Journal of Digital Landscape Architecture

Seite drucken

Training A Machine Learning Model to Identify Historic Wetlands in an Area of Critical Environmental Concern

Historic land use and topographic maps have been used to inform design at the scale of discrete sites by revealing features that may have been obscured or buried, in the case of wetlands and floodplains, over time. Using a case study in the Great Marsh, an Area of Environmental Concern in the Commonwealth of Massachusetts in the United States of America, we provide a review of U-Net deep learning methods for archival cartographic products and investigate a machine learning approach to identifying wetlands from historic maps produced by the United States Geological Survey (USGS). The case study uses a UNet deep learning workflow to operationalize hand-drawn bespoke raster patterns into vector geospatial data. The results of this initial methodology reveal the potential of using machine learning to digitize other hand-drawn raster cartographic features in USGS maps to inform landscape design at both a site and regional scale.

Autor / Author: Clingen, Kira; Booz, Justin
Institution / Institution: Harvard University, Massachusetts/USA; Harvard University, Massachusetts/USA
Seitenzahl / Pages: 9
Sprache / Language: Englisch
Veröffentlichung / Publication: JoDLA – Journal of Digital Landscape Architecture, 10-2025
Tagung / Conference: Digital Landscape Architecture 2025 – Collaboration
Veranstaltungsort, -datum / Venue, Date: Dessau Campus of Anhalt University, Germany 04-06-25 - 07-06-25
Schlüsselwörter (de):
Keywords (en): Machine learning, climate change, wetlands, adaptation
Paper review type: Full Paper Review
DOI: doi:10.14627/537754031
8374 -