). Hurtt et al. (2011) assumed a mean residence time of 15 years for shifting cultivation for tropical regions in the LUH1 reconstruction data. Based on these reports, we assume forest clearance for shifting cultivation to occur primarily in secondary forests and treat it similarly as secondary forest harvest when allocating the prescribed LUC area into different cohorts (Table 2). The only difference is that the destination land cover remains forest in the case of forest harvest but is agricultural land in the case of shifting cultivation. For all other land transfers in shifting cultivation (e.g. pasture to forest), we start exclusively from the oldest cohort and move downwards to younger ones (Table 2). For net LUC, priority is again given to older cohorts followed by younger ones (Table 2).
Fractions of above-ground woody biomass lost immediately to the atmosphere during a forest clearing and channelled to 10- and 100-year turnover wood product pools. These fractions are different depending on forest biomes.
Finally, we still need to downscale the LUC area in each cohort to its component CFTs. This is done by allocating the LUC area in each cohort to its member CFTs in proportion to the existing area of each CFT.
ORCHIDEE simulates two wood product pools with a turnover length of 10 years
and 100 years. Fractions of above-ground woody biomass as
instant on-site losses (
Other processes relevant to LUC are left unchanged with the original model
version. In particular, crop harvest is applied to cropland CFTs with a
fraction of 45
The land carbon balance simulated by ORCHIDEE-MICT v8.4.2 (i.e. net biome
production or NBP), when land use change is included, is defined as
The LUC emissions (
Instantaneous fluxes refer to the carbon emissions directly arising from
LUC, often occurring within the first year since LUC (
The model developments presented here enable us to make two parallel
simulations that include LUC: with and without sub-grid age dynamics. Their
simulated
We conducted an idealized grid cell simulation with prescribed land cover
and LUC matrices to compare in detail the simulated carbon pools and fluxes
between
Subsequently, the model behaviour has been documented for a real-world case over the region of southern Africa (south from the Equator of the African continent). All three LUC types occurred historically in this region, making it ideal to demonstrate model behaviour regarding forest cohort dynamics as presented in Fig. 5. This regional simulation serves a single purpose – to further exemplify model features that cannot be sufficiently demonstrated over a grid cell.
The regional simulation is performed at 2
Factorial simulations to examine forest cohort dynamics when
including different LUC processes: net land use change, land turnover, and
wood harvest. The plus signs (
Each forest MTC has six CFTs to represent six cohorts. The woody mass
thresholds are set in a way that they correspond roughly to the woody masses
at ages of 3, 9, 15, 30, 50 years, and the mature or primary forest (with an
age greater than 50 years) during the spin-up simulation for
Cohort
Biomass carbon stock as simulated by two model configurations without
(
Figure 6a and b exhibit the evolution of above- and below-ground biomass for
both
More differences emerge when entering the transient simulation. Above-ground
biomass in
While the above-ground biomass continuously grows during the spin-up, the
below-ground biomass first increases with time and then slightly declines
before reaching the equilibrium value. This is because ORCHIDEE-MICT has a
preferential allocation of NPP to below-ground sapwood when forests are
young. The small decline in below-ground biomass in the late spin-up stage
thus results from an almost stabilized NPP (under a big-leaf approximation),
a reduced below-ground allocation, and a constant mortality. Because of this
feature, ORCHIDEE-MICT creates a higher below-ground biomass in younger
forest cohorts (e.g. Cohort
As shown in Fig. 7a, in
Mean annual carbon flux differences between the LUC and control
simulations over 100 years for an annual forest–cropland turnover with 5
As defined in Eq. (4), the net LUC carbon emission (
First of all,
NPP is higher in LUC simulations than in the control. This is because young forests are established in the former case (either by merging with existing forest patch or not), leading to a younger leaf age than in the control simulation, which is parameterized to have a higher photosynthetic capacity than older leaves in the model. This suggests the model can somewhat integrate the effect of recovering young forests or intermediately aged forests with a higher productivity than the old-growth forests, as reported by Tang et al. (2014) using observation data.
