Forest elephant movement and habitat use in a tropical forest-grassland mosaic in Gabon
Forest elephant movement and habitat use in a tropical forest-grassland mosaic in Gabon
Emily C. Mills 1 2 3
John R. Poulsen 1 2 3
J. Michael Fay 0 1 3
Peter Morkel 1 3
Connie J. Clark 1 2 3
Amelia Meier 1 2 3
Christopher Beirne 1 2 3
Lee J. T. White 0 1 3
0 Agence Nationale des Parcs Nationaux , Batterie IV, Libreville, Gabon, 3 Independent Researcher, Karasburg , Namibia , 4 Institut de Recherche en EÂcologie Tropicale, Libreville, Gabon, 5 African Forest Ecology Group, School of Natural Sciences, University of Stirling , Scotland , United Kingdom
1 Data Availability Statement: The data are politically sensitive because public availability of the data could allow poachers to target study elephants. As such, data cannot be made publicly available, however the data can be made available on request from Duke University (Contact: John Poulsen
2 Nicholas School of the Environment, Duke University , Durham, North Carolina , United States of America
3 Editor: Elissa Z. Cameron, University of Tasmania , AUSTRALIA
Poaching of forest elephants (Loxodonta cyclotis) for ivory has decimated their populations
in Central Africa. Studying elephant movement can provide insight into habitat and resource
use to reveal where, when, and why they move and guide conservation efforts. We fitted 17
forest elephants with global positioning system (GPS) collars in 2015 and 2016 in the
tropical forest-grassland mosaic of the Wonga WongueÂ Presidential Reserve (WW), Gabon.
Using the location data, we quantified movement distances, home ranges, and habitat use
to examine the environmental drivers of elephant movements and predict where elephants
occur spatially and temporally. Forest elephants, on average, traveled 2,840 km annually
and had home ranges of 713 km2, with males covering significantly larger home ranges than
females. Forest elephants demonstrated both daily and seasonal movement patterns.
Daily, they moved between forest and grassland at dawn and dusk. Seasonally, they spent
proportionally more time in grassland than forest during the short-wet season when grasses
recruit. Forest elephants also traveled faster during the short-wet season when fruit
availability was greatest, likely reflecting long, direct movements to preferred fruiting tree
species. Forest elephants tended to select areas with high tree and shrub density that afford
cover and browse. When villages occurred in their home ranges elephants spent a
disproportionate amount of time near them, particularly in the dry season, probably for access to
agricultural crops and preferred habitat. Given the importance of the grassland habitat for
elephants, maintenance of the forest-grassland matrix is a conservation priority in WW. Law
enforcement, outreach, and education should focus on areas of potential human-elephant
conflict near villages along the borders of the reserve. GPS-tracking should be extended
into multi-use areas in the peripheries of protected areas to evaluate the effects of human
disturbance on elephant movements and to maintain connectivity among elephant
populations in Gabon.
Gabon Parks Agency (Agence Nationale des Parcs
Funding: The authors received no specific funding
for this work.
Poaching of forest elephants, Loxondonta cyclotis, for ivory is decimating their populations [
Between 2002 and 2011, populations in Central Africa decreased by 62% and lost 30% of their
geographical range due to global demand for ivory [
]. The densely forested country of Gabon
is one of the last strongholds of forest elephants, housing about half of the world's surviving
forest elephants [
]. But poaching has taken a toll on Gabon's elephant
populations±approximately 25,000 elephants were lost from the MinkeÂbeÂ National Park in a decade .
With forest elephants under intense poaching pressure, information on their habitat use,
movements, and ecology is necessary to maximize the effectiveness of conservation efforts.
Movement ecology can provide insights into species' resource requirementsÐfood, water, and
spaceÐand elucidate temporal and spatial patterns of habitat use. Using global positioning
systems (GPS) technology, animal movement studies have guided management strategies for a
variety of terrestrial species [
], including identifying critical habitat for endangered species
] and understanding resource-driven animal migrations . Tracking movement can also
detect changes to animal behavior in landscapes disturbed by urban development, extractive
land use, and recreation, with implications for managing these activities [
data from wide-ranging animals, such as wolves and elephants, have provided the empirical
basis for identifying and securing movement corridors, particularly between large, protected
wilderness areas [
Most current knowledge of the movement ecology of elephants comes from studies of
savanna elephants, Loxidonta africana, in southern [13±15] and eastern Africa [16±18].
Vegetation (tree cover and food resources) and water limitation during the dry seasons are the
main drivers of savanna elephant movements [13±15,17,18]. Savanna elephants are most active
at night when temperatures fall [
], avoiding areas of high human density outside of
protected areas [
]. In areas of low human density, elephants tend to avoid settlements during
the daytime, with males more likely to approach villages than females at night [
By contrast, studies of forest elephant movement are rare because of the difficulty of
collaring elephants in dense forests and, until recently, the challenge of reliably transmitting GPS
signals through the canopy. Only five studies to date have used GPS collars to monitor forest
elephant movements: a preliminary study successfully tracked one female elephant [
two additional studies derived descriptive metrics of home ranges and activity patterns
]. Another study used random walk models to characterize forest elephant movements
. The largest study, consisting of 28 forest elephants, is the only study to use GPS collar
data to model determinants of elephant movements, finding that unprotected roads acted as
major barriers to movement [
]. These studies demonstrate that forest elephant movements
are constrained by human disturbance except in the rare areas where elephants are safe from
], and that home range size varies greatly across both protected and human-use
zones. Much remains to be learned regarding the relative effects of ecological and
anthropogenic drivers of forest elephant movements and how they change across seasons, sites, and
levels of protection.
