In tidal wetlands, small differences in ground elevation can have a large impact on hydrology, vegetation, and habitat (e.g., assessment of wetland health and stability, habitat, flood risk, and coastal inundation). Estimating ground elevation using LiDAR (Light Detection and Ranging) is desirable in tidal wetlands because it can cover large areas (statewide) and produce high spatial densities of elevations. Unfortunately, in tidal wetlands, the high density of vegetation cover inhibits penetration of LiDAR pulses through the vegetation to the ground surface and low vegetation heights inhibit classification of ground point elevations. It is also rare to constrain or verify LiDAR data with ground measurements in tidal wetlands. This typically results in a positive bias in elevations in LiDAR-derived spot elevations and gridded digital elevation models (DEMs) in tidal wetlands of 10 to 50 cm. Removal of this positive bias is required to effectively use LiDAR-derived elevations for many tidal-wetland applications.

GPS-RTK measurements and GIS techniques are employed to better understand and reduce vertical bias in the 2014 Delaware LiDAR-derived DEM for saltwater wetlands in the watersheds containing the St. Jones and Blackbird Creek research reserves.   Approximately 425 points were surveyed covering five major vegetation communities that make up the vast majority of the reserves vegetation: low salt marsh (spartina alterniflora), high salt marsh (spartina patens), tidal reed marsh (phragmites), Eastern tidal salt shrub (shrub), and mesohaline seepage marsh (other). 

Overall, the 2014 Delaware DEM was found to have a 13 centimeter (cm) to 26 cm positive elevation when compared to GPS-RTK points and the magnitude of the bias was dependent on vegetation communities. Bias correction factors were developed and applied to the USGS 2014 DEM to produce a corrected DEM for each reserve.  Three correction methods were evaluated in the current study: mean overall bias, mean bias per vegetation community, and minimum-value bins derived from the LiDAR point cloud, with the mean bias per vegetation type method reducing the error the most with mean absolute errors (MAE) down to 4 cm.