Introduction
Understanding the structure of natural ecosystems forms the basis for understanding the processes within those ecosystems, including the transmission of infectious and vector-borne diseases. Remotely sensed datasets and geographical information systems (GIS) have been widely used to further our understanding of these systems. This technology has not only helped in the study of the global drivers of ecological change, but is also invaluable for understanding the biotic and abiotic factors influencing ecosystems at much smaller scales. GIS technology has become an integral component of many conservation programmes and the development of trans-disciplinary approaches such as Conservation Medicine, One Health and EcoHealth have highlighted its utility. However, recent moves to adopt ecosystem-based approaches within conservation and development programmes have highlighted frequent deficiencies in baseline ecological data, particularly in developing countries (Rapport et al.
1998).
For an area with such an internationally acclaimed biodiversity, relatively little ecological data exist for the Luangwa Valley in eastern Zambia (latitude −10.4° to −15.6° and longitude 30.2° to 33.1°). Although many valuable mapping and vegetation studies have been conducted, they either lack detail (Trapnell
1950; Naylor et al.
1973; Phiri
1989; Marks
2005) or have restricted geographical coverage (Astle et al.
1969; Smith
1998; Yang and Prince
2000). Similarly, published faunal surveys for the Luangwa Valley are rare and no peer-reviewed published data are available for many areas. Studies have been conducted in game management areas (GMAs) surrounding some of the national parks (Ndhlovu and Balakrishnan
1991; Lewis et al.
2011), and many of the species recorded historically in the mid-Luangwa Valley have been documented (Astle
1999). Aerial surveys have been conducted in the core parts of the Luangwa Valley on behalf of the Zambian Wildlife Authority (ZAWA) (Simukonda
2011) and as part of the Community Markets for Conservation Programme (COMACO) (Olive et al.
2012; Frederick
2013). The population of hippopotamus (
Hippopotamus amphibious) has recently been surveyed and extensively studied (Wilbroad and Milanzi
2010; Chansa et al.
2011a; Chansa et al.
2011b). However, there is a clear need for more high-resolution data to enable active monitoring of ecosystem health in the valley.
There has been much interest in the role of warthogs (
Phacochoerus africanus) as natural reservoir hosts for African swine fever (Plowright et al.
1969; Wilkinson et al.
1988), trypanosomiasis (Dillmann and Townsend
1979; Claxton et al.
1992) and bovine tuberculosis (Bengis et al.
2002; Michel et al.
2006). Warthog burrows not only provide a refuge for warthogs from predators and extremes of temperature, but they also provide a refuge for many parasites (Cumming
1975; Somers et al.
1994). The cool shady conditions in the entrance to warthog burrows provide an ideal refuge for tsetse flies during the heat of the day (Pilson and Pilson
1967), and the burrows are important sites for larviposition by female flies (Leak
1998). Warthogs are also a preferred host for
Glossina morsitans species of tsetse flies, and a close ecological association between tsetse and warthog has been proposed (Pilson and Pilson
1967; Torr
1994; Leak
1998). A study of tsetse ecology in Luambe National Park (LNP) revealed that
Combretum-Terminalia vegetation supports the highest apparent density of
G. m. morsitans and thicket the highest apparent density of
G. pallidipes (Anderson
2009). Warthogs have been shown to carry a moderate prevalence of trypanosomes and the human-infective
Trypanosoma brucei rhodesiense, the cause of human African trypanosomiasis (HAT), has been identified in warthogs in the Luangwa Valley (Dillmann and Townsend
1979; Anderson et al.
2011). As a wide variety of other hosts are fed on by tsetse to varying degrees (Clausen et al.
1998) and are components of the natural reservoir community for trypanosomiasis in the Luangwa Valley (Anderson et al.
2011), it is important to understand more about the density and distribution of wild animal hosts within these ecosystems.
The majority of investigations into trypanosomiasis in wildlife have focussed on estimation of the prevalence of infection. Historically, prevalence was largely interpreted in terms of host susceptibility to infection and, to a lesser extent, host preference by tsetse. However, the importance of ecological and behavioural factors in the transmission of wildlife disease is now recognised (Cross et al.
2009). Factors such as habitat preference, resource use, territoriality, group size and group density contribute to a complex social and spatial structure for wildlife disease. Understanding the structure and distribution of both plant and animal communities is therefore critical for clarifying the nature of contact between hosts and vectors, and its impact on disease transmission. In a detailed review of the ecological factors influencing the epidemiology of trypanosomiasis in the Luangwa and Zambezi Valley ecosystems, Munang’andu et al. (
2012) identified host distribution and abundance as having a significant influence on the survival of tsetse and therefore on trypanosomiasis epidemiology. Many other factors are also important including daily activity patterns of hosts and seasonal migration behaviour. Tsetse distribution and abundance is largely driven by climatic factors, host abundance and vegetation (Robinson et al.
