CALIFORNIA DEPARTMENT OF FISH AND GAME GIS METADATA California Coast Kelp Survey -- 1999 METADATA FILE NAME: dfgkelp_1999_readme.txt METADATA DATE: August 19, 2002 Range of Data Acquisition Dates/Times: Beginning Date: 19991013 Ending Date: 19991222 Geospatial Data Presentation Forms: Aerial and Satellite Imagery, ArcView shapefiles Direct Spatial Reference Method: Raster, Vector Raster Object Type: Grid Cell Cell Size: 2 meters Vector Object Types: Polygons, points PROJECT DESCRIPTION Kelp beds along the coast of California are a critical habitat for many important sport and commercial species of invertebrates and fishes. Kelp is also harvested by commercial and sport concerns. Until now, kelp bed area figures used in harvest regulatory processes by the State have been based upon hardcopy maps, created from 1989 aerial photography, displaying hand drawn delineated features of the kelp canopy (see \metadata\kelp1989work.txt). The Fall 1999 project digitally remeasured the 1989 kelp maps and established new digital methods to calculate the area of the 1999 kelp canopy based on aerial, color-infrared photographs. The objective is to create a baseline that could be used to assess the effects of current and future use of coastal kelp. This document describes various procedures and datasets assembled to-date. DATA ORGANIZATION The State of California Fish and Game Commission established and mapped kelp bed boundaries for the purpose of administering kelp harvest leases and harvest regulations (Title 14, California Code of Regulations, Section 165.5). Subdirectories under the 1999 main project directory were established to store various kinds of datasets. These are described below. Directory: KELP_1999 Subdirectory Name: Description CoastPhotos: Contains scanned, georeferenced images of 1999 coastal aerial photography that did not contain kelp. Images of photos with kelp are stored in separate subdirectories as described below. Images: Each administrative kelpbed is a directory, i.e. akbed001, containing photos, classified kelp, mask files, control points, area stats and files associated with that particular kelpbed Shapefiles: All associated ArcView shapefiles, such as kelp bed boundaries, actual kelp canopy delineations, photo-center indexes, etc., for use with ArcView GIS software. Assoc_imagery: other images (i.e. satellite) associated with project. Metadata: All metadata (documentation) text files, spreadsheets, etc. Tide_stages: Graphs illustrating the tide stages at the time the photography was acquired. Wave_data: Wave height and wind speed data from the National Data Buoy Center at the time of photo acquisition. OceanImaging: Sample of digital, multi-spectral video imagery of Point Loma, La Jolla, and Carlsbad area kelp beds. Prepared for DFG by OceanImaging, Inc. (October 22, 1999). ===== PROJECT ENTITIES 1) PHOTOGRAPHIC DATA SOURCES AND PROCEDURES 2) KELP CLASSIFICATION METHODS 3) KELP AREA STATISTICS 4) DATA QUALITY 5) APPENDIX -DATA CONTACTS -SPATIAL REFERENCE INFORMATION -RELATED DATASETS -METHODOLOGY DOCUMENTS -LESSONS LEARNED 1) PHOTOGRAPHIC DATA SOURCES & PROCEDURES Kelp aerial photography was acquired from three sources: DFG Air Services: Aerial photos were taken in October, November, and December of 1999 with a medium format (60 mm) Hasselblad camera with an 80mm lens from an altitude of 11,000 feet above sea level. A Global Positioning System receiver (Garmin GPS12XL) was used to capture the aircraft's position at the exposure station of each photo. The developed photo frames (scale 1:41000 or 1" = approximately 3400', color, infrared transparency) were annotated with consecutive ID numbers. Every fifth frame was additionally labeled with local time and date of exposure (from the GPS data). An ArcView shapefile (see \shapefiles\KELP99FG.shp) including the plotted photo points and table of photo data was created to index all photography. A logical (yes/no) value was also added to the table for identifying kelp presence or absence in each photo. I.K. Curtis Services Aerial photos were taken in October and November of 1999 with Wild RC10 and RC20 aerial mapping cameras (industry-standard, 6" lens with 9"x9" film format). The resultant color infrared contact prints were supplied to DFG labeled with date, original scale (1:24000 or 1"=2000'), time, and photo ID. A GPS-based, photo-center index was also supplied by the contractor. DFG converted the index into an ArcView shapefile (KELP99IK.shp) and added photo ID numbers and other data (as above) to the attribute table, including the administrative kelp bed number (see \metadata\kelpadmbed.txt, \shapefiles\kelpbed2.shp) that a given photo covered all or a portion thereof. Pacific Aerial Surveys Photos were taken in November 1999 with an industry standard Zeiss RMK TOP mapping camera (6" lens, 9"x9" format). The resultant color, infrared contact prints were supplied to DFG labeled with date, scale (1:24000), time and photo ID