Impoundments in the Chesapeake Bay coastal zone, 2016

Migration of salt marshes into adjacent uplands presents an opportunity to maintain ecosystem resilience in the face of sea level rise. However, infrastructure installed in low-lying coastal areas can both intentionally and unintentionally act as barriers to marsh migration. In the Chesapeake Bay, impoundments are commonly installed on salt-impacted agricultural fields to create habitat for waterfowl. Most of these structures are privately built and owned, with no centralized, public record, which makes it difficult to assess ecosystem impacts. We use deep learning tools in ArcGIS Pro 3.2 to identify impoundments installed within the Chesapeake Bay coastal zone (0 - 5 m above sea level). We trained a Mask Region-based Convolutional Neural Network (R-CNN) model to detect impoundments using a dataset of slope generated from the U.S. Geological Survey (USGS) Coastal National Elevation Database (CoNED) high resolution (1 m cell width) Topobathymetric Digital Elevation Model (TBDEM). Training samples were delineated in Somerset County, Maryland because of the extensive and rapid salinization of coastal farmland which has led to the widespread installation of waterfowl impoundments. The final set of training samples contained 3,900 images (512 x 512 m cell chip size), of which 90% were used for training and 10% were set aside for validation. The final model selected for impoundment detection had a precision score of 0.9609, which suggested that it performed well over the training area, and was then applied to the entire study region. We conducted extensive post-processing and visual examination of identified impoundments to account for any errors associated with applying the model to a broader region. The final dataset contained 1,684 impoundments which cover 6.6 km^2 (1,627 acres). The CoNED TBDEM was published in 2016, making these results a conservative estimate of impoundments in the Chesapeake Bay today.