The impact of cattle ranching on large-scale vegetation patterns in a coastal savanna in Tanzania


  • M. W. Tobler,

    1. Geobotanical Institute, Swiss Federal Institute of Technology, 8092 Zurich, Switzerland
    Search for more papers by this author
  • R. Cochard,

    1. Geobotanical Institute, Swiss Federal Institute of Technology, 8092 Zurich, Switzerland
    Search for more papers by this author
  • P. J. Edwards

    Corresponding author
    1. Geobotanical Institute, Swiss Federal Institute of Technology, 8092 Zurich, Switzerland
      P. J. Edwards, Geobotanical Institute ETH, Zurichbergstrasse 38, CH-8044 Zurich, Switzerland (fax +411 632 12 15; e-mail
    Search for more papers by this author

P. J. Edwards, Geobotanical Institute ETH, Zurichbergstrasse 38, CH-8044 Zurich, Switzerland (fax +411 632 12 15; e-mail


  • 1The success of large-scale cattle ranching in African savanna vegetation has often been limited by problems of bush encroachment and disease (in particular trypanosomiasis spread by tsetse flies). Mkwaja Ranch, occupying an area of 462 km2 on the coast of Tanzania, is a recent example of a large ranching enterprise that failed within the savanna environment. It was closed in 2000 after 48 years of operation. In this paper we describe the main vegetation types of the area (excluding closed forest vegetation) and relate their patterns of distribution to the former use of the ranch for cattle.
  • 2The study area comprised the former ranch and parts of the adjacent Saadani Game Reserve, which had not been grazed by cattle for many years and had never been used for large-scale ranching. Following field surveys, 15 distinct types of grassland and bush vegetation were defined and a vegetation map was created using a Landsat TM satellite image. A multispectral classification using the maximum likelihood algorithm gave good results and enabled all 15 vegetation types to be distinguished on the map.
  • 3Two main spatial trends were detected in the vegetation. One was a large-scale decrease in the cover of bushland from the most intensively used parts of the ranch through more extensively used areas to the game reserve; this trend was attributed to differences in management history as well as to climatic and topographic factors. A second trend was a radial vegetation pattern associated with the enclosures where cattle were herded at night. High amounts of three bushland types [dominated by (i) Acacia zanzibarica, (ii) Dichrostachys cinerea, Acacia nilotica or Acacia mellifera and (iii) Terminalia spinosa] occurred in a zone between 300 and 2500 m from the paddocks, with a peak in bush density at about 900 m (mean value for 18 paddocks). In contrast, bushland dominated by Hyphaene compressa was scarce close to the paddocks and became more abundant with distance. There was also a radial trend in the grassland communities: close to the paddocks there was short grass vegetation containing many ruderals and invasive weedy species, while the tall grassland types with species such as Hyperthelia dissoluta and Cymbopogon caesius occurred further away in the areas less affected by cattle.
  • 4Synthesis and applications. The intensive modern livestock ranching as practised on Mkwaja Ranch proved to be unsustainable both economically and ecologically. In the end, the biggest problem faced by the ranch managers was not controlling disease, as had originally been feared, but preventing the spread of bush on pasture land. The results of our study demonstrate just how severe the problem of bush encroachment was, especially in areas close to paddocks. An important lesson for management is that grazing patterns need to be taken into consideration when determining the sustainable stocking rate for an area. To reduce the risk of bush encroachment in grazing systems with focal points such as paddocks or watering points, stocking rates need to be lower than in systems with a more uniform grazing distribution.


Savanna vegetation is dominated by varying proportions of mainly grasses and woody plants (Walker 1987). The characteristics of natural savannas are determined by a strongly seasonal climate, limited availability of moisture and nutrients, fire and herbivory (Scholes & Walker 1993). Grazing by domestic livestock can have a dramatic impact on savanna ecosystems and is often responsible for extensive bush encroachment (Brown & Archer 1989; Behnke, Scoones & Kerven 1993; Perkins & Thomas 1993a; Hudak 1999). There appear to be various reasons for this effect. One is that livestock reduces the vigour of grasses while promoting seed germination, establishment and survival of woody species (Brown & Archer 1989; Van Staden, Kelly & Bell 1994; Brown & Carter 1998). Secondly, livestock may act as dispersal agents for seeds of woody species (Reid & Ellis 1995; Miller 1996; Brown & Carter 1998). Thirdly, intensive grazing may diminish the frequency and intensity of fires by reducing the grass biomass, which also favours the woody component of the vegetation (Scholes & Walker 1993; Roques, O’Connor & Watkinson 2001). Finally, domestic livestock tend to be predominantly grazers but they displace a range of native browsers, such as impala Aepyceros melampus (Lichtenstein, 1812) and kudu Tragelaphus strepsiceros Pallas, that control tree seedlings more effectively (Cole 1986; Du Toit & Cumming 1999).

During the past few decades, many African governments have supported the development of large livestock enterprises (Okigbo 1985; Du Toit & Cumming 1999; Hudak 1999; Dahlberg 2000). This is because modern ranching methods, which make use of fences, artificial water holes and a variety of measures to control pests and diseases, allow a much higher stocking rate. However, cattle tend to feed on a limited number of grass species that can soon be overexploited, particularly during severe drought years, while bush encroachment may reduce the area of available pasture (Skarpe 1991). Partly for these reasons, and also because of problems of disease, many intensive ranching enterprises in tropical Africa have failed after a few years (Okigbo 1985).

