Content uploaded by Jessica Bernal Borrego
Author content
All content in this area was uploaded by Jessica Bernal Borrego on Apr 20, 2023
Content may be subject to copyright.
HUMAN-WILDLIFE CONFLICT
IN THE LIMPOPO NATIONAL
PARK
An approach through habitat suitability modeling
Jessica Bernal Borrego
Jessicabernal2012@gmail.com
Advisors: Guillermo Palacios Rodríguez and Miguel Ángel Lara Gómez
Final Master's Project
MSc in Geomatics, Remote Sensing and Spatial Models Applied to Forest Management
University of Cordoba (UCO)
ii
Human-wildlife conflict in the Limpopo National Park. An approach
through habitat suitability modeling.
Abstract
Among the large herbivores that inhabit the Mozambican Limpopo National Park are the
elephant (L. africana), the buffalo (S. caffer), and the hippo (H. amphibius). Although
since the moment of declaration of the territory as a National Park in 2001 the existing
population within the park and adjoining areas have been relocating outside of it, the
negative human interaction with the species continue to occur, a conflict exacerbated by
drought and increasing pressure on freshwater resources. In this work, we approach this
conflict through ensemble modeling of species and interactions, exploring the statistics
to eventually provide heat maps of potential conflict areas in and around the park. The
ensemble statistics showed in general good results in TSS (0.72-0.78), ROC (0.91-0.95),
and Kappa (0.4-0.65) values. RF, Maxent, and GAM show better performance, with the
variables of most importance in the models being altitude and mean temperature of the
warmest month for L. africana; altitude and annual precipitation for S. caffer; and the dry
season NDWI and annual precipitation for H. amphibius. Potential conflict points are
found around the park, mainly distributed in the western half of it, if we consider the
probability gradient of occurrence of the species. Our results illustrate the potential ap-
plication of habitat distribution models in the study of human-wildlife conflicts.
keywords: Conflict points, ensemble modeling, habitat distribution model, human-wild-
life conflict, large herbivores, Limpopo.
iii
Conflicto entre humanos y vida silvestre en el Parque Nacional de Limpopo.
Una aproximación a través de la modelización de idoneidad del hábitat.
Resumen
Entre los grandes herbívoros que habitan en el Parque Nacional de Limpopo de Mozam-
bique se encuentran el elefante (L. africana), el búfalo (S. caffer) y el hipopótamo (H.
amphibius). Si bien desde el momento de la declaración del territorio como Parque Na-
cional en 2001 la población existente dentro del parque y áreas aledañas se ha ido reubi-
cando fuera del mismo, la interacción humana negativa con las especies continúa ocu-
rriendo, conflicto agudizado por la sequía y la creciente presión sobre los recursos de agua
dulce. En este trabajo, abordamos este conflicto a través de modelos de conjuntos de es-
pecies e interacciones, explorando las estadísticas para eventualmente proporcionar ma-
pas de calor de áreas de conflicto potenciales dentro y alrededor del parque. Las estadís-
ticas del conjunto mostraron en general buenos resultados en los valores TSS (0,72-0,78),
ROC (0,91-0,95) y Kappa (0,4-0,65). RF, Maxent y GAM muestran un mejor desempeño,
siendo las variables de mayor importancia en los modelos la altitud y la temperatura me-
dia del mes más cálido para L. africana; altitud y precipitación anual para S. caffer; y la
estación seca NDWI y la precipitación anual para H. amphibius. Los potenciales puntos
de conflicto se encuentran alrededor del parque, distribuidos principalmente en la mitad
occidental del mismo, si consideramos el gradiente de probabilidad de ocurrencia de la
especie. Nuestros resultados ilustran la aplicación potencial de los modelos de distribu-
ción de hábitats en el estudio de los conflictos entre humanos y vida silvestre.
Palabras clave: Conflicto entre humanos y vida silvestre, grandes herbívoros, Limpopo,
modelado de conjuntos, modelo de distribución de hábitat, puntos de conflicto.
0
Índex
1. Introduction ......................................................................................................................... 1
1.1. Justification and Objectives ........................................................................................... 1
1.2. Documental Analysis ..................................................................................................... 2
2. Theoretical Framework ...................................................................................................... 4
2.1. Human-wildlife conflict ................................................................................................. 4
2.2. Species under study ...................................................................................................... 4
2.2.1. Loxodonta africana................................................................................................ 4
2.2.2. Syncerus caffer ...................................................................................................... 5
2.2.3. Hippopotamus amphibius ..................................................................................... 6
2.3. The role of Habitat Suitability Modeling ....................................................................... 7
3. Methodology ........................................................................................................................ 9
3.1. Potential distribution and Habitat Suitability Models................................................... 9
3.1.2. Conceptualization or overview ............................................................................ 10
3.1.3. Data preparation ................................................................................................. 11
3.1.4. Model fitting or calibration ................................................................................. 13
3.1.5. Assessment or model evaluation ......................................................................... 13
3.1.6. Spatial predictions ............................................................................................... 14
3.2. Heat Maps with the areas of a higher risk of conflict ................................................. 14
4. Results and Discussion ...................................................................................................... 16
4.1. HSM and Heat Maps for Limpopo National Park HWC ............................................... 16
4.2. Tools in the conflict mitigation strategy ..................................................................... 35
5. Conclusions ........................................................................................................................ 40
6. Bibliography ...................................................................................................................... 42
Annex I ....................................................................................................................................... 50
Annex II...................................................................................................................................... 51
Annex III .................................................................................................................................... 54
1
1. Introduction
1.1. Justification and Objectives
The Limpopo National Park is a protected area of more than 10.000 km2 located in the
southwest of Mozambique, Gaza province, on the border with South Africa (Figure 1). It
forms part of the Greater Limpopo Transfrontier Park together with the Kruger (South
Africa) and Gonarezhou (Zimbabwe) National Parks. The declaration of the territory, un-
til then hunting preserve (1961), as a National Park, occurred in 2001
1
.
Figure 1. Boundary of the Limpopo National Park. Source: Own elaboration in QGIS 3.16 from official cartography.
Full-extent map in Annex II.
The climate of Mozambique is influenced by the monsoons of the Indian Ocean and by
the hot current of the Mozambique Channel and can be broadly defined as tropical and
humid. However, it has a dry season that tends to be longer in the south, where it can last
between six and nine months, which places us before a subregionalization with a dry
tropical climate, although the orography (54-527 m.a.s.l.) plays a key role in defining a
1
World Conservation Monitoring Center, UNEP-WCMC
2
high-altitude tropical climate. Average temperatures are around 20ºC in the south, with
maximum temperatures in the rainy season
2
.
Among the large species that inhabit this park are the elephant (Loxodonta africana), the
buffalo (Syncerus caffer) and the hippopotamus (Hippopotamus amphibius). Although
from the moment of declaration of the National Park, the human population centers still
existing within the park and adjoining areas have been relocating outside of it, the prob-
lems derived from human interaction with the species continue to occur, a conflict exac-
erbated by drought and increasing pressure on freshwater resources.
The main objective of this work is the identification, through the mapping of interactions
and the use of statistical analysis, of the areas with a higher risk of conflicts between
humans and wildlife under study. Once this main objective has been achieved, an analysis
of different tools utilized in the conflict mitigation strategy is carried out to eventually
offer an overview of their effectiveness and the status of the situation.
1.2. Documental Analysis
For the development of this work, a bibliographic analysis of peer-reviewed publications
has been carried out, mainly using the Scopus search engine, and specifically PubMed,
ScienceDirect, ResearchGate and Google Academic. At first, terms related to human-
wildlife conflict were combined, including keywords related to remote sensing, satellites
and the species object of the study, for the period 2012-2022. Specifically, the search was
carried out using the following combination of keywords:
(human-wildlife OR elephant OR loxodonta OR hippopotamus OR buffalo OR syncerus)
AND (Conflict) AND ("remote sensing" OR "remotely sensed" OR sensor OR satellite OR
"unmanned aerial vehicle" OR spectral OR "map") AND PUBYEAR > 2011 AND
PUBYEAR < 2023
In Scopus, a total of 6221 publications matched our search criteria. A limitation to books
and articles and a refinement by theme was applied to this result, so that the search was
focused on the areas of "Agricultural and Biological Sciences", "Environmental Sciences"
2
https://www.mozambique-emb.es/clima (Accessed August 4, 2022)
3
and "Earth and Planetary Sciences". This refinement limited the results to 2996 publica-
tions. Next, to keep the workload manageable, the results were limited by publication
source to focus only on high-impact remote sensing journals: Biological Conservation,
African Journal of Ecology, European Journal Of Wildlife Research, Remote Sensing,
Ecological Modelling, International Journal Of Geo Information, Remote Sensing In
Ecology And Conservation, Applied Ecology And Environmental Research, and Remote
Sensing Applications Society And Environment. In this way, the result was limited to 232
documents. These publications were individually reviewed to identify and keep only
those relevant to our objectives, that is, articles or books that use remote sensing data,
preferably in modelling and monitoring species to identify spatial human-wildlife conflict
(HWC) patterns. In this way, we exclude:
(i) Works that only focused on mapping species richness distribution, since they
do not evaluate the ability of habitat modeling to identify nor quantify taxa-
environmental relationships, nor the influence of the human species on his
conflicts,
(ii) Those that are focused on historical and multitemporal analyses to monitoring
changes in the habitat of the species, and,
(iii) Those studies carried out at very broad spatial scales or at coarse resolution
(>10 km) since these pixel sizes make little ecological sense at the local scale
of the area under study.
In addition, an analysis of tools implemented in the conflict mitigation strategy is per-
formed by looking at publications focused on “Limpopo” or “Mozambique" regardless of
its relationship with geomatics techniques
3
. From this search, 27 documents resulted.
Finally, a total of 110 articles were selected for the development of this work.
3
Keywords: (human-wildlife OR elephant OR loxodonta OR hippopotamus OR buffalo OR syncerus
) AND ( conflict ) AND ( "Limpopo" OR "Mozambique" ) AND PUBYEAR > 2010 AND PUBYEAR <
2025 AND ( LIMIT-TO ( AFFILCOUNTRY , "Mozambique" ) ) AND ( LIMIT-TO ( DOCTYPE , "ar" ) OR
LIMIT-TO ( DOCTYPE , "bk" ) ) AND ( LIMIT-TO ( SUBJAREA , "AGRI" ) OR LIMIT-TO ( SUBJAREA ,
"ENVI" ) OR LIMIT-TO ( SUBJAREA , "EART" ) )
4
2. Theoretical Framework
2.1. Human-wildlife conflict
The IUCN Species Survival Commission (SSC) Human-Wildlife Conflict & Coexistence
Specialist Group defines human-wildlife conflict as: “struggles that emerge when the
presence or behavior of wildlife poses an actual or perceived, direct and recurring threat
to human interests or needs, leading to disagreements between groups of people and neg-
ative impacts on people and/or wildlife” (IUCN Issues Brief. Human-Wildlife Conflict.,
2022).
