Shapefile
Tags
intertidal oyster, Crassostrea virginica, NOAA, DNR, washed shell
Summary
This dataset was produced as part of the South Carolina Oyster Beds Imaging effort per USGS Contract # This dataset was produced as part of the South Carolina Oyster Beds Imaging effort per USGS Contract # 01CRCN0012 for NOAA, to support the mapping of intertidal oyster reefs for management of the state's shellfish resources.
Description
This dataset contains digitized boundaries of washed oyster shell deposits found along the South Carolina coast. The deposits were digitized through a combination of automated and manual techniques using 4-band (blue, green, red, near-infrared) digital orthophotos with a theoretical ground resolution of 0.25 meters. The photos are dated from 2003 to 2006. The project area was selected specifically to cover those sections of the SC coastal critical zone where oysters had historically been mapped by the SC Department of Natural Resources (SCDNR). The area spans 122 USGS quarter quadrangles (DOQQs). Sixty of the DOQQs were ground-truthed by boat to assess accuracy. Some areas were verified through photographs taken from low-altitude helicopter flights conducted from 2006 to 2008 by SCDNR. The initial digitization process has met with the minimum accuracy requirements of the project (80% correct classification) and was completed through a joint effort between Photo Science Inc. and SCDNR. The entire dataset has been reviewed by SCDNR for quality using all known information through 2010. Edits and improvements were completed by SCDNR on April 28, 2011. Updated versions of this dataset will be produced as more information becomes available. Information on accuracy and completeness of the data are contained in the Data Quality Sections of this report.
Credits
None. The South Carolina Department of Natural Resources should be acknowledged in products derived from these data.
Access and use limitations
Neither the State of South Carolina, the Department of Natural Resources, nor any of its employees, is responsible for any improper or incorrect use of the information described and/or contained herein, and assume no responsibility for the use of the information. The South Carolina Department of Natural Resources should be acknowledged as the data source in products derived from these data.
This dataset contains digitized boundaries of washed oyster shell deposits found along the South Carolina coast. The deposits were digitized through a combination of automated and manual techniques using 4-band (blue, green, red, near-infrared) digital orthophotos with a theoretical ground resolution of 0.25 meters. The photos are dated from 2003 to 2006. The project area was selected specifically to cover those sections of the SC coastal critical zone where oysters had historically been mapped by the SC Department of Natural Resources (SCDNR). The area spans 122 USGS quarter quadrangles (DOQQs). Sixty of the DOQQs were ground-truthed by boat to assess accuracy. Some areas were verified through photographs taken from low-altitude helicopter flights conducted from 2006 to 2008 by SCDNR. The initial digitization process has met with the minimum accuracy requirements of the project (80% correct classification) and was completed through a joint effort between Photo Science Inc. and SCDNR. The entire dataset has been reviewed by SCDNR for quality using all known information through 2010. Edits and improvements were completed by SCDNR on April 28, 2011. Updated versions of this dataset will be produced as more information becomes available. Information on accuracy and completeness of the data are contained in the Data Quality Sections of this report.
This dataset was produced as part of the South Carolina Oyster Beds Imaging effort per USGS Contract # This dataset was produced as part of the South Carolina Oyster Beds Imaging effort per USGS Contract # 01CRCN0012 for NOAA, to support the mapping of intertidal oyster reefs for management of the state's shellfish resources.
Neither the State of South Carolina, the Department of Natural Resources, nor any of its employees, is responsible for any improper or incorrect use of the information described and/or contained herein, and assume no responsibility for the use of the information. The South Carolina Department of Natural Resources should be acknowledged as the data source in products derived from these data.
None. The South Carolina Department of Natural Resources should be acknowledged in products derived from these data.
Internal feature number.
ESRI
Feature geometry.
ESRI
Each reef was identified based on the DOQQ imagery it was digitized from and the year of data publication. A "W" was placed in front of the ID number if the reef was initially characterized as dead washed shell (ex. adamrNE_W00001).
Area of feature in internal units squared.