The changes that occur in savanna vegetation as a result of cattle ranching have been described in various field studies (Brown & Archer 1989; Skarpe 1990; Friedel 1997; Schultka & Cornelius 1997; Bossdorf, Schurr & Schumacher 2000). A common finding is that the impact of grazing declines with increasing distance from focal points such as night paddocks and permanent watering points (Perkins & Thomas 1993a; Young, Patridge & Macrae 1995; Turner 1998; Rietkerk et al. 2000). Among the few studies of bush encroachment at a landscape scale, Perkins & Thomas (1993b) described the spatial pattern of bush encroachment in relation to grazing gradients around waterholes in Botswana, and Jeltsch et al. (1997) developed a simulation model to describe this process in South African savannas. Both of these studies were conducted in semi-arid savannas and there are no comparable large-scale studies of bush encroachment in more humid savannas. The distinction is important because in more humid savannas the most important factors affecting the equilibrium between grasses and woody plants are often herbivores and fire, rather than water availability (Walker 1985; Frost & Robertson 1987; Hopkins 1992; Valone & Kelt 1999).

The study of vegetation patterns at a landscape scale requires appropriate methods for the large-scale capture and analysis of data. Satellite images from the Landsat Thematic Mapper (TM) have proved a valuable tool for mapping vegetation at such a scale (Goodchild 1994; Wyatt 2000). At 30 m, the resolution of the TM sensor is high enough to distinguish the small patches of woody vegetation that frequently occur in savannas, but coarse enough not to be influenced by individual trees. The reflectance of vegetation in the six spectral bands recorded by TM depends on the abundance of woody plants, canopy structure, grass biomass and on the presence of particular species (Fuller, Prince & Astle 1997). These are the same criteria that are used for classifying savanna vegetation in conventional field surveys (White 1983).

The aims of this study were to: (i) describe the vegetation types and their patterns of distribution on a large cattle ranch on the coast of Tanzania; (ii) show the distribution of bush around former paddocks; (iii) demonstrate the usefulness of remotely sensed data for investigating the influence of ranching on vegetation.

Study area

Mkwaja Ranch (5°43′S, 38°47′E), occupying an area of 462 km2 on the Tanzanian coast (Fig. 1), provides a recent example of a technology-based, big ranching enterprise that failed within the savanna environment. Amboni Estates Limited acquired the land in 1953 and began to stock it with East African zebu cattle and boran bulls Bos indicus L. (Ford & Blaser 1971). In the following years Mkwaja Ranch developed into one of the largest private ranches in Tanzania, supporting over 13 000 head of cattle during the 1970s. However, problems of disease control and bush encroachment meant that the ranch was never profitable. Aerial photographs from 1954 show that initially the savanna was much more open than it is today. During the 1970s and early 1980s, extensive use was made of brush cutting but the practice was finally given up as it became clear that it was not only expensive but also exacerbated the problem of encroachment. After 1980, the number of animals was gradually reduced and parts of the ranch were abandoned; it was finally closed in August 2000 following years of financial deficits.

Figure 1.

Map showing the location of the study area on the coast of Tanzania.

Mkwaja Ranch is divided into two parts, known as Mkwaja North (240 km2) and Mkwaja South (222 km2). The whole area was organized into 18 paddock systems (Fig. 2) and 40 dams were built to ensure an adequate water supply. Up to 1500 cows were herded into each paddock for the night. In the morning the cattle were divided into herds of 200–400 animals and led by a herdsman to pasture areas and later to a nearby dam. Initially, daily grazing routes were not fixed and tended to differ between seasons. In 1976, a rotational system was introduced to let the herds graze in different areas for a limited amount of time and so reduce the impact on any one area (Ford & Blaser 1971; Gates et al. 1983; Trail et al. 1985).

Figure 2.

The locations of the night paddocks (triangles) on Mkwaja Ranch. The Thiessen polygons define the land associated with each paddock system.

In addition to the livestock, Mkwaja Ranch supported mostly non-migratory populations of several species of herbivores, including warthog Phacochoerus africanus Gmelin, waterbuck Kobus ellipsiprymnus Orgilby, bush pig Potamochoeros larvatus F. Cuvier, reedbuck Redunca redunca Pallas, bushbuck Tragelaphus scriptus Pallas, buffalo Syncerus caffer Sparrman, duiker Silvicapra grimmia L., sable antelope Hippotragus niger Harris, hartebeest Alcelaphus buselaphus Pallas, giraffe Giraffa camelopardis L., hippopotamus Hippopotamus amphibius L., kudu and elephant Loxodonta africanus Blumenbach (Gates et al. 1983). Little information is available on population sizes of these native herbivores but Trail et al. (1985) estimate that the total biomass of wildlife was up to one-quarter of that of the livestock.