Averaged over the LUC simulation period of 100 years, both
Areas subject to historical land use change and the resulting modelled
temporal changes in areas of different forest cohorts in southern Africa.
Decreases in fire carbon emissions (
Overall, the lower
One of the useful features of our model development is to account for
sub-grid forest age dynamics as a result of historical LUC, as
illustrated in Fig. 9 for southern Africa. When no LUC is
included (S0, the control simulation shown in light blue), the areas of all
forest cohorts are constant over time. Except that younger cohorts have a
very small area (
In the S2 simulation with both net LUC and land turnover, large
areas of younger forests, in particular of Cohort
DGVMs, either used in an offline mode or coupled with climate models, are powerful tools to investigate the role of past and future LUC in the global carbon cycle perturbed by human activities (Arneth et al., 2017; Le Quéré et al., 2016). Therefore, a more realistic representation of LUC processes in these models is a scientific priority. We included two new features in ORCHIDEE-MICT v8.4.2: gross LUC and forest wood harvest, and sub-grid vegetation cohorts. In a recent review (Prestele et al., 2017), proper representation of gross LUC or sub-grid bidirectional land turnover has been identified as one of the three major challenges in implementing LUC in DGVMs for credible climate assessments, despite that these have already been pioneered by some models (Table 1). Large underestimation of LUC emissions would occur when gross LUC is ignored, as is shown by several model results reviewed in Arneth et al. (2017).
Shifting cultivation, or forest wood harvest, or more forest management in general, often involves a stable fallow length or rotation cycle, which involves secondary forests rather than primary ones. In tropical regions, fallow lengths in shifting cultivation range from 10 to 40 years (Bruun et al., 2006; Mertz et al., 2008; Thrupp et al., 1997; van Vliet et al., 2012), with a tendency of reduction in fallow length. In Latin American tropics, agricultural abandonment has already led to prominent growth of secondary forests (Chazdon et al., 2016; Poorter et al., 2016). Forest management, including wood harvest, is more common in temperate and boreal regions. In European forests, rotation lengths depend on tree species, regional climate, and management purposes, ranging from 8 to 20 years in coppicing systems in southern Europe to 80–120 years in northern countries (McGrath et al., 2015). The prevalence of secondary forests associated with land use and LUC therefore calls for their representation in DGVMs, especially when modelling LUC.
To our knowledge, Shevliakova et al. (2009) performed the first study to include both sub-grid secondary lands and gross transitions in the LM3V model, but the number of PFTs and secondary land tiles are limited in their study (up to in total 12 secondary land tiles compared with 50 in our study). Stocker et al. (2014) included secondary land in LPX-Bern 1.0 but only one tile of secondary land is available. Yang et al. (2010) examined the contribution of secondary forests to terrestrial carbon uptake using the ISAM model by explicitly including secondary forest PFTs, but they did not include the dynamic clearing of secondary forests nor shifting cultivation in LUC. Therefore, none of these studies have included a dynamic decision rule regarding the ages of cohorts to be targeted in different LUC processes or the possibility of targeting different cohort ages in different geographical regions. ORCHIDEE-CAN is especially designed to address forest management and species change. Although certain LUC such as wood harvest and net land cover changes are included, a more comprehensive LUC scheme addressing gross change is missing (Naudts et al., 2015).
