Year of Award

2017

Document Type

Thesis

Degree Type

Master of Science (MS)

Degree Name

Geography

Department or School/College

Department of Geography

Committee Chair

Anna Klene

Commitee Members

Faith Ann Heinsch, Ragan Callaway

Keywords

Blackfoot Swan Landscape Restoration Project; MaxEnt; Species Distribution Modeling; Rare Plants

Abstract

This study predicted rare-plant habitat at the landscape scale. Using the Maximum Entropy (MaxEnt) algorithm, relationships to and between each environmental variable were quantified for five species in the Blackfoot Swan Landscape Restoration Project (BSLRP) study area in western Montana. This project is part of a greater vegetation assessment for BSLRP that utilizes remotely sensed products for planning and management purposes. The five rare plant species studied in this analysis were common camas (Camassia quamash), clustered lady’sslipper (Cypripedium fasciculatum), western pearlflower (Heterocodon rariflorum), Howell’s gumweed (Grindelia howellii), and crested shieldfern (Dryopteris cristata). Rare plant models typically do not address dispersal mechanisms in conceptual design. This analysis built dispersal mechanisms into model design by buffering the project area based upon dispersal potential. Plant population data is typically stored as polygons in state and federal databases. This data is usually condensed into a single point before entry into modeling algorithms. This analysis addressed this issue by proportionately placing multiple points inside the polygons. In addition, this analysis considered the effects of using different regularization parameter values in MaxEnt and how it affected model performance. For one species, the efficacy of including LiDAR-derived canopy cover to enhance discrimination of understory communities and its effect on improving model performance was examined. Accuracy assessments were used to better understand predictions and statistical relationships between environmental variables. Lastly, predicted habitat maps were overlaid to identify areas of high probability habitat for multiple species across the project area. The field surveys identified thirteen new populations of plants. Overall accuracy for the predictions ranged from 26 to 69%. Comparison between predictions based upon centroids versus distributed points yielded an improved AUC and reduced standard deviation. The LiDAR data defined a narrower niche and had an improved AUC and lower standard deviation but did not lower assessed accuracy for crested shieldfern. Species distribution model studies are one way for resource managers to identify potential habitat for rare species before going into the field. They can use this information to prioritize field surveys and inform management decisions. Also, SDM studies provide information on species’ environmental associations and can be used to further understand species ecology.

Download

Share

COinS
 

© Copyright 2017 Annalisa Suzan Ingegno