To the south of Mkwaja Ranch lies the Saadani Game Reserve (209 km2, 6°02′S, 38°45′E), which was gazetted in 1974. The same species of wildlife are present as in Mkwaja, although mostly in much larger numbers (Robinson 1999). There has been no cattle grazing in the Saadani area since it became a game reserve, and only small areas around the villages are used for collecting firewood. The most important factors affecting the present vegetation are wild herbivores and frequent fires. It therefore provides a useful reference site when investigating the influence of cattle on Mkwaja Ranch. In 1996, Mkwaja South was incorporated as an extension to the Saadani Game Reserve. Now that ranching at Mkwaja has ceased, it is planned to establish a new national park comprising the Saadani Game Reserve, the adjacent Zaraninge Forest Reserve and the whole of the former Mkwaja Ranch (Fig. 1).

While Saadani is mostly flat, Mkwaja has an undulating topography with several small hills of mesozoic limestone (Klötzli 1980). Alluvial floodplains with recent sedimentary deposits occur along the larger rivers and estuaries, and the zone immediately adjacent to the coast is composed of fairly recent marine sediments such as coral sand and clay (Milewski 1993). The relatively nutrient-poor soils consist of greyish fine sand or loamy sand in the flats and reddish loamy sand over clay on slopes and hilltops (Klötzli 1980; Milewski 1993).

The mean annual temperature recorded at the ranch complex in Mkwaja North (1973–98) is 25 °C, with an annual range of 5 °C and a daily range of 8 °C. The annual rainfall has varied between 500 mm and 1700 mm during the last 50 years, with a mean of around 900 mm. There is a short rainy season from October to December during which monthly averages exceed 100 mm. January and February are usually rather dry. Rains start again in March and continue until the beginning of June, followed by 4 drier months. Due to its location on the coast, Mkwaja Ranch has very few months with no rain. However, August and September are often dry enough to allow extensive bush fires, many of which are started deliberately by local people.


field data

The main goals of the field survey were to classify the different vegetation types and to collect training sites for a multispectral classification of the satellite image. The field survey was carried out between September and November 2000. The vegetation was first classified according to the major structural types as given by White (1983), i.e. forest, woodland, bushland, thicket, grassland and wooded grassland. A stratified sampling scheme was then established, with 70 sample plots representing all of the vegetation types except closed forest, which was not used by cattle. The percentage cover and height of the most abundant grass species were recorded in a 5 × 5-m subplot at each sampling site. Bush or tree density was measured using the point-centred quarter method (Mueller-Dombois & Ellenberg 1974). On the basis of this information, the classification of vegetation types was further subdivided. For woodlands and bushlands, the dominant woody species were the most important criterion. The grasslands and wooded grasslands were classified by the dominant grass species because these contribute most to the reflectance of these vegetation types. The range of vegetation types finally recognized was restricted by the requirement that each type must be distinguishable in a satellite image. The vegetation types were named by adding the dominant species or an adjective to the basic structural types as defined by White (1983) (e.g. Acacia zanzibarica bushland or deciduous woodland).

As a next step, training sites for a multispectral classification were selected in the field. For each vegetation type at least 15 sites in different parts of the study area were chosen. The location of all sites was determined using a handheld GPS receiver (Garmin 12XL, GARMIN International Inc., Kansas City, USA).

image data

No recent Landsat TM image was available in which the entire study area was cloud-free. The best image available was taken on 15 December 1994, when Mkwaja North was almost free of clouds but there was rather high cloud cover over Saadani and some areas of Mkwaja South. The fact that the image was taken towards the end of the short rain season, when the vegetation was already regenerating, made it suitable for vegetation mapping.

The image was registered and geo-referenced to Universal Transverse Mercator (UTM) coordinates using a topographic map as reference. After a preliminary inspection of the image, it was decided to keep six of the seven bands for further analyses (omitting band 6). As the field survey took place 6 years after the satellite image was recorded, we were concerned that possible changes in the vegetation would influence the classification. We therefore collected training data only in areas that, to judge from the present vegetation, were unlikely to have changed significantly during the last few years. As a further precaution, outliers were removed using a correspondence analysis of all the training sites with the spectral information for each of the six bands as variables.

The maximum-likelihood algorithm was used for classifying the image. Clouds, shadows and the ocean were classified first, expanded by two pixels and then used as a mask for the classification to avoid misclassification at the edge of a cloud or shadow. A total of 380 training points was used for the final classification of 15 vegetation types. Additional classes for water, sand and bare soil were defined from the satellite image. Due to the small-scale heterogeneity of the landscape, single-pixel training was chosen to avoid using mixed pixels. This method also reduces effects of spatial autocorrelation and can therefore improve the classification (Gong & Howarth 1990).

No independent accuracy assessment was performed because a sample size of 30–60 samples per vegetation type, as suggested by Richards (1986), was beyond the scope of this study. Nevertheless, some idea of the accuracy can be gained by comparing the classification result for the training sites with the vegetation recorded at these sites.

data analysis

A geographical information system (ArcView 3.2; ESRI, Redlands, CA) was used to investigate the spatial distribution of the vegetation. Unless otherwise stated, the whole of Mkwaja Ranch excluding areas of closed forest (and excluding unclassified pixels) was used for the analyses. Cover of a vegetation type is defined here as the percentage of the total classified area occupied by that vegetation type; it does not refer to the cover of individual plants within vegetation types.

