The African spurred tortoise (Centrochelys sulcata) is one of the largest-bodied tortoise species (up to 100 kg) and has a wide and scattered distribution across the African Sahel (Branch 2008). Despite its wide distribution, free-ranging populations of African spurred tortoises are thought to be declining (Vetter 2005; Burney et al. 2012) and are listed as vulnerable by the International Union for Conservation of Nature (2018). It is likely that competition with cattle and other domestic animals, along with human-initiated seasonal fires, have reduced populations of African spurred tortoises. Their presence is negatively correlated with high cattle density and high incidence of fires at the regional scale (Petrozzi et al. 2017a). Centrochelys sulcata density (tortoises per hectare) is lower than that of most other terrestrial chelonians (i.e., median = 0.001 [Petrozzi et al. 2018] vs., for instance, 0.16 [Kinixys spekii], 3.9 [Testudo horsfieldi], or 0.7–2.7 [Gopherus polyphemus]; see review in Luiselli 2006), with the highest and lowest densities observed in Teneré-Termit region of Niger and in Burkina Faso, respectively (Petrozzi et al. 2018). In addition, the current distribution of African spurred tortoises is characterized by wide gaps, which are likely due to human influences (above) and species-specific preferences for sites with intermittent streams (called “kori” in the Sahel region) and stabilized dunes (Petrozzi et al. 2017b). Therefore, it appears that a combination of ecological factors and anthropogenic disturbances (e.g., agriculture, pastoralism, habitat degradation, and collection for the pet trade) have affected the location and size of African spurred tortoise populations in the Sahel (Vetter 2005; Petrozzi et al. 2017a, 2017b).

We lack information relating C. sulcata distributions to landscape features (besides kori and dunes) such as land-use, density of and type of vegetation, presence of roads or human settlements, or climate (e.g., rainfall quantity). For instance, it is unknown whether vegetation (i.e., the density of trees) affects C. sulcata in the same way that it affects the presence and site selection of other tortoise species (e.g., Del Vecchio et al. 2011 for the case of Testudo hermanni).

To ascertain whether African spurred tortoise distribution is associated with specific landscape and climatic variables, we statistically tested for the effects of rainfall, vegetation type and structure, land use, presence of roads, and presence of human settlements on C. sulcata presence at sites across the West African Sahel. Our goal was to increase understanding of the habitat requirements of this vulnerable species whose area of occupancy and extent of occurrence are poorly defined (Petrozzi et al. 2016).

Methods

Field Protocol. — We used survey data from 6 Sahel countries: Mauritania, Mali, Niger, Nigeria, Burkina Faso, and Chad, in 1994–1995, 2000, 2005, 2007, and 2013–2017. Our original field investigations were carried out mostly during the heaviest phase of the short-wet season, when above-ground activity of the spurred tortoise is at its maximum (Petrozzi et al. 2017b). For detailed description of the field methodology, see Petrozzi et al. (2016, 2017a, 2017b, 2018). Unfortunately, because of social tensions, political instability, and security issues (e.g., risk of terrorist attacks) in the studied countries, we were unable to systematically explore all the territories that would have been valuable to survey. For instance, our access to certain areas of Burkina Faso, Mali, and Niger was prohibited; thus, sampling was limited resulting in a sample size of n = 36 presence sites. Despite these limitations, ours is the most spatially extensive study ever attempted for this species. We used the Global Positioning System (GPS) to record the geographic coordinates (± 25 m) of all sites where free-living tortoises were encountered. Geographic coordinates of these locations are not presented in this article for conservation reasons.

We restricted our analyses to accurate, geo-referenced GPS C. sulcata location data that we collected ourselves or were collected by trusted sources (e.g., management personnel of protected areas, data repositories in environmental ministries, and from other governmental and nongovernmental organizations). For C. sulcata data collected by others, we verified that each GPS-recorded sighting referred to tortoises observed in the wild by inspecting associated photos or other voucher materials (e.g., shells or shell fragments). However, some records could refer to released or escaped captives (Minuth and Thieme 2004; Petrozzi et al. 2016); therefore, we tried as much as possible to explore whether each record (especially the indirect records, such as those from the databases of local institutions) referred to native tortoise populations by interviewing shepherds, cattle farmers, farmers, and hunters. We then removed from subsequent analyses all records (n = 17) that appeared to be surely or likely based on captive or released individuals.

