South African Journal of Botany 116 (2018) 86–95

Contents lists available at ScienceDirect

South African Journal of Botany

journal homepage: www.elsevier.com/locate/sajb

How vulnerable are ecosystems in the Limpopo province to

climate change?
S. Scheiter a, * , C. Gaillard a , C. Martens b , B.F.N. Erasmus c , M. Pfeiffer a
a
Senckenberg Biodiversity and Climate Research Centre (SBiK-F), Senckenberganlage 25, Frankfurt am Main 60325, Germany
b
Institute of Physical Geography, Goethe University Frankfurt am Main, Altenhoeferallee 1, Frankfurt am Main 60438, Germany
c
Global Change Institute, University of the Witwatersrand, 1 Jan Smuts Ave, Braamfontein, Johannesburg, Gauteng 2050, South Africa

A R T I C L E I N F O A B S T R A C T

Article history: South Africa’s biomes are characterized by their exceptional biodiversity and they provide important ecosys-
Received 2 September 2017 tem services such as food, livestock production, medical plants or fuel wood to people. However, during
Received in revised form 31 January 2018 recent decades, vegetation in South Africa experienced substantial changes and loss of biodiversity due to
Accepted 13 February 2018 habitat loss, intensification of land use and climate change. The development of sustainable management
Available online 20 March 2018
policies requires an understanding of interactions between vegetation, climate change as well as land use
Edited by T Kraaij
and an identification of the areas most vulnerable to vegetation change. Here, we use the aDGVM, a dynamic
vegetation model for tropical ecosystems, to investigate the risk of biome shifts in South Africa’s Limpopo
province under a set of IPCC climate change trajectories. The Limpopo province exemplifies an area highly
Keywords:
susceptible to climate and land use change, where people in rural areas heavily rely on natural resources.
Limpopo province
aDGVM We found a general trend towards more tree-dominated ecosystems and a particularly high risk of vegeta-
Dynamic vegetation model tion shift in more open grassland and savanna areas. The rate of biome shift is strongly linked to the IPCC
Savanna scenario applied with the highest risk of biome shifts in the RCP 8.5 scenario. We conclude that, irrespective
Biome shift of future climate trajectories, management and conservation initiatives should particularly focus on these
Climate change more open grassland and savanna ecosystems.
CO2 fertilization © 2018 SAAB. Published by Elsevier B.V. All rights reserved.

1. Introduction to livestock grazing, fuel wood collection and conversion of (semi-)

During recent decades, South African ecosystems experienced
shifts in vegetation and loss of biodiversity (Biggs et al., 2008; Chown,
2010). These transformations often have been attributed to climate
change, elevated CO2 , land use change and habitat loss. Tempera-
tures in South Africa have increased since the 1950s (Kruger and
Shongwe, 2004; DEA, 2011; MacKellar et al., 2014) and climate
model simulations conducted for IPCC assessment reports predict
further changes in the climate system for the next decades (IPCC,
2013, 2014a,b). Projected increases in temperature and extreme
events such as heat waves, droughts and El Niño events amplify
stress on vegetation and may threaten biodiversity (Ogutu and
Owen-Smith, 2003; Mooney et al., 2009; Archer et al., 2017). Ele-
vated CO2 can serve as fertilizer for vegetation growth (Wigley et
al., 2010), and Buitenwerf et al. (2012) suggest that elevated CO2
had a significant impact on savanna vegetation in South Africa, in
particular as driver of woody encroachment. Land use impacts due
* Corresponding author.
E-mail address: simon.scheiter@senckenberg.de (S. Scheiter).
natural ecosystems into cropland are widespread in many regions of
South Africa and entail changes in vegetation structure and function-
ing that may lead to loss of biodiversity and soil erosion (Matsika
et al., 2013; Twine and Holdo, 2016). Land use may mediate the
magnitude of the CO2 fertilization effects on vegetation (Stevens et
al., 2016).
Alterations in vegetation distribution and biodiversity induced by
climate and land use change feed back on socio-ecological systems
and thereby affect people’s livelihoods. In South Africa, many peo-
ple in rural areas, and in particular in poor households, directly rely
on ecosystem services and goods provided by natural resources, such
as livestock production, wild food, medical plants or wood as energy
source for cooking and heating (Le Maitre et al., 2007). Yet, the
Millennium Assessment Report (Millennium Ecosystem Assessment,
2005) highlights that intensification of land use already reduced
flows of important ecosystem services in many regions globally, and
it also manifests locally (Coetzer-Hanack et al., 2016). South Africa’s
exceptional biodiversity, comprising three of the global biodiver-
sity hotspots and a high degree of endemism (Mittermeier et al.,
2005), is target of many conservation efforts (e.g., Reyers et al., 2007).
However, the development of management actions to preserve

