Predicting the Habitat Suitability of Asian Elephants in 2070 with Bayesian Models
Loading...
URL
Journal Title
Journal ISSN
Volume Title
Perustieteiden korkeakoulu |
Master's thesis
Unless otherwise stated, all rights belong to the author. You may download, display and print this publication for Your own personal use. Commercial use is prohibited.
Authors
Date
2022-06-13
Department
Major/Subject
Ryoko Noda
Mcode
SCI3044
Degree programme
Master’s Programme in Computer, Communication and Information Sciences
Language
en
Pages
63 + 14
Series
Abstract
Pleistocene rewilding is an ambitious approach to conservation in which extinct megafauna are replaced by extant relatives to fulfill missing ecological roles. While controversial, it has the potential to repair diminishing ecosystems with minimal human intervention. However, careful inspection and simulation is required before the project can take place as this involves introducing species to a new habitat. For this purpose, species distribution models are expected to be a powerful tool as they predict species presence under various climate and habitat conditions. Therefore, this thesis presents a case study of iterative modeling and Bayesian workflow of species distribution models under a hypothetical Pleistoscene rewilding plan. The aim of this project is to predict a suitable habitat for Asian elephants (Elephas Maximus) that remains suitable until the year 2070. All models in the workflow use predicted climate features under the representative concentration path- way 8.5 scenario to produce scores of future habitat suitability (range [0, 1]. 0: not suitable/1: suitable). The iterative model building starts with non-Bayesian machine learning, logistic regression and random forest. These models are then used as benchmarks for two different Bayesian models, a Bayesian generalized linear model and a Bayesian generalized additive model. While building and exploring models, this research explores the effects of feature selection, feature scaling, and for the Bayesian models, the priors and non-linearity settings. The model exploration process was able to identify a model that gives convincing predictions for present-day and future conditions. The outputs from this model implied that possible rewilding sites would include northern South America, sea-facing regions of east and west Africa, and the north shoreline of Australia.Description
Supervisor
Vehtari, AkiThesis advisor
Žliobaitė, IndrėKeywords
Species Distribution Models, Bayesian modeling, Bayesian workflow, machine learning, pleistocene rewilding, elephas maximus