|UFDC Home||myUFDC Home | Help|
This item has the following downloads:
1 E LEPHANT BROWSING IN MAJETE WILDLIFE RESE RVE, SOUTH WESTERN MALAWI By CAROLINE GHILAINE STAUB A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGRE E OF MASTER OF SCIENCE UNIVERSITY OF FLORIDA 2011
2 2011 Caroline Ghilaine Staub
3 To my parents and my friends in Majete Wildlife Reserve
4 ACKNOWLEDGMENTS I would like to thank the African Parks Network and the Department of National Park s and Wildlife of Malawi for sponsoring this research and allowing me to use Majete as my study site I am indebted to Anthony Hall Martin, Cornell Dudley, Julian Bayliss, Jes Gruner and Harvey Mtete for technical and logistical support, Hassam Patel for h is exp ertise in plant identification Michael Binford, Brian Child, Sabine Grunwald Cornell Dudley and Andrea Gaughan for their help with the manuscript, Forrest Stevens, Yasar Yesilcay and Tim Fik for assistance with the statistical analysis, Ghilaine St aub, Paul Garside, Jeanette Batiste Mtete, Sarah Hall and Hanneke Hogerheijde for moral support.
5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 6 LIST OF FIGURES ................................ ................................ ................................ .......... 7 ABSTRACT ................................ ................................ ................................ ..................... 8 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .... 10 2 MATERIALS AND METHODS ................................ ................................ ................ 16 Study Area ................................ ................................ ................................ .............. 16 Measurement of Parameter Variables ................................ ................................ .... 18 Statistical Procedure ................................ ................................ ............................... 19 Model Assessment and Interpretation ................................ ................................ .... 20 3 RES ULTS ................................ ................................ ................................ ............... 22 4 DISCUSSION ................................ ................................ ................................ ......... 25 5 MANAGEMENT APPLICATIONS ................................ ................................ ........... 28 6 CONCLUSION ................................ ................................ ................................ ........ 30 APPENDIX A CORRELATION MATRIX ................................ ................................ ....................... 31 B LOGISTIC REGRESSION ................................ ................................ ...................... 32 C CROSS VALIDATION ................................ ................................ ............................. 38 D REGRESSION DIAGNOSTICS ................................ ................................ .............. 39 LIST OF REFERENCES ................................ ................................ ............................... 40 BI OGRAPHICAL SKETCH ................................ ................................ ............................ 46
6 LIST OF TABLES Table page 2 1 Browsing measured in Majete according to vegetation type. ............................. 22 2 2 Numerical independent variables measured in Majete Wildlife Reserve. .......... 22 2 3 Results of the logistic regression. ................................ ................................ ...... 23 A 1 Pearson correlation matrix. ................................ ................................ ................ 31 B 1 Logistic regression: data summary. ................................ ................................ ... 32 B 2 Baseline model (intercept only) ................................ ................................ ......... 33 B 3 Baseline classification table ................................ ................................ ............... 33 B 4 Fo rward stepwise regression including independent variables .......................... 34 B 5 Stepwise regression including variables: model summary ................................ 35 B 6 Forward stepwise regression classification table ................................ ............... 35 B 7 Model fit statistics following the addition of interaction terms ............................ 36 B 8 Stepwise regression including interaction terms: classification table ................. 36 B 9 St epwise regression including interaction terms: variable selection .................. 37 C 1 Cross validation model summary ................................ ................................ ....... 38 C 2 Cross validation model classification table ................................ ........................ 38 C 3 Cross validation model variable selection ................................ .......................... 38 D 1 Regression diagnostics: model fit statistics ................................ ....................... 39 D 2 Regression diagnostics: classification table ................................ ...................... 39 D 3 Regression diagnostics: variable selection ................................ ........................ 39
7 LIST OF FIGUR ES Figure page 2 1 Majete Wildlife Reserv e and simplified topography major vegetation types, permanent water sources, and sampling plots. ................................ .................. 17 2 2 Probability of the presence of browsing as a function of distance from permanent wate r, and diameter at breast height ................................ ................ 24
8 Abstract o f T hesis Presented to the Graduate School of the University of Fl orida in Partial Fulfillment of the Requirements for the Degree of Master of Science ELEPHANT BROWSING IN MAJETE WILDLIFE RESE RVE, SOUTH WESTERN MALAWI By Caroline Ghilaine Staub May 2011 Chair: Michael William Binford Major: Geography The African elephant ( Loxodonta africana ) has a major influence on vegetation structure, composition and ecosystem processes, and is a primary agent of habitat change in Africa. Elephants at moderate to high population densities can damage vegetation, esp ecially when enclosed in protected areas. Knowledge of time and site specific factors affecting elephant browsing can be used to forecast future habitat transformations. This study examines the effects of elephant browsing on woody trees with a diameter a t breast height of 10 cm and above in Majete Wildlife Reserve (WR), south western Malawi, where a population of elephants was recently reintroduced after 14 years of absence. The research questions are that g iven a set of factors known to drive el ephant b rowsing in other areas, w hich ones have the most inf luence on browsing at Majete WR, and how are these factors related to browsing in the reserve? Twenty five sampling plots were set up within the three main vegetation types, individual cm were tagged, browsing was quantified and factors that were potentially associated with browsing were measured. Logistic regression was used to develop a model that identifies factors that influence browsing occurring on trees within the reserve Tw enty four per cent of tagged trees had been subject to elephant
9 browsing. Elephants mostly favored riparian woodlands, followed by Acacia dominated woodland and Brachystegia dominated woodland. Vegetation type, stem diameter and distance from permanent water co mbined, explained 80% of browsing occurrence. Browsing occurrence was negatively related to distance from permanent water as well as diameter at breast height. Managers should consider these factors to forecast future trajectories of habitat transformation and to manipulate the range of the elephants within the reserve.
10 CHAPTER 1 I NTRODUCTION H abit at loss and fragmentation are among the most important threats to biodiversity around the world ( Myers, 2000, Brooks et al ., 2002 ). The miombo woodlands of sou th central Africa are currently experiencing extensive deforestation and land degradation as a result of increased human populations (Walker & Desanker, 2004 ). Also affected by these changes is the African elephant ( Loxodonta africana (Blume nbach ), whose r ange in Southern African countries such as South Africa and Malawi is now often limited to protected area boundaries ( Guldemond & van Aarde, 2008 ). Elephants are among the most important agents of vegetation change in African woodlands, especially when t heir movements are confined (Laws, 1970, Guy, 1976, Cumming et al. 1997, Mapaure & Campbell, 2002, Guldemond & van Aarde, 2008 ). Elephant activity has a major influence on ecosystem structure and function, primarily on vegetation ( Laws, 1970, Barratt & Ha ll Martin, 1991, Huntly, 1991, Mapaure & Campbell, 2002, Conybear, 2004 ) While other herbivores also shape ecosystem structure (van Aarde, 1994 ), the absolute amount of vegetation consumed per elephant on a daily basis is greater than any other because of their large body size (Owen Smith, 1988). Elepha hey have been found to feed on 146 different plant species in Addo Elephant National Park, South Africa (Kerley et al., 2006 ). Elephants have also been recorded break ing branc hes, stripping bark, felling and uprooting trees without consuming them, a phenomenon that has been attributed to social behavior (Hendrichs, 1971).
