Range collapse of a tropical cervid (Cervus eldi) and the extent
of remaining habitat in central Myanmar
INTRODUCTION
The conservation of large mammals in tropical ecosys-
tems is a difficult task, due to many factors, including
fragmentation of remaining refuges (Laurence, 1991;
Newmark, 1996), and poaching for food (Alvard et al.,
1997; Bodmer, Eisenberg & Redford, 1997), trophies, or
medicinal products (Geist, 1994; Heinen &
Srikosamatara, 1996). Consensus on protecting large
mammals is complicated by conflicts between animals and
agriculture (Fox, Yonzon & Podger, 1996; but see
Saberwal, 1996), their distribution across political and
legal boundaries (Keiter & Locke, 1996), and by compe-
tition between conservation organizations (Rabinowitz,
1995). Increasing human populations threaten tropical for-
est ecosystems and all large mammals associated with
these systems. This problem is compounded by the rela-
tive lack of protection and funding for many tropical coun-
tries (Balmford & Long, 1995). We report on our effort
to assess the remaining range and available habitat for a
rapidly disappearing cervid, the thamin or Eld?s deer
(Cervus eldi thamin), now restricted to Myanmar, a trop-
ical country experiencing many of the problems that
impede conservation efforts for large mammals. Our pur-
pose is to use the decline of thamin as a case study of
which landscape and habitat features are tied to the decline
of a large mammal species in tropical forests.
The Eld?s deer (C. eldi eldi) was first discovered in
the Manipur Valley of India in 1838 by a British army
officer, Lieutenant Percy Eld. Three subspecies are rec-
ognized as occurring in India, Myanmar and Indo?China
(Whitehead, 1972). During this century, the once wide-
spread species has disappeared throughout most of its
range (Fig. 1). By 1945 it was described as rare or endan-
gered throughout the southern portions of its range
(Harper, 1945), and the western subspecies in Manipur
(C. eldi eldi) was considered extinct until a small pop-
ulation was located in the 1950s (Ranjitsinh, 1978). In
1986, C. eldi eldi was restricted to 100 individuals in
Indian zoos and 50 individuals in Keibul Lamjao
National Park, Manipur (Salter & Sayer, 1986). The
Indo?Chinese subspecies, C. eldi siamensis, is on the
brink of extinction (Humphrey & Bain, 1990), with only
Animal Conservation (1999) 2, 173?183  ? 1999 The Zoological Society of London  Printed in the United Kingdom
William J. McShea1, Peter Leimgruber1, Myint Aung2, Steven L. Monfort1 and Christen Wemmer1
1Conservation and Research Center, National Zoological Park, 1500 Remount Rd., Front Royal, VA 22630, USA
2Department of Wildlife, West Gyogone, Insein, Yangon, Myanmar
(Received 5 July 1998; accepted 1 December 1998)
Abstract
The thamin (Cervus eldi) is an endangered species of deer whose present range is greatly reduced
from its original distribution covering the deciduous forests throughout south-east Asia. We mapped
the present distribution of thamin using ground surveys and tied this information to habitat types
derived from satellite images in order to detect patterns that might indicate the landscape features
which precipitated the decline. We conducted a survey of 24 out of 28 Myanmar townships that were
reported to contain thamin in 1992, and evidence of thamin were found in 23 of these townships, pre-
dominately in mixed deciduous forests where dipterocarp trees were present. There was no signifi-
cant correlation between the number of thamin detected and forest remaining in the township, or the
size of the human or livestock population. A landcover classification of Landsat Thematic Mapper
images indicated 58% of the study area contained deciduous forest, of which 12% was dipterocarp
forest. Forest tracts containing thamin were digitized and landscape analyses were conducted on a
resampled habitat map that emphasized dry and dipterocarp deciduous forest. Of six landscape vari-
ables measured only core area size was a significant predictor for the presence of thamin. None of
the unsurveyed indaing forest tracts possessed a core area large enough to support thamin. The pat-
tern of thamin decline matches predictions that peripheral, rather than central, populations are more
likely to persist in declining species. 
All correspondence to: William J. McShea. Tel: 540-635-6563;
Fax: 540-635-6506; E-mail: wmcshea@crc.si.edu.
one known population of approximately 150 individuals
within a fenced reserve in Hainan (Song, 1996) and pos-
sibly a population in northern Cambodia (Humphrey &
Bain, 1990). Populations of the third subspecies, C. eldi
thamin, occur in scattered areas of central Myanmar
(Tun Yin, 1967; Salter & Sayer, 1986). Within
Myanmar, thamin were considered common throughout
the 1940s, and their distribution was described as patchy,
but abundant, in 1967 and 1986 (Tun Yin, 1967; Salter
& Sayer, 1986).
Thamin are grazers and opportunistic browsers, who
supplement their diet with wild fruit and cultivated
crops, particularly rice (Lekagul & McNeely, 1977).
Probably few pristine thamin habitats remain, and pop-
ulations have been reported to occupy areas ranging
from dry scrub and thorn forest to open deciduous forest
(Salter & Sayer 1986; Wemmer, 1987). The least dis-
turbed of these habitat types is deciduous forest.