These negative outcomes of human–wildlife interactions have been reported with the Af-
rican savanna elephant (Loxodonta africana), the buffalo (Syncerus caffer) and the hip-
popotamus (Hippopotamus amphibius), which are being considered as threats to people,
crops, and livestock at places where both species co-occurred. At the same time, these
mega-fauna species are themselves particularly threatened with extinction, mainly by
overexploitation, agricultural activities associated with crop and livestock production,
and by illegal hunting and retaliatory killings. As such, HWC, exacerbated by drought
and the increasing pressure on freshwater resources, may eventually undermine the efforts
of conservation, which highlights the need to understand HWC determinants (Meyer &
Börner, 2022).
2.2. Species under study
2.2.1. Loxodonta africana
The African Savanna elephant, Loxodonta africana (Blumenbach, 1797), is an endan-
gered
4
mammal species that can be found over a wide latitudinal (16° N - 34° S) and
altitudinal range in Africa, occupying a variety of habitats ranging from forests, savanna,
shrubland, grassland, inland wetlands, and even arid deserts. However, today L. africana
occupies an estimated 15% of their historic pre-agricultural range (Chase et al., 2016).
Naturally, L. Africana populations move greater or shorter distances in response to cli-
matic conditions (e.g., seasonality and drought), exhibiting different migratory, semi-mi-
gratory, or residence range patterns behaviors. This last is the case of the subpopulation
4
The IUCN Red List of Threatened Species: https://www.iucnredlist.org/ (Accessed August 4, 2022)
5
occurring in Limpopo National Park, where plant productivity and water availability have
provided vital resources for entire generations of this big herbivore species, which counts
with an intergenerational interval of 24-25 years (Gobush et al., 2021).
Besides the well-known charismatic role of wild animals such as the elephant in attracting
tourists to national parks, a much more important ecological purpose exists behind the
scenes for this species. Between the ecosystem services provided by L. Africana are their
functions, for example, as pollinators, and as land engineers by interfering in the tree-
grass ratio, processing plant material, trampling, and maintaining and creating water
paths, all of which make room for other smaller important species.
They also play an important role in soil fertility by transporting nutrients from eutrophic
or richer sources to dystrophic or poorer soils (Doughty et al., 2016; Gobush et al., 2021;
Malhi et al., 2016).
On the other hand, although some of the major causes for population reduction are human
triggered (e.g., habitat loss due to human population expansion and illegal elephant killing
for ivory trading), L. Africana holds a symbolic significance within the culture of many
communities, and there are African conservation organizations engaged on their protec-
tion
5
.
2.2.2. Syncerus caffer
The African buffalo, Syncerus caffer (Sparrman, 1779) is a mammal species of the Bo-
vidae family, categorized as Near Threatened by the IUCN Red List. Given the existence
of significant morphological variability and the recognition of various subspecies, two
distinct groups are distinguished, i.e., a West and Central African group and an East and
Southern African group (Smitz et al., 2013). In the Limpopo National Park this last group
is found, comprised of a single subspecies, the Southern Savannah Buffalo or Syncerus
caffer caffer (hereafter, S. caffer). S. caffer is twice the size of forest buffalo, with large
downward curved horns and a brownish to black coloration (Smitz et al., 2013)
S. caffer can thrive in a wide range of non-desertic habitats with annual precipitation
above 250 mm, from coastal savannas, semi-arid shrublands, Acacia and miombo Bra-
chystegia woodlands, to montane grasslands and forests up to 4000 m.a.s.l., passing
5
https://www.awf.org/ (Accessed August 12, 2022)
6
through moist lowland rainforests. However, during the 19th and 20th centuries, the distri-
bution of the African Buffalo has been drastically reduced, and is predicted to continue
to decline because of different factors, i.e.: severe climatic events such as droughts, that
in Gonarezhou and Kruger have already strongly impacted buffalo populations (IUCN
SSC Antelope Specialist Group, 2019) (Cornélis et al., 2014); habitat loss due land con-
version, including forest clearance and the expansion of human settlements, agriculture,
and livestock grazing; and poaching, as S. caffer is also a target prized for bushmeat and
as a trophy for hunters.
On the other hand, disease outbreaks, like the rinderpest epidemic suffered on the late
19th century, are a major issue that concentrate efforts around the species (e.g., culling and
fences), since the transmission of pathogens may cause high reductions in wild and do-
mestic buffalo populations, cattle, and other species.
All the above are major threats outside protected areas like the Limpopo National Park,
where human-wildlife interactions become stronger, more reason why it is essential to
seek solutions that involve conservation and can help local communities in the sustainable
management and protection of their ecosystems, even more so considering the impacts of
climate change.
Regarding their ecosystem roles, S. caffer species are important and cyclic grazers in their
habitats, they open vegetation patches to other species that can follow or graze with the
herd. Some of these species (e.g., zebras and wildebeest) may also be benefited from the
possible protection against lions by associating with the more aggressive African buffalo
(Ng, 2015).
2.2.3. Hippopotamus amphibius
Hippopotamus amphibius (Linnaeus, 1758), commonly called common hippopotamus,
hippopotamus, large hippo, or simply, hippo, is a mammal species of the Hippopotamidae
family, categorized as Vulnerable by the IUCN Red List. H. amphibius is a semi-aquatic
species that can be found in wetlands, rivers, shallow lakes, and swamps in which sub-
merge its entire body and retreat after sunset, when it usually emerges from water to feed.
Normally, hippos forage on dense, grassy grazing areas during night hours up to several
7
kilometers from the water source. It is likely that the distance H. amphibius travel to
grazing areas varies seasonally and among different areas (Lewison & Pluháček, 2017).
The species still occurs in much of its former range from the 20th century, although pop-
ulation sizes have declined. Illegal hunting and poaching, habitat loss and fragmentation
by fences or other barriers, and reduction of water quality and quantity are the primary
causes of this situation. Despite of that, Kruger National Park is believed to be home to
3,000 individuals, showing general positive trends in its subpopulations. In Limpopo,
there is an estimated 700 animals, with another 700 residing in the Olifants and Letaba
Rivers, at the south-west of the study area (Lewison & Pluháček, 2017).
H. amphibius habitat loss is particularly linked with agricultural development, as hippos
rely on freshwater habitats and water diversion increase pressures over populations, grad-
ually isolating many of them into protected areas (Lewison, 2007). At the same time,
given the growing pressure on freshwater resources across Africa, protected areas like the
Limpopo National Park need effective management to avoid the growing pressure from
local communities, as the increase of hippo’s population in wetlands areas may fuel the
HWC (Mackie et al., 2013).
Lastly, note that similarly to L. Africana and S. caffer, H. amphibius plays a key role in
the habitat it belongs to. In its daily routine in and out of the water, H. amphibius opens
and clears paths that when flooded, create most of the water lagoons and side pools that
many small organisms use especially during drought conditions (Mason, 2013; Mosepele
et al., 2009).
2.3. The role of Habitat Suitability Modeling
For any of the species’ object of this study, all major threats currently in place include
habitat destruction, modification, and fragmentation, together with the growing intensity
of droughts as part of the threats posed by climate change. As these escalate, so do HWC.
These threats are commonly estimated using multi-scale assessments based on IUCN Red
List criteria such as geographic extent or area of occupancy (Guisan et al., 2017). Alt-
hough this system is still essential for decision-making, new sound methods, that take
advantage of technological advances and high spatiotemporal resolution datasets, are
needed to forecast, monitor, and inform decisions at the local level. For this matter, the
8
development habitat suitability modeling approaches have demonstrated their key role in
improving our understanding of species’ ecological patterns and on testing different en-
vironmental and biogeographical hypothesis (Guisan & Thuiller, 2005; Guisan & Zim-
mermann, 2000; Matawa et al., 2012, 2012; Meynard & Kaplan, 2012; Scharf et al., 2018;
Thuiller, 2014).
Habitat suitability models (HSM), also called Species distribution models (SDM) or Eco-
logical niche models (ENM), refer to ecological modeling approaches in which species
observations data are related to a set of environmental (abiotic and biotic) predictors var-
iables to quantify the realized environmental niche and project it in geographic space to
predict species distribution (Guisan et al., 2017). Working on HWC with HSM, as with
any other mathematical model, requires awareness about its limitations and initial as-
sumptions before any practical applications are made, whenever they represent a simpli-
fication of the real world.
One important assumption concerns the system's consideration of being at equilibrium or
pseudo-equilibrium. This implies that the data accurately represent species-environment
relationships in a way the resulting model could be projected at another place or period.
There are many contextual reasons why this assumption cannot hold (Wiens & Graham,
2005). Nevertheless, an organism’s niche can be fitted successfully for its current distri-
bution if all the environmental combinations that make its niche are captured by the spe-
cies’ occurrences (Guisan et al., 2014; Pearman et al., 2008). Moreover, we need to know
what the model will be eventually used for, as we also assume that the species occurrences
we use (densities, frequencies, simple observations, etc.) are appropriate to fit it. Another
assumption is that all-important environmental predictors required are assumed to be
available at the resolution relevant for the species being modeled, since an absence of
important variables when modeling will lead us to unexplained variance (Guisan et al.,
2017; Mod et al., 2016).
In addition to the above, the appropriateness of the statistical modelling methods, the
avoidance of error in predictors measures, the absence of biased species data, and the
independence of observations, are all critical methodological assumptions that will be
dealt with in the next section.
There are also additional assumptions required to be considered, for example when pro-
jecting HSM to future scenarios, such as whether the ecological niches fitted by these
9
models are fully captured and assumed to be stable in time and space (Guisan et al., 2017;
Guisan & Thuiller, 2005; Thuiller, 2014). For this work, these future projections would
be useful for integrate climate change into conservation management of the Limpopo Na-
tional Park, but, after careful consideration, and giving the current lack of information on
aspects considered key (multi-scale species migration patterns, subpopulations dynamics,
interspecific interactions, etc.), it has been decided to follow this path in future research.
In any case, on a subject as complex as the HWC, the study presented here has no more
desire than to carry out a first exercise in the search for ecological axioms or principles
from which to build new hypotheses and that serve to better understand the system in
which these conflicts occur. Therefore, HSM, as ecological models, do not only have the
value of their predictive capacity, but also that of addressing a cleaner discourse of ideas
(Rodríguez Martínez, 2013), with concepts and processes that are as well defined as pos-
sible, which eventually allow further improvement of its performance.
