Degree

Doctor of Philosophy (PhD)

Department

Geography and Anthropology

Document Type

Dissertation

Abstract

Barrier islands provide important ecosystem services, including storm protection and erosion control to the mainland, habitat for fish and wildlife, and tourism. As a result, natural resource managers are concerned with monitoring changes to these islands and modeling future states of these environments. Landscape position, such as elevation and distance from shore, influences habitat coverage on barrier islands by regulating exposure to abiotic factors, including waves, tides, and salt spray. Geographers commonly use aerial topographic lidar data for extracting landscape position information. However, researchers rarely consider lidar elevation uncertainty when using automated processes for extracting elevation-dependent habitats from lidar data. Through three case studies on Dauphin Island, Alabama, I highlighted how landscape position and treatment of lidar elevation uncertainty can enhance habitat mapping and modeling for barrier islands. First, I explored how Monte Carlo simulations increased the accuracy of automated extraction of intertidal areas. I found that the treatment of lidar elevation uncertainty led to an 80% increase in the areal coverage of intertidal wetlands when extracted from automated processes. Next, I extended this approach into a habitat mapping framework that integrates several barrier island mapping methods. These included the use of landscape position information for automated dune extraction and the use of Monte Carlo simulations for the treatment of elevation uncertainty for elevation-dependent habitats. I found that the accuracy of dune extraction results was enhanced when Monte Carlo simulations and visual interpretation were applied. Lastly, I applied machine learning algorithms, including K-nearest neighbor, support vector machine, and random forest, to predict habitats using landscape position information extracted from topobathymetric data. I used the habitat map to assess the accuracy of the prediction model and I assessed the ability of the model to generalize by hindcasting habitats using historical data. The habitat model had a deterministic overall accuracy of nearly 70% and a fuzzy overall accuracy of over 80%. The hindcast model had a deterministic overall accuracy of nearly 80% and the fuzzy overall accuracy was over 90%. Collectively, these approaches should allow geographers to better use geospatial data for providing critical information to natural resource managers for barrier islands.

Date

8-5-2019

Committee Chair

Wang, Lei

Share

COinS