Deciduous dipterocarp (indaing) forest is the most
abundant forest type within south-east Asia (Rundel &
Boonpragob, 1995). In 1990, Myanmar was approxi-
mately 50?58% forested, with 14% covered with some
form of deciduous forest (UNEP, 1995; Achard &
Estreguil, 1995). Rainfall, soil type and terrain account
174 W. J. MCSHEA ET AL.
Fig. 1. Historical and present range of Cervus eldi in south-east Asia. Historical range, indicated by diagonal stripes, is derived
from maps by Whitehead (1972) and shows the range of the three subspecies C. eldi eldi (A), C. eldi thamin (B) and C. eldi
siamensis (C). The present range, indicated by stippling, is adapted from Tun Yin (1967) and Salter & Sayer (1986) and includes
two semi-captive populations in India and Hainan Island, and a postulated (Humphrey & Bain, 1990) population in northern
Cambodia (?).
for most of the variation in forest types across south-
east Asia (Richards, 1979; Rundel & Boonpragob,
1995). Although rainfall amounts are variable across the
historical thamin range, the remaining thamin range in
central Myanmar receives 100 cm of rain during the
monsoon season (Salter & Sayer, 1986). Thamin are
considered to occur only in deciduous forests (Tun Yin,
1967; Salter & Sayer, 1986; Humphrey & Bain, 1990),
of which three types are found in this region; diptero-
carp (indaing), dry (thandahat) and mixed deciduous
(teak) (Stamp, 1925). All three types are confined to the
region that annually receives 100?200 cm of rainfall
(Stamp, 1925). Determining the abundance and distrib-
ution of these forest types is a focus of our research.
Myanmar has one of the highest proportions of forest
cover in the Asian?Pacific region (UNEP, 1995). This
forest inventory was based on NOAA AVHRR satellite
data and it estimated cover of closed forest at 43.3%
(29.33 million ha) of the country. Although forest cover,
in general, is extensive in Myanmar, there may be sig-
nificant losses of deciduous forest types. In Thailand, the
only south-east Asian country for which reliable data are
available, indaing forest has declined 45?60% over the
last 20 years (Rundel & Boonpragob, 1995). Overall
deforestation rates in Myanmar were estimated at one-
third those of Thailand for 1980?1991 (Achard &
Estreguil, 1995).
The thamin population in Myanmar was estimated to
be 4000 animals in the 1970s (Whitehead, 1972), but the
first countrywide questionnaire distributed by the
Wildlife Department in 1992 estimated that 2200 deer
remained within Myanmar, with the largest population
(>1200 deer) in Chatthin Wildlife Sanctuary (Myint
Aung, 1994). This survey reported thamin in 28 town-
ships, six townships fewer than reported in 1967 by
Tun Yin (1967) (Fig. 2). Chatthin Wildlife Sanctuary
(268 km2) is composed of secondary indaing and mixed
deciduous forest, and was originally established as a fuel
reserve forest in 1941. Sight surveys for thamin along
65 km of line transects were conducted at Chatthin
beginning in 1982 by the Food and Agriculture
Organization (FAO: Salter & Sayer, 1986), and were
repeated annually from 1992?1996 by forest rangers of
the Wildlife Department (Table 1). These surveys esti-
mated a 40% decline in thamin abundance since 1983.
Despite declines in the abundance of Eld?s deer, both
across its entire range and within Myanmar, no system-
atic field studies have been conducted. With respect to
Myanmar, this is largely because it is one of the least
externally-funded and internally-protected tropical coun-
tries in Asia (Balmford & Long, 1995). Our goals were
to conduct a ground survey for thamin across the cen-
tral plains of Myanmar, and to relate information on
thamin abundance and distribution to maps of forest
abundance and distribution. The result may serve both
as a management guide for proposed protected areas, and
to help determine landscape patterns that might explain
the decline in the species.
METHODS
To determine the status of thamin and the extent of
potential remaining habitat in central Myanmar, we con-
ducted a series of field surveys and used the survey data
to delineate forest stands on landcover maps derived
from satellite imagery.
Thamin surveys
Surveys were conducted by Wildlife Department per-
sonnel in spring 1997. All 28 townships identified in
1992 as containing deer were contacted, and 24 town-
ships were visited by survey teams. The survey team first
conducted an interview with the Forest Department per-
sonnel responsible for the inventory and protection of
all forest tracts within each township. On Forest
Department maps, forestry personnel indicated the
175Conservation of thamin
Fig. 2. Townships in Myanmar that have been reported to con-
tain thamin in the last 30 years. The diagonally striped town-
ships were listed as containing thamin by Tun Yin (1967), but
did not contain thamin in 1992. The other townships were con-
sidered likely to contain thamin by Wildlife Department per-
sonnel based on questionnaires completed in 1992. Dark grey
townships are those surveyed in 1997. All the townships sur-
veyed in 1997 contained some evidence of thamin with the
exception of Bundalin township (black area on map).
Table 1. Annual estimates of thamin abundance within Chatthin
Wildlife Sanctuary based on line transect surveys conducted during
the first week of April
1983 1991 1992 1993 1994 1995 1996
Kilometers walked 81 94 76 73 81 174 87
Animals sighted 147 187 89 72 80 137 111
Density (deer/km2)? 8.3 9.3 7.8 4.7 4.8 4.1 4.7
? Haynes estimator, Buckland et al. (1993).
extent of forests within the township and which forest
tracts contained thamin based on recent sightings. A sec-
ond interview was conducted in each township with
villagers who were considered knowledgeable about
wildlife. Using the maps of forest cover produced during
the first interview, the villagers were also asked to iden-
tify forest tracts that may contain thamin. The adminis-
trative office of each township provided 1997 census
records on population size of humans and livestock, and
estimated forest cover. Forestry personnel estimated the
total number of thamin within their township.
