Reintroducing a large herbivore: a remote sensing and modeling approach to determine the mountain bongo

1
Reintroducing a large herbivore a remote sensing
and modeling approach to determine the mountain
bongos (Tragelaphus euryceros isaaci) past and
present critical habitatL.D. Estesa,b, A.G.
Mwangic, G.S. Okind, P.R. Reilloa, H.H. Shugartb,
H.M. Wilburb aRare Species Conservatory
Foundation, bDepartment of Environmental
Sciences, University of Virginia, cSchool of
Natural Resource Management, Moi University,
dDepartment of Geography, University of
California Los Angeles
Background and Introduction The mountain bongo
(Figure 1), an endangered antelope endemic to
East African montane forests, has declined
dramatically over the past 40 years in the face
of increasing human pressure and habitat change.
To reverse this decline, an ambitious
international reintroduction project seeks to
re-establish a viable wild population on the
Mount Kenya World Heritage Site using captive
bongo from North America.

• Predictive models
• Initially, logistic regression models will be
  developed with variables selected using an
  information theoretic approach (4).
• These will be compared with enhanced models that
  account for false absences arising from
  variations in bongo detectibility between seasons
  and habitats (5).
• Model will be developed and verified with the
  Aberdares dataset, and validated with the smaller
  Mount Kenya dataset.
• If microsatellite DNA analysis is feasible, the
  resulting abundance and range size information
  will allow a richer ecological assessment.

• The bongo data points (n 56) were standardized
  and converted to PC scores for the first three
  PCs using the eigen vectors from Table 1.
• Figure 3 offers a graphical presentation of the
  distribution of grid and bongo data in PC space.

• Results to Date
• Field data
• Three field seasons (1st 6-8/2005 2nd
  2-3/2006 3rd 5-6/2006) comprising 92 expedition
  days have been completed. The timing of these
  coincided with the three major montane climatic
  seasons (rainy, dry, and misty).
• Two additional expeditions (12 days total) were
  made to southern Mount Kenya during June and
  July, 2006.
• A total of 56 bongo data points and 94 grid
  points were collected from the Aberdares, and 8
  bongo points (with a reduced variable set) from
  Mount Kenya (Figure 2). The nature and location
  of human sign were also recorded where
  encountered. 88 dung samples were collected (and
  have been combined with 144 collected by the BSP
  for the population genetics analysis).

Many slender trees
The mountain bongo is poorly understood, and the
success of this conservation effort depends on
answering two key questions

1. What are the bongos important habitat factors,
   and how are these configured?
2. How have these habitat factors changed?

However, studying this rare, shy animal in its
mountain habitat presents challenges

• Image analysis and modelling
• An analysis of imagery to determine vegetation
  structure is in progress. Once complete, the
  analysis to determine the distribution of forest
  types will commence.
• Predictive models will be developed following the
  completion of the satellite image analysis.

• collecting large and spatially comprehensive
  field datasets is difficult
• key habitat factors are likely to operate at
  different scales (e.g. 1).
• Field data will be related to current and
  historical satellite imagery to overcome these
  limitations. Variables derived from these data
  will be used in generating predictive models that
  will provide insight into the current and
  historical configuration of bongo habitat. This
  approach will allow two further questions to be
  addressed
• 1) Can remotely sensed data be successfully used
  to map the key habitat variables of a large
  herbivore in a rugged, forested landscape?
• 2) Given a relatively small dataset, can a
  probabilistic habitat model successfully predict
  a large, mobile, and rare organisms habitat?
• A population genetics study of collected faecal
  DNA being conducted at Cardiff University is
  expected to yield demographic and range size data
  that will enhance habitat modelling.

• Anticipated Outcomes
• This applied conservation ecology research will
  provide the science for a regionally important
  conservation project by
• 1) delineating suitable bongo habitat
• 2) identifying the management actions necessary
  for conserving the Aberdares population and
  creating conditions for a viable wild Mount Kenya
  population.
• The broader contribution of this approach to the
  fields of conservation biology and ecology will
  be to
• Further demonstrate the ability of remotely
  sensed data to 1) characterize important
  ecological factors 2) enhance the
  spatio-temporal coverage of field data where it
  is difficult to collect
• Evaluate the ability of one of the latest
  quantitative habitat modeling techniques to
  predict habitat utilization of a rare and highly
  mobile organism
• Improve understanding of the role of spatial
  configuration and complexity in habitat
  selection
• Enhance understanding of the relationship between
  changing habitat configurations and large
  herbivores
• Devise an approach applicable to other
  reintroduction and ecosystem management programs.

