www.publish.csiro.au/journals/wr Wildlife Research, 2008, 35, 409–416
Predicting habitat suitability for the endemic mountain nyala
(Tragelaphus buxtoni ) in Ethiopia
Paul H. EvangelistaA,B,D, John Norman IIIA, Lakew BerhanuC, Sunil Kumar A and
ANatural Resource Ecology Laboratory, Colorado State University, Fort Collins, CO 80523-1499, USA.
BThe Murulle Foundation, PO Box 1442, Fort Collins, CO 80522, USA.
CEthiopian Wildlife Conservation Department, PO Box 386, Addis Ababa, Ethiopia.
DCorresponding author. Email: email@example.com
Abstract. The use of statistical models to predict species distributions and suitable habitat has become an essential tool
for wildlife management and conservation planning. Models have been especially useful with rare and endangered wildlife
species. One such species is the mountain nyala (Tragelaphus buxtoni), a spiral-horned antelope endemic to the Ethiopian
highlands. The full range of the species has never been adequately defined and recent discoveries of new populations
suggest that others may exist undetected. To identify potential mountain nyala occurrences, we used classification tree
analysis to predict suitable habitat using 76 climatic, topographical and vegetative variables. Three model evaluation
methods showed a strong performance of the final model with an overall accuracy of 90%, Cohen’s maximised κ of 0.80
and area under the receiver operating characteristic curve (AUC) value of 0.89. Minimum temperature and maximum pre-
cipitation generally had the greatest predictive contributions to suitable mountain nyala habitat. The predicted habitat
covered an area of 39378 km2, with the majority occurring in remote forests on the southern escarpment of the Bale
Mountains. Other areas within the predicted range may be too impacted by human and livestock populations to support
mountain nyala; however, the model will be useful in directing future surveys for new populations while offering clues to
the species historical range.
Additional keywords: classification tree, ecological niche model, habitat suitability, WorldClim.
Tragelaphus buxtoni (mountain nyala) is a large spiral-horned open spaces, such as cultivated fields, grasslands, wet
antelope endemic to Ethiopia’s highlands. First identified by meadows, fens and the Afro-alpine habitats to feed and drink
the scientific community in 1908 (Lydekker 1910, 1911), very (Evangelista 2006b).
little is known about the biology, range and population of the Habitat loss and land degradation are the most significant
species. Mountain nyala are believed to inhabit only the south- threats to mountain nyala and the majority of Ethiopia’s
ern highlands east of the Rift Valley. Remnant populations exist wildlife. Ethiopia’s forests once covered 65% of the country
in the Chercher and Arussi mountains; however, most inhabit and 90% of the highlands. Today, Ethiopia’s forests cover only
the Bale Mountains (Fig. 1; Brown 1969a; Yalden and Largen 2.2% of the country and 5.6% of the highlands (FAO 2006). As
1992; Evangelista 2006b; Sillero-Zubiri 2007). Mountain nyala forest habitats are reduced by the growing human population
are generally found between 1800 and 4000 m, but have been and need for natural resources (e.g. agriculture, livestock
observed at elevations as low as 1500 m (K. Wakijira, pers. grazing, fuel-wood), mountain nyala populations are becoming
comm.). Generally, the species prefers dense highland forests fragmented and increasingly confined to small areas
for concealment, thermal regulation and the availability of sea- (Evangelista 2006b). Although it is evident that habitat loss is
sonal forage. In particular, mountain nyala prefer the upper and reducing the known range of mountain nyala, the full distri-
lower Afro-montane zones, which are dominated by Hagenia bution of the species has never been adequately determined. As
abyssincia, Hypericum revolutum, Juniperus procera and a result, total population estimates remain inconclusive, hinder-
Sinarundinaria alpina (Bekele-Tesemma et al. 1993; Miehe ing effective management strategies and skewing conservation
and Miehe 1994; Bussman 1997). Historic records suggest that policy (East 1999; Refera and Bekele 2004; Evangelista 2006b;
the ericaceous belt (sub-alpine zone dominated by Erica Evangelista et al. 2007; Sillero-Zubiri 2007). For the Ethiopian
trimera and E. arborea) was at one time prime mountain nyala Wildlife Conservation Department (EWCD), identifying the
habitat; however, much of this habitat type has been converted full range and total population of mountain nyala is a high
for agriculture and livestock grazing (Brown 1969a, 1969b; priority and a requirement for implementing effective manage-
Malcolm and Evangelista 2005; Evangelista et al. 2007). At ment plans, facilitating conservation strategies and formulating
night, mountain nyala have been observed congregating in wildlife policy (T. Hailu, pers. comm.).
© CSIRO 2008 10.1071/WR07173 1035-3712/08/050409
410 Wildlife Research
In recent years, several populations of mountain nyala have
been documented for the first time (Evangelista 2006a, 2006b).
In 2000 and 2001, the species was found to be widespread
throughout the eastern slopes of the Bale Mountains
(Evangelista 2006b; Evangelista et al. 2007). In response to
apparent land degradation and poaching activities, the EWCD
established three controlled hunting areas (CHA) in the Bale
Mountains to intensively manage the species through limited
trophy hunting. In 2005, the EWCD discovered two additional
populations: the first was located in the south-western region of
the Bale Mountains and the second on the north-eastern slope of
the Arussi Mountains (Fig. 1). Populations in these two areas
have been surveyed by the EWCD and are currently under
review for management through controlled hunting.