As part of each township survey a field census was
conducted, and consisted of at least one set of transects
traversing a forested area. The first transect of each set
traversed the forest tract, the second transect was laid
out perpendicular to the first, and the third transect
returned in the opposite compass direction of the first
transect. When forest tracts were contiguous between
adjoining townships, the information was combined
prior to analysis (Table 2). Transect surveys for each
township ranged from 9?236 km (X? = 44 km), depend-
ing on the amount of forest cover. One walker was
responsible for detecting tracks and faecal pellet groups
along each transect, and the second walker concentrated
on sighting deer. There was a significant correlation
between the number of signs recorded and the number
of deer observed (linear regression: F = 4.48, d.f. = 1,18,
r2 =0.20, P = 0.048). We used the number of signs
detected along the transects, rather then the thamin sight-
ing records, for an abundance index because these mea-
sures were less sensitive to variability in vegetation
density across the habitats. The surveyors recorded total
distance walked, and the distance walked in each habi-
tat type, along each transect by counting their steps and
converting that measure into meters. All forest types that
covered >10% of the transect length are listed for the
township (Table 2). For each thamin sighting the walker
recorded the number of deer, their distance from the
walker, and the angle of the sighting relative to the tran-
sect line (Buckland et al., 1993). At Chatthin Wildlife
Sanctuary (Kambalu and Kawlin townships) and
Shwesettaw Wildlife Sanctuary (Minbu township) 
the transects are permanently marked and separated by
1.5 km.
The number of tracks and faecal pellet groups
observed along a transect were combined to create a
single abundance index. Both the number of signs and
deer observations along a transect were divided by the
number of kilometers walked. Estimates of population
and livestock from census records were divided by the
size of the township. Percent forest cover within each
township was arcsin transformed before analysis, and
abundance measures were log transformed if they were
not normally distributed. Stepwise linear regression
(backwards) and ANOVA were performed using SAS
(SAS, 1987) on human census information that might be
correlated with the abundance of thamin within a town-
ship. The presence/absence of thamin within forest tracts
and landscape variables for each stand was examined
using discriminant function analysis. The relationship
between the number of thamin estimated by foresters
and the number detected along the transects was tested
using Pearson?s correlation coefficients.
Landscape analyses and regional assessment
We used Arc/Info to digitize the outline of forest tracts,
as drawn by Wildlife Department personnel on govern-
ment township maps. We registered the final map to a
landcover map produced from remote sensing analyses
of Landsat TM imagery. To include the data from the
field surveys for thamin in the digital map, we assigned
176 W. J. MCSHEA ET AL.
Table 2. Survey information for townships
Township Area Population/ Livestock/ Forest Forest type? Distance Distance Number of 
(km2) km2 km (%) walked (km) sign/km deer/km
Thegon & Paungde 199 1258 522 23 In forest 11 1.1 0
Pyi 79 281 850 13 In forest 9 1.8 0
Minbu 167 846 413 48 Dry/In forest 37 0.19 0
Nga-Phae 131 281 214 68 Dry forest 53 0.3 0
Pwintbyu 122 1168 525 39 Dry forest 25 0.72 0
Salin 232 886 537 51 Dry forest 13 0.15 0
Mahlaing 111 595 657 25 Dry forest 17 0.05 0
Pauk & Pakokku 375 129 517 24 Dry/In forest 73 0.23 0.12
Nahtogyi 125 1493 1112 4 Dry forest 26 0.61 0
Ye-Sa-Kyo 100 2372 1118 12 Dry forest 27 0.63 0.03
Salingyi & Yinmabin 162 1511 1006 10 Dry forest 35 1.17 0.85
Wetlat 156 2047 785 6 In forest 20 1.65 0.3
Bundalin 107 1171 674 14 Dry forest 10 0 0
Depeyin 133 1019 507 48 Dry/In forest 29 0.4 0.03
Kane 336 406 234 52 Md/In forest 30 0.4 0.1
Khin-Oo 104 2507 1439 19 In forest 70 0.47 0.03
Yae-Oo 145 721 399 81 In forest 49 1 0
Tanse 186 765 406 49 In/Md forest 91 0.48 0
Kyunha 262 253 184 33 Md/In forest 15 2 0
Kambalu & Kawlin 531 632 405 42 In/Md forest 236 3.38 0.44
Townships were combined if forest tracts surveyed crossed township boundaries. Population, livestock and % forest values are from 1997 township census records. Forest
type is dominant type in township, when mixed the more common is listed first. The townships are listed from south to north.
? In, indaing forest type; Md, mixed deciduous forest type
each forest polygon attributes for forest type and the
presence or absence of thamin.