3. Methodology
3.1. Potential distribution and Habitat Suitability Models
The steps followed in the development of this work are based on the methodology pro-
posed in previous works consulted and especially the protocols established by Guisan et
al. (2017) and Zurell et al. (2020) (Figure 2). These steps can be summarized in (1) Over-
view or conceptualization, (2) Data preparation, (3) Model fitting or calibration, (4) As-
sessment or model evaluation, and (5) Spatial predictions
6
.
6
ODMAP: Overview, Data, Model, Assessment and Prediction (Zurell et al., 2020)
10
Figure 2. The five main modelling steps in the species distribution modelling cycle. Source: Zurell et al., 2020
3.1.2. Conceptualization or overview
In previous chapters (Introduction and Theoretical Framework), this first step has been
mostly settled: focal taxa, location, ecological data overview, underlying assumptions,
etc. As the goal of the analysis is not to project species habitat suitability to future climate
or extrapolate it to other regions, nor to test an ecological hypothesis, but to locally predict
habitat suitability under current environmental conditions for species management pur-
poses in the framework of HWC, the scale of the analysis is defined by the spatial extent
the Limpopo National Park. Considering the ecology of the species, a buffer zone of 30
km was established around the protected area, so the datasets also considered those of the
countries bordering to the west and northwest of Mozambique, i.e., South Africa and
Zimbabwe, respectively.
Geodata on the range of presence of the species, after evaluation of the regional geo-
graphic information provided by the IUCN, have been compiled from the Global Biodi-
versity Information Facility (GBIF). The predictors considered for the models are all
compiled or calculated from open existing databases and include bioclimatic variables,
human footprint, and vegetation indexes. All spatial data were downloaded or converted
to 1 km2 resolution and projected as WGS 84 coordinate system. For final maps, all layers
were assigned with the final Coordinate Reference System of the project, UTM 36 S.
11
All models were generated in R software (R Core Team, 2014) through RStudio environ-
ment. The algorithms used are: Generalized Linear Model (GLM), Generalized Boosting
Model (GBM), Generalized Additive Model (GAM), Classification Tree Analysis (CTA),
Artificial Neural Network (ANN), Surface Range Envelop (SRE), Flexible Discriminant
Analysis (FDA), Multiple Adaptive Regression Splines (MARS), Random Forest (RF)
and Maximum Entropy (MAXENT Phillips.2). These are modeling techniques currently
available with the R package Biomod2. From the individual models obtained with the
different methods a consensus model is generated
7
.
3.1.3. Data preparation
Obtaining and preparing data on the presence of the species has been carried out using
the R packages rgbif and spThin. The first library includes functions to search for biodi-
versity data in the GBIF database. Only georeferenced data was used, with a return limit
of 5000 records to avoid spatial autocorrelation. This step includes the elimination of
excess information (NAs), duplicates, and the removal of outliers, when present. The
check and removal of outliers are carried out with QGIS Desktop 3.16. With the spatial
thinning of species occurrence records, problems associated with spatial sampling biases
are minimized, at the time the fewest records necessary are removed to retain the greatest
amount of useful information (Aiello-Lammens et al., 2015).
The bioclimatic variables were obtained from the WorldClim database, where the raster
values are calculated from several major climate databases, while elevation data comes
from the SRTM database (NASA). WorldClim contains interpolated climate data availa-
ble at 4 different spatial resolutions, from 30 seconds (~1 km2), the resolution of interest,
to 10 minutes (~340 km2). As this study is not built on a priori hypotheses about which
variables could explain the species distributions, but is rather exploratory, all variables
for the current climate were downloaded (Table 1). However, to reduce the level of un-
certainty in our predictions, the variables BIO 3, BIO14 and BIO15 were excluded from
the models, following the recommendations of Varela et al. (2015), since these variables
have been shown to carry a high level of discrepancy between general circulation models.
WorldClim data are downloaded with the geodata R package
8
.
7
The scripts are available at https://gitfront.io/r/user-8903261/UwyHqoc69fPR/HWC-LNP-TFM/
8
https://rdrr.io/cran/geodata/man/geodata-package.html
12
Table 1. WorldClim' Bioclimatic variables. Source: Own elaboration based on https://www.world-
clim.org/data/bioclim.html
Environmental Variable
Description
BIO1
Annual Mean Temperature
BIO2
Mean Diurnal Range (Mean of monthly (max temp - min temp))
BIO3
Isothermality (BIO2/BIO7) (×100)
BIO4
Temperature Seasonality (standard deviation ×100)
BIO5
Max Temperature of Warmest Month
BIO6
Min Temperature of Coldest Month
BIO7
Temperature Annual Range (BIO5-BIO6)
BIO8
Mean Temperature of Wettest Quarter
BIO9
Mean Temperature of Driest Quarter
BIO10
Mean Temperature of Warmest Quarter
BIO11
Mean Temperature of Coldest Quarter
BIO12
Annual Precipitation
BIO13
Precipitation of Wettest Month
BIO14
Precipitation of Driest Month
BIO15
Precipitation Seasonality (Coefficient of Variation)
BIO16
Precipitation of Wettest Quarter
BIO17
Precipitation of Driest Quarter
BIO18
Precipitation of Warmest Quarter
BIO19
Precipitation of Coldest Quarter
As remote sensing-based predictors, Normalized Difference Vegetation Index (NDVI),
Soil Adjusted Vegetation Index (SAVI) and the Normalized Difference Water Index
(NDWI) are used. Average vegetation indices for the wet and dry seasons are calculated
from satellite images Sentinel 2 (level 2) collection through the geospatial analysis plat-
form Google Earth Engine.
The human footprint, an indicator on the relative human influence in each terrestrial bi-
ome, is obtained from the Global Human Footprint Index dataset of the Last of the Wild
Project, Version 2, 2005 (Wildlife Conservation Society-WCS & Center For International
Earth Science Information Network-CIESIN-Columbia University, 2005). The dataset is
produced by the Wildlife Conservation Society and the Columbia University Center for
13
International Earth Science Information Network (CIESIN/SEDAC)
9
. This dataset, which
includes the Human Influence Index normalized by biome and realm, is a global dataset
of 1 km grid cells, created from nine global data layers of human population pressure,
human land use and infrastructure (built-up areas, night-time lights, land use/land cover),
and human access (coastlines, roads, railroads, navigable rivers).
3.1.4. Model fitting or calibration
Before the selections of definitive variables for calibration, bioclimatic data, as well as
vegetation indices NDVI, SAVI, and NDWI, are tested for multicollinearity through the
variance inflation factor (VIF), exploring the results from applying the functions vifcor
and vifstep, and comparing them with the VIF calculated (vif function) after variables
evaluation with the Spearman correlation coefficient.
For the calibration of each model, 10 sets of pseudoabsence data were generated by ran-
dom pseudo absences selection. The aim is to explore the performance of all the methods
(i.e., GLM, GBM, GAM, CTA, ANN, SRE, FDA, MARS, RF and MAXENT Phillips.2),
only changing the GAM default parameter “k = -1” to “k = 4” to avoid too complex
models.
The calibration options are settled through the function BIOMOD_Modeling, choosing
20 evaluation rounds, an 80% of data used to calibrate the models (the remaining part use
for testing), a prevalence of 0.5 to build weighted response weights, a number of permu-
tations to estimate the variable importance of 1, and setting all models to be calibrated
and evaluated with the entire dataset.
3.1.5. Assessment or model evaluation
Once evaluation values of the models are obtained, all variables importance and evalua-
tion metrics are saved for each model as a .csv data file. Performance statistics are esti-
mated on training data, on validation data and on test data through sensitivity and speci-
ficity. We distinguish True Skill Statistic (TSS), and the area under receiver operating
9
https://sedac.ciesin.columbia.edu/data/collection/gpw-v4/whatsnewrev11
14
characteristic curve (ROC or AUC), so models whose ROC scores are above 0.7 are con-
sidered good, while those above 0.9 are considered very good.
For the ensemble modeling, all partial models were assembled using the “top 5" models
(95% probability) with the ROC as the threshold metric. The ensemble estimation meth-
ods chosen are mean, median, and weighted mean of the probabilities across predictions,
with a 95% confidence or credible interval (level of significance of 0.05) around the
mean, getting the evaluations for ensemble models, Kappa statistics are also considered.
The relative importance of the weights is settled to “proportional”, so the attributed
weights are proportional to the evaluation scores given by weight method.
After Obtaining a habitat model for each species the resulting graphs include the assembly
model performance graph (TSS vs. ROC) and a graph of the response curves of the vari-
ables that contribute the most to the model.
3.1.6. Spatial predictions
The prediction output includes the projection of models over the study area under current
environmental conditions (models considered in the assembly), averaged to the criteria
selected (mean, median, or weighed).
Maps with the potential distribution (probability of presence) of each species and maps
of habitat availability for each species are generated, checking for ecological plausibility
against available biological knowledge before continuing to explore for hotspots HWC
maps (next section). For the predicted habitat plot, the necessary cut-off threshold point
depends on the technique finally selected so as not to lose predictive capacity.
Maps of potential species distributions are accompanied by equivalent “maps of devia-
tion”, so to inform the magnitude and extent of prediction uncertainty, hence supporting
a more correct and fair interpretation.
3.2. Heat Maps with the areas of a higher risk of conflict
In addition to the HSM model, the generation of HWC heat maps needs also geodata
representative of the variables involved in the conflict. For this purpose, population den-
sity, livestock and farming data are gathered and processed.
15
Crop data is obtained from the Copernicus Global Land Cover map at ~0.001° (~100 m)
resolution. For livestock (cattle, chicken, ducks, goats, horses, pigs, sheep, and domestic
buffaloes), the most recent version of the Gridded Livestock of the World database has
been used, a subnational livestock dataset generated using Random Forests, the machine-
learning technique that recently shown to provide more accurate gap filling and disaggre-
gation of livestock data than did the previously-used multivariate regression methods
(Gilbert et al., 2018). In this database species distributions are available in two represen-
tations, termed dasymetric and areal weighted. The first one contains different animal
densities assigned to different pixels within a given census polygon according to the RF
models, while the seconds simply spread individuals of a census polygon evenly, and the
density of animals in each pixel corresponds to the average number of animals per km2
of suitable land in the census unit. Therefore, of those available, the dasymetric model is
used.
