Estuarine Shoreline Mapping using Object-based Ensemble Analysis, Aerial Imagery, and LiDAR: A Case Study in the Neuse River Estuary, NC

Abstract

Estuarine shorelines are highly dynamic due to their unique geological history, wave and weather conditions, and human modifications to the shoreline. These interactions are heightened as sea level rise intensifies and extreme storms become more frequent due to climate change. Estuarine shoreline classification maps are critical to understanding the context and magnitude of storm-induced erosion as well as ad hoc efforts to shoreline stabilization. Here, an object-based ensemble analysis is used to map natural and engineered shoreline types observed within the Neuse River Estuary (NRE), NC. Object-based ensemble analysis has emerged as a successful framework to improve image classification but has yet to be tested in classifying an estuarine shoreline environment. This approach used in-situ reference data, high-resolution aerial imagery, and LiDAR point data to train an ensemble of five machine learning algorithms (Random Forest, Support Vector Machine, LibLINEAR, Artificial Neural Network, and k-Nearest Neighbors). The object-based ensemble produced the highest overall classification accuracy at 76.4% (Kappa value = 0.66), 6.3% higher than the top performing pixel-based model, justifying its use to produce the final shoreline classification map. NRE shoreline change and erosion vulnerability were classified using the object-based image analysis and produced comparable erosion rates to those observed in past studies. The object-based ensemble approach was an effective way to map shoreline classifications in the NRE and should continue to be explored within other shoreline management applications.