ORI GIN AL PA PER
Evaluating the protection of wildlife in parks:
the case of African buffalo in Serengeti
K. L. Metzger • A. R. E. Sinclair • Ray Hilborn • J. Grant C. Hopcraft •
Simon A. R. Mduma
Received: 16 July 2009 / Accepted: 21 July 2010 / Published online: 3 August 2010
� The Author(s) 2010. This article is published with open access at Springerlink.com
Abstract Human population growth rates on the borders of protected areas in Africa are
nearly double the average rural growth, suggesting that protected areas attract human
settlement. Increasing human populations could be a threat to biodiversity through
increases in illegal hunting. In the Serengeti ecosystem, Tanzania, there have been marked
declines in black rhino (Diceros bicornis), elephant (Loxodonta africana) and African
buffalo (Syncerus caffer) inside the protected area during a period when there was a
reduction of protection through anti-poaching effort (1976–1996). Subsequently, protec-
tion effort has increased and has remained stable. During both periods there were major
differences in population decline and recovery in different areas. The purpose of this paper
is to analyse the possible causes of the spatial differences. We used a spatially structured
population model to analyze the impacts of three factors—(i) hunting, (ii) food shortage
and (iii) natural predation. Population changes were best explained by illegal hunting but
model fit improved with the addition of predation mortality and the effect of food supply in
areas where hunting was least. We used a GIS analysis to determine variation in human
settlement rates and related those rates to intrinsic population changes in buffalo. Buffalo
populations in close proximity to areas with higher rates of human settlement had low or
K. L. Metzger (&) � A. R. E. Sinclair
Centre for Biodiversity Research, University of British Columbia, 6270 University Blvd,
Vancouver, British Columbia V6T 1Z4, Canada
e-mail: metzger@zoology.ubc.ca
R. Hilborn
School of Aquatic and Fishery Sciences, University of Washington, Box 355020, Seattle,
WA 98195-5020, USA
J. G. C. Hopcraft
Frankfurt Zoological Society, Serengeti National Park, Box 14935, Serengeti, Tanzania
J. G. C. Hopcraft
Arusha, Tanzania & Community and Conservation Ecology, University of Groningen, PO Box 14,
9750AA Haren, The Netherlands
S. A. R. Mduma
Serengeti Biodiversity Programme, Tanzania Wildlife Research Institute, Box 661, Arusha, Tanzania
123
Biodivers Conserv (2010) 19:3431–3444
DOI 10.1007/s10531-010-9904-z
negative rates of increase and were slowest to recover or failed to recover at all. The
increase in human populations along the western boundary of the Serengeti ecosystem has
led to negative consequences for wildlife populations, pointing to the need for enforcement
of wildlife laws to mitigate these effects.
Keywords African buffalo � Animal monitoring data � Enforcement � Illegal hunting �
Instantaneous rate of population change � Serengeti National Park
Introduction
Conservation has traditionally responded to potential loses of biota by establishing legally
protected areas (Pressey 1994; Rodrigues et al. 2004). Wittemyer et al. (2008) have shown
that average human population growth rates on the borders of protected areas in Africa and
Latin America were nearly double the average rural growth, suggesting that protected areas
attracted human settlement. People perceive or obtain benefit from their proximity to such
areas (de Sherbinin and Freudenberger 1998; Scholte 2003) but, there could be a con-
comitant threat to biodiversity within them. Many species are continuing to decrease
within protected areas (Brashares et al. 2001; Newmark 2008) often due to the illegal
wildlife harvesting for meat and trophies (Milner-Gulland et al. 2003). This is particularly
true for African nature reserves where local species extinctions are directly linked to
human population proximity, high reserve perimeter to area ratios, and bushmeat hunting
(Brashares et al. 2001; Ogutu et al. 2009).
In the Serengeti ecosystem, Tanzania, there have been marked declines in black rhino
(Diceros bicornis), elephant (Loxodonta africana) and African buffalo (Syncerus caffer)
inside the protected area (Dublin et al. 1990b; Metzger et al. 2007; Sinclair et al. 2007).
Declines in the numbers of large herbivores were attributed to cessation of anti-poaching
activities during a period of economic decline. Analysis of the trends in the buffalo
population over the whole area has suggested that population change was primarily due to
illegal hunting, and that enforcement of wildlife laws reduced the illegal offtake (Hilborn
et al. 2006) a conclusion also reached for other areas (Hilborn et al. 2006; Jachmann and
Billiouw 1997; Keane et al. 2008; Leader-Williams and Milner-Gulland 1993). Using
50 years of buffalo census data, Hilborn et al. (2006) established that illegal hunting and
enforcement activities could account for the overall trends in buffalo population yet
examination of the buffalo total counts indicated variation in the buffalo population
recovery; some areas have almost completely recovered from the population low of 1994
and other areas have failed to recover. Therefore, the main purpose of this paper is to
analyse the possible causes of these spatial differences.