number. As in the previous two photography sets, an ArcView point shapefile was created (\shapefiles\KELP99PA.shp) to index each photo; the date and time were added manually to the table as well as a logical (yes/no) value for kelp presence/absence per frame. AERIAL PHOTO SCANNING PROCEDURES DFG photos were scanned at 700 dots-per-inch (dpi), whereas the IK Curtis and Pacific Aerial photos were scanned at 400 dpi. All were saved as standard .jpg files. While the scan resolution was greater (and the original scale smaller) for the DFG photos versus the contractor photos, the approximate ground pixel size is 2 meters for all photography. The raw digital images were saved to CDROM disks (see \metadata\cd_photolist.xls). Processing priority was given to kelp-bearing photos, so presently, there are some gaps in the coastwide, digital versions of the photography. GEOREFERENCING THE AERIAL PHOTOGRAPHS The process of using digital aerial photography in a GIS requires a processing step known as registration or georeferencing. Aerial photographs have a nominal scale, but that scale is not constant throughout the frame. This registration process attempts to correct for scale changes, as well as to add an earth-coordinate system, thereby allowing other digital layers with similar spatial coordinate systems to be co-registered with the photography. There are several methods for performing the registration process and various levels of accuracy associated with each method. The 1999 kelp survey employed two registration methods. In the initial phase of the project, photogrammetric processes were used ERDAS Orthobase, version 8.4. This process employs camera parameters(see \metadata\camera parameters), camera orientation and position, and corrects for scale changes due to terrain elevational changes. Excellent results in the spatial accuracies throughout the photos were achieved with this technique. The process was difficult to implement because of the nature of the photos, which were kelp-centered. This off-shore centering of the kelp photography resulted in a minimal amount of land along the edge of the photos having identifiable features suitable as control points. To compensate for the control point alignment along the edge of photos and parallel to the flight line, extra control points were selected which required more processing time. Some of the pattern-matching algorithms were not as automatic as hoped, so manual selection of additional points (known as tie points) was required. Typical Root Mean Square Errors (RMSE) on the control points was 1 - 2 meters. Erdas Orthobase was used to perform the photogrammetric corrections. As an alternative approach, an affine transformation was used (ArcView v3.1/Image Analysis v1.0). This reduced the preparation time because fewer inputs were required, the user interface was easier to operate, and because the image would shift onto a basemap in real time as control point selections were made, shortening the feedback loop from input to results. Control point selection was limited to near sea level to help ensure the target of interest (the kelp forests surface canopy) was in the same plane as were the control points. Each control point was saved in a shapefile that also shows RMSE for each point and total RMSE value. Typical values were in the range of 4 - 12 meters. 6 - 15 points were selected from each photo. The photo registration was less accurate on land areas, particularly areas with steep terrain, than along the shoreline and in kelp areas. IMAGE NAMING CONVENTIONS 1) IK CURTIS: Coastal Mainland = CALKELP.img; Islands = .img 2) DFG AIR SERVICES: <"k" for kelp>.img; i.e. k99fg200.img 3) Pacific Aerial: pa.img 4) Orthobase corrected images: ortho->prefix followed by normal naming convention. 5) Control points = either by image name or "con_pts".shp DFG prepared a separate ArcView shapefile for each series of photos. A polygon (or "wire-frame") delineating the "footprint" or net coverage of each photo frame was screen-digitized after each photo registration was completed. These wire-frame indexes were prepared in addition to the initial photo-center-only spot indexes described above. K99FG_INDX.shp = wire-frame index of each georeferenced DFG Air Services photo K99PA_INDX.shp = wire-frame index of each georeferenced Pacific Aerial Survey photo K99IK_INDEX.shp = wire-frame index of each georeferenced I.K.Curtis photo Latest Index (work in progress): k99_index_all_images4.shp = wire-frame index of each georeferenced I.K.Curtis photo 2) KELP CLASSIFICATION METHODS: IMAGE NAME: k99aknnn_4cls.img, where nnn = 3 digit numeric DFG Administrative Kelp Bed Identifier. COVERAGE DESCRIPTION These ERDAS images represent kelp extent within each Administrative Kelp Bed, identified by number in their respective filenames. ERDAS Imagine and ArcView Image Analysis were used to create an iterative, self-organizing clustering (Isodata) process to segment the image based on