Collectively, the five new CHAs are estimated to have between
1750 and 2000 mountain nyala based on line-transect surveys
conducted by the EWCD (Evangelista 2006b). In 2006, moun-
tain nyala were found to be prevalent in the higher elevations of
the Harenna Forest, a cloud forest that covers the southern
escarpment of the Bale Mountains (P. Evangelista, pers. obs.
2006–2007; Evangelista et al. 2007). Population surveys have
not yet been conducted for the area, but the presence of moun-
tain nyala has been confirmed in four regions inside and outside
of the Bale Mountains National Park by EWCD and the authors.
These recent discoveries and other undocumented reports
suggest that additional populations of mountain nyala are likely
to exist and add to knowledge of the species’ habitat require-
ments. However, the Ethiopian highlands are vast and often
P. H. Evangelista et al.
inaccessible, creating many logistical challenges in the search
for new populations.
Fortunately, advancements in computer technology and soft-
ware are allowing ecologists to integrate field observations,
remotely sensed data, geographical information systems (GIS)
and digital image processing to define species distributions and
identify undiscovered populations. Specifically, probability
models are employed by ecologists to predict species occur-
rence (Corsi et al.1999; Pearson et al. 2007), migratory patterns
(Boone et al. 2006), critical habitats (Hatten and Paradzick
2003; Turner et al. 2004), risk of diseases (Broadfoot et al.
2001; Pfeiffer and Hugh-Jones 2002), invasion of non-native
species (Ficetola et al. 2007; Evangelista et al. in press) and
management priorities (Clevenger et al. 2002; Felix et al. 2004).
One of the most widely used model applications is classification
and regression tree (CART) analysis. CART analysis is a com-
monly used non-parametric modelling technique that predicts
the response of a dependent variable through a series of simple
regression analyses (Breiman et al. 1984; Hansen et al. 1996;
Lewis 2000). Unlike other regression approaches that conduct
simultaneous analyses, CART models statistically partition the
dependent data into two homogenous groups at a node, repeat-
ing the procedure for each group in a continuous process that
forms a hierarchal tree. Classification trees are used when the
dependent variables are categorical (i.e. presence, absence) and
regression trees are used when the dependent variables are con-
tinuous (e.g. percent basal cover, species richness). Several
characteristics of this modelling approach appeal to researchers
Fig. 1. Map of Ethiopia’s southern highlands and
known populations of mountain nyala.
Mountain nyala habitat Wildlife Research 411
and resource managers. First, the analyses explicitly allow for
non-linear relationships between the dependent and inde-
pendent variables. These methods make no a priori assumptions
about the distribution of the data, the relationships among inde-
pendent variables or relationships between the dependent and
independent variables. Second, they are well suited to handle
non-homogenous datasets (i.e. unbalanced sample sizes, high
variability). Finally, the results are easily interpreted and the
predictive strength of each independent variable is explicitly
reported in the results (Michaelsen et al. 1994; Andersen et al.
2000; Evangelista et al. 2004).
The aim of our study is to use classification tree analyses at
landscape and regional scales to identify areas suitable for
mountain nyala and to direct future survey efforts that may lead
to the discovery of undocumented populations. We rely on
known distribution information with a suite of environmental
variables to relate species occurrence and absence to geograph-
ical and topographical features. In addition, to model evalua-
tions that are built into classification tree analyses (i.e.
predictive strength of independent variables, number of parti-
tions and terminal nodes), we evaluate model performances
with three proven statistical tests.
Although mountain nyala are believed to inhabit only a small
portion of Ethiopia’s southern highlands, the extent of our initial
analyses included the entire county, which encompasses
1.12 million km2 (MoWR 2001). Ethiopia is located in East
Africa and often described as the ‘roof of Africa’ because of its
rugged mountain topography, which has the largest expanse of
Afro-alpine habitats on the continent (Gamachu 1988; Uhlig
and Uhlig 1991). Over 50% of Africa’s land mass above 2000 m
and over 80% of the land mass above 3000 m is found in
Ethiopia (McClanahan and Young 1996). The Ethiopian high-
lands are divided by the Rift Valley into two dominant massifs:
the Bale Mountains in the south-eastern block and the Simien
Mountains in the north-western block. Elevation in Ethiopia
ranges from 120 m below sea level to 4620 m above sea level.
The contrasting topography and Ethiopia’s situation within the
Intertropical Convergence Zone result in varying climates
across the country. Temperatures in the arid regions of the
country may be as high as 37°C, and temperatures at higher ele-
vations are known to dip as low as –15°C (MoWR 2001).
Annual precipitation also fluctuates across Ethiopia’s land-
scape, occurring in two distinct seasons: the Bega, a dry season
from October to May, and the Kiremt, a long rainy season from
June to September. A small rainy season called the Belg occurs
in April and May at the end of the Bega. Mean annual precipita-
tion ranges from 2000 mm in areas of the south-west to less than
200 mm over the Afar lowlands (MoWR 2001). Based on
Koppen’s climate classification system, 10 climate zones define
the Ethiopian landscape (Gonfa 1996). These climate types
harbour many unique and diverse ecosystems that include
desert, savanna grasslands, shrubland, savanna woodlands
(Acacia sp.), tropical forests, montane forests, bamboo
(Yushane alpine and Oxytenanthera abyssinica), heathlands
(Erica sp.) and Afro-alpine (Logan 1946; Von Breitenbach
1961; Egziabher 1988; Miehe and Miehe 1994; Bussman 1997;
Carr 1998; Embaye 2000).