Buffalo are known to be targeted by illegal hunters (Sinclair 1977). Park rangers who
actively search for snares and signs of illegal hunting have identified buffalo carcasses in
the field (Hilborn personal observation) and buffalo meat appears in villagers bushmeat
diets (Ndibalema and Songorwa 2007). Illegal hunting remains a large threat to conser-
vation efforts in the Serengeti (Holmern et al. 2007; Kaltenborn et al. 2005; Loibooki et al.
2002) and, therefore, we determined whether illegal hunting was a contributing factor to
the spatial differences in buffalo recovery.
Many factors can contribute to variation in animal population change including disease,
food supply, drought, and natural predation. Disease has been well monitored in the buffalo
and we have no indication that the observed changes in population have been caused by
epizootics. Rinderpest virus originally caused major declines in buffalo numbers after 1890
3432 Biodivers Conserv (2010) 19:3431–3444
123
but the virus has not caused declines since the 1960s (Dobson 1995; Dublin et al. 1990a;
Rossiter et al. 1983; Sinclair et al. 2008), and indeed it is now globally extinct (Normille
2008). Bovine tuberculosis (Myobacterium bovis), although prevalent in South Africa
(Cross et al. 2009), has not been found in Serengeti buffalo (Cleaveland et al. 2008;
Sinclair 1977). Drought can be a major controlling factor and drought induced mortality
occurred in 1993 causing approximately 40% mortality in the buffalo population. This
mortality was equally distributed across the ecosystem and therefore cannot be responsible
for the spatial patterns in recovery (Dublin et al.1990a; A. Sinclair unpublished data).
While it is possible that other factors may contribute to the spatial variation of buffalo
recovery, the major controlling factors are likely to be food supply, natural predation and
illegal hunting. We analyzed the impacts of these three factors—hunting, food supply and
natural predation—using a spatial analysis to separate out their effects. Thus, human
population density and rate of increase, which we show are related to hunting within the
reserve (Campbell and Hofer 1995; Hofer et al. 2000), are greatest in the west and
northwest. In contrast, food limitation, which is a function of rainfall (Sinclair 1977), is
most severe in the east and south, while predation is evenly spread over the buffalo range.
The greatest food supply is in the north where rainfall is highest (Fig. 1). The next highest
food levels are in the west, while the lowest food supplies are in the east (Sinclair 1977).
During the 1960s, prior to the population collapse, these northern areas supported the
highest densities of buffalo recorded in Africa, and in general Serengeti buffalo are limited
by food and not by predation (Sinclair 1977).
Materials and methods
Study area
The Serengeti-Mara ecosystem is located east of Lake Victoria and northwest of the
Ngorongoro highlands and the Rift Valley (Fig. 1) and is described elsewhere (Sinclair and
Arcese 1995b; Sinclair et al. 2007; Sinclair and Norton-Griffiths 1979). Serengeti National
Park is designated IUCN land category II and is managed for ecosystem protection and
recreation. A network of game reserves and conservation areas are located to the west and
east of Serengeti National Park (Fig. 1). This whole area is known as the Greater Serengeti
Ecosystem. The east of the national park boundary is settled by Maasai pastoralists who
rarely hunt for wild meat and their lifestyles tend to be consistent with conservation of
wildlife (Polansky et al. 2008). In contrast, human settlements to the west of the park
boundary do consume game meat regularly (Holmern et al. 2006; Loibooki et al. 2002;
Nyahongo et al. 2005).
Buffalo total counts
Beginning in the early 1960s, buffalo populations were censused by aerial survey every
few years. A detailed description of methods is given in Sinclair (1977). In 1970 all
observations of buffalo (individuals and herds) in the Greater Serengeti Ecosystem were
plotted on a map of the ecosystem. These observations were later incorporated into a GIS
using the Universal Transverse Mercator (UTM) coordinates. From the 1992, 1998, 2000,
2003, and 2008 censuses similar data were obtained using global positioning system (GPS)
technology. The buffalo population was close to its maximum in 1970 and this census was
Biodivers Conserv (2010) 19:3431–3444 3433
123
therefore used as the baseline with which we compared the following years. We deter-
mined the instantaneous rate of change in the buffalo population from 1970 to 2008 by
zone. Zones within the park (Fig. 1) represent distinct geographical and ecological areas.
