Informations et ressources scientifiques
sur le développement des zones arides et semi-arides

Accueil du site → Master → Finlande → Species distribution models explaining human-wildlife conflicts in Taita Taveta County, Kenya

University of Helsinki (2017)

Species distribution models explaining human-wildlife conflicts in Taita Taveta County, Kenya

Äärilä, Sakari

Titre : Species distribution models explaining human-wildlife conflicts in Taita Taveta County, Kenya

Auteur : Äärilä, Sakari

Université de soutenance : University of Helsinki,

Grade : Master 2017

Résumé
Human-wildlife conflict (HWC) is an incidence where wildlife’s needs become incompatible with those of human populations, with costs both to humans and wild animals. As such, it is a global challenge that is seen impossible to eradicate. People affected most by HWC are the already vulnerable communities in the Global South. Loss of household member, crops or livestock can have dire impact on already stressed food security. Conflicts may result in antipathy towards the animals, which also complicates wildlife conservation. For these reasons, research on HWC is needed.
In this thesis, species distribution modelling (SDM) approaches are used to examine and predict HWCs in Taita Taveta County in Southern Kenya. Data for SDM was collected in February-September 2016 from compensation forms filed by victims of HWC and it consists of all the reported conflicts in 2014 and 2015. Temporal distribution of HWCs is also examined through analysis made with another dataset obtained from Kenya Wildlife Service including reported HWCs between 1990–2016. The study focuses on conflicts involving elephants, lions, leopards, hyenas and cheetahs.Species distribution modelling combines ecological theory and mathematical methods basing their results on the relationship between species and environment. This study uses biomod2 package in R, which includes 10 state-of-the-art modelling techniques : Generalized linear models (GLM), Generalized Additive Models (GAM), Generalized Boosted Models (GAM), Random Forest (RF), MaxEnt, Multiple Adaptive Regression Splines (MARS), Artificial Neural Networks (ANN), Classification Tree Analysis (CTA), Flexible Discriminant Analysis (FDA), Surface Range Envelope (SRE). Model prediction accuracy was estimated with True Skill Statistic (TSS) and Area Under the Curve (AUC). Most of the models reached moderate to good accuracy. Only human-cheetah conflicts were left out of the final analysis due to poor prediction accuracy.Explanatory variables used in the final models were : distance to protected area, annual average precipitation, NDVI, population density, distance to water point, distance to river, distance to house and distance to road. HWCs involving different species were seen to be driven by different factors. Overall, distance to protected area, annual average precipitation and population density were selected as the most important variables determining the distribution of conflicts. The models were seen to be accurate and realistic in most cases. However, the models’ ability to be generalized in different areas is debatable and the models must be tuned for distinct regions separately

Présentation

Version intégrale (4,02 Mb)

Page publiée le 6 avril 2018