On the other hand, human population layers used were retrieved from the GHS-POP spa-
tial raster product (GHSL Data Package 2022), a product produced in World Mollweide
at 100 m that depicts the distribution of human population expressed as the number of
people per cell. The base source for population estimates is the raw dataset of the Gridded
Population of the World, version 4.11, harmonized by CIESIN at polygon level. The dis-
aggregation from census or administrative units to grid cells is informed by the distribu-
tion, classification, and density of built-up as mapped in the Sentinel/Landsat based
GHSL global layers per corresponding epoch, as described in Freire et al., 2016
(Schiavina et al., 2022).
These datasets are aggregated to 1 km2 resolution and warped to WGS 1984 coordinate
system. For final maps, all layers were assigned with the final Coordinate Reference Sys-
tems of the project, UTM 36 S.
To obtain heat maps the R package spdep is run in RStudio. The analysis takes advantage
of the statistical index Getis-Ord Gi* (Getis & Ord, 1992), which evaluates the spatial
aggregation of data through the comparison of local (a site and its surroundings) and
global (the entire study area) averages. This comparison is based on the calculation of the
standardized value z or z-score. The z (standard deviations) and p values indicate whether
the features aggregate statistically over a given distance. Important functions used for this
analysis are the dnearneigh function, to identify neighbours of region points by euclidean
16
distances, the nb2listw function, to calculate the weights assigned to the neighbors of each
point, and the localG function, to calculate the local spatial statistic G or z value for each
zone.
The resulting maps include (1) the heat maps with the areas of a greater risk of conflict
according to the potential distribution of each species and of the three species, and (2) the
heat maps with the areas of a greater risk of conflict according to the availability of habitat
for each species and for the three species.
4. Results and Discussion
4.1. HSM and Heat Maps for Limpopo National Park HWC
The mean distance observed between points was, for the three species studied, less than
that expected for a random distribution, with the nearest neighbor index obtained being
less than 1, which indicates the presence of clusters in the data sets. Likewise, the z-scores
obtained were very low (negative) and associated with very small p-values, indicating
that, with a confidence level greater than 99%, it is possible to reject the null hypothesis
of randomness in our samples.
Through the spatial thinning tool and using a minimum distance of 3 km and 10 replicates,
we reduced these clusters to a maximum of 172 (L. africana, La), 38 (H. amphibius, Ha),
and 96 (S. caffer, Sc) records, obtaining the maximum frequencies shown in Figure 3 and
Figure 4. With each script a text file is created with a summary of the results
10
.
10
Included as supplemental material in https://gitfront.io/r/user-8903261/UwyHqoc69fPR/HWC-LNP-
TFM/
Figure 3. Frequency of maximum records for H. amphibius (left), and for L. Africana (right).
17
The calculation of the variance inflation factor (VIF) based on the Vifcor function gave a
greater statistical significance for the independent variables BIO6, BIO10, BIO12,
BIO13, BIO18, BIO19 and ALT (Figure 5), while with Vifstep the greater statistical sig-
nificance was only for the independent variables BIO10, BIO12, BIO13, BIO18 and ALT.
Figure 5. Most significant environmental variables according to Vifcor: bio6 (Minimum temperature of the coldest
month), bio10 (Average temperature of the warmest quarter), bio12 (Annual Precipitation), bio13 (Precipitation of the
rainiest month), bio18 (Precipitation of the warmest quarter) , bio19 (Precipitation of the coldest quarter) and altitude.
Figure 4. Frequency of maximum records for S. caffer (Sc).
18
However, even if the variables with less collinearity were selected, they could still be
correlated, since through these algorithms the VIF does not specifically evaluate the re-
lationship between all pairs of variables
11
. For this reason, the VIF was also calculated
"by hand" using the Spearman correlation coefficient between variables and selecting the
variables with a correlation less than 0.8 (Figure 6 and Figure 7).
Figure 6. Representative dendrogram of the distance matrix between variables.
Figure 7. Correlation between environmental variables.
11
https://www.rdocumentation.org/packages/usdm/versions/1.1-18/topics/vif
19
In this way, the variables finally selected are the mean temperature of warmest quarter or
BIO10, the annual precipitation or BIO12, the precipitation of wettest month or BIO13,
and the elevation or ALT. Figure 8 summarizes the VIF values of the variables resulting
from this process, obtained with the function vif.
Figure 8. Boxplot with VIF of selected environmental variables
Regarding the vegetation indices NDVI, SAVI and NDWI, two averaged images were
generated for each index through Google Earth Engine. The dry (_DS) period was calcu-
lated between June 1 and August 31, 2020 (Figure 9), and the wet (_WS) period between
October 1, 2020 and February 26, 2021 (Figure 10). Only the images with clouds cover-
ing less than 5% were analyzed.
21
Figure 10. Screenshot of the code used to calculate the NDVI, SAVI and NDWI indices for the wet season (WS) in the
AOI. The code snapshot can be consulted at:
https://code.earthengine.google.com/495078e067ef8812223a36f2d61c197e
After the subsequent multicollinearity tests in R, the analysis leads us to select the NDWI
of the wet season (NDWI_WS) and the one of the dry season (NDWI_DS) as the most
suitable for our study (Figure 11 and Figure 12).
22
Figure 11. Representation of the vegetation indices considered (RStudio)
Figure 12. Correlation between vegetation indices.
23
From the collinearity analysis performed for the whole dataset of RS-based indices, cli-
mate-selected variables, together with the human footprint (hfp_PN), low correlation val-
ues were obtained, with which the calibration of the models continued
12
. Figure 13,
Figure 14, and Figure 15 resume the 10 sets of pseudoabsence data generated for each
model.
Figure 13. Sets of pseudo-absences (PAs) generated for L. africana.
Figure 14. Sets of pseudo-absences (PAs) generated for H. amphibius.
12
Final files with the selected climatic variables, Human Footprint and NDWI in ascii format are provided
with the supplementary material https://gitfront.io/r/user-8903261/UwyHqoc69fPR/HWC-LNP-TFM/
24
Figure 15. Sets of pseudo-absences (PAs) generated for S. caffer.
After running the GLM, GBM, GAM, CTA, ANN, SRE, FDA, MARS, RF, and
MAXENT Phillips.2 algorithms from the biomod2 R package, the following TSS vs.
ROC assembly model performance plots are obtained, as well as the graphs of the re-
sponse curves of the variables that contribute the most to each model (Figure 16, Figure
17 and Figure 18, see enlarged figures in Annex III). The ensemble statistics shown in
general good results in TSS (0.72-0.78), ROC (0.91-0.95) and Kappa (0.4-0.65) with a
95% confidence or credible interval (level of significance of 0.05). Ensemble model sta-
tistics by ROC threshold are attached as Annex I.
Figure 16. TSS vs ROC assembly model performance (left) and variable response curves (right) for L. africana.
25
In general, better performance has been observed between the RF, Maxent, and GAM
algorithms, with the variables that show greater importance in the models being altitude
and BIO10 (mean temperature of the warmest month) for L. africana; altitude and BIO12
(annual precipitation) for S. caffer; and the dry season NDWI and again BIO12 for H.
amphibius, which is consistent with the biology of each species.
However, a closer look at the data shows that each model gives relatively different
weights to the study variables. For example, in L. africana, both GAM and RF models
detect greater importance of BIO10 and ALT, with a preponderance of the latter (Figure
20), observing that BIO10 might become more important with altitude (Figure 19). Also,
in both, the more relative importance of the climatic variables is detected compared to the
vegetation indices, although with greater importance of BIO13 compared to BIO12 in
Figure 17. TSS vs ROC assembly model performance (left) and variable response curves (right) for H. amphibius.
Figure 18. TSS vs ROC assembly model performance (left) and variable response curves (right) for S. caffer.
26
GAM that does not occur in RF. On the other hand,
in both models, NDWI_DS has a greater weight
than NDWI_WS, although while in GAM a signif-
icantly low contribution of hfp_PN to the model is
found, in RF this is above the metrics based on re-
mote sensors.
We see the case of H. amphibius, in which the best individual models are Maxent and RF.
Although in both models, the greater importance of NDWI_DS and BIO12 is detected
(without observing a relationship between them, Figure 21), the Maxent model grants
much greater relative importance to the first (Fig-
ure 22). In this case, it is noteworthy that, in gen-
eral, the relative importance of the variables is less
pronounced in RF, where, for example, BIO10 and
NDWI_WS have a relatively equivalent weight
(and just slightly bigger than BIO13), something
that contrasts with the results of Maxent, notice the
relative importance of NDWI_WS in this last.
Figure 19. 3D representation of the re-
sponse curve for altitud and BIO10 in the en-
semble model of L. africana.
Figure 20. Variable importance for GAM (left) and RF (right) models for L. africana.
Figure 21. 3D representation of the re-
sponse curve for BIO12 and NDWI_DS in
the ensemble model of H. amphibius.
27
Finally, we look at the best models for S. caffer, GAM, and Maxent. It can be seen that
both detect the more relative importance of the BIO12 and ALT variables (see the relation
of the assembled model in Figure 23), though the first
variable has even more weight in the GAM model
(Figure 24). GAM also shows the less relative dis-
tance between the variables NDWI, BIO13, and
hfp_PN. It can be noticed that among these four vari-
ables, Maxent gives a greater relative contribution to
hfp_PN. On the other hand, both models coincide in
the low relative contribution of BIO10 to the distribu-
tion models of the species.
Figure 22. Variable importance for Maxent (left) and RF (right) models for H. amphibius.
Figure 23. 3D representation of the re-
sponse curve for ALT and BIO12 in the
ensemble model of S. caffer.
Figure 24. Variable importance for GAM (left) and Maxent (right) models for S. caffer.
28
From the individual models obtained with the different methods the consensus models
were generated, from which the following maps of potential distribution and habitat avail-
ability of the species are obtained (Figure 25, Figure 26, and Figure 27).
Figure 25. Potential distribution (left) and habitat availability (right) of L. africana.
Figure 26. Potential distribution (left) and habitat availability (right) of H. amphibius.
29
To obtain the heat maps, the variables were previously normalized and reclassified for
their weighted sum (Figure 28, see enlarged figures and classification scheme in Annex
III).
Finally, from the subsequent operations and application of algorithms to get the statistical
index Getis-Ord Gi*, the following heat maps are obtained with the areas a greater con-
flict according to the potential distribution and the availability of habitat for each species
and the set of species (Figure 29, Figure 30, Figure 31, and Figure 32).
Figure 27. Potential distribution (left) and habitat availability (right) of S. caffer.
Figure 28. Representation of normalized and reclassified variables. Ls = livestock, Ch = chickens, Ct = cows, Dk =
ducks, Gt = goats, Pg = pigs, Sh = sheep (left), and sum of variables (right).