11 The effect of elephants on canopy trees is of particular concern (Cumming et al ., 1997, Owen Smith, 1988, Conybeare, 2004). Elephan ts cause declines and deaths of large trees along with other animal and plant species that depend on the trees (Cumming et al ., 1997 ). Large trees also have important ecosystem functions (Treydte et al ., 1997). In semi arid systems where water and nutrients often limit vegetation productivity (Scholes, 1990 ) large trees have better hydraulic lift properties than grasses, thus, shrubs and small trees extract more water and minerals from soils (Ludwig et al ., 2003 ). Their extensive root system stabilizes the soil, preventing soil erosion. Leguminous trees such as Acacia species also fix nitrogen, increasing soil fertility (Owen Smith, 1988 ). Once felled, mature canopy trees take longer to regenerate than grasses and shrubs and their dominance structure is sometimes altered (Conybeare, 2004 ). As the canopy layer is thinned, fuel accumulates on the ground, leading to an increase in fire frequency and intensity (van Wyk & Fairall, 1969, Guy, 1989, Mapaure, 2001 ). This, in turn, may promo te the proliferation of brows ing resistant species that may not be favored as much by herbivores (Mapaure 2001 ). A recent rise in elephant densities, coupled with a compression in the size of their home range have made elephants responsible for causing increasing change in associated vegetation structure and composition, and the situation now requires management (Guldemond & van Aarde 2008, Guldemond & van Aarde2010 ), especially in South Africa and neighboring Botswana and Zimbabwe where populations are among the highest in Southern Africa. Long term population control measures such as translocation and contraception are currently expensive and ethically controversial (Slotow et al ., 2008, Ltter et al ., 2008, Grobler et al ., 2008, Bertschinger et al ., 2008)
12 In Southern Africa, the majority of the budget allocation, donor funding, private investment and other financing mechanisms are directed towards fighting extreme poverty and population density related issue s (UNDP UNEP, 2010) Under such circumstan ces, resources for conservation and habitat management are limited, and conservation measures that focus on assessing the risk of habitat conversion, cost of protection and/or biological value are becoming increasingly popular in conservation practice and research (Mills et al ., 1993, Pearce et al ., 1994, Stephens et al ., 2008 ). Probability based models are proving especially useful to identify variables that can be used to forecast future trajectories of habitat transformations as well as land use change ( Rouget et al ., 2003, Newburn et al ., 2005, Gude et al ., 2007 ) Identifying the factors associated with elephant browsing in conservation areas could be used to suggest adaptive habitat management options that focus on changing the intensity of landscape u se either by preventing access to certain types of vegetation through the use of en closures (Grant et al ., 2008 ) or by altering the distribution of key resources (e.g.water) both spatially and over time, to direct elephants away from areas that are under t oo much pressure (Grant et al ., 2008 ) A number of different site and time specific factors, including elevation, proximity to water and vegetation characteristics have been shown to drive the distribution of elephants, and thus their effects on vegetatio n (Anderson & Walker, 1974, Ben Shahar, 1993, Campbell et al ., 1995, Van de Koppel & Prins, 1998, Chafota, 2007 ) Elephants are selective feeders and prefer certain plant genera, including Acacia, Colophospermum Brachystegia and Adansonia (Child 1968, And erson & Walker, 1974, Guy, 1989 ) The pattern of species distribution, the vegetation type, as well as the
13 species richness and diversity in an area may therefore potentially have an important influence on elephant browsing characteristics in an area. Vege tation characteristics such as tree density and stem diameter have been reported to influence browsing. I n Ruaha National Park, browsing intensity on Faidherbia albida decreased with a corresponding decline in tree density (Barnes, 1982 ). A lthough elephant s have been observed to push over trees up of to 60 cm diameter (Chafota, 2007 ) they have the tendency to utilize trees that have a diameter of 30 cm or smaller (Laws, 1970 ). The proximity of certain vegetation types to permanent water has also been assoc iated with an increased susceptibility to elephant br owsing, especially in areas where palatable species are present (van Wyk & Fairall, 1969, Conybeare, 1991, Ben Shahar, 1993, Fullman, 2009 ). Riparian woodlands tend to be favored for these reasons (Child 1968, Anderson & Walker, 1974 ) Finally, elephants tend to avoid climbing and pref er areas where topography is flat presumably to conserve energy (Whyte et al ., 1996 ). In Majete Wildlife Reserve the study site used in this study, elephant browsing is expected to vary according to a number of factors including vegetation characteristics, tree stem diameter, and distance from roads as well as from per manent water. dominant tree genera such as Acacia Brachystegia and Combretum have been favored by elephants in different areas (Anderson & Walker, 1974, Barnes, 1982). However, b rowsing is expected to be highest in riparian woodland followed by Acacia dominated woodland and Brachystegia dominated woodland, mostly due to water accessibility and topo graphic differences. Both riparian and Acacia woodland grow at low altitude, riparian vegetation grows along rivers and streams, while Brachystegia woodland grows in the higher, wetter parts of the reserve. Whether elephant browsing varies according
14 to ve getation species richness and diversity in an area is unclear. However, species richness was expected to be positively associated with browsing, since a higher number of plant species was likely to result in the occurrence of a higher number of favored spe cies. Similarly an increase in species diversit y in an area featuring favored species was expected to be subject to more browsing than others. Elephants seek water daily therefore browsing tends to be higher in areas located close to water sources. Howeve r, vegetation utilization by elephants does not always decrease with distance from water (Fullman, 2009, Holdo, 2006). Also unclear is whether there is a significant relationship between browsing, distance from the road and tree density Areas where t ree d ensity is low were expected to be more favored than others, since these are presumably easier to navigate through. E lephants were often seen walking along and browsing vegetation along vehicle tracks, particularly during the rainy season, when the vegetati on is densest presumably f or the same re ason Browsing was therefore expected to be high er near vehicle tracks than further away. Finally, we assumed that increasing stem diameter would be nega tively associated with browsing. Holdo (2006) observed that i n H wang e National Park, Zimbabwe, stem diameter was the single strongest predictor of browsing presence Tall, l arge trees tend to be hard to break or push over and their canopy hard er to reach by elephants. Although all the above factors have been extens ively studied in different areas over the years, research focusing on the relative importance of key variables and their interactions with elephant browsing in different areas is relatively scarce (Holdo, 2006 ). The objective of this study is to quantify elephant browsing within a sample of medium to large trees in Majete Wildlife Reserve (hereafter referred to as Majete), in
15 southwestern Malawi, and examine the occurrence of browsing as a function of key environmental variables known to influence browsing in other areas. The Lower Shire valley at the southern end of the African Rift Valley and where Majete is located is poorly documented in terms of habitat change ( Timberlake et al ., 2000). Mammal populations in Majete and almost completely wiped out, with elephants disappearing between 1986 and 1992 (Sherry, 1995 ) bu t reintroduced beginning in 2003 As part of a joint effort to rehabilitate the area, the African Parks Network and the Department of National Parks of Ma lawi improved law enforcement, constructed an elephant proof fence around the reserve and brought animals back f rom surrounding parks. Two thousand five hundred and fifty of animals including elephants had been reintroduced by 2008. Majete offered an oppor tunity to analyze and monitor the impact of this new elephant population at the early stage of two years after reintroduction, with a view to identifying the drivers of elephant damage, and managing habitat transformation within the reserve by manipulating the range of the elephants, until an acceptable long term population control method becomes implemented.