ESRI
This dataset contains digitized boundaries of washed oyster shell deposits found along the South Carolina coast. The deposits were digitized through a combination of automated and manual techniques using 4-band (blue, green, red, near-infrared) digital orthophotos with a theoretical ground resolution of 0.25 meters. The photos are dated from 2003 to 2006. The project area was selected specifically to cover those sections of the SC coastal critical zone where oysters had historically been mapped by the SC Department of Natural Resources (SCDNR). The area spans 122 USGS quarter quadrangles (DOQQs). Sixty of the DOQQs were ground-truthed by boat to assess accuracy. Some areas were verified through photographs taken from low-altitude helicopter flights conducted from 2006 to 2008 by SCDNR. The initial digitization process has met with the minimum accuracy requirements of the project (80% correct classification) and was completed through a joint effort between Photo Science Inc. and SCDNR. The entire dataset has been reviewed by SCDNR for quality using all known information through 2010. Edits and improvements were completed by SCDNR on April 28, 2011. Updated versions of this dataset will be produced as more information becomes available. Information on accuracy and completeness of the data are contained in the Data Quality Sections of this report.
This dataset was produced as part of the South Carolina Oyster Beds Imaging effort per USGS Contract # This dataset was produced as part of the South Carolina Oyster Beds Imaging effort per USGS Contract # 01CRCN0012 for NOAA, to support the mapping of intertidal oyster reefs for management of the state's shellfish resources.
ground condition
None
Neither the State of South Carolina, the Department of Natural Resources, nor any of its employees, is responsible for any improper or incorrect use of the information described and/or contained herein, and assume no responsibility for the use of the information. The South Carolina Department of Natural Resources should be acknowledged as the data source in products derived from these data.
None. The South Carolina Department of Natural Resources should be acknowledged in products derived from these data.
The Accuracy of the mapped oyster reefs was assessed for 60 of the 122 DOQQs using mapping grade GPS and video transects captured by boat along the length of the reef. Approximately 100 reefs including some washed shell deposits were mapped per DOQQ, and 30 textured mud areas were mapped to check for errors of omission. The horizontal accuracy of the GPS measurements was 0.3-0.5 meters. The spatial accuracy of the images from which the reefs were digitized was 4 meters. Two metrics were used to assess accuracy. The first metric assessed was Presence/Absence, or the correct identification of oyster or textured mud. Four possible scoring categories were available for this metric: 1) Correct positive (correctly classified shell), 2) Correct negative (correctly classified as mud), 3) False positive (mud incorrectly classified as shell), False Negative (shell incorrectly classified as mud). If 25% of the measured transect contained the correct classification, it was scored as correct (Present). The second metric assessed was Extent. It was defined as the correct delineation of the entire length of an oyster reef or mud patch. The extent of the reef was considered incorrect if the length of the reef was different from the length of the transect by more than 10 meters in the case of a fringing reef. In the case of a hummock or "patch reef" if there was a gap within the reef greater than 10 m2, it was considered incorrect. It is important to note that although correct classifications tend to be 80% or above, these measurements are primarily for oysters along shorelines. Reefs in mud flats were difficult to assess by boat, so DOQQs containing large areas of flats could have poorer accuracy than reflected by the score. The flats were difficult to digitize due to the patchiness of the shell in the mud, and it was often hard to see. Manual editing is often required in these areas. Helicopter photos were used to correct some of the accepted product, especially in areas inaccessible to boats. Accuracy numbers for the helicopter process are not available, but the reefs are easily visible on the low altitude photos. For DOQQs not assessed for accuracy, SCDNR used their knowledge of the resource and photo interpretation skills to improve them as much as possible, but no accuracy numbers will be available. SCDNR is also correcting DOQQs that have been assessed for accuracy, so the accuracy scores reported should be considered as the minimum accuracy. Before SCDNR edits, the cumulative metric scores of this data set were 87% correct for Presence/Absence and 84% correct for Extent. Individual DOQQ scores ranged from 71% to 100% correct for Presence/Absence classification. Scores ranged from 50% to 97% correct for Extent. Accuracy values are available at the DOQQ level in the associated table: SCDNRoyster2010MetadataTable.pdf
Photo Science Inc. was responsible for oyster reef digitization of 75 DOQQs or 61% of the data. The remaining 47 DOQQs were mapped at SCDNR. The techniques used by Photo Science and SCDNR for initial digitization using Feature Analyst are similar with a few exceptions. Photo Science trained two types of oyster reefs, fringing and patch reefs (flats), and took the best training set from each to create one file with minimal manual editing. The resulting files were manually edited by SCDNR. When SCDNR digitized reefs, more training sets were created to better capture variations in the appearance of the shell. Instead of picking the best single output, SCDNR then used parts from all training sets to piece together the final product. This second method is more time consuming, but manual editing could be completed at the same time.
