Tinley also identified four thicket types and two forest types from the valley floor area. All thicket types (riverine, alluvial fan, tree-base, and termitaria thickets) appear to occur in the Muaredzi area, but the forest types appear to be absent.

As it is situated within the national park, the village is exposed to wildlife. Elephant move within the village area and surrounds and clearly do cause some destruction to crops. A number of smaller animals are also commonly seen close to the village, including nyala, impala, bushbuck, oribi, warthog, and wild pig. Lake Urema is reported to harbor a healthy population of crocodiles, and hippos are also present.

The Nhanchururu site is situated astride the western boundary of GNP, some 15 km southeast of Gorongosa Mountain, and some 25 km northeast of Villa Gorongosa (Fig. A1.2). It is part of the Barue Plateau, the altitude of which varies between about 200 and 340 m above sea level. The terrain is deeply dissected, with rivers draining south to the Mucodza River and north or northeast to the Vunduzi River. The community lies on the upper portion of the rift escarpment, on the watershed between the Mucodza and Vunduzi Rivers.

The vegetation of the Nhanchururu area is largely miombo savanna woodland, but with some evergreen thickets on the deeper sands of the interfluve crests. The dominant woodland species are Brachystegia boehmii, B. spiciformis, Erythrophloeum africanum, Julbernardia globiflora, and Pterocarpus angolensis. There are some narrow patches of thick riverine forest along the Vunduzi and Mucodza Rivers but these are very limited in extent.

Sketch maps drawn by community members provided more specific background data for Nhanchururu. The village area is roughly rectangular in shape, about 10 km south to north and 8 km east to west. Nhanchururu is bounded to the east by the national park, to the west by Nhangeia village, to the south by Nhanthemba village, and to the north by Safumira village. The boundaries with adjacent villages appear to be reasonably clear. These comprise the Mucodza River to the south, the Vunduzi River to the north, and a minor drainage called the Rio Nhachituzui to the west.

To the east, the boundary between the village and the park is less clear. The community members were adamant that the entire village was outside of the park, and that the park started immediately to the east of the village, with the boundary being marked by a line of low hills and the small Rio Nhachiru. However, at the approach to the village along the main access road from the west, shortly after entering the village area, there is an official sign stating that one is now entering Gorongosa National Park. According to this, the bulk of the village falls within the national park. Regardless of this situation, the community members seemed to feel much more secure than the Muaredzi residents, and there was never any suggestion of fears that the park may in future attempt to move them.

In terms of roads and major paths, the main access road follows the watershed between the Vunduzi and Mucodza Rivers, bisecting the village into southern and northern portions. It leads through the village to the Rangers' post, and then continues east into the park (and in former times apparently all the way through to Chitengo). There were no other significant tracks to the east. To the south, there are two routes that cross the Mucodza River, both of which are located towards the western end of the village. One of these is a shortcut to Villa Gorongosa, if traveling by foot or bicycle. As far as vehicles are concerned, this route appears not to have been used for some time, is in a very poor state of repair, and the crossing over the Mucodza would not be passable until late into the dry season. To the west, in addition to the main access road, there is one other footpath that crosses the Rio Nhachituzui and continues to the neighboring village. To the north, there are a number of routes that lead off the main access road towards the Vunduzi River. Two of these reach to the Vunduzi, but neither appears to cross the river.

A total of 107 households were identified within the village, split roughly equally to either side of the main access road. Households tend to be scattered individually rather than clumped. Nhanchururu has four fumos. Of these, Fumo Almeida appears to be the most influential, and the other three of lesser significance. The responsible Regulo lives outside of the village to the south of the Mucodza River.

People were moved from the rift valley areas of Gorongosa National Park in the 1950s to the Barue plateau area, including what is now Nhanchururu. Further disruptions and movements occurred during the war for independence and the subsequent period of continued fighting.

APPENDIX 2.