16 CHAPTER 2 M ATERIALS AND METHODS Study Area Majete Wildlife Reserve is a 691 km protected area in the Lower Shire Valley, the southern tip of the African Great Rift Valley in the south west of Malawi (Figure 1). The Lower Shire Valley is characterized by a semi arid climate, where annual rainfall (~700 mm per annum ), is seasonal and mostly occurs between November and May and average temperature is moderate (~23 C). Altitude in Majete varies from 150 m on the Shire in the East, to 90 0 m in the West. The soils are L ithosols, shallow stony, ferruginous and of low fertility, with limited areas of more fertile alluvial soil occurring along some of the r ivers (Clarke, 1983). There are two permanent water sources within the reserve; the Shire and Mkurumadzi Rivers, located on the north eastern border. Five artificial waterholes were constructed between 2003 and 2009, and their locations along with the maj or vegetation types, permanent water sources, and sampling plots are shown in Figure 1 ( map adapted from Sherry, 1989). The western part of the reserve where a series of hills form the western escarpmen t of the African Rift Valley, is covered by tall, cl osed Miombo woodland dominated by Brachystegia and Julberna r dia spp. Altitude decreases toward th e east, where the vegetation changes into open Acacia dominated mixed woodland, and strips of riparian woodland/thicket association, dominated by Markhamia obt usifolia Cleistochlamys kirkii and Di chrostachys cinerea growing along the seasonal and permanent water sources. Detailed descriptions of the vegetation were given by Dowsett Lemaire & Dowsett, (2002). The average elephant density at the time of this stud y was 0.2 km (total: 144
17 elephants 82 within and 62 outside the sanctuary located in the north east of the reserve ). Figure 2 1. Location of Majete Wildlife Reserve and simplified topographic map, showing the location of major vegetation types, perma nent water sources, and sampling plots ( map adapted from Sherry, 1989).
18 Measurement of Parameter Variables Twenty five 0.1 hectare rectangular plots (20 m x 50 m) were established within the reserve, and measurements (adapted from Dudley, 1993) were taken between June 2008 and May 2009. Plots were distributed within the three major vegetation types and in areas that were accessible from the road. Plot locations were recorded using a hand held global positioning system (G PS ) cm) were marked with aluminum tags and identified to species. A sketch was produced to show the location of tagged trees within the plots to ensure the same trees could be measured over time. Presence or absence of elephant browsing was determined by inspe ction. Elephant browsing is distinguished from that caused by other herbivores by the appearance of torn branches (Nellemann et al ., 2002, Holdo, 2003). Major vegetation types were separated into Riparian woodland, Acacia dominated and Brachystegia domina ted dry woodland (Dowsett Lemaire & Dowsett, 2002). The d istance from roads and water sources was measured from the center of the plot to the nearest dirt track or the two rivers and the five artificial waterholes. Tree density was measured as the number o plot, and species diversity was calculated using the Shannon = p i ln p i ) where p i is the proportion of individuals of tree species i to the total number of individuals in the community (Wiener, 1949) Interactions between vegetation types and distance f rom water were considered to be potentially important factors of elephant browsing and were included in the model
19 Statistical Procedure Stepwise binary logistic regression (SPSS ver. 17.0) was conducted to identify factors that are related to elephant br owsing within the reserve. In this study, the logistic regression model evaluates the influence of a set of independent (predictor) variables on the probability of browsing occurring on trees in Majete. Logistic regression takes the following form: Logit (p) = b 0 + b 1 X 1 + b 2 X 2 + b 3 X 3 k X k Where p represents the probability of browsing presence, b 0 is the intercept, and b 1 b 2 k are the coefficients estimated for each independent variable, X 1, X 2 k independent variables for multicollinearity. High correlation between covariates may potentially inflate standard errors and bias the estimates of logistic regression coefficients, justifying the removal of one of the highly correlated variables from the m odel (Field, 2005). A threshold level of 0.8 was selected, above which one of the correlated variables was excluded. All predictors were continuous, with the exception of vegetation type which consisted of three ca tegories and was converted in to 0 1 dummy variables. Riparian woodland, being the category where most of the browsing was expected to come from, was assigned the non coded category, while Brachystegia woodland (BW) and Acacia woodland (AW) were coded dummies 1 and 2 respectively. A two stage stepw ise regression procedure was used to first select model main effects and then to select interactions of those main effects (Meents et al ., 1983). A variable was selected if it was statistically significant at p<0.05. A threshold for the final model was se lected to maximize true positives and minimize false negatives. Sensitivity (true positives) is the percentage of trees correctly classified by the model as being browsed while specificity is the percentage of
20 observations correctly predicted as not being browsed. From the conservation management perspective used in this study, which was to protect habitats from excessive than those that resulted from classifying a non (170) (170) observations. This was done to ensure that the differen ce between the number of browsed and non browsed observations would not bias the coefficient estimates and model predictions (Noon, 1986, Field, 2005). Model Assessment and Interpretation We used the model chi square statistic to determine whether there w as a significant improvement in the explanatory power of the model upon the addition of the independent variables, while the log likelihood statistic provided an estimate of the variance unaccounted for after the model had been fitted (Field, 2005). The N agelkerke R, a Pseudo Pseudo R 2 value, was used to assess how well the model fit the data (Field, 2005). A Nagelkerke R value larger than 0.2, which is equivalent to a regular R value of 0.5 suggests a satisfactory model fit R (Domencich & McFadden, 19 75, Serneels & Lambin, 2001, Peterson et al ., 2009). The presence of potential outliers or and leverage statistics. From the resulting model, estimated coefficients pr ovided by the estimated model were used to measure the association of independent variables with browsing (Field 2005). Negative exponential Beta values were inverted to produce values larger than one, which are easier to interpret. Sensitivity, specificit y and overall classification
21 accuracy were used to evaluate the predictive value of the model (Lindenmayer et al., 1992).