In 2015 and 2016, the Gabon Parks Agency (ANPN) collared 17 forest elephants in the
Wonga WongueÂ Presidential Reserve (WW) to assess their movements in relation to
environmental and anthropogenic variables. Situated on the western coast of Gabon, WW consists of
a forest-grassland matrix and is surrounded by 59 villages. Using hourly GPS locations, we
characterize movements, estimate home ranges, and model the drivers of movements to test
the following hypotheses:
1. Male forest elephants travel farther and cover larger home ranges than females because of
their greater exploratory movements and lower risk avoidance;
2 / 17
2. Forest elephants spend more time in forest than grassland during the dry season and
daytime to avoid exposure to sunlight, high temperatures, and poaching;
3. Forest elephants avoid roads and villages because of poaching threats, with females
demonstrating greater avoidance than males;
4. During the dry season when water is limiting, forest elephants stay relatively close to
permanent water sources, particularly streams and lakes.
In this paper, we improve our understanding of the movement ecology of forest elephants
in a relatively well-protected reserve in Gabon. Based on our results, we propose management
strategies for the conservation of forest elephants.
Materials and methods
Forest elephant data
In October 2015, PM led a field team to fit 12 forest elephants with GPS collars in WW and
collared five additional elephants in April 2016 (S1 Table). The Gabon Parks Agency (ANPN)
and Center for Scientific Research and Technology (CENAREST) reviewed and approved all
darting and collaring methods, including the research ethics, before the work was conducted.
The Duke Institutional Animal Care and Use Committee (IACUC) also reviewed the research,
but no approval number was obtained because Duke's role in the research began after the
elephants were collared by ANPN. All the elephants were darted with a standard dose of 5mg of
etorphine hydrochloride (Captivon). Sterile water was added to the etorphine to achieve a dart
volume of 2 ml. The darts had a Cap-Chur 2cc aluminium barrel, neoprene plunger, and 1-3cc
internal charge. We employed a Joubert Capture Equipment 45 mm x 3.5 mm barbed stainless
steel needle and plastic flight. To fire the darts, we used a Dan-inject JM Special dart gun,
powered by compressed carbon dioxide, with a 13 mm smooth barrel and a red-dot sight.
A standard dose of 5 mg was used for all adult elephants. When darting elephants in the
forest with low visibility, the sex and age class of the animal are difficult to ascertain. Thus, we
selected a dose that would stop an adult male, but not overdose an adult female. If darted in a
good muscle mass, the elephant was usually recumbent within seven minutes. If the animal
was lying on its sternum, the field team would push it onto its side. Most elephants were
immediately given 10 mg butorphanol intravenously to improve ventilation and blood oxygenation.
We carefully observed blood oxygenation and respiratory rate and depth and monitored heart
rate with a pulse oximeter. Care was taken to ensure that airflow through the trunk was
unobstructed and the trunk was held above pooled water if necessary. The dart wound was treated
with 4 ml of 100 mg/ml oxytetracycline, and after fitting the collar, the elephants were woken
up with 12 mg diprenorphine (Activon) and 50 mg of naltrexone (Trexonil) given
The focal elephants consisted of seven females and 10 males, with age categories ranging
from juveniles to old adults (S1 Table). The GPS collars transmitted coordinates hourly, with a
successful transmission rate of approximately 90%. In this study, we use data from November
4, 2015 to March 4, 2017, consisting of 158,755 locations.
Within WW (425,000 ha), habitats vary from white sand beaches and mangrove wetlands on
the Atlantic coast to a mosaic of open grasslands and tropical forest in the interior (Fig 1).
Grassland covers 15% of the reserve, with the central grassland (64,000 ha) forming a tropical
forest-grassland mosaic. ANPN conducts prescribed burns annually during the long-dry
3 / 17
Fig 1. Study area. (A) Location of Wonga WongueÂ Presidential Reserve in Gabon. (B) Elevation map of WW: elevation is low
except for the raised central grassland. (C) Location of collaring sites (yellow points) for 17 forest elephants in Wonga WongueÂ
Presidential Reserve. Green land cover is forest, brown land cover is grassland, and small white dots indicate clouds. Human
population is low in the reserve, but 59 villages (red points) are located within 10 km of its border.
season to maintain the grasslands, which would otherwise be overtaken by forest
encroachment. Human population density is 0.2 people km-2, but 59 villages are located within 10 km
of the reserve's southwest and southeast borders (Fig 1). Elephants are free-roaming in the
area and the reserve is well protected±no elephants had been poached within WW from 2014
to 2016 (pers. comm. David Fine).
Land cover classification
To create a land cover map and examine landscape characteristics such as vegetation and
wetness, we obtained Landsat 8 OLI/TIRS imagery (30 m resolution; S2A Table) from the U.S.
Geological Survey (USGS) [
]. We used four images to cover the study area, one pair from
the long-dry season and one pair from the long-wet season (S2B Table). All remote sensing
analyses were performed in ENVI version 5.3 [
], unless otherwise specified. To ground truth
4 / 17
the remote sensing imagery, ECM collected habitat and vegetation data in WW from May-July
2016, characterizing 220 recent elephant locations (S1 Fig; S3A Table). We created a thematic
land cover map for the study area using the pair of images taken during the dry season because
the ground truth data was collected during the dry season. To distinguish land cover
categories, we employed maximum likelihood supervised classification using regions of interest
(ROIs) created from the ground truth points as a training sample (S3B Table). We performed
an accuracy assessment on the final classification using randomly generated ground truth
points and Google Earth to evaluate the performance of the supervised classification technique.
We excluded unclassified and rare pixels (those containing sand, chalk, or other habitat)
because we were primarily interested in correctly classifying forest vs. grassland types. For
both dry and wet season images, we created spectral enhancement bands for enhanced
vegetation index (EVI), using coefficients adopted from the MODIS-EVI algorithm [
Tasseled Cap Transformation (TCT) bands for Brightness, Greenness, and Wetness, using
coefficients derived by Baig et al. [
]. We used these rasters as habitat covariates in habitat
modeling (see Drivers of forest elephant movements methods). Detailed classification methods
are presented in S1 Text.