As it was impossible to determine the exact grazing pattern, it was assumed that any particular point was grazed by cattle from the nearest paddock. The area of each paddock system was therefore defined using Thiessen polygons (Fig. 2). To investigate the abundance of different vegetation types in relation to the distance from a paddock, the vegetation within a polygon was analysed in annuli of 100 m width centred on the paddock. At first, the grazing area within these annuli increases with distance from a paddock (Fig. 3); however, due to the proximity of neighbouring paddocks (mostly 4–8 km) the mean available area decreases beyond 2000 m. On average, 80% of the available area lies within 3500 m of the nearest paddock. Areas more than 4000 m away from any paddock are only found in some areas in Mkwaja South and were not included in the analyses.

Figure 3.

Relationship between the available area and the distance from the nearest paddock calculated in annuli of 100-m width.

The ratio of bushland to grassland (B/G ratio) was calculated as a means of quantifying bush encroachment. In calculating this ratio, ‘grassland’ was taken to include both grasslands and thicket–grassland mosaics. Panicum wooded grasslands were excluded because they occur on wetter sites and are not susceptible to encroachment by the main bushland species. Polynomial regression was used to find relationships between the distance to the paddocks and the intensity of bush encroachment.

To investigate the influence of grazing intensity on the spread of bushland, the B/G ratio was related to the stocking rate and area of grazing land in each paddock system. For this purpose, grazing land included all bushland and grassland within the Thiessen polygon. The average stocking rates in hectares per livestock unit (ha lsu−1) between 1989 and 1994 were calculated for 11 paddock systems using information from the monthly reports provided by the ranch managers (Table 1). Where two paddocks were very close together it is likely that areas were grazed by cattle from both paddocks. For this reason paddocks 8, 11, 12 and 13 were not included in the analyses. Paddocks 1, 17 and 18 were excluded because of a high cloud cover.

Table 1.  Area, stocking rate and bushland/grassland (B/G) ratios within Thiessen polygons defining the various paddock systems at Mkwaja Ranch, Tanzania. The grazing area includes all grassland, thicket–grassland mosaic, wooded grassland and bushland vegetation types. Average number of cattle and stocking rate as ha per livestock unit (lsu) between 1989 and 1994 are also given. Grassland includes grasslands and thicket–grassland mosaics; wooded grassland was not included as these moist grasslands are less susceptible to bush encroachment
Paddock systemArea (ha)Grazing area (ha)Number of cowsStocking (ha lsu−1)Bushland (%)Grassland (%)B/G ratio
21476 8116341·30·30·211·43
52180 9014641·90·220·141·57


vegetation types

The vegetation of Mkwaja Ranch and Saadani was classified into 15 different vegetation types. Names for grasses follow Ibrahim & Kabuye (1988), names for trees and shrubs follow Beentje, Adamson & Bhanderi (1994). A map showing the distribution of different vegetation types is shown in Fig. 4.

Figure 4.

Vegetation map of Mkwaja North, Mkwaja South, Saadani Game Reserve and Zaraninge Forest Reserve in Tanzania based on interpretation of a Landsat TM satellite image.


Four different types of bushland were identified. Acacia zanzibarica was the main tree species in Acacia zanzibarica bushland, where it could form dense, almost mono-specific stands with up to 1500 trees ha−1. This bushland type was commonly found close to paddocks where the soil and grass layer had been severely disturbed by cattle grazing and brush cutting. In the Saadani Game Reserve dense stands of Acacia zanzibarica occurred mainly along the coast, but some recent bush development seemed to have occurred in heavily grazed areas within the reserve.

The DichrostachysAcacia bushland was defined mainly on structural grounds as a mixed bushland of various small (1·5–2·5 m in height) woody species. It commonly occurred in homogeneous patches of several hectares on well-drained soils near paddocks. Abundant species included Dichrostachys cinerea, Acacia nilotica and Acacia mellifera, all of which are known to be encroaching species (Klötzli 1980; Fritz, Garine-Wichatitsky & Letessier 1996; Skowno et al. 1999). Other commonly found tree species in this vegetation type were Catunaregam nilotica, Piliostigma thonningii, Harrisonia abyssinica, Balanites aegyptiaca, Commiphora africana, Terminaliaspinosa, Annona senegalensis and Acacia nigrescens. Tree densities were between 400 and 900 ha−1.

Terminalia spinosa, a common tree in many savanna vegetation types, was the dominant species in Terminalia bushland, which was frequent on well-drained soils in disturbed areas such as close to paddocks. In this bushland type Terminalia spinosa could form very dense, homogeneous stands of < 1000 trees ha−1. In the shade of older Terminalia trees young evergreen species were found, suggesting that Terminalia bushland represents a successional stage from open savanna to forest.

Hyphaene bushland had a high abundance of the palm Hyphaene compressa and several woody species also found in DichrostachysAcacia bushland. On Mkwaja Ranch it occurred chiefly in parts of the fly belt (a strip of land that was regularly cleared of woody plants to prevent the spread of tsetse flies) and in some areas where Hyphaene-dominated grassland was the main surrounding vegetation type. Hyphaene compressa is known to form dense stands in response to brush cutting (Klötzli 1980).