22 CHAPTER 3 R ESULTS Six hundred and ninety six plots. 53.5% of browsing was concentrated within Riparian woodland, followed by 39.4% in Acacia woodland and 7.06% in Brachystegia woodland (Table 2 1 ). Table 2 1. Browsing m easured in Majete according to v egetation type. Acacia woodland Riparian woodland Brachystegia w oodland Total No. of plots 13 6 6 25 Browsing Absent (0) 195 125 206 526 Present (1) 67 91 12 170 Total 262 216 218 696 The mean and standar d deviation of the numerical independent variables are shown in Table 2 2 All variables were measured for each plot except for diameter at breast height, which was measured on individual trees. All the independent variables displayed multicollinearity, bu t there were no significant correlations of 0.8 or higher at the 0.05 level of statistical significance (Appendix A). Table 2 2 Numerical independent variables measured in Majete Wildlife Reserve. Variable Mean S.D Distance from the road (m) 80.020 6.75 0 Distance from permanent water (km) 2.129 1.486 Tree density 31.110 1.106 Diameter at breast height (cm) 22.400 1.343 Species richness 10.740 3.567 Species diversity 0.7623 1.809 The model selected vegetation type, distance from water and DBH as significant browsing factors in the reserve (Table 2 3). The interaction between distance to water and both dummy variables were significant but their addition resulted in poorer model fit
23 and classification (Appendix B, Tables B6 B8) Interaction terms we re therefore left out of the final model. The classification probability threshold or cutoff point was manipulated to obtain the highest sensitivity and overall classification accuracy values, and was set to p=0.25. Regression diagnostics revealed that 1. 9% of observations in the data were potentially influential. These extreme cases were part of the same population as the other cases, so their removal could not be justified and they were left in the final mode l, which is displayed in Table 2 3 Table 2 3. Results of the logistic regression. Odds S.E. p < Exponential Beta 95% C.I.for EXP(B) Lower Upper Brachystegia woodland 2.313 .339 0.0001 .099 .051 .192 DBH_cm .026 .008 .001 .975 .959 .990 Acacia Woodland .921 .209 0.0001 .398 .264 .60 0 Dist_fm_water .439 .083 .001 .645 .548 .758 Constant 1.075 .274 .002 2.929 Log likelihood statistic = 640.078, Nagelkerke R = 0.26 Classification cutoff value = 0.25. Sensitivity = 80%, specificity = 63.9 % overall model accuracy = 67.8 % Brachys tegia woodland (BW) Acacia woodland (AW) DBH and distance from water are statistically significant independent variables at p < 0.05. The odds are negative for all independent variables, indicating that the likelihood of browsing decreased with a corresp onding increase in the selected independent variables. The odds ratio indicates that the odds of browsing increased by 1.5 times per unit in distance from water (1/.645) (Figure 2 2 A ). DBH had the smallest effect on the odds, with every unit increase resu lting in a drop by a factor of 1.02 (Figure 2 2 B ). BW and AW, being dummy variables, are interpreted in comparison with the reference category (Riparian woodland). At the time of the study, a tree was 10.10 times (1/0.099) more likely to be
24 browsed in Rip arian woodland than in Brachystegia Woodland. The odds of browsing were also higher for Riparian woodland trees than for those located in Acacia w oodland, but only by a factor of 2.5 (1/.398). Figure 2 2 Probability of the presence of browsing as a fu nction of distance from permanent water (A), and DBH (B ), with all other variables being held constant. A cross validation operation using randomly selected equal size subsets of ere similar to the ones produced using the full sample (Nagelkerke R = 0.278 ) and the same four vari ables were selected (Appendix C, Tables C1 C3 ), which confirmed the reliability of the model.
25 CHAPTER 4 D ISCUSSION Two years following their reintrod uction, elephant browsing was already high, and probably has an effect on vegetation structure and composition in Majete WR, as has been reported to occur in other areas (Kerley et al. 2008, Anderson & Walker, 1974, Barnes, 1985). The model helped identif y and characterize a number of factors that influence the spatial heterogeneity of browsing in the reserve. An association between browsing and vegetation type within the reserve was clear. Browsing was most concentrated in riparian woodland, where over 50 % of the browsing recorded in this study occurred, and least concentrated in Brachystegia woodland (7.06%). This was also shown by the logistic regression model, where the addition of vegetation type improved the model by reducing unexplained variance. Nu merous studies acknowledge that riparian woodlands, in addition to being located in the proximity of water, are favored for their combination of shade, palatable species and nutrient levels (Child, 1968, Conybeare, 2004, Anderson & Walker, 1974). Second to riparian woodlands, elephants show a preference for Acacia over Brachystegia dominated woodlands. Elephants have been attracted to both these vegetation types in different areas (Anderson & Walker, 1974, Barnes, 1982). This finding may therefore be attri buted to species composition, but it may also partly be attributed to topographic relief, which is known to be negatively associated with browsing (Jachmann & Bell, 1985, Whyte et al. 1996, Conybeare, 2004), and/or to the location of permanent water sourc es, most of which are located within low altitude Acacia dominated woodlands. There was no need for ele phants to travel uphill, to Brachystegia dominated woodland in search for water.
26 My findings also showed that browsing increases with proximity to water which was already known to be the case, since these animals are water dependent (van Wyk & Fairall, 1969, Conybeare 1991, Ben Shahar 1993). Howe ver, Fullman (2009) r eported that in Chobe National Park, Botswana, only one form of utilization (debarking) d ecreased with distance from Chobe river, while total utilization did not. Furthermore, Holdo (2006) showed that distance to water may not be statistically significant when sampled trees are too far from water. This suggests that in Majete, the sa mpling plo ts are located close enough to permanent water sources for wa ter to have an important effect on elephant browsing. DBH was the fourth and last variable to be selected as having a significant influence on browsing in Majete WR. The model suggests that it h as a smaller effect on browsing than the other variables, which is opposite to what occurred in H wang e National Park, Zimbabwe, where stem diameter was found to be the single strongest predictor of browsing presence (Holdo 2006). That result may be due to the difference in the DBH range considered in both studies. Hol do (2006) considered trees with a DBH of 5 cm and below, while we considered a DBH of 10 cm and above. A smaller DBH range may have made a different contribution to the final model. Our study a lso reports a negative relationship between browsing and stem diameter, which concurs with findings from Laws (1970), but not with Holdo (2006) who foun d that with trees with that elephants do favor a certain diameter range, and that trees may be rejected for being too small or too large.
27 The 2 log likelihood value produce d in the final model indicated the presence of unexplained variance in the model. This may be explained by the absence of other important variables and/or interactions, and also the presence of the outliers (Holdo 2006). Individual species such as Acacia t ortillis and Lonchocarpus bussei which appeared to be browsed more than others, should be included in the model to determine whether their relationship with browsing is significant. Elephants are selective feeders and prefer certain plant genera, includin g Acacia, Colophospermum Brachystegia and Adansonia (Child 1968, Anderson & Walker, 1974, Barnes 1982) The amount of elephant activity on the plots is also likely to be a significant factor. Another likely contributing factor is prior disturbance (e.g. fi re or disease), which has the potential to affect plant palatability, hence their attractiveness to elep hants (Bergstrom et al ., 2000; Du Toit et al ., 1990) In this study, variable interactions caused a decrease in model fit and classification accuracy, w hich may be due to the tradeoff that sometimes occurs between model complexity and the addition of potentially important variables (Bozdogan, 1987). Other interactions may have resulted in model improvement. For example, an interaction between plot locatio n and the past fire events may have result ed in variations in elephant browsing.