1997). A better understanding of the distribution and characteristics of both mammal and plant communities is therefore likely to improve our management of HAT.
Here we generate accurate high-resolution datasets of vegetation and large mammal density in the remote, relatively inaccessible LNP within the Luangwa Valley, in order to investigate the ecology of warthogs and to further our understanding of their interactions with Glossina spp., vectors of trypanosomiasis.
Discussion
The overall accuracy of 71.2% for the final classified image was considered to be good and the fuzzy logic algorithm presented statistically significant improvements over the conventional maximum likelihood algorithm. The presence of mixed pixels in the image is likely to account for much of the difference in performance between the two. Mixed pixels have been identified as a major source of error in traditional ‘hard’ classifications that assign only one class to each pixel (Wang
1990; Foody
1996; Benz et al.
2004) and as the most important cause of misclassifications (Foody
2002). Detailed information on joint membership by other classes, particularly around boundary areas, is lost. There is no doubt the training data will have contained some mixed pixels as vegetation exists as a continuum in LNP (as in most natural ecosystems) and classes overlap. Indeed, in his detailed floristic study of North Luangwa National Park, Smith (
1998) grouped his vegetation categories into mosaics of vegetation types that could not be mapped separately at his chosen resolution. Although attempts were made to collect homogenous reference data in this study, it will have contained some heterogeneity and represented, in effect, ‘fuzzy’ ground-truth data. An important feature of fuzzy classifiers is that homogenous reference data are not needed.
Most of the classes within the classification scheme performed well, with the exception of the hill scrub miombo class. A small area of this class was identified during the ground survey on the hills in the south eastern section of the park, but only at an altitude of 660 m or greater. The highest point in LNP, based on the 1:250,000 topographical maps (Surveyor-General
1972), is 680 m meaning this class was only present over a very restricted area. As it was exerting a deleterious effect on the accuracy of the rest of the classification, it was removed. Most of this area is mapped by the classification as
Combretum woodland with some scrub mopane woodland. In reality, it is likely to represent more of a transition zone from
Combretum woodland to scrub miombo woodland rather than just the latter.
Vegetation Composition of the Park
As discussed earlier, much of the vegetation of LNP exists in a natural mosaic of vegetation types. However, at a larger scale, several classes occur in fairly discreet zones, notably mopane woodland, mopane scrub and grassland. The two forms of
Colophospermum mopane vegetation together are dominant over large areas of the park covering 37% of the total area. The large grassland habitats formed by the floodplains of the Luangwa River tributaries are a significant component of the park covering 14%. Thicket vegetation also forms fairly clear zones in places, but interdigitates with
Combretum woodland in others. Aquatic grasslands, in the form of permanent or semi-permanent lagoons, account for a much smaller proportion of the total park area, but are a very characteristic feature of LNP. Riverine woodland mainly flanks the Luangwa River, but is also found in patches by the larger tributaries and lagoons. Detailed descriptions of the vegetation classes in this study may be found elsewhere (Anderson
2009).
Species Densities
Despite its small size, LNP has some distinctive wildlife habitats and supports populations of several globally threatened or endemic mammal species. The wildlife density estimates presented here represent the most detailed published information to date and can act as a baseline for on-going research and monitoring. Overall mammal densities were relatively low with some notable exceptions such as puku (Kobus vardonii).
Of the most abundant species, only warthog are known to be preferred hosts for tsetse (Clausen et al.
1998). Warthogs are generally a successful species and density estimates vary from 15 km
−2 (Cumming
1975) to 30 km
−2 (Estes
1993) in the best habitat (fertile alluvial soils), and less than 1 km
−2 (Cumming
1975) in less favourable areas. Densities are highest in short grassland or wooded grassland areas (Rodgers
1984) and mosaics of suitable wet and dry season habitat are important (Cumming
1975). The density of warthog estimated in this study (3.14 km
−2, 95% CI 1.93–5.98) was comparable with the density recorded in nearby Upper Lupande GMA (2.2 km
−2) and the Zambezi Valley, Zimbabwe, but towards the lower end of reported densities (Cumming
1975; Rodgers
1984; Ndhlovu and Balakrishnan
1991). The relatively high proportion of warthog clusters observed in aquatic grassland is notable as it covered only 4% of the transect area. Aquatic grassland presents a reliable dry season source of forage for warthogs which were frequently observed digging for rhizomes of grasses and sedges in this habitat. Observations were comparatively frequent in the
Combretum woodland habitat, which has a well-developed grass layer. Also notable were high densities of warthog in areas with new grass appearing after a bush fire. Marked local increases in warthog density after dry season fires have been noted before (Cumming
1975). Uncontrolled burning is a regular occurrence in LNP and is also likely to exert a significant selection pressure on the vegetation, especially given the resistant nature of
Combretum species in particular to fire (Smith
1998). In turn it is likely to have significant effects on the diversity and abundance of fauna.