• Methods
• Field data
• Data were collected from the bongo population in
  the Aberdare mountains. A rediscovered herd on
  Mount Kenya has also yielded a small dataset.
• Expert trackers from the local Bongo Surveillance
  Programme (BSP) were employed to identify points
  of bongo habitat use (e.g. feeding, resting).
• Vegetation structural and compositional data
  (DBH, shrub density, ground and canopy cover,
  canopy height, herbaceous height) were collected
  within 11.3 m radius fixed plot centered on bongo
  sign.

Acknowledgements Field collaboration S. Gichure,
K. Gichuri, M. Gichuri, J. Kariuki, L. Kariuki,
J. Korage, P. Mwangi, B. Nderitu, and M.G.
Prettejohn of the Bongo Surveillance Programme.
The Kenya Wildlife Service, particularly Asst.
Dir. J. Warutere, Senior Warden K. Wambani, Asst.
Warden Gichohi, and Dep. Dir. R. Bagine. J. and
H. Henley of Honi Farm. Support NASA (Earth
System Science Fellowship), Wildlife Conservation
Society (Conservation Research Fellowship),
Explorers Club Washington Group (Exploration and
Field Research Grant), and U. Virginia Dept.
Environmental Science (Exploratory Research
Award).

• Exploratory data analysis
• To examine data patterns at the microhabitat
  level, a principal components analysis (PCA) was
  performed using all 8 variables on the grid data
  set (n 94).
• The first 3 PCs account for 66 of the variance
  (see Table 1 for eigen vectors).

• The same information was recorded at pre-assigned
  points centered in 1 km2 grid cells, collected as
  survey tracks passed within 200 m.

• Detectibility was assessed with repeat counts of
  proxy species sign.
• DNA extracted from dung samples collected during
  fieldwork and by BSP will be used to confirm
  identity of species and to determine demography.
• Remote sensing
• Spectral mixture analysis of ASTER, Landsat, and
  MODIS data and texture analysis of SPOT 5m
  panchromatic data will be used to map vegetation
  structure in study area.
• Vegetation types will be classified with a
  technique that incorporates prior probabilities
  (e.g. 2), which will be generated with a digital
  elevation model based model (e.g. 3).
• Information on seasonal variation in forage
  quality will be extracted from multi-date MODIS
  scenes with NDVI.
• Landsat TM data will be used to determine the
  historical distribution of successfully mapped
  variables.

 Variables PC1 PC2 PC3 PC4 PC5 PC6 PC7 PC8
Basal area 0.41 0.36 -0.02 0.12 -0.52 0.40 0.46 -0.22
Trees/ha 0.12 0.57 -0.37 -0.07 0.53 -0.04 0.29 0.40
Canopy hgt 0.48 0.26 0.27 -0.11 -0.23 -0.05 -0.60 0.45
Bamboo/ha 0.30 -0.51 0.30 0.04 0.33 0.50 0.25 0.36
Shrubs/ha -0.33 0.29 0.10 0.66 0.18 0.47 -0.34 -0.05
Canopy cover 0.48 0.10 0.31 0.12 0.48 -0.24 -0.06 -0.60
Ground cover -0.27 0.27 0.29 -0.71 0.14 0.43 -0.09 -0.22
Herb height -0.29 0.23 0.71 0.13 -0.08 -0.35 0.40 0.24
References (1) Johnson, C. J., Seip, D.R., Boyce,
M.S. 2004. A quantitative approach to
conservation planning using resource selection
functions to map the distribution of mountain
caribou at multiple spatial scales. J Appl
Ecology 41238-251. (2)Horsch, B. 2003. Modelling
the spatial distribution of montane and subalpine
forests in the central Alps using digital
elevation models. Ecological Modelling
168267-282 (3)Pedroni, L. 2003. Improved
classification of Landsat Thematic Mapper data
using modified prior probabilities in large and
complex landscapes. International Journal of
Remote Sensing 24(1)91-113. (4)Burnham, K.P.,
Anderson, D.R. 2002. Model Selection and
Multi-Model Inference A Partical Information
Theoretic Approach (2nd Ed.). Springer-Verlag,
New York. 488 pp. (5)MacKenzie, D.I., Nichols,
J.D., Royle, J.A, Pollock, K.H., Bailey, L.L.,
Hines, J.E. 2006. Occupancy Estimation and
Modeling Inferring Patterns and Dynamics of
Species Occurrence. Elsevier, London. 324 pp.