28 CHAPTER 5 M ANAGEMENT APPLICATIO NS Now that we know some of the factors that influence elephant browsing in Majete, it may be possible to alter the intensity at which eleph ants use the landscape. One way of the elephants that lived in Majete was reported to be larger than the area now enclosed within the protected area boundaries (She rry, 1989). As of 2009, up to 1 30 000 people lived on the perimeter of the reserve 1 therefore extending the elephants range outside the park boundary is not justified because of the risk of human elephant conflict However, the Northern part of Majete co ntains a sanctuary fence that is supposed to be taken down in the future As a result elephants will be able to disperse further during the wet season and the sanctuary area would be under less pressure from elephants. Managers should also consider alteri vegetation) in certain areas. Since e lephant reintroduction is relatively recent, parts of the reserve still experience low levels of browsin g relative to others E lephant proof vegetation en closures could be co nstructed to protect areas of higher biological and or economic value, and act as benchmark sites against which to compare the vegetation in elephant browsed areas, as was done in Addo Elephant National Park in the 1970 s ( Moolman & Cowling 1994 cited in Jo hnson et al ., 1999 ). The third option, which is currently being implemented both in Kruger National Park and Addo Elephant National Park in South Africa, involves altering the distribution of water both spatially and over time, to direct elephants away fro m areas that are under too much pressure (Grant et 1 Mzumara per. comm., 2009
29 al ., 2008). The Shire River will remain the main source of water for elephants with in Majete but the manipulation of artificial waterhole s may prove effective in directing elephants towards ce rtain areas and away from others (more vulnerable) ones especially during the dry season Management should ensure that both quantitative and qualitative measure s of elephant browsing are collected over time. Vegetation change should be monitored using a combination of spatial (GIS and remote sensing) and ground vegetation monitoring techniques to obtain results that are representative of browsing in the reserve as a whole, and to identify spatial patterns in landscape use by elephants.
30 CHAPTER 6 C ONCLUSION Two years after the re introduction of African elephants in Majete Wildlife reserve, certain areas already show much evidence of browsing on woody trees Among factors that influence the spatial heterogeneity of browsing within the reserve are vegetation type, distance from permanent water and stem diameter. Elephants show strong preference for riparian woodlands and Acacia dominated woodlands, while Brach ystegia dominated woodlands were relatively unaffected. Browsing increased with proximity to water and decre ased as a function of stem diameter. A larger number of trees, sampled at random and covering a larger portion of the reserve would provide more reliable estimates of browsing and related factors. A potential association between browsing, individual plant species prior disturbance and levels of elephant activity may improve our understanding of elephant browsing patterns in Majete. A short term management strategy that this study could be used to develop involves manipulating the intensity of landscape use by elephants, either by preventing access to certain types of vegetation through the use of enclosures or by altering the distribut ion of key resources both spatially and over time
31 APPENDIX A CORRELATION MATRIX Table A 1. Pearson correlation matrix. DBH ( cm ) Sp_rich Sp_div Dist_fm_water (k m) Tree density Dist_fm_rd (m) B W AW DBH ( cm ) 1 .003 .089 .045 .133 ** .048 .168 ** .084 .934 .020 .240 .000 .203 .000 .027 Sp_rich .003 1 .339 ** .194 ** .360 ** .009 .142 ** .424 ** .934 .000 .000 .000 .811 .000 .000 Sp_div .089 .339 ** 1 .305 ** .170 ** .121 ** .258 ** .208 ** .020 .000 .000 .000 .001 .000 .000 Dist_fm_water (km) .045 .194 ** .305 ** 1 .020 .072 .392 ** .273 ** .240 .000 .000 .605 .056 .000 .000 Tree density .133 ** .360 ** .170 ** .020 1 .279 ** .293 ** .575 ** .000 .000 .000 .605 .000 .000 .000 Dist_fm_rd (m) .048 .009 .121 ** .072 .279 ** 1 .426 ** .297 ** .203 .811 .001 .056 .000 .000 .000 BW .168 ** .142 ** .258 ** .392 ** .293 ** .426 ** 1 .525 ** .000 .000 .00 0 .000 .000 .000 .000 AW .084 .424 ** .208 ** .273 ** .575 ** .297 ** .525 ** 1 .027 .000 .000 .000 .000 .000 .000 ** Correlation is significant at the 0.05 level (2 tailed) *. Correl ation is significant at the 0.01 level (1 tailed)
32 APPENDIX B LOGISTIC REGRESSION Table B 1. Logistic r egression : data summary. Unweighted Cases a N Percent Selected Cases Included in Analysis 696 100.0 Missing Cases 0 .0 Total 696 100.0 Unselected Cases 0 .0 Total 696 100.0 a. If weight is in effect, see cla ssification table for the total number of cases.
33 Table B 2. Baseline model ( intercept only) Iteration 2 Log likelihood Coefficients Constant Step 0 1 775.342 1.023 2 773.859 1.127 3 773.858 1.130 4 773.858 1.130 a. Const ant is included in the model. b. Initial 2 Log Likelihood: 773.858 c. Estimation terminated at iteration number 4 because parameter estimates changed by less than .001. Table B 3 : Baseline classification table Observed Predicted Browsing Percentage Correct .00 1.00 Step 0 Browsing .00 526 0 100.0 1.00 170 0 .0 Overall Percentage 75.6 a. Constant is included in the model. b. The cut value is .500
34 Table B 4 Forward stepwise regression including independent variables B S.E. D f Sig. Exp(B) 95% C.I.for EXP(B) Lower Upper Step 1 a BW 2.137 .312 1 .000 .118 .064 .218 Constant .706 .097 1 .000 .494 Step 2 b Dist_fm_water (km) .391 .079 1 .000 .676 .579 .790 B W 1.735 .322 1 .000 .176 .094 .332 Constant .077 .154 1 .617 .926 Step 3 c Dist_fm_water (km) .441 .082 1 .000 .643 .548 .756 BW 2.182 .335 1 .000 .113 .058 .217 AW .919 .207 1 .000 .399 .266 .598 Constant .483 .202 1 .017 1.620 Step 4 d DBH ( cm ) .026 .008 1 .001 .975 .959 .990 Dist_ fm_water (km) .439 .083 1 .000 .645 .548 .758 B W 2.313 .339 1 .000 .099 .051 .192 AW .921 .209 1 .000 .398 .264 .600 Constant 1.075 .274 1 .000 2.929 a. Variable(s) entered on step 1: Brachystegia. b. Variable(s) entered on step 2: Dist_fm_wat er. c. Variable(s) entered on step 3: LowAltitude. d. Variable(s) entered on step 4: DBH_cm.