Forest elephant movements and home ranges
To characterize elephant movements, we mapped and analyzed track distances across
temporal variables (time of day and season) and elephant sex. We used generalized linear mixed
models (GLMM) to evaluate whether these variables influence hourly rates and daily distances
of movement (see Model selection details below). For hourly movement rates, we discarded
coordinates recorded more than one hour after the previous coordinate, and treated individual
elephant and hour as nested random effects. For daily movements, we only used days with at
least 23 recorded locations, and treated individual elephant and day as random effects. Because
of our small sample size (17 elephants), we tested for differences in total distance traveled by
sex and season by bootstrapping the data 1,000 times with replacement and calculating 95%
confidence intervals. To examine the spatial extent of individual forest elephant movements
across time and to visualize high-use habitat areas, we generated home range polygons (see
below) and again used bootstrapping to test for differences in home range area across sex and
We defined `home range' as the area used by a given individual during its routine activities
(including, but not limited to, foraging, mating, and predator avoidance) [
] over a specified
period of time (in this caseÐannual or seasonal). We employed two different methods to
estimate and visualize home ranges: minimum convex polygon (MCP) and kernel utilization
distribution (KUD) [
]. We produced 100% and 95% MCPs as well as 95% and 50% KUD
home ranges (using the reference bandwidth for the smoothing parameter, h; see  for
equations). The MCP and KUD analyses were conducted in the adehabitat package in R
version 3.3.1 [
]. The minimum bounding geometry and kernel density tools in ArcMap
version 10.4.1  were used for mapping MCP and KUD home ranges.
Forest elephant habitat use
To examine patterns of forest elephant habitat use, we extracted the land cover type at each
hourly GPS location of the collared elephants. Time of day was split into four equal categories,
using breaks at the approximate sunrise and sunset times: 0:00±5:59, 6:00±11:59, 12:00±17:59,
and 18:00±23:59. We defined seasons using precipitation data from the Tropical Rainfall
Measuring Mission (TRMM) from November 2015 to November 2016, with dry seasons
characterized by < 60mm total rainfall as defined by the Koppen climate classification. The long-dry
5 / 17
Fig 2. Forest elephant habitat use by season. (Above) Average precipitation (mm) by month, from which seasons
were defined as the short-wet season (Oct. 3 ±Dec. 6), short-dry season (Dec. 7 ±Jan. 13), long-wet season (Jan. 14 ±
May 6), and long-dry season (May 7 ±Oct. 2). Note that depicting rainfall by month fails to show the daily variation in
precipitation that determines the seasons. (Below) Percentage of elephant locations in each land cover type by season.
Pairwise comparisons with different letters indicate significant differences in proportion of use by elephants across
seasons within land cover type. Error bars represent 95% confidence intervals.
season (May 7 ±Oct. 2) had no single rain event > 10 mm, the short-dry season (Dec. 7 ±Jan.
13) had no single rain event > 20 mm, and the remaining two periods were designated as the
long- (Jan. 14 ±May 6) or short- (Oct. 3 ±Dec. 6) wet seasons based on their duration (Fig 2).
We used one-way ANOVA with post-hoc pairwise comparisons (Bonferroni correction) to
analyze differences in land cover types at elephant locations across seasons and times of day.
Drivers of forest elephant movements
To determine the main drivers of forest elephant movements in WW, we first assembled 11 a
priori candidate variables hypothesized to influence elephant movement (S4 Table). Variables
included distance and slope rasters, which we developed using ArcMap and original spatial
data provided by ANPN. We also included spectral enhancement bands (as described above in
Land cover classification methods) and created a raster as a measure of forest openness, by
estimating the proportion of pixels classified as forest within a 90 m circular neighborhood.
6 / 17
We then built a habitat model using a generalized linear mixed model (GLMM) with a
presence/pseudoabsence design [
]. Presence points were GPS locations of elephants and
pseudoabsence points±a representative sample of available habitat not used within home ranges±
were randomly selected points within the 100% MCP home range of each elephant.
Pseudoabsence points could be randomly selected from the same pixel (900 m2) as other pseudoabsence
points, but a 50 m buffer was maintained around presence points to prevent contamination of
presence and pseudoabsence points (within the same 30 m cell). We thinned presence points
to one point per day to minimize temporal and spatial interdependence of consecutive points
] and generated a single pseudo-absence point per day.
We modeled wet and dry seasons separately to account for seasonal differences in
availability of resources, such as grass, fruiting trees, and precipitation. We used mixed-effects logistic
regression to model elephant presence for dry and wet seasons, with candidate variables
treated as fixed effects and elephant identity treated as a random effect to account for lack of
independence of points and unequal sample sizes among individuals [
]. To avoid flawed
inference arising from multicollinearity, we evaluated correlations between all candidate
variables (n = 11), removing one of the variables from any highly correlated pair (where r > 0.7).
This resulted in a final set of 6 candidate variables: distance to nearest road (km), distance to
the nearest village (km), distance to the nearest stream (km), slope (m), Enhanced Vegetation
Index (EVI), and sex. Continuous covariates were standardized as z-scores to facilitate
comparison of effect sizes. Given that all remaining covariates were uncorrelated, we tested all
combinations of covariates. This resulted in a suite of 64 candidate models for each season (see
Model selection details below). For each best-supported seasonal habitat model, we also
employed k-fold cross validation with a testing ratio of 20% to evaluate model performance
Model selection details. We applied an information theoretic approach to model
] whereby candidate models were ranked using Akaike's Information Criterion
corrected for small sample size (AICc). For each season, we defined a `top model set' which
included all models with a ΔAIC 6 from the best supported model, after excluding any
models of which a simpler nested version attained stronger support (following the `nesting rule' of
]). This approach avoids the issue of selecting and interpreting spurious covariates.
Informative covariates are discussed in terms of their relative importance, defined here as the sum
of Akaike weights (SW) across all of the `top-set' models in which the covariate occurs, and
absolute importance, defined here as the model averaged effect-size for each covariate.
GLMMs were fitted in the lme4 package in R [
Land cover classification
The land cover map accurately classified the three most common land cover categories (forest,
grassland, water) with 98.4% accuracy (Kappa = 0.977; S5 Table; S2 Fig). Mangroves and
swamps were poorly distinguished from forest because of their restricted representation and
small number of ground truth points.