Geographic Information System (GIS) Procedures. — We imported coordinates of our presence records into GIS software (QGIS2.18) saving them as a shape file, and then creating a minimum convex polygon shape file for all points. We created a buffer of 30 km around the most external presence sites of the convex polygon (Supplemental Fig. S1; all supplemental material is available at https://doi.org/10.2744/CCB-1302.1.s1) to increase its size and avoid any potential edge effect. Using the Random Points algorithm, we created a second shape file with random points (n = 250; Supplemental Fig. S1) inside the increased minimum convex polygon shape file area. Then, by using both the presence records and the random points, we created a single shapefile translating the coordinates from WGS 84 to UTM 31 N and created buffer areas with 1-km radii around each presence and random point.

We then uploaded the following raster data using the GIS software.

1. Tree cover raster (Hansen et al. 2013). This raster has a spatial resolution of 1 arc-second (∼ 30 m at the equator) per pixel, and represents the percentage of forest; it ranges between 0 and 100 per output grid cell (Supplemental Table S1).

2. Land cover raster (http://www.fao.org/geonetwork/srv/en/main.home). With a grid resolution of 1 × 1 km, this raster represents the characteristics of the land grouped in different categories (Supplemental Table S2).

3. Rainfall (Food and Agriculture Organization of the United Nations–United Nations Environment Pro-gramme 1984). This raster represents the average annual rainfall, and is expressed in terms of millimeters (mm) per year (Supplemental Table S1).

The spatial distribution of these variables within the increased minimum convex polygon is displayed in Supplemental Fig. S2. We extracted all the available information about these 3 variables within the buffer areas, both for the presence records and the random buffer areas, by using the zonal statistic tool of the above-mentioned GIS package. Concerning the observed (bio)physical cover on the land surface (Land cover—LCT), we analyzed the data by using also the LCT classification categories (Supplemental Table S2).

We analyzed the variable “distance from the nearest road (km)” using a shapefile obtained from the Center for International Earth Science Information Network (CIESIN) at Columbia University, and Information Technology Outreach Services (ITOS) from the University of Georgia (2013). We also used the variable “distance from the nearest human settlement (km)” using the CIESIN et al. (2017) shapefile. We calculated the linear distances of presence and random points from settlements and roads using the “QGIS NNJoin” plugin (Dharmawan and Farda 2017). The spatial distribution of roads across the increased minimum convex polygon area is given in Supplemental Fig. S2.

Statistical Analyses. — We tested all variables for normality and homoscedasticity (by Shapiro-Wilk W test) before applying any test, and we used nonparametric tests when their distribution was nonnormal (p < 0.05). We used Mann-Whitney U-tests to compare 1) the average distances from village, 2) the difference in the linear distance from the nearest road, 3) the average tree-density categories, and 4) the rainfall class, of the various presence vs. random points. We used χ² tests to analyze the differences between presence and random sites in terms of 1) frequency of the Corine Land Cover (CLC) categories (http://www.fao.org/geonetwork/srv/en/main.home), and 2) frequency of the sites with roads in the surroundings. In the case of CLC analysis, we first excluded all the zero result categories to avoid violating convention for χ² frequency cell minima.

Means are presented with ± 1 standard deviation. All statistical analyses were performed with Predictive Analytics SoftWare version 11.0 software, with alpha set at 5%.

RESULTS

Thirty-six presence records of free-ranging African spurred tortoises (5 from Burkina Faso, 12 from Mali, 6 from Mauritania, 8 from Niger, 1 from Nigeria, and 4 from Chad) were retained for analyses and compared with 250 random sites. Presence sites differed significantly from random sites for 3 variables: CLC, distance from human settlements, and distance from the nearest road, of the presence sites.

Concerning CLC, the presence sites had significantly higher frequencies of C12 (Shrub cover, Deciduous) and C13 (High Herbaceous Cover) categories, and significantly lower frequency of C19 (Bare areas), than random points (χ² = 16.5, df = 6, p < 0.01; Fig. 1).

View Figure

• Download as Powerpoint
• Download Figure

The average distance from human settlements of presence sites (x̄ = 71,317.2 ± 96,762.9 m, n = 36) was significantly greater than that of random points (x̄ = 90,004 ± 85,276.1 m, n = 36; Mann-Whitney U-test, z = –2.22, p = 0.023). In addition, the frequency distribution of the presence records in relation to the range of distances from villages was significantly nonnormal (Shapiro-Wilk W = 0.688, p < 0.001), with the great majority of the presence records being situated at a distance of > 10 km from the closest human settlement (Fig. 2).

