Development of a habitat suitability model to determine the potential distribution of Klipspringer (Oreotragus Oreotragus subsp. Oreotragus) in Table Mountain National Park

Abstract

The klipspringer (Oreotragus oreotragus subsp. oreotragus) population became extinct on the Cape Peninsula in 1930. Being re-introduced into Table Mountain National Park (TMNP) in 1999 it became one of the species of special conservation concern to monitor in the Park. Most klipspringer territories are known by Park management but the distribution of all potentially suitable habitats for this species in the Park is not known. The main aim of this study is to produce a distribution range map that is representative of all potentially suitable habitats for the klipspringer within TMNP, through the use of a species distribution modelling tool. Since only presence data were available for this study, a popular presence-only modelling tool namely maximum entropy (MaxEnt) was used. The use of MaxEnt in species distribution modelling has become popular as it has proven to provide robust predictions of a species’ geographic distribution. Klipspringer occurrence data and five environmental variables namely altitude, slope, aspect, vegetation, and distance to urban edge were used as model input. Occurrence data were sourced through existing databases and employing a stratified random sampling technique of dividing the Park into different habitat subtypes to survey the Park for more klipspringer occurrences. These habitat subtypes consisted of a variety of vegetation communities or vegetation types and altitudinal and slope ranges available in the Park. Grid size for all the raster layers used was 10x10 m. Spatial filtering of one point per 100 m² grid was used to eliminate clumping of points. Six models were run at different regularisation multiplier (RM) values namely 0.25, 0.5, default (1), 2, 4 and 7. To assist in better understanding of the spatial extent of the occurrence data and the areas inhabited by the klipspringer, home range analyses were carried out. This was done through kernel density estimation in the Geospatial Modelling Environment (GME). All six bandwidth parameters in GME namely smoothed cross validation (SCV), biased cross-validation (BCV), a second BCV algorithm, plug-in estimator, least squares cross validation and the likelihood cross validation (CVh) were tested. The smoothed cross validation and likelihood cross validation bandwidth algorithms provided the best visual output of klipspringer home ranges and territories. Home range sizes from the SCV output ranged from about 3 – 11 ha across the study area, and home range size for the CVh output ranged from 0.6 – 2.5 ha. The output from the CVh algorithm was interpreted as territories rather than home ranges, as it is based on a univariate kernel unlike, the SCV algorithm that produces rotated bivariate kernels. iv The default regularisation multiplier of 1 provided the best probability distribution output, whilst values lower than the default tended to underestimate the prediction and those values higher than the default were tending towards overestimations. Response curves for the default RM also gave the most ecologically meaningful responses of the klipspringer to each environmental variable. Model evaluation in the form of area under the receiver operating characteristic curve (ROC AUC) showed that all models performed well. Therefore, the choice of the “best” model was based on the ability to provide ecological interpretation, on the shape of the response curve and the probability distribution maps. Consequently, the default RM model was considered the best, with an AUC score of 0.903. Altitude and vegetation contributed the most to suitable habitat and therefore indicates that klipspringer in the Park do prefer high altitudinal areas with the right vegetation to feed on. Suitable altitudinal ranges are from 400 m.a.s.l. and higher and ericaceous fynbos is the most preferred vegetation community. Slope, aspect and distance to urban edge played a less important role in suitable klipspringer habitat. The probability map and an additional binary map produced at the 10 percentile training logistic presence threshold showed that suitable habitat for the klipspringer occurs in all three sections of the Park in different proportions. These maps can be used by Park management to prioritise conservation efforts and future re-introductions.