35 Table B 5 Stepwise regression including variables: model summary Step 2 Log likelihood Cox & Snell R Square Nagelkerke R Square 1 699.559 a .101 .151 2 672.166 b .136 .203 3 651.868 b .161 .240 4 640.078 b .175 .261 a. Estimation terminated at iteration number 5 because parameter estimates changed by less than .001. b. Estimation terminated at iteration number 6 because parameter estimates changed by les s than .001. Table B 6 Forward stepwise regression classification table Observed Predicted Browsing Percentage Correct .00 1.00 Step 1 Browsing .00 526 0 100.0 1.00 170 0 .0 Overall Percentage 75.6 Step 2 Browsing .00 526 0 100.0 1.0 0 170 0 .0 Overall Percentage 75.6 Step 3 Browsing .00 494 32 93.9 1.00 128 42 24.7 Overall Percentage 77.0 Step 4 Browsing .00 482 44 91.6 1.00 123 47 27.6 Overall Percentage 76.0 a. The cut value is .500
36 Table B 7 Model fit stat istics following the addition of interaction terms Step 2 Log likelihood Cox & Snell R Square Nagelkerke R Square 1 694.818 a .107 .160 2 659.870 a .151 .225 3 646.062 a .168 .250 4 641.792 a .173 .258 a. Estimation terminated at iteration number 6 becau se parameter estimates changed by less than .001. Table B 8 Stepwise logistic regression including interaction terms: classification table Observed Predicted Browsing Percentage Correct .00 1.00 Step 1 Browsing .00 526 0 100.0 1.00 170 0 0 Overall Percentage 75.6 Step 2 Browsing .00 526 0 100.0 1.00 170 0 .0 Overall Percentage 75.6 Step 3 Browsing .00 514 12 97.7 1.00 158 12 7.1 Overall Percentage 75.6 Step 4 Browsing .00 505 21 96.0 1.00 149 21 12.4 Overall Perce ntage 75.6 a. The cut value is .500
37 Table B 9 Stepwise logistic regression including interaction terms: variable selection B S.E. df Sig. Exp(B) 95% C.I. for EXP(B) Lower Upper Step 1 a BW by Dist_fm_water (km) .836 .144 1 .000 .433 .327 .575 Constant .728 .096 1 .000 .483 Step 2 b BW by Dist_fm_water (km) 1.014 .157 1 .000 .363 .267 .494 Dist_fm_water (km) by AW .597 .120 1 .000 .550 .435 .696 Constant .322 .117 1 .006 .725 Step 3 c DBH (cm) .027 .008 1 .001 .973 .958 .988 BW by Dist_fm_water (km) 1.083 .163 1 .000 .339 .246 .466 Dist_fm_water (km) by AW .620 .121 1 .000 .538 .425 .681 Constant .318 .215 1 .138 1.375 Step 4 d DBH (cm) .026 .008 1 .001 .974 .959 .989 BW by Dist_fm_water (km) 1.020 .172 1 000 .361 .257 .506 Dist_fm_water (km) .176 .086 1 .041 .839 .709 .993 Dist_fm_water (km) by AW .575 .129 1 .000 .563 .437 .725 Constant .535 .241 1 .026 1.708
38 APPENDIX C CROSS VALIDATION Table C 1. Cross validation model summary Step 2 Log l ikelihood Cox & Snell R Square Nagelkerke R Square 1 391.973 a .208 .278 a. Estimation terminated at iteration number 4 because parameter estimates changed by less than .001. Table C 2. Cross validation model classification table Observed Predicted Selected Cases a Unselected Cases b Browsing Browsing .00 1.00 Percentage Correct .00 1.00 Percentage Correct Step 1 Browsing .00 105 65 61.8 233 123 65.4 1.00 36 134 78.8 0 0 Overall Percentage 70.3 65.4 a. Selected cases Filter_0 EQ 1 b. Unselected cases Filter_0 NE 1 c. The cut value is .500 Table C 3. Cross validation model variable selection B S.E. Wald df Sig. Exp(B) 95% C.I.for EXP(B) Lower Upper Step 1 a DBH_cm .027 .009 8.331 1 .004 .973 .955 .991 Dist_fm_wate r .428 .097 19.404 1 .000 .652 .539 .789 Brachystegia 2.006 .382 27.576 1 .000 .134 .064 .284 LowAltitude .632 .271 5.451 1 .020 .532 .313 .904 Constant 2.001 .349 32.792 1 .000 7.394 a. Variable(s) entered on step 1: DBH_cm, Dist_fm_water, Brac hystegia, LowAltitude.
39 APPENDIX D REGRESSION DIAGNOSTICS Table D 1. Regression diagnostics: model fit statistics Step 2 Log likelihood Cox & Snell R Square Nagelkerke R Square 1 620.207 a .189 .282 a. Estimation terminated at iteration number 6 becau se parameter estimates changed by less than .001. Table D 2. Regression diagnostics: classification table Observed Predicted Browsing Percentage Correct .00 1.00 Step 1 Browsing .00 336 189 64.0 1.00 31 136 81.4 Overall Percentage 68.2 a The cut value is .250 Table D 3. Regression diagnostics: variable selection B S.E. Df Sig. Exp(B) 95% C.I.for EXP(B) Lower Upper Step 1 a DBH (cm) .027 .008 1 .001 .974 .958 .990 Dist_fm_water (km) .446 .084 1 .000 .640 .542 .755 BW 2. 616 .378 1 .000 .073 .035 .153 AW .931 .210 1 .000 .394 .261 .595 Constant 1.117 .278 1 .000 3.056
40 LIST OF REFERENCES A NDERSON G.D. & W ALKER, B.H. (1974) Vegetation composition and elephant damage in the Sengwa Wildlife Research Area, Zimbabwe. J. S. Afr. Wildl. Manage. Assoc. 4 1 14. B ARRAT D.G. & H ALL M ARTIN A.J. (1991) The effects of indigenous herbivores on Valley Bushveld in the Addo Elephant National Park. In: Proceedings of the First Valley Bushveld/Subtropical Thicket Symposium (eds P. J.K. Z ACHARIAS & G.C. S TUART H ILL ). Grassland Societ y of Southern Africa, Howick. B ARNES R.F.W. (1982). Elephant feeding behaviour in Ruaha National Park, Tanzania. Afr. J.Ecol. 20, 123 136. B ARNES R.F.W. (1985) Woodland changes in the Ruaha National Pa r k (Tanzania) between 1976 1982. Afr. J. Ecol. 23, 215 222.B EN SHAHAR R. (1993) Woodlands dynamics under the influence of elephants and fire in Northern Botswana. Vegetatio 123, 153 163. B ERTSCHINGER H., D ELSINK A., VAN A LTENA J.J., K IRKPATRICK J., K I LLIAN H., G ANSWINDT A., S LOTOW R. & C ASTLEY G. (2008) Reproductive control of elephants. In: Elephant Management: A Scientific Assessment of South Africa (eds R. J. S CHOLES and K. G. M ENNELL ). Witwatersrand University Press, Johannesburg. B ERGSTROM R. S KARPE C., & D ANNELL K. (2000). Plant responses to herbivory following simulated browsing and stem cutting of Combretum apiculatum. J. veg Sci. 11, 409 414. B OZDOGAN general theor y and its analytical extensions. Psychometrika 52, 345 370. B ROOKS T., M ITTERMEIER R.A., M ITTERMEIER C.G., D A F ONSECA G.A.B., R YLANDS A.B., K ONSTANT W.R., F LICK P., P ILGRIM J., O LDFIELD S., M AGIN G. & H ILTON T AYLOR C. (2002) Habitat loss and ext inction in the hotspots of Biodiversity. Conserv. Biol. 16, 909 923. C AMPBELL B.M., C UNLIFFE R.N. & G AMBIZA J. (1995) Vegetation structure and small scale pattern in miombo woodland, Marondera, Bothalia 25, 121 126. CHAFOTA, J. (2007) Factors governing selective impacts of elephant on woodland. Ph.D. Thesis University of the Witwatersrand, Johannesburg. C HILD G.F.T. (1968) An ecological survey of north eastern Botswana Unpublished report for the Government of Botswana. FAO, Rome. C LARKE J.E. (1983d) Protected Areas Master Plan for Southern Region Unpublished report for the Department of National Parks and Wildlife, Lilongwe.
