To assess the accuracy of the LIDAR-derived DEM the DEM was compared to the RTK ground truth training data at 297 locations to calculate the mean vertical error. We developed habitat-specific correction factors for six cover classes: three height classes of Spartina alterniflora (tall, medium, short); Juncus roemerianus/Schoenoplectus spp.; Spartina cynosuroides/Schoenoplectus tabernaemontani; and marsh meadow (Salicornia virginica(Sarcoconia sp.), Distichlis spicata, Batis maritima). Class-specific correction factors were derived from the RTK training data set by subtracting the surveyed RTK elevation from the DEM elevation at the corresponding x/y coordinate of each GCP and represented the mean error for each cover class. The, overall mean vertical error the unmodified DEM was was 0.44 ft (fundamental vertical accuracy (FVA) of 1.54 ft), over predicting ground elevations. The mean errors for the different dominant species cover classes ranged from -0.13 (marsh meadow) to 0.90 ft (J. roemerianus/ Schoenoplectus spp.), with taller vegetation having larger errors.

The habitat classification was combined with the unmodified LIDAR-derived DEMs and the DEM correction factors to correct the DEMs (on a habitat classification pixel-pixel basis). Following the method described in Hladik and Alber (2012), the habitat classification raster was brought into ArcGIS. Next, the random forest classified raster was reclassed by assigning a habitat-specific correction factor to each dominant class. The mud and unvegetated classes were given a correction factor of zero. This produced a “Correction Factor” DEM with values corresponding to the dominant class-specific correction factors. The “Correction Factor” DEM was then subtracted from the “Unmodified” DEM using the Raster Math tool in ArcGIS (Spatial Analyst toolbox) to produce a “Modified” DEM. The application of the derived correction factors and subsequent DEM modification based on the random forest classification were successful and greatly improved the accuracy of the LIDAR-derived DEM, reducing the overall mean DEM error from 0.40 to -0.07 ft and the FVA from 1.68 to 1.42 ft based on the reserved RTK validation data (N=299),. The slight negative value for the overall error indicates that the correction factors produced a DEM surface that was slightly lower than RTK elevations, but no classes were significantly different in comparison to RTK GCPs.

To assess the accuracy of the LIDAR-derived DEM the DEM was compared to the RTK ground truth training data at 297 locations to calculate the mean vertical error. We developed habitat-specific correction factors for six cover classes: three height classes of Spartina alterniflora (tall, medium, short); Juncus roemerianus/Schoenoplectus spp.; Spartina cynosuroides/Schoenoplectus tabernaemontani; and marsh meadow (Salicornia virginica(Sarcoconia sp.), Distichlis spicata, Batis maritima). Class-specific correction factors were derived from the RTK training data set by subtracting the surveyed RTK elevation from the DEM elevation at the corresponding x/y coordinate of each GCP and represented the mean error for each cover class. The, overall mean vertical error the unmodified DEM was was 0.44 ft (fundamental vertical accuracy (FVA) of 1.54 ft), over predicting ground elevations. The mean errors for the different dominant species cover classes ranged from -0.13 (marsh meadow) to 0.90 ft (J. roemerianus/ Schoenoplectus spp.), with taller vegetation having larger errors.

The habitat classification was combined with the unmodified LIDAR-derived DEMs and the DEM correction factors to correct the DEMs (on a habitat classification pixel-pixel basis). Following the method described in Hladik and Alber (2012), the habitat classification raster was brought into ArcGIS. Next, the random forest classified raster was reclassed by assigning a habitat-specific correction factor to each dominant class. The mud and unvegetated classes were given a correction factor of zero. This produced a “Correction Factor” DEM with values corresponding to the dominant class-specific correction factors. The “Correction Factor” DEM was then subtracted from the “Unmodified” DEM using the Raster Math tool in ArcGIS (Spatial Analyst toolbox) to produce a “Modified” DEM. The application of the derived correction factors and subsequent DEM modification based on the random forest classification were successful and greatly improved the accuracy of the LIDAR-derived DEM, reducing the overall mean DEM error from 0.40 to -0.07 ft and the FVA from 1.68 to 1.42 ft based on the reserved RTK validation data (N=299),. The slight negative value for the overall error indicates that the correction factors produced a DEM surface that was slightly lower than RTK elevations, but no classes were significantly different in comparison to RTK GCPs.