Using predicted locations and an ensemble approach to address sparse data sets for species distribution modelling : Long-horned beetles (Cerambycidae) of the Fiji islands

Loading...
Thumbnail Image
Other Title
Authors
Aguilar, Glenn
Waqa-Sakiti, Hilda
Winder, Linton
Author ORCID Profiles (clickable)
Degree
Grantor
Date
2016-12-09
Supervisors
Type
Journal Article
Ngā Upoko Tukutuku (Māori subject headings)
Keyword
Fiji
Long-horned beetles (Cerambycidae)
GIS for conservation
Giant Fijian Beetle (Xixuthrus heros)
GIS mapping
ANZSRC Field of Research Code (2020)
Citation
Aguilar, G. D., Waqa-Sakiti, H. & Winder, L. (2016). Using predicted locations and an ensemble approach to address sparse data sets for species distribution modelling : Long-horned beetles (Cerambycidae) of the Fiji islands. Research report & emedia teaching resource. Unitec Institute of Technology. Unitec ePress. Retrieved from: http://www.unitec.ac.nz/epress
Abstract
Several modelling tools were utilised to develop maps predicting the suitability of the Fiji Islands for longhorned beetles (Cerambycidae) that include endemic and endangered species such as the Giant Fijian Beetle Xixuthrus heros. This was part of an effort to derive spatially relevant knowledge for characterising an important taxonomic group in an area with relatively few biodiversity studies. Occurrence data from the Global Biodiversity Information Facility (GBIF) and bioclimatic variables from the WorldClim database were used as input for species distribution modelling (SDM). Due to the low number of available occurrence data resulting in inconsistent performance of different tools, several algorithms implemented in the DISMO package in R (Bioclim, Domain, GLM, Mahalanobis, SVM, RF and MaxEnt) were tested to determine which provide the best performance. Occurrence sets at several distribution densities were tested to determine which algorithm and sample size combination provided the best model results. The machine learning algorithms RF, SVM and MaxEnt consistently provided the best performance as evaluated by the True Skill Statistic (TSS), Kappa and Area Under Curve (AUC) metrics. The occurrence set with a density distribution of one sampling point per 10km2 provided the best performance and was used for the final prediction model. An ensemble of the best-performing algorithms generated the final suitability predictive map. The results can serve as a basis for additional studies and provide initial information that will eventually support decision-making processes supporting conservation in the archipelago.
Publisher
Unitec ePress
DOI
Copyright holder
Unitec ePress
Copyright notice
Creative Commons Attribution-NonCommercial 4.0 International License
Copyright license