41 C ONYBEAR A.M. (2004) Elephant impacts on vegetation and other biodiversity in the broadleaved woodlands of south central Afri ca. In: Biodiversity of the Four Corners Area : Technical Reviews, Volume Two (Chapters 5 15) (eds J. R. T IMBERLAKE & S. L. C HILDES ). Occasional Publications in Biodiversity No 15, Biodiversity Foundation for Africa, Bulawayo/Zambezi Society, Harare, Zimbab we. C UMMING D.H.M., F ENTON M.B., R AUNTENBACH I.L., T AYLOR R.D., C UMMING G.S., C UMMING M.S., D UNLOP J.M., F ORD A.G., H OVORKA M.D., J OHNSON D.S., K ALCOUNIS M., M AHLANGU Z. & P ORTFORS V.R. (1997) Elephants, woodlands and biodiversity in souther n Africa. S. Afr. J. Sci. 93, 231 236. DOMENCICH, T.A. & MCFADDEN, D.L. (1975) Urban Travel Demand: A Behavioral Analysis. North Holland Publishing Co. Amsterdam. D OWSETT L EMAIRE F. & D OWSETT R.J. (2002) Biodiversity surveys and the development of new r esearch and monitoring strategies for the lower Shire protected areas Technical report for the Department of National Parks and Wildlife of Malawi and USAID COMPASS Malawi. D U T OIT J.T., B RYANT J.P. & F RISBY K. (1990) Regrowth and palatability of Acac ia shoots following pruning by African savanna browsers. Ecology 71, 149 154. D UDLEY C.O. (1983) Liwonde National Park vegetation study Unpublished Technical Report for the Department of National Parks and Wildlife, Malawi. F IELD A. (2005) Discovering s tatistics using SPSS 2nd edn. Sage Publications Ltd, London. F ULLMAN T. (2009) Spatial dynamics of elephant impacts on trees in Chobe National Park, Botswana MSc Thesis. University of Florida, Gainesville, FL. G RANT C.C., B ENGIS R., B ALFOUR D., P EEL M, D AVIES M OSTERT W., K ILLIAN H., L ITTLE R., S MIT I., G ARAI M., H ENLEY M., A NTHONY B., H ARTLEY P., P RINSLOO M., B ESTER I., A DEDORF J., H AVEMANN P., H OWELL B., M ACFADYEN D. & P ARKER T. (2008) Controlling the distribution of elephants. In: El ephant Management: A Scientific Assessment of South Africa (eds R. J. S CHOLES and K. G. M ENNELL ). Witwatersrand University Press, Johannesburg. G ROBLER D., VAN A LTENA J.J., M ALAN J.H., M ACKEY R.L. (2008) Elephant translocation. In: Elephant Managemen t: A Scientific Assessment of South Africa (eds R. J. S CHOLES and K. G. M ENNELL ). Witwatersrand University Press, Johannesburg. G UDE P.H., H ANSEN A.J. & J ONES A.D. (2007) Biodiversity consequences of alternative future land scenarios in Greater Yellowst one. Ecol. Appl. 17, 1004 1018.
42 G ULDEMOND R. & VAN A ARDE R. (2008) A meta analysis of the impact of African elephants on savanna vegetation. J. Wildl.Manage. 72 892 899. G ULDEMOND R. & VAN A ARDE R. (2010) The influence of elephants and canopies on su b canopy vegetation in a savannah. Afr. J. Ecol. 48 180 189. G UY P.G. (1989) The influence of elephants and fire on a Brachystegia Julbernardia woodland in Zimbabwe. J. Trop. Ecol 5, 215 226. G UY P.R. (1976) The feeding behaviour of elephant ( Loxodonta africana ) in the Sengwa area, Rhodesia. S. Afr. J. Wildl. Res. 6, 55 63. H ENDRICHS H. (1971) Freilandbeobachtungen zum Sozialsystem des afrikanishen Elefanten, Loxodonta africana (Blumenbach, 1797). In: Dikdik und Elefanten (eds H. & U. H ENDRICHS R. Pipe r, Munich. H OLDO R.M. (2003) Woody plant damage by African elephants in relation to leaf nutrients in western Zimbabwe. J. Trop. Ecol 19, 189 196. HOLDO, R.M. (2006) Elephant herbivory, frost damage, and topkill in Kalahari sand woodland savanna trees. J Veg. Sci. 17, 509 518. H UNTLY N. (1991). Herbivores and the dynamics of communities and ecosystems. Annu. Rev. Ecol. Syst 22 447 503. J ACHMANN H. & B ELL R.H.V. (1985) Utilization by elephants of the Brachystegia woo dlands of the Kasungu National P ark Malawi. African Journal of Ecology 23, 245 258. J OHNSON C.F., C OWLING R.M. & P HILLIPSON P.B. (1999) The flora of the Addo National Park, South Africa: are threatened species vulnerable to elephant damage? Biodivers. Conserv. 8 1447 1456. K ERLEY G. I.H. & M. L ANDMAN (2006) The impacts of elephants on biodiversity in the Eastern Cape Subtropical Thickets. S. Afr. J. Sci. 102, 395 402. K ERLEY G.I.H., L ANDMAN M., O WEN SMITH N., B ALFOUR D., D E B OER W.F., G AYLARD A., L INDSAY K. & S LOTOW R. (2008) Effects of elephants on ecosystems and biodiversity (eds R. J. S CHOLES and K. G. M ENNELL ). In: Elephant Management: A Scientific Assessment of South Africa Witwatersrand University Press, Johannesburg. L AWS R.M.(1970) Elephants as agents of habitat and landscape change in East Africa. Oikos 21, 1 15. L INDENMAYER D., T ANTON M., N ORTON T. (1992) Identification of the forest habitats of possums and gliders in Central Victoria Panther Publishing and Printing, Camberra.