reflected light signatures. These signatures were then interpreted for kelp canopy. The parameters used in the classification include 20 spectral clusters, .95 convergence, and initial classes to be allocated along the axis of greatest variability (principle component). Twenty spectral classes means that the "image colors" in and around the kelp bed are divided into twenty different units. Because the algorithm is iterative in nature, the spectral locations of the clusters are adjusted with each iteration to maximize the number of pixels contained close to a spectral cluster center. When the amount of adjustment is minimized to meet the .95 convergence and the twenty classes are created, the iteration stops. The classes were evaluated for statistical separability. Some adjacent classes were not separable, which is not uncommon in this type of routine. However, the twenty classes will be merged into 4 classes. The 1989 data was derived from methods based on already interpreted maps and methods using a binary level of kelp or not kelp. The following describes the initial 20 class values and an interpretation based on examining the photos, and 1989 kelp information and the comparison of how those 20 classes were merged in order to generalize the classification scheme to 4 classes. As indicated by the initial scheme, more classes were created than were needed. This was done intentionally so that the classes would have a smaller variance, understanding that class merging would take place. Twenty Classes Four classes Class# Description Class# Description 1-9 Open Water 1 Open Water 10 Submerged Kelp 2 Low Density or Submerged Kelp 11 Open Water 1 Open Water 12 Submerged Kelp 2 Low Density or Submerged Kelp 13 Water w/ high sediment 1 Open water 14 Submerged Kelp 2 Low Density or Submerged Kelp 15 Submerged Kelp 2 Low Density or Submerged Kelp 16-18 Low-Medium Density 3 Medium Density or or submerged. submerged. 19-20 Dense Kelp at Surface 4 Dense Kelp at Surface. The clustering routine was initially run on mosaicked images in southern beds. In northern beds, too much variability between photos, due to higher sediment in the water column, prevented this approach. The clustering was run on each frame, resulting in a set of twenty classes that had unique interpretations. The twenty classes were grouped into four and then mosaicked. Final steps involved a conversion to polygon format in order to test some statistical properties. To prepare for the conversion, the reclassified image (4 classes) was filtered. Any isolated pixels in "bunches" of less than 8 contiguous pixels were set to background using the ERDAS sieve function. This process was only performed on bed 3 only. Some statistics were created for patch size analysis. DATA DICTIONARY Gridcode: Kelp density class described above. Value range 0 - 4 (0=background). meters_sq: Area of the polygon in square meters. lnmeters: Natural log of square meters, only for high density kelp, class 4. IMAGE NAMING CONVENTION Photo#_20cls.img = 20-class image Photo#_4cls.img = 4-class image k99_4cls_mos.img = Final classified image for bed 3) KELP AREA STATISTICS Area statistics are an intrinsic aspect of a GIS database. For the 1999 kelp survey, the data are in a raster format. The process of calculating area is simply a product of the number of pixels of kelp times the area of each pixel (4 square meters). Conversion factors were applied to convert square meters to acres and square miles. All of these values are calculated in the GIS environment using ERDAS Imagine. A Microsoft Excel spreadsheet was developed to compile the area statistics for all kelp beds (see\metadata\k99ak_areastats.xls). Area is reported by administrative kelp bed number, and by kelp density, in acres and square miles. These values were exported from the Erdas Raster Attribute Editor to a comma delimited text file, then imported into a spreadsheet. A total of kelp (all densities) is recorded for each kelp bed. A second spreadsheet (see\metadata\kelp1989_1999stats.xls) shows the area of kelp by bed, as surveyed in 1989 and 1999. The reader is cautioned against making direct comparisons between the two surveys for the following reasons: 1) Timing of the survey is important, particularly with respect to growing season conditions in the ocean, and storms and harvest levels preceding the dates of survey photography. Seasonal variability may account for differences in surveys, which may not reflect a change in the bed's extent, productivity, or harvest level. 2) Statistical significance in change of area should be evaluated. To do this, a variance parameter is needed, which is obtained by repeated measurements. For example: Data for the Palos Verdes Peninsula is available, on a 4 survey per year interval, dating back to 1974 (see metadata\areastats_palosverdes.xls). Most of the coastline, however, has been surveyed only on three occasions (1967, 1989, and 1999). 