30
Figure 29. Heat maps with the greater conflict areas (green) according to potential distribution (left) and habitat
availability (right) for L. africana.
Figure 30. Heat maps with the greater conflict areas (green) according to potential distribution (left) and habitat
availability (right) for H. amphibius.
Figure 31. Heat maps with the greater conflict areas (green) according to potential distribution (left) and habitat
availability (right) for S. caffer.
31
We explore these areas in more detail through a GIS and Google Satellite Imagery
(Google tile map service) (Figure 33, Figure 34, Figure 35, Figure 36, Figure 37, Figure
38 and Figure 39):
Figure 33. Representation in QGIS 3.16 of the map of potential HWC areas.
Figure 32. Heat maps with the greater conflict areas (green) according to potential distribution (left) and habitat availa-
bility (right) for all three species of the study.
32
Figure 34. First point. The congruence of water points and natural vegetation with cultivated areas, adjacent to a
school center and a vehicle transit road is observed.
Figure 35. Second point. Urban settlements less than 2 km from the Great Limpopo River (National Park border).
33
Figure 36. Point 3. Urban settlements adjacent to one of the most significant water points south of the Limpopo Na-
tional Park are observed. There are highways and numerous dirt paths that can be used as wildlife corridors.
Figure 37. Point 4. Belonging to the Kruger National Park, south of the Olifants river. Area with numerous tourist
routes that intersect with watercourses.
34
Figure 38. Point 5. Urban settlements adjacent to natural watercourses are observed, together with numerous dirt
paths and human-made ponds.
Figure 39. Note the river basin that forms the border between Zimbabwe and South Africa (Greater Limpopo Trans-
frontier Park) and its abundantly vegetated plains, adjacent to which agricultural activities take place.
35
As can be seen, there are many potential conflict points around the Limpopo National
Park, especially distributed in the western half of it, if we consider the greater probability
of the presence of the species in this area. We know from the official livestock data ana-
lyzed in this work that much of the goat and sheep farming is concentrated in this western
half, the latter even more concentrated in the southwest.
Considering the availability of habitat, however, we observe increasing potential conflict
points that are also distributed to the east and southeast of the study area. In these areas,
even with a lower theoretical encounter probability (more significant in the case of the
hippopotamus), the conflicts can get relevant mainly due to the higher density of the ex-
isting population and the high concentration to the south of most livestock species (see
Figure 28).
4.2. Tools in the conflict mitigation strategy
The development effective tools for ecosystem-based solutions are a very important part
of HWC mitigation strategies, and local communities can play a central role in its imple-
mentation. There are many researchers and stakeholders working throughout the world to
promote effective practices for a sustainable coexistence and development, so, although
is not the main purpose of this work, it couldn’t be complete without a reference to some
of their last results directly linked to the Limpopo National Park.
Many authors address the effectiveness of wildlife crossing structures and/or road miti-
gation strategies (Green et al., 2018; Okita‐Ouma et al., 2021; Roque et al., 2022; Sch-
midt et al., 2021). For instance, Roque et al. (2022) studied the historical distribution and
movement patterns of seven large herbivore species in Limpopo National Park, among
them L. africana and S. caffer, founding evidence of the functioning of proposed wildlife
corridors in the protected area. However, the results give reason to assume that restoration
of populations is still in a very early and vulnerable state and that further efforts are nec-
essary to strengthen the slowly increasing populations.
Other research have investigated the functioning of translocation and resettlement
(Neelakantan et al., 2019; Roque et al., 2021; Tiller et al., 2022). Roque et al. (2021)
found species-specific and guild-specific responses, as well as an association of most re-
introduced large herbivores community parameters with habitat types rather than distance
36
to initial release, highlighting the importance of post-release monitoring of reintroduced
wildlife in the Limpopo National Park. On the other hand, Tiller et al (2022) tracked by
GPS five translocated elephants for 1 year on an hourly basis, analyzing home range,
displacement rates, problematic behavior, and group size. Three of the five elephants were
illegally killed, one continues breaking fences and raiding crops, and only one stayed
away from human settlement. These researchers also recommend careful consideration
of elephant social systems, age and timing, together with release site and proximity to
human settlements.
Regarding the use of fences to protect households, different outcomes has been reported
(Di Minin et al., 2021; Honda, 2022; Virtanen et al., 2021). Virtanen et al. (2021), after
three years studying this measure as a solution for the human-elephant conflict, pointed
out a reduction in the effectiveness of fences in the long term. The suggested underlying
reasons are the failure to establish a common understanding between the local population
and authorities about the risks versus benefits involved, and the elements for an accepta-
ble solution, including a stronger commitment with significantly more resources from the
government.
Another great proportion of research has address the social perspective of the conflict in
more depth, either studying compensation and other financial systems (Karanth et al.,
2013; Muriuki et al., 2017), the role of the media (van Houdt et al., 2021), or the risk
perception (Read et al., 2021; Sage et al., 2022) among others (Meyer & Börner, 2022;
Pereira et al., 2021; Sage et al., 2022). Pereira et al. (2021), studied the livelihood vul-
nerability index of rural households in Quirimbas National Park (north-eastern Mozam-
bique) finding a link between this index and human-wildlife interactions, with more vul-
nerable households taking greater risks and encountering wildlife (e.g., when fetching
water from rivers). They recommend consideration of livelihood strategies and commu-
nity vulnerability when designing conflict mitigation schemes and implementing conser-
vation measures.
Precisely regarding management frameworks, Gross et al. (2022), after reviewing the
management of conflicts with L. Africana through qualitative expert interviews, proposed
a management framework that was validated and adjusted with stakeholder participation
in two southern African projects, one of them in Mozambique. This framework considers
environmental, legal, socio-political, technical, and financial factors, as well as monitor-
ing steering all processes.
37
Finally, a lot of work on monitoring and evaluation projects has been carried out, whether
through individual-based movement or agent-based modeling (Diaz et al., 2021; He et
al., 2022, p.; Unnithan Kumar et al., 2022), models predicting shifts in distribution under
climate change and management (Fullman et al., 2017), modeling behavior on crop dam-
age or patterns in access to water sources (Buchholtz et al., 2020; Pozo et al., 2018; Se-
lebatso et al., 2018; Vogel et al., 2020), aerial sample surveys (Dunham, 2012), camera
trapping monitoring (Agha et al., 2018; Gaynor et al., 2018), among many others (Table
2).
Table 2. Summary of tools used in the HWC mitigation following the topic classification of the IUCN Species Survival
Commission
13
. *General applies when other species or all the three species are the focus of the study.
Topic
Tool
Referencias
LA / HA / SC / Ge-
neral*
Engaging with stakeholders
Effectiveness of wildlife
crossing structures and/or road
mitigation strategies
Okita-Ouma et al.,
2021
LA
Schmidt et al., 2021
General
Green et al., 2019
LA
Roque et al., 2022
LA, SC, General
Management of corridors
Gara et al., 2021
LA
Resettlement and landscape-
level conservation
Neelakantan et al.,
2019
General
Biologically relevant scales in
large mammal management
policies
Delsink et al., 2013
LA
Management framework
Gross et al., 2022
LA
Social research methods
Spatial analysis of households
(HHs) vulnerability level out-
comes
Meyer & Börner, 2022
General
Determine factors facilitating
and motivating unauthorised
hunting
Kisingo et al., 2021
SC, General
Mapping the spatial pattern of
attitudes toward wildlife
Sage et al., 2022
General
Stakeholder attitudes toward
the incentives used to mitigate
HWC
van Houdt et al., 2021
LA
Human movement influenced
by perceived risk of wildlife
encounters
Read et al., 2021
General
13
https://www.hwctf.org/document-library
38
Topic
Tool
Referencias
LA / HA / SC / Ge-
neral*
Livelihood Vulnerability In-
dex in mitigation schemes
Pereira et al., 2021
General
Social context of poaching
Lunstrum & Givá,
2020
General
Socioeconomic analysis
Virtanen et al., 2020
LA
Mbanze et al., 2019
General
Historical perspectives
Historical distribution and
movement patterns
Roque et al., 2022
LA, SC, General
Behaviour change & social
marketing
Open-access online courses
TRAFFIC
General
Guidance documents and key
papers
IUCN Species Survival
Commision
General
Compensation & other fi-
nancial instruments
The cost of livestock lost
Muriuki et al., 2017
General
Patterns of human-wildlife
conflicts and compensation
Karanth et al., 2013
General
Fences
Height and tension adjustment
Honda, 2022
General
Effectiveness
Di Minin et al., 2021
LA, General
Virtanen et al., 2020
LA
Livestock guarding
Guidance documents and key
papers
IUCN Species Survival
Commision
General
Translocation
GPS-tracking data of translo-
cated African savanna ele-
phants
Tiller et al., 2022
LA
Conflict analysis & theory
Framework for diagnosing
complex conservation con-
flicts
Harrison & Loring,
2020
General
Political ecology of conflicts
Guidance documents and key
papers
IUCN Species Survival
Commision
General
Cultural dimensions
Guidance documents and key
papers
IUCN Species Survival
Commision
General
Human dimensions theory
Guidance documents and key
papers
IUCN Species Survival
Commision
General
Role of the media
News media content analysis
van Houdt et al., 2021
LA
Monitoring & evaluation
Individual-Based Movement
Model for Connectivity Mod-
elling
Unnitah Kumar et al.,
2022
General
Predicting shifts in distribu-
tions under climate change and
management
Fullman et al., 2017
General
39
Topic
Tool
Referencias
LA / HA / SC / Ge-
neral*
Evaluation of Nature Reserve
Requirements
Guan et al., 2015
General
Gunshots passive acoustic re-
corders in combination with
occupancy models
Pardo et al., 2022
General
Interaction amongst wildlife,
livestock, and outdoor recrea-
tionists
Marion et al., 2022
General
Formozov–Malyshev–Pere-
leshin (FMP) method to con-
vert spoor counts into density
estimates
Ahlswede et al., 2018
General
Aerial sample surveys
Dunham, 2012
LA, SC, General
Agent-based modelling
(ABM)
He et al., 2022
LA
Diaz et al., 2021
LA
Anti-poaching efforts
Ngotima et al., 2022
LA
Spatial and temporal patterns
in access to water resources
Buchholtz et al., 2020
LA
Selebatso et al., 2018
General
Changing in crop-raiding
trends y HWC hotspots
Tiller et al., 2021
LA
Crop consuming behaviour
Vogel et al., 2020
LA
Camera trapping monitoring
Agha et al., 2018
General
Roque et al., 2022,
2021
LA, SC, General
Space use predictions and
crops damage
Pozo et al., 2018
LA
Approach to detect and dis-
criminate the objective species
from other species
Szenicer et al., 2021
LA
Governmental databases of
wildlife-vehicle collisions
Snow et al., 2015
General
Effects of human settlement
and roads on diel activity pat-
terns of elephants
Gaynor et al., 2018
LA
Other barriers
Guidance documents and key
papers
IUCN Species Survival
Commision
General
Deterrents & repellents
Photo trapping evaluation of
deterrents
Laguna et al., 2022
General
Law enforcement strategies
Ngorima et al., 2022
LA
40
One issue that threats to hamper the progress on conservation and human-wildlife positive
interaction is the continuing illegal and unregulated hunting for meat and ivory (found
not only in elephants but also in the canine teeth of hippos). Between 1980 and 1992,
most large herbivore species were absent in Limpopo National Park following the civil
war in Mozambique (Mackie et al., 2013; Roque et al., 2021, 2022). This situation were
followed by a period of intense poaching that keep being a problem in recent days, alt-
hough slight recovery on some populations has been observed since the designation of
Great Limpopo Transfrontier Park (Roque et al., 2022). Some authors pointed out the
necessity of strengthening law enforcement efforts (Ngorima et al., 2022; Roque et al.,
2022), having these shown promising results where improved. Ngorima et al. (2022)
highlights that these results come from the evolution and growing use of the technology
to increase coverage and patrol efforts, i.e., GPS collars, spatial monitoring and reporting
tools and Unmanned Aerial Systems (UAS).