Field data sources
Our first model relied on mountain nyala observations collected
at regional levels from: (1) line-transect surveys conducted by
the EWCD between 2002 and 2006 in nine CHAs (Evangelista
2006b); (2) population estimates from observations at Kuni-
Muktar Wildlife Reserve and Galama Mountains Forest Priority
Area (Malcolm and Evangelista 2005; F. Kebede, pers. comm.);
(3) direct counts in the northern portion of Bale Mountains
National Park (Refera and Bekele 2004); and (4) field observa-
tions using global positioning system (GPS) collected from the
Senetti Plateau and the Harenna Forest in Bale Mountains
National Park (P. Evangelista, pers. obs. 2002–2006). Most
presence data did not include specific coordinates of observa-
tions. For these, we created boundary layers of management
areas, CHAs and study sites in ArcGIS 9.2 (ESRI 2006) using
GPS surveys, paper maps of CHAs provided by the EWCD
(Evangelista 2006b), Refera and Bekele’s (2004) map of their
study site and Landsat7 TM satellite images (GLCF 2006).
Within each boundary layer, we randomly generated n presence
points based on the most current population estimate available
for each defined area. All presence points were integrated into a
single GIS shapefile and converted to a raster format with a
pixel size of 1 km2. When more than one presence point fell
within a single raster pixel, that pixel would only represent a
single occurrence for our analyses. This approach was necessary
to avoid habitat bias of mountain nyala in areas with unnaturally
high numbers (e.g. Kuni-Muktar; Evangelista et al. 2007) or
areas where animals are easily or regularly observed (e.g.
Gaysay; Refera and Bekele 2004). The total number of presence
points summed from regional population estimates was 4766
and the total number of presence cells used in our first analyses
was 1443. Additionally, we generated 4766 pseudo-absence
points throughout Ethiopia with the criteria that each point had
to be below 1500 m in elevation and it could not occur within
the known mountain nyala populations (shapefiles previously
described). We repeated the conversion from point data to raster
cells leaving 4756 absence cells for our analyses.
For our second model, we generated 3000 random points
within the predicted range of mountain nyala occurrence from
the first model. Based on previous mountain nyala population
field surveys and observations (Evangelista et al. 2007), we
defined each random point as a presence or absence and omitted
any points where occurrence was uncertain. Following this
process, 471 points (181 presence and 290 absence) remained
for the second analyses with 75% (144 presence and 233
absence points) used to train the model and 25% (37 presence
and 57 absence points) withheld to validate the model.
Spatial data sources
To predict suitable habitat for mountain nyala, we used 76 inde-
pendent spatial variables in our analyses that were derived from
various remotely sensed data and GIS analyses. Data at coarse
resolutions were re-sampled to 90 m and projected in the
Universal Transverse Mercator system (WGS 84, Zone 37N).
Continental and global datasets were reduced to the extent of
Ethiopia. Using a 90 m Digital Elevation Model (DEM) from
412 Wildlife Research
the National Aeronautics and Space Administration’s (NASA)
Shuttle Radar Topography Mission (SRTM; Jarvis et al. 2006;
Farr 2007), we generated slope in degrees, aspect, surface
roughness index, soil wetness index and solar insolation raster
layers. Slope in degrees and aspect were generated in ArcGIS
9.2 Spatial Analyst (ESRI 2006). Soil wetness index was calcu-
lated using the formula [ln(A/tan β)], where ln(.) is the natural
logarithm, A is the area drained per unit contour or specific area
and tan β is the slope (Moore et al. 1991; Wolock 1993). The
solar insolation grid was generated using the Shortwave
program developed by Kumar et al. (1997).
Three global vegetation indices (GVI) and 12 normalised
difference vegetation index (NDVI) scenes from moderate reso-
lution imaging spectroradiometer (MODIS) instruments aboard
NASA’s Terra satellite were acquired from the Global Land
Cover Facility (GLCF. 2006). GVIs included woody tree cover,
herbaceous vegetation cover and bare ground cover at 500-m
resolution. Collectively, these indices total 100% of ground
cover and can be properly displayed in red, green and blue band
combinations (Hansen et al. 2003a, 2003b). NDVI scenes were
acquired for the 15th day of each month for 2003 at 250 m res-
olution (Kidwell 1990, 1991; Carrol et al. 2004). NDVI is cal-
culated by [(NIR – red)/(NIR + red)], where NIR is band 2 (near
infrared) and red is band 1 (Sellers 1985; Myneni et al. 1995).
Nineteen bioclimatic parameters were acquired from the
WorldClim (Hijmans et al. 2005; http://www.worldclim.org/).
These fine resolution (~1 km) spatial data were interpolated
from weather stations across the globe, with averages calcu-
lated from at least 10 years of data (Hijmans et al. 2005, 2006).
