Digital elevation models for the South Atlantic Bight with elevation uncertainty treatment, overview

This data release includes digital elevation models (DEMs) with elevation uncertainty treatment of the South Atlantic Bight. The DEMs were created using Monte Carlo simulations targeted towards limiting overestimation in elevation. For each DEM, we used Monte Carlo simulations with 1,000 iterations to create new DEM realizations. We then summarized the 1,000 DEM realizations by calculating the 5th through 95th percentiles (at an increment of 5 percentiles) of elevation by pixel across all the iterations. Point data observations from several sources were used to calculate the mean bias error (MBE) to identify a lower, best, and upper estimate for each DEM. For areas classified as vegetated wetland land cover types (that is, emergent and scrub/shrub wetlands and estuarine forested wetland classes), excluding palustrine forested wetlands, the recommended DEMs were determined based off the MBE using validation data. The "best" DEM was determined by the percentile that had a MBE closest to zero and the upper and lower recommended DEMs were determined based on a general rule to select the percentile furthest from the “best” percentile that had a MBE within +/-5 cm. For the SAB region, the "best" percentile was the 35th percentile and the upper percentile was the 65th and the lower percentile was the 20th. Due to lack of validation points outside of emergent wetlands, areas classified as upland vegetated, non-vegetated, and palustrine forested wetland land cover types were set to the 25th, median, and 75th percentiles for the “lower”, “best”, and “upper” DEMs respectively. For each recommended DEMs we used an approach that allowed for smoothing transitions between areas where the percentiles differed. To do this, we developed land cover masks that were used to isolate each grouping of land cover classes (that is, vegetated wetlands, excluding palustrine forested wetlands; upland vegetation and palustrine forested wetlands; and non-vegetated). These masks were expanded by two pixels. For this study area, we used the 2016 NOAA C-CAP 30-meter land cover product. The recommended percentiles for each DEM (lower, best, or upper) were extracted to the corresponding masked extent. The extracted DEM surfaces were then mosaicked into a single, seamless DEM where areas of overlap were resolved using the mean mosaic operation (that is, overlapping area [2-pixel areas on either side of transitions] contained the mean of the DEM percentiles used for the specific DEM [for example, the best would have the mean of the 50th percentile and the 35th percentile). The resulting DEMs preserve the optimal percentiles for vegetated wetlands while still reducing possible overestimation issues in upland vegetated areas and palustrine forested wetland areas. This metadata file describes the overview of this data release.