Date of Award:

5-2002

Document Type:

Thesis

Degree Name:

Master of Science (MS)

Department:

Biology

Committee Chair(s)

James W. Haefner

Committee

James W. Haefner

Committee

Edmund D. Brodie, Jr.

Committee

John A. Bissonette

Abstract

As geographical information systems and spatial data become more accessible, predictive spatial modeling in ecology is becoming more common. Unfortunately, not all ecologists possess the necessary skills to successfully combine statistical models and geographical information systems. In response to this problem, I wrote an extension for ArcView® GIS called StatMod. This extension interfaces ArcView GIS with SAS® and S-PLUS® statistical software and walks the user through creating and mapping logistic regression and classification and regression tree models.

StatMod was then used to run a series of analyses that would have been difficult without such a tool. A hypothetical density distribution for a nonexistent species was created, and then the effects of sample size and sampling regime on predictions of this density distribution were investigated. The effect of ground-truthed sample size on model accuracy estimates was also examined.

As expected, model accuracy generally improved as sample size increased. However, the majority of the improvement was seen as the sampled area approached one percent of the study area. Simple random sampling usually performed better than stratified random sampling, although the difference was most apparent when one density classification dominated the landscape. Estimates of model accuracy also improved as the number of ground-truthed sample points increased. Samples stratified by model predictions performed better than simple random samples at small sample sizes, although the reverse was true at large sample sizes. Again, this interaction was most apparent when one density classification dominated the landscape.

Checksum

1e79e76a3b456f25c3be4db54fa19c59

Included in

Biology Commons

Share

COinS