The 19 BioClim variables are derived from mean monthly
and quarterly climate estimates (e.g. minimum, maximum,
mean temperature and precipitation) to approximate energy
and water balances at a given location (Nix 1986;
Statistical analyses and model evaluation
All statistical analyses were conducted using S-Plus 3 statistical
software (Insightful 2000). We generated two classification tree
models: the first to model suitable mountain nyala habitat for all
of Ethiopia and the second to refine the predicted suitable
habitat of the first. Classification trees were pruned using
10-fold cross-validation (Breiman et al. 1984). We used three
methods to evaluate the performance of each model: (1) a con-
fusion matrix that shows specificity and sensitivity and overall
accuracy; (2) Cohen’s maximised κ; and (3) area under the
receiver operating characteristic curve (AUC) (Fielding and
There are two possible errors that may occur in prediction
models: false negatives (under prediction or under-fit models)
and false positives (over predictions or over-fit models; Fielding
and Bell 1997). Using the validation data, we presented the rel-
ative proportions of these errors in a confusion matrix.
Specificity (the proportion of true-positive and false-positive
absences) and sensitivity (the proportion of true-positives and
false positive presences) are reported for the final model with
overall accuracy reported as a percentage (Fielding and Bell
1997). Threshold dependent evaluation measure Cohen’s max-
imised κ (Cohen 1960), was calculated using a cut-off threshold
determined by plotting sensitivity against specificity (Fielding
P. H. Evangelista et al.
and Bell 1997). The κ statistic measures the proportion of cor-
rectly classified points (i.e. presence, absence) after accounting
for the probability of chance agreement. κ statistic values range
from –1 to +1, where +1 would be perfect agreement and any
values less than 0 would indicate a performance no better than
random (Cohen 1960; Allouche et al. 2006). Landis and Koch
(1977) ranked analysis performances as poor when κ values are
<0.40; good when the κ values range from 0.40 to 0.75; and
excellent when κ values are >0.75
The AUC is calculated by generating a receiver operating
characteristic (ROC) curve to plot the sensitivity to 1 – speci-
ficity for all possible thresholds (Pearce and Ferrier 2000).
Commonly used in predictive models (Phillips et al. 2006;
Russell et al. 2007; Evangelista et al. in press), the AUC is a
measure of probability that a random positive point falls within
the predicted range of occurrence and a random negative point
falls outside (Pearce and Ferrier 2000). Strength of predictabil-
ity ranges from weak to strong or 0 to 1.0 respectively. An AUC
value of 0.5 represents complete random predictions, whereas a
value of 1.0 shows perfect discriminatory ability (Pearce and
Our first model had nine terminal nodes with a residual mean
deviance of 0.01 (Fig. 2). Of the 76 potential independent vari-
ables, seven predictors were selected in the model results with
precipitation of the warmest quarter (BIO18) occurring twice in
the tree. The probability of mountain nyala occurrence was
greatest when precipitation of the warmest quarter (BIO8) was
>270 mm (79% present, 21% absent), precipitation of the
wettest month (BIO13) was <241 mm (88% present, 12%
absent), minimum temperature of the coolest month (BIO6) was
<10.1 C° (100% present, 0% absent) and temperature seasonal-
ity (BIO4) was <1279.6 (100% present, 0% absent). Other pre-
dictor variables of mountain nyala occurrence include
maximum temperature of the warmest month (BIO5), precipita-
tion seasonality (BIO15) and NDVI for the month of September
The final model, analysed within the boundaries of our first
model, had 10 terminal nodes with a residual mean deviance of
0.1008 (Fig. 3). Of the potential independent variables, nine
were significant in the model results with surface roughness
index occurring twice in the tree. The probability of mountain
nyala occurrence was greatest when minimum temperature of
the coolest month (BIO18) was <3.3 C° (92% present, 8%
absent) and precipitation of the warmest quarter (BIO18) was
>281 mm (98% presence, 2% absent). Other significant vari-
ables in the model include surface roughness, precipitation of
the wettest quarter (BIO16), minimum temperature for
December, mean temperature for March, precipitation of the
coldest quarter (BIO19) and herbaceous cover (Fig. 3). The pre-
dicted suitable habitat had an area of 39378 km2, with ~97% of
the predicted area in the Chercher, Arussi and Bale mountains
and nearly 2% of the predicted area in the Gurage Massif on the
west side of the Rift Valley (Fig. 4).
Evaluation methods showed that the final model’s perfor-
mance was strong for training and validation datasets. The train-
ing model had an AUC of 0.93 and κ value of 0.88. The
confusion matrix showed the training model to have an overall
Mountain nyala habitat Wildlife Research 413
Precip. of Warmest Quarter (BIO18)
< 270 mm, 0.01 > 270 mm, 0.79
Max. Temp. of Warmest Month (BIO5) Precip. of Wettest Month (BIO13)
< 12.9 C°, 0.88 > 12.9 C°, 0.01 < 241 mm, 0.88 > 241 mm, 0.01
Precip. Seasonality (BIO15) 0 Min. Temp. of Coolest Month (BIO6) 0
< 70.6, 0.90 > 70.6, < 0.01 < 10.1 C°, 1.00 > 10.1 C°, 0.28
Temp. Seasonality (BIO4) Precip. of Warmest Quarter (BIO18)
0 < 1279.6, 1.00 > 1279.6, 0.28 < 38 mm, < 0.01 > 38 mm, 0.06
1 0 0 < 3293.1, 0.99 > 3293.1, 0.06
accuracy of 94% with specificity at 99% and sensitivity at 87%.
