Automated Wet/Dry Shoreline Delineation Using Deep Learning

Keywords: Deep Learning, GeoAI, Remote Sensing
Session Type: Virtual Paper Abstract
Day: Tuesday
Session Start / End Time: 3/1/2022 11:20 AM (Eastern Time (US & Canada)) - 3/1/2022 12:40 PM (Eastern Time (US & Canada))
Room: Virtual 21

Abstract

The wet/dry shoreline is the intersection of water and land surfaces. Georeferencing the precise location of the wet/dry shoreline is very helpful for beach management operations for biodiversity conservation in coastal areas. This research proposes a general deep learning method to automate the detection of the wet/dry shoreline using UAV imagery. This approach uses the HED (Holistically-Nested Edge Detection) architecture to obtain an accurate wet/dry shoreline delineation from the georeferenced UAV imagery. The deep learning model has been trained and tested with data from over 10 UAV flights over various beaches over the Gulf of Mexico.
The imagery consists of georeferenced RGB high-resolution imagery, which we then processed as an orthomosaic and used as inputs for training the deep learning model. Once the prediction of the deep learning model has been made, the output is able to automatically extract the exact wet/dry shoreline coordinates in the coastal imagery.
This deep learning GeoAI model, once trained and optimized, enables biologists, geologists, and other coast scientists to apply morphological analysis to their coastal ecosystem conservation efforts using UAV imagery. Thus far, this research has only been applied to the Gulf of Mexico, but we intend to generalize the method for any coastal topology. Moreover, detecting the wet/dry shoreline boundary using deep learning is the initial step to develop models to predict future peak water levels, critical to our research application to protect sea turtle nests from inundation.