Date Approved

8-31-2022

Graduate Degree Type

Thesis

Degree Name

Biology (M.S.)

Degree Program

Biology

First Advisor

Robert D. Hollister

Second Advisor

Alexandra Locher

Third Advisor

Sergio A. Vargas Zesati

Academic Year

2021/2022

Abstract

Plot photography can provide a quick, robust method to measure vegetation, especially in polar environments where logistics can be expensive and challenging. The success and widespread adoption of plot photography in the Arctic hinges on the accuracy of image analysis and data product interpretation. The relative cover of eight vegetation classes was estimated using a point frame and digital camera across thirty, 1-m2 plots at Utqiaġvik, Alaska from 2012 to 2021. Geographic object-based image analysis (GEOBIA) was applied to generate objects and classify the three band (red, green, blue) images. Machine learning classifiers (random forest, gradient boosted model, classification and regression tree, support vector machine, k-nearest neighbor) were applied, and random forest performed the highest (60.5% overall accuracy). Objects were classified reliably in six out of the eight vegetation classes using the random forest classification, including bryophytes, forbs, graminoids, litter, shadow and standing dead. Deciduous shrubs and lichens were not reliably classified. We also assessed whether estimates of relative vegetation cover from plot photography were comparable to estimates using the point frame. Based on Spearman-Rank correlations within each year, graminoid cover was consistently, positively correlated. Most of the remaining vegetation classes showed moderate positive associations except for litter and standing dead, which showed a negative association. We then used multinomial regression models to gauge if the cover estimates from plot photography could accurately predict the abundance estimates from the point frame across space or time. Currently, our approach to image analysis is best suited to detect large shifts in composition over spatial gradients rather than the more subtle temporal shifts in vegetation over time. Together these results suggest that plot photography coupled with semi-automated image analysis maximizes time, funding, and available technology to monitor vegetation cover in the Arctic.

Download

Share

COinS