https://doi.org/10.1016/j.sajb.2018.02.394
0254-6299/ © 2018 SAAB. Published by Elsevier B.V. All rights reserved.
 S. Scheiter et al. / South African Journal of Botany 116 (2018) 86–95
biodiversity and ecosystem services requires an understanding of
vegetation dynamics and tools to project trajectories of potential
future vegetation.
The Limpopo Province in the North-East of South Africa exem-
plifies anthropogenic threats to biodiversity and ecosystem ser-
vices (Reyers, 2004). In particular, rural smallholder farmers in the
region depend heavily on natural resources and are therefore prone
to environmental, economic and social impacts caused by climate
change and land use intensification (Twine et al., 2003; Gbetibouo,
2009). Large areas of the province are highly impacted by com-
mercial farming as well as by smallholder and subsistence farm-
ing (Lehohla, 2012). Its abundant agricultural resources make the
Limpopo province one of the country’s prime agricultural regions
with respect to production of livestock, fruits, vegetables, cereals
and tea. The Limpopo province is characterized by large gradients
in rainfall, temperature, soil quality and elevation, allowing for high
biodiversity. It hosts conservation areas and national parks, e.g.,
part of the Kruger National Park. Conservation initiatives resulted
in the designation of South Africa’s largest biosphere reserve, the
Vhembe Biosphere Reserve, in 2009 (Pool-Stanvliet, 2013). These
parks and nature reserves are not only important from a con-
servation point of view, but also from an economic perspective,
because tourism substantially contributes to South Africa’s economy
(Lehohla, 2015).
To understand the feed-backs between climate, vegetation and
land use in complex socio-ecological systems and to identify sustain-
able land use opportunities, it is necessary to assess the vulnerability
of vegetation to climate change and the risk of undesired biome
shifts. Dynamic global vegetation models (DGVMs, Prentice et al.,
2007) can help to analyze these interactions as they provide a
process-based representation of ecosystem dynamics and therefore
allow us to simulate how climate change may influence vegetation
patterns on large spatio-temporal scales. Published DGVM studies
for Africa typically simulated more tree biomass under future con-
ditions (e.g., Scheiter and Higgins, 2009; Higgins and Scheiter, 2012;
Sato and Ise, 2012). However, these studies focused on the continen-
tal scale, a single future scenario, and did not provide a likelihood
assessment for projected biome shifts.
Here, we use the aDGVM (adaptive Dynamic Global Vegetation
Model, Scheiter and Higgins, 2009) to investigate the likelihood of
regional-scale biome shifts until 2030, 2050 and 2100 under differ-
ent IPCC climate change projections (IPCC, 2013, 2014a,b) at a high
spatial resolution. The aDGVM is a particularly well-suited tool to
study biome shifts in the region of interest as it has been explic-
itly developed and tested for tropical ecosystems characterized by
C4 grasses and trees, such as grasslands or savannas. It integrates
important features of savanna dynamics such as grass-tree compe-
tition and fire impacts on vegetation structure (Higgins et al., 2000;
Baudena et al., 2015). Specifically, we ask (1) what are likely trajecto-
ries of future vegetation in the Limpopo Province? (2) How uncertain
are these projections of potential future vegetation? (3) Which areas
and biome types are at high risk of vegetation change?
2. Methods
2.1. The aDGVM
We used the aDGVM (adaptive Dynamic Global Vegetation
Model), a dynamic vegetation model designed for tropical grass-tree
ecosystems (for details see Scheiter and Higgins, 2009; Scheiter et
al., 2012). The aDGVM integrates plant physiological processes com-
monly implemented in DGVMs (Prentice et al., 2007) with additional
processes that allow plants to dynamically adjust leaf phenology
and carbon allocation to environmental conditions. The aDGVM is
individual-based and simulates state variables such as photosyn-
thetic rates, biomass or height of individual plants. This approach
87
represents impacts of herbivory (Scheiter and Higgins, 2012) and
fire (Scheiter and Higgins, 2009) on vegetation as a function of
individual plant height. Grasses are simulated as two types of super-
individuals to distinguish grasses growing beneath or between tree
canopies.
The aDGVM simulates four vegetation types: C3 grasses, C4
grasses, forest trees and savanna trees (Scheiter et al., 2012). Differ-
ences between C3 and C4 grasses are due to the distinctive physiolog-
ical characteristics of C3 and C4 photosynthesis. Savanna and forest
tree types are different with respect to fire tolerance and shade toler-
ance (Bond and Midgley, 2001; Ratnam et al., 2011). While the forest
tree type is shade-tolerant but fire-sensitive, the savanna tree type
is shade-intolerant but fire-resistant. Forest trees dominate in dense
communities and in the absence of fire while savanna trees dominate
in more open, fire-driven communities.
In the aDGVM, fire intensity is modeled as a function of fuel load,
fuel moisture and wind speed (Higgins et al., 2008). Fire spreads
when (1) an ignition occurs, (2) the fire intensity exceeds a thresh-
old value of 300 kJ m−1 s−1 , and (3) the likelihood for a fire to spread,
pfire , is exceeded. We use a constant value of pfire = 1% as previous
simulations show that this value ensures good agreement between
observed and simulated fire patterns (Scheiter and Higgins, 2009).
Ignition sequences, which indicate days with fire ignitions, are ran-
domly generated. This approach implies that fire ignitions and the
ignition probability are not directly linked to factors such as region,
climate seasonality, lightning strikes or other ignition sources. How-
ever, fire regimes in aDGMV’s fire model are indirectly determined
by climate, as climate influences biomass growth, fuel accumulation,
and fuel moisture. These variables are used to calculate fire intensity
(Higgins et al., 2008).
The proportion of aboveground grass biomass removed by fire is
a function of burn patchiness, which is calculated using fire inten-
sity (Williams et al., 1998). The response of trees to fire is a function
of tree type, tree height and fire intensity (Higgins et al., 2000).
Seedlings and juveniles of both savanna and forest trees are in the
flame zone and are damaged by each fire. Adult savanna trees are
largely fire resistant and get only damaged by intense fires. Adult
forest trees are damaged by each fire. The critical height defining if
a savanna tree is damaged by fire or if it is resistant is calculated
from plant height and fire intensity (Higgins et al., 2000; Scheiter
and Higgins, 2009). Grasses and damaged savanna trees can regrow
from root reserves after fire (Bond and Midgley, 2001) while forest
trees cannot regrow. In aDGVM, tree death following fire is indirect
and occurs when the carbon balance becomes negative, a factor that
increases a tree’s probability of mortality.
The performance of the aDGVM was evaluated in previous stud-
ies. Scheiter and Higgins (2009) and Scheiter et al. (2012) show
that the aDGVM can simulate the current distribution of vegeta-
tion in Africa better than alternative dynamic vegetation models.
Scheiter and Higgins (2009) demonstrate that the aDGVM can repli-
cate biomass observed in a long-term fire manipulation experiment
in the Kruger National Park (Experimental Burn Plots, Higgins et al.,
2007). Scheiter and Savadogo (2016) showed that a slightly adjusted
version of the aDGVM can reproduce vegetation dynamics observed
in a long-term experiment in Burkina Faso (Savadogo et al., 2008;
Savadogo et al., 2009). In the current study, we did not perform
a quantitative comparison between simulated and observed veg-
etation for several reasons: (1) the entire Limpopo province was
classified as savanna in a recent biome map (except small grass-
land patches Rutherford et al., 2006). (2) The Limpopo province is
strongly influenced by land use, which is reflected in remote sensing
products but not considered in our simulations; this implies biases
in data-model comparisons. (3) We conducted simulations at high
spatial resolution. Accurate model testing at this resolution would
require more detailed information on topography, soils and climate
than data sources we used for this study.
88 S. Scheiter et al. / South African Journal of Botany 116 (2018) 86–95
2.2. Study area
The Limpopo Province is located in the North-East of South Africa
bordering Botswana, Zimbabwe and Mozambique, and covers an
area of 12.5 million hectares. Rainfall in the Limpopo province cov-
ers a gradient from less than 200 mm in the Limpopo valley to more
than 1000 mm in mountainous areas and is typically highly seasonal
with rainfall in summer and dry periods in winter (New et al., 2002).
Mean annual temperature varies between approximately 18◦ C and
28◦ C and elevation is between ca. 200 m and ca. 2100 m (New et al.,
2002). Environmental gradients allow diverse vegetation types in the
Limpopo Province. Mucina and Rutherford (2006) found a high diver-
sity of different vegetation units and bioregions ranging from moist
grasslands to bushveld and savannas to forests.
Approximately 81% of the total area of the Limpopo province is
used for grazing and livestock, and 10.5% for agriculture. In total,
approximately 90% of the area of the Limpopo province is utilized,
both by commercial and by small farmers (Maluleke et al., 2016).
The remaining area is used for conservation, urban settlements, min-
ing and other land cover types. Only around 0.7% of the area are
covered by wetlands and indigenous forests. Human population in
the Limpopo province is currently 5.4 million people and growing
(Lehohla, 2012), resulting in a rapid expansion of rural settlements
and urbanization in some areas. This development requires policies
that arbitrate between different groups of interests such as sub-
sistence farmers, commercial farmers, mining industry or tourism.
Socio-economic problems persist in rural areas with high unemploy-
ment and Limpopo remains South Africa’s province with the lowest
average annual household income (StatisticsSouthAfrica, 2015).
2.3. Simulation design
We simulated vegetation in the Limpopo Province with the
aDGVM at 0.125◦ spatial resolution. To ensure that model variables
are in equilibrium with environmental conditions we first conducted
a 200-year model spin-up using monthly data from the CRU refer-
ence climatology for the period between 1961 and 1990 (New et al.,
2002). Previous simulations showed that a 200-year period is suffi-
cient to ensure that model variables stabilize and that the model is
well initialized. Following the spin-up period, the vegetation model
was forced with CO2 , temperature and precipitation trends from the
IPCC (2013, 2014a,b) RCP 2.6, 4.5 and 8.6 scenarios until 2100. For
each respective RCP scenario, we used the simulated climate output
data from MPI-ESM-LR, the Earth system model developed by the
Max Planck Institute for Meteorology in Hamburg (Giorgetta et al.,
2013), and calculated precipitation and rainfall anomalies relative to
the CRU reference climatology (New et al., 2002). The MPI-ESM-LR
was included in the Third National Communication to the UNFCCC,
and in conjunction with an RCM, was found to have a high degree
of agreement within an ensemble. Time series including the anoma-
lies were then used as climate forcing for aDGVM. We conducted 30
replicate simulations per RCP scenario to account for stochasticity
in the model, which for example occurs in sub-routines simulating
fire occurrence or demography of the tree population. By default, the
aDGVM simulates potential vegetation for given biotic and abiotic
conditions and we ignore impacts by animals or land use such as crop
production, plantations or pastures (but see Scheiter and Higgins
(2012), Scheiter et al. (2015) and Scheiter and Savadogo (2016) for
studies where animals, land use and management were considered).
2.4. Analysis of simulation results
In this study, we quantify the risk of biome shifts for South
Africa’s Limpopo Province until 2100. We used simulated tree
cover and grass biomass to classify simulated vegetation into five
biome types: C4 grasslands, C3 grasslands, savannas, woodland and
forests. Following the classification scheme presented in Scheiter
and Higgins (2012) (Table A.1), we classify vegetation as grassland
if tree cover is less than 10% and grass biomass exceeds 1.5 t/ha. In
C4 grasslands, C4 grass biomass exceeds C3 grass biomass and vice
versa. Vegetation is classified as savanna if tree cover is between 10%
and 80% and the cover of savanna trees is higher than the cover of
forest trees, irrespective of grass biomass. Simulated vegetation is
categorized as woodland if tree cover is between 10% and 80% and
the cover of savanna trees is lower than the cover of forest trees,
irrespective of grass biomass. We classify vegetation as forest if tree
cover exceeds 80%, irrespective of grass biomass and tree type. In this
scheme, savannas and woodlands can have similar tree cover and
grass biomass, but differences in the relative abundance of savanna
and forest trees reflect differences in fire activity. A higher abun-
dance of savanna trees indicates high fire activity, characteristic for
savanna systems. As C3 grassland only occurs with very low abun-
dance in two of the future scenarios, we decided to aggregate C3 and
C4 grasslands in our analysis.
Note that the definition of the thresholds used for the classifi-
cation influences simulated biome patterns and may explain fine-
scale mismatch between simulated and observed biome types. To
account for this uncertainty, we classified simulated vegetation with
different tree cover thresholds to explore how the classification
scheme influences biome patters. More specifically, we used val-
ues of 5%, 10% and 20% for the threshold between grassland and
savanna/woodland and values of 60%, 80% and 95% for the thresh-
old between savanna/woodland and forest and plotted maps for
all combinations of these values (see Fig. A.1). We decided to re-
use the classification scheme described in the previous paragraph,
because previous studies using this scheme at large spatial scales
showed good agreement between simulated and observed biome
types (Scheiter and Higgins, 2009; Scheiter et al., 2012), and to
ensure consistency with previous studies.
We generated maps of dominant biome types for selected years
(2016, 2030, 2050 and 2100). The dominant biome type of each
grid cell was obtained by assigning the biome type simulated in the
majority of the 30 replicate simulation runs. As a measure of model
robustness, we generated maps that show the proportion of replicate
runs that simulate the same dominant biome type within each grid
cell. A value close to one indicates high consistency between simu-
lation runs and therefore robust results, while a low value indicates
disagreement between replicate runs and more uncertainty in our
simulation results.
Maps were generated to indicate areas where biomes remain
stable between 2016 and future years (2030, 2050 and 2100, respec-
tively) and areas where aDGVM projects biome shifts. To create these
maps, we used the most frequent transition simulated in the 30 repli-
cate model runs in each grid cell. In addition to the maps, we used
the ‘circlize’ R package (Gu et al., 2014) to illustrate transitions of
biome types between different years and calculated transition matri-
ces indicating the percentages of the area of the Limpopo province
that remain in a stable biome state or shift between biome states.
To quantify the risk of biome shift for each individual grid cell in
the study area, we counted the number of simulation runs for which
the model projected biome shifts between 2016 and 2030, between
2016 and 2050 and between 2016 and 2100. We then divided these
counts by 30, the number of replicate simulation runs, to obtain the
risk as a proportion for each grid cell. We define four categories for
the risk of biome shift: values between 0 and 0.3 indicate low risk,
values between 0.3 and 0.6 medium risk, values between 0.6 and 0.9
indicate high risk, and values above 0.9 critical risk. We calculated
the area of the Limpopo Province covered by different risk categories
in 2030, 2050 and 2100.
To analyze the vulnerability of different biome types in the entire
Limpopo province to biome shifts, we calculated for each biome
type the absolute and relative number of grid cells that experience
 S. Scheiter et al. / South African Journal of Botany 116 (2018) 86–95 89

