Object-based machine learning correction of LiDAR using RTK-GNSS to model the potential effects of sea-level rise in Swanquarter National Wildlife Refuge, North Carolina

Abstract

Coastal wetland systems are a vital habitat that provide many beneficial services; however, the complexity of these habitats makes it difficult for conservation managers to preserve these environments and predict future changes. Sea-level rise (SLR) is a growing and accelerating threat to coastal wetlands making its predictability essential for conservation planners. Light Detection and Ranging (LiDAR) Digital Elevation Models (DEMs) have become an important component in monitoring coastal wildlife refuges and are implemented into models like Sea Level Affecting Marshes Model (SLAMM) to produce SLR vulnerability assessments. Although, with dense vegetation in these environments LiDAR penetration is reduced and DEMs in turn are less accurate. This study implemented an Object-Based Machine Learning (OBML) technique to improve DEM accuracy at Swanquarter National Wildlife Refuge (SNWR) and was implemented into SLAMM to provide land cover maps of the year 2050 for land cover change analysis. The corrected OBML DEM was compared with the original LiDAR DEM obtained from North Carolina Floodplain Mapping Program (NCFMP), which found the OBML DEM to provide a more reliable depiction of the potential impacts of future SLR on the coastal wetlands in North Carolina. Conservation managers may find the OBML approach in this study to be a useful option for SLR analysis.