The validation data had an AUC value of 0.89 and κ value of
0.80. Overall accuracy of the validation model was 90% with
specificity at 95% and sensitivity at 84%.
The results of our analyses support earlier assessments of the
habitat specialist characteristics of the mountain nyala and its
limited range within the Ethiopian highlands. At the landscape
scale, mountain nyala habitat is heavily reliant on precipitation
and temperature, which are known to have direct influences on
Min. Temp. of Coolest Month (BIO6)
< 3.3 C°, 0. 92 > 3.3 C°, 0.13
Precip. of Warmest Quarter (BIO18 ) Surface Roughness
< 281 mm, < 0.01 > 281 mm, 0.98 < 119.5; 0.02 > 119.5; 0.32
0 1 0 Precip. of Wettest Quarter
< 396 mm, 0.96 > 39 6 mm, 0.22
Fig. 2. Classification tree analysis of
suitable mountain nyala habitat for
Ethiopia. Bold numbers represent the
percentage of presence points that fell
within the given criteria of significant
vegetation structure and diversity (Bekele-Tesemma et al.
1993). These environmental variables also played significant
roles at a regional scale in addition to topographical and vegeta-
tion variables. Although the species’ range presented by the
final model is highly restricted, it is substantially larger than the
speculated range reported in the literature (Malcolm and
Evangelista 2005; Evangelista 2006b; Sillero-Zubiri 2007).
Suitable habitat was predicted in areas that have not been
explored during previous population research efforts. These
areas are largely found in the south-western slopes of the Bale
Mountains and coincide with Brown’s surveys in the 1960s,
Fig. 3. Classification tree analysis of suit-
1 Surface Roughness
able mountain nyala habitat within the pre-
< 332, 0.14
Mean Temp. (March)
< 16.5 C°, 0.30 > 16.5 C°, 0.22
Precip. of Coldest Quarter (BIO19) 0
< 215 mm, 0.40 > 215 mm, < 0.01
Herbaceous Cover 0
< 55.4 %, 1.00 > 55.4 %, 0.25
> 332, 0.65
dicted range of the first model (Fig. 2). Bold
numbers represent the percentage of pres-
ence points that fell within the given criteria
Min. Temp. (Dec.)
of significant independent variables.
< 6.6 C°, 1.00 > 6.6 C°, 0.35
414 Wildlife Research P. H. Evangelista et al.
Potential Mou ntain Nyala Habita t
0 25 50 75 100 km12.5
Rift Valley Lakes
Fig. 4. Map of suitable mountain nyala habitat based
on the second classification tree analysis.
occurrences reported by Yalden and Largen (1992) and several
recent discoveries we previously reported. In 2006, two of the
authors attempted to explore this region’s interior but found the
terrain difficult to navigate and traverse (Evangelista et al.
2007). The survey team was able to confirm the presence of
mountain nyala in much of the area, and local people that were
interviewed consistently reported that significant populations
were common throughout the large expanses of forest
(Evangelista 2006b; Evangelista et al. 2007).
The final model also predicts that several regions outside the
Bale Mountains may have suitable habitat for mountain nyala.
Of particular interest are the tributaries and headwaters of the
Wabe Shebele River located east of the Arussi Mountains
(Fig. 1). Although we have never adequately explored these
regions from the ground, several aerial surveys revealed pockets
of intact forests with steep topography that is prohibitive to
human settlement and livestock grazing. Additionally, we have
received reports from local people that mountain nyala are
found in these areas. The northern tributaries of the Wabe
Shebele River, which flow from the Arussi and Chercher moun-
tains, are currently managed for mountain nyala and include one
of the recently discovered populations in 2005 (Evangelista
2006b). A second area of interest is the Gurage Massif located
on the west side of the Rift Valley. Although mountain nyala
have never been documented beyond the southern highlands and
east side of the Rift Valley, it is possible that the species may
have historically occurred there. Aerial surveys of the Gurage
Massif reveal only remnant forested areas remain with large
expanses of heathlands densely populated with settlements and
livestock (P. Evangelista, pers. obs. 2007). Mountain nyala are
known to persist in areas heavily impacted by land-use activities
(e.g. Galama Mountains, Gaysay Valley); however, we find it
unlikely that the species can presently be found there. Future
field surveys and in-depth interviews with the local people are
warranted, and could reveal significant information on the
present and historical status of the mountain nyala throughout
the Gurage Massif.
Our models were fitted with environmental variables that
included climate, topography and vegetation indices, but lack
data on human and livestock populations, land-use and vegeta-
tion types. We believe that model performances could be
improved if such data were available and integrated into our
analyses. Similarly, we have compiled numerous reports from
local people regarding mountain nyala sightings, but we elected
to exclude this information from the analyses because it either
has not been adequately confirmed or the locations were some-
times vague. Although the results from the models suggest that
the distribution of mountain nyala and suitable habitat are sig-
nificantly greater than previously thought, it should be noted
that human population densities, agriculture, livestock grazing
and deforestation will greatly reduce the actual area of occu-
pancy (Evangelista et al. 2007; Sillero-Zubiri 2007).