Buffalo herds are relatively sedentary, confine themselves to a home range of less than
20 km in diameter, and so rarely cross over zone boundaries (Sinclair 1977). These zones
were the north, far east, far west, center, south and short grass plains. Because buffalo do
not use the short grass plains we did not include this area in our analysis. We summed
buffalo numbers within each zone for each year that we had census data and compared
these numbers with those in 1970 to show the relative change. A major drought in 1993
affected all zones and caused a 40% mortality (Sinclair et al. 2007, 2008).
Spatial population dynamics model
We used a spatially structured population dynamics model to determine the trends in
buffalo abundance in the five different regions between 1965 and 2008 (Hilborn et al.
2006). We examined a range of possible influences on abundance. These factors included
carrying capacity, which is a function of size of zone times rainfall (a surrogate for food
supply, Sinclair and Arcese 1995a), lion predation, and hunting effort. The population N^a;y,
the predicted number of buffalo in zone a year y is given by;
Fig. 1 The Serengeti ecosystem in Tanzania, East Africa includes the Serengeti National Park as a
protected area, and the game reserves and conservations areas. These are the Ikorongo Game Reserve,
Grumeti Game Reserve, Maswa Game Reserve, Ngorongoro and Loliondo Conservation Areas, which
surround Serengeti National Park and have restrictions on settlement within their borders. The Serengeti
National Park is divided up into zones (north, far west, centre, far east and south). Rainfall isohyets, showing
the highest rainfall in the northwest and the lowest in the southeast. Rainfall data collected at local rainfall
stations across the Serengeti ecosystem has been interpolated to produce the isohyets
3434 Biodivers Conserv (2010) 19:3431–3444
123
N^a;yþ1 ¼ N^a;y þ rN^a;y 1 � N^a;y
ka
� �� �
Sy � ua;yN^a;y � Ea;y
while
ua;y ¼ qvaPy
and
Ea;y ¼ N^a;y 1 � exp zLy
� ��
ð1Þ
where r is an intrinsic rate of growth assumed to be the same in all zones. Carrying
capacity for zone a is ka and Sy is survival from drought in year y, assumed to be 1.0 for all
years except 1993, the year of the drought. The exploitation rate from hunting in zone
a and year y is ua, Py is the relative hunting effort in year y, va is the relative hunting effort
for zone a, and q is a scalar relating hunting effort and area specific vulnerability to the
exploitation rate. Eay is the number of buffalo in zone a killed by lions in year y, Ly is an
index of the number of lions in buffalo habitat in year y, and z scales the lion abundance
index to lion mortality rate.
We explored a range of nested models, in various configurations that either included or
excluded hunting, lion predation, and rainfall. We estimated the parameters using census
data for each of five zones assuming a lognormal likelihood
L Na;yjparameters
� � ¼ 1
r
ffiffiffiffiffiffi
2p
p exp � ln Na;y � N^a;y
� �� 2
2r2
!
ð2Þ
where Nay is the observed number of buffalo in zone a, year y, and r the standard deviation
of the lognormal observation process.
The relative hunting effort (P) is poachers arrested per number of patrols day-1 (see
Hilborn et al. 2006. Figure 1b). The zone specific vulnerability parameters (va) were
estimated relative to that in the north which was fixed at 1.0. The parameter q is the harvest
rate per unit of hunting effort (P) in a zone with v = 1.
Food supply and rainfall
We also considered a range of hypotheses regarding carrying capacity. First, we assumed
all zones had the same carrying capacity. Secondly, we assumed that carrying capacity in
each zone (ka) was proportional to the size of the zone and the rainfall. Thus,
ka ¼ pAaRa ð3Þ
where Aa is the area in square km of zone a, Ra is the average dry season rainfall in zone a,
and p is a scalar to relate the product of area and rainfall to the carrying capacity.
While rainfall was the primary determinant of the food supply in most of Serengeti
(Fig. 1), the far east differed by lacking riverine grassland. In this zone rainfall was less
suitable as a predictor of resources (Sinclair 1977). Hence, thirdly we estimated the
carrying capacity for each zone independent of its size and rainfall.