The same approach was followed for both sites: a traditional ceremony was held first, followed by an open community meeting; then, a representative group of community informants (community resource use assessment team, or CRUAT) was selected, a modus operandi was established with the informant group and, thereafter, the process of data collection was begun. For Muaredzi, this was achieved over a series of three field trips (September 2001, November 2001, and April 2002). For Nhanchururu, the traditional ceremony was held in April 2002 and the remainder of the activities and collection of community livelihood data were carried out during a single field trip in May 2002.

The initial community meetings provided an opportunity to explain the aims and needs of the project to those present. The community members were told that the project team sought an improved understanding of household and community livelihoods. We explained that we wished to work with a limited group of informants, and that these informants should be representative of the major socio-economic groups within the community. These representatives would form the CRUAT. In Muaredzi, the CRUAT comprised 14 men and 8 women; in Nhanchururu, 10 men and 8 women.

Three basic tools were used to conduct the analysis: spidergrams, sketch mapping, and open discussion. As each of these have been described in detail elsewhere (Lynam 1999, 2001), they are not described here.

In the initial model, the importance of a landscape unit to the community was expressed as a simple ratio of benefits to costs (i.e., benefits divided by costs—B:C). Thus, the larger the B:C ratio, the more important the landscape unit or location was expected to be. The benefit side of the model was defined as a function of three inputs: i) the relative importance or preference for each of the goods and services (GS) derived from a given landscape unit or location; ii) the number of such GS; and iii) the density of GS per unit area in the landscape unit. Thus, the gross benefit derived from a unit of the landscape was a simple weighted sum of the importance score and density across all GS (Eq. 1).

Where:

B = the total, gross benefit derived from a landscape unit or element;
P_i = the preference weighting (RIW) for a good or service_i;
D_i = the density of a good or service _i, where density ranges between 0 (none) and 1 (maximum).

The cost component of the model was deemed to be a function of three major cost sources. First, the distance traveled to obtain the good or service, where this distance was the weighted sum of distances along major routes and distances off routes. The off-route distances would be more costly. The second cost source was physical barriers, such as rivers, wetlands or steep terrain. The third cost-contributing source was the institutional barriers or rules and regulations governing access to a given resource or landscape unit. This latter group was complicated by the elements associated with institutional costs—in the context of this project, the probability of transgressions being discovered and then the associated fine or punishment for deviations. This was simplified in the model to reflect only an opportunity cost associated with regulations—the resource use opportunities forgone due to the regulations.

The conceptual model defined our expectations of the determinants of landscape importance. Explicitly, the expectations derived from the model were that the importance of landscape units would be highest where there were multiple goods and services of high importance that were not governed by limiting institutions, which were close to the household or community, and which had no barriers impeding access. Landscape units or locations of low importance would occur under the reverse conditions.

A computer implementation of the model was developed as a Bayesian Belief Network (BBN) using Netica (www.norsys.com).

Information obtained from the CRUATs was subsequently used to shape and update the model for each study site. In particular, this enabled the detailing of goods, services, and cost functions for each site, and the assignment of relative weights to each of these factors. The result was the development of specific prior models for either site. These models were at a stage where, when information regarding the status of each of the peripheral nodes (goods and services and cost functions) for a particular point location was input, the model would provide an estimate of the most probable landscape importance for that location.

The final step in terms of collection of field data was to carry out a sampling process, in order to generate field data with which to confront the model, and to provide the basis for further refinement and updating of the model. The general approach was to visit a number of locations within each village, together with CRUAT members and, for each location, to record their scores for each of the goods and services present at the site, for all cost factors, and then an overall landscape unit importance score. The scores for goods and services and for cost factors were subsequently fed into the model, and the model then generated an estimated value for each sample. These estimates were then compared with the CRUAT importance scores for each sample.

In order to increase the number of samples possible within the available time, the CRUAT group at each site was split into three or four subgroups. Each subgroup comprised several community members, plus a data recorder (facilitator). For Muaredzi, each subgroup comprised two men and four women; for Nhanchururu, two men and two women.