a biome shift. We used the number of simulated grid cells covered
by each biome type in 2016 as reference and calculated how many
cells of different biomes are projected to shift towards another biome
type until 2030, 2050 and 2100, respectively. We used results from
all replicate simulation runs to calculate these numbers. To obtain
relative numbers of change, we divided the absolute number of grid
cells that change by the total number of cells covered by a specific
biome in 2016.
Classification of simulated vegetation into different biomes
implies loss of information inherent to aDGVM model results. We
therefore created maps of mean tree cover and mean grass biomass
of the 30 replicate runs for 2016, 2030, 2050 and 2100. We calcu-
lated differences between 2016 and future years (2030, 2050 and
2100, respectively) to track changes in these variables. As a measure
of model uncertainty, we calculated standard deviation of tree cover
and grass biomass of all 30 replicate simulation runs for 2016, 2030,
2050 and 2100.
Note that we provide only figures for the RCP 4.5 scenario in
the main text; figures for RCP 2.6 and 8.5 are provided in the
Supplementary Materials.
3. Results
The aDGVM simulates that in 2016 most of the area of the
Limpopo Province could potentially be covered by savannas (63.8%
of the study area, Fig. 1, Table 1), with C4 grasslands in the North and
West (21.4%), forests in the East (4.8%) and woodland areas in the
South (10%). Simulated vegetation patterns correlate well with mean
annual precipitation (not shown), with grasslands in the more arid
regions, savannas in intermediate rainfall regions and forest in high
rainfall regions. This strong response to rainfall explains why aDGVM
does not simulate savanna in the entire Limpopo Province (Ruther-
ford et al., 2006). Agreement of the projected biome types between
replicate simulation runs varies in the study region (Fig. 1). It is the
highest in grassland areas in the North and savanna areas in the East
of the Limpopo province while it is lower in the western part. High
agreement indicates that biome types projected by replicate runs are
robust and that simulations are only weakly influenced by model
stochasticity.
Under future conditions, the aDGVM projects woody encroach-
ment, i.e. increases in tree cover (Figs. 2, A.3 and A.8) and associated
biome shifts (Figs. 1, A.2, A.7, Table 1). Savannas tend to expand
into grassland areas, and woodlands and forests are apt to spread
into savanna areas (Figs. 3, A.4 and A.9, Table 2). These patterns
are consistent for all RCP scenarios considered in this study, but
the woody encroachment effect in year 2100 becomes increasingly
pronounced from RCP 2.6 to RCP 4.5 and to RCP 8.5 (Table 1). For
instance, in the RCP 4.5 scenario, the area covered by grasslands
decreases from 21.4% in 2016 to less than 1% in 2100 while forested
areas increase from 4.8% to 24.6%. The area covered by savannas is

Biome type Biome shifts Agreement

2016 2016−2030 2016

2030 2016−2050 2030

xx xx xx

2050 2016−2100 2050

xx xx xx

Biome types and biome shifts Agreement

2100 C4 grassland/C4 grassland stable 0−20% 2100
xx xx
C4 grassland − savanna/woodland 20−40%
Savanna/savanna stable 40−60%
Savanna − forest 60−80%
Woodland/woodland stable 80−100%
Woodland − forest
Forest/forest stable
Forest − woodland