The present forest cover maps may be suited to assess
trends in habitat fragmentation and conversion for south-
east Asia (UNEP, 1995), but are not suitable for deter-
mining the remaining indaing and dry forest habitats in
Myanmar. In the past, Landsat TM imagery has been
successfully used for vegetation and landcover mapping
(Lillesand & Kiefer, 1994), so we used this imagery to
produce an accurate map of deciduous forest for central
Myanmar. We obtained two sets of mosaicked, multi-
spectral Landsat TM 4 images from the USGS EROS
Data Center. Both sets of images were acquired in
January 1989, the first covered the northern and central
parts of our study region, while the second set covered
the southern section of our study region. Images were
geometrically corrected, projected to a UTM grid, and
resampled using a nearest neighbour algorithm. Spatial
resolution, the size of the picture elements (pixels), was
28.5 m ? 28.5 m.
The classification of satellite imagery into a landcover
map is usually performed with a set of target classes in
mind that are described in a hierarchical classification
system (Lillesand & Kiefer, 1994; ERDAS, 1997). Our
landuse classification emphasized deciduous forest
types, because we were interested in an accurate delin-
eation of actual and potential habitat for thamin, specif-
ically indaing and dry forest. Our landuse classes were:
? Evergreen Forest: Evergreen forests are characterized
by diverse overstories of evergreen trees composed of
several Dipterocarp species with sparse understories
that may contain ferns or Selaginella (Stamp, 1925).
In our study region, the forests are moist and are gen-
erally found only in mountainous areas, specifically
along stream banks and ravines.
? Mixed Deciduous Forest: These are diverse forests
that contain teak (Tectona grandis), along with Xylia
dolabriformis and Pterocarpus macrocarpus.
Different species of bamboo (Bambosa sp.) are com-
mon in the understory. Generally, the vegetation in
this forest type is not as dense as in evergreen forests.
Mixed deciduous forests are found in moist areas and
along streams in central Myanmar, as well as on the
slopes of the mountainous regions.
? Indaing Forest: In central Myanmar, indaing forest is
found in sandy, flat terrain that seasonally floods and
is dominated by Dipterocarpus tuberculatus, Shorea
obtusa, Terminalia tomentosa and Melanorrhoea
usitata (Stamp, 1925). These forests are open with
sparse understories and abundant growth of grasses
(Stamp, 1925). Indaing forest is commonly found in
central Myanmar below an elevation of 1000 m
(UNEP, 1995).
? Dry Forest: This category comprises dry deciduous
forests and thorn forests, with widely spaced trees and
little ground vegetation. Dry forest (thandahat) is char-
acterized by Tectona hamiltonii, Terminalia oliveri
and Acacia catechu (Stamp, 1925).
? Grassland: This category comprises seasonally
flooded grasslands, but also pastures, and lwins in the
open indaing forests.
? Agriculture: This category comprises areas used in
cultivation, including crops, degraded shrublands, or
pastures.
? Barren: This category comprises rocky outcrops,
sandbanks in large rivers, developed lands, and heav-
ily eroded areas that were previously degraded by
agriculture and logging.
? Water: This category comprises lakes, reservoirs and
rivers.
Indaing and dry forest are restricted to the low elevation
areas (UNEP, 1995) and thamin are not reported in town-
ships whose average elevation is above 1000 m. In order
to reduce topographic effects that tend to complicate the
differentiation of forest types in regions with strong
topographic relief (Lillesand & Kiefer, 1994; ERDAS,
1997), areas above 1000 m were removed from our
images using digital elevation data from the Digital
Charts of the World data set (ESRI, 1993).
Using ERDAS Imagine image processing software,
we applied ISODATA clustering algorithms on bands 3,
4 and 5 of the Landsat TM images to produce 50 spec-
tral clusters in the initial classification (ERDAS, 1997).
Each of the 50 clusters was assigned to one of our land-
cover categories based on one investigator?s (W.J.M.)
knowledge of the country. In 1997, two investigators
(W.J.M. and P.L.) collected ground-truth information at
Chatthin Wildlife Sanctuary, Alaungdaw Kathapa
National Park and the intervening roadways, to evaluate
the accuracy of the landcover map. We used Global
Positioning Systems to determine the exact geographic
location of the reference points. After refining the clas-
sification, we calculated the overall accuracy of our land-
cover map using information from reference points that
were collected during the previous site visit. Overall
accuracy of the final classification was 79 %, with high
accuracies for indaing forest (85 %).
The landcover classification of central Myanmar was
overlaid with a map of the forest stands surveyed for
thamin. Using these digitized forest polygons and the
landcover map, we determined the % cover of each land-
cover type encompassed by the polygon. To facilitate
landscape analyses of thamin habitats, we removed all
landcover categories other than indaing or dry forest, and
resampled the image to a spatial resolution of 5 km ? 5
km. During resampling, each cell in the new raster map
that comprised at least 30 % of either indaing or dry for-
est was classified accordingly. Therefore, the smallest
patches of indaing or dry forest on the resampled map
are at least 25 km2 and are covered by no less than 
30 % of either dry or indaing forest. The minimum size
of the sampling unit was chosen because a current radio-
tracking study of thamin at Chatthin Wildlife Sanctuary
revealed a relatively large home range (6 km2; W. J.
McShea, pers. comm.) indicating the deer require large
areas, and anecdotal evidence of deforestation since the
Landsat images were taken in 1989 led us to believe
many of the smaller patches no longer existed.
177Conservation of thamin
178 W. J. MCSHEA ET AL.
Fig. 3. A landcover map based on five 1989 Landsat TM images. The boundaries (in blue) of the four established protected
areas within the study area are shown.