Warthog Burrow Distribution
Mapping of the warthog burrows over the classified dataset allowed the spatial pattern and habitat preference for burrow location to be examined. The clear pattern revealed in Figure
3 may be explained by the drainage of the soils, the ease of excavation and the provision of cover from predation.
Combretum woodland and thicket are generally found on more sandy soils with better drainage in the rains and easier excavation. In contrast, mopane woodland and mopane scrub generally occur on clay soils, prone to seasonal flooding (Smith
1998) and difficult to excavate in the dry season. Warthog have been reported to thrive in areas of wooded grassland bounding suitable floodplain grassland (Rodgers
1984), a situation which occurs especially towards the south of the transect study area in LNP. The close proximity of patches of aquatic grassland to burrows makes suitable dry season grazing readily available. The close ecological association between warthog and tsetse was outlined earlier in the Introduction including the observation that apparent densities of
G. m. morsitans tsetse are greatest in
Combretum woodland and
G. pallidipes in thicket (Anderson
2009). It is very notable, therefore, that the majority of warthog burrows are located within these two habitats.
Habitat Densities
The use of the classified dataset also allowed the aggregate density of wild mammals to be examined by vegetation class. Not surprisingly, the highest densities were recorded in grassland with nutritious herbage providing for large densities of puku in particular. Although lower densities of large mammals were recorded in the riverine woodland and aquatic grassland classes, these habitats are likely to be very important ecologically, especially in the dry season as a source of forage and water. They may support a wide diversity of other species not included in the survey such as birds, amphibians and invertebrates. Acacia woodland forms only a very small component of the vegetation in LNP and animals were rarely observed in the dense stands of Acacia kirkii, but were more commonly seen in more open acacia woodland near the Luangwa River.
It would have been desirable to estimate individual species density by habitat type, but the data were not robust enough to allow this. The large survey effort required (approximately 60 observations per habitat type for each species) makes this difficult to achieve across all habitat types, especially in environments with low mammal densities. Similarly, four land cover classes (riverine woodland, thicket, mopane scrub and Combretum woodland) did not have sufficient observations to enable the use of the stratum detection function in the analysis. Although preferable to using the global detection function, an accurate estimate was made possible by pooling the class in question with the class with the most similar visibility characteristics and using the global detection function for the two classes combined to estimate density. Riverine woodland was pooled with thicket for this purpose, and mopane scrub was pooled with Combretum woodland. Although species and cluster size may have confounded the detection probabilities to some degree, the estimated densities provide a useful indication of the general distribution of mammals.
Size of the Park
Calculation of the area of LNP using the classified image (331 km
2) produces a considerably different value to the official figure for the park area (254 km
2). Unfortunately, it is not clear where the official figure used by the ZAWA originates from. Changes in the course of the Luangwa River forming the western boundary of the park are occurring continuously, but will not account for such a large discrepancy. Although the exact boundary of the park is disputed by the local community, the shapefile used in this study was created through digitisation of high resolution topographical maps (Surveyor-General
1972) based on the original gazetting of the park, which suggests the national park area figures used by ZAWA may be incorrect.
Conclusion
This study provides a reliable framework for ecological monitoring of vegetation composition and wild mammal densities in remote, relatively inaccessible environments. Information generated can be used as a baseline for further study into wildlife disease systems. The use of classification algorithms based on fuzzy set theory enables accurate classification of vegetation classes despite the presence of natural mosaics and mixed pixels. The datasets created are ideal for use as a GIS base layer for the design, implementation and analysis of ecological and epidemiological studies. The distance sampling technique utilising a ground survey allows for reliable estimation of densities of smaller mammal species and important hosts of trypanosomes such as warthogs. The large survey effort required to estimate species density accurately in areas with relatively low wild mammal densities may limit the usefulness of this technique for health research in some environments.
Despite decades of research into trypanosomiasis our understanding of disease transmission in wildlife hosts is limited by the complexity and large size of the reservoir host community, and the many factors that influence it. Accurate description of the structure and distribution of communities is necessary to further our understanding and will enable better management of health relationships in remote environments such as those described in this study. Data such as these will help to enable improved modelling of disease systems with a consequential improvement in our understanding of the effects of interventions in biodiverse ecosystems.