Fig. 1. Dominant biome types, biome shifts and agreement of projections for IPCC SRES 4.5 scenario. The dominant biome type (left column) is the biome simulated in most of the
30 replicate simulation runs. The biome shifts (middle column) indicate areas where different biomes are stable or where they may shift into another biome state. The maps show
biome shifts that occur in most of the 30 replicate simulation runs. Agreement (right column) represents the proportion of 30 replicate simulation runs that simulate the dominant
biome type. High numbers indicate high agreement between replicate simulation runs whereas low numbers indicate that replicate simulation runs diverge. See Supplementary
Figs. A.2 and A.7 for RCP 2.6 and RCP 8.5.
90 S. Scheiter et al. / South African Journal of Botany 116 (2018) 86–95

Table 1 areas of the study region (Figs. 1 and A.2), whereas it decreases in
Percentage of area covered by different biome types in different years and different many regions in the RCP 8.5 scenario (Fig. A.7). Tree cover stan-
future scenarios. See also Figs. 1, A.2 and A.7.
dard deviation of replicate runs generally decreases under future
Scenario Year Grassland Savanna Woodland Forest conditions (Figs. 2, A.3 and A.8).
2016 21.4 63.8 10.0 4.8 The risk of biome shifts increases between 2016 and 2100 with
RCP 2.6 2030 12.7 72.2 9.5 5.7 the highest risk in year 2100 in all scenarios (Table 3, Figs. 4, A.5 and
RCP 2.6 2050 6.8 77.0 7.9 8.3 A.10). Until 2030, the risk of biome shifts is similar in all three RCP
RCP 2.6 2100 6.3 77.6 9.3 6.8
scenarios. The model projects low risk for a large proportion of the
RCP 4.5 2030 12.7 72.5 9.6 5.2
RCP 4.5 2050 5.8 76.0 7.9 10.3 area, with a small number of grid cells being at medium risk. Until
RCP 4.5 2100 0.6 68.1 6.7 24.6 2050, the area exposed to medium risk expands and areas with high
RCP 8.5 2030 12.7 71.0 11.2 5.1 risk (or even critical risk, only in RCP 8.5) emerge. Until 2100, areas
RCP 8.5 2050 4.5 76.6 8.4 10.5 with high or critical risk expand and in the RCP 8.5 scenario, most of
RCP 8.5 2100 0.0 55.2 7.3 37.6
the Limpopo province is projected to be at medium, high or critical
risk of a biome shift in reference to the 2016 baseline.
For all scenarios and considered time periods, grasslands are
relatively stable and increases from 63.8% to 68.1% (Table 1). This
the biome type most likely to experience a shift, followed, in this
indicates that savannas are likely to experience spatial shifts until
order, by savannas, forests and woodlands (Figs. 5, A.6 and A.11). For
2100. Savannas spread into new areas at the arid biome boundary
example, in the RCP 4.5 scenario, the aDGVM projects that almost
while they are replaced by woodland and forest at the humid biome
all grasslands simulated in 2016 will shift to other biome types
boundary.
until 2100, whereas the proportion is between ca 30% and 60% for
We found that within-grid cell agreement of simulated biome
savannas, woodlands and forests.
states in replicate simulation runs changes under future conditions.
Agreement is typically the lowest within grid cells that experienced
a biome shift between 2016 and the considered year (2030, 2050 4. Discussion
or 2100, Figs. 1, A.2 and A.7). How agreement changes over time
also depends on the tested RCP scenario. In the RCP 2.6 and RCP 4.5 In this study, we show that large proportions of Limpopo’s
scenarios, within-grid cell agreement increases over time in large grasslands and savannas are susceptible to experience a cascade of

Tree cover Tree cover change Tree cover deviation

2016 2016−2030 2016

2030 2016−2050 2030

xx xx xx

2050 2016−2100 2050

xx xx xx

0% cover −100% change 0% deviation

2100 2100
xx xx

100% 100% 35%

Fig. 2. Tree cover, tree cover change and tree cover standard deviation for 2016, 2030, 2050 and 2100 for the RCP 4.5 scenario. Tree cover and tree cover change represent the
averages of all 30 replicate simulation runs, tree cover deviation represents standard deviation of all 30 replicate runs. See Supplementary Figs. A.3 and A.8 for RCP 2.6 and RCP
8.5, respectively.
 S. Scheiter et al. / South African Journal of Botany 116 (2018) 86–95 91

Fig. 3. Biome shifts between different time slices for the IPCC RCP 4.5 scenario. The origin of a shift is indicated by the larger distance between the outer circle and the line and by
the color of the respective biome type. The width of the lines represents the number of transitions. Note that these figures only represent biome shifts but not situations where
biomes are stable. Hence, width between panels is not comparable. See Table 2 and Supplementary Figs. A.4 and A.9 for RCP 2.6 and RCP 8.5.

biome shifts towards wood-dominated biome types, and that current area in the Limpopo Province decreased by almost 800,000 ha while
biomes may be lost until 2100. For grasslands, the risk of biome shift the areas covered by woodlands and open bush increased by more
is particularly high between 2016 and 2050 and levels off between than 800,000 ha.
2050 and 2100 because most grassland areas are already converted Biome shifts simulated by aDGVM can be explained by CO2 fer-
prior to this period. Until 2100, grasslands may almost disappear and tilization effects. In the model, elevated CO2 increases growth rates,
be replaced by savannas with a higher fraction of woody plants. The reproduction rates and water use efficiency of modeled trees, which
area covered by savannas is relatively stable until 2050 showing that triggers a shift in the competitive balance between grasses and trees
savanna areas will be prone to spatial shifts. Between 2050 and 2100, towards trees (Scheiter and Higgins, 2009; Higgins and Scheiter,
large savanna areas show a high probability of biome shift and they 2012). As a consequence, grasses and fire activity are suppressed
are replaced by woodland or forest. This trend agrees with Maluleke which further promotes tree growth (Higgins and Scheiter, 2012;
et al. (2016), showing that between 1996 and 2014 the grassland Hoffmann et al., 2012). Previous studies showed that aDGVM is
more sensitive to changes in CO2 than to changes in other climate
variables. A full factorial modeling experiment for Australian savan-
Table 2
nas with CO2 , temperature and precipitation kept at ambient levels
Transitions (in %) between different biome types in the RCP 2.6, 4.5 and 8.5 scenarios
between 2016 and 2100. Transitions correspond to the graphs in Figs. 3, A.4 and A.9. or changed according to the SRES A1B scenario showed that the effect
size of elevated CO2 was 0.97, while it was 0.17 and 0.11 for temper-
From/to Grassland Savanna Woodland Forest
ature and precipitation change, respectively (Beringer et al., 2015;
RCP 2.6 Scheiter et al., 2015). When ignoring CO2 increases until 2100 and
Grassland 26.6 70.1 3.2 0.1
hence CO2 fertilization effects, the aDGVM simulates increases or
Savanna 2.6 88.8 3.5 5.1
Woodland 1.9 32.2 51.7 14.2
decreases in woody biomass, depending on increasing or decreasing
Forest 0.0 9.3 42.8 47.9 rainfall and rainfall seasonality (Scheiter et al., 2015). We therefore
expect that simulation results for an ensemble of future projec-
RCP 4.5 tions simulated by different general circulation models (IPCC, 2013,
Grassland 3.8 88.2 5.5 2.5 2014a,b) would influence the absolute rates of simulated biome
Savanna 0.1 73.1 3.5 23.3 shifts but not the general pattern that we found in this study.
Woodland 0.1 19.6 44.7 35.7
Forest 0.0 6.8 31.8 61.5
Scenarios with extreme rainfall decreases may be an exception.
Although CO2 fertilization effects on vegetation are highly
RCP 8.5
debated (e.g. Körner et al., 2005) and the strength is ecosystem
Grassland 1.2 82.7 6.9 9.2 specific and related to absolute values of CO2 (Bond et al., 2003),
Savanna 0.2 53.1 3.2 43.5 empirical studies document the occurrence of woody encroachment
Woodland 0.3 11.9 41.6 46.3 in many southern African savanna ecosystems. Our model pro-
Forest 0.0 2.4 25.3 72.3
jections agree with this observed trend. Remote sensing products
92 S. Scheiter et al. / South African Journal of Botany 116 (2018) 86–95