Forest elephant movements and home ranges
There was no support for sex differences in hourly movement rates (Table 1). Elephants
traveled significantly faster in grassland than in any other land cover type (Table 1). Elephants
moved faster near dawn (6:00±8:00) and dusk (17:00±20:00) than other time periods (S3 Fig),
but hourly movement rates varied significantly with season (significant interaction between
time of day and season, Table 1). Similarly, female and male elephants did not exhibit
7 / 17
Sums of weight (SW), coefficients (Coef), and confidence intervals (CI), from the model averaged top model set GLMM of hourly (above) and daily (below) movement
rates of elephants. Intercepts relate to grassland, dry season, and the 00:00±05:59 time period for the hourly movement analysis, and dry season for the daily movement
analysis. Contrasts are only shown for fixed effects that have some support under model selection (SW > 0). For full model selection output see S6 Table.
significant differences in daily movement rates, although elephants traveled significantly
greater distances during the wet season than the dry season (Table 1; S4 Fig).
Elephants traveled an average of eight km per day and 3,444 km over the study period (Fig
3; S7 Table). The 12 elephants with at least one full year of location data traveled 2,840 km
annually on average. Total distance traveled by elephants did not differ significantly by sex or
season (Table 2). The average 100% MCP home range, which has been most commonly used
in previous studies, was 713 km2 (Table 2; S8 Table). Using 100% MCP, male elephants
covered significantly larger home ranges than female elephants, but home ranges did not differ
significantly between wet and dry seasons (Fig 3; Table 2; S9 Table; S5 Fig). The size of core
areas, represented by 50% KUD areas, did not differ between sexes (Table 2; S10 Table).
Forest elephant habitat use
Forest elephants spent 62% of their time in forests and 33% in grasslands (males: 78% forest,
20% grassland; females: 68% forest, 30% grassland; Fig 2). Use of forest varied among seasons,
with significantly more locations in forest during the long-dry than the short-wet season
(ANOVA: F3,64 = 4.2, p = 0.009; Fig 2). During daytime hours, when sun exposure and
8 / 17
Fig 3. Forest elephant tracks and home ranges. Movement tracks (A and C) and 100% MCP home ranges (B and D) for female elephants (top row) and a subset of
male elephants (bottom row). Female elephants tended to stay closer to the central grassland, whereas half of the males traveled up to 110 km from the central
grassland to core home range areas in the reserve periphery.
temperatures are highest, elephants spent significantly more time in forest than grassland
(ANOVA: F3,64 = 24.4, p < 0.001; S6 Fig). Peak usage of forest occurred between 6:00±12:00,
and peak usage of grassland occurred between 18:00±24:00 (S6 Fig).
Drivers of forest elephant movements
During the dry season, there was full support for distance to villages and EVI in predicting
forest elephant habitat use (Table 3), whereas distance to streams and slope had weaker support.
There was no support for distance to roads or sex in influencing elephant habitat use (Table 3).
Distance to village most strongly predicted elephant presence, with twice the effect size of EVI.
Elephant presence decreased with increasing distance from the village, whereas EVI, distance
to stream, and slope had weaker, positive effects on elephant presence.
During the wet season, there was full support for EVI, distance to streams, and distance to
roads in predicting forest elephant habitat use, and weaker support for distance to village
9 / 17
Bootstrapped 95% confidence intervals (CI) for total distance traveled and various home range metrics. Total distance traveled was calculated over the 10 months for
which data was available for all elephants (May 2016-March 2017). Bolded metric names denote statistical significance between variables based on non-overlapping 95%
CIs. See supporting information for summary distance metrics (S7 Table) and home range metrics (S8, S9 and S10 Tables) by individual elephant.
(Table 3). There was no support for slope or sex. EVI most strongly influenced habitat use,
with the probability of elephant presence increasing 25% with each one-unit increase in
standardized EVI. The next strongest effect was distance to stream, which positively affected
elephant presence. Distance to road and distance to village had weaker, negative effects on
Our study highlights distinct differences in temporal and seasonal use of habitat by forest
elephants in a forest-grassland matrix. Forest elephants spend most of their time in forest, but
selectively use grassland during nighttime hours and during the short-wet season, when young
grass shoots provide browse. Grassland habitat is likely important for elephant resource use
and social interactions, like the use of bais (forest clearings) for forest elephant social
]. In the Wonga WongueÂ Presidential Reserve, whether environmental factors such
as food, water, and habitat drive elephant movements more than anthropogenic disturbance
was dependent on the seasonÐelephants come into closer contact with villages in the dry
season compared to the wet season. Elephants are free-roaming in the area and the reserve is well
10 / 17
protected with low rates of poaching compared to other sites (e.g. ). Whereas previous
studies of forest elephant movement focused on the importance of unprotected habitat or roads,
we demonstrate that environmental variables can also strongly influence elephant movements
where they are protected. Even so, forest elephants spent disproportionate time near villages in
their home ranges, particularly in the dry season, underscoring that conservation efforts
should focus on areas of potential human-elephant conflict along the populated peripheries of
Despite our first hypothesis that male forest elephants would travel farther than females,
both sexes travelled on average 7±8 km daily and 2,840 km annually. Peak elephant
movements occurred around dawn and dusk, likely representing movements from grassland to
forest in the morning and from forest to grassland in the evening. While this finding concurs
with activity patterns of savanna elephants [
], it contrasts with the diurnal activity patterns
of a female elephant in Congo that was most active between 12:00±21:00, with a peak at 15:00
]. Forest cover in closed canopy forest may protect elephants from heat [
elephants in a forest-grassland mosaic likely adapt their activity patterns to use the open grassland
when temperatures are cooler. Whereas male and female elephants did not differ in total
distances moved, they differed significantly in home range sizes. The average 100% MCP home
range area for WW forest elephants was 713 km2, with males averaging 965 km2 and females
averaging 354 km2Ðgreater and less, respectively, than the average home range of 546.8 km2
across six sites in Central Africa [
]. These differences are consistent with male elephants
using larger home ranges than females because they engage in exploratory movements outside
core areas of habitat useÐeven outside park boundariesÐto find forage and mates. Males are
often solitary and have more fluid social interactions unconstrained by offspring [
females are more likely to be in small family groups with dependent offspring. Females
constrained by offspring may behave more cautiously, remaining in known areas with reliable
resources where it may also be easier to protect their young from predation.
