1970s Shorelines for Vieques and Culebra, Puerto Rico

Images were collected from the Puerto Rico Highway Authority, Photogrammetry Office.
The images were georeferenced in ArcGIS Pro to extract the shoreline positions. Aerial images were purchased from the Puerto Rico Highway Authority. For Culebra, we received a total of 13 images. For Vieques, we received 33 images. The timeframe of the images is from 1964, 1972, and 1978. The scale of all the images was 1:20,000. The 2006-2007 orthophotos were used as reference. The ground control points (GCPs) for each image used varied according to the transformation method used: first order polynomial, second order polynomial, and spline. After the rectifications were completed, we checked each of the images for quality control.
A geodatabase (GDB) was created to mask or clip the images to the area of interest and create the final mosaic. The clipping of the images was done to erase black areas, labeled areas or undesired coverage improves the quality of the mosaics. To best classify the images, different datasets were used as reference: land cover and benthic areas. Also, the DEM was collected to calculate the slope. This helped to identify cliff areas in the classification.
GROUND CONTROL POINTS AND ROOT MEAN SQUARE ERROR (RMSE) The main control points were streets, highways, and bridges. However, on some occasions, some rocky features were used in areas where no anthropogenic structures were found. Also, due to bombing in Vieques and Culebra, another reliable source to rectify the images was the army bombarding target points, especially along the Vieques coast. To select the control points, the map scale in ArcGIS Pro was zoomed in between 1:100 and 1:500. A total of four points were selected per image in most cases. These points were well-distributed across the image. In other cases, more points were used due to the image quality. A maximum RMSE of 4.0 was achieved for each image.
All GCPs in text files were saved individually for backup and reference purposes. All images were exported as “.tif” before being added into the GDB.
TRANSFORMATION Different raster transformations were used to accomplish the best quality possible; these include first order polynomial, second order polynomial, and spline. Different transformations were applied based on the number of available GCPs.
REVISIONS The rectifications were revised several times to achieve the best possible RMSE while ensuring the aerial image matched closely with the reference orthophotos. The GCPs and different transformations were tested for the aerial images individually.
MOSAIC CREATION Once all the images were rectified and the best images were chosen, we mosaicked the images. First, an individual GDB was created for Vieques and Culebra. All images were inserted in the GDB.
Each of the images was masked before creating the final mosaics. By creating a mask of the images, we eliminated the black borders and labels on the photos. For these purposes, a polygon was created in the desired area to cut the image and the image was exported to a different GDB to create the mosaics. Once all images were clipped by the mask, the mosaic process was started.
Culebra Mosaic As mentioned, several attempts were conducted to create the final mosaics. For Culebra, all the rectified images were used. In some cases, the images were clipped twice to improve the quality of the transformation. For Culebra, the VORONOI transformation was chosen. After testing other methods, the Voronoi transformation was optimal as it divides the images into different sections and analyzes the similarity in each of the images to construct the final mosaics. Therefore, the seamlines inside the mosaic dataset was created using the Voronoi effect (see Build Seamlines: https://pro.arcgis.com/en/pro-app/2.6/tool-reference/data-management/build-seamlines.htm). Due to technical limitations with the software, we created first the mosaic using the Voronoi and merged the resulting image with a small section that was not being exported from the original mosaic (see Create Mosaic Dataset: https://pro.arcgis.com/en/pro-app/latest/help/data/imagery/creating-mosaic-datasets-wf.htm; and Merge: https://pro.arcgis.com/en/pro-app/latest/arcpy/image-analyst/merge.htm). We collected each image RMSE and averaged these values to establish a final RMSE for the whole mosaic of 0.8309.
Vieques Mosaic Vieques aerial images mosaic was approached differently. In this case, each image was analyzed and were organized in a certain order using the field “ZOrder.” The ZOrder field is automatically included in the mosaic dataset and gives the user the ability to control the display order of the images. Most of the images were organized from west to east. We collected each image RMSE and averaged these values to establish a final RMSE for the whole mosaic of 0.804.