Tall grasses such as Hyperthelia dissoluta and Cymbopogon caesius dominated the grass layer of the two ‘long grassland’ vegetation types. Hyphaene-dominated grassland could be dominated by either of these grasses; the woody stratum in this vegetation type was mainly represented by Hyphaene compressa with a density of 20–80 individuals ha−1. Towards the south of the study area, Hyphaene-dominated grassland graded into Cymbopogon grassland. This grassland type was dominated by Cymbopogon caesius and palms were mostly replaced by scattered trees (< 30 ha−1), mainly Terminalia spinosa and Acacia zanzibarica. Fire seemed to be an important determinant of these vegetation types.

The thicket–grassland mosaic was a common vegetation type composed of small clumps of dense, semi-evergreen trees and bushes dispersed in an open grassland of short to medium height. There was a high diversity of woody plants including a mix of savanna and forest species. Tree and bush density was between 20 and 100 ha−1. The species composition of the grass layer varied. In Saadani, Cymbopogon caesius was the dominant species, but was replaced by Echinochloa haploclada on moister soils. Andropogon gayanus, which often occurred interspersed with Cymbopogon caesius, became dominant and replaced Cymbopogon caesius in the north. Although these two grasses are tall, they occurred intermixed with short grass species such as Heteropogon contortus, Panicum infestum, Digitaria milanjiana and Eragrostis superba. For this reason, the thicket–grassland mosaic vegetation types are summarized under ‘short grasslands’.

The grasses Echinochloa haploclada and Panicum maximum were typical representatives of ‘moist grassland’. Panicum maximum formed homogeneous and almost impenetrable stands in Panicum wooded grassland on floodplains and along riverine forests. The bush Cordia ovalis was the main component of the woody stratum, with a density between 30 and 100 trees ha−1. In drier areas Panicum maximum was gradually replaced by tall-growing Echinochloa haploclada (Echinochloa wooded grassland). This was also the dominant grass in the almost treeless Echinochloa grassland, a vegetation type occurring mainly on pure black cotton soils in the south. Other typical grasses in this type were Setaria incrassata, Sporobolus spp. and Bothriochloa insculpta.

Forests and woodlands

The evergreen or largely evergreen coastal forests were divided by Klötzli (1980) in two subclasses: hilltop forests and riverine forests. Forest species lists can be found in Burgess et al. (1992), Sheil (1992), Burgess, Dickinson & Payne (1993), Burgess, Clark & Rodgers (1998) and Mwasumbi, Burgess & Clarke (1994). Deciduous woodland or miombo was a common vegetation type in the western part of Mkwaja and outside the ranch area. It was an open woodland dominated by deciduous trees and a continuous grass layer. It occurred mainly on sandy soils. A listing of common species of this vegetation type can be found in Knapp (1973). Semi-evergreen woodland was characterized by a fairly homogeneous tree layer of semi-evergreen, medium-sized (3–4 m) species such as Piliostigma thonningii, Commiphora africana and Annona senegalensis. Pancium maximum, which is known to be shade tolerant (Boonman 1993; Scholes & Walker 1993), was the main grass in the undergrowth.

The overall classification accuracy for the training sites was 77% (kappa coefficient 0·75). Producer accuracy for the individual vegetation types ranged from 50% to 100%, and user accuracy from 57·7% to 100%. It was lowest for deciduous woodland, which was confused with semi-evergreen woodland, DichrostachysAcacia bushland, confused with other bushland types, and for Echinochloa thicket grassland mosaic, which was confused with Dichrostachys–Acacia bushland.

spatial distribution of vegetation types

Table 2 summarizes the differences in the distribution of the main structural types of vegetation. Grassland cover increased by a factor of four from Mkwaja North to Saadani. Thicket–grassland mosaic also increased, while bushland and woodland decreased towards the south. As a result the ratio of bushland to grassland (B/G ratio) decreased by a factor of three from Mkwaja to Saadani. The B/G ratio also varied considerably between different parts of the ranch, from 0·68 for paddock system 16 to 2·67 for paddock system 9 (Table 1).

Table 2.  Cover of the main structural vegetation types in Mkwaja North, Mkwaja South and the Saadani Game Reserve, Tanzania. The B/G ratio for each area is also shown (for this purpose grassland includes thicket–grassland mosaics)
Vegetation typeCover (% of classified)
Mkwaja NorthMkwaja SouthSaadani
Grassland 6·315·123·9
Thicket–grassland mosaic19·329·429·8
Wooded grassland12·1 9·210·7
Bushland29·728·42 1·7
Woodland20·3 9·5 3·4
Forest 9·7 6·3 6·8
Bare soil 2·6 1·9 3·6
B/G ratio 1·2 0·6 0·4

Bush vegetation was most strongly developed around paddocks (Fig. 5). Figures 6 and 7 depict the changes in mean cover of the main vegetation types in relation to distance from the paddocks. In the first few hundred metres the impact of cattle (mainly trampling) was so high that there was bare soil. Bush cover increased to reach a peak at 500–1000 m. For some paddocks (e.g. 3, 4 and 6) cover then gradually decreased, whereas for others (e.g. 7 and 16) it declined to a more or less constant minimum level. Polynomial regression using the data for all paddocks showed that grassland increased linearly with distance from the paddock (F = 242·78, d.f. = 39, P < 0·001, R2 = 0·865). The relationship between the B/G ratio and distance from the paddock after 900 m is best described by a quadratic equation (F = 2690, d.f. = 31, P < 0·001, R2= 0·995; Fig. 8). There was a significant negative correlation between the B/G ratio and the available grazing area within a paddock system (F = 15·80, d.f. = 9, P= 0·004, R2 = 0·662; Fig. 9). There was also a statistically significant (α = 0·05) correlation between stocking rate and the B/G ratio (F = 6·13, d.f. = 9, P = 0·038, R2 = 0·434).

