Global Biogeochemical Cycles
Relevance of methodological choices for accounting of land
use change carbon fluxes
Eberhard Hansis
1
, Steven J. Davis
2
, and Julia Pongratz
3
1
Hamburg, Germany,
2
Department of Earth System Science, University of California, Irvine, Croul Hall Irvine,
California, USA,
3
Max Planck Institute for Meteorology, Hamburg, Germany
Abstract Accounting for carbon fluxes from land use and land cover change (LULCC) generally requires
choosing from multiple options of how to attribute the fluxes to regions and to LULCC activities. Applying
a newly developed and spatially explicit bookkeeping model BLUE (bookkeeping of land use emissions),
we quantify LULCC fluxes and attribute them to land use activities and countries by a range of different
accounting methods. We present results with respect to a Kyoto Protocol-like “commitment” accounting
period, using land use emissions of 2008–2012 as an example scenario. We assess the effect of accounting
methods that vary (1) the temporal evolution of carbon stocks, (2) the state of the carbon stocks at the
beginning of the period, (3) the temporal attribution of carbon fluxes during the period, and (4) treatment of
LULCC fluxes that occurred prior to the beginning of the period. We show that the methodological choices
result in grossly different estimates of carbon fluxes for the different attribution definitions.
1. Introduction
Emissions from land use and land cover change (LULCC) have contributed about one third of cumulative
anthropogenic CO
2
emissions [Houghton, 2003] and represent about 10% of current annual CO
2
emissions [Le
Quéré et al., 2014]. This flux of emissions is usually called “net LULCC flux,” because it consists of both source
terms, e.g., biomass burnt or being decomposed after clearing of natural vegetation, and sink terms, e.g.,
regrowth of forest when agricultural land is abandoned.
The nonnegligible size of the net LULCC flux and the corresponding potential for emission reduction and miti-
gation make investigating its causes relevant. The net LULCC flux may be attributed to specific LULCC activities,
such as clearing and wood harvest; to their underlying physical processes, such as soil decomposition and
regrowth; and to the region where the LULCC takes place. The geography of LULCC emissions is relevant for
policies aimed at avoiding the emissions. Under the Kyoto Protocol and subsequent UNFCCC (United Nations
Framework Convention on Climate Change) decisions, emissions related to land use, land use change, and
forestry are included in evaluating the Annex I Parties’ commitments [e.g., Birdsey et al., 2001].
However, such attribution requires model simulations because the net LULCC flux is not directly observable.
The net exchange between atmosphere and land can be inferred as residual from carbon stocks and fluxes
of atmosphere and ocean. However, models are needed to split the net exchange into natural sinks (and
sources) on the one hand and the net LULCC flux on the other hand. As LULCC, unlike the burning of fossil
fuels, may result in carbon fluxes that occur over many years after the LULCC event, analysts must make vari-
ous methodological choices of how to simulate these delayed fluxes and how to attribute them in time. In the
present study, we apply our model “Bookkeeping of Land Use Emissions” (BLUE), which extends the widely
used bookkeeping approach of estimating carbon fluxes, to illustrate and quantify the relevance of different
methodological choices in attribution of the net LULCC flux.
An LULCC activity leads to delayed carbon fluxes because it usually alters the relationship between CO
2
taken
up by photosynthesis (the net primary production) and decomposition of organic carbon in litter, soils, and
product pools. Both uptake and release act on various timescales, leading to complex temporal patterns of
carbon fluxes following a given LULCC activity, which may be modified further by subsequent LULCC events.
Multiple approaches exist to model the net LULCC flux. The simplest approach ignores temporal dynamics of
delayed processes and assumes that carbon stocks before and after an LULCC event are at equilibrium. Under
this approach, the net LULCC flux can be simply derived from information on carbon stocks of each land use
RESEARCH ARTICLE
10.1002/2014GB004997
Special Section:
Global Land-Use Change and
Carbon/Climate Dynamics
Key Points:
• We use a spatially explicit model to
attribute C fluxes to land use activities
• We compare several accounting
options for a post-Kyoto climate
agreement
• The different choices result in grossly
different estimates of carbon fluxes
Supporting Information:
• Table S1
• Text S1, Figures S1 and S2,
and Table S1 caption
Correspondence to:
J. Pongratz,
julia.pongratz@mpimet.mpg.de
Citation:
Hansis, E., S. J. Davis, and
J. Pongratz (2015), Relevance
of methodological choices for
accounting of land use change
carbon fluxes, Global Bio-
geochem. Cycles, 29, 1230–1246,
doi:10.1002/2014GB004997.
Received 1 OCT 2014
Accepted 20 JUL 2015
Accepted article online 21 JUL 2015
Published online 24 AUG 2015
©2015. American Geophysical Union.
