Study area
Maze National Park is located at (6°25′N, 37°14′E) in southern Ethiopia (see Fig. 1). The Park covers an area of 175 km2 and was established in 2005 to conserve the rare and endangered Swayne’s hartebeest, which is considered a flagship species for the Park. The elevation of the study area ranges between 900–1,300 m asl. It is semi-arid and drought prone area with low and erratic rainfall (mean annual rainfall is below 800 mm) with high mean monthly temperature not less than 30°C. The Park has sufficient water sources for wildlife. The Maze River and several small tributaries, such as Daho, Lemasea and Domba flow throughout the year in the Park.
Maze National Park has a variety of habitat types, including riverine forests, plain grassland habitats with scattered trees (hereafter called grassland), steep bushland habitat above 15° slope (hereafter sloppy bushland habitat, see Additional file 1), plain bushland habitat, riverine forest, rugged bushland habitat with small valleys and neighborhood agricultural land (Fig. 1). The Park is surrounded by mountains, agricultural land and communal grazing lands. The grasslands are primarily dominated by annual grass species, such as Exotheca abyssinca, Heteropogon contortus, Loudentia spp., Setaria incrassate, and Hyparrhenia filipendula with scattered woody plants such as Combretum terminalia. Burning of the grasslands has occurred by the wildlife managers since the Park was established.
Swayne’s hartebeest sampling design
The Park was initially divided into 10 blocks using features such as roads, rivers, vegetation cover and valleys for a total count of Swayne’s hartebeest in each of habitat types and burned/unburned grassland patches. In each block, the habitat types and burned/unburned grassland patches were demarcated by using GPS within approximately 30 m accuracy and the extent was estimated using ArcGIS 10.3. In each block, we established permanent parallel transect lines spaced approximately 150–200 m apart. In the plains areas (i.e. open grassland and plain bushland areas), 37 transect lines were spaced at 200 m gap, whereas in the forest and rugged bushland areas where observation from distance was impossible, 15 transect lines were spaced 150 m apart. The length of transects varied due to the difference in size of each habitat types with average length of 5.9 km (±1.5 SD). In each transect, we randomly established 4 m × 5 m plots for Swayne’s hartebeest pellet presence/absence detection. We established plots systematically along transect lines at every 100 m regular intervals (thus, the total is 10 plots per 1 km). A total of 400 plots in the grassland, 100 plots in the plain bushland, 119 plots in the sloppy bushland, 191 plots in the rugged bushland, 148 plots in the riverine forest habitat and 44 plots in the agricultural land adjacent to the Park boundary were permanently established. The GPS coordinates and habitat types were recorded at each plot.
Swayne’s hartebeest habitat selection
The general habitat use of Swayne’s hartebeest from the available six habitat types were conducted for one year (i.e. from December 2016 – November 2017). Since the grass height varied before and after burning the grassland habitat, we divided the dry season into early dry season (before burning) and dry season (after burning). During the dry season (i.e. from December – May), we counted the pellet samples across the 1,002 plots established in the whole Park. Pellet-groups that were more than 50 cm apart in a plot were recorded as pellet from different individuals. Freshness of pellet samples also gave clues for being defected from different individuals. We visited each plot for an average of 36 times during the dry season. After a pellet-group was detected, it was removed from each plot to avoid redetecting during the subsequent surveys.
In the wet seasons (i.e. from June – August) and early dry (i.e. from September – November), we used direct observation of Swayne’s hartebeest along transect lines as pellet sampling was difficult due to dense habitat cover. During both seasons, habitat use of the Swayne’s hartebeest was estimated through transect counting aided with 10x42 binoculars. Whenever the Swayne’s hartebeests were observed, habitat types and abundance of the Swayne’s hartebeests were recorded [39]. We surveyed each transect 12 times during each season, and to avoid double detections of individuals, all transect lines of a block were surveyed at the same time. The surveys were carried out at early morning from 6:00 – 10:00 a.m. and late evening from 3:00 – 6:00 p.m. when Swayne’s hartebeests were active [66].
Because Swayne’s hartebeests were found in concentrated in three patches in grassland habitat during the wet season, we delineated the area by using GPS coordinates with 30 m intervals resulting 0.7 km2, 2.3 km2, and 2.5 km2 (see Additional file 2). During the early dry season, we also found the Swayne’s hartebeests in the peripheral part of the park leaving the three patches used during the wet season and covering 3.4 km2, 4.7 km2 and 5.3 km2 areas (see Additional file 2). The density was then derived by dividing the population estimate of the Swayne’s hartebeest during the transect count to the area where they found in the wet and early dry season.