Table 3
Percentage of area covered by different categories of risk of biome shift in different low
years and different scenarios. See also Figs. 4, A.5 and A.10. medium
Scenario Year Low Medium High Critical high
critical
RCP 2.6 2030 86.3 13.7 0 0

2030
RCP 2.6 2050 62.2 32.3 5.5 0
RCP 2.6 2100 52.0 33.9 13.7 0.4
RCP 4.5 2030 85.7 14.3 0 0
RCP 4.5 2050 58.2 32.3 9.5 0
RCP 4.5 2100 33.3 30.4 29.7 6.6
RCP 8.5 2030 87.6 12.2 0.2 0
RCP 8.5 2050 59.0 29.1 11.6 0.3
RCP 8.5 2100 13.2 29.6 34.8 22.4

(Donohue et al., 2013; Zhu et al., 2016; Skowno et al., 2017) and xx
empirical studies (Buitenwerf et al., 2012; O’Connor et al., 2014;
Stevens et al., 2017) show greening and woody encroachment in
savanna areas and suggest that the effect of elevated CO2 concentra-

2050
tions is a main driver. Kgope et al. (2010) showed in a greenhouse
experiment how root growth of two Acacia species that are typical
for South African savannas increases with CO2 . Midgley and Bond
(2015) highlight that to be able to understand future vegetation
dynamics of savannas, CO2 needs to be considered together with
climate and disturbances.
We calculated average tree cover changes in the Limpopo
province for the period between 2016 and 2030 and found an
increase of 0.39% per year for the RCP 4.5 and 8.5 scenarios and
0.36% per year for the RCP 2.6 scenario. These values are in the xx
range of observations. Stevens et al. (2017) report values between
0.13% per year in areas used for subsistence grazing, 0.4% per year in
conservation areas without elephants and an average woody cover
2100