Other studies of forest elephants have found that home ranges vary widely across sites. A
female elephant in Ndoki National Park had an MCP home range of 2,226 km2 [
three female elephants in an oil concession in southwest Gabon covered an average home
range of 212 km2, perhaps constrained by human presence [
]. Home range sizes vary
strongly across protected areas in Gabon, from an average MCP of 75.6 km2 in Loango
National Park to 568.2 km2 in MinkeÂbeÂ National Park and 623.2 km2 in Ivindo National Park
]. Six females in Loango National Park had an average 95% kernel home range of only 51.7
]. From this small number of studies, it appears that apart from Loango, where home
ranges are geographically constrained by the coastline and a large lagoon in the north of the
park, forest elephants across Gabon have average home range areas of 500 to 600 km2.
Consistent with elephants in Loango [
], females in WW maintained more separated home ranges
than males. The variation in home range areas across sites underscores the need for detailed
studies of elephant habitat to determine how fine scale differences in land use history, geology,
soils, nutrient/mineral availability, and rainfall affect habitat quality, elephant resource use,
Consistent with our second prediction that forest elephants would spend more time in the
forest than grassland habitat during the daytime and dry season, we found strong evidence for
seasonal shifts in activity patterns and habitat use. Overall, forest elephants spend most of their
time in forests, with approximately one-third of their time in grassland. During the dry season,
elephants occur in the forests adjacent to the central grassland, but rarely venture into the
grassland. Use of grassland habitat only surpassed forest in the short-wet season, when
elephants congregate in the central grassland to browse on young grass following prescribed
burns. Studies on savanna elephants have similarly shown that elephants select habitats with
11 / 17
greater tree cover in the dry season, but not the wet season [
]. Reduced distances moved
during the dry season could reflect the aggregation of water and fruit resources, such that: 1.
Elephants stay near perennial water sources during times of relative water scarcity; and/or 2.
Elephants stay near patches of Sacoglottis gabonensis trees, which supply a keystone fruit
[41,42] and are abundant in WW during the long-dry season. By comparison, the greater
hourly and daily distances moved in the wet season are likely a result of the onset of new grass
growth in grasslands and fruiting of most tree species in the forest [
]. Elephants likely move
greater distances as they travel between forest and grassland and as they make long direct
movements to less common preferred fruiting species [
Elephants tend to visit grasslands at night, moving into forest during the daytime for
protection from the sun and high temperatures in the grassland. Nighttime use of grasslands in
WW is consistent with research on forest clearings: 79% of visits to the Dzanga-Sangha Bai in
Central African Republic occurred at night [
]. Daytime use of forest habitat could also be an
evolutionary adaptation to avoid predation or hunting. Relatively low rates of poaching in
WW compared to other sites (e.g. [
]) over the last few years, however, suggests that
environmental factors may play a stronger role than anthropogenic factors in driving elephant
movements between forest and grassland.
Contrary to our third hypothesis that elephants would avoid villages, we found that
elephants tended to spend time near villages in their home ranges, particularly in the dry season.
Elephants may be attracted to secondary habitat and transitional areas near villages cleared for
agriculture, logging, and road building [
]. The attraction of villages was particularly strong
in the dry season, when fewer tree species produce fruits and new leaves. Although we did not
detect a significant difference between sexes in average distance from villages, three male
elephants noticeably ventured within 5 km of villages at the edges of the reserve (S7 Fig). Villagers
from GongoueÂ, on the west coast of WW, observed a collared male elephant raiding their
crops. As mentioned above, female elephants may be intrinsically more cautious because of
offspring, whereas young, subordinate males may be more likely to raid crops or abandoned
fields and lose fear of people, causing damage and conflict. Monitoring the distance of
elephants from villages is important to assess the severity of crop raiding in the periphery of WW,
to pinpoint target areas for conflict mitigation, and perhaps to identify problem elephants that
need to be removed or relocated [
]. Repeated crop raiding could prompt villagers to kill
elephants in retaliation. In February 2017, poachers killed a collared male elephant 5 km from
a village southwest of the reserve. It is unknown whether the incident was motivated by crop
raiding or the ivory trade, but bullets recovered from the carcass and the removal of tusks
suggest the latter.
Similar to their attraction to villages, elephant presence increased near roads during the wet
season and was unaffected by roads during the dry season. Although forest elephants avoid
unprotected roads in areas where poaching and human disturbance is high [
or abandoned roads with little traffic may act as convenient movement routes for elephants. In
a protected oil concession in Gabon, elephants occurred just 516 m from roads on average
. In WW, elephants tended to be closer to roads during the wet season, in contrast to our
hypothesis that elephants would avoid roads. Male and female elephants crossed roads
frequently, with males averaging 503 crossings and females averaging 342 crossings over the
study period, although they crossed roads at significantly higher speeds than non-crossing
movements (e.g., [
Consistent with our final hypothesis, during the wet season, presence of elephants increased
at greater distances from perennial water sources. This result is consistent with elephants
moving farther during the wet season (>1 km per day) than in the dry season. The greater
12 / 17
abundance of small, temporary water sources and decreased water limitation likely allows
elephants to move farther away from permanent water supplies.
Our study provides insights into the habitat use and behavior of forest elephants in a
forestgrassland mosaic, but there is still a great deal to be learned about the movement ecology of
forest elephants to effectively conserve their populations. Even with 17 GPS-collared elephantsÐ
the second-largest study of forest elephant movementÐour sample size is low and additional
tracking of elephants is necessary to reach robust conclusions regarding their movement, home
range, and habitat use through space and time.