The initial digitization process has been completed for all contracted 122 DOQQs or 443 mosaiked images. All of the DOQQs have been manually edited by SCDNR to correct known and/or clearly visible errors. The edits were generally done at the DOQQ level, but low-altitude helicopter flights were conducted to photograph additional areas at the image level (one quarter of a DOQQ). The majority of the photographs from flights conducted between 2004-2008 have been used to assist with error correction. The editing status of each DOQQ is contained in the associated table: SCDNRoyster2010MetadataTable.pdf. Any additional information gathered from low-altitude photographs will be incorporated in a new data file.
The data was digitized from 0.25 m resolution multispectral digital imagery with reported horizontal accuacy of 4 meters or less.
Initial Digitization: Each photo (one fourth of a DOQQ), was interpreted and had its own training samples due to the variable appearance of the oyster and the imagery itself. Training polygons were created for each photo and input into Feature Analyst for automated feature extraction/creation. Feature Analyst is an extension developed by Virtual Learning Systems for ESRI's ArcGIS desktop. Feature Analyst Versions 4.0-4.2 were used and the ArcGIS versions ranged from 9.0-9.2. Training sets were developed separately for fringing reefs, patch reefs and washed shell due to the difference in appearance of the three types of shell. Starting in June 2007, additional training sets were also used when the color or texture of shell differed widely within an image due to: light angles, dryness of the shell, or degree of contrast with surrounding mud. Masks were created to eliminate dry land and large areas of water from analysis. Within Feature Analyst's Set-Up Learning step, the Image Resolution was set at 1 foot. The Manhattan 5x5 Input Representation was determined to be the most versatile and used for most extractions. Within Learning Options, "Aggregate areas" was set at a minimum area of 64 pixels and "Find rotated features" was checked. All primary extractions were run with the learning algorithm "Approach 1". When layers were filtered for incorrect features (clutter), "Approach 2" was used. Manual editing or clean-up of shapes was not generally done by Photo Science. The manual clean-up step was completed by SCDNR and is listed as a separate step.
Ground-truthing by Boat: Sixty of the 122 DOQQs were ground-truthed by boat for accuracy assessments. Using a Trimble XR GPS unit, approximately 100 reefs in each DOQQ were measured. The areas had to be assessed before the imagery was available, so efforts were made to randomly select reefs spread out over the DOQQ using historical maps of oyster as a guide. Some DOQQs did not have 100 accessible reefs to measure, so in these cases as many as were accessible were measured. Reefs also had to be 10 m2 or more in size per the contract agreement with Photo Science. Using the GPS unit, transects were measured lengthwise for each reef. A video of the entire length of the reef was also filmed as the reef's approximate width and condition were recorded. In addition to the oyster reefs, approximately 30 areas of textured mud or sand were marked to check for errors of commission. Ground-truthing efforts were conducted no more than two hours before or after low tide. Data collected from these efforts were used to score the accuracy of the digitization process.
QA/QC for Accuracy: The accuracy of the product was scored using two metrics: 1) the presence/absence of oyster adjoining each transect measured, an 2) the extent of the reef that was captured. For presence/absence, a matrix of four possible scoring categories were available: 1) Correct positive (correct shell), 2) Correct negative (correct mud), 3) False positive (mud incorrectly classified as shell), False Negative (shell incorrectly classified as mud). If 25% of the measured transect contained the correct classification, it was scored as correct. The score reported for presence/absence was calculated as the number of correct classifications divided by the total number of observations. To score extent, if there was an error in extent of more than 10m, it was scored as incorrect. If the extent of the mapped reef exceeded that of the field transect, we reviewed the video records. If the tide in the video appeared to be covering shell that was captured on the photo, the length was not graded. Oyster reefs that extended past the transect could be graded as correct if notes were made in the field record about the inaccessibility of the remaining shell from the boat. Reefs that extend past a transect can only be graded as incorrect if there is no doubt on the error. Extent was scored as number of correct lengths divided by total number of observations. These scores are listed in an associated table (SCDNRoyster2010MetadataTable.pdf) for the corresponding DOQQs. The images were processed based on batches of 15 DOQQs. Half of the DOQQs in each batch were ground-truthed. For each DOQQ scored, the two metrics were averaged for an overall score. If the cumulative score of the batch was greater than 80%, it was accepted. If not, the failing DOQQs were returned for reprocessing, as well as any images that were not ground-truthed and had obvious problems. Individual DOQQs that failed in an "accepted batch" were edited by SCDNR.
