Purpose: Wetland Conservation with Earth Observation Data and Deep Learning

Semester: Fall, 2024

Supervisors: Leila Hashemi-Beni, Mulham Fawakherji

Software: ArcGIS Pro, Google Earth Engine, Google Earth Pro, Visual Studio Code

Study Area

The coastal areas of Wilmington, New Hanover are endowed with different classes of wetlands including salt marshes, brackish marshes, and tidal creeks, which play a critical role in filtering pollutants, providing nursery habitats for marine species, and protecting the shoreline from erosion and flooding .

While many of Wilmington’s wetlands face threats from urban development, ongoing restoration and conservation efforts aim to preserve and enhance wetland ecosystems for flood control and biodiversity

Multispectral data from remote sensing systems and the recent adoption of AI models that implement neural networks have provided a powerful method for detecting complex patterns in landcover data without manual feature engineering. Neural networks are efficient at detecting non-linearities and complex interactions in data, making them ideal for wetland mapping

Data Inputs

Labelled Data

Data consists of wetlands in hydrogeomorphic classes based on landscape position, water resources and hydrodynamics (Brinson, 1993)

National Wetland Inventory data for the Hannover study site were last updated in 1999 by the North Carolina Coastal Region Evaluation of Wetland Significance (NC-CREWS)

Preliminary Results

Presentation