Population size and structure of Grant’s gazelle and lesser kudu in Geralle National Park, Southeastern Ethiopia

Description of the study area

Geralle National Park (GNP), which was established in 2006 (Asefa, Mengesha & Aychew, 2017), is located in the south-eastern part of Ethiopia in the Somali National Regional State (SNRS) in Dawa administrative zone, Hudet woreda (district) (Fig. 1). The National Park is bounded to the north by the Liban woreda of the Guji zone (Oromia Region), to the west by the Arero woreda of Borena zone (Oromia Region), to the east by Mubarak Woreda of Dawa zone (SNRS), and southeast by Moyale woreda of the Dawa zone (SNRS). The park is located 709 km from Addis Ababa via the Nagelle Borena highway and 901 km via Moyale. The park is divided into west and east parts. The west part covers around 692 km² and the east part covers 1,042 km² including the adjacent area of the Dawa ecosystem (Heaven, 2014). West GNP, where the present study was conducted, is situated between 4^∘0′0″ to 4^∘30′0″N latitude and 39^∘30′ to 39^∘50′E longitude (Fig. 1).

The topography of the National Park is mostly characterized by low land and plains. The altitude ranges as low as 800 m above sea level on the banks of the Dawa River and as high as 1,380 m above sea level on top of the escarpments (Asefa, Mengesha & Aychew, 2017). The rainfall regime in GNP is bimodal, with a long rainy season between March and May with a peak in April, and a short rainy season between September and November, with a peak in October. The mean annual rainfall for the period 2005–2018 was 594.9 mm (NMA, 2021). The peak mean monthly rainfall was in April (132.5 mm) and October (121.0 mm). The least mean monthly rainfall was in January (8.9 mm) (NMA, 2021). The hottest months are January and February, and temperatures fluctuate between 28.3 °C and 29.1 °C. The mean annual minimum temperature was 12 °C. The mean monthly maximum temperature was 29.1 °C., while the minimum was 23.9 °C (NMA, 2021).

The most dominant habitat types in the park are: grassland, woodland, wooded grassland, and open bushland (Asefa, Mengesha & Aychew, 2017). The dominant vegetation type is Acacia-comiphora, Balanites aegyptiaca, Boswellia spp, Acacia mellifera, Acacia brevispica, Acacia oerfata, and Grewia spp are the dominant tree species in the park (Asefa, Mengesha & Aychew, 2017). About 42 mammalian species were recorded from the park (Heaven, 2014). Some of the most common mammals include, Beisa Oryx (Oryx beisa), Grant’s gazelle (Nanger granti), Gerenuk (Litocranius walleri), lesser kudu (Tragelaphus imberbis), cheetah (Acinonyx jubatus), lion (Panthera leo) and leopard (Panthera pardus). Moreover, endangered and critically endangered species such as African Elephant (Loxodonta africana), and African wild dog (Lycaon pictus) inhabit the national park. Two hundred species of birds have been identified in the park (Alemu, 2004; Heaven, 2014). Increasing human population around Geralle National Park and consequent expansion of settlement, and grazing are threatening the survival of species and their habitats (Asefa, Mengesha & Aychew, 2017).

Methods

Data collection

We laid transects according to the available habitat types namely; grassland, bushland, wooded grassland, and woodland. Transect lines were established based on the stratified habitat types in a systematic random sampling design and proportional to the size of each habitat type to minimize sampling bias and achieve representativeness. Accordingly, there were a total of 36 transect lines: five (5) in grassland, seven (7) in bushland, ten (10) in wooded-grassland, and fourteen (14) in woodland (Fig. 2). A transect line varied in length from 2.3 km to 6.8 km (totally 165.4 km transect line distance was determined). Adjacent transects were placed with a minimum of 1,500 m apart based on vegetation type, and transect lines were roughly parallel to each other and their ends were at least 500 m from the habitat edges. The 1500 m of space left between consecutive transects was used to avoid double counting. QGIS was used to plot selected transects using systematic random sampling onto the study area map (Fig. 2; QGIS Development Team, 2024). The same transects were used for surveys that were conducted during the wet and dry seasons. This method enabled an all-seasons survey that was uniform across all habitats.