On the other hand, Lunstrum & Givá (2020) captured the social context to better under-
stand the drivers behind poaching. Through fieldwork study consisting of interviews
within communities located in the southern section of the Limpopo National Park directly
involved in (rhino) poaching recruits, they show that economic factors are the most cen-
tral drivers of poaching on the ground-level and that, rather than mere poverty per se, they
are better captured in the concept of economic inequality.
5. Conclusions
Using the occurrence data and the most representative environmental variables as predic-
tors, consensus models have been generated using the Biomod tool in R from the individ-
ual models obtained with different methods. Efforts have been made to reduce the limi-
tations of this type of model by using a series of rules described in the scripts generated
14
.
However, nature is complex, and when generating models, a series of space-time prem-
ises are assumed that limit their accuracy.
The ensemble statistics shown in general good results in TSS (0.72-0.78), ROC (0.91-
0.95) and Kappa (0.4-0.65) values, with a 95% confidence or credible interval. RF,
Maxent, and GAM are the algorithms that shown better performance, with the variables
14
https://gitfront.io/r/user-8903261/UwyHqoc69fPR/HWC-LNP-TFM/
41
of most importance in the models being altitude and BIO10 (mean temperature of the
warmest month) for L. africana; altitude and BIO12 (annual precipitation) for S. caffer;
and the dry season NDWI and again BIO12 for H. amphibius, which is consistent with
the biology of each species.
Finally, after preparing and processing the elements involved in interactions with humans
(population and land use factors), heat maps have been obtained with the areas of a greater
risk of conflict according to the potential distribution and habitat availability of each spe-
cies and of the three species. We have found many potential conflict points around the
Limpopo National Park, mainly distributed in the western half of it, if we consider the
probability gradient of occurrence of the species.
This work has also reviewed the main tools that are being used to mitigate these conflicts,
having seen that human activity may be simultaneously associated with risk and reward
for animals (access to cultivated crops, and use roads as movement corridors). All species,
including human, try to avoid threats while exploiting opportunities along the boundaries
of the Limpopo National Park. This is consistent with the relatively low importance given
in the models to the human footprint factor, which a priori indicates us that it is not de-
terminant to explain the distribution of the large herbivores studied.
On the other hand, it has been seen that illegal hunting and poaching continues to be an
element that conservation cannot solve on its own, since it is a problem rooted not only
in law enforcement but also in economic inequality, and that as such must be addressed
(Lunstrum & Givá, 2020).
In any case, the use of new technologies shows a fundamental role (Ngorima et al., 2022)
not only to guide planning decisions but also to provide better means to surveillance pa-
trols in wild areas, increase the effectiveness of monitoring and the motivation and in-
volvement of the local population, which in turn leads to better communication and better
data to inform new decisions, so that positive human-wildlife interaction can ultimately
become the norm and not the exception.
42
6. Bibliography
Agha, M., Batter, T., Bolas, E. C., Collins, A. C., Gomes da Rocha, D., Monteza-Mo-
reno, C. M., Preckler-Quisquater, S., & Sollmann, R. (2018). A review of wild-
life camera trapping trends across Africa. African Journal of Ecology, 56(4),
694-701. https://doi.org/10.1111/aje.12565
Aiello-Lammens, M. E., Boria, R. A., Radosavljevic, A., Vilela, B., & Anderson, R. P.
(2015). spThin: An R package for spatial thinning of species occurrence records
for use in ecological niche models. Ecography, 38(5), 541-545.
https://doi.org/10.1111/ecog.01132
Buchholtz, E. K., Stronza, A., Songhurst, A., McCulloch, G., & Fitzgerald, L. A.
(2020). Using landscape connectivity to predict human-wildlife conflict. Biolog-
ical Conservation, 248, 108677. https://doi.org/10.1016/j.biocon.2020.108677
Chase, M. J., Schlossberg, S., Griffin, C. R., Bouché, P. J. C., Djene, S. W., Elkan, P.
W., Ferreira, S., Grossman, F., Kohi, E. M., Landen, K., Omondi, P., Peltier, A.,
Selier, S. A. J., & Sutcliffe, R. (2016). Continent-wide survey reveals massive
decline in African savannah elephants. PeerJ, 4, e2354.
https://doi.org/10.7717/peerj.2354
Cornélis, D., Melleti, M., Korte, L., Ryan, S. J., Mirabile, M., Prin, T., & Prints, H. H.
T. (2014). African buffalo (Syncerus caffer Sparrman, 1779). En M. Melleti & J.
Burton (Eds.), Ecology, Evolution and Behaviour of Wild Cattle: Implications
for Conservation (pp. 326-372). Cambridge University Press.
Di Minin, E., Slotow, R., Fink, C., Bauer, H., & Packer, C. (2021). A pan-African spa-
tial assessment of human conflicts with lions and elephants. Nature Communica-
tions, 12. https://doi.org/10.1038/s41467-021-23283-w
Diaz, S. G., DeAngelis, D. L., Gaines, M. S., Purdon, A., Mole, M. A., & van Aarde, R.
J. (2021). Development and validation of a spatially-explicit agent-based model
for space utilization by African savanna elephants (Loxodonta africana) based
on determinants of movement. Ecological Modelling, 447, 109499.
https://doi.org/10.1016/j.ecolmodel.2021.109499
Doughty, C. E., Roman, J., Faurby, S., Wolf, A., Haque, A., Bakker, E. S., Malhi, Y.,
Dunning, J. B., & Svenning, J.-C. (2016). Global nutrient transport in a world of
43
giants. Proceedings of the National Academy of Sciences, 113(4), 868-873.
https://doi.org/10.1073/pnas.1502549112
Dunham, K. M. (2012). Trends in populations of elephant and other large herbivores in
Gonarezhou National Park, Zimbabwe, as revealed by sample aerial surveys. Af-
rican Journal of Ecology, 50(4), 476-488. https://doi.org/10.1111/j.1365-
2028.2012.01343.x
Fullman, T. J., Bunting, E. L., Kiker, G. A., & Southworth, J. (2017). Predicting shifts
in large herbivore distributions under climate change and management using a
spatially-explicit ecosystem model. Ecological Modelling, 352, 1-18.
https://doi.org/10.1016/j.ecolmodel.2017.02.030
Gaynor, K. M., Branco, P. S., Long, R. A., Gonçalves, D. D., Granli, P. K., & Poole, J.
H. (2018). Effects of human settlement and roads on diel activity patterns of ele-
phants ( Loxodonta africana ). African Journal of Ecology, 56(4), 872-881.
https://doi.org/10.1111/aje.12552
GBIF. (s. f.). Global Biodiversity Information Facility. Recuperado 15 de agosto de
2022, de https://www.gbif.org
Getis, A., & Ord, J. K. (1992). The Analysis of Spatial Association by Use of Distance
Statistics. Geographical Analysis, 24(3), 189-206.
https://doi.org/10.1111/j.1538-4632.1992.tb00261.x
Gilbert, M., Nicolas, G., Cinardi, G., Boeckel, T. P. V., Vanwambeke, S. O., Wint, G.
R. W., & Robinson, T. P. (2018). Data Descriptor: Global distribution data for
cattle, buffaloes, horses, sheep, goats, pigs, chickens and ducks in 2010. SCIEN-
TIFIC DATA, 12.
Gobush, K. S., Edwards, C. T. T., Balfour, D., Wittemyer, G., Maisels, F., & Taylor, R.
D. (2021). Loxodonta africana (amended version of 2021 assessment). The
IUCN Red List of Threatened Species.
https://dx.doi.org/10.2305/IUCN.UK.2021-
2.RLTS.T181008073A204401095.en
Green, S. E., Davidson, Z., Kaaria, T., & Doncaster, C. P. (2018). Do wildlife corridors
link or extend habitat? Insights from elephant use of a Kenyan wildlife corridor.
African Journal of Ecology, 56(4), 860-871. https://doi.org/10.1111/aje.12541
44
Gross, E. M., Pereira, J. G., Shaba, T., Bilério, S., Kumchedwa, B., & Lienenlüke, S.
(2022). Exploring Routes to Coexistence: Developing and Testing a Human–El-
ephant Conflict-Management Framework for African Elephant-Range Countries.
Diversity, 14(7), 525. https://doi.org/10.3390/d14070525
Guisan, A., Petitpierre, B., Broennimann, O., Daehler, C., & Kueffer, C. (2014). Unify-
ing niche shift studies: Insights from biological invasions. Trends in Ecology &
Evolution, 29(5), 260-269. https://doi.org/10.1016/j.tree.2014.02.009
Guisan, A., & Thuiller, W. (2005). Predicting species distribution: Offering more than
simple habitat models. Ecology Letters, 8(9), 993-1009.
https://doi.org/10.1111/j.1461-0248.2005.00792.x
Guisan, A., Thuiller, W., & Zimmermann, N. E. (2017). Habitat Suitability and Distri-
bution Models: With Applications in R. Cambridge University Press.
https://doi.org/10.1017/9781139028271
Guisan, A., & Zimmermann, N. E. (2000). Predictive habitat distribution models in
ecology. Ecological Modelling, 135(2-3), 147-186.
https://doi.org/10.1016/S0304-3800(00)00354-9
He, H., Buchholtz, E., Chen, F., Vogel, S., & Yu, C. A. (Alex). (2022). An agent-based
model of elephant crop consumption walks using combinatorial optimization.