Sampling was done along line transects. Each subgroup covered a single transect per day. Transects were selected based on overall coverage of each village area, coupled with logistical constraints, notably the existence of potential access paths and roads (the start and end points for each transect needed to be accessible by path or road). The length of transects was decided according to the estimated time available for sampling, i.e., total working time of 6 hours each day, less time required to travel to the starting point and to return from the end point. Lengths varied from about 1.5 to 4.5 km. The sampling interval and number of samples per transect were decided in the field, once at the starting point for each transect, and were based on the estimated time available for sampling and the time it was likely to take to traverse the transect. Sampling intervals ranged from about 250 to 600 m and the number of samples per transect from four to 12.

Transects covered the principal land types within each area, and all different combinations of distances along paths and off paths (as the prior versions of the model had shown a high sensitivity to these parameters). Actual positions were first selected on satellite imagery. Thereafter, the coordinates of the start and end points were read off the GIS maps and entered into GPS units.

Sample areas were circles, roughly 30 m in radius (i.e., 0.28 ha in extent). When a group arrived at a sample point, they scored the necessary factors based on consideration of the resources, etc. apparent within a 30-m radius. Sample areas were not systematically searched either before, or during, the scoring.

Scoring of landscape values was open ended, and relative to the least important locality within the village area, which was allocated a score of one point. For either site, the reference point of lowest importance was identified at the outset of the sampling process, by the entire CRUAT group together. For Muaredzi, the CRUAT identified a certain occurrence of chipale, known as nteca, as being the site of lowest value. For Nhanchururu, the CRUAT identified a certain range of hills within the national park area as being the lowest value. CRUAT members reported that they were familiar with these sites and the types of resources to be found there. However, in neither case had all the informants, particularly the women, ever been to these places, nor did they visit them as part of this exercise.

For Muaredzi, a total of 75 samples was recorded from ten transects over a 3-day period. For Nhanchururu, 82 samples were obtained from 13 transects, recorded over 7 days. There are several reasons for the lower rate of sampling for Nhanchururu compared with Muaredzi: Nhanchururu was the first site sampled and the procedure was still unfamiliar; the terrain was more difficult (broken and hilly); and rain caused disruptions.

Field sample data were subsequently entered onto a spreadsheet to form a case file for each site. Each case file consisted of the total number of samples (75 for Muaredzi and 82 for Nhanchururu), with each sample having scores for all goods and services and for all cost factors. Based on these data, the model generated estimated landscape values for each sample location. These values were then compared with the values given for each sample by the CRUAT members.

The case files were then used to confront the models for each site. In each case, the models were first confronted with the data in the case files, and the same case files were then used to update the probability structure of the model. The resulting (posterior) models were subsequently used to explore the sensitivity of the models to the collection of further information for each node, and also to explore the implications of the understanding gained for land-use planning and policy decision making.

The extent of the sample area for each site was selected based on initial discussions with the Muaredzi community on how far they traveled to collect or use resources, together with subsequent discussions among the research team. For both sites, this comprised a square, centered on each respective community, and 20 km on a side (giving a total sample area of 400 km² for either site). These areas dictated the extent of the vegetation assessments and the development of spatial data sets for the two sites.

Topographic maps at 1:250 000 and 1:50 000 were obtained from the Mozambique government, and data sets were digitized within the 400-km² area around either community. The 1:50 000 maps were based on rather old air photography from 1958 to 1960. Additional data were obtained through field mapping using a number of handheld Garmin GPS units (using the WGS 84 datum). For both sites, the positions of households and all major roads and paths were recorded.

As far as possible, the BBN models developed for each site were used to guide the generation of final landscape importance maps for each site. Although the calculations for generating the final maps were simplifications of the calculations used in the models, the general approach and principles were the same; the final landscape importance maps were developed as ratios of the benefits to the costs. Benefits were calculated as the weighted sum of scores of the benefits that the CRUATs identified as derived from each vegetation type. The weightings were the CRUAT relative importance weights (RIW) as used in the BBN. For each site, the developed vegetation maps were converted into raster maps with square cells. These dimensions were selected because they were the same size as the sample plots for field confrontation.