We used the FRAGSTATS (McGarigal & Marks,
1995) program to quantify landscape characteristics and
patterns of indaing and dry forest patches (see
Appendix). Patches are defined as a group of pixels that
are of one forest type, either indaing or dry forest, and
that are connected by at least one corner or one other
pixel. We were interested in the spatial relationships
between patches, their configuration in the landscape and
what role these landscape characteristics play in the dis-
tribution of thamin. Based on the results from our land-
scape analyses, we predicted where thamin are likely to
occur and which areas may have suitable habitat for re-
introduction and protection. We standardized the results
from FRAGSTATS to a mean of zero and a standard
deviation of ?1. Using the digitized maps from field and
township surveys, we identified patches that were sur-
veyed for thamin (n = 63). In a discriminant function
analysis we used this set of patches as training data to
identify the linear combinations of variables that best
separated patches with thamin from patches without
thamin. We then used the identified discriminant func-
tions to classify all unsurveyed patches into patches 
that either were likely or unlikely to support thamin
populations.
RESULTS
The classified landcover map incorporated 123 415 km2
of the central plains of Myanmar (Fig. 3). Forest cover
encompassed 60% of the area (Table 3). Of the three
deciduous forest types preferred by thamin, mixed decid-
uous (teak) forest was the most abundant (35%), fol-
lowed by dry forest (16%) and indaing forest (7%).
Using the resampled map that included only large (> 25
km2) forest tracts of predominantly (> 30 %) indaing or
dry forest, resulted in a lower estimate of dry forest (7%),
but a similar estimate for indaing forest (8%).
Surveys
The 24 townships were surveyed along 876 km of tran-
sects (see Table 2). Sign of thamin (i.e. faecal pellet
groups or tracks) were observed in 23 townships, with
actual sightings of deer in 12 townships (Fig. 4). Within
forest stands that contained thamin, the number of
sign/km of transect was not significantly correlated with
human population density (partial F = 0.02, d.f. = 1,18,
P > 0.1), livestock density (partial F = 1.17, d.f. = 1,18,
P > 0.1), or % forest cover (partial F= 0.23, d.f. = 1,18,
P > 0.1).
For 16 townships we have an estimate of thamin num-
bers obtained during the interview with local foresters.
The number of deer signs detected/km of transect was
correlated with the number of deer estimated by the
township foresters during interviews (r = 0.69, n = 16,
P < 0.002). However, the highest estimate for a town-
ship (Kambalu estimate = 1590) was more than 3 stan-
dard deviations (SD) above the other townships and
when this estimate is deleted there is no significant cor-
relation between forester estimates and estimates based
on the actual transects (r = ?0.12, n = 15, P >0.1 ).
Thamin were not evenly distributed across forest types
(ANOVA F = 3.16, d.f. = 3,16, P = 0.05). The highest
density of sign/km transect was observed in indaing/
mixed deciduous forest (X? = 1.56 (? 1.42) SD) and
indaing forest (X? = 1.24 (? 0.59) SD), while the
indaing/dry forest (X? = 0.27 (? 0.11) SD) and dry for-
est (X? = 0.45 (? 0.4) SD) types contained fewer deer.
Landscape analyses
There were 63 forest stands categorized by the field
teams that occurred within the boundaries of the land-
cover map. We used the classified landcover map to
determine forest composition within these stands. Forest
stands with thamin contained more indaing/dry forest,
179Conservation of thamin
Table 3. Landuse and forest type classification of study area based
on five 1989 Landsat TM images
Land class Area (km2) Land cover (%)
Water 1230 1
Barren 1988 1.6
Agriculture 41 323 33.5
Grassland 4269 3.5
Forest 74 606 60.4
Dry 19 224 15.6 (25.8)?
Indaing 8691 07 (11.6)?
Mixed deciduous (teak) 43 603 35.3 (58.4)?
Evergreen 3088 2.5 (4.1)?
Total 123 416
? Percentage of forest cover.
Fig. 4. The forest stands surveyed by Wildlife Department per-
sonnel in spring 1997. The stands containing thamin are indi-
cated in dark grey.
more agriculture and less mixed deciduous forest then
forest stands that did not have thamin (discriminant
function analysis, F = 8.04, d.f. = 3, 63, P < 0.0001,
r2 = 0.15).
The landcover map was resampled to contain only
stands of indaing or dry forest greater than 25 km2 in
size (Fig. 5). We found 9166 km2 of dry forest distrib-
uted within 254 stands and 9532 km2 of indaing forest
distributed in 120 stands. A stepwise discriminate func-
tion analysis (backwards) revealed that stands containing
thamin had a larger core area than stands that did not
contain thamin (Table 4: F = 4.0, d.f. = 1,62, r2 = 0.06,
P = 0.05). If dry and indaing forest landscape variables
were analyzed separately, there were no significant land-
scape predictors for dry forest stands, but the size of the
core area was highly significant for stands of indaing
forest (Table 4: F = 15.8, d.f. = 1,21, r2 = 0.42, P =
0.0007).
The landscape discriminant functions that differenti-
ated between thamin and non-thamin stands were used
to predict which of the unsurveyed stands fit the crite-
ria established to contain thamin (Table 4). For the func-
tion derived from both dry and indaing forest (P = 0.05),
14 out of 213 unsurveyed stands have a core area
sufficient in size to potentially contain a thamin popu-
lation. For the function derived for indaing forest only
(P = 0.0007), none of the 97 unsurveyed stands pos-
sessed a core area of sufficient size to support a thamin
population.
