SupClas, GeoSet, SubType, VegDen, VegType: Categorical landcover rasters of landcover, geomorphic setting, substrate type, vegetation density, and vegetation type: Cobb Island, VA, 2014

Cobb14_SupClas.tif
Using the orthoimage, we conducted a supervised classification to delineate substrate and vegetation characteristics. Full methods are provided in the associated Methods OFR (Zeigler and others, 2019). For additional information and example figures, see Zeigler and others (2019). All steps were undertaken in ArcGIS and ArcToolbox version 10.4.1.
Because the signatures of some features closely resembled those of others (e.g., water and the reflection from buildings), we ran the classification in three stages and combined results into a single classification of landcover. We created sub-classifications for (1) bare sand and water, (2) marsh, and (3) vegetation and other features in the barrier’s interior.
To divide the landscape into these three sub-classifications, we hand-digitized masking polygons in ArcGIS. All digitization was performed with the orthoimage for reference. For the marsh classification, we hand-digitized polygons of marsh platforms. For vegetation and inland features, we hand-digitized a polygon that encompassed vegetation and the barrier’s interior by delineating the boundary of open sand and vegetation. Finally, we hand-digitized polygons to encompass areas of human development (Cobb14_Development.shp in larger work). ‘Human development’ could include housing communities, commercial infrastructure, recreational facilities, major roads, and shoreline armoring (e.g., jetties). In sites without any human development, such as in the Virginia Coast Reserve, we omitted the development component of the following steps.
We then conducted the classification for marsh. We used the Classification toolbar to hand-digitize training polygons for the following classes: (1) Water; (2) Marsh, vegetation or unknown cover; (3) Marsh, unvegetated sediment; and (4) Marsh, shrub or forest. We ran the interactive supervised classification routine in the Classification toolbar and clipped the resulting classification image along the boundaries of the marsh mask. In the Image Analysis window, we masked out areas that overlapped with development polygons, if present. The resulting image matched the extent of the marsh polygons and contained cells at the original resolution of the orthoimage (0.35 m) with one of the four values represented by the training polygons.
Next, we created training polygons in the same manner that represented different types of vegetation in the inland portion of the barrier. Training samples represented the following classes: (1) Water; (2) Sand; (3) Herbaceous Vegetation or Shrub, assumed sparse to moderate cover (< 20% cover); (4) Herbaceous Vegetation or Shrub, assumed moderate to dense cover (>20% cover); and (5) Shrub or Forest, assumed moderate to dense cover. We ran the interactive supervised classification routine in the Classification toolbar and clipped the resulting classification image along the boundaries of the vegetation mask. In the Image Analysis window, we masked out areas that overlapped with the marsh or development mask. The resulting image matched the extent of the vegetation mask, minus areas overlapping with marsh and development, and contained cells at the original resolution (0.35 m) of the orthoimage. Cells had one of the five values represented by the training polygons.
We created training polygons in the same manner for the final sub-classification. These samples represented the following classes: (1) Water and (2) Sand. We ran the interactive supervised classification routine in the Classification toolbar. We masked out areas that overlapped with the marsh, development, or vegetation masks in the Image Analysis window. The resulting image covered areas not already covered by the marsh, vegetation, or development and contained cells at the aerial imagery’s original resolution (0.35 m) with a value for either sand or water.
Using the Merge tool in the Image Analysis window in ArcGIS, we combined the rasters for (1) marsh, (2) vegetation, (3) sand/water sub-classifications with the (4) rasterized development layer into a single raster (hereafter, the ‘supervised classification’). We changed the supervised classification’s resolution from the aerial imagery’s original cell size (0.35 m) to a 5 m cell size with the Resample tool in ArcToolbox. We used ‘majority’ as the resampling technique, where the new 5 m cell took on the most common value within the 5x5 m area.
In many instances, cells took on the value of ‘NoData’ around the boundaries of the marsh, vegetation, and development masks when we merged individual sub-classifications. We replaced these NoData values using statistical information from surrounding cells according to the following code in the Raster Calculator tool in ArcToolbox:
CON(isnull([raster]), FOCALMAJORITY([raster], rectangle, 5, 5), [raster]).
This code indicates that, if the value of a cell in the supervised classification is NoData, then replace that value with the value held by the majority of cells in the surrounding 5x5 cell window.
The final supervised classification was a raster layer with a 5 m cell size clipped to the extent of the study area with cells taking on one of nine possible values: (1) water, value = 1; (2) sand, value = 2; (3) herbaceous vegetation or low shrub, sparse to moderate cover, value = 9; (4) herbaceous vegetation or low shrub, moderate to dense cover, value = 3; (5) high shrub or forest, moderate to dense cover, value = 4; (6) marsh, sediment, value = 11; (7) marsh, vegetation or unknown cover, value = 7; (8) marsh, high shrub or forest, value = 14; and (9) development, value = 10. Note, this list contains all possible values in the supervised classification we considered within the larger research program (see Zeigler and others, 2019). Not all values may be represented in this supervised classification specifically.