View Figure

• Download as Powerpoint
• Download Figure

The distance from the nearest road of the presence sites (x̄ = 16,630.7 ± 43,249 m, n = 36) was significantly lower (Mann-Whitney U-test, z = –2.09, p = 0.038) than that of random points (x̄ = 27,600.1 ± 47,746.7 m, n = 250).

None of the other variables were significantly different between presence and random sites. Frequency of the sites with roads in the surroundings did not differ between presence and random points (χ² = 0.77, df = 1, p = 0.678). In addition, annual rainfall was similar between presence sites (x̄ = 14.97 ± 8.84 cm; n = 36) and random sites (x̄ = 12.65 ± 9.3 cm; n = 250; Mann-Whitney U-test, z = –1.61, p = 0.108), and the same was true for the average tree density of presence vs. random points (respectively, x̄ = 1.18 ± 3.43 vs. x̄ = 0.58 ± 2.4; Mann-Whitney U-test, z = –1.83, p = 0.068).

DISCUSSION

Among the independent variables studied in this article, our statistical analyses showed that only CLC, the distance from the nearest village of the presence sites, and the distance from the nearest road, were significantly different from those of random sites. Overall, the fact that frequencies of CLC categories of presence sites differed significantly from those of random sites suggests that C. sulcata is not a generalist species, but that it inhabits specific areas with distinctive landscape and vegetation characteristics. Thus, our analyses do not support Vetter's (2005) opinion that this species is a habitat generalist in the Sahel. Contingency table analysis also showed that the semideciduous shrublands and high herbaceous areas were positive for the presence of the tortoise, compared with other CLC categories that are widespread in the region (i.e., cultivated and managed areas and bare areas). The significant correlation between trees–shrubs and tortoise presence mirrors the anecdotal information suggesting that it inhabits areas with Sahelian ephemeral herbaceous plants with an exclusively wet season phenology (Cenchrus biflorus prairies; Scherman and Riveros 1989) and scattered trees (e.g., Acacia spp., Adansonia digitata, and Grewia bicolor; Vetter 2005). The statistically significant avoidance of bare areas (called “zipele” in the region) by spurred tortoises also mirrors small-scale habitat data provided by Petrozzi et al. (2017b). Previous literature also suggested that these tortoises also avoid cultivated areas in general (Petrozzi et al. 2017b).

Our study also revealed that the distance from the human settlements was a good predictor of tortoise presence. Indeed, not only the great majority of our field records occurred > 10 km away from the nearest village (with only 6 presence sites being situated at ≈ 2 km distance from villages), but also the mean distance from human settlements of presence sites was statistically greater than random. These findings suggest that African spurred tortoises presently inhabit predominantly remote sites where the human activities are greatly limited. It is highly unlikely that the records from such isolated sites refer to nonnative tortoise individuals. On the other side, we would suggest that future studies should be cautious concerning records of C. sulcata near urban settlements because these may possibly refer to released or escaped captives. Indeed, it is unlikely that these tortoises still survive in human-populated areas because 1) C. sulcata certainly require large home ranges of suitable habitat (e.g., radiotelemetric studies on herbivorous tortoises from arid environments, with ecology similar to C. sulcata but of smaller body size, revealed average short-term home ranges of 10–26 ha; Hailey and Coulson 1996; Franks et al. 2011), and 2) C. sulcata represent a valuable target for human hunters. Indeed, their meat is considered a local delicacy by some Sahelian human populations (Petrozzi et al. 2017a, 2018), and C. sulcata is also heavily traded for export to distant global pet markets as a valuable income source for the poor communities of the Sahel.

Our study revealed that the distance from the nearest road was lower than expected in presence sites. This statistical evidence might appear in contradiction with the previous considerations on the distance from human settlements. However, we think that this is an outcome of the extreme difficulty in exploring very remote sites (i.e., without access roads) in this semidesert region where movements of researchers are heavily constrained by safety concerns (presence of Islamic terrorist groups, particularly in the remote areas). Thus, we think that the positive effect of road distance with presence sites was merely a proxy for ease of sighting. If our reasoning is correct, more presence sites should be expected if remote areas, far away from roads, can be further explored.

In conclusion, more high-quality occurrence records are needed to develop a good understanding of this species' habitat needs in the western Sahel. We would argue for increased work to collect more distribution data, with special attention given to evidence indicating whether populations are native or introduced.