Data collection was conducted during both the wet and dry seasons. The wet season data was collected from April to May in 2020, while the dry season data was collected from August to September in 2021. Surveys were carried out during the early morning hours (from 06:30 to 10:30 a.m.) and late in the afternoon (from 03:00 to 6:30 p.m.), when most mammals are believed to be active (Buckland et al., 2009). This schedule was maintained to ensure the accuracy and relevance of the data collected. We used Global Positioning System (GPS) (Garmin 60/78, Oregon, USA) and handheld bearing compass (Suunto KB-14/360RG, Vantaa, Finland) to navigate to each line transect by walking at a constant speed of 1.25 km/hr (Peres, 1999). The data collection was made through direct observations with the naked eye and aided by a 7 ×35 binoculars. A motorbike was used in this survey by riding slowly at speed of 20–30 km/hr (Keeping & Pelletier, 2014) in most habitat types and transects. A motorbike survey was used due to the plain topography and open habitats. Grassland, wooded grassland, and woodlands are relatively open and have more or less plain topography, so motorbike was used. However, out of seven transects in bushland habitat three transects were covered by thick bushes and it was not possible to transverse the transects using motorbike, hence those transects were walked on foot. Since transects were surveyed predominantly using motorbike, there were not as many variations derived from sampling method. Hence, the cumulative effect on animal behavior and detection functions remain more or less constant. Binoculars were used for proper sex and age identification. The individuals seen within less than 50 m of the nearby group were recorded as members of the same group (Hillman, 1987).

A total of six (6) people (a minimum of seven years of experienced scouts and degree-holder experts in wildlife and related disciplines) and a researcher were involved during the survey and transects nearby with no clear geographic boundaries were surveyed simultaneously. Two days of data collection training were provided for field assistants on how to operate the GPS, count animals, identify sex and age, and record data on datasheets. Per transect, three experienced people were assigned to each transect in the survey. The start and end of the coordinates of each transect were saved in GPS to ensure transects were not repeated during each counting session. Double counting of the same individual herd was avoided using easily recognizable features of an individual’s herd size (total numbers in the herd), age, distinct body deformations, cut tail and ear, and composition (Wilson & Delahay, 2001).

The process of determining age and sex in lesser kudu and Grant’s gazelles was conducted using a variety of physical characteristics. For the lesser kudu, physical characteristics such as body size and the presence or absence of horns were considered (notably, the female lesser kudu, or ewe, is smaller and lacks horns compared to the male, or ram). Both sexes have white markings on the throat and inner legs. In the case of Grant’s gazelles, similar physical characteristics were used for age and sex determination. The body size of Grant’s gazelles can provide an initial clue, with males generally being larger than females. Both sexes carry long, slender horns (Stuart & Stuart, 2006). Morphological development, coloration, and structural organs like the tail, and ears were used to determine the approximate age (Sutherland, 2006). Following the categorization proposed by Lewis & Wilson (1979), the age and sex composition of the species were recorded. This classification system comprises three primary age groups: juveniles, sub-adults, and adults. The juveniles represent the youngest group, while the sub-adults serve as an intermediate group. The adult category encompasses mature individuals. During the census, each individual was sexed as either male or female, and their age was assessed in line with above three age categories. However, it is noteworthy that the sex determination of juveniles and calves posed a significant challenge due to the difficulty in identifying their genital organs. This complexity often results in these groups being categorized without a specific sex assignment.

Data analysis

The data was organized into a data frame, encompassing variables such as stratum, area, perpendicular distance, transect length, cluster size, species, session, and season. The data manipulation and cleaning were performed using R software version 4.3.3 (R Core Team, 2024). For the analysis of population size, density, and model selection procedures, the distance sampling software version 7.5 (Thomas et al., 2010) supplemented by the MRDS package in R software 4.3.3 was used (Laake et al., 2023; Burt et al., 2014). Variance estimation was grounded on the principles outlined by (Buckland et al., 2001) employing the delta method with empirical variance. Each habitat type was analyzed as a separate stratum, and each species and season were analyzed using a data filtering method in a distance software analysis window (Thomas et al., 2010). Population density for each species was calculated as follows: (1)${\displaystyle \hat{D}=\frac{n}{2wL{\hat{P}}_a}}$ (Buckland et al., 2015; Buckland et al., 2001; Buckland et al., 1993) where, ${\hat{P}}_a$ = probability of detection, n number of observations, w = width of the transect, and L = transect distance. 0 < Pa < 1. Pa = μ/w where μ = ${\int }_0^w\left(gx\right)$)dx. assume that animals on the line are certain to be detected at g(0) = 1