Despite the caveats, the models have significant value in
identifying suitable habitat for mountain nyala and adding to the
Mountain nyala habitat Wildlife Research 415
scientific knowledge of this endemic species. Specifically, our
results provide the first map of probable range of the mountain
nyala based on field observations since Brown (1969a). Our
analyses also demonstrate the strong correlation between the
species’ range and specific climate and topographic conditions.
This information not only provides wildlife managers with
important geo-spatial data to support management and conser-
vation decisions, but can also guide future surveys to new areas
where the mountain nyala may likely persist. Finally, our results
offer clues to the of mountain nyala’s historic range, which may
have once included areas west of the Rift Valley. The current
known distribution of the mountain nyala and our model results
support the belief that the species’ range has always been iso-
lated to the Ethiopian highlands.
The authors would like to thank the Ethiopian Wildlife Conservation
Department and Oromia Regional Land and Natural Resource Department
for their support of this important work and for sharing their survey data. We
also would like to thank The Murulle Foundation for funding and logistical
support; the USA Geological Survey, Fort Collins Science Center and
Natural Resource Ecology Laboratory, Colorado State University, for tech-
nical support; ESRI® GIS and Mapping Software for software support; and
Ethiopian Rift Valley Safaris for use of their facilities and field personnel.
Additional gratitude is expressed to Banovich Wildscapes Foundation,
Conklin Foundation, Dallas Ecological Foundation, R. Baker, G. Bond,
A. Randell, P. Ripepi, J.C. and N.A. Roussos, A. Sackman, W. Stout, SCI
Pittsburg Chapter, Shikar Safari Club and P. Swartzinski.
Allouche, O., Tsoar, A., and Kadmon, R. (2006). Assessing the accuracy of
species distribution models: prevalence, Kappa and the true skill
statistic (TSS). Journal of Applied Ecology 43, 1223–1232.
Andersen, M. C., Watts, J. M., Freilich, J. E., Yool, S. Y., Wakefield, G. I.,
McCauley, J. F., and Fahnestock, P. B. (2000). Regression-tree modeling
of desert tortoise habitat in the central Mojave Desert. Ecological
Applications 10(3), 890–900. doi:10.1890/1051-0761(2000)010
Bekele-Tesemma, A., Birnie, A., and Tengnas, B. (1993). ‘Useful Trees and
Shrubs for Ethiopia. Regional Soil Conservation Unit.’ (Swedish
International Development Authority: Stockholm.)
Boone, R. B., Thirgood, S. J., and Hopcraft, J. G. C. (2006). Serengeti wilde-
beest migratory patterns modeled from rainfall and new vegetation
growth. Ecology 87(8), 1987–1994. doi:10.1890/0012-9658(2006)87
Breiman, L., Friedman, J. H., Olshen, R. A., and Stone, C. J. (1984).
‘Classification and Regression Trees.’ (Wadsworth and Brooks/Cole
Advanced Books and Software: Pacific Grove, CA.)
Broadfoot, J. D., Rosatte, R. C., and O’Leary, D. T. (2001). Raccoon and
skunk population models for urban disease control planning in Ontario,
Canada. Ecological Applications 11(1), 295–303. doi:10.1890/1051-
Brown, L. (1969a). Observations on the status, habitat, and behavior of the
mountain nyala (Tragelaphus buxoni) in Ethiopia. Mammalia 33,
Brown, L. (1969b). Ethiopia’s elusive mountain nyala. Animals 12,
Bussman, R. W. (1997). The forest vegetation of the Harenna escarpment
(Bale Province, Ethiopia)-syntaxonomy and phytogeographical affini-
ties. Phytocoenologia 27(1), 1–23.
Carr, C. L. (1998). Patterns of vegetation along the Omo River in south-
western Ethiopia. Plant Ecology 135, 135–163. doi:10.1023/
Carrol, M. L., DiMiceli, C. M., Sohlberg, R. A., and Townshend, J. R. G.
(2004). 250m MODIS Normalized Difference Vegetation Index,
250ndvi28920033435, Collection 4, University of Maryland, College
Park, MD, Day 289, 2003.
Clevenger, A. P., Wierzchowski, J., Chruszcz, B., and Gunson, K. (2002).
GIS-generated, expert-based models for identifying wildlife habitat
linkages and planning migration passages. Conservation Biology 16(2),
Cohen, J. (1960). A coeff icient of agreement of nominal scales. Educational
and Psychological Measurement 20, 37–46. doi:10.1177/00131644600
Corsi, F., Dupre, E., and Boitani, L. (1999). A large-scale model of wolf
distribution in Italy for conservation planning. Conservation Biology
13(1), 150–159. doi:10.1046/j.1523-1739.1999.97269.x
East, R. (1999). ‘African Antelope Database 1998.’ (IUCN/SSC Antelope
Specialist Group: Gland, Switzerland.)
Egziabher, T. B. (1988). Vegetation and environment of the mountains of
Ethiopia: implications for utilization and conservation. Mountain
Research and Development 8(2/3), 211–216. doi:10.2307/3673449
Embaye, K. (2000). The indigenous bamboo forests of Ethiopia:
an overview. Ambio 29(8), 518–521.
Environmental Systems Research Institute (ESRI) (2006). ArcGIS software,
version 9.2. Available online at http://www.esri.com/ [Verif ied July
Evangelista, P. (2006a). Status report on the mountain nyala (Tragelaphus
buxtoni). Technical Report to the IUCN Antelope Specialist Group.
