In the Hawaiian Archipelago, reef-building corals exist from as far north as Kure atoll at 29° north latitude (Darwin's point) to 19° north latitude (South Point on the island of Hawaii). This wide geographic range dictates the need for a combination of remote sensing and field surveys to map coral reef habitat in the Archipelago. It is estimated that scientists have visited only about 10% of the reefs in the Pacific region, and the majority of these sites were not subject to any detailed surveying, assessment, mapping or extensive examination (Ginsberg & Glynn, 1994). This lack of current information on reefs in the Pacific can be addressed by benthic habitat maps created from remotely sensed data (Mumby et al., 1997, NOAA, 1999).
Benthic habitat maps have proven to be an integral tool in establishing resource management policy over large geographic areas (Holthus & Maragos, 1995; Mumby et al., 1997; NOAA, 2003). Accurate habitat maps are instrumental for resource managers to make informed decisions about the protection and use of these areas (Mumby et al., 1999; NOAA, 1996; NOAA, 1999). For these maps to be of utmost utility, however, their thematic accuracy needs to be soundly determined. This thematic map accuracy should also meet minimum specified requirements (Aronoff, 1982; Chauvaud et al., 1998). Although cartographers can use remotely sensed data to make maps of both known and unknown thematic accuracy, resource managers need maps of known accuracy to facilitate confident management decisions.
The creation of the Northwest Hawaiian Islands Marine Reserve in December 2000, which was declared a monument in 2006, raised an urgent need for comprehensive benthic habitat maps for the entire Hawaiian Archipelago. The United States Coral Reef Task Force (CRTF) placed a high priority on mapping of Pacific reefs that is reflected in several million "new" dollars directed at this effort (NOAA, 1999; Griffis & Koltes, 1999; NOAA, 2003). The results of similar mapping efforts in Florida and the Caribbean have produced the Benthic Habitats of the Florida Keys CD-ROM and Atlas (Kendall et al, 2001; Sheppard et al., 1998), representing over seven years of cooperative efforts and seven million dollars funding (FRMI, 1998). Many of the lessons learned in the Florida and Caribbean mapping projects can be applied to future mapping in Hawai‘i and the Pacific, greatly reducing the expenses associated with this daunting task (NOAA, 1999).
The United States Coral Reef Task Force presented a detailed Mapping Implementation Plan (MIP) at their October 1999 Worldwide meeting asking that local and regional geographic priorities be defined to support the mapping of U.S. coral reefs (NOAA, 1999). The Hawaiian Archipelago was designated as one of these priority areas. NOAA requested that research in support of this mapping effort be conducted by local universities and private contractors in the State of Hawaii. This thesis is a result of that request. In October 2001, NOAA funded a cooperative agreement between the Hawai‘i Coral Reef Assessment and Monitoring Program (CRAMP) and NOAA’s Center for Coastal Monitoring and Assessment (CCMA) for research facilitating the mapping of Hawaiian coral reef habitat. Another partner involved in this mapping effort was Analytical Laboratories Hawaii, Inc (ALH), one of NOAA’s private contractors. ALH did the primary map production and worked closely with NOAA and CRAMP on the fieldwork and accuracy assessment analysis.
ALH, CRAMP and NOAA worked together to produce coral reef habitat maps for approximately 60% of the main Hawaiian Islands. These maps were created by photo interpretation techniques from imagery produced by three different types of remotely sensed data. The accuracy of these thematic maps were tested at four independent test areas and subsequently combined for an overall State-wide accuracy assessment for each of the three types of remotely sensed imagery.
Benthic Habitat Classification Scheme
Before thematic habitat map production can begin, cartographers must have a habitat classification scheme. The classes of the scheme will be used to attribute the various polygons in the final map product. Previous experience has shown that the classification scheme should be hierarchal, so levels of detail in each class can be expanded or collapsed based on the cartographer’s abilities to distinguish habitats and the map user’s needs (Congalton, 1991; Congalton, 1999). The classification scheme should also be mutually exclusive and totally exhaustive so that all habitat types encountered can be distinguished from each other, delineated and classified (Congalton, 1991; Mumby & Harborne, 1999). A NOAA peer-reviewed hierarchal benthic habitat classification scheme for Hawai‘i was created for this project which meets these needs. It has four first-level major classes (unconsolidated sediments, submerged aquatic vegetation, coral reef and hardbottom, other delineations) and 28 detailed classes. It was originally based on the benthic habitat classification scheme used for other NOAA mapping projects in Florida and the U.S. Virgin Islands (FMRI, 1998; Kendall et al., 2001) and has been modified to reflect habitats in Hawai‘i and the greater Pacific. These peer-reviewed modifications occurred over a two-year period in conjunction with marine cartographers, Hawai‘i marine biologists, state and federal coastal resource managers, and local Geographic Information Systems (GIS) specialists. This classification scheme was used for all map production and accuracy assessment associated with this research.
