Study site
Our study was conducted at the Yankari Game Reserve (longitude 9°45'N latitude 10°30’E, Fig. 1), located ~ 100 km southeast of Bauchi town in east-central Nigeria. Yankari Game Reserve (here after referred to as Yankari) covers an area of 2244 km² consisting primarily of Sudan Savanna vegetation. The reserve has experienced several proprietary transitions that have likely impacted management goals, approaches, and expectations. First created in 1957 as the Bauchi Native Authority Forest Reserve by the then Northern Nigerian colonial administration, the area received a national park status and was managed by the Federal Government of Nigeria through its National Park Service from 1991–2006. In 2006, ownership was handed over to the Bauchi State Government to increase local revenue and maintain cultural heritage. Lying within the West Sudanian zone of Africa, upland areas in the reserve comprise primarily of grasses and thick bushes interspaced by savanna woodland. Lowland habitat is mainly floodplains, woody vegetation, and a well-kept gallery forest along the river Gagi (Geerling 1973; Abdullahi et al. 2009). Average rainfall is ~ 900 mm annually occurring between April and October with peak records typically in August. Prevailing winds are primarily from the southwest with temperatures ranging between 18°C to 33°C during the wet season and as much as 45°C in the dry season (Green and Amance 1987). Having a rich fauna diversity that includes scores of charismatic native species e.g., West African Lion (Panthera leo leo), African Buffalo (Syncerus caffer brachyceros), Hippopotamus (Hippopotamus amphibius), Roan Antelope (Hippotragus equinus), and Hartebeest (Alcelaphus buselaphus major) (Afolayan and Ajayi 1980); Bergl et al. 2011; (Omotoriogun et al. 2011), Yankari is the prime destination for ecotourists in Nigeria - a top priority of the current management (Olokesusi 1990).
Yankari is an ‘island’ without a buffer zone as it is surrounded by many villages just outside the boundary of the reserve (Fada 2015). These villages are populated by farmers and herders whose impact on the reserve includes poaching, lopping of tree species, and illegal livestock grazing (Metilelu et al. 2022; Fada 2015; Bergl et al. 2011). With no clear fire management plan and poorly enforced fire suppression policy, incidental fires are common occurrences in the reserve, often with severe consequences. Many of these fires are attributed to nomadic herders from the Fulani tribe of Northern Nigeria and are prevalent during the dry season (November – April) with most fires occurring between December and January (Fada 2015). Although there are anecdotal indications that Yankari game rangers occasionally set fires early in the dry season to improve foliage for herbivores (Fada 2015), there are no evidence that these burns are planned or targeted at specific areas.
Bird Capture Technique and Radiotelemetry
We trapped Stone Partridge during the dry season (November–April), which also corresponds to the fire season in Yankari. We coupled two trapping approaches – walk-in trap (Stoddard 1931); and leg nooses to maximize capture rates (Fig. S1 - Supplemental materials). Indeed, leg nooses have been shown to be effective in capturing ground-dwelling birds with a reduced chance of stress myopathy, injuries, and death (Chiozzi et al. 2015; Ponjoan et al. 2008). Nooses used in our study were lightweight, inexpensive, easy to assemble, and virtually invisible to most birds (Fig. S2 - Supplemental materials). Trapping occurred at 7 locations in Yankari (Fig. 1) selected based on cues such as birds’ fecal dropping, footprint, and repeated partridge sightings. Each site was pre-baited with Acha (Digitaria exilis) grains two weeks prior to the commencement of trapping and re-baited 2 to 3 times a week thereafter. We preferred Acha grains over more popular grains such as maize, guinea corn, and groundnut to reduce the chance of attracting non-target species such as baboons, hornbills, squirrels, and larger ungulates that prefer these grains. At all locations, trapping was initially done with walk-in traps but later replaced with leg noose traps after approximately four weeks of poor captures. When used, walk-in traps were camouflaged with dry branches and leaves to provide concealment and shelter for trapped birds. All traps were assembled daily before sunrise (6:00 am) and dissembled at 5:00 pm during the final check of the day to reduce the chance of trapping non-target species and leaving animals in a trap for long periods. We inspected traps at 2-hour intervals to minimize stress and predation risk for captured birds. Entrapped birds were weighed, aged, and tagged with a 6-g necklace-style VHF radio transmitter (Advanced Telemetry Systems, Isanti, MN, USA). In all cases, transmitters were ≤ 4% of the bird’s body mass to ensure that life history processes of tagged individuals are not altered by transmitter weight. Our trapping and handling protocol was approved by the Southeast Missouri State University Institutional Animal Care and Use Committee (approval number 202202-002).
We relocated radio-tagged birds 24 hours after tagging and 4–5 times a week upon release. Relocation was achieved using a hand-held 3-element folding Yagi antenna and an ATS receiver (Model R410, Advanced Telemetry Systems, Inc., Isanti, MN). To ensure homing, we triangulated VHF signals at several separate angles to determine an individual’s location. Triangulation was used to approach within 10–5 m of a radio-marked individual (White and Garrott 1990) and coordinates of sighted individuals were collected using a handheld Garmin Oregon® 750 GPS unit (Garmin, Ltd., Olathe, KS). Coordinates were either collected at locations of first detection prior to flushing or at locations of feces and depressions left by loafing individuals. Tacking was carried out at different hours of the day and rotated among tagged individuals to capture variation in activity patterns and ensure an even distribution of time at which we tacked each individual. Bird relocation continued until transmitters failed, fell off, bird was killed by predators, or when we could not detect radio signals for over one week.