Edge Matching: Edge matching of the shellfish polygons was performed on both live and washed shell layers produced from the Feature Analyst software for each DOQQ (Digital Ortho Quarter Quad). There were three steps to editing these layers. Step A: Merging the QQQ shapefiles to QQ feature classes within a geodatabase, Step B: Editing within DOQQs, and Step C: Working between DOQQs. For each of these steps, the washed shell layer was processed before the live shell layer. Step A: Feature Analyst produced a series of shapefiles for each DOQQ. These were merged into two feature classes (one washed, one live) per DOQQ (e.g. NW, NE, SW, SE), and they were merged into a geodatabase. Step B: Starting with the washed shell, the polygons were checked for obvious errors (e.g. polygons on houses, streets). Then the interior lines of each DOQQ were checked for overlapping or duplicate polygons. These polygons were either merged or one was deleted. When choosing between polygons, we chose the one the more attribute information (e.g. helicopter data) or the one associated with the photo within the same DOQQ. Then the washed shell layer was used to clip the live shell layer using the Erase tool. The wash and live shell layers were both checked for polygon size. Any polygons smaller than 1 meter were deleted. Also any polygons with a gridcode of 0 AND no neighboring 'ADD' comment field were deleted. Finally, after the wash and live layers had been edited, a new ID column was created that identified the quad, the quarter quad, the year, and whether it was live or wash shell (e.g. adamrNE2008oysterW). Step C: Starting with the wash shell, several adjacent DOQQs were opened and compared. There were three possible issues to fix. First, any polygons that crossed DOQQ boundaries were cut and then were appended into one or the other DOQQ wash/live feature class. Second, due to the fact that polygons were digitized from several photos, some polygons were associated with the wrong quad. Any polygons that appeared in the wrong quarter quad were selected, appended to the proper file, and then deleted out of the original file. Finally, some polygons were duplicated or overlapping in multiple quarter quads. The polygons were either merged or one of the duplicate polygons was deleted. If necessary, the polygons were then appended into the proper quarter quad feature class. When choosing between polygons that represent the same shellfish bed, we always choose the polygon that was associated with the photo in that quad, not the adjacent quad and photo, unless the combination of the polygons better represented the shellfish bed, in which case, we merged the two polygons.
Low Altitude Helicopter -Based Validation: Photographs have been taken of areas not accessible by boat or not previously ground-truthed. The photos were taken from a helicopter at an altitude of 200-400 feet. A Trimble Pro XRS GPS antenna was attached to the helicopter to collect continuous points along the path of the helicopter. The photos were collected using a Canon EOS 30 D digital camera with 10 megapixel resolution. All photos were collected during a negative low tide within +/- 1 hour of noon to reduce shadow effects. Using the GPS points as a guide and going through the chronological file of photos, all reefs along the flight patch were checked and edited in ArcGIS 9.2 for accuracy. Photographs were taken from 2004-2010 and were used to validate oyster presence/absence. Photographs used to validate this dataset were dated between 2004-2008, however, not all available photographs from these years have been viewed to date. Information pertaining to areas flown and validated using these methods are available in the associated file: SCDNRoyster2010MetadataTable.pdf.
Manual Edits and Updates of Initial Digitization: SCDNR manually checked and edited all products completed by Photo Science using known information about oyster resources and through visual interpretation of the image. ArcGIS 9.2-9.3 was used. Only clear errors were corrected, so as not to decrease any accuracy scores an image received during the QA/QC assessment. Visual interpretation was aided through comparisons with low altitude oblique aerial imagery available through Bing Maps from 2008-2010, and with SCDNR photographs taken by helicopter (see the Low Altitude Helicopter-Based Validation step). Edits were finalized for this version in April 2011. Any additional edits to the data will be contained in a new file.
Metadata imported.
For additional information, refer to DNR web site: http://www.dnr.sc.gov/gis.html
In no event shall the creators, custodians, or distributors of this information be liable for any damages arising out of its use (or inability to use it).