3) Survey methods have not been consistent. Some method of calibration between the methods needs to be performed in order to insure a change of area is not due to survey instrumentation, and not misinterpreted as a biological change. As an example, administrative kelp bed 205 was surveyed both by Ecoscan using manual techniques and by PG&E using digital techniques. The photos were acquired in 1989 for both surveys. The Ecoscan survey calculated 153 acres of kelp. In contrast, the PG&E survey calculated 88 acres of kelp. We believe this difference in area is due to methodology differences. (see \metadata\dcpp\dcpp.txt) 4) DATA QUALITY The following are subjective remarks regarding the image data quality. The quality of the 9 x 9, large format aerial photography is excellent. The quality of the 60mm, medium format photography ranges from fair to excellent due to weather and other logistical problems. Shapefile indexes, spatial accuracy, and attribute completeness is very good, except for latitude/longitude values in some point attribute tables. The quality of the georeferenced images is slightly degraded in spatial resolution due to a required resampling and smoothing of the source image during the registration process. Viewing of the original film under magnification will always be superior to digital imagery in terms of interpretive detail available. Accuracy of registered images using ArcView Image Analysis extension is assessed based on the Root Mean Square Error (RMSE) of the affine transformation used to reproject the images. This value was in the range of 4 - 12 meters. Users may encounter greater registration errors withing the coastwide photography set. Control points were selected at sea level or as near to sea level as possible. Selection of control points in elevations above sea level introduced greater errors in the target areas of interest, the kelp, due to terrain-based photo distortions. Large (tens of meters) horizontal registration errors may be detected in upland areas of the photography. Accuracy of the registered images using ERDAS Orthobase is much better. RMSE is in the neighborhood of 1-2 meters in shoreline and kelp areas. Orthobase uses a block bundle adjustment, which is a very rigorous photogrammetric model. This method allows for the selection of a greater variety of control points, corrects for lens distortion, camera attitude, and terrain distortion effects. Spatial errors even in the upland areas of the photos were minimized. 5) APPENDIX DATA CONTACT Fred Wendell California Department of Fish and Game Nearshore Ecosystem Manager Marine Region 20 Lower Ragsdale Drive Monterey, CA 93940 (831) 649-2893 fwendell@dfg.ca.gov DFG MARINE REGION GEOGRAPHIC INFORMATION SYSTEM CONTACTS Nancy Wright California Department of Fish and Game GIS Analyst Marine Region 20 Lower Ragsdale Drive Monterey, CA 93940 (831) 649-2942 nmwright@dfg.ca.gov Paul Veisze California Department of Fish and Game Regional GPS/GIS Coordinator Information Technology Branch GIS 1807 13th Street, Suite 201 Sacramento, CA 95814 (916) 323-1667, Fax: -1431 pveisze@dfg.ca.gov SPATIAL DATA ORIGINATORS Alan Kilgore California Department of Fish and Game GIS Research Analyst II Information Technology Branch 1807 13th Street, Suite 201 Sacramento, CA 95814 (916) 445-6264, Fax: (916) 323-1431 email: akilgore@dfg.ca.gov Mark Lampinen California Department of Fish and Game GIS Research Analyst I Office of Spill Prevention and Response 1700 K Street, Suite 150 Sacramento, CA 94244-2090 (916) 322-4777, Fax: (916) 324-8829 mlampinen@dfg.ca.gov Carol Clark California Department of Fish and Game GIS Student Assistant Information Technology Branch 1807 13th Street, Suite 201 Sacramento, CA 95814 Fax: (916) 323-1431 cclark@dfg.ca.gov SPATIAL REFERENCE INFORMATION Datum: NAD 27 Projection: Albers (standard Teale parameters) Units: Meters 1st Std. Parallel: 34 00 00 N 2nd Std. Parallel: 40 30 00 N Longitude of Origin: -120 00 00 W Latitude of Origin: 00 00 00 False Easting: 0.0 False Northing: -4,000,000 RELATED DATASETS 1) \metadata\klpareachrts67_89(folder) = images of scanned bar charts comparing kelp bed areas from 1967 vs. 1989 2) \shapefiles\kelpadmbed.shp = administrative kelpbed boundaries, their associative CCR's, status and ID number (see \metadata\kelpadmbed.txt) 3) \tide_stages(folder)= graphs illustrating the tide stages at the time the photography was acquired 4) \assoc_imagery\adminkelp\southcoast_kelp.tif and central_coast.tif = administrative kelpbed boundary maps, scanned and georeferenced from original hardcopy 5) \metadata\drg24.txt = metadata for USGS 7.5 min quadrangles used as base maps for geo-referencing(see \shapefiles\drg24.shp) 6) 1989 kelp canopy grids: \kelp_1989\grids\k89akcp where n=administrative kelp bed number (see\metadata\kelp89work.txt) 7) \assoc_imagery\tm = Landsat thematic mapper satellite imagery acquired in summer of 1999. (see readme_L7_ETM.txt) Kelp beds display well in a false color infrared (band 4,3,2)at 30 meter resolution, or in the panchromatic (wideband truecolor, and near infrared) at 15 meters. 