Forest-grassland mosaics pose some unique challenges for management of elephant
populations, but protecting elephants from poaching is still the key to their conservation. Unlike the
arid savannas of southern and eastern Africa where water management is a primary concern,
in humid tropical forest the chief issues are maintaining grassland openness, promoting grass
turnover, and preserving fruiting tree phenology and diversity. Prescribed burns are an
important tool for maintaining grasslands that serve as sources of food resources and social
interactions. The interior of WW is currently well protected from poachers, but conflict with humans
could threaten elephants along its boundaries. Conservation monitoring and law enforcement
should be focused in the periphery zone, especially near villages where elephants might raid
crops. Future GPS-tracking efforts should focus on elephants, particularly males, in the
periphery of protected areas to evaluate their interactions with villages and to warn park
management, e.g. via SMS text alerts, when elephants approach plantations [
]. Raising awareness of
local communities about the GPS monitoring program and elephant ecology, and investing in
methods to prevent crop raiding could further mitigate threats.
Comparing elephant habitat use inside, outside, and near the borders of protected areas will
help to assess the effects of anthropogenic disturbance on elephant movements. As economic
development progressesÐparticularly with the growth of extractive industriesÐnatural
landscapes are fragmented, confining large, mobile species to islands of suitable habitat.
Maintaining connectivity between protected areas will enhance the capacity of Gabon's protected area
network to conserve forest elephant populations and ensure they remain a long-term
component of Central African forests.
S1 Table. Seventeen forest elephants collared in Wonga WongueÂ Presidential Reserve.
S2 Table. Landsat information.
S3 Table. Ground truth points and regions of interest.
S4 Table. Environmental variables of interest included in habitat modeling.
S5 Table. Confusion matrices in pixels and percentages.
S6 Table. Full model selection outputs for hourly and daily movements.
13 / 17
S7 Table. Total track distance summaries for 17 GPS-collared forest elephants in WW.
S8 Table. Area of MCP home ranges.
S9 Table. 100% MCP home range areas for WW by elephant and season.
S10 Table. Area of KUD home ranges.
S11 Table. Full model selection output for the factors influencing elephant movement in the dry season.
S12 Table. Full model selection output for the factors influencing elephant movement in the wet season.
S1 Fig. GPS locations and ground truth points.
S2 Fig. Thematic land cover map for the study area.
S3 Fig. Boxplots showing distances moved between consecutive GPS points for all hours of the day.
S4 Fig. Boxplots showing distributions of distance between consecutive GPS points by month.
S5 Fig. 95% MCP and 95% KUD home ranges for forest elephants in WW.
S6 Fig. Percentage of elephant locations in each land cover type by time of day.
S7 Fig. Distance to nearest villages by month (above) and by elephant (below).
S1 Text. Land cover classification: Detailed protocol.
We are grateful to the Gabonese government, including the Agence Nationale des Parcs
Nationaux (ANPN) and the Centre National de la Recherche Scientifique et Technologique
(CENAREST), for permission to conduct this research and administrative and logistical
support. We thank the managers and staff of the Wonga WongueÂ Presidential Reserve, and
A.D. Banguiya, M. Malouata, and A. Awoubazo for their skilled work in the field. We
dedicate this paper to the memory of David Fine who lost his life in service to conservation in
14 / 17
Data curation: Emily C. Mills.
Conceptualization: Emily C. Mills, John R. Poulsen, J. Michael Fay, Connie J. Clark.
Formal analysis: Emily C. Mills, John R. Poulsen, Christopher Beirne.
Funding acquisition: J. Michael Fay, Connie J. Clark.
Investigation: Emily C. Mills, Peter Morkel, Amelia Meier.
Methodology: Emily C. Mills, John R. Poulsen, Christopher Beirne.
Project administration: Emily C. Mills, John R. Poulsen, J. Michael Fay.
Supervision: John R. Poulsen, J. Michael Fay, Connie J. Clark, Lee J. T. White.
Writing ± original draft: Emily C. Mills, John R. Poulsen.
Writing ± review & editing: Emily C. Mills, John R. Poulsen, J. Michael Fay, Peter Morkel,
Connie J. Clark, Amelia Meier, Christopher Beirne, Lee J. T. White.
15 / 17
16 / 17
1. Blake S , Strindberg S , Boudjan P , Makombo C , Bila-Isia I , Ilambu O , et al. Forest elephant crisis in the Congo Basin . PLoS Biol . 2007 ; 5 ( 4 ):e111. https://doi.org/10.1371/journal.pbio. 0050111 PMID: 17407383
2. Maisels F , Strindberg S , Blake S , Wittemyer G , Hart J , Williamson EA , et al. Devastating decline of forest elephants in Central Africa . PLoS One . 2013 ; 8 ( 3 ):e59469. https://doi.org/10.1371/journal.pone. 0059469 PMID: 23469289
3. IUCN African and Asian Specialist Groups . Elephant database . In: IUCN African and Asian Specialist Groups. Elephant database [Internet] . 2013 . Available from: http://www.elephantdatabase.org/
4. Poulsen JR , Koerner SE , Moore S , Medjibe VP , Blake S , Clark CJ , et al. Poaching empties critical Central African wilderness of forest elephants . Curr Biol . 2017 ; 27 ( 4):R134±R135 . https://doi.org/10.1016/j. cub. 2017 . 01 .023 PMID: 28222286
5. Hebblewhite M , Haydon DT . Distinguishing technology from biology: a critical review of the use of GPS telemetry data in ecology . Philos Trans R Soc B Biol Sci . 2010 ; 365 ( 1550 ): 2303 ± 2312 . https://doi.org/ 10.1098/rstb. 2010 .0087 PMID: 20566506
6. Sawyer H , Nielson RM , Lindzey F , McDonald LL . Winter habitat selection of mule deer before and during development of a natural gas field . J Wildl Manage . 2006 ; 70 ( 2 ): 396 ± 403 . https://doi.org/10.2193/ 0022 - 541x ( 2006 ) 70 [ 396 :whsomd]2.0.co;2
7. Scientific review for the identification of critical habitat for Woodland caribou (Rangifer tarandus caribou), Boreal Population in Canada . Ottowa; 2009 .
