Isolating Anthropogenic Wetland Loss by Concurrently Tracking Inundation and Land Cover Disturbance across the Mid-Atlantic Region, U.S.

Abstract

Global trends in wetland degradation and loss have created an urgency to monitor wetland extent, as well as track the distribution and causes of wetland loss. Satellite imagery can be used to monitor wetlands over time, but few efforts have attempted to distinguish anthropogenic wetland loss from climate-driven variability in wetland extent. We present an approach to concurrently track land cover disturbance and inundation extent across the Mid-Atlantic region, United States, using the Landsat archive in Google Earth Engine. Disturbance was identified as a change in greenness, using a harmonic linear regression approach, or as a change in growing season brightness. Inundation extent was mapped using a modified version of the U.S. Geological Survey’s Dynamic Surface Water Extent (DSWE) algorithm. Annual (2015–2018) disturbance averaged 0.32% (1095 km2 year-1) of the study area per year and was most common in forested areas. While inundation extent showed substantial interannual variability, the co-occurrence of disturbance and declines in inundation extent represented a minority of both change types, totaling 109 km2 over the four-year period, and 186 km2, using the National Wetland Inventory dataset in place of the Landsat-derived inundation extent. When the annual products were evaluated with permitted wetland and stream fill points, 95% of the fill points were detected, with most found by the disturbance product (89%) and fewer found by the inundation decline product (25%). The results suggest that mapping inundation alone is unlikely to be adequate to find and track anthropogenic wetland loss. Alternatively, remotely tracking both disturbance and inundation can potentially focus efforts to protect, manage, and restore wetlands.

DSWEmod — The Production of High‐Frequency Surface Water Map Composites from Daily MODIS Images

“…Our extensive assessment of the products also showed that "high confidence" water pixels were the most reliable, while classification errors were common in moderate (Classes 2 and 3) and lower confidence water classes (Class 4), including confusion with urban areas. These errors, along with similar findings in other DSWE investigations (Vanderhoof et al 2020), informed our decisions to focus solely on "high confidence" water pixels (Class 1) for our analyses. The accuracy assessment, based on JRC reference data (Pekel et al 2016), of water presence (Class 1) and water absence (all other classes), suggests that Class 1 pixels mapped in MODIS images do an effective job of identifying surface water patterns mapped in moderately sized (approximately 25 ha) water bodies located in California.…”

“…Qualitative assessments of the DSWEmod product reveal accuracy characteristics similar to our and previous DSWE classification efforts (see Figure 2d). As confirmed by comparison with a contemporaneous, higher-resolution image (not shown), "high confidence" water (Class 1) matches open water well, while Classes 3 and 4 overlap with a range of other land cover and land use types, such as mixed vegetation and urban, also observed by Vanderhoof et al (2020). Due to concerns about mapping accuracy, we follow and and focus the remainder of our analyses on the ability to capture "high confidence" water pixels.…”

“…DSWEmod can be fused with Landsat DSWE methods or extended to include Sentinel-2 to either generate even more complete records or to potentially help identify finer-scale features (Zhang et al 2014;Singh et al 2020;. National wetland mapping (Vanderhoof et al 2020), large-scale flood hazard assessments (Tellman et al 2020), and global raster water masks (Carroll et al 2009) are a few of the additional applications that can be readily generated from publicly available DSWEmod code and/or products (Soulard et al 2021).…”

“…also created a multi-decade record of Landsatbased DSWE and concluded that "high confidence" water area summarized at the subbasin scale in parts of California correlated well with streamgage discharge records, but the relationship did not improve when lower confidence classes were included. An application of DSWE focused on the Mid-Atlantic Region of the U.S. observed that lower confidence classes were often confused with upland forests and built-up lands (Vanderhoof et al 2020).…”

Assessment of carbon storage potential of forested wetland soils as affected by urbanization degree in two different physiographic provinces of Virginia, USA

“…Urbanization has becoming a pressing environmental and social issue across the globe, as the outward growth of urban development into former natural and agricultural lands has had lasting impacts on natural ecosystem services. Within the Chesapeake Bay Watershed (CBW) of the Mid-Atlantic United States, for example, the impacts of urbanization on the loss of natural ecosystems like forests and forested wetlands have been palpable, with over 100 km 2 disturbed between 2015 and 2018 (Dahl 2011;Fretwell et al 1996;Homer et al 2020;Johnson and Lichter 2020;Lacher et al 2019;Scanes 2018;Vanderhoof et al 2020). Despite such losses, urban-speci c environmental problems-speci cally, high atmospheric greenhouse gas concentrations-have been addressed in part by prioritizing urban green spaces like urban forests and wetlands (Bae and Ryu 2015;Pulighe et al 2016;Säynäjoki et al 2018;Xue et al 2019).…”