All Rights Reserved.
HANSIS ET AL. ACCOUNTING OF LAND USE CARBON FLUXES 1230
Global Biogeochemical Cycles 10.1002/2014GB004997
state and the change in area. This approach is most commonly used in combination with remote sensing
data [e.g., Fearnside, 1997; Harris et al., 2012] and reflects a form of “committed flux,” which attributes both
instantaneous and delayed emissions related to a specific LULCC event to the time when the event occurred.
As an alternative to attributing the difference in equilibrium carbon fluxes to the time when the LULCC event
occurs, the fluxes can also be spread uniformly over some time period. Such distribution is also conceptually
simple (it only introduces one additional parameter, the choice of time horizon) but may be advantageous
in the case of LULCC types that can be anticipated to succeed each other [Davis et al., 2014]. If, for example,
forest is cleared for a certain type of cultivation (e.g., soybean) that is later transformed to another type (e.g.,
wheat), such a uniform distribution over time allows carbon fluxes to be attributed to both crops. Analysts
could thus conceivably distribute emissions in time according to whether and to what extent the successive
uses are foreseeable or intended by the parties who are clearing land [Davis et al., 2014].
A physically more accurate representation of the distribution of delayed carbon fluxes in time can be produced
by process-based or bookkeeping models. Process-based models, such as dynamic global vegetation mod-
els, simulate carbon stocks and fluxes as a result of photosynthetic and decomposition processes interacting
with environmental conditions. However, the current generation of dynamic global vegetation models does
not allow for attributing the resulting carbon fluxes to an individual LULCC event because of computational
constraints. Fulfilling this task requires bookkeeping models capable of tracking the area and type of LULCC
and combining these with empirical response curves [e.g., Houghton et al., 1983; Reick et al., 2010; Gasser and
Ciais, 2013]. We call a spatially and temporally explicit modeling of carbon stocks, which also accounts for the
succession of LULCC events, the “legacy scheme.” See section 4 for details on how the succession of LULCC
events leads to redistribution of carbon fluxes between land uses.
The above-cited references illustrate two important choices when attributing carbon fluxes to specific LULCC
activities: first, whether the temporal evolution of carbon stocks is simulated (as in a legacy scheme) or not
(as when the difference of equilibrium states is assumed) and, second, to which point or period in time the
modeled carbon fluxes are attributed to: as they are simulated to occur in time, as committed fluxes at the
time of the LULCC event, or as committed fluxes distributed over time (e.g., a uniform distribution over a
given time span). Further choices emerge when only specific periods of LULCC are of interest. This becomes
particularly relevant in the context of the Kyoto Protocol and follow-up UNFCCC decisions [United Nations
Framework Convention on Climate Change (UNFCCC), 2011, 2012]. These require parties to count only LULCC
during the “commitment period” toward their debits or credits and thereby exclude any carbon fluxes from
LUCC events preceding that period. Investigating LULCC of a specific time period—whether an arbitrary
commitment period, the year 1850 (the typical start date of the simulations for the coupled model intercom-
parison project contributing to the reports of the Intergovernmental Panel on Climate Change), or even earlier
years in the history of LULCC—always requires a further choice on how to initialize carbon stocks. If earlier
LULCC is known, simulations can start earlier than the time period of interest to represent the actual state of
carbon stocks when entering the period; a simpler method is to assume that carbon stocks are in equilibrium
with the existing vegetation distribution [e.g., DeFries et al., 2002a].
This list of choices is not comprehensive but represent typical and, as we will show later, important choices that
have to be made when setting up a model for attribution studies. Only some of them have been discussed in
earlier studies, mostly on the issue of temporal attribution of fluxes: Fearnside [1997] has discussed the differ-
ence between committed and actual emissions, which was later quantified exemplarily for tropical emissions
by Ramankutty et al. [2007]. Davis et al. [2014] compared conceptually several ways of temporal attribution,
including committed, actual, and uniformly distributed fluxes. Ramankutty et al. [2007] further showed that
estimates of tropical emissions in the 1990s differ substantially when the beginning of the simulation (which
assumes equilibrium carbon stocks) is placed at 1961, 1981, or 1991. However, a consistent comparison of
the effects of these methodological choices on attributed emissions using the same model does not yet exist.
Our study fills this gap and shows for historical LULCC that the choices lead to vastly different results. We sim-
ulate carbon fluxes due to LULCC since the year 1500 but focus on the consequences of different choices in
attribution during a recent time period of 5 years (2008–2012), which is a typical time frame for a UNFCCC
commitment period [UNFCCC, 2011].