change of 0.25% per year for African savannas. Yet, a direct compari-
son between observed and simulated rates of woody encroachment
is not possible with our model setup, because aDGVM was forced
with a climatology and anomalies representing RCP scenarios. A
quantitative comparison would require simulations with time series
of historic climate data and detailed knowledge and control of all
drivers influencing empirical studies such as climate, CO2 , herbivory
and land use.
The risk of critical biome shifts is the highest for the RCP 8.5
scenario because this scenario shows the most pronounced increase
in CO2 concentrations until 2100 and hence the strongest CO2 fer-
tilization effect. In this scenario, disagreement between replicate
simulation runs is the highest because CO2 is still in a transient
state between 2080 and 2100. Hence, vegetation is also transient and
replicate simulation runs disagree. In contrast, in the RCP 2.6 and
4.5 scenarios, CO2 stabilizes between 2080 and 2100. Consequently,
vegetation approaches an equilibrium in the RCP 2.6 and 4.5 sce-
narios between 2080 and 2100, resulting in an agreement between
results from replicate simulation runs. This model behavior indicates
that especially in the RCP 8.5 scenario, vegetation is not yet in equi-
librium with environmental conditions and therefore committed to
further change (Jones et al., 2009), irrespective of changes in climatic
conditions after 2100.
Simulated biome shifts are a consequence of changes in vegeta-
tion composition; biomes are not directly simulated by aDGVM, but
are derived from simulation results by a posteriori classification. For
instance, climate change and CO2 fertilization increase tree num-
bers and tree cover (Figs. 2, A.3 and A.8), modify grass biomass (Figs.
A.12, A.13 and A.14), and thereby fine fuel load, and fire activity
(Scheiter and Higgins, 2009; Higgins and Scheiter, 2012; Scheiter et
al., 2015). If these changes are large enough to let tree cover exceed
the threshold used to distinguish between savanna and forest in our
biome classification scheme, the biome type flips from savanna state
to forest state. However, areas that are, based on the classification
Fig. 4. Risk of biome shift between 2016 and 2030, 2050 or 2100, respectively for
the IPCC RCP 4.5 scenario. The categories indicate the percentage of the 30 replicate
simulation runs where biome shifts are projected: low - between 0 and 30% of the
simulation runs project biome shift; medium - between 30 and 60%, high - between
60 and 90%; critical - between 90 and 100%. See Table 3 for areas covered by different
risk categories and Supplementary Figs. A.5 and A.10 for RCP 2.6 and RCP 8.5.
thresholds, not subject to biome shifts can still experience substan-
tial modifications of vegetation such as increases in woody biomass
or woody cover (Figs. 2, A.3 and A.8). Therefore, depending on the
variables and thresholds of applied biome classification schemes,
changes in vegetation composition and structure may be hidden if
they do not translate into a biome shift.
Mismatches between observed and simulated biome patterns
may partially result from differences in classification schemes. For
example, if we use a lower tree cover threshold to differentiate
between grassland and savanna, then fractions of the grassland areas
simulated in the North and West of the study area are classified
as savanna (Fig. A.1) and the simulated probability of biome shifts
would decrease in these areas. This re-classification increases agree-
ment with the Rutherford et al. (2006) biome map that categorizes
most of the Limpopo Province as savanna. Accordingly, the simu-
lated biome distribution and biome shifts would change by using
state variables such as leaf area index or productivity for the classi-
fication; such an approach has for example been used by Sato and
Ise (2012). We use a scheme based on biomass and vegetation cover
 S. Scheiter et al. / South African Journal of Botany 116 (2018) 86–95
100
Percentage biome shift (%)
80
60
40
20
0
C4 grassland
Area biome shifts (no. cells)
120
80
40
0
C4 grassland
Fig. 5. Relative and absolute rate of biome shift for different biome types in the IPCC RCP 4.5 scenario. Bars indicate the percentage or absolute number of the grid cells covered
by different vegetation states in 2016 that shift into other vegetation states until 2030, 2050 and 2100. Percentages and absolute numbers were calculated using all 30 replicate
simulation runs. See Supplementary Figs. A.6 and A.11 for RCP 2.6 and RCP 8.5.
because it implicitly considers information on vegetation struc-
ture and composition and because vegetation classification schemes
based on remote sensing products often use vegetation cover (see
also Scheiter and Higgins, 2009).
Vegetation structure is a key component influencing biodiversity
across different trophic levels and, to a great extent, defines habi-
tat suitability for animals (Tews et al., 2004). For example, Chown
(2010) showed that rodent numbers are the highest at approxi-
mately 10% shrub cover while they decrease for higher shrub cover.
Simulated increases of tree cover under future conditions therefore
could imply that habitats become unsuitable for rodents. Parr et al.
(2012) showed that assemblages of ant species differ between open
and closed habitats in the Hluhluwe Game Reserve, South Africa. Sir-
ami et al. (2009) found that bird species diversity is related to shrub
cover and concluded that shrub encroachment is likely to impact
bird community composition. Hence, biome shifts can have cascad-
ing effects on higher trophic levels if habitats become unsuitable for
certain animal species that would then need to migrate to find a new
habitat. If this is not possible due to environmental barriers or mobil-
ity of animals, species may go locally extinct. By contrast, changes in
habitat structure may release other species from competitive pres-
sure and promote their growth, and may also facilitate invasion of
other animal species that are adapted to this habitat. In this context,
process-based vegetation models are valuable tools to investigate
trophic interactions and biodiversity of both flora and fauna in more
detail (Scherer et al., 2016). In particular, recent DGVM develop-
ments allow an improved representation of diversity in vegetation
(Scheiter et al., 2013; Moncrieff et al., 2015; Langan et al., 2017).
93
until 2030
until 2050
until 2100
Savanna Woodland Forest
until 2030
until 2050
until 2100
Savanna Woodland Forest
Biome type in 2016
Our results identify those areas of the Limpopo Province where
the risk of biome shifts until 2100 is high and where model pro-
jections are uncertain. In particular, our study reveals that savannas
and grasslands are prone to potential biome shifts, with grass-
lands turning into savanna and savannas turning into woodlands
or into forests. Within each biome type, woody biomass or tree
cover increase. In addition, biome types may shift their spatial loca-
tion. This transformation has important implications for land use
in the Limpopo province because losses or spatial shifts of grass-
lands and savannas may decrease the size and shift the location
of areas suitable for livestock husbandry. On the other hand, our
results show expansion of forest which suggests a high potential for
afforestation and carbon sequestration. In this study, we focused on
natural vegetation and ignored land use and management. However,
both empirical studies (Savadogo et al., 2009; Wessels et al., 2013;
Fisher et al., 2014; Mograbi et al., 2017) and modeling studies
(Scheiter et al., 2015; Scheiter and Savadogo, 2016) show that
activities such as grazing, fuel wood collection and fire manage-
ment may have substantial impacts on carbon stocks, vegetation
patterns and diversity and accelerate or inhibit undesired vegeta-
tion change. To allow for explicit management recommendations
for the Limpopo province, we also need to understand interactions
between grazing, wood harvesting, shrub encroachment and fire
and how these factors interact with climate and CO2 fertilization
effect to define future vegetation patterns, community composition
and socio-ecological systems in savanna rangelands (Pfeiffer et al.
unpublished, Gaillard et al. unpublished). Le Maitre et al. (2007)
argued that links between climate, biodiversity, land use, ecosystem
94 S. Scheiter et al. / South African Journal of Botany 116 (2018) 86–95
services and socio-economic systems need to be improved and cou-
pled vegetation-land use models are important tools to fill this
research gap.
5. Conclusions
We conclude that under future climate conditions and elevated
CO2 concentrations, large proportions of Limpopo’s grasslands could
develop into savannas and woodland while today’s savannas may
transition into forests. These biome shifts could entail substantial
impacts on higher trophic levels, biodiversity and the flux of ecosys-
tem services important to humans in the Limpopo province. Our