Figure 5.

Bush encroachment on Mkwaja Ranch as indicated by the percentage of bushland. The map was created applying a 10 × 10 mean filter to the vegetation map (Fig. 4); bush pixels were assigned a value of 1 and all the other pixels 0. Darker areas have a higher bush cover. Triangles show the location of the paddocks; white areas are unclassified.

Figure 6.

(a) Changes in the mean cover of different grassland types and all bushland with distance from the paddocks. (b) Changes in the cover of various bushland types with distance from the paddocks.

Figure 7.

Changes in the cover of three bushland types and in all bushland with distance from individual paddocks on Mkwaja Ranch.

Figure 8.

Changes in the mean bushland/grassland ratio (y) with distance from the paddocks (x). The quadratic equation describing this trend is y= 1·42486E-7* × 2 − 0·001136*x + 2·6852 (R2 = 0·995, P < 0·001).

Figure 9.

Relationship between the area available for grazing and the bushland/grassland ratio for 10 paddocks on Mkwaja Ranch. The regression line is y=−0·0005x + 1·859 (R2 = 0·662; P= 0·004).

There were interesting differences in the spatial patterns of the different types of bushland (Figs 6b and 7). Dichrostachys–Acacia bushland had a marked peak at around 250 m from the paddock. In contrast, amounts of Acacia zanzibarica and Terminalia bushland at first increased rapidly with distance and peaked between 800 and 1100 m from the paddock. Dichrostachys–Acacia bushland and Terminalia bushland then decreased steadily, while Acacia zanzibarica bushland decreased up to 2000 m and then remained constant at around 10%. Hyphaene bushland exhibited a completely different pattern, being scarce close to the paddocks and becoming more abundant with distance (Fig. 6b); this pattern was similar to that for long grassland (Fig. 6a). Acacia zanzibarica bushland was widespread and was only missing from moister areas occupied by wooded grasslands or from areas where forests or woodlands predominated. In these areas, Terminalia bushland was more common. Terminalia bushland could also be found mixed in with forest and woodland vegetation. Dichrostachys–Acacia bushland showed a pattern similar to that of Acacia zanzibarica bushland but with a greater concentration around paddocks.


vegetation mapping

Previous studies using satellite imagery to investigate the ecological impact of ranching have characterized the condition of the vegetation by means of indices based on combinations of different spectral bands (Hanan et al. 1991; Pickup 1994). Such indices are sensitive to above-ground biomass and are therefore useful for monitoring the degradation of grasslands; they are less useful in structurally heterogeneous vegetation with a complex mix of grassland and bush. In this study we preferred to use the remote-sensed image to make a vegetation map. Although this is a more difficult and time-consuming procedure, such a map provides more useful information. In practice, several characteristics of savanna vegetation can make the classification of remotely sensed images difficult. For example, areas that have been recently burnt or intensively grazed may have a lower biomass than undisturbed areas and thus a different reflectance. As a result, savanna types such as long- and short-grass savannas may be confused. To minimize such problems one could use an image taken in the late rainy season when the vegetation has had time to recover from disturbances, or develop a map from a series of two or more images taken a few months apart (Grignetti et al. 1997). In our study area, however, these solutions were not feasible due to the high cloud cover for much of the year.

Despite the 6-year interval between the dates of the image and the field survey, the multispectral classification of the Landsat TM image produced good results. The vegetation map we obtained gave reasonably reliable information on the abundance and distribution of the main vegetation types, and provided clear evidence about how cattle ranching had influenced the vegetation. An overall accuracy of 77% is sufficient for many uses and lies within the range obtained in other studies (e.g. 76·7% for blanket bog in Scotland, Reid & Qarmby 2000; 85% for Mediterranean vegetation; Grignetti et al. 1997; 40% for an arid grassland shrubland vegetation in Australia, Lewis 1998). The main problems were in distinguishing structurally similar vegetation types. For example, species dominating one type of bushland also occur in other types, while Terminalia bushland has many of the evergreen and semi-evergreen species found in semi-evergreen woodland. Relatively open bushland can be confused with grassland or thicket–grassland mosaics. Echinochloa wooded grassland and Panicum wooded grassland both occur in moist areas and are structurally similar. Tall stands of Andropogon gayanus, which can grow to > 2 m, may resemble Hyperthelia dissoluta and so thicket–grassland mosaic can be wrongly classified as Hyphaene-dominated grassland. On the other hand, if the Hyphaene-dominated grassland is grazed the grass layer is similar to that of the thicket–grassland mosaic.

distribution of bushland

The two main trends in the distribution of bushland are a decrease in abundance from north to south, and a concentration of bush in the neighbourhood of paddocks. While the decrease of bushland and woodland types from Mkwaja North to Mkwaja South could be partly a result of north–south gradients in rainfall, topography and soils, the local patterns of bush encroachment around former paddocks are undoubtedly of anthropogenic origin. The B/G ratio decreases at distances above 900 m from a paddock and reaches a value of 0·4 at 4000 m (i.e. the same value as the average for the Saadani Game Reserve in this study). This suggests that bush encroachment in Mkwaja is chiefly the result of ranching, and that the vegetation structure more than 3500–4000 m from a paddock has not been significantly affected by livestock. On Mkwaja Ranch the distances between the paddocks are mainly between 4 and 8 km, so that > 55% of the area of the ranch lies within the zones mostly affected by bush encroachment (300–2500 m). The significant negative correlation between the B/G ratio and the available grazing area per paddock may be best explained by the fact that, with increasing grazing area, the maximum distance to the paddock also increases.