HANSIS ET AL. ACCOUNTING OF LAND USE CARBON FLUXES 1231
Global Biogeochemical Cycles 10.1002/2014GB004997
Figure 1. Illustration of the BLUE modeling scheme. (a) Data arrays and model features and (b) model cycle. See Text S1
for details on model implementation.
2. Methods
2.1. BLUE Model
The following is a brief description of the BLUE model. The modeling scheme is illustrated in Figure 1. A
detailed model documentation is included in Text S1 in the supporting information.
BLUE largely follows the bookkeeping approach as developed by Houghton et al. [1983] and Houghton [2003]
but adds several features required for our further analysis: the model is spatially explicit (as is the bookkeep-
ing approach by Reick et al. [2010], but not the regional model by G asser and Ciais [2013]). It further tracks
individual histories of successive LULCC events in each grid cell, including their interactions. Here, with its
approach of cumulating excess carbon pools by LULCC activity, as explained below and in Text S1, BLUE is
computationally much more efficient than the model by Reick et al. [2010], which tracks individual histories
by splitting each grid cell into individual plots of land. Unlike these previous approaches, BLUE is capable of
HANSIS ET AL. ACCOUNTING OF LAND USE CARBON FLUXES 1232
Global Biogeochemical Cycles 10.1002/2014GB004997
tracking the carbon fluxes caused by each year’s LULCC events through time (see “temporal accounting” later
in this section).
The model runs on a grid of half-degree cells. Land properties and LULCC transitions are prescribed per grid
cell. Typical of a bookkeeping model, BLUE tracks the areas undergoing transitions and the types of change
and combines these with empirical data on densities of soil and vegetation carbon stocks. Upon a transition,
carbon is transferred between pools (including product pools and atmosphere) with prescribed fractions.
Response curves for growth and decomposition track subsequent changes in carbon pools. A major model-
ing challenge is posed in that LULCC transitions generally affect only a fraction of each cell, such that different
plots of land within a cell undergo different transition sequences. Explicitly modeling the history of each sin-
gle plot of land would require a model resolution resolving the typical minimum plot size, which is on the
order of 1 ha [Bruun et al., 2006; Lojkaetal., 2011]. This is inefficient in terms of computation time and memory.
Instead, we adopted a modeling scheme with exponential response curves, which enables accurate represen-
tation of multiple LULCC histories within a grid cell using a finite number of carbon pools only. This is made
possible by the fact that annual changes from relaxation processes assuming exponential response curves are
directly proportional to the “excess” carbon present—the amount of carbon separating a pool’s state from
its equilibrium. Relaxation fluxes can therefore be computed accurately by accumulating changes in excess
pools resulting from successive LULCC events, without storing information about when these events hap-
pened. This makes modeling of separate plots of land within a grid cell unnecessary. The BLUE setup thus
accurately models legacy effects and process interdependencies, as illustrated further in Figure 2.
The model tracks carbon stocks in a number of discrete “pools” for each combination of (1) cover type (pri-
mary land, secondary land, crop and pasture), (2) transition type (i.e., LULCC activity: harvest, clearing to crop,
clearing to pasture, and abandonment), (3) pool type (vegetation biomass, soil carbon with rapid or slow
relaxation processes, product pools with 1, 10, and 100 year life times, and atmosphere, i.e., emissions), and
(4) plant functional type (11 plant functional types are distinguished, see Table S1).
Emissions from each combination of cover type, transition type, pool type, and plant functional type can be
extracted separately. For an LULCC transition, the affected cover types and plant functional types are pre-
scribed. However, land of the respective cover type and plant functional type within a grid cell may have
undergone different LULCC sequences in the past, and the LULCC transition data set used here [Hurtt et al.,
2011] does not specify the history of subgrid cell areas to which new transitions should be applied. The BLUE
modeling approach corresponds to distributing each new LULCC event proportionally by area across the
different histories present in the cell.
Historical attribution studies require attributing carbon fluxes to specific transition years. For this purpose,
we added a “temporal accounting” layer to the model that tracks on a per-country basis (spatially explicit
temporal accounting would be too memory intensive) the contribution of each past year’s LULCC events to
the current year’s carbon fluxes. The excess carbon caused by LULCC events of a year is stored for each of
the carbon pools defined above, per country. Resulting carbon fluxes in subsequent years are computed
by applying exponential response curves. Consecutive LULCC events are also accounted for in the temporal
accounting—see Text S1 for details.
Figure 2a shows model output from an exemplary single-point run: a pixel of 0.5
∘
×
0.5
∘
, located at 50
∘
N,
10
∘
E and with potential vegetation of temperate/boreal deciduous broadleaf forest, is initially covered by
secondary forest. In model year 10, the land is harvested (remaining secondary land); in year 25, it is cleared
with a transition to crop; in year 40, it is abandoned back to secondary land. Each LULCC event affects the entire
grid cell. The figure shows the progress of the different carbon pools, as well as emissions attributed to each
of the LULCC events. In addition to the modeled legacy emissions, a second set of curves shows emissions for
scenarios disregarding LULCC events before or after each of the three events (Figures 2b–2d).