DISCUSSION
We found thamin only in the north-east fringe of the
geographic range identified by Salter & Sayer (1986).
This fringe encompasses all the remaining large tracts
of dry and indaing forest within central Myanmar. The
persistence of peripheral populations supports the pat-
tern of range collapse observed for other species of large
vertebrates (Lomolino & Channell, 1995, 1998). They
found that species experiencing a range collapse persist
not at the centre of their former range, but along the
periphery in remnant populations. They attribute the
180 W. J. MCSHEA ET AL.
Table 4. Results of discriminant function analysis (DFA) for landscape attributes of forest patches derived from reclassification of Landsat
images into dry and indaing forest patches
Landscape variable? Forest patches
All Dry forest                              Indaing forest
Partial r2 F Partial r2 F Partial r2 F
Area 0.016 0.966 0.029 1.036 0.066 1.126
Fractile dimension 0.040 2.386 0.045 1.613 0.103 1.848
Shape index 0.023 1.355 0.038 1.348 0.059 1.004
Core area 0.047 2.791 0.045 1.622 0.238 5.002?
Nearest neighbour distance 0.007 0.403 0.048 1.723 0.067 1.856
Number of core areas 0.013 0.740 0.024 0.859 0.014 0.230
Surveyed patches
Number of patches 64 41 23
Number correctly classified using DFA 34 18 21
Unsurveyed patches
Number of patches 310 213 97
Number potentially suitable using DFA 14 25 0
Partial r2 and F values are given for initial loading of variables prior to stepwise DFA (backwards) Core area of indaing forest patches was the only significant landscape
variable according to the stepwise DFA (see results). Classification scores using DFA are given for each subset of survey data, as well as the number of potential thamin
patches using the DFA to predict suitability of unsurveyed patches.
? For explanation of landscape variables see Appendix.
? P < 0.05.
Fig. 5. Forest stands greater than 25 km2 and containing at
least 30% dry or indaing forest types based on a resampling
of the landcover map shown in Fig. 3. The boundaries of 
the four established protected areas within the study area are
outlined.
persistence of these populations to their isolation from
decimating forces and the general characteristics of iso-
lated populations, which may make them more resilient
to change. The pattern observed with thamin is probably
less due to these sites being isolated islands, but rather
to the sites constituting the last large tracts of forest on
the edge of an advancing tide of human cultivation.
The decimating forces (Lomolino & Channell, 1995)
at work in the decline of thamin are probably anthro-
pogenic disturbances. Agricultural subsidies and policies
that encourage deforestation are a primary concern
throughout south-east Asia (Braatz et al., 1992), and the
Myanmar government views increased rice production
as a major potential export. The central plain of
Myanmar is shaped like a bowl, with mountains to the
north, east and west providing natural barriers to the cul-
tivation of rice (Fig. 1). Indaing and dry forests persist
along the edges of this bowl (Fig. 3), but as advances
in human cultivation enable farmers to push into the sur-
rounding hills, the remaining forest is removed. Large
portions of Myanmar are still forested (> 55%: Achard
& Estreguil, 1995), but the forest type present in these
mountains does not appear to be suitable for sustaining
viable populations of thamin. Four protected areas are
situated within the range of thamin and all are forested,
but only Chatthin Wildlife Sanctuary contains signifi-
cant amounts of indaing forest, and only Shwessetaw
Wildlife Sanctuary contains dry forest.
The loss of forest habitat is compounded by the lack
of thamin in small forest patches. A large core area was
the only landscape measure that was significant in the
discriminant function describing deciduous forest patches
that contained thamin. The loss of deer from small
patches may be due to reduced productivity within these
deciduous forest patches. The prolonged dry season in
central Myanmar produces a ?nutrient winter? during the
hot?dry season (March?May), after the crops have been
harvested but before the monsoon rains begin. In savanna
ecosystems of West Africa, productivity is lowest during
the hot?dry season, and bovids focussed on grass pro-
ductivity following intermittent fires (Schuette et al.,
1998). Indaing forest is fire-adapted (Stott, 1986; Rundel
& Boonpragob, 1995) and man-made fires are common
during the hot?dry season to clear crops and to stimulate
the growth of grass for livestock.
Indaing forest also responds to fire with a vigorous
growth of grasses in the understory (Rundel &
Boonpragob, 1995). Folk wisdom in the villages asserts
that a strong fire, one capable of generating grass regen-
eration, is not possible in either dry or indaing forest that
has been degraded. A pattern of forest degradation that
progresses from select logging and fuel-wood removal,
to intensive cultivation/grazing and decreased fire inter-
val, has been described for Thailand (Blasco, 1983) and
applies to Myanmar. With fuel-wood consumption being
a major cause of deforestation in this region (Braatz et
al., 1992), the degraded habitat that results from wood
extraction and overgrazing along the periphery of each
forest stand may not generate sufficient fires to trigger
grass regeneration.