A single GDB and feature class was created in ArcGIS Pro to digitize the 1970s shoreline for Vieques and Culebra Shoreline. Shoreline types (SHORE_TYPE) and descriptions (SHORE_DESC) were used to increase the quality of the shoreline classification.
For the 1970s shoreline, 1964, 1972 and 1978 aerial images were used. The total shoreline digitized per year was: 1964 (2%), 1972 (49%) and 1978 (49%).
To best classify the shoreline, different datasets were used as reference: land cover and benthic areas. Also, the DEM was collected to calculate the slope. This helped to identify cliff areas in the classification. The land cover and the benthic areas were used to understand the shoreline areas.
The digitizing scale was at 1:500 and visualization of 1:1000. The digitization was conducted clockwise. The digitized line was continuous line with their respective exceptions (rocky barrier, sand bars, etc.). When digitizing the type of beach coast ("Beach"), the high-water line (HWL) will be used as a proxy (Boak and Turner, 2005). If the HWL is not visible, the shoreline proxies will be digitized following this order: Wet/Dry Line or Run Up Maxima, Groundwater Exit, Instantaneous Line.
To merge the digitization and avoid overlaps, transition line features were made when each person finished their assigned classification.
Digitization Decisions: Summary The main requirement to classify the shoreline features was to answer the question: “What is in contact with the highest water level?” Using this information, the shoreline was digitized as beach, concrete forms, vegetation, and others described in the attribute information. However, the variability of Vieques and Culebra's coasts do not make it possible to answer this question with a single answer for each type of coastline. Therefore, some decisions were taken to homogenize the digitization of Vieques and Culebra. One of the goals was to digitize a continuous line describing the coastal areas and the shoreline. Some exceptions were considered when digitizing rocky barriers, vegetation formations in the ocean such as mangroves, and rock formations off the coast. This summary was generated mainly from the inconsistencies that were found during the digitization process.
Beach High Water Line In beaches with coarser grain size, the HWL was defined following the line of sediment accumulated along the beach. For this analysis, the benthic layers (2000-2002 Benthic Habitat) helped to identify features like coral reefs and understand the granulometry of the beach. When bluff areas were seen, the HWL was digitized in the color contrast between the bluff and beach berm. If the HWL was not visible, areas with shadows were classified as “No Visibility>Shadows” When the HWL was not visible, and sand accumulation was seen in the shoreline, other proxies were digitized: instantaneous line, wet/dry line, or groundwater exit point. For beaches with terrigenous sand, the HWL was identified by the highest contrast between wet and dry areas. Tombolo sand formations were digitized using the instantaneous line or HWL indicators if visible. When sargasso was present, the continuous line of sargasso accumulated around the last highest tide mark was digitized, avoiding sargasso deposited by a storm or extreme event.
Rocky The coast is characterized as rocky when water is observed to impact the rock. The instantaneous lines were digitized in the Rocky formation. The DEM for 2015-2017 was used to calculate the slope in the coastal areas in order to classify the cliff areas. Areas with a slope greater than 40 degrees digitized as “CLIFF.” However, some cells between "cliff" areas could be 20 degrees. In cliff areas where the instantaneous line was seen on the ocean side, the area was digitized as "Bare Land" with a description of "Rocky land." For the shoreline type of "Rocky," the area can have a description of "Cliff," "Bare Land” or "Rocky Barrier." Rocky areas off the coast within a 300 meters buffer of the shoreline were digitized as “Bare Land” or “Cliff.” The slope information and Google Earth were used to help distinguish the vegetated shoreline amongnst rocky areas. Rocky Barriers The rocky barriers were identified as areas where rocky land and sandy beaches were divided by water. This feature was classified as a reference for further shoreline change analysis. To identity rocky barriers, the benthic layers helped to identify features like coral reefs. Pocket beaches mostly between rocky areas were mostly digitized using the instantaneous line proxy.
Anthropogenic Marina areas like boating docks were digitized up to 500 meters inland of the shoreline area. Tourism areas were classified using other sources such as Puerto Rico Land Cover 2010. When no information was available, the concrete forms were classified as “Housing.”
Vegetation To classify the vegetation, the land cover classification was followed. In small sandy areas around mangroves, the HWL was digitized to capture sand accumulation.
No Visibility Palms trees or shadows created by them were classified as “Shadows.” In cases where areas mostly shadows were present, even small sections of the HWL were digitized if visible. If the HWL or any other indicator was not visible, it was classified as "Other." When no orthophoto was available, the area was classified as "Not Surveyed." In areas where the HWL was not visible, but sand accumulation was seen, other proxies were digitized: instantaneous line, wet/dry line, or groundwater exit point.
Hard Structures When multiple hard structures were seen, it was classified as “Multiple Structures.” For example, when mitigation structures and construction filling areas were seen, it was classified as “Multiple Structures." Rip-rap covered by water was also digitized if visible.
Other Considerations River mouth areas were digitized up to 500 meters inland and where some of the water or vegetation connect the shoreline area. The sandy areas around the river mouth were digitized as HWL. If a continuous beach was seen along the river mouth areas, the river mouth was not digitized.

The geodatabase was transferred to USGS for final processing steps. The geodatabase feature class was exported as a shapefile (ArcToolbox >> Conversion >> To Shapefile >> Feature Class to Shapefile (multiple)), ensuring that the "Transfer field domain descriptions" was checked in the Environment Settings.
A "YEAR" field was added and calculated based on the IMG_DATE field so shorelines could be easily grouped by year.
Finally, the shapefile was projected (ArcToolbox >> Data Management >> Projections and Transformations >> Project) to WGS 1984 using the Puerto_Rico_To_NAD_1983 + Puerto_Rico_to_WGS_1984_4 geographic transformation.