8. Leimgruber P , McShea WJ , Brookes CJ , Bolor-Erdene L , Wemmer C , Larson C. Spatial patterns in relative primary productivity and gazelle migration in the Eastern Steppes of Mongolia . Biol Conserv. 2001 ; 102 ( 2 ): 205 ± 212 . https://doi.org/10.1016/s0006- 3207 ( 01 ) 00041 - 6
9. Dyer SJ , O'Neill JP , Wasel SM , Boutin S. Avoidance of industrial development by Woodland Caribou. J Wildl Manage . JSTOR; 2001 ; 65 ( 3 ): 531 ± 542 . https://doi.org/10.2307/3803106
10. Hebblewhite M , Merrill E. Modelling wildlife-human relationships for social species with mixed-effects resource selection models . J Appl Ecol . 2008 ; 45 ( 3 ): 834 ± 844 . https://doi.org/10.1111/j.1365- 2664 . 2008 . 01466 .x
11. Graham MD , Douglas-Hamilton I , Adams WM , Lee PC . The movement of African elephants in a human-dominated land-use mosaic . Anim Conserv . 2009 ; 12 ( 5 ): 445 ± 455 . https://doi.org/10.1111/j. 1469- 1795 . 2009 . 00272 .x
12. Chester C . Landscape vision and the Yellowstone to Yukon Conservation Initiative . Conservation across borders: biodiversity in an interdependent world . Washington DC, USA: Island Press; 2006 . pp. 134 ± 157 .
13. Loarie SR , van Aarde RJ , Pimm SL . Fences and artificial water affect African savannah elephant movement patterns . Biol Conserv . 2009 ; 142 ( 12 ): 3086 ± 3098 . https://doi.org/10.1016/j.biocon. 2009 . 08 .008
14. ChamailleÂ-Jammes S , Mtare G , Makuwe E , Fritz H . African elephants adjust speed in response to surface-water constraint on foraging during the dry-season . PLoS One . 2013 ; 8 ( 3 ):e59164. https://doi.org/ 10.1371/journal.pone. 0059164 PMID: 23554989
15. Cook RM , Henley MD , Parrini F . Elephant movement patterns in relation to human inhabitants in and around the Great Limpopo Transfrontier Park . Koedoe. 2015 ; 57 ( 1 ). https://doi.org/10.4102/koedoe. v57i1. 1298
16. Douglas-Hamilton I , Krink T , Vollrath F . Movements and corridors of African elephants in relation to protected areas . Naturwissenschaften . 2005 ; 92 ( 4 ): 158 ± 163 . https://doi.org/10.1007/s00114-004 -0606-9 PMID: 15770465
17. Ngene SM , Van Gils H , Van Wieren SE , Rasmussen H , Skidmore AK , Prins HHT , et al. The ranging patterns of elephants in Marsabit protected area, Kenya: the use of satellite-linked GPS collars . Afr J Ecol . 2010 ; 48 ( 2 ): 386 ± 400 . https://doi.org/10.1111/j.1365- 2028 . 2009 . 01125 .x
18. Roever CL , van Aarde RJ , Leggett K. Functional connectivity within conservation networks: delineating corridors for African elephants . Biol Conserv . 2013 ; 157 : 128 ± 135 . https://doi.org/10.1016/j.biocon. 2012 . 06 .025
19. Blake S , Douglas-Hamilton I , Karesh WB . GPS telemetry of forest elephants in Central Africa: results of a preliminary study . Afr J Ecol . 2001 ; 39 ( 2 ): 178 ± 186 . https://doi.org/10.1046/j.1365- 2028 . 2001 . 00296 .x
20. Kolowski JM , Blake S , Kock MD , Lee ME , Henderson A , Honorez A , et al. Movements of four forest elephants in an oil concession in Gabon, Central Africa . Afr J Ecol . 2010 ; 48 ( 4 ): 1134 ± 1138 . https://doi.org/ 10.1111/j.1365- 2028 . 2009 . 01204 .x
21. Schuttler SG , Blake S , Eggert LS . Movement patterns and spatial relationships among African forest elephants . Biotropica . 2012 ; 44 ( 4 ): 445 ± 448 . https://doi.org/10.1111/j.1744- 7429 . 2012 . 00889 .x
22. Yackulic CB , Blake S , Deem S , Kock M , Uriarte M. One size does not fit all: flexible models are required to understand animal movement across scales . J Anim Ecol . 2011 ; 80 : 1088 ± 1096 . https://doi.org/10. 1111/j.1365- 2656 . 2011 . 01851 .x PMID: 21521216
23. Blake S , Deem SL , Strindberg S , Maisels F , Momont L , Isia I-B , et al. Roadless wilderness area determines forest elephant movements in the Congo Basin . PLoS One . 2008 ; 3 ( 10 ):e3546. https://doi.org/ 10.1371/journal.pone. 0003546 PMID: 18958284
24. Landsat 8 (L8) Data Users Handbook Version 2.0 . Sioux Falls , South Dakota: Department of the Interior, U.S. Geological Survey; 2016 .
25. ENVI 5 .3. Harris Geospatial . Boulder, Colorado: Exelis Visual Information Solutions, Harris Geospatial Solutions; 2015 .