As with domestic livestock, thamin are primarily graz-
ers (Lekagul & McNeely, 1977) and may depend on the
pulse in grass productivity generated by fires in intact
indaing forests. Fire as a wildlife management tool has
been proposed for other open forest/savanna ecosystems
in Nepal (Dinerstein, 1979, 1987). Based on the popu-
lation estimates for the townships surveyed, no forests
in the study area were immune to the influence of farm-
ers. Larger forest stands may provide enough isolation
from the daily scouring for fuel-wood and fodder to gen-
erate grass-producing fires during the dry season. It may
not be necessary to restrict all access to these wildlife
sanctuaries, but use should be managed, particularly
along the sanctuary boundaries. Saberwal (1996) found
that moderate livestock grazing in Indian parks was com-
patible with conservation needs, and advocated working
with villages in developing wise use of limited sanctu-
ary resources.
Our landuse classification is based on a 1989 Landsat
image and how much forest has been lost since that time
is conjecture. Human population growth and associated
environmental demands have been mounting in
Myanmar, with a major focus on expanding rice pro-
duction to generate export capital. It is our impression
that the loss of forest since 1989 has concentrated on the
smaller patches embedded within the agricultural matrix.
Larger patches of forest still persist, but few of these
large patches exist within the present protected area
framework (Fig. 5).
Within the study area, there were few unsurveyed
stands of dry or indaing forest that could potentially con-
tain thamin. However, the existing reserves encompass
only small portions of the largest forest tracts. Our first
recommendation would be to expand existing reserves
in the northern townships to incorporate as much remain-
ing indaing forest as possible. This may be done by
enlarging the gazetted reserves of Chatthin Wildlife
Sanctuary and Alaungdaw Kathapa National Park, or by
establishing a new sanctuary between these two reserves
in the Maha-myaing region along the northern reaches
of the Chindwin River. The last remaining large dry for-
est stands exist around the Shwesettaw Sanctuary in the
southern part of the study area, although this forest type
did not possess the densities of thamin observed in
indaing forest. Our second recommendation would be to
clarify the role of fire in maintaining both indaing and
thamin populations. Our third recommendation is that
thamin conservation is not incompatible with farming
activities, but there needs to be a recognition that the
remaining dry and indaing forest serve a function within
these rural communities both for livestock and thamin.
Conservation methods that serve to prevent the degra-
dation of these habitats should be a priority.
Acknowledgements
This work was supported by restricted endowment
funds from the Smithsonian Institution, and by funds
from Friends of the National Zoo, the Wildlife
Conservation Society and the National Geographic
181Conservation of thamin
Society. The field work was supported by U U Ga and
U Than Nwai in the Myanmar Wildlife Department and
by the hard efforts of the Thamin Ecology Team. The
GIS work was assisted by Tom Small and Jeff Diez.
REFERENCES
Achard, F. & Estreguil, C. (1995). Forest classification of
Southeast Asia using NOAA AVHRR data. Rem. Sens. Environ.
24: 198?208.
Alvard, M. S., Robinson. J. G., Redford, K. H. & Kaplan, H.
(1997). The sustainability of subsistence hunting in the neotrop-
ics. Conserv. Biol. 11: 977?982.
Balmford, A. & Long, A. (1995). Across-country analyses of bio-
diversity congruence and current conservation effort in the trop-
ics. Conserv. Biol. 9: 1539?1547.
Blasco, F. (1983). The transition from open forest to savanna in
continental Southeast Asia. Tropical savannas: 167?182.
Bouliere, F. (Ed.). Amsterdam: Elsevier.
Bodmer, R. E., Eisenberg, J. F. & Redford, K. H. (1997). Hunting
and the likelihood of extinction of Amazonian mammals.
Conserv. Biol. 11: 460?466.
Braatz, S., Davis, G., Shen, S. & Rees, C. (1992). Conserving
biological diversity: a strategy for protected areas in the
Asia?Pacific region. World Bank Techn. Pap. 193: 1?66.
Buckland, S. T., Anderson, D. R., Burnham, K. P. & Laake, J. L.
(1993). Distance sampling: estimating abundance of biological
populations. New York: Chapman and Hall.
Dinerstein, E. (1979). An ecological survey of the Royal
Karnali?Bardia Wildlife Reserve, Nepal. Part 1: vegetation,
modifying factors and successional relationships. Biol. Conserv.
15: 127?150.
Dinerstein, E. (1987). Deer, plant phenology, and succession in
the lowland forests of Nepal. In Biology and management of the
Cervidae: 272?284. Wemmer, C. (Ed.). Washington, DC:
Smithsonian Institution Press.
ERDAS (1997). ERDAS field guide. 4th edn. Atlanta, GA:
ERDAS, Inc.
ESRI (1993). Digital charts of the world. Sunnydale, CA:
Environmental Systems Research Institute, Inc.
Fox, J., Yonzon, P. & Podger, N. (1996). Mapping conflicts
between biodiversity and human needs in Langtang National
Park, Nepal. Conserv. Biol. 10: 562?569.
Geist, V. (1994). Wildlife conservation as wealth. Nature, Lond.
368: 491?492.
Harper, F. (1945). Extinct and vanishing mammals of the Old
World. Special Publication 12. New York: American Committee
for International Wild Life Protection, New York Zoological
Park.
Heinen, J. T. & Srikosamatara, S. (1996). Status and protection
of Asian wild cattle and buffalo. Conserv. Biol. 10: 931?935.
Humphrey, S. R. & Bain, J. R. (1990). Endangered animals of
Thailand. Gainesville, FL: Sandhill Crane Press.