8) \shapefiles\kelpbed_masks.shp: ArcView polygons created by digitizing the edges of the kelp boundaries as viewed in color infrared aerial photos. The masks were created as a step during the processing of the infrared aerial photographs. After the photographs were georeferenced, the masks were digitized to limit the analysis to only those areas of the photographs containing kelp. 9) metadata\areastats_palosverdes.xls = kelp survey statistics for Palos Verdes Penninsula. The range of dates is 1974 - 1999. Survey frequency: quarterly 10) \metadata\kelp1995ceqa(folder): Prepared in accordance with the California Environmental Quality Act (CEQA), these files describe the kelp regulation process, environmental setting, environmental impacts, mitigation, alternatives, and conclusions. 11) \wave_data (folder) = wave height and wind speed data from the Monterey NDBC buoy station in the last quarter of 1999 used to investigate a correlation between amount of kelp showing in photography to damaging wave action at the time METHODOLOGY DOCUMENTS 1) \metadata\Classification of Kelp.doc = describes step by step procedures of classification to 20 classes 2) \metadata\k99reclass_20_4.doc: Document describing kelp classification recoding, how to reduce the number of kelp classes from 20 to four 3) \metadata\ortho_steps_rev2.xls: Orthobase Photo Registration Process LESSONS LEARNED The methods used during this project are much different than previous kelp area projects. Care should be used in comparing the acreages acquired by varying types of data interpretation. There is great variability within kelp area as recorded by color infrared film. This can be attributed to seasonal changes, tidal changes, water clarity, timing of the aerial photography in relationship to storms and kelp harvesting. Also the identification of kelp from aerial infrared photography should be checked with field observations. The photographs are difficult to georeference when the percentage of ocean area is greater than visible shoreline. Accurate GPS center points of the images would decrease the time involved in processing. The time involved in the scanning, georeferencing and categorization of the color infrared images needs to be compared to other methods of imagery acquisition including satellite imagery. The satellite imagery would contain a larger footprint of the coastline resulting in fewer images to process. New satellites are capable of providing a one to four meter resolution image, much higher than previous satellite imagery which was in the 30-meter range. Landsat satellite imagery, which was acquired in the summer of 1999, was used to cross-reference and check kelp mapping from the IK Curtis photos. Even though the 30-meter resolution was lower on the satellite imagery, than that of the scanned aerial photos, the kelp signatures were very distinct and in some cases pointed to an under representation by photography acquired in late November. Two, and in some areas, three sets of aerial photographs were acquired. This turned out to provide a good means to replicate, or cross-check, individual findings within the survey. Initial visual comparisons of photography showed a very high correlation of spatial patterns of kelp occurrence. In some cases, beds were resurveyed based on the photography acquired by DFG Air Services. The situation occurred when the IK Curtis photography was acquired in late November after a series of storms, whereas the DFG photography had been acquired in early November. The measured kelp area between the two dates of photography is markedly different, with the earlier photography (pre-storm) flown by DFG Air Services, depicting a much greater amount of kelp. Santa Rosa Island, San Miguel Island, Administrative Beds 221, 220, and 219 (Santa Cruz, Monterey, Carmel, respectively), were all surveyed or resurveyed using DFG photography. Satellite imagery was used to confirm the kelp bed extents. Area calculations were not made from satellite imagery, because of the inconsistency of resolution (30 meters) with the photography employed for the rest of the coastline (2 meters). In the future, satellite imagery may provide a good means of conducting the entire survey. Since processing is repeated for each photo, using satellite images would translate into a reduced level of effort to perform the survey. It could reduce the amount of work in several ways. 1) Nine LANDSAT scenes cover the entire coast, compared to hundreds of aerial photographs. 2) Satellite imagery may be purchased with various levels of spatial referencing already performed. 3) The footprint of a LANDSAT image is about 100 miles on a side. This makes the images useful for organizations, such as OSPR, who may need coverage of both ocean areas as well as coastal marshes, bays, and estuaries. -- end of file dfgkelp_1999_readme.txt --