Date of Graduation
2007
Document Type
Thesis
Degree Type
MA
Committee Chair
Jennifer Miller
Abstract
The distribution of plants and animals in space and time has long been a focus of many biogeographical and ecological studies. The various approaches for modeling species distributions are rooted in the quantification of the species-environment relationship, where bioclimatic variables are used to explain the distribution of species and communities. Predictive modeling of species distribution has become a widespread tool in the areas of conservation biology, climate change research, land-use/land-cover change assessment, and biodiversity estimates. Although many statistical methods are now available, previous model comparison studies have found little difference in prediction accuracy when models were compared using the same data. Therefore, there is a need to explore ways to maximize prediction accuracy with multiple models since comparison studies have not found a ìbestî model. One alternative for increasing predictive accuracy among modeling techniques has been through combining models. Using a dataset consisting of presence/absence data of four vegetation alliances from the Mojave Desert, CA, and twelve environmental predictor variables, I compare different model combination techniques of three different types of parametric and non-parametric statistical models: classification trees, generalized linear models, and multivariate adaptive regression splines. Each techniqueís classification accuracy is assessed by using threshold-independent receiver-operating characteristic (ROC) plots and threshold-dependent Kappa.
Recommended Citation
Gilmer, Benjamin F. IV, "Predictive vegetation models: A comparison of model combination approaches." (2007). Graduate Theses, Dissertations, and Problem Reports (ETD). 10973.
https://researchrepository.wvu.edu/etd/10973