Vegetation measurement
We sampled environmental covariates based on structural vegetation and fire complexities at Yankari. To do this, we randomly selected three relocation points (used points) per individual and measured vegetation and fire-related covariates at two spatial extents. At the first and larger extent, we established a 20-m2 plot centered on a used point to assess prevailing environmental covariates that could reflect Stone Partridge’s broad-scale habitat in this landscape. Covariates collected here included the number, height, and diameter of trees; number, height, and diameter of tussocks; height and number of shrubs; rock cover; time since last fire; and an index of fire severity obtained by observing physical elements of vegetation after fire (Luke and McArthur 1978). Each plot was assigned a fire return interval of 1, 2, or > 2 years depending on the last time it experienced fire. In rare cases where a plot encapsulates different fire return intervals, the year with the most (> 80%) burnt patch was inferred for the plot. Within each 20-m2 plot, we randomly placed four 2-m2 plots to measure fine-scale selection covariates to include ocular measurements for ground and litter cover to the nearest 5%, grass height measured with a centimeter calibrated vegetation stick. The choice of measured covariates at both spatial extents (2-m2 and 20-m2 plots) was guided by ecological literature and management plans on Galliformes habitat requirements. We sampled for habitat availability by pairing each used location with 3 non-used (available) locations selected randomly from 4 cardinal directions of the used location using a compass bearing. We specified 100-m between used and available locations to minimize the chance of available points falling within the immediate influence of used locations. At each available location, we collected the same suite of environmental covariates measured in used locations.
Fire severity and frequency
Knowledge of fire history and distribution was based on observations made during our study, information obtained from the Yankari management, and data from two researchers (Dr. Da’an and Dr. Lontong) with long-term knowledge of fire histories in Yankari. Notwithstanding, we verified all fire records by examining fire scars on the back of trees, woody debris, as well as grass and shrub remains. Because records of fire history in reserve were sometimes sketchy, we opted to minimize the change of misclassifying severity index for burn patches by including only fires that occurred within the two years preceding our study. We categorized a patch (i.e., 20-m2 plot) as a “high severity burnt” if ≥ 50% of the patch had severe fire scares, loss of vegetation, or had woody debris, otherwise, we classified it as a “low severity burnt”. Patches with no evidence of fire within the two years preceding our study were classified as “unburnt”. Relatedly, we classified fire return intervals into three categories – 1-year, 2-years, or > 2-years that can be easily varied without compromising data quality.
Statistical Analysis
We recorded environmental covariates at 60 used locations (three locations per bird) and 180 available (three per each used location) to fit a third-order Resource Selection Function (RSF) models (Johnson 1980) that assessed how partridges will select habitat variables relative to their availability (Manly et al. 2002; Boyce et al. 2002). We used point-based RSFs in a Generalized Linear Model framework to evaluate Stone Partridge habitat choices (Manly et al. 2002; Boyce et al. 2002; Lautenbach et al. 2021) at two spatial extents.
At the first extent, we fitted GLM models with a logit link function (Hilbe 2010) to identify fine-scale covariates that may drive partridge space use. Our response variable was a binary outcome representing opposite ends of used and available habitats within a partridge foraging range. “Used points” were coded “1” in our model and referred to a Stone Partridge GPS location in space, whereas “available points” were coded “0” and represented locations that were theoretically available for selection by the individual utilizing the GPS location at a particular time. At this level, we fitted three models: 1) a global model containing all explanatory variables (i.e., ground cover, litter cover, and grass height); 2) a reduced model containing litter cover and ground cover, and 3) an intercept-only model. Grass height was not statistically significant (at 0.05 level of significance) in the global model; hence, it was removed resulting in a reduced model. To assess the goodness-of-fit of the three fitted models and evaluate the contribution of subset parameters, we performed the likelihood ratio test (Dobson. and Barnett. 2018) comparing the full model to the null model.
We followed a similar approach to evaluate the contributions of 9 environmental and fire-related covariates (number of trees, tree height, diameter of trees, number of tussocks, normalized tussock height, diameter of tussocks, number of shrubs, shrub height, diameter of shrubs, rock presence, time since fire, and fire severity) to habitat selection by partridges at the broad spatial extent. Prior to modeling, we followed a two-step approach to test for multicollinearity among explanatory variables at the two spatial extents of our analysis explained above. First, we conducted a pairwise correlation analysis using the Pearson Product Moment (PPM) correlation coefficient and removed one colinear variable (usually the least biologically important variable) when PPM coefficient was ≥ |7|. For example, the number of tussocks correlated significantly with the diameter of tussocks, hence we dropped the diameter of tussocks from our analysis. Since three or more variables within a model can be correlated and a pairwise test typically tests correlations between two variables, we subjected the remaining variables to a variance inflation factor (VIF) test that checks for the independence of each variable in the model. Typically, a VIF value > 5 suggests severe collinearity and redundancy of variables in a model (Craney and Surles 2002). In our case, all explanatory variables in the VIF test were less than 2.5 suggesting that little or no collinearity existed among explanatory variables. At both spatial extents of our analysis, competing models were merely simplifications of the global model. We compared and selected models using an information-theoretical approach (Akaike Information Criterion - AIC) that measures the degree of information lost in each model across a given set of candidate models (Burnham and Anderson 2002). We interpreted the importance of each explanatory variable based on the direction of estimated coefficient, effect size, and uncertainty (using 95% confidence intervals). We considered variables with 95% confidence intervals overlapping zero to be insufficient in explaining the variation in the response variable.
To evaluate how partridges utilized habitat patches across different fire return intervals, we collapsed all used points by the month that a survey was conducted. Further, we conducted a one-way analysis of variance (ANOVA) with time since the last fire as a response variable and used points as explanatory variables. We performed all statistical analyses using R Statistical Software version 4.3.0.