The cost maps were a little more difficult to generate. First, the distance from household cost raster was generated as a buffer raster of distance from the households noted in each site. The distance classes that were used were the same as those used in the field confrontation estimates. As in the BBN, distances were estimated along paths and assigned cost values based on the proportional costs allocated to this distance function by the CRUAT at each site. Then, distance from paths was estimated using a buffer from the mapped paths (these were mapped using handheld GPS units and the results converted into vectors). Again, the distance classes developed in the field were used to assign costs to these buffers. The total distance cost was then estimated as the sum of the costs of the on-path and off-path distance maps.

APPENDIX 3.

Landsat images (Scene 167/73; 22 August 1999) and aerial photographs were interpreted by carefully examining paper copies, as well as on-screen images. The study areas were initially demarcated on the image to form a 10 x 10 km square but were revised according to boundaries indicated by communities in the respective areas. Image and aerial photograph interpretation was used to produce preliminary vegetation associations evident from differences in color and texture on the aerial photographs and images. This formed the basis upon which the vegetation was stratified and enabled sampling within each vegetation stratum. Field work was carried out in the Muaredzi area between 3 and 14 September 2001, and in the Nhanchururu area between 6 and 18 May 2002. Ground truthing of vegetation boundaries and further assessments were done between 8 and 19 April 2002 in the Muaredzi area. The ground-truthing exercise was deemed unnecessary for Nhanchururu because of the simplicity of the mapping units.

Vegetation survey

In Muaredzi, four main transects covering the area were identified according to the directions of the main access roads. Taking the Rangers’ Post as a reference point, these were: the track toward the confluence of the Urema and Muaredzi Rivers (western direction); the road toward Goinha village (northern direction); the road toward Muanza town (eastern direction); and the road toward the Urema crossing to Chitengo (southern direction). In addition, a number of smaller access tracks were followed but much of the inventory was done along the main roads. The positioning of the roads seemed to adequately cover much of the variation in the vegetation evident on the Landsat image.

In Nhanchururu, there was better access to places compared with Muaredzi. The former site, with its widely scattered homesteads, had more tracks and paths branching through the area, allowing better access to sample areas. A number of these paths were followed and assessments of vegetation done.

For both sites, a plot-less sampling procedure similar to that of Timberlake et al. (1993) was followed when inventorying the vegetation. Sites were selected within the stratified zones according to how representative they were of the vegetation type under consideration. At each site, a starting point was randomly selected and a circular area covered around this central point, recording all plant species until no new species were encountered within the defined area, which was usually between 0.25 and 0.5 ha, depending on species richness. This approach follows the concept of the species-area curve (Connor and McCoy 1979), which ensures that an adequate area to record all species is sampled. Care was taken to avoid roadside margins and to ensure that no obvious environmental boundaries were traversed to avoid straying into different vegetation units. A cover abundance value for each species was estimated according to the Braun-Blanquet scale (Mueller-Dombois and Ellenberg 1974). Average heights of the canopy, sub-canopy, shrub, and grass layers were estimated and the dominant species noted. Forty-seven sample points (including nine on termite mounds) were inventoried in Muaredzi and 50 sample points (including five on termite mounds) were inventoried in Nhanchururu. In addition, notes were taken at various other points at the two sites. The location of each sample point was entered into a global positioning system.

Data analyses

Hierarchical Cluster Analysis (HCA) using average linkage method (van Tongeren 1995) was performed on a matrix of 47 plots by 228 species for Muaredzi, and a matrix of 50 plots by 246 species for Nhanchururu, using species cover-abundance data. This was done to produce a classification of the vegetation based on floristic and structural similarities/dissimilarities among them. The HCA was performed using MINITAB version 13.1 statistical software (Minitab Inc. 2000). Detrended Correspondence Analysis (DCA) (ter Braak 1986, 1995; Gauch 1982), an indirect gradient analysis technique, was performed on species cover abundance data to elucidate relationships among the various plant associations and underlying environmental gradients. CANOCO Version 4 for Windows package (ter Braak 1988, 1991; ter Braak and Smilauer 1997) was used for this analysis. CANODRAW package, available in CANOCO, was used to calculate the Shannon diversity and richness indices (Ludwig and Reynolds 1988, Magurran 1988) for each inventoried site. The absolute richness values for each site were calculated as the total number of species recorded at the site.