Dune Toe Delineation Utilizing Machine LearningÂ
Standing
Undergraduate
Type of Proposal
Oral Research Presentation
Challenges Theme
Open Challenge
Faculty Sponsor
Chris Houser, Alex Smith
Proposal
Accurately identifying landform boundaries in coastal systems can improve our understanding on process-form relationships that develop in response to variable environmental and anthropogenic controls. With increasing access to large topographic databases (e.g., NOAA’s Digital Coast, NRCAN HRDEM Dataset) it is critical to increase the consistency of landform extraction and our interpretation of coastal dynamics at multiple spatiotemporal scales. Morphometric based and Machine Learning (ML) models have been used to delineate boundaries in coastal systems, but few can produce accurate and consistent results across sites and through time. This research aims to create a form-based approach that integrates a ML model through a deep neural network to extract the dune toe position, which is often used as a proxy for evolution of the landscape in response to changing climate controls including storm impacts and sea level rise. Dune toe predictions were made using pre-existing methods and compared to the predictions made by the three-dimensional Minimum Averaged Relative Relief (MARR) approach introduced in this study. Preliminary results show MARR has significantly less alongshore variability and produces statistically different estimates of horizontal, vertical, and volume change of the foredune system. This work will ultimately provide a more consistent and accurate fully open-source methodology for dune toe extractions, enhancing the transferability of coastal research and our understanding on the resiliency of coastal systems in response to climate change.
Grand Challenges
Viable, Healthy and Safe Communities
Dune Toe Delineation Utilizing Machine LearningÂ
Accurately identifying landform boundaries in coastal systems can improve our understanding on process-form relationships that develop in response to variable environmental and anthropogenic controls. With increasing access to large topographic databases (e.g., NOAA’s Digital Coast, NRCAN HRDEM Dataset) it is critical to increase the consistency of landform extraction and our interpretation of coastal dynamics at multiple spatiotemporal scales. Morphometric based and Machine Learning (ML) models have been used to delineate boundaries in coastal systems, but few can produce accurate and consistent results across sites and through time. This research aims to create a form-based approach that integrates a ML model through a deep neural network to extract the dune toe position, which is often used as a proxy for evolution of the landscape in response to changing climate controls including storm impacts and sea level rise. Dune toe predictions were made using pre-existing methods and compared to the predictions made by the three-dimensional Minimum Averaged Relative Relief (MARR) approach introduced in this study. Preliminary results show MARR has significantly less alongshore variability and produces statistically different estimates of horizontal, vertical, and volume change of the foredune system. This work will ultimately provide a more consistent and accurate fully open-source methodology for dune toe extractions, enhancing the transferability of coastal research and our understanding on the resiliency of coastal systems in response to climate change.