Geomorphic settings raster (Cobb14_GeoSet.tif)
We created the geomorphic settings raster dataset (Cobb14_GeoSet.tif) by first delineating the boundaries of individual features (e.g., the edges of washover fans), combining these features into a single shapefile, and converting that shapefile to a raster. Each individual geomorphic feature was given a value for identification purposes (‘value’) as well as a priority number (‘priority’) on which to base the merging of all features into a single layer.
For additional information and example figures, see Zeigler and others (2019). All steps were conducted in ArcGIS and ArcToolbox version 10.4.1.
We began by hand-digitizing the boundaries for marsh (value = 6, priority = 6) based on the visual inspection of the orthoimage. The rear-edges of marsh boundaries were drawn such that they extended out past the study area boundary and included areas of open water that were directly across from marsh platforms.
We created a polygon for beach (value = 1, priority = 5) using the shoreline (cobb14_shoreline.shp in the larger work) and custom study area boundary polygons. The custom study area boundary spans the entire coverage of the Cobb Island site relevant to the broader research program (see Zeigler and others, 2019 for more details). The boundary polygon’s spatial extent does not affect the quality or completeness of the dataset.
To create the beach polygon, we used the ‘Erase’ tool in ArcToolbox to mask out the shoreline polygons and the hand-digitized marsh polygons from a copy of the full study area polygon. This process essentially removed the features of marsh and anything interior of the shoreline from the study area boundary, leaving only the beach geomorphic setting in the final polygon. Together, the beach and marsh polygons covered the outermost edges of the study area.
We also hand-digitized the boundaries of dunes (value = 3, priority = 7) using the DEM (displayed in 1 m intervals), dune crest and dune toe points (cobb14_DCpts.shp and cobb14_DTpts.shp in larger work and Doran and others, 2017), and the orthoimage as guides. We digitized the rear of dune complexes such that the boundary fell landward of dune crest points (if present) and along a clear decrease in elevation (according to visual inspection of the source DEM). The front of the dune was also drawn such that the boundary fell along a clear decrease in elevation and passed through dune toe points. We used the ‘Snapping’ feature in the Editing toolbar of ArcGIS to ensure that the front boundary of dune complexes coincided with the dune toe points. In some instances, particularly in washovers, dune crest points were present without obvious changes in elevation (in the DEM) or the presence of dune toe points. In these cases, we hand-digitized the dune boundary such that it tightly encompassed dune crest points.
Boundaries of the remaining geomorphic features were hand-digitized according to visual inspection of the orthoimage and in reference to boundaries already created for beach, marsh, and dunes. The rear, landward boundary of the backshore (value = 2, priority = 4) was digitized such that it overlapped with dune polygons (but did not extend past the landward-most boundary of the dune). When dunes were not present, the landward boundary of the backshore was drawn where non-vegetated sand ended and dense vegetation began. We extended the backshore polygon boundary seaward beyond the shoreline (or the boundary of marsh in the case of the back-barrier).
The characteristic fan-shaped boundaries of washovers (value = 4, priority = 3) were hand-digitized such that they overlapped with the boundaries of marsh along the back-barrier and with the boundaries of dune complexes on the seaward shore. In instances where dunes were not present seaward of a washover, we digitized the washover boundary along wrack if visible in the orthoimage.
We hand-digitized polygons for ridge/swale complexes (value = 7, priority = 2), when present, at the boundary of open sand and vegetation according to the orthoimage. Note: this geomorphic setting was not present at all sites and may not be represented.
For barrier interior (value = 5, priority = 1), the final geomorphic setting, we used the Erase tool in ArcToolbox to remove all other geomorphic settings (i.e., beach, backshore, dune, washover, marsh, and ridge/swale) from a copy of the study area polygon (not published). In this way, the barrier interior setting occurred only in the absence of any other geomorphic features.
Using the Merge tool in ArcToolbox, we combined the individual polygon shapefiles for beach, backshore, dune, washover, barrier interior, marsh, and ridge/swale into a single shapefile. We used the Integrate tool in ArcToolbox (distance = 1 m) to close small gaps between polygons.
We then converted the merged polygon shapefile to a raster with a 5 m cell size, using the priority value for each geomorphic setting type in the attribute table to determine which geomorphic setting type took precedence when two or more settings overlapped (where a raster cell takes the value of the feature with the highest priority value). We selected ‘maximum area cell assignment’ to determine cell values. The extent was set to the study area extent in the geoprocessing Environment Settings window so that the converted raster would be clipped to the extent of the study area.
The final raster provided a categorical map of geomorphic features with every 5x5 m cell categorized as one of the seven possible geomorphic settings.