Evangelista, P. (2006b). Range and distribution of the mountain nyala
(Tragelaphus buxtoni). Technical Report to the Ethiopian Wildlife
Conservation Department and Oromia Rural Land and Natural
Evangelista, P., Stohlgren, T., Guenther, D., and Stewart, S. (2004).
Vegetation response to fire and postburn seeding treatments in juniper
woodlands of the Grand Staircase-Escalante National Monument, Utah.
Western North American Naturalist 64(3), 293–305.
Evangelista, P., Swartzinski, P., and Waltermire, R. (2007). profile of the
mountain nyala (Tragelaphus buxtoni). African Indaba 5.
Evangelista, P., Kumar, S., Stohlgren, T., Jarnevich, C. S., Crall, A.,
Norman J. B. III, and Barnett, D. (in press). Modelling invasion for a
habitat generalist and a specialist plant species. Diversity &
Farr, T. G. (2007). The Shuttle Radar Topography Mission Reviews of
Geophysics 45, RG2004. doi:10.1029/2005RG000183
Felix, A., Campa, H., Millenbah, K., Winterstein, S., and Moritz, W. (2004).
Development of landscape-scale habitat-potential models for forest
wildlife planning and management. Wildlife Society Bulletin 32(3),
Ficetola, G. F., Thuiller, W., and Miaud, C. (2007). Prediction and validation
of the potential global distribution of problematic alien invasive species
– the American bullfrog. Diversity & Distributions 13, 476–485.
Fielding, A. H., and Bell, J. F. (1997). A review of methods for the assess-
ment of prediction errors in conservation presence/absence models.
Environmental Conservation 24(1), 38–49. doi:10.1017/S0376892997
Food and Agriculture Organization of the United Nations (FAO) (2006).
FAO Forestry Country profiles. Available online at http://www.fao.org/
[Verified July 2008]
Gamachu, D. (1988). Some patterns of altitudinal variation of climatic ele-
ments in the mountainous regions of Ethiopia. Mountain Research and
Development 8(2/3), 131–138. doi:10.2307/3673439
Global Land Cover Facility (GLCF) (2006). Global Land Cover Facility.
Available online at http://glcf.umiacs.umd.edu/ [Verified July 2008].
Gonfa, L. (1996). ‘Climate Classifications of Ethiopia.’ (National
Meteorological Service Agency: Addis Ababa.)
416 Wildlife Research
Hansen, M., Dubayah, R., and Defries, R. (1996). Classification trees: an
alternative to traditional land cover classifiers. Journal of Remote
Sensing 17, 1075–1081. doi:10.1080/01431169608949069
Hansen, M., DeFries, R., Townshend, J. R., Carroll, M., Dimiceli, C., and
Sohlberg, R. (2003a). Vegetation Continuous Fields MOD44B, 2001
Percent Tree Cover, Collection 3, University of Maryland, College Park,
Hansen, M., DeFries, R. S., Townshend, J. R. G., Carroll, M., Dimiceli, C.,
and Sohlberg, R. A. (2003b). Global percent tree cover at a spatial reso-
lution of 500 meters: first results of the MODIS vegetation continuous
fields algorithm Earth Interactions 7(10), 1–15. doi:10.1175/1087-
Hatten, J. R., and Paradzick, C. E. (2003). A multiscaled model of south-
western flycatcher breeding habitat. The Journal of Wildlife
Management 67(4), 774–788. doi:10.2307/3802685
Hijmans, R. J., Cameron, S. E., Parra, J. L., Jones, P. G., and Jarvis, A.
(2005). Very high resolution interpolated climate surfaces for global
land areas. International Journal of Climatology 25, 1965–1978.
Hijmans, R. J., Cameron, S. E., and Parra, J. L. (2006). WorldClim Global
Climate Layers Version 1.4, available from WorldClim database:
http://www.worldclim.org [Verified July 2008].
Insightful (2000). S-PLUS software, Professional Release 3. Available
online at http://www.insightful.com/products/splus/default.asp
[Verified July 2008].
Jarvis, A., Reuter, H. I., Nelson, A., and Guevara, E. (2006). Hole-filled
SRTM for the globe Version 3, available from the CGIAR-CSI SRTM
90m Database: http://srtm.csi.cgiar.org [Verified July 2008].
Kidwell, K. B. (Ed.) (1990). ‘Global Vegetation Index User’s Guide.’
(NOAA/NESDros. Inf. Serv.: Washington, DC.)
Kidwell, K. B. (Ed.) (1991). ‘NOAA polar orbiter data (TIROS-N,
NOAA-6, NOAA-7, NOAA-8, NOAA-9, NOAA-10, NOAA-11, and
NOAA-12) Users Guide.’ (NOAA/NESDros. Inf. Serv.: Washington,
Kumar, L., Skidmore, A. K., and Knowles, E. (1997). Modeling topographic
variation in solar radiation in a GIS environment. International Journal
of Geographical Information Science 11(5), 475–497. doi:10.1080/
Landis, J. R., and Koch, G. C. (1977). The measurement of observer agree-
ment for categorical data. Biometrics 33, 159–174. doi:10.2307/
Lewis, R. J. (2000). ‘An introduction to classification and regression
(CART) analysis. Presented at the 2000 Annual Meeting of the Society
for Academic Emergency Medicine in San Francisco, CA.’