Remotely Sensed Imagery
NOAA's National Ocean Service (NOS) and National Geodetic Survey (NGS) have acquired a combination of aerial color photography, IKONOS multispectral satellite imagery, and airborne hyperspectral data for the near shore waters of about sixty percent of the shoreline of the eight main Hawaiian Islands (Coyne et al., 2003). These images were used by ALH to create draft benthic habitat maps of the region's marine resources, including coral reefs and other important habitats for fisheries, tourism and other aspects of the coastal economy. The draft version of these maps was released by NOAA as an interim product in June 2003 (Coyne et al., 2003). Before this mapping product was released, however, the thematic accuracy of the maps was assessed and the error described and quantified. Because different types of imagery were used to map different areas around the state, it was necessary to determine whether maps made from these different types of imagery were statistically comparable to each other (Congalton, 1991; Congalton & Green, 1999).
These accuracy assessments are the primary focus of this research. They were conducted at four test areas as part of a pilot mapping project conducted by NOAA and CRAMP. The assessments were completed prior to a two-year State-wide mapping effort scheduled to begin in the spring of 2005. The results of the analyses – the thematic accuracy and comparability of the interim habitat maps made from different types of imagery – will be used by NOAA to help decide which types of imagery to acquire for the 2005-2006 mapping efforts. The results will also be used to predict approximate thematic map accuracy that can be expected from the next round of mapping.
Four geographically separate independent test areas were selected to evaluate and compare thematic habitat maps created from the three different types of remotely sensed imagery. Their locations were distributed statewide (Figure 1). They were specifically chosen to be representative of the various habitat types that would be encountered throughout the remainder of the state. They ranged from the west coast of the island of Hawai‘i to the east coast of O‘ahu. The boundaries of the test areas extended from at least 50 meters shoreward of the mean high tide line to a depth of at least 30 meters off shore.
Figure 1: Overall study site showing the geographic extent of the four test areas (shaded).
Accuracy Assessment Field Data
Ground truth data were collected at field stations within these test areas using a stratified random sampling design. The geographic location of the field samples were randomly generated within the mapped polygons of each habitat type (strata). At each sampling station GPS data were collected and the major and detailed benthic habitat classes were recorded. These map classes were based on the NOAA peer-reviewed hierarchical Hawai‘i coral reef habitat classification scheme.
Map Accuracy Assessment Analyses
These ground truth field data were used for accuracy assessments based on a combination of conventional image interpretation methods formerly used in aerial photographic interpretation and more recently applied to automated classifications (Congalton, 1991; Hudson & Ramm, 1987; Ma & Redmond, 1995). The analyses were derived from several error matrices. Creation of these error matrices was facilitated by the use of Geographic Information Systems (GIS) software (Congalton & Green, 1999; Redmond & Ma, 1995; Richards & Jia, 1999). The analyses included producer’s and user’s map accuracy, overall map accuracy, the Kappa statistic, and the Tau coefficient.
The producer’s accuracy is a measurement of how well a photointerpreter can classify a given habitat type. The user’s accuracy is a measurement of how often map polygons of a certain habitat type are classified correctly. They are both calculated from error matrices, and are presented as the percentage of sampling stations correctly classified in the map. A sampling station was considered correctly classified if the habitat type identified in the field matched the habitat type mapped by the photointerpreter.
Overall accuracy is determined by dividing the total correctly classified field samples in the error matrix by the total number of field samples. This overall accuracy percentage is based on the diagonal of the error matrix. It does not take into account errors of commission and omission not in the diagonal. The Kappa statistic and Tau coefficient, however, do incorporate the off-diagonal values of the error matrix (Cohen, 1960; Ma & Redmond, 1995). These statistics give a single proportion of accuracy agreement above that of a random assignment of classes.
Last Update: 04/21/2008
By: Lea Hollingsworth
Hawai‘i Coral Reef Assessment & Monitoring Program
Hawai‘i Institute of Marine Biology
P.O. Box 1346
Kāne‘ohe, HI 96744