Date of Award

2013

Degree Type

Thesis

Degree Name

Master of Science (MS)

Department

Ecological Sciences and Engineering

First Advisor

Songlin Fei

Committee Chair

Songlin Fei

Committee Member 1

Jeffrey Dukes

Committee Member 2

Jane Frankenberger

Abstract

Invasive species have significant ecological and economic impacts. To control species' invasion, risk assessment provides the most essential information for identification and evaluation of the potential risk of the invasive species, especially in their early invasion stages. Species distribution models (SDMs) is the foundation for risk assessment, in terms of both the practical and theoretical interest in our understanding of species invasion process. SDMs contribute to the proactive invasion management and the test of ecological or biogeographical hypotheses about species distributions in relation to their environment.

However, modeling of invasive species at large spatial scale (i.e., cross-continental) is rarely discussed. Besides, sampling bias of the presence-only occurrence data can seriously reduce the performance of SDMs, but no quantitative method is available to assess data quality. In my thesis, I used MaxEnt to build bioclimatic envelope models for 39 high-risk invasive plant species and predicted their potential invasion ranges in the U.S., based on their global-scale and presence-only occurrence data. I used optimized-parameterization techniques such as `target-group' background selection and regularization value optimization to improve model performance. I also created an acceptability criterion, proximity to ideal completeness and evenness (PICE), to evaluate the quality of species occurrence data in terms of representativeness and equilibrium. Results indicated that southern Florida and southwestern U.S. have high probability of invasion for most of the 39 invasive plant species. Meanwhile, the quality of species occurrence data has greater influences on model performance than parameterization. Generally, model performance (reliability and accuracy) stabilizes when PICE is greater than 0.40. The predicted results can assist early detection and monitoring of exotic invasion via regional-level prevention, proactive management, and policy-making. PICE can be used as a model-independent acceptability criterion to evaluate model performance with a given set of species occurrence data. The computational efficiency of PICE benefits the SDM community by preventing the utilization of biased, poor quality species occurrence data.

Download

Share

COinS