Similar bushland distribution patterns have been found by Perkins & Thomas (1993a, 1993b) around artificial water holes in a semi-arid savanna rangeland in Botswana. They distinguished four more or less distinct zones: a ‘sacrifice’ zone, with low vegetation cover (0–400 m from water holes), a nutritious grass zone (200–800 m), a zone of high bush encroachment (200– 2000 m) and a zone of shrub savanna. In a simulation model, Jeltsch et al. (1997) generated comparable spatial patterns using parameters appropriate for a semi-arid grazing system in the Kalahari (385 mm average annual rainfall). After a simulation period of 16 years, the pattern of bush cover strongly resembled that found in our study. From this we may conclude that the pattern of bush encroachment in this humid savanna is similar to that found in semi-arid savannas despite the expected differences in the balance between grasses and woody plants. Using the same model as Jeltsch et al. (1997), Weber et al. (1998) investigated the influence of stocking rate on bush encroachment and found that there was a threshold density of livestock below which bush encroachment was very low and above which it increased rapidly. A simulation of vegetation development over 50 years with an annual average rainfall of 300–400 mm and a stocking rate of > 10 ha lsu−1 resulted in a maximum bush cover of  >  60%, a level similar to that recorded at Mkwaja at an average distance of 900 m from paddocks. The stocking rates on Mkwaja were 1–7 ha lsu−1, significantly higher than in many semi-arid grazing systems, for which 9–16 ha lsu−1 are recommended (Perkins & Thomas 1993b), but the carrying capacity at Mkwaja may be higher due to higher rainfall (annual average of 900 mm). However, average stocking rate can be a poor predictor of the grazing pressure in an area. Particularly in grazing systems that are based on central points like paddocks or waterholes, the impact close to those points is significantly higher than it is in areas further away.

Other management factors have also influenced the development of scrub at Mkwaja. On Mkwaja North the brush cutter was used regularly from the 1960s to the 1980s to fight bush encroachment. According to several sources (Klötzli 1980; Lupi & Walther 1994; ranch reports), this had the effect of promoting both regeneration from seed and regrowth, particularly of Acacia zanzibarica, and actually promoted bush encroachment. Although brush cutting was finally abandoned in the 1980s, the effects of this management on the vegetation are still evident. For example, near paddocks 2 and 3 (Fig. 7) the area with the highest bush density probably represents the zone that was first invaded by Acacia zanzibarica as a result of intensive cattle grazing and was then repeatedly cut. In contrast, the paddocks in Mkwaja South were spared from brush cutting and it can be assumed that the bush pattern is primarily grazing-induced. For example, around paddocks 14 and 16 the peak in bush density is less marked and closer to the paddocks than is the case for most northern paddocks.

Although there was a general increase in bush associated with ranching, there were noticeable differences in the distribution of the various types of bushland, which can be explained by differences in the ecology of the dominant bush species. For example, around paddocks 5, 6, 7 and 16 the mixed Dichrostachys–Acacia bushland is more abundant than Acacia zanzibarica bushland. The seed pods of Dichrostachys cinerea and Acacia nilotica, the most important species in this bushland type, are eaten by cattle and the seeds are thus dispersed in places where animals gather (cf. Van Staden, Kelly & Bell 1994; Miller 1996; Brown & Carter 1998; R. Cochard personal observation). In certain places (such as in paddock 7) high densities of seeds of Acacia nilotica and Dichrostachys cinerea could still be found 2 years after the paddocks were abandoned. Acacia zanzibarica has dehiscent pods that are not consumed by cattle. It appears to be favoured in areas of high to intermediate grazing, not because of enhanced dispersal but as a result of increased seedling recruitment and survival. Acacia zanzibarica seems to be well adapted to a wide range of conditions in this area and bushlands are not restricted to the vicinity of the paddocks. Acacia zanzibarica bushlands also occur in Saadani, where there has been no cattle grazing for many years but native herbivores are relatively abundant. The factors that led to the formation of these stands are unknown, but grazing by animals such as hartebeest Alcelaphus buselaphus Pallas and the introduced wildebeest Connochaetes taurinus Burchell and/or fire management may be involved.

Terminalia spinosa has winged seeds that are wind-dispersed. It is a successful colonizing species, particularly in undulating terrain within the Andropogon thicket–grassland mosaic where soils are often visibly degraded by trampling and grazing (e.g. around paddocks 4, 9 and 14). In some areas that were cleared of forest or woodland (e.g. around paddocks 4, 7 and 9) it seems to be the first woody species to recolonize, perhaps restoring the shade and nutrient conditions needed for the establishment of forest species.