Keiter, R. B. & Locke, H. (1996). Law and large carnivore con-
servation in the Rocky Mountains of the U.S. and Canada.
Conserv. Biol. 10: 1003?1012.
Laurence, W. F. (1991). Ecological correlates of extinction prone-
ness in Australian tropical rain forest mammals. Conserv. Biol.
5: 79?89.
Lekagul, B. & McNeely, J. A. (1977). Mammals of Thailand.
Bangkok, Thailand: Kurusapha Ladprao Press.
Lillesand, T. M. & Kiefer, R. W. (1994). Remote sensing and
image interpretation. New York: John Wiley and Sons, Inc.
Lomolino, M. V. & Channell, R. (1995). Splendid isolation: pat-
terns of range collapse in endangered mammals. J. Mammal. 76:
335?347.
Lomolino, M. V. & Channell, R. (1998). Range collapse, re-intro-
ductions, and biogeographic guidelines for conservation.
Conserv. Biol. 12: 481?484.
McGarigal, K. & Marks, B. J. (1995). FRAGSTATS: spatial pat-
tern analysis program for quantifying landscape structure. U.S.
Department of Agriculture, General Technical Report PNW-
GTR-351.
Myint Aung (1994). Field notes on thamin in Myanmar. Yangon,
Myanmar: Wildlife and Sanctuaries Division, Forest
Department.
Newmark, W. D. (1996). Insularization of Tanzanian parks and
the local extinction of large mammals. Conserv. Biol. 10:
1549?1556.
Rabinowitz, A. (1995). Helping a species go extinct: the Sumatran
rhino in Borneo. Conserv. Biol. 9: 482?488.
Ranjitsinh, S. (1978). The Manipur brow-antlered deer (Cervus
eldi eldi) ? a case history. Proceedings of the IUCN Threatened
Deer Programme. Morges, Switzerland: International Union for
Conservation of Nature and Natural Resources.
Richards, P. W. (1979). The tropical rainforest. Cambridge:
Cambridge University Press.
Rundel, P. W. & Boonpragob, K. (1995). Dry forest ecosystems
of Thailand. In Seasonally dry tropical forests: 93?124. Bullock,
S. H., Mooney, H. A. & Medina, E. (Eds). Cambridge:
Cambridge University Press.
Saberwal, V. K. (1996). Pastoral politics: gaddi grazing, degra-
dation and biodiversity conservation in Himachal Pradesh, India.
Conserv. Biol. 10: 741?749.
Salter, R. E. & Sayer, J. A. (1986). The brow-antlered deer in
Burma; its distribution and status. Oryx 20: 241?245.
SAS (1987). SAS/STAT guide for personal computers, version 6
edition. Cary, NC: SAS Institute, Inc.
Schuette, J. R., Leslie, D. M. Jr, Lochmiller, R. L. & Jenks, J. A.
(1998). Diets of hartebeest and roan antelope in Burkina Faso:
support of the long-faced hypothesis. J. Mammal. 79: 426?436.
Song, Yan-Ling (1996). Population viability analysis for two iso-
lated populations of Haianon?s Eld?s deer. Conserv. Biol. 10:
1467?1472.
Stamp, L. D. (1925). The vegetation of Burma from an ecologi-
cal standpoint. Calcutta: Thacker Spink & Co.
Stott, P. (1986). The spatial pattern of dry season fires in the
savanna forests of Thailand. J. Biogeogr. 13: 345?58.
Tun Yin (1967). Wild animals of Burma. Rangoon, Burma:
Rangoon Gazette Ltd.
UNEP (1995). Land cover assessment and monitoring; Myanmar.
UNEP/EAP. TR/95-06. Bangkok, Thailand: UNEP Environment
Assessment Programme for Asia and the Pacific.
Wemmer, C. (1987). Burmese brow-antlered deer (Cervus eldi
thamin). In North American regional studbook: 1?6. Bethesda,
MD: American Association of Zoological Parks and Aquariums.
Whitehead, G. K. (1972). Deer of the world. New York: Viking
Press.
182 W. J. MCSHEA ET AL.
APPENDIX
Landscape metrics for analyses of differences in patch and landscape characteristics between forests with and without thamin. 
Adapted from McGarigal & Marks (1995).
183Conservation of thamin
Landscape metric Description Formula?
Area Area of patch is dependent on the size of pixels and is calculated in hectares.
Fractal dimension Measurement of complexity of the perimeters of patches. Fractal dimensions in 
FRAGSTATS are treated as perimeter-area relations.
Core area Core area is the area that is more than 1 pixel away from the patch perimeter 
and is calculated in hectares.
Number of core areas If a patch has no area that is more than 1 pixel away from the perimeter, this number 
equals zero. Otherwise the number of core areas for a patch equals the number of 
disjunct areas that are more than 1 pixel away from the perimeter.
Nearest neighbour distance Distance to the nearest patch of the same type. Measured as shortest possible 
distance between edges in metres.
Shape index Compares the shape of a patch to a circle and increases in value as patches 
become more irregular in shape.
? Equation symbols: pij, perimeter of patch ij, measured in metres; aij, area of patch ij, measured in hectares.
FRACT
p
a
y
ij
=
( )2 0 25In
In
.
SHAPE
p
a
y
ij
=
0 25.