Intrinsic rate of increase and lion predation
While we could, in theory, estimate the intrinsic rate of increase (r) from the spatial data
using the likelihood in Eq. 2 we found that the estimates obtained in that fashion were
much lower than the total population growth rate in the 1960s and 1970s. This is because
Biodivers Conserv (2010) 19:3431–3444 3435
123
the variability of the data by zone is much higher than the variability for the total popu-
lation. We estimated the intrinsic rate of increase (r = 0.092) from the total census
between 1965 and 1976. We fixed r in all scenarios except that when lion predation was
added we needed to increase the r value to account for additional lion mortality
r ¼ 0:092 þ 1 � exp zL1958ð Þ½ � ð4Þ
Each model scenario included different parameters estimated or fixed. In all cases we
estimated the number of buffalo in each zone in the first year, and the parameter r. Thus,
the simplest model has 6 parameters, plus a single carrying capacity (k) for a total of 7
parameters. As r is fixed in all models it is not considered an estimated parameter.
Fine scale buffalo population rate of increase (1970–1998 and 2000–2008)
The spatial trend in buffalo population was examined by comparing two time periods;
1970–1992 and 2000–2008 by creating a fine resolution map of buffalo population change
across the park. To do this we first constructed a buffalo density map. In the GIS we
divided the Serengeti National Park into 5 9 5 km areas and all observations within each
25 km2 area were summed. These numbers were then transformed to density (animal
km-2) within each 25 km2 area. In order to smooth across the 25 km2 cell boundary the
whole park was subdivided into 1 km2 units. The 30 nearest neighbor 1 km2 cells were
averaged for each 1 km2 cell using the neighborhood analysis tool in ArcGIS 9.2. This
allowed us to reduce the heterogeneity created from the clumping effect of large herds in
some grid cells adjacent to empty cells. The 30 km2 area was of a similar magnitude to the
maximum home range of buffalo (Sinclair 1977). We calculated the instantaneous rate of
population change per year (r) using the raster calculator tool in ArcGIS 9.2 spatial analyst.
Instantaneous rate of population change is defined as:
r ¼ ln Nt=N0ð Þ=t ð5Þ
where Nt is the population size at time t, N0 is the population size at the start of the time
period, and t is the number of years between the two. The r calculation was performed on
each cell in the density map for the two time periods 1970–1992 and 2000–2008.
Relation between buffalo numbers and human densities
We calculated the distance of each buffalo observation to the nearest edge of the park
where there was human settlement in 1970, 1992, 1998, 2000, 2003 and 2008. Using
Pearson’s correlation coefficient we determined the spatial correlation between buffalo
counts and distance to humans (see below).
Hunter population estimates
We used two years of human census data, 1978 and 2002 (Bureau of Statistics, Dar es
Salaam) for the area west of the Serengeti National Park boundary to Lake Victoria.
Census data were organized by local areas called wards (similar to US counties). The area
(km2) of each ward was known and we converted the ward population to density (humans
km-2). From the human density we calculated the hunter density. Hunter density is a
proportion of human density, which changes with the distance from the protected area
boundary. The equation is that given in Campbell and Hofer (1995). For each ward, we
determined the instantaneous rate of change of hunter density between 1978 and 2002.
3436 Biodivers Conserv (2010) 19:3431–3444
123
Results
Total population numbers of buffalo in the protected area
Figure 2 shows the changes in total numbers of buffalo in the Serengeti National Park
since 1965. At that time the population was recovering from the impacts of the viral
disease, rinderpest, and numbers subsequently increased to a peak of 74,237 in 1975
(Sinclair 1977). Shortly after that, in 1977, anti-hunting activities were severely restricted
by an economic crisis in Tanzania (Hilborn et al. 2006; Sinclair and Arcese 1995b) and
widespread hunting on this species (and others) followed (Dublin et al. 1990a). By 1992
anti-hunting efforts had returned but the population had been reduced to 36,119 animals,
some 49% of the peak number. The sharp decline to 21,186 buffalo in 1994 reflects the
effect of a severe drought amounting to an additional mortality of 42% of the remaining
population. Since 1998 the population has slowly increased. The most recent census (2008)
of buffalo recorded 28,524 individuals in Serengeti National Park. Because these are total
counts there are no sampling errors associated with the data. However, bias errors have
been calculated, accommodated by technique design and kept constant over the years
(Mduma and Hopcraft 2008; Sinclair 1972).
Buffalo population trends by region
Figure 3a, b presents the distribution of buffalo herds in 1970 before the main hunting
period, and in 2003 during the recovery phase. These show that northern and western parts
of the protected area have lost herds while the center and east have developed larger herds.
Figure 4 shows the proportional changes in buffalo population in each zone (see Fig. 1),
relative to 1970, the year when we have complete spatial distribution of animals prior to
the onset of hunting. By 1992 the north had lost 84% (±5% (95%C