26. Baig MHA , Zhang L , Shuai T , Tong Q. Derivation of a tasselled cap transformation based on Landsat 8 at-satellite reflectance . Remote Sens Lett . 2014 ; 5 ( 5 ): 423 ± 431 . https://doi.org/10.1080/2150704x. 2014 .915434
27. Burt WH . Terrioriality and home range concepts as applied to mammals . J Mammal . 1943 ; 24 ( 13 ): 346 ± 352 . https://doi.org/10.2307/1374834
28. Wall J , Wittemyer G , Klinkenberg B , LeMay V , Douglas-Hamilton I . Characterizing properties and drivers of long distance movements by elephants (Loxodonta africana) in the Gourma , Mali. Biol Conserv . 2013 ; 157 : 60 ± 68 . https://doi.org/10.1016/j.biocon. 2012 . 07 .019
29. Calenge C. The package ªadehabitatº for the R software: a tool for the analysis of space and habitat use by animals . Ecol Modell . 2006 ; 197 ( 3 ±4): 516 ± 519 . https://doi.org/10.1016/j.ecolmodel. 2006 . 03 .017
30. R Core Team. R: a language and environment for statistical computing . R Found Stat Comput . 2013 .
31. ArcMap 10.4.1 ESRI. ArcGIS Desktop: Release 10 . Redlands, CA: Environmental Systems Research Institute; 2011 .
32. Boyce MS , Vernier PR , Nielsen SE , Schmiegelow FKA . Evaluating resource selection functions . Ecol Modell . 2002 ; 157 : 281 ± 300 . https://doi.org/10.1016/s0304- 3800 ( 02 ) 00200 - 4
33. Thomas DL , Taylor EJ . Study designs and tests for comparing resource use and availability . J Wildl Manage . 1990 ; 54 ( 2 ): 322 ± 330 . https://doi.org/10.2307/3809050
34. Burnham K , Anderson DR . Model selection and multi-model inference: a practical information-theoretic approach [Internet]. Springer; 2002 . Available: http://cds.cern.ch/record/1664022
35. Richards S , Whittingham M , Stephens P . Model selection and model averaging in behavioural ecology: the utility of the IT-AIC framework . Behav Ecol Sociobiol . 2011 ; 65 ( 1 ): 77 ± 89 . https://doi.org/10.1007/ s00265-010-1035-8
36. Bates D , Maechler M , Bolker B , Walker S. Fitting linear mixed effects models using lme4 . J Stat Softw . 2015 ; 67 ( 1 ):1± 48 .
37. Wrege PH , Rowland ED , Bout N , Doukaga M. Opening a larger window onto forest elephant ecology . Afr J Ecol . 2011 ; 50 ( 2 ): 176 ± 183 . https://doi.org/10.1111/j.1365- 2028 . 2011 . 01310 .x
38. Turkalo AK , Wrege PH , Wittemyer G . Long-term monitoring of Dzanga Bai forest elephants: forest clearing use patterns . PLoS One . 2013 ; 8 ( 12 ):e85154. https://doi.org/10.1371/journal.pone. 0085154 PMID: 24386460
39. Schuttler SG , Philbrick JA , Jeffery KJ , Eggert LS . Fine-scale genetic structure and cryptic associations reveal evidence of kin-based sociality in the African forest elephant . PLoS One . 2014 ; 9 ( 2 ):e88074. https://doi.org/10.1371/journal.pone. 0088074 PMID: 24505381
40. Roever CL , van Aarde RJ , Leggett K. Functional responses in the habitat selection of a generalist mega-herbivore, the African savannah elephant . Ecography . 2012 ; 35 ( 11 ): 972 ± 982 . https://doi.org/10. 1111/j.1600- 0587 . 2012 . 07359 .x
41. White LJT . Sacoglottis gabonensis fruiting and the seasonal movements of elephants in the LopeÂ Reserve, Gabon . J Trop Ecol . 1994 ; 10 ( 1 ): 121 ± 125 . https://doi.org/10.1017/s0266467400007768 42. Morgan BJ. Sacoglottis gabonensis- a keystone fruit for forest elephants in the ReÂserve de Faune du Petit Loango, Gabon . Afr J Ecol. 2009 ; 47 ( 2 ): 154 ± 163 . https://doi.org/10.1111/j.1365- 2028 . 2007 . 00897 .x
43. White LJT . Patterns of fruit-fall phenology in the LopeÂ Reserve, Gabon . J Trop Ecol . 1994 ; 10 ( 3 ): 289 ± 312 . https://doi.org/10.1017/s0266467400007975
44. Terborgh J , Davenport LC , Niangadouma R , Dimoto E , Mouandza JC , Scholtz O , et al. Megafaunal influences on tree recruitment in African equatorial forests . Ecography . 2015 ; 39 ( 2 ): 180 ± 186 . https:// doi.org/10.1111/ecog.01641
45. Poulsen JR , Clark CJ , Bolker BM . Decoupling the effects of logging and hunting on an Afrotropical animal community . Ecol Appl . 2011 ; 21 ( 5 ): 1819 ± 1836 . https://doi.org/10. 1890 /10- 1083 .1 PMID: 21830721
46. Guerbois C , Chapanda E , Fritz H. Combining multi-scale socio-ecological approaches to understand the susceptibility of subsistence farmers to elephant crop raiding on the edge of a protected area . J Appl Ecol . 2012 ; 49 ( 5 ): 1149 ± 1158 . https://doi.org/10.1111/j.1365- 2664 . 2012 . 02192 .x
47. Graham MD , Douglas-Hamilton I , Adams WM , Lee PC . The movement of African elephants in a human-dominated land-use mosaic . Animal Conservation . 2009 ; 12 ( 5 ): 445 ± 455 . https://doi.org/10. 1111/j.1469- 1795 . 2009 . 00272 .x
48. Yackulic CB , Strindberg S , Maisels F , Blake S. The spatial structure of hunter access determines the local abundance of forest elephants (Loxodonta africana cyclotis) . Ecol Appl . 2011 ; 21 ( 4 ): 1296 ± 1307 . https://doi.org/10.2307/23022997 PMID: 21774431
49. Wall J , Wittemyer G , Klinkenberg B , Douglas-Hamilton I . Novel opportunities for wildlife conservation and research with real-time monitoring . Ecol Appl . 2014 ; 24 ( 4 ): 593 ± 601 . https://doi.org/10. 1890 /13- 1971 .1 PMID: 24988762