Evaluation of Approaches for Mapping Tidal Wetlands of the Chesapeake and Delaware Bays

• Main Authors: Brian T. Lamb, Maria A. Tzortziou, Kyle C. McDonald
• Format: Article
• Language: English
• Published: MDPI AG 2019-10-01
• Series: Remote Sensing
• Subjects: tidal wetlands; synthetic aperture radar; random forest; r; google earth engine; chesapeake bay; delaware bay; sentinel-1a; sentinel-2a; palsar; palsar-2; landsat 8
• Online Access: https://www.mdpi.com/2072-4292/11/20/2366

The spatial extent and vegetation characteristics of tidal wetlands and their change are among the biggest unknowns and largest sources of uncertainty in modeling ecosystem processes and services at the land-ocean interface. Using a combination of moderate-high spatial resolution (≤30 meters) optical and synthetic aperture radar (SAR) satellite imagery, we evaluated several approaches for mapping and characterization of wetlands of the Chesapeake and Delaware Bays. Sentinel-1A, Phased Array type L-band Synthetic Aperture Radar (PALSAR), PALSAR-2, Sentinel-2A, and Landsat 8 imagery were used to map wetlands, with an emphasis on mapping tidal marshes, inundation extents, and functional vegetation classes (persistent vs. non-persistent). We performed initial characterizations at three target wetlands study sites with distinct geomorphologies, hydrologic characteristics, and vegetation communities. We used findings from these target wetlands study sites to inform the selection of timeseries satellite imagery for a regional scale random forest-based classification of wetlands in the Chesapeake and Delaware Bays. Acquisition of satellite imagery, raster manipulations, and timeseries analyses were performed using Google Earth Engine. Random forest classifications were performed using the R programming language. In our regional scale classification, estuarine emergent wetlands were mapped with a producer’s accuracy greater than 88% and a user’s accuracy greater than 83%. Within target wetland sites, functional classes of vegetation were mapped with over 90% user’s and producer’s accuracy for all classes, and greater than 95% accuracy overall. The use of multitemporal SAR and multitemporal optical imagery discussed here provides a straightforward yet powerful approach for accurately mapping tidal freshwater wetlands through identification of non-persistent vegetation, as well as for mapping estuarine emergent wetlands, with direct applications to the improved management of coastal wetlands.