results suggest that management and mitigation efforts should par-
ticularly focus on more open ecosystems because the likelihood of
biome shifts is critical in these regions. Yet, further studies are neces-
sary to understand how management and land use may interact with
climate change to define the future vegetation state in the Limpopo
province.
Acknowledgments
SS and MP thank the Deutsche Forschungsgemeinschaft (DFG)
for the funding (Emmy Noether programme, grant SCHE 1719/2-1),
MP, CG and CM thank the German Federal Ministry of Education
and Research (BMBF) for funding (SPACES initiative, ‘Limpopo Living
Landscapes’ project, grant 01LL1304B and ‘ARS AfricaE’ project, grant
01LL1303C). BFNE is supported by the Exxaro Chairman’s Fund.
Appendix A. Supplementary data
Supplementary data to this article can be found online at https://
doi.org/10.1016/j.sajb.2018.02.394.
References
Archer, E.R.M., Landman, W.A., Tadross, M.A., Malherbe, J., Weepener, H., Maluleke,
P., Marumbwa, F.M., 2017. Understanding the evolution of the 2014–
2016 summer rainfall seasons in Southern Africa: key lessons. Climate
Risk Management 16, 22–28. http://www.sciencedirect.com/science/article/pii/
S2212096316300857. https://doi.org/10.1016/j.crm.2017.03.006.
Baudena, M., Dekker, S.C., Van Bodegom, P.M., et al. 2015. Forests, savannas, and
grasslands: bridging the knowledge gap between ecology and dynamic global
vegetation models. Biogeosciences 12, 1833–1848. http://www.biogeosciences.
net/12/1833/2015/. https://doi.org/10.5194/bg-12-1833-2015.
Beringer, J., Hutley, L.B., Abramson, D., et al. 2015. Fire in Australian savannas: from
leaf to landscape. Global Change Biology 21, 62–81. https://doi.org/10.1111/gcb.
12686.
Biggs, R., Simons, H., Bakkenes, M., Scholes, R.J., Eickhout, B., Van Vuuren, D., Alkemade,
R., 2008. Scenarios of biodiversity loss in Southern Africa in the 21st century.
Global Environmental Change 18, 296–309.
Bond, W.J., Midgley, G.F., Woodward, F.I., 2003. The importance of low atmospheric
CO2 and fire in promoting the spread of grasslands and savannas. Global Change
Biology 9, 973–982.
Bond, W.J., Midgley, J.J., 2001. Ecology of sprouting in woody plants: the persistence
niche. Trends in Ecology & Evolution 16, 45–51.
Buitenwerf, R., Bond, W.J., Stevens, N., Trollope, W.S.W., 2012. Increased tree densities
in South African savannas: > 50 years of data suggests CO2 as a driver. Global
Change Biology 18, 675–684. https://doi.org/10.1111/j.1365-2486.2011.02561.x.
Chown, S., 2010. Temporal biodiversity change in transformed landscapes: a South-
ern African perspective. Philosophical Transactions of the Royal Society B 365,
3729–3742.
Coetzer-Hanack, K.L., Witkowski, E.T.F., Erasmus, B.F., 2016. Thresholds of change in
a multi-use conservation landscape of South Africa: historical land-cover, future
transformation and consequences for environmental decision-making. Environ-
mental Conservation 43, 253–262. https://doi.org/10.1017/S0376892916000084.
DEA, 2011. South Africa’s Second National Communication under the United Nations
Framework Convention on Climate Change. Department of Environmental Affairs,
Republic of South Africa, Pretoria.
Donohue, R.J., Roderick, M.L., MT, R., Farquhar, G.D., 2013. Impact of CO2 fertiliza-
tion on maximum foliage cover across the globe’s warm, arid environments.
Geophysical Research Letters 40, 1–5. https://doi.org/10.1002/grl.50563.
Fisher, J., Erasmus, B., Witkowski, E., Van Aardt, J., Asner, G., Wessels, K., Mathieu,
R., 2014. Management approaches of conservation areas: differences in woody
vegetation structure in a private and a national reserve. South African
Journal of Botany 90, 146–152. http://www.sciencedirect.com/science/article/
pii/S0254629913004523. https://doi.org/10.1016/j.sajb.2013.10.016.
Gbetibouo, G., 2009. Understanding farmers perceptions and adaptations to climate
change and variability: the case of the Limpopo Basin farmers South Africa. IFPRI
Discussion Paper 849. International Food Policy Research Institute, Washington,
D.C. (USA).
Giorgetta, M.A., Jungclaus, J., Reick, C.H., et al. 2013. Climate and carbon cycle changes
from 1850 to 2100 in MPI-ESM simulations for the Coupled Model Intercompari-
son Project phase 5. Journal of Advances in Modeling Earth Systems 5, 572–597.
https://doi.org/10.1002/jame.20038.
Gu, Z., Gu, L., Eils, R., Schlesner, M., Brors, B., 2014. Circlize implements and enhances
circular visualization in R. Bioinformatics 30, 2811–2812.
Higgins, S.I., Bond, W.J., Trollope, W.S., 2000. Fire, resprouting and variability: a recipe
for grass-tree coexistence in savanna. Journal of Ecology 88, 213–229.
Higgins, S.I., Bond, W.J., Trollope, W.S.W., Williams, R.J., 2008. Physically motivated
empirical models for the spread and intensity of grass fires. International Journal
of Wildland Fire 17, 595–601.
Higgins, S.I., Kantelhardt, J., Scheiter, S., Boerner, J., 2007. Sustainable management of
extensively managed savanna rangelands. Ecological Economics 62, 102–114.
Higgins, S.I., Scheiter, S., 2012. Atmospheric CO2 forces abrupt vegetation shifts locally,
but not globally. Nature 488, 209–212.
Hoffmann, W.A., Geiger, E.L., Gotsch, S.G., et al. 2012. Ecological thresholds at the
savanna-forest boundary: how plant traits, resources and fire govern the distri-
bution of tropical biomes. Ecology Letters 15, 759–768. https://doi.org/10.1111/
j.1461-0248.2012.01789.x.
IPCC, 2013. Climate Change 2013: The Physical Science Basis. Contribution of Work-
ing Group I to the Fifth Assessment Report of the Intergovernmental Panel
on Climate Change. Cambridge University Press, Cambridge, United Kingdom
and New York, NY, USA. http://www.climatechange2013.org. https://doi.org/10.
1017/CBO9781107415324.
IPCC, Field, C.B., Barros, V.R., Dokken, D.J., Mach, K.J., Mastrandrea, M.D., Bilir, T.E.,
Chatterjee, M., Ebi, K.L., Estrada, Y.O., Genova, R.C., Girma, B., Kissel, E.S., Levy,
A.N., MacCracken, S., Mastrandrea, P.R., White, L.L. (Eds.), 2014a. Climate change
2014: impacts, adaptation, and vulnerability. Part A: global and sectoral aspects.
Contribution of working group II to the fifth assessment report of the intergovern-
mental panel on climate change. Cambridge University Press, Cambridge, United
Kingdom and New York, NY, USA.(1132 pp).
IPCC, Barros, V.R., Field, C.B., Dokken, D.J., Mastrandrea, M.D., Mach, K.J., Bilir, T.E.,
Chatterjee, M., Ebi, K.L., Estrada, Y.O., Genova, R.C., Girma, B., Kissel, E.S., Levy, A.N.,
MacCracken, S., Mastrandrea, P.R., White, L.L. (Eds.), 2014b. Climate change 2014:
impacts, adaptation, and vulnerability. Part B: regional aspects. Contribution of
working group II to the fifth assessment report of the intergovernmental panel
on climate change. Cambridge University Press, Cambridge, United Kingdom and
New York, NY, USA.688 pp.
Jones, C., Lowe, J., Liddicoat, S., Betts, R., 2009. Committed terrestrial ecosystem
changes due to climate change. Nature Geoscience 2, 484–487. https://doi.org/
10.1038/NGEO555.
Kgope, B.S., Bond, W.J., Midgley, G.F., 2010. Growth responses of African savanna trees
implicate atmospheric [CO2 ] as a driver of past and current changes in savanna
tree cover. Austral Ecology 35, 451–463.
Körner, C., Asshoff, R., Bignucolo, O., et al. 2005. Carbon flux and growth in mature
deciduous forest trees exposed to elevated CO2 . Science 309, 1360. http://science.
sciencemag.org/content/309/5739/1360.abstract.
Kruger, A.C., Shongwe, S., 2004. Temperature trends in South Africa: 1960–2003.
International Journal of Climatology 24, 1929–1945.
Langan, L., Higgins, S.I., Scheiter, S., 2017. Climate-biomes, pedo-biomes or
pyro-biomes: which world view explains the tropical forest - savanna boundary
in South America? Journal of Biogeography 44, 2319–2330.
Le Maitre, D.C., O’Farrell, P.J., Reyers, B., 2007. Ecosystems services in South Africa: a
research theme that can engage environmental, economic and social scientists in
the development of sustainability science? South African Journal of Science 103,
367–376.
Lehohla, P., 2012. Census 2011 Census in brief. Report no.: 03-01-41, Statistics South
Africa, Pretoria.978-0-621-41388-5.
Lehohla, P., 2015. Tourism Satellite Account for South Africa, Final 2011 and Pro-
visional 2012 and 2013. Report No. 04-05-07. Statistics South Africa, Statistics
South Africa, Pretoria.
MacKellar, N., New, M., Jack, C., 2014. Observed and modelled trends in rain-
fall and temperature for South Africa: 1960–2010. South African Journal of
Science http://www.sajs.co.za/observed-and-modelled-trends-rainfall-and-
temperature-south-africa-1960%E2%80%932010/neil-mackellar-mark-new-
chris-jack. https://doi.org/10.1590/sajs.2014/20130353.
Maluleke, M., Malungani, T., Steenkamp, K. (Eds.), 2016. Limpopo Environment Out-
look Report 2016. Limpopo Provincial Government, Department of Economic
Development, Environment and Tourism, South Africa.
Matsika, R., Erasmus, B.F.N., Twine, W.C., 2013. A tale of two villages: assessing the
