Models that are used for predicting species' potential distributions are important tools that have found applications in a number of areas of applied ecology. The majority of these models can be classified as correlative, as they rely on strong, often indirect, links between species distribution records and environmental predictor variables to make predictions. Correlative models are an alternative to more complex mechanistic models that attempt to simulate the mechanisms considered to underlie the observed correlations with environmental attributes. This study explores the influence of the type and quality of the data used to calibrate correlative models. In terms of data type, the most popular techniques in use are group discrimination techniques, those that use both presence and absence locality data to make predictions. However, for many organisms absence data are either not available or are considered to be unreliable. As the available range of profile techniques (those using presence only data) appeared to be limited, new profile techniques were investigated and evaluated. A new profile modelling technique based on fuzzy classification (the Fuzzy Envelope Model) was developed and implemented. A second profile technique based on Principal Components Analysis was implemented and evaluated. Based on quantitative model evaluation tests, both of these techniques performed well and show considerable promise. In terms of data quality, the effects on model performance of false absence records, the number of locality records (sample size) and the proportion of localities representing species presence (prevalence) in samples were investigated for logistic regression distribution models. Sample size and prevalence both had a significant effect on model performance. False absence records had a significant influence on model performance, which was affected by sample size. A quantitative comparison of the performance of selected profile models and group discrimination modelling techniques suggests that different techniques may be more successful for predicting distributions for particular species or types of organism than others. The results also suggest that several different model design! sample size combinations are capable of making predictions that will on average not differ significantly in performance for a particular species. A further quantitative comparison among modelling techniques suggests that correlative techniques can perform as well as simple mechanistic techniques for predicting potential distributions.
Identifer | oai:union.ndltd.org:netd.ac.za/oai:union.ndltd.org:rhodes/vital:5816 |
Date | January 2003 |
Creators | Robertson, Mark Peter |
Publisher | Rhodes University, Faculty of Science, Zoology and Entomology |
Source Sets | South African National ETD Portal |
Language | English |
Detected Language | English |
Type | Thesis, Doctoral, PhD |
Format | 252 leaves, pdf |
Rights | Robertson, Mark Peter |
Page generated in 0.0016 seconds