43 L OTTER H.P.P., H ENLEY M., F AKIR & S., P ICKOVER M. (2008) Ethical considerations in elephant management. In: Elephant Management: A Scientific Assessment of South Africa (eds R. J. S CHOLES and K. G. M ENNELL ). Witwatersrand University Press, Johannesburg. L UDWIG F., D AWSON T.E., DE K RO ON N., B ERENDSE F. & P RINS H.H.T. (2003) Hydraulic lift in Acacia tortillis trees on an East African savanna. Oecologia 134 293 300. M APAURE, I. 2001. Small sc ale variations in species composition of miombo woodland in Sengwa, Zimbabwe: the influence of edaphic factors, fire and elephant herbivory. Syst. Geogr. Pl. 71, 935 947. M APAURE I.N. & C AMPBELL B.M. (2002) Changes in miombo woodland cover in and around Sengwa Wildlife Research area, Zimbabwe in relation to elephants and fire. Afr. J. Ecol. 40 167 185. M EENTS J. K., R ICE J., A NDERSON B. W. & O HMART R. D. (1983). Nonlinear relationships between birds and vegetation. Ecology 64, 1022 1027. M OOLMAN H.J. & C OWLING R.M. (1994) The impact of elephant and goat grazing on the endemic flora of S outh African succulent thicket. Biol. Conserv. 68 53 61 M ILLS L.S., F REDERICKSON R.J. & M OORHEAD B.B. (1993) Characteristics of old growth forests associated with northern spotted owls in Olympic National Park. J. Wild. Manage. 57, 315 321. M YERS N. M ITTERMEIER R.A., M ITTERMEIER C.G., DA F ONSECA G.A.B., K ENT J. (2000) Biodiversity hotspots for conservation priorities. Nature 403, 853 858 NEWBURN, D.S., REED, S., BRECK, P & MERENLENDER, A. (2005) Economics and land use change in prioritizing pri vate land conservation. Conserv. Biol 19, 1411 1420. N ELLEMANN C., M OE S.R. & R UTINA L.P. (2002) Links between terrain characteristics and forage patterns of elephants ( Loxodonta africana ) in northern Botswana. J. Trop. Ecol 18, 835 844. N OON B.R. ( 1986) Summary: biometric approaches : the researcher's viewpoint. In: Wildlife 2000: modeling habitat relationships of terrestrial vertebrates (eds. J. V ERNER M. L. M ORRISON and C. J.R ALPH ). University of Wisconsin Press, Madison. O WEN SMITH N. (1988) Me gaherbivores: the influence of very large body size on ecology Cambridge university Press, Cambridge. P EARCE J. L., B URGMAN M. A., F RANKLIN D. C. (1994) Habitat selection by helmeted honeyeaters. Wildlife Res. 21 53 63.
44 P ETERSON L. K., B ERGEN K. M., B ROWN D. G., V ASHCHUK L. & B LAM Y. (2009), Forested land cover patterns and trends over changing forest management eras in the Siberian Baikal region. Forest Ecol. Manag.. 257, 911 922. R OUGET M.D., R ICHARDSON M., C OWLING R.M., L LOYD J.W. & L OMBARD A.T. (2003) Current patterns of habitat transformation and future threats to biodiversity in terrestrial ecosystems of the Cape Floristic Region, South Africa. Biol. Conserv. 112, 63 85. S CHOLES R.J. (1990) The influence of soil fertility on the ecolo gy of Southern African savannas. J. Biogeogr 17, 415 419. S ERNEEL S. & L AMBIN E.F. (2001) Proximate causes of land use change in Narok District, Kenya: a spatial statistical model, Agr. Ecosyst. Environ.. 85 65 81. S HERRY B.Y. (1989) Aspects of the ec ology of the elephant Loxodonta Africana (Blumenbach, 1797 ) in the Middle Shire Valley MSc Thesis. University of Malawi, Zomba, Malawi. S HERRY B.Y. (1995) The demise of the Elephants of the Middle Shire Valley, southern Malawi. Report to FFPS (UK), Londo n. S LOTOW R., W HYTE I., H OFMEYR M., K ERLEY G.H.I., C ONWAY T., S CHOLE S, R.J. (2008) Lethal management of elephants. In: Elephant Management: A Scientific Assessment of South Africa (eds R. J. S CHOLES and K.G. M ENNELL ). Witwatersrand University Press, J ohannesburg. T REYDTE A.C., H EITKONIG I.M.A., P RINS H.H.T. & L UDWIG F. (2007) Trees improve grass quality for hervivores in African savannas. Pers. Plant. Ecol. Syst. 8, 197 205. S TEPHENS S., W ALKER J.A., B LUNCK D.R., J AYARAMAN A., N AUGLE D.E., R INGELMAN J.K. & S MITH A.J. (2008) Predicting risk of habitat conversion in native temperate grasslands. Conserv. Biol. 5, 1320 1330. T IMBERLAKE J., C OTTERILL F., P ITMAN D. (2000) Biodiversity of the Zambezi Wetlands Consultancy Report for the IUCN, t he World Conservation Union, Regional Office for Southern Africa, Harare. U NDP U NEP Poverty Environment Initiative Annual Progress Report 2010, Malawi Country Fact Sheet. Available online: http://www.unpei.org/PDF/PEI annualprogress report2009.pdf W ALKER S.M., & D ESANKER P.V. (2004) The impact of land use on soil carbon in Miombo Woodlands of Malawi. Forest Ecol. Manag. 203, 345 360. VAN A ARDE R. (1994) Aspects of habitat disturb ance by Cape porcupines in a savanna ecosystem. S. Afr. J. of Zool. 29, 217 220.
45 V AN D E K OPPEL J. & P RINS H.T. (1998) The importance of herbivore interactions for the dynamics of African savanna woodlands: An hypothesis. J. Trop, Ecol. 71, 935 947. V AN W Y K P. & F AIRALL N. (1969) The influence of the African elephant on the vegetation of the Kruger National Park. Keodoe 12 57 89. WIENER, N. (1949) The interpolation, extrapolation, and smoothing of stationary time series Wiley, New York. W HYTE I.J., N EL P.J., S TEYN T.M. & W HYTE N.G. (1996) Baobabs and elephants in the Kruger National Park. Preliminary Report, Kruger National Park.
46 BIOGRAPHICAL SKETCH Caroline G. Staub received a Bachelor of Science degree in Environmental Biology from Curtin Univers ity of Technology, Perth, Western Australia in 2005 and a Master of Science degree in Geography from the University of Florida in the spring of 2011. Her work focuses on the ecological aspects of landscape change in Southern Africa. Prior to pursuing her Caroline worked as a wildlife monitoring officer for the African Parks Network (Majete) in Chikwaw a, Malawi from 2008 to 2009