Habitat suitability modelling and implications for management of roan antelopes in Kenya

Abstract

The roan antelope is locally endangered in Kenya with <50 individuals remaining in Ruma National Park (RNP). An understanding of the habitat requirements is essential to formulate habitat management strategies for species recovery. This study identifies habitat and management features affecting roan distribution in different seasons using information-theoretic and multimodel inference (MMI) techniques. MMI averaged models were coupled with GIS data to develop habitat suitability maps. Results showed that roan habitat suitability is determined by different factors in different seasons but overall habitat was more important than management. Best predictors of roan incidence were wooded grassland, open grassland and water points. Water was more significant in the dry season whilst vegetation burned status, slope, soil type, distances to snares, park fence and security gates were important predictors at other seasons. Receiver operating characteristic (ROC) plots confirmed that the models fitted well to the data (AUC>0.9). Sustainable conservation of roans in RNP can be achieved via long-term habitat management including prescribed burning and construction of more water points and short-term urgent solutions to key management issues, especially control of poaching.R esum e L'antilope rouanne est localement en danger au Kenya o u il reste moins de 50 individus dans le Parc National de Ruma (RNP). Il est essentiel de bien comprendre ses pr ef erences en mati ere d'habitat pour concevoir des strat egies de gestion de l'habitat propices a la restauration de cette esp ece. Cette etude identifie les caract eristiques de l'habitat et de sa gestion qui influencent la distribution de la rouanne selon les saisons au moyen de techniques de th eorie de l'information et d'inf erence multi-mod eles (MMI). Des mod eles de MMI moyens ont et e coupl es a des donn ees SIG pour dresser des cartes du caract ere plus ou moins appropri e des habitats. Les r esultats ont montr e que la pertinence de l'habitat des rouannes est d etermin ee par diff erents facteurs selon les saisons, mais aussi que l'habitat en g en eral est plus important que sa gestion. Les meilleurs pr edicteurs de la pr esence de rouannes etaient des prairies bois ees, des prairies ouvertes et des points d'eau. L'eau etait plus importante en saison s eche alors que l' etat plus ou moins brûl e de la v eg etation, la pente et la nature des sols, la distance par rapport a des pi eges, des clôtures du parc et des portes de s ecurit e etaient les meilleurs pr edicteurs pendant les autres saisons. Des graphiques ROC (Caract eristiques du fonctionnement du r ecepteur) ont confirm e que les mod eles concordaient avec les donn ees (Aire sous la courbe : AUC>0.9). La conservation durable des rouannes dans le RNP peutêtre r ealis ee par une gestion de l'habitat a long terme, y compris des feux programm es et la construction de nouveaux points d'eau, et par la r esolution urgente des probl emes de gestion essentiels, sp ecialement le contrôle du braconnage.

Roan antelope Hippotragus equinus in Africa: a review of abundance, threats and ecology

“…, Martin , Kimanzi & Wanyingi , Kimanzi et al. ). Threats faced by various subpopulations of roan antelope, the majority of which were documented in southern Africa, include predation (McLoughlin & Owen‐Smith ), poaching (Kimanzi et al.…”

“…Roan antelope are a targeted source of meat in the Ivory Coast (Fischer ), and incidental poaching with snares is the primary cause of roan antelope decline in Ruma National Park, Kenya (Kimanzi & Wanyingi , Kimanzi et al. ).…”

Habitat suitability modelling for Hirola (Beatragus hunteri) in Tsavo East National Park, Kenya

“…Logistic regression was used to predict the presence or absence of hirola (Estes et al ., ; Kimanzi, Sanderson & Rushton, ; Sarhangzadeh et al ., ) using the equation: where the dependent variable ( y ) was coded as 1 for presence and 0 for absence while x 1 , x 2 , … x n were independent predictor distance raster maps, and β 0 , β 1 , … β n were logistic regression coefficients (Hosmer & Lemeshow, ; Sarhangzadeh et al ., ).…”

“…) which served as a pseudo‐absence hirola data as described by Greaves, Sanderson & Rushton (), Barbet‐Massin et al . () and (Kimanzi, Sanderson & Rushton, ).…”

“…The presence and absence data were split into two data sets: training and evaluation data (Sarhangzadeh et al ., ). Percentage of evaluation data was calculated as [1 + (p − 1) 1/2 ] −1 × 100 (Huberty, ; Kimanzi, Sanderson & Rushton, ), where p is the number of predictors. Consequently, 44 points were used in evaluation of the model which was selected randomly using QGIS (Lyon, ), and the remaining 102 points were used to develop the model.…”