The harvest event depletes vegetation biomass (green curve) and reduces the “slow” soil pool (red curve)
stock, while depositing large amounts of dead vegetation and soil biomass in the “rapid” soil pool (pink curve)
and adding to the product pools (turquoise curve). Each pool relaxes toward its equilibrium state in the fol-
lowing period. Rapid release of carbon from the soil and product pools causes emissions into the atmosphere
(blue curve), which are only partially countered by uptake as biomass regrows and the slow soil pool recov-
ers. The subsequent clearing to crop again lowers carbon in the biomass and slow soil pools, while increasing
carbon in the rapid soil and product pools. The clearing event results in biomass and slow soil pools that
HANSIS ET AL. ACCOUNTING OF LAND USE CARBON FLUXES 1233
Global Biogeochemical Cycles 10.1002/2014GB004997
Figure 2. (a) Model output from an exemplary single-point run. Land cover types and transition events are noted at the
figure top. Depicted are carbon pool stocks for vegetation biomass (green curve), slow-process soil pool (red curve),
rapid-process soil pool (pink curve), product pool (turquoise curve; for display purposes, all three product pools are
combined in one curve), and accumulated emissions to the atmosphere (blue curve), plotted over simulation time.
(b–d) Annual carbon fluxes (green) and cumulative carbon fluxes (yellow) attributed to each of the three LULCC events,
shown for the legacy modeling scheme (Legacy) and for simulations with realistic temporal evolution of carbon stocks
but disregarding changes in carbon stocks from LULCC events preceding and following the respective transition (No
Past, No Future). The latter set of curves is shown for reference. Dots (for Legacy) and squares (No Past, No Future) on the
right vertical axes denote cumulative carbon fluxes reached when running the model until infinity (without additional
LULCC events occurring). See text for further details.
are already close to the equilibrium values of cropland, so that subsequent relaxation results in small carbon
stock changes.
Inspecting the cumulative carbon fluxes (yellow curves) attributed to the harvest event in Figure 2b, one sees
a clear difference between a legacy accounting scheme (solid yellow curve) and one disregarding both past
and future LULCC events (dotted yellow curve). In the legacy scheme, the clearing event prevents relaxation
of biomass and slow soil carbon pools after harvest to equilibrium values of secondary land, leaving cumula-
tive fluxes attributed to the harvest event. Annual carbon fluxes (green curves) change from carbon uptake to
a release at the time of the clearing event: in BLUE, carbon fluxes from the rapid soil and product pools con-
tinue to be accounted toward the harvest event but are no longer counteracted by uptake in biomass and
slow soil carbon. When disregarding the clearing event, cumulative emissions for the harvest event return
to zero as the secondary land completely recovers. Correspondingly, carbon fluxes attributed to the clearing
event, Figure 2c, are smaller in a legacy scheme taking into account the reduced biomass and slow soil carbon
HANSIS ET AL. ACCOUNTING OF LAND USE CARBON FLUXES 1234
![Table 2. Comparison to Other Recent Studies (Modified From [Schneck et al., 2015])a](/figures/table-2-comparison-to-other-recent-studies-modified-from-31ni6w3m.webp)
![Figure 6. Comparison of vegetation and soil carbon stocks simulated by BLUE to those simulated by Coupled Model Intercomparison Project 5 (CMIP5) models and reference data sets [taken from Anav et al., 2013]. Default values of BLUE for carbon densities are based on Houghton et al. [1983], and lower carbon densities based on Reick et al. [2010]. Stocks refer to present day (CMIP5 and BLUE output is averaged over 1986–2005); additionally, initial carbon stocks for BLUE are shown.](/figures/figure-6-comparison-of-vegetation-and-soil-carbon-stocks-1eyqif30.webp)
![Figure 5. Comparison of the net LULCC flux simulated by BLUE to other recent studies; only studies including gross transitions and wood harvest are included. A 5 year running mean is applied. The studies by Gasser and Ciais [2013], Gasser (personal communication), and Houghton (personal communication) apply bookkeeping models on the regional level, and those by Shevliakova et al. [2013], Stocker et al. [2014], andWilkenskjeld et al. [2014] process-based models. The study by Gasser and Ciais [2013] applied the data set by Hurtt et al. [2011] but scaled transitions down to the regional values by Houghton et al. [2012], while the original data set was kept in the estimate by Gasser (personal communication).](/figures/figure-5-comparison-of-the-net-lulcc-flux-simulated-by-blue-9rh7kwoc.webp)
