Developing and Assessing Modeling Strategies in Plant Population and Community Ecology

Date of Completion

January 2012


Biology, Botany|Biology, Ecology




Ecosystems are changing rapidly in response to anthropogenic climate change, land use change and nonnative species. Understanding these changes requires a synthesis of concepts from ecological theory and robust modeling strategies to help identify their causes and effects. I focus on developing and evaluating novel modeling strategies for understanding spatial patterns of species occurrence to determine their strengths and weaknesses, and the ecological insights they facilitate. ^ In Chapter 1, I use Entropy Maximization (EM) models to link functional traits to community abundance patterns in South African fynbos communities. I develop a complementary suite of tests to examine the strengths and limitations of EM and the community-aggregated traits on which it depends. I show that EM can characterize ecological niches by quantifying constraints on complex trait relationships in local communities, demonstrate how similarity in species' traits confounds predictions, and provide guidelines for applying EM. ^ Species distribution models (SDMs) are a fundamental tool in ecology, conservation biology and biogeography and are typically static and phenomenological. In Chapter 2, I demonstrate the importance of complementing correlative SDMs with spatially explicit, dynamic, mechanistic models. I develop general, grid-based, pattern-oriented models incorporating three mechanisms—plant population growth, local dispersal and long distance dispersal—to predict broad scale spread patterns in heterogeneous landscapes. As a case study, I examine the spread of the invasive Celastrus orbiculatus across northeastern North America. ^ In Chapter 3, I provide guidelines for building robust SDMs using the software package Maxent. Very little guidance exists on how alternative formulations and assumptions of Maxent models relate to specific ecological questions, so default settings are often chosen. I relate a detailed explanation of how Maxent works to decisions necessary for fitting SDMs. I focus on building simple, interpretable models that contrasts with current machine learning approaches aimed at complex pattern recognition. I discuss how choice of background samples and accounting for sampling bias reflect prior assumptions, the challenges of evaluating presence-only models, the importance of accounting for prediction uncertainty in models, how priors can be used to represent ecological information, and demonstrate the methods using a case study on the Proteaceae of South Africa. ^