Predicting the habitat usage of African black rhinoceros (Diceros bicornis) using random forest models Lucy Lush 1,2 *, Martin Mulama 3 and Martin Jones 2 1 Centre for Environmental and Marine Sciences, University of Hull, Scarborough campus, Filey Road, YO11 3AZ., Scarborough, U.K., 2 Division of Biology and Conservation Ecology, Faculty of Science & Engineering, Manchester Metropolitan University, John Dalton Building, Chester Street, Manchester, M1 5GD, U.K. and 3 Ol Pejeta Conservancy, Private Bag, Nanyuki, 10400, Kenya Abstract Species distribution models are often used in ecology to ascertain relationships between environmental variables and species presence. Modelling to understand this rela- tionship can be used to aid conservation management strategies. In this paper, we applied the random forest classification method to predict habitat used by black rhino for browsing. The random forest model was created using detailed habitat data collected from Ol Pejeta Conservancy in Kenya. Variables from plots where rhino had been present were compared to those not used by rhino. Independent data were used to test the predictive accuracy of the rules generated. The model performed well with the independent test data, correctly classifying 69% of the sampling plots where black rhino were present. Important habitat features that affected rhino presence were browse availability and density of vegetation, with Vachellia drepanolobium (formerly Acacia) and Euclea divinorum being important components. The analysis also highlighted areas of potential high browse pressure, which should be the focus of continued monitoring and management. Key words: black rhino, browsing, habitat, random forest, savannah Resume En ecologie, on utilise souvent des modeles de distribution des especes pour etablir les relations entre des variables environnementales et la presence d’especes. Pour com- prendre cette relation, la modelisation peut ^ etre utilisee pour aider les strategies de gestion de la conservation. Dans cet article, nous avons applique la methode de classement par for^ et aleatoire pour predire quel habitat frequente le rhino noir pour se nourrir. Le modele de classement par for^ et aleatoire a ete cree en utilisant des donnees detaillees sur l’habitat, collectees dans l’Aire de conservation d’Ol Pejeta, au Kenya. Les variables de parcelles ou le rhino etait present furent comparees a celles d’autres, non frequentees par le rhino. Des donnees independantes ont permis de tester la justesse predictive des regles generees. Le modele donnait de bons resultats avec les donnees independantes du test, classant correct- ement 69 % des parcelles echantillons ou le rhino etait present. D’importantes caracteristiques de l’habitat qui affectaient la presence du rhino etaient la disponibilite de la nourriture adequate et la densite de la vegetation, dont Vachellia drepanolobium (anc. Acacia) et Euclea divinorum sont des elements importants. L’analyse a aussi mis en lumiere les zones ou la pression de la consommation des rhinos risque d’^ etre elevee, ce qui devrait ^ etre le point focal de la poursuite du suivi et de la gestion. Introduction Modelling to understand the presence of animals based on habitat data has been utilized extensively by ecologists to aid conservation management strategies (Manel, Williams & Ormerod, 2001; Stockwell & Peterson, 2002). However, errors in the accuracy of predicting species presence have been found in various studies, resulting in inappropriate management strategies (Manel, Williams & Ormerod, 2001; Loiselle et al., 2003; Araujo & Guisan, 2006). The ability to accurately predict suitable habitat for endangered species becomes more important as habitat fragmentation increases, and suitable habitat for many taxa is dramat- ically reduced. The techniques most often used to model species distribution are artificial neural networks, discriminant *Correspondence: E-mail: llush@hotmail.co.uk © 2015 John Wiley & Sons Ltd, Afr. J. Ecol. 1