Logan, W. E. (1946). An introduction to the forests of central and southern
Ethiopia. Paper No. 24, Imperial Forestry Institute, University of
Lydekker, R. (1910). The spotted kudu. Nature 84, 396–397. doi:10.1038/
Lydekker, R. (1911). On the mountain nyala, Tragelaphus buxtoni.
Proceedings of the Royal Zoological Society of London 1911, 348–353.
Malcolm, J., and Evangelista, P. (2005). The range and status of the moun-
tain nyala (2004). Technical Report to the Ethiopian Wildlife
Conservation Department and Oromia Rural Land and Natural
McClanahan, T. R., and Young, T. P. (Eds) (1996). ‘East African Ecosystems
and their Conservation.’ (Oxford University Press: New York.)
Michaelsen, J., Schimel, D., Friedl, M., Davis, F., and Dubayah, R. (1994).
Regression tree analyses of satellite and terrain data to guide vegetation
sampling and surveys. Journal of Vegetation Science 5, 673–686.
Miehe, S., and Miehe, G. (1994). ‘Ericaceous Forests and Heathlands in the
Bale Mountain of South Ethiopia.’ (Stiftung Walderhaltung in Afika:
P. H. Evangelista et al.
Ministry of Water Resources (MoWR) (2001). Initial National
Communication of Ethiopia to the United Nations Framework
Convention on Climate Change. National Meteorological Services
Agency, Addis Ababa.
Moore, I. D., Grayson, R. B., and Ladson, A. R. (1991). Terrain based catch-
ment partitioning and runoff prediction using vector elevation data.
Water Resources Research 27, 1177–1191. doi:10.1029/91WR00090
Myneni, R. B., Hall, F. G., Sellers, P. J., and Marshak, A. L. (1995). The
interpretation of spectral vegetation indexes. IEEE Transactions on
Geoscience and Remote Sensing 33, 481–486. doi:10.1109/36.377948
Nix, H. A. (1986). biogeographic analysis of Australian elapid snakes. In
‘Australian Flora and Fauna’. (Ed. R. Longmore.) Series 8, pp. 4–15.
(Australian Government Printing Service: Canberra.)
Pearce, J., and Ferrier, S. (2000). Evaluating the predictive performance of
habitat models developed using logistic regression. Ecological
Modelling 133, 225–245. doi:10.1016/S0304-3800(00)00322-7
Pearson, R. G., Raxworthy, C. J., Nakamura, M., and Townsend Peterson, A.
(2007). Predicting species distributions from small numbers of occur-
rence records: a test case using cryptic geckos in Madagascar. Journal
of Biogeography 34, 102–117. doi:10.1111/j.1365-2699.2006.01594.x
Pfeiffer, D. U., and Hugh-Jones, M. (2002). Geographical information
systems as a tool in epidemiological assessment and wildlife disease
management. Office International des Epizooties Scientific and
Technical Review 21(1), 91–102.
Phillips, S., Anderson, R., and Schapire, R. (2006). Maximum entropy
modeling of species geographic distributions. Ecological Modelling
190(3–4), 231–259. doi:10.1016/j.ecolmodel.2005.03.026
Refera, B., and Bekele, A. (2004). Population status and structure of moun-
tain nyala in the Bale Mountains National Park, Ethiopia. African
Journal of Ecology 42, 1–7. doi:10.1111/j.0141-6707.2004.00437.x
Russell, R., Saab, V., and Dudley, J. (2007). Habitat-suitability models for
cavity-nesting birds in postfire landscape. The Journal of Wildlife
Management 71(8), 2600–2611. doi:10.2193/2007-034
Sellers, P. J. (1985). Canopy reflectance, photosynthesis, and transpiration
International Journal of Remote Sensing 6, 1335–1372. doi:10.1080/
Sillero-Zubiri, C. (2007). Tragelaphus buxtoni. In ‘IUCN Red List of
Threatened Species’. (IUCN: Gland, Switzerland.) Available online at
http://www.iucnredlist.org [Verif ied July 2008].
Turner, J. C., Douglas, C. L., Hallum, C. R., Krausman, P. R., and
Ramey, R. R. (2004). Determination of critical habitat for the
endangered Nelson’s bighorn sheep in southern California. Wildlife
Society Bulletin 32(2), 427–448. doi:10.2193/0091-7648(2004)32[427:
Uhlig, S., and Uhlig, K. (1991). Studies on the altitudinal zonation of forests
and alpine plants in the central Bale Mountains, Ethiopia. Mountain
Research and Development 11(2), 153–156. doi:10.2307/3673574
Von Breitenbach, F. (1961). Forests and woodlands of Ethiopia. Ethiopian
Forestry Review 1, 5–16.
Wolock, D. M. (1993). TOPMODEL, a topography based watershed model.
In ‘Proceedings of the Federal Interagency Workshop on Hydrologic
Modeling Demands for the 90’s, USGS Water-Resources Investigations
Report 93–4018’. pp. 8–57.
Yalden, D. W., and Largen, M. J. (1992). The endemic mammals of Ethiopia.
Mammal Review 22(3–4), 115–150. doi:10.1111/j.1365-2907.1992.
Manuscript received 12 November 2007, accepted 9 May 2008