Reclassification for the substrate type layer (Cobb14_SubType.tif):
All steps were undertaken in ArcGIS and ArcToolbox version 10.4.1. For additional information and example figures, see Zeigler and others (2019).
Each of the classes contained in the supervised classification is associated with a substrate type, vegetation type, and vegetation density for later use in creating raster coverages for these characteristics. For substrate type, we reclassified the original supervised classification value to the following substrate types and associated values:
(1) water -> Water, 4444; (2) sand -> Sand or ShellGravelCobble, 7777; (3) herbaceous or shrub, sparse to moderate cover -> Sand or ShellGravelCobble, 7777; (4) herbaceous or shrub, moderate to dense cover -> Sand or ShellGravelCobble, 7777; (5) shrub or forest -> Unknown, 9999; (6) marsh, sediment -> Sand or MudPeat, 1000; (7) marsh, vegetation or unknown cover -> MudPeat, 3333; (8) marsh, shrub or forest -> Unknown, 9999; (9) development -> Development, 6666.
Note, this list contains all possible values for substrate types considered within the larger research program (see Zeigler and others, 2019). Not all values may be represented for this site.
We made one manual change to the reclassification of the supervised classification to create the substrate type layer. For raster cells that were classified as ‘beach’ in the geomorphic settings layer (Cobb14_GeoSet.tif), we manually reclassified the substrate type for these cells to ‘Water’ with 'None' for both vegetation density and vegetation type. The beach geomorphic setting represented all raster cells outside of the shoreline polygons (cobb14_shoreline.shp in larger work), and we assumed that these areas would be underwater at some point depending on the tide. We reclassified everything outside of the full shoreline polygons as Water to maintain consistency in landcover classifications relative to tide fluctuations.

Reclassification for the vegetation density layer (Cobb14_VegDen.tif):
All steps were undertaken in ArcGIS and ArcToolbox version 10.4.1. For additional information and example figures, see Zeigler and others (2019).
For vegetation density, we reclassified the original supervised classification value to the following vegetation densities and associated values:
(1) water -> None, 111; (2) sand -> None or Sparse, 666; (3) herbaceous or shrub, sparse to moderate cover -> Sparse or Moderate, 777; (4) herbaceous or shrub, moderate to dense cover -> Moderate or Dense, 888; (5) shrub or forest -> Moderate or Dense, 888; (6) marsh, sediment -> None or Sparse, 666; (7) marsh, vegetation or unknown cover -> Unknown, 9999; (8) marsh, shrub or forest -> Moderate or Dense, 888; (9) development -> Development, 555.
Note, this list contains all possible values for vegetation density classes considered within the larger research program (see Zeigler and others, 2019). Not all values may be represented in this dataset.
We made one manual change to the reclassification of the supervised classification to create the vegetation density layer. For raster cells that were classified as ‘beach’ in the geomorphic settings layer (Cobb14_GeoSet.tif), we manually reclassified the substrate type for these cells as ‘Water’ with 'None' for both vegetation density and vegetation type. The beach geomorphic setting represented all raster cells outside of the shoreline polygons (cobb14_shoreline.shp in larger work), and we assumed that these areas would be underwater at some point depending on the tide. This also reduced misclassifications where beach wrack (or dead materials washed up on to the beach) was incorrectly identified as vegetation, which was a common issue in the supervised classification process. Thus, we reclassified everything outside of the shoreline polygons as 'None', or lacking terrestrial vegetation.

Reclassification for the vegetation type layer (Cobb14_VegType.tif):
All steps were undertaken in ArcGIS and ArcToolbox version 10.4.1. For additional information and example figures, see Zeigler and others (2019).
For vegetation type, we reclassified the original supervised classification value to the following vegetation types and associated values:
(1) water -> None, 11; (2) sand -> None or Herbaceous, 77; (3) herbaceous or shrub, sparse to moderate cover -> Herbaceous or Shrub, 88; (4) herbaceous or shrub, moderate to dense cover -> Herbaceous or Shrub, 88; (5) shrub or forest -> Shrub or Forest, 99; (6) marsh, sediment -> None or Herbaceous, 77; (7) marsh, vegetation or unknown cover -> Unknown, 9999; (8) marsh, shrub or forest -> Shrub or Forest, 99; (9) development -> Development, 55.
Note, this list contains all possible values for vegetation types considered within the larger research program (see Zeigler and others, 2019). Not all values may be represented for this site.
We made one manual change to the reclassification of the supervised classification to create the vegetation type layer. For raster cells that were classified as ‘beach’ in the geomorphic settings layer (Cobb14_GeoSet.tif), we manually reclassified the substrate type for these cells as ‘Water’ with 'None' for both vegetation density and vegetation type. The beach geomorphic setting represented all raster cells outside of the shoreline polygons (cobb14_shoreline.shp in larger work), and we assumed that these areas would be underwater at some point depending on the tide. We reclassified everything seaward of the full shoreline as Water to maintain consistency in landcover classifications relative to tide fluctuations. This also reduced misclassifications where beach wrack (or dead materials washed up on to the beach) was incorrectly identified as vegetation, which was a common issue in the supervised classification process. For these reasons, we reclassified everything outside of the shoreline polygons as none, or lacking in terrestrial vegetation.

Added keywords section with USGS persistent identifier as theme keyword.