Conventional distance sampling (CDS) is a widely used methodology for estimating the density or abundance of a population. Specifically, distance sampling method major outputs such as (i) detection probability, (ii) population density, and (iii) model selection criteria were used. For determining the detection probability of the two species the Multiple-observer Distance Sampling (MRDS) analysis engine were used. The MRDS engine is an R package for use primarily with double-platform line transect data, where the assumption of certain detection on the line can be relaxed (Laake & Borchers, 2004). Double-platform methods are widely used in both aerial and ground survey of mammals (Borchers & Burnham, 2004), but are potentially useful in many situations where objects at zero distance are difficult to detect. The MRDS analysis engine, which operates on a single observer model, was utilized with a constant parameter from the CDS engine. Despite the MRDS single observer model not introducing additional features compared to the CDS engine, it generates comprehensive summary estimates at both the global and stratum levels when a set of data records is used. These estimates are considered more insightful and detailed than those produced by the CDS engine (Thomas et al., 2010). Both grouped and un-grouped perpendicular distances were performed during model selection procedures to consider categorical and non-categorical data. The distance model selection procedure was between half-normal and hazard-rate key functions for grouped and ungrouped perpendicular distances in the MRDS engine with constant CDS parameters. The best-fitted model was selected according to Akaike’s Information Criterion (AIC) and Chi-square goodness test. Observation differences among habitat, season, and species were performed using two-way analysis of variance (ANOVA). Type III sum of squares was used to overcome unequal sample size designs. The Poisson regression model, which is recommended for analyzing discrete distributions (count data), was used (Tutz, 2011). In this model, transect distances, which vary in length, were used to find the effective strip width as shown in Eq. (2). This approach was employed to analyze the deviance table (ANOVA). We fitted quasi poisson regression to reduce overdispersion. The optimal Poisson regression model was identified using a stepwise selection process, which simplified the predictor variables in the full model (James et al., 2021). (2)${\displaystyle \mathrm{n}=e^{\alpha +{\beta }_1\cdot \text{species}+{\beta }_2\cdot \text{season}+{\beta }_3\cdot \text{habitat}+log\left(\text{Transect distances}\right)}}$

where α represents the intercept, β stands for the coefficient, and ‘n’ denotes the number of observations. The factors being considered are habitat type, season, and species. The offset is transect distances.

Density and abundance

During the dry and wet seasons 125 and 134 individuals of Grant’s-gazelle (Nanger granti) and 100 and 134 individuals of lesser kudu (Traglaphus imberbis), respectively, were recorded (Table 1). Species, habitat types and seasonal variations had significant and non-significant associations effect of Poisson regression model. Chi-square test analysis on habitat vs species (χ² = 397.71,  DF = 3, P-value < 0.001) and habitat vs season (χ² = 8.37 DF = 3, P-value = 0.039) showed significant associations. Species vs season (χ² = 6.80 DF = 3, P-value = 0.07) showed an insignificant association. Species (χ² = 14.3 DF = 3, P-value = 0.0024), habitat (χ² = 218.04 DF = 3, P-value < 0.0001) and season (χ² = 6.31 DF = 1, P-Value = 0.012) show significant associations. The wet season showed a higher record of observation than the dry season (P-value < 0.0001). In habitat comparison, there were significantly higher numbers of observations in grassland than in the three habitats (P-value <0.0001). bushland and wooded grassland have insignificant records (P-value = 0.3418), however, these two habitats showed a higher record than woodland habitat (P-value <0.0001). For pair wise multiple comparison, there was higher observation in grassland habitat in the wet than grassland in the dry season (P-value < 0.001).

There were significantly higher population records of Grant’s gazelle in the grassland than other habitat types (P-value < 0.001). Lesser kudu population record in bushland was significantly higher than woodland habitat (P-value = 0). Lesser kudu exhibited a higher population record than Grant’s gazelle in bushland habitat. Gazelle had a higher population observation record than lesser kudu in grassland habitats, whereas lesser kudu showed higher records in bushland, wooded-grassland, and woodland habitats (Table 1). The minimum population observations were 100 and 125 individuals for lesser kudu and Grant’s gazelle, respectively, during the dry season (Table 1). The maximum number of observations was 134 individuals for both species during the wet season.