Ecological Modelling, 464, 109852.
https://doi.org/10.1016/j.ecolmodel.2021.109852
Honda, T. (2022). Height and tension of electric lines: How should an electric fence be
installed to effectively mitigate human-wildlife conflict? European Journal of
Wildlife Research, 68(5), 60. https://doi.org/10.1007/s10344-022-01606-6
IUCN Issues Brief. Human-Wildlife Conflict. (2022). https://www.iucn.org/resources/is-
sues-brief/human-wildlife-conflict
IUCN SSC Antelope Specialist Group. (2019). Syncerus caffer, African Buffalo. The
IUCN Red List of Threatened Species.
http://dx.doi.org/10.2305/IUCN.UK.2019- 1.RLTS.T21251A50195031.en
Karanth, K. K., Gopalaswamy, A. M., Prasad, P. K., & Dasgupta, S. (2013). Patterns of
human–wildlife conflicts and compensation: Insights from Western Ghats pro-
tected areas. Biological Conservation, 166, 175-185. https://doi.org/10.1016/j.bi-
ocon.2013.06.027
45
Lewison, R. (2007). Population responses to natural and human-mediated disturbances:
Assessing the vulnerability of the common hippopotamus (Hippopotamus am-
phibius). African Journal of Ecology, 45(3), 407-415.
https://doi.org/10.1111/j.1365-2028.2006.00747.x
Lewison, R., & Pluháček, J. (2017). Hippopotamus amphibius, Hippopotamus. The
IUCN Red List of Threatened Species.
http://dx.doi.org/10.2305/IUCN.UK.2017- 2.RLTS.T10103A18567364.en
Lunstrum, E., & Givá, N. (2020). What drives commercial poaching? From poverty to
economic inequality. Biological Conservation, 245, 108505.
https://doi.org/10.1016/j.biocon.2020.108505
Mackie, C. S., Dunham, K. M., & Ghiurghi, A. (2013). Current status and distribution
of the Vulnerable common hippopotamus Hippopotamus amphibius in Mozam-
bique. Oryx, 47(1), 70-76. https://doi.org/10.1017/S0030605311001554
Malhi, Y., Doughty, C. E., Galetti, M., Smith, F. A., Svenning, J.-C., & Terborgh, J. W.
(2016). Megafauna and ecosystem function from the Pleistocene to the Anthro-
pocene. Proceedings of the National Academy of Sciences, 113(4), 838-846.
https://doi.org/10.1073/pnas.1502540113
Mason, K. (2013). Hippopotamus amphibius. Animal Diversity Web. https://animal-
diversity.org/accounts/Hippopotamus_amphibius/
Matawa, F., Murwira, A., & Schmidt, K. S. (2012). Explaining elephant (Loxodonta af-
ricana) and buffalo (Syncerus caffer) spatial distribution in the Zambezi Valley
using maximum entropy modelling. Ecological Modelling, 242, 189-197.
https://doi.org/10.1016/j.ecolmodel.2012.05.010
Meyer, M., & Börner, J. (2022). Rural livelihoods, community-based conservation, and
human–wildlife conflict: Scope for synergies? Biological Conservation, 272,
109666. https://doi.org/10.1016/j.biocon.2022.109666
Meynard, C. N., & Kaplan, D. M. (2012). The effect of a gradual response to the envi-
ronment on species distribution modeling performance. Ecography, 35(6), 499-
509. https://doi.org/10.1111/j.1600-0587.2011.07157.x
Mod, H. K., Scherrer, D., Luoto, M., & Guisan, A. (2016). What we use is not what we
know: Environmental predictors in plant distribution models. Journal of Vegeta-
tion Science, 27(6), 1308-1322. https://doi.org/10.1111/jvs.12444
46
Mosepele, K., Moyle, P. B., Merron, G. S., Purkey, D. R., & Mosepele, B. (2009). Fish,
Floods, and Ecosystem Engineers: Aquatic Conservation in the Okavango Delta,
Botswana. BioScience, 59(1), 53-64. https://doi.org/10.1525/bio.2009.59.1.9
Muriuki, M. W., Ipara, H., & Kiringe, J. W. (2017). The cost of livestock lost to lions
and other wildlife species in the Amboseli ecosystem, Kenya. European Journal
of Wildlife Research, 63(4), 60. https://doi.org/10.1007/s10344-017-1117-2
Neelakantan, A., DeFries, R., & Krishnamurthy, R. (2019). Resettlement and landscape-
level conservation: Corridors, human-wildlife conflict, and forest use in Central
India. Biological Conservation, 232, 142-151. https://doi.org/10.1016/j.bio-
con.2019.01.033
Ng, C. (2015). Syncerus caffer. Animal Diversity Web. https://animaldiversity.org/ac-
counts/Syncerus_caffer/
Ngorima, P., Mpakairi, K. S., Kavhu, B., Gara, T. W., Ndaimani, H., & Chakuya, J.
(2022). Trends in elephant poaching in the MID‐ZAMBEZI Valley, Zimbabwe:
Lessons learnt and future outlook. African Journal of Ecology, aje.13039.
https://doi.org/10.1111/aje.13039
Okita‐Ouma, B., Koskei, M., Tiller, L., Lala, F., King, L., Moller, R., Amin, R., &
Douglas‐Hamilton, I. (2021). Effectiveness of wildlife underpasses and culverts
in connecting elephant habitats: A case study of new railway through Kenya’s
Tsavo National Parks. African Journal of Ecology, 59(3), 624-640.
https://doi.org/10.1111/aje.12873
Pearman, P. B., Guisan, A., Broennimann, O., & Randin, C. F. (2008). Niche dynamics
in space and time. Trends in Ecology & Evolution, 23(3), 149-158.
https://doi.org/10.1016/j.tree.2007.11.005
Pereira, J., Rosalino, L. M., Mucova, S., Massangue, Y., Abdulrazak, M., Vahossa, S.,
Selemane, M., Fonseca, C., & Santos, M. J. (2021). Livelihood vulnerability in-
creases human–wildlife interactions. Environmental Conservation, 48(4), 301-
309. https://doi.org/10.1017/S037689292100028X
Pozo, R. A., Cusack, J. J., McCulloch, G., Stronza, A., Songhurst, A., & Coulson, T.
(2018). Elephant space-use is not a good predictor of crop-damage. Biological
Conservation, 228, 241-251. https://doi.org/10.1016/j.biocon.2018.10.031
R Core Team. (2014). R: A language and environment for statistical computing. R
Foundation for Statistical Computing, Vienna, Austria. http://www.R-pro-
ject.org/.
47
Read, D. J., Habib, B., Stabach, J., & Leimgruber, P. (2021). Human movement influ-
enced by perceived risk of wildlife encounters at fine scales: Evidence from cen-
tral India. Biological Conservation, 254, 108945. https://doi.org/10.1016/j.bio-
con.2020.108945
Rodríguez Martínez, J. (2013). Ecología (3a). Ediciones Pirámide.
Roque, D. V., Göttert, T., Macandza, V. A., & Zeller, U. (2021). Assessing Distribution
Patterns and the Relative Abundance of Reintroduced Large Herbivores in the
Limpopo National Park, Mozambique. Diversity, 13(10), 456.
https://doi.org/10.3390/d13100456
Roque, D. V., Macandza, V. A., Zeller, U., Starik, N., & Göttert, T. (2022). Historical
and current distribution and movement patterns of large herbivores in the Lim-
popo National Park, Mozambique. Frontiers in Ecology and Evolution, 10,
978397. https://doi.org/10.3389/fevo.2022.978397
Sage, A. H., Hillis, V., Graves, R. A., Burnham, M., & Carter, N. H. (2022). Paths of
coexistence: Spatially predicting acceptance of grizzly bears along key move-
ment corridors. Biological Conservation, 266, 109468.
https://doi.org/10.1016/j.biocon.2022.109468
Scharf, A. K., Belant, J. L., Beyer, D. E., Wikelski, M., & Safi, K. (2018). Habitat suita-
bility does not capture the essence of animal-defined corridors. Movement Ecol-
ogy, 6(1), 18. https://doi.org/10.1186/s40462-018-0136-2
Schiavina, M., Freire, S., & MacManus, K. (2022). GHS-POP R2022A - GHS popula-
tion grid multitemporal (1975-2030). European Commission, Joint Research
Centre (JRC). https://data.jrc.ec.europa.eu/dataset/d6d86a90-4351-4508-99c1-
cb074b022c4a
Schmidt, G. M., Lewison, R. L., & Swarts, H. M. (2021). Pairing long-term population
monitoring and wildlife crossing structure interaction data to evaluate road miti-
gation effectiveness. Biological Conservation, 257, 109085.
https://doi.org/10.1016/j.biocon.2021.109085
Selebatso, M., Bennitt, E., Maude, G., & Fynn, R. W. S. (2018). Water provision alters
wildebeest adaptive habitat selection and resilience in the Central Kalahari. Afri-
can Journal of Ecology, 56(2), 225-234. https://doi.org/10.1111/aje.12439
Smitz, N., Berthouly, C., Cornélis, D., Heller, R., Van Hooft, P., Chardonnet, P., Caron,
A., Prins, H., van Vuuren, B. J., De Iongh, H., & Michaux, J. (2013). Pan-Afri-
48
can Genetic Structure in the African Buffalo (Syncerus caffer): Investigating In-
traspecific Divergence. PLoS ONE, 8(2), e56235. https://doi.org/10.1371/jour-
nal.pone.0056235
Thuiller, W. (2014). Editorial commentary on ‘Patterns and uncertainties of species’
range shifts under climate change’. Global Change Biology, 20(12), 3593-3594.
https://doi.org/10.1111/gcb.12727
Tiller, L. N., Humle, T., Amin, R., Deere, N. J., Lago, B. O., Leader-Williams, N., Si-
noni, F. K., Sitati, N., Walpole, M., & Smith, R. J. (2021). Changing seasonal,
temporal and spatial crop-raiding trends over 15 years in a human-elephant con-
flict hotspot. Biological Conservation, 254, 108941. https://doi.org/10.1016/j.bi-
ocon.2020.108941
Tiller, L. N., King, L. E., Okita‐Ouma, B., Lala, F., Pope, F., Douglas‐Hamilton, I., &
Thouless, C. R. (2022). The behaviour and fate of translocated bull African sa-
vanna elephants ( Loxodonta africana ) into a novel environment. African Jour-
nal of Ecology, aje.13038. https://doi.org/10.1111/aje.13038
Unnithan Kumar, S., Kaszta, Ż., & Cushman, S. A. (2022). Pathwalker: A New Individ-
ual-Based Movement Model for Conservation Science and Connectivity Model-
ling. ISPRS International Journal of Geo-Information, 11(6), 329.
https://doi.org/10.3390/ijgi11060329
van Houdt, S., Brown, R. P., & Traill, L. W. (2021). Stakeholder attitudes toward the
incentives used to mitigate human-elephant conflict in southern Africa: A news
media content analysis. Journal for Nature Conservation, 61, 125982.
https://doi.org/10.1016/j.jnc.2021.125982
Varela, S., Lima-Ribeiro, M. S., & Terribile, L. C. (2015). A Short Guide to the Cli-
matic Variables of the Last Glacial Maximum for Biogeographers. PLOS ONE,
10(6), e0129037. https://doi.org/10.1371/journal.pone.0129037
Virtanen, P., Macandza, V., Goba, P., Mourinho, J., Roque, D., Mamugy, F., & Langa,
B. (2021). Assessing tolerance for wildlife: Human-elephant conflict in Chi-
manimani, Mozambique. Human Dimensions of Wildlife, 26(5), 411-428.
https://doi.org/10.1080/10871209.2020.1834648
Vogel, S. M., Blumenthal, S. A., de Boer, W. F., Masake, M., Newton, I., Songhurst, A.