Although Hyphaene compressa is cited in the ranch reports and in the work of Klötzli and others (Klötzli 1980; Lupi & Walther 1994) as one of the most severe ‘problem’ species, Hyphaene bushland is not prominent near paddocks. This suggests that it was promoted less by cattle than by brush cutting; this is consistent with its abundance in those parts of the fly belt that were probably Hyphaene-dominated grasslands before the area became a ranch. According to ranch reports, infestations by Hyphaene compressa were intensively treated with arboricides followed by manual clearing. This could also explain why in 1994 there was relatively little Hyphaene bushland near the paddocks.

distribution of grassland

The abundance of long grassland increases gradually from < 5% cover near paddocks to about 15% at 2800 m distance (Fig. 6a). Fire tends to favour long grassland; Hyperthelia dissoluta, the major species of long grasslands, is only grazed when young (Ibrahim & Kabuye 1988) while it is promoted by intense fires (Knapp 1973; Klötzli 1980). Fire also suppresses most bush species other than the fire-adapted Hyphaene (Knapp 1973). Both factors, management for cattle pastures and subsequent fire control, may thus explain why this grassland type was replaced by short-grass types under heavy grazing. Short grassland was probably more susceptible to bush encroachment because it provides better pasturing for cattle, and occasional fires were probably not intense enough to kill off bush regrowth. Short grassland covers almost 30% of the areas close to paddocks and this proportion does not increase recognizably up to about 3000 m away from paddocks (Fig. 6a).

The short grass vegetation in the vicinity of paddocks contains many ruderals and invasive species (Barker, Thurow & Herlocker 1990; Young, Patridge & Macrae 1995; Cornelius & Schultka 1997; Rietkerk et al. 2000). The paddocks themselves were highly disturbed environments that received high inputs of animal excreta. The grass community within paddocks clearly represents a floristically distinct vegetation type, but due to its very confined distribution it was not distinguished from other short grassland types in this study.


The results of this study help us to understand why intensive livestock ranching as practised on Mkwaja Ranch is unsustainable. The serious problems facing the ranch managers at Mkwaja became evident at a relatively early stage in the history of the ranch, and were described by Ford & Blaser (1971) writing just 16 years after the enterprise began: ‘It is still not possible to say whether management will succeed in controlling vegetation and disease at the same time or whether, in the end, these obstacles to cattle raising will prove too expensive for the ranch to continue’. For a few years, a stocking rate of 1–7 ha lsu−1 could be maintained. If, however, carrying capacity implies a ‘continuing yield without environmental damage’ (Allaby 1994), an appropriate stocking rate for this type of single-species livestock system is probably much lower and may be similar to that recommended for more arid savanna ecosystems. The climate in humid savannas is generally less erratic than in arid ecosystems. This implies that the potential for degradation of grass and soil resources is reduced, as degradation in rangelands is often aggravated during episodic drought events (Rietkerk 1998; Fynn & O’Connor 2000; Sternberg et al. 2000). However, the potential for gradual bush encroachment promoted by livestock may be higher in humid savannas that have been derived from woodland and forest and are mainly fire controlled.

In theory, the highest stocking rates may be sustained under uniform grazing pressure over the whole area. In practice, however, grazing impact is never uniform, even in natural grazing systems. A discussion about an ecologically appropriate and sustainable stocking rate should therefore account for the spatial component of any ranching system. In this context, the advantages of traditional, nomadic livestock systems over modern intensive ranching are again being rediscovered (Behnke, Scoones & Kerven 1993). Even on confined private ranches there are ways to achieve a more regular grazing distribution. For example, paddocks should only be used for a limited period (1–2 years); grazing should then be transferred to an area that has not been recently used, thus reducing the dangers of long-term overgrazing in the surroundings of paddocks. Clearly this option is feasible for paddocks but not for dams or boreholes. The daily management of the herd is also important; herdsmen should quickly lead cattle away from the paddock in the morning, so that areas further away from the paddock are grazed with similar intensity.

More research on grazing patterns as well as on the ecological factors determining humid savanna vegetation is needed to understand their dynamics under different management regimes. The survey technique presented here provides a convenient way of monitoring the process of bush encroachment. Further, a comparison of the distribution patterns of the various bush species at the landscape level may be useful in developing improved management regimes targeted on particular species. In the particular case of Mkwaja Ranch, which will soon become part of the new Saadani National Park, it will also be interesting to observe how the vegetation changes when natural factors such as fire and a diverse assemblage of herbivores resume their former importance.


Financial support for this project was provided by Dr U. Albers, Novartis, the Swiss Development Cooperation (DEZA) and the Swiss Federal Institute of Technology (ETH). We thank Jennifer Marion Adeney, Christoph von Känel and André Wehrli for their help with the fieldwork, the logistics and for their comments on the manuscript. We thank Professor Frank Klötzli for his help and advice, and for providing us with his records and reports on Mkwaja Ranch. We also thank the staff of the Swiss embassy in Dar es Salaam, especially Marianne Wyss and Titiano Bassi, Dr Rolf Baldus from the GTZ in Tanzania, Claire-Lise Reift from the Swiss Development Cooperation in Tanzania, Dr S. L. S. Maganga from the Sokoine University of Agriculture in Morogoro, and our assistants on the ranch, Benjamin and John. We are grateful to Roland Brun and Sabine Guesewell for statistical advice and valuable comments on the manuscript.