The lowest AIC, ΔAIC (close to zero) and Chi-square tests (P-value > 0.05) showed that the hazard rate key function with un-equal interval group model was selected (Table 2). The P-value for the Chi-square tests for all species under two seasons was fitted (P-value > 0.05). P-value for Grant’s gazelle, and lesser kudu were 0.07 and 0.09 in the dry season and 0.17 and 0.16 in the wet season, respectively. The estimated detection probability (${\hat{P}}_a$) (percentage) for Grant’s gazelle varied between seasons; in the wet season the detection probability was ${\hat{P}}_a$ = 71 ± 9 Standard Error (SE), while in the dry season it increased to a maximum of ${\hat{P}}_a$ = 80 ± 5 (SE). For lesser kudu the detection probability was ${\hat{P}}_a$ = 77 ± 5 (SE) in the dry season and ${\hat{P}}_a$ = 81 ± 7 (SE) was recorded in the wet season (Table 2). Grant’s gazelle estimates of density and abundance coefficient variation in two habitat types were greater than 30% and less than 65%; however, the overall coefficient variation (CV%) was less than 35% (30%) during the dry season and greater than 35% (36%) during the wet season. Grant’s gazelle occurred in low densities in bushland and woodland habitats. Lesser kudu estimates of density and abundance coefficient variation (CV%) in three habitat types were greater than 25% and less than 45%; however, the overall coefficient variation was less than 30% (22%) for both seasons. The estimated overall total population size and density of each species in each habitat type of the species in the GNP during each season is shown in Table 3 below. The highest density of Grant’s gazelle and lesser kudu were recorded in grassland and bushland habitats, respectively (Table 3). The density of Grant’s gazelle in the grassland habitat during the dry season averaged 11.23 (95% CI [5.06–24.92]-; p < 0.001, DF = 10), whereas averaged 15.4 (95% CI [6.31–37.59]; p < 0.001, DF = 11) during the wet season (Table 3). The density estimate of lesser in the bushland habitat during the dry season averaged 2.37 (95% CI [1.1–5.11]; p < 0.001, DF = 15), whereas averaged 3.43 (95% CI [1.49–7.9]; p < 0.001, DF = 14) during the wet season (Table 3).

For both seasons, all observations beyond the 150 m perpendicular distance were discarded, which is the right truncation. Although individuals could still be observed well above 150 m in open habitats such as grasslands, identifications of individuals by age and sex classes were not apparent. As shown in Fig. 3, excessive detection occurs at 45 to 60 m and 85 to 105 m transect distances. For the wet season, the graph shows that there was no detection at 0 to 15 m perpendicular distance and less detection between 60 to 75 m and 90 m to 105 m (Fig. 4). For Grant’s gazelle in the dry season, the detection function during the dry season showed very low close to the transect and increased with an increase in distance and reaching its highest at a distance 80–100 m, while it recorded fewer and fewer observations as the distance gets longer at 150 m (Fig. 3). For the wet season, there were few observations beyond the 60 m to 80 m interval, which is a high observation (Fig. 4).

The detection function for kudu in the dry season shows that there was heaping at the interval 69 to 85 m and little detection at the interval 85 to 96 m perpendicular distance. However, most distance interval observations are close to the detection function curve (Fig. 5). In the wet season, there was more discrepancy at the interval 75 to 96 m and 96 to 110 m perpendicular distances (Fig. 6). The irregularity in the detection function curve was due to some topographic and vegetation dispersion irregularities. In some of the cases, valleys and hills and bush tickets were limiting detections even at the nearest detection distances.

Population structure

Age structure

Adult female comprised the highest proportions for both species during both dry and wet seasons (Fig. 5). Adult female comprised the highest female proportion for Grant’s gazelle during both dry (49.60%) and wet (41.04%) seasons. On the other hand, sub adult male comprised the least percent proportion during both dry (2.40%) and wet (7.46%) seasons (Fig. 7). Adult female comprised the highest female proportion of lesser kudu population during both dry (47.01%) and wet (52.00%) seasons and there was no record of young individuals during dry season (Fig. 7).