Swayne’s hartebeest grass height preferences
To estimate average grass height in the grassland habitat during each season, grass heights were measured for 464, 193 and 133 central points of random plots of 1 m2 area during the dry, wet and early dry seasons, respectively. The average grass height was varied across seasons in the Park. From the randomly measured grass heights, overall grass height for the survey year was 56.8±60.4 cm (mean ± SD); while for dry, wet and early dry season was 32±39.9 cm, 70.2±51 and 121.7±76 cm, respectively. Based on this estimate, we subjectively categorized the grass heights as below 30 cm, 31-50 cm, 51-100 cm, and above 100 cm.
During the three seasons, the grazing events of Swayne’s hartebeests were recorded to determine the grass height preferred by Swayne’s hartebeest by walking on the transect lines established in the grassland areas. The surveys were carried out for 5–8 days in every month for one year (i.e., from December 2016 – November 2017). Whenever an individual or a herd was observed on the transect walk within 150 m of either side of a transect line (i.e., 300 m width) for open habitats (i.e., grassland, plain bushland, sloppy bushland and agricultural land), while within 100 m (i.e., 200 m width) for riverine forest and rugged bushland habitats, the total number of Swayne’s hartebeests was recorded. The first grazing Swayne’s hartebeest seen was chosen as a focal animal and its feeding location was identified using the nearby landmarks like trees or bushes. It was then displaced and fresh bites were identified at the site. Fresh bites were identified by the white coloration at the bite, whereas old bites turn brown [39]. Once the bites were identified, a 1 m2 quadrat was placed over the grass patch. Within each quadrat, heights of the preferred grasses by Swayne’s hartebeest were measured, but only those escaped from fresh grazing during the observation time.
Effect of fire on Swayne’s hartebeest habitat use
The Maze National Park management conducted controlled burning on some parts of the grassland habitat at the end of the wet season every year (mostly from October – November, depending on when the rain ends). Only some portion of the grassland habitat is burned in every year. Burning practice in the Park is mainly maintained by the Park managers with scheduled time in a year for herbivores use. However, in some places mostly at the periphery the local farmers also conduct burning. During this study period, the burning time was end of November. During the survey year, 21.4 km2 of the grassland area was burned while 30.2 km2 remained unburned. In both habitat types, we carried out 36 times transect count (a transect might cross both grassland types) from the first date of the burning (i.e. from the beginning of December – to mid-May and count the number of Swayne’s hartebeest individuals). In both grassland areas, we counted the Swayne’s hartebeests twice (two days) every week to record how long Swayne’s hartebeests were attracted in those areas. We summed the number of observed individuals for each surveying days in the burned and unburned grassland areas, separately. Counting was conducted in the morning 6:00–10:00 a.m. and late evening from 3:00–6:00 p.m. [66].
Data analysis
General habitat use
We used Ivlev’s selectivity calculations as a measure of relative habitat selection of Swayne’s hartebeest among the different habitat types using pellet presence data. Following [39], we used the equation Ei = (ri – ni) / (ri+ni) where ri is the proportion pellet detected in each habitat types within the survey period and ni is the proportion of plots in each habitat types during the surveying period available from the total area represented by the survey period.
We used linear mixed effect model from the package lme4 [67] to evaluate the relationship between density of Swayne’s hartebeest pellet (response variable) and habitat types during the dry season. We also used linear mixed effect model to evaluate the relationship between density of Swayne’s hartebeest (response variable) and time (i.e. Julian date as explanatory variable) during early dry and wet seasons, separately. We used generalized a linear model to estimate the relationship between grass height (response variable) and Julian date (explanatory variable) for one year. We used generalized linear mixed model for Swayne’s hartebeest seasonal grass height preference using density of Swayne’s hartebeest as a response variable with season (at three levels: wet, early dry and dry) and grass height as predictor variables. Block and transects were used as random factors to account for variations among areas and transects for the above models [68]. We also used generalized linear mixed model to estimate Swayne’s hartebeest abundance (response variable) in relation to burning (categorical variable at two levels: burned and unburned), and days since burning as predictor variables. Block was used as random factor to account for variations among areas [68]. We checked residuals and all the models met the assumption of normality. All analyses were done in R version 3.5.1 [69].