The gross LUC combined with sub-grid cohorts presented here has
shown some promising results. We first confirmed that including gross LUC leads to additional carbon emissions. However, these additional
emissions tend to be overestimated when secondary forests are not explicitly
accounted for. The idealized grid cell simulation explained the
mechanism driving such overestimation in
As a preliminary effort to demonstrate the model behaviour, the land
turnover parameterization is heavily tied with the input LUC forcing data
(LUH1), so that the age of Cohort
In the following paragraphs we will discuss the decisions that were marked as deliberate and their potential impacts on modelled LUC stocks and fluxes. First, the LUC module developed is intended for usage within DGVMs and forced with external data sets that provide information on land flows between different land cover types. It is not intended to supersede a LUC model per se, which simulates LUC using other available social and economic information such as population, food demand, wood demand, etc. (Hurtt et al., 2011). In this sense, the LUC module implementation has to inevitably take into account the details of information in forcing data that are available and to reconcile the potential mismatch between the model and forcing data. For example, the LUC module presented here can accommodate forest wood harvest from primary and secondary forests when these two sources are distinguished in the forcing data, but hierarchical decision rules are also made when the model and forcing data disagree (e.g. Fig. 5), for example, when prescribed secondary forest wood harvest can actually harvest a primary forest in the model if all younger cohorts are exhausted.
Second, because of this clearly defined border of the LUC module to use land
areas as the input information, model output from ORCHIDEE-MICT can
potentially disagree with the socio-economic information used to generate
the LUC forcing data. For instance, crop yield simulated by ORCHIDEE may
differ with that used to convert food demand and consumption to cropland area,
so that simulated crop output or food production may disagree with
historical food demand in the real world. The same applies to forestry wood
production: simulated harvest wood volume might disagree with the wood
volume actually used to generate the harvest area information – the
harvested wood biomass information is provided in the LUH1 data set but not used
as an input in the current stage of model development. This largely raises
the issue of to what extent the information that drives LUC
decisions can be internally integrated into DGVMs, for example, to directly use crop
production, rather than cropland area, or wood volume, rather than forest
harvest area as the model input. One potential obstacle is that statistical
information (e.g. on wood volume demand) is often available on a regional
basis (FAO global forest resource assessment,
The developments presented here mainly build on a model structure that distinguishes differently aged cohorts. Nonetheless, we have built a better tool to address the impacts of historical LUC on carbon cycle and climate with these developments. Forest demographics, which are shown to have great impact on the current Northern Hemisphere carbon sink (Pan et al., 2011; Piao et al., 2009b), either as a result of active afforestation, agricultural abandonment, or natural regeneration, could then be explicitly investigated. These developments also make it possible to verify modelled global and regional forest age distribution using independent age information from either forest inventory or remote sensing. The model version used here has incorporated the developments in pasture and cropland modules (Chang et al., 2015; Wang et al., 2017). On a regional scale such as Europe, where the comprehensive forcing data are available, it is possible to go beyond the carbon emissions only by LUC activities, but also to include LUC-induced changes in emissions of other greenhouse gases such as methane and nitrogen oxide.
We have presented new developments made in a global vegetation model to include gross LUC and forest wood harvest, in combination with explicit representation of sub-grid forest age dynamics. Furthermore, a set of decision-making rules regarding the land cohorts to be targeted in different LUC processes have been implemented. The presented simulation results are specific of the ORCHIDEE-MICT model, but the methods are generic for other DGVMs. We demonstrated through an idealized pixel simulation that gross LUC leads to additional emissions but accounting for sub-grid land cohorts yields lower emissions than not. Over the region of southern Africa, the model is able to account for changes in different forest cohort areas along with the temporal changes in different LUC processes, including regrowth of old forests when LUC area decreases. Our developments provide the possibility of accounting for forest demography when evaluating LUC impacts on global carbon cycle and climate.
The source code for ORCHIDEE-MICT version 8.4.2 is available online
(
Primary data and scripts used in the analysis and other supplementary information that may be useful in reproducing the authors' work can be obtained by contacting the corresponding author.
The authors declare that they have no conflict of interest.
Chao Yue and Wei Li acknowledge the European Commission-funded project LUC4C (no. 603542). Philippe Ciais acknowledges the support from the European Research Council through Synergy grant ERC-2013-SyG-610028 “IMBALANCE-P”. We thank Benjamin Stocker, Julia E. M. S. Nabel and the anonymous reviewer for their valuable comments in the review process that helped improve the quality of the paper. Edited by: Min-Hui Lo Reviewed by: Benjamin Stocker and two anonymous referees