dynamics of fuelwood supply in communal landscapes in South Africa. Environ-
mental Conservation 40, 71–83. https://doi.org/10.1017/S0376892912000264.
Midgley, G.F., Bond, W.J., 2015. Future of African terrestrial biodiversity and ecosys-
tems under anthropogenic climate change. Nature Climate Change 5, 823–829.
https://doi.org/10.1038/nclimate2753.
Millennium Ecosystem Assessment, 2005. Ecosystems and Human Well-Being:
Synthesis. Island Press, Washington, DC.
Mittermeier, R.A., Gil, P.R., Hoffman, M., et al. 2005. Hotspots Revisited: Earth’S
Biologically Richest and Most Endangered Terrestrial Ecoregions. Cemex Books,
Mexico City.
 S. Scheiter et al. / South African Journal of Botany 116 (2018) 86–95
Mograbi, P.J., Asner, G.P., Witkowski, E.T.F., Erasmus, B.F.N., Wessels, K.J., Mathieu, R.,
Vaughn, N.R., 2017. Humans and elephants as treefall drivers in African savannas.
Ecography 40, 001–011. https://doi.org/10.1111/ecog.02549.
Moncrieff, G.R., Scheiter, S., Slingsby, J.A., Higgins, S.I., 2015. Understanding global
change impacts on South African biomes using dynamic vegetation models. South
African Journal of Botany 101, 16–23. https://doi.org/10.1016/j.sajb.2015.02.004.
Mooney, H., Larigauderie, A., Cesario, M., et al. 2009. Biodiversity, climate change, and
ecosystem services. Current Opinion in Environmental Sustainability 1, 46–54.
https://doi.org/10.1016/j.cosust.2009.07.006.
Mucina, L., Rutherford, M.C., 2006. The Vegetation of South Africa, Lesotho and
Swaziland, chap. Biomes and bioregions of Southern Africa. South African
National Biodivesity Institute, Pretoria., pp. 748–789.
New, M., Lister, D., Hulme, M., Makin, I., 2002. A high-resolution data set of surface
climate over global land areas. Climate Research 21, 1–25.
O’Connor, T.G., Puttick, J.R., Hoffman, M.T., 2014. Bush encroachment in Southern
Africa: changes and causes. African Journal of Range & Forage Science 31, 67–88.
https://doi.org/10.2989/10220119.2014.939996.
Ogutu, J.O., Owen-Smith, N., 2003. ENSO, rainfall and temperature influences on
extreme population declines among African savanna ungulates. Ecology Letters
6, 412–419. https://doi.org/10.1046/j.1461-0248.2003.00447.x.
Parr, C.L., Gray, E.F., Bond, W.J., 2012. Cascading biodiversity and functional conse-
quences of a global change-induced biome switch. Diversity and Distributions 18,
493–503. https://doi.org/10.1111/j.1472-4642.2012.00882.x.
Pool-Stanvliet, R., 2013. A history of the UNESCO man and the Biosphere Programme
in South Africa. South African Journal of Science 109, https://doi.org/10.1590/sajs.
2013/a0035.
Prentice, I., Bondeau, A., Cramer, W., 2007. Terrestrial Ecosystems in a Changing World.
Springer., pp. 175–192.
Ratnam, J., Bond, W.J., Fensham, R.J., 2011. When is a ‘forest’ a savanna, and why does it
matter? Global Ecology and Biogeography 20, 653–660. https://doi.org/10.1111/
j.1466-8238.2010.00634.x.
Reyers, B., 2004. Incorporating anthropogenic threats into evaluations of regional bio-
diversity and prioritisation of conservation areas in the Limpopo Province, South
Africa. Biological Conservation 118, 521–531. https://doi.org/10.1111/j.1466-
8238.2010.00634.x.
Reyers, B., Rouget, M., Jonas, Z., Cowling, R.M., Driver, A., Maze, K., Desmet, P.,
2007. Developing products for conservation decision-making: lessons from a
spatial biodiversity assessment for South Africa. Diversity and Distributions 13,
608–619. https://doi.org/10.1111/j.1472-4642.2007.00379.x.
Rutherford, M.C., Mucina, L., Powrie, L.W., 2006. The Vegetation of South Africa,
Lesotho and Swaziland. Biomes and bioregions of Southern Africa, South African
National Biodivesity Institute, Pretoria. , pp. 748–789. chap.
Sato, H., Ise, T., 2012. Effect of plant dynamic processes on African vegetation
responses to climate change: analysis using the spatially explicit individual-
based dynamic global vegetation model (SEIB-DGVM). Journal of Geophysical
Research-Biogeosciences 117, G03017. https://doi.org/10.1029/2012JG002056.
Savadogo, P., Tigabu, M., Sawadogo, L., Oden, P.C., 2009. Herbaceous phytomass and
nutrient concentrations of four grass species in Sudanian savanna woodland sub-
jected to recurrent early fire. African Journal of Ecology 47, 699–710. https://doi.
org/10.1111/j.1365-2028.2008.01027.x.
Savadogo, P., Tiveau, D., Sawadogo, L., Tigabu, M., 2008. Herbaceous species responses
to long-term effects of prescribed fire, grazing and selective tree cutting in the
savanna-woodlands of West Africa. Perspectives in Plant Ecology Evolution and
Systematics 10, 179–195. https://doi.org/10.1016/j.ppees.2008.03.002.
Scheiter, S., Higgins, S.I., 2009. Impacts of climate change on the vegetation of Africa:
an adaptive dynamic vegetation modelling approach (aDGVM). Global Change
Biology 15, 2224–2246.
Scheiter, S., Higgins, S.I., 2012. How many elephants can you fit into a conserva-
tion area. Conservation Letters 5, 176–185. https://doi.org/10.1111/j.1755-263X.
2012.00225.x.
95
Scheiter, S., Higgins, S.I., Osborne, C.P., 2012. Fire and fire-adapted vegetation pro-
moted C4 expansion in the late Miocene. New Phytologist 195, 653–666. https://
doi.org/10.1111/j.1469-8137.2012.04202.x.
Scheiter, S., Higgins, S.I., Beringer, J., Hutley, L.B., 2015. Climate change and long-term
fire management impacts on Australian savannas. New Phytologist 205, 1211–
1226. https://doi.org/10.1111/nph.13130.
Scheiter, S., Langan, L., Higgins, S.I., 2013. Next generation dynamic global vegeta-
tion models: learning from community ecology. New Phytologist 198, 957–969.
https://doi.org/10.1111/nph.12210.
Scheiter, S., Savadogo, P., 2016. Ecosystem management can mitigate vegetation shifts
induced by climate change in West Africa. Ecological Modelling 332, 19–27.
https://doi.org/10.1016/j.ecolmodel.2016.03.022.
Scherer, C., Jeltsch, F., Grimm, V., Blaum, N., 2016. Merging trait-based and individual-
based modelling: an animal functional type approach to explore the responses
of birds to climatic and land use changes in semi-arid african savannas. Eco-
logical Modelling 326, 75–89. http://www.sciencedirect.com/science/article/pii/
S0304380015003117. https://doi.org/10.1016/j.ecolmodel.2015.07.005.
Sirami, C., Seymour, C., Midgley, G., Barnard, P., 2009. The impact of shrub encroach-
ment on savanna bird diversity from local to regional scale. Diversity and
Distributions 15, 948–957. https://doi.org/10.1111/j.1472-4642.2009.00612.x.
Skowno, A.L., Thompson, M.W., Hiestermann, J., Ripley, B., West, A.G., Bond, W.J., 2017.
Woodland expansion in South African grassy biomes based on satellite observa-
tions (1990-2013): general patterns and potential drivers. Global Change Biology
23, 2358–2369. https://doi.org/10.1111/gcb.13529.
StatisticsSouthAfrica, 2015. Living conditions of households in South Africa. Tech. rep.,
STATISTICS SOUTH AFRICA.,
Stevens, N., Erasmus, B.F.N., Archibald, S., Bond, W.J., 2016. Woody encroachment over
70 years in South African savannahs: overgrazing, global change or extinction
aftershock? Philosophical Transactions of the Royal Society of London B: Biologi-
cal Sciences 371, 20150437. http://rstb.royalsocietypublishing.org/content/371/
1703/20150437. https://doi.org/10.1098/rstb.2015.0437.
Stevens, N., Lehmann, C.E.R., Murphy, B.P., Durigan, G., 2017. Savanna woody
encroachment is widespread across three continents. Global Change Biology 23,
235–244. https://doi.org/10.1111/gcb.13409.
Tews, J., Brose, U., Grimm, V., Tielbörger, K., Wichmann, M.C., Schwager, M., Jeltsch,
F., 2004. Animal species diversity driven by habitat heterogeneity/diversity: the
importance of keystone structures. Journal of Biogeography 31, 79–92. https://
doi.org/10.1046/j.0305-0270.2003.00994.x.
Twine, W.C., Holdo, R.M., 2016. Fuelwood sustainability revisited: integrating size
structure and resprouting into a spatially realistic fuelshed model. Journal Of
Applied Ecology 53, 1766–1776. https://doi.org/10.1111/1365-2664.12713.
Twine, W., Moshe, D., Netshiluvhi, T., Siphugu, V., 2003. Consumption and direct-use
values of savanna bio-resources used by rural households in Mametja, a semi-
arid area of Limpopo province, South Africa. South African Journal of Science 99,
467–473. https://journals.co.za/content/sajsci/99/9-10/EJC97689.
Wessels, K.J., Colgan, M.S., Erasmus, B.F.N., 2013. Unsustainable fuelwood extraction
from South African savannas. Environmental Research Letters 8, 014007. http://
stacks.iop.org/1748-9326/8/i=1/a=014007.
Wigley, B.J., Bond, W.J., Hoffman, M.T., 2010. Thicket expansion in a South African
savanna under divergent land use: local vs. global drivers? Global Change Biology
16, 964–976.
Williams, R.J., Gill, A.M., Moore, P.H.R., 1998. Seasonal changes in fire behaviour in a
tropical Savanna in Northern Australia. International Journal of Wildland Fire 8,
227–239.
Zhu, Z., Piao, S., Myneni, R.B., 2016. Greening of the Earth and its drivers. Nature
Climate Change 6, 791–795. https://doi.org/10.1038/nclimate3004.