C., McCulloch, G., Stronza, A., Henley, M. D., & Coulson, T. (2020). Timing of
49
dietary switching by savannah elephants in relation to crop consumption. Bio-
logical Conservation, 249, 108703. https://doi.org/10.1016/j.bio-
con.2020.108703
Wiens, J. J., & Graham, C. H. (2005). Niche Conservatism: Integrating Evolution, Ecol-
ogy, and Conservation Biology. Annual Review of Ecology, Evolution, and Sys-
tematics, 36(1), 519-539. https://doi.org/10.1146/annurev.ecol-
sys.36.102803.095431
Wildlife Conservation Society-WCS, & Center For International Earth Science Infor-
mation Network-CIESIN-Columbia University. (2005). Last of the Wild Project,
Version 2, 2005 (LWP-2): Global Human Footprint Dataset (Geographic) [Data
set]. Palisades, NY: NASA Socioeconomic Data and Applications Center
(SEDAC). https://doi.org/10.7927/H4M61H5F
Zurell, D., Franklin, J., König, C., Bouchet, P. J., Dormann, C. F., Elith, J., Fandos, G.,
Feng, X., Guillera‐Arroita, G., Guisan, A., Lahoz‐Monfort, J. J., Leitão, P. J.,
Park, D. S., Peterson, A. T., Rapacciuolo, G., Schmatz, D. R., Schröder, B.,
Serra‐Diaz, J. M., Thuiller, W., … Merow, C. (2020). A standard protocol for
reporting species distribution models. Ecography, 43(9), 1261-1277.
https://doi.org/10.1111/ecog.04960
Table index
Table 1. WorldClim' Bioclimatic variables. Source: Own elaboration based on
https://www.worldclim.org/data/bioclim.html ............................................................... 12
Table 2. Summary of tools used in the HWC mitigation following the topic
classification of the IUCN Species Survival Commission. *General applies when other
species or all the three species are the focus of the study............................................... 37
Index of Figures
Figure 1. Boundary of the Limpopo National Park. Source: Own elaboration in QGIS
3.16 from official cartography. Full-extent map in Annex II. .......................................... 1
Figure 2. The five main modelling steps in the species distribution modelling cycle.
Source: Zurell et al., 2020 .............................................................................................. 10
Figure 3. Frequency of maximum records for H. amphibius (left), and for L. Africana
(right). ............................................................................................................................. 16
Figure 4. Frequency of maximum records for S. caffer (Sc). ........................................ 17
Figure 5. Most significant environmental variables according to Vifcor: bio6 (Minimum
temperature of the coldest month), bio10 (Average temperature of the warmest quarter),
50
bio12 (Annual Precipitation), bio13 (Precipitation of the rainiest month), bio18
(Precipitation of the warmest quarter) , bio19 (Precipitation of the coldest quarter) and
altitude. ........................................................................................................................... 17
Figure 6. Representative dendrogram of the distance matrix between variables. ......... 18
Figure 7. Correlation between environmental variables. ............................................... 18
Figure 8. Boxplot with VIF of selected environmental variables .................................. 19
Figure 9. Screenshot of the code used to calculate the NDVI, SAVI and NDWI indices
for the dry season (DS) in the AOI. ................................................................................ 20
Figure 10. Screenshot of the code used to calculate the NDVI, SAVI and NDWI indices
for the wet season (WS) in the AOI. .............................................................................. 21
Figure 11. Representation of the vegetation indices considered (RStudio)................... 22
Figure 12. Correlation between vegetation indices. ...................................................... 22
Figure 13. Sets of pseudo-absences (PAs) generated for L. africana. ........................... 23
Figure 14. Sets of pseudo-absences (PAs) generated for H. amphibius. ....................... 23
Figure 15. Sets of pseudo-absences (PAs) generated for S. caffer. ............................... 24
Figure 16. TSS vs ROC assembly model performance (left) and variable response
curves (right) for L. africana. ......................................................................................... 24
Figure 17. TSS vs ROC assembly model performance (left) and variable response
curves (right) for H. amphibius. ..................................................................................... 25
Figure 18. TSS vs ROC assembly model performance (left) and variable response
curves (right) for S. caffer............................................................................................... 25
Figure 19. 3D representation of the response curve for altitud and BIO10 in the
ensemble model of L. africana. ...................................................................................... 26
Figure 20. Variable importance for GAM (left) and RF (right) models for L. africana.
........................................................................................................................................ 26
Figure 21. 3D representation of the response curve for BIO12 and NDWI_DS in the
ensemble model of H. amphibius. .................................................................................. 26
Figure 22. Variable importance for Maxent (left) and RF (right) models for H.
amphibius. ...................................................................................................................... 27
Figure 23. 3D representation of the response curve for ALT and BIO12 in the ensemble
model of S. caffer. .......................................................................................................... 27
Figure 24. Variable importance for GAM (left) and Maxent (right) models for S. caffer.
........................................................................................................................................ 27
Figure 25. Potential distribution (left) and habitat availability (right) of L. africana. .. 28
Figure 26. Potential distribution (left) and habitat availability (right) of H. amphibius.
........................................................................................................................................ 28
Figure 27. Potential distribution (left) and habitat availability (right) of S. caffer. ....... 29
Figure 28. Representation of normalized and reclassified variables. Ls = livestock, Ch =
chickens, Ct = cows, Dk = ducks, Gt = goats, Pg = pigs, Sh = sheep (left), and sum of
variables (right)............................................................................................................... 29
Figure 29. Heat maps with the greater conflict areas (green) according to potential
distribution (left) and habitat availability (right) for L. africana. .................................. 30
Figure 30. Heat maps with the greater conflict areas (green) according to potential
distribution (left) and habitat availability (right) for H. amphibius. ............................... 30
Figure 31. Heat maps with the greater conflict areas (green) according to potential
distribution (left) and habitat availability (right) for S. caffer. ....................................... 30
51
Figure 32. Heat maps with the greater conflict areas (green) according to potential
distribution (left) and habitat availability (right) for all three species of the study. ....... 31
Figure 33. Representation in QGIS 3.16 of the map of potential HWC areas. ............. 31
Figure 34. First point. The congruence of water points and natural vegetation with
cultivated areas, adjacent to a school center and a vehicle transit road is observed. ..... 32
Figure 35. Second point. Urban settlements less than 2 km from the Great Limpopo
River (National Park border). ......................................................................................... 32
Figure 36. Point 3. Urban settlements adjacent to one of the most significant water
points south of the Limpopo National Park are observed. There are highways and
numerous dirt paths that can be used as wildlife corridors............................................. 33
Figure 37. Point 4. Belonging to the Kruger National Park, south of the Olifants river.
Area with numerous tourist routes that intersect with watercourses. ............................. 33
Figure 38. Point 5. Urban settlements adjacent to natural watercourses are observed,
together with numerous dirt paths and human-made ponds. .......................................... 34
Figure 39. Note the river basin that forms the border between Zimbabwe and South
Africa (Greater Limpopo Transfrontier Park) and its abundantly vegetated plains,
adjacent to which agricultural activities take place. ....................................................... 34
50
Annex I
ENSEMBLE MODEL STATISTICS BY ROC THRESHOLD 0.95
Testing data Cutoff Sensitivity Specificity Testing data Cutoff Sensitivity Specificity Testing data Cutoff Sensitivity Specificity
KAPPA 0.406 731 61.404 89.894 0.417 690 83.041 84.22 0.409 746 61.404 89.894
TSS 0.722 578 94.152 78.014 0.736 573 95.906 77.541 0.722 598 93.567 78.546
ROC 0.911 583.5 94.152 78.487 0.909 580 95.906 78.132 0.912 596.5 94.152 78.546
KAPPA 0.624 774 65.789 96.579 0.614 811 65.789 96.316 0.646 758 63.158 97.632
TSS 0.763 591 92.105 83.947 0.768 528 92.105 80.526 0.766 583 92.105 84.474
ROC 0.946 595 92.105 84.211 0.947 615 92.105 85.526 0.945 585.5 92.105 84.737
KAPPA 0.508 730 74.737 90.928 0.527 779 74.737 91.667 0.508 739 74.737 90.928
TSS 0.776 481 97.895 79.747 0.781 563 94.737 83.228 0.776 487 97.895 79.747
ROC 0.939 515.5 96.842 80.907 0.938 566 94.737 83.544 0.939 518.5 96.842 80.802
S. caffer
Median
Mean
Weighed mean
L. Africana
H. amphibius
52
53
57
Figure 28. Representation of normalized and reclassified variables. Ls = livestock, Ch = chick-
ens, Ct = cows, Dk = ducks, Gt = goats, Pg = pigs, Sh = sheep, and sum of variables.
RECLASSIFIED VARIABLES
Classification scheme
0-0.5 0 0-0.1 0 0-20 0
0.5-30 1 0.1-30 1 20-40 1
30-40 2 30-60 2 40-60 3
40-60 4 60-100 3 60-80 0
60-100 6 80-117 1
117-121 0
121-127 1
127-250 0
Population density
Livestock
Landcover
