Coastal Services Center

National Oceanic and Atmospheric Administration

Search >>

NOAA Coastal Change Analysis Program (C-CAP): Guidance for Regional Implementation

Note of Clarification – January 13, 2003

DOCUMENT CONTAINS DATED MATERIAL. The information contained within this document, NOAA Coastal Change Analysis Program (C-CAP): Guidance for Regional Implementation (1995), provides technical and historical background on C-CAP. Since the publication date, technological and programmatic advancements have occurred. The program itself, for instance, is now part of the NOAA Coastal Services Center's Coastal Remote Sensing program. This document remains available for public consumption because it also provides valuable information on change detection algorithms, classification schemes, and the use of remote sensing for documenting regional coastal changes. For additional information about C-CAP products, please review the NOAA Coastal Services Center's Coastal Remote Sensing Web site (http://www.csc.noaa.gov/crs/) or contact the Center by e-mail at csc@csc.noaa.gov

NOAA Technical Report NMFS 123 – Department of Commerce

J. E. Dobson and E. A. Bright
Oak Ridge National Laboratory
Oak Ridge, TN 37831
R. L. Ferguson, D. W. Field*, and L. L. Wood
NOAA/NMFS Beaufort Laboratory
Beaufort, NC 28516
K. D. Haddad
Florida Dept. of Environmental Protection, Marine Research Institute
St. Petersburg, FL 33701
H. Iredale III
NOAA/NESDIS National Oceanographic Data Center
Washington, DC 20235
J. R. Jensen
Dept. of Geography
University of South Carolina
Columbia, SC 29208
V. V. Klemas
College of Marine Studies, University of Delaware
Newark, DE 19716
R. J. Orth
Virginia Institute of Marine Science
College of William and Mary
Gloucester Point, VA 23062
J. P. Thomas
NOAA/NMFS Office of Protected Resources
Silver Spring, MD 20910

Table of Contents

Preface
Executive Summary

Chapter 1. Introduction

Chapter 2. The C-CAP Coastal Land Cover Classification System

Chapter 3. Guidelines for Monitoring Uplands and Wetlands Using Satellite Remote Sensor Data

Chapter 4. Guidelines for Monitoring Submerged Land Using Aerial Photography

Chapter 5. Quality Assurance and Control

Chapter 6. Product Availability

Chapter 7. Users and Information Needs

Chapter 8. Regional Participation

Acknowledgments

References

Appendices

Preface

This unique document represents a first attempt to develop guidelines that will allow researchers and resource managers alike to quantitatively monitor changes that are occurring in the abundance of emergent and submergent wetlands and adjacent uplands in coastal regions. Such information is essential in order to effectively relate changes in coastal land use to changes in the productivity of estuaries and coastal waters on a regional scale.

This is a document that was developed from input from approximately 200 research scientists and resource managers that attended five regional workshops and several topical interagency meetings. Thus, we believe it represents a general consensus of how to approach the issue of quantifying land cover and wetland change in coastal regions. Because improvement in existing technologies and in our understanding of how to measure habitat change on a regional scale undoubtedly will occur, we intend to update this document periodically. These updates, however, require time to publish, so anyone planning to use these guidelines should contact the corresponding author to obtain drafts of any revised chapters that have not yet been published.

Finally, I would like to express my appreciation to the authors for their fine effort and to Dr. Don Scavia, Director of NOAA's Coastal Ocean Program, for his support, both financial and moral, during the development of this document. I believe we have made a significant step in addressing an important coastal issue.

Ford A Cross
Manager, Coastal Change Analysis Program (C-CAP)
National Marine Fisheries Service/NOAA
Beaufort Laboratory
Beaufort, North Carolina 28516

Executive Summary

The Coastal Change Analysis Program1 (C-CAP) is developing a nationally standardized database on land cover and habitat change in the coastal regions of the United States. C-CAP is part of the Estuarine Habitat Program (EHP) of NOAA's Coastal Ocean Program (COP). C-CAP inventories coastal submersed habitats, wetland habitats and adjacent uplands and monitors changes in these habitats on a 1 to 5 year cycle. This type of information and frequency of detection are required to improve scientific understanding of the linkages of coastal and submersed wetland habitats with adjacent uplands and with the distribution, abundance and health of living marine resources. The monitoring cycle will vary according to the rate and magnitude of change in each geographic region. Satellite imagery (primarily Landsat Thematic Mapper), aerial photography, and field data are interpreted, classified, analyzed, and integrated with other digital data in a geographic information system (GIS). The resulting land cover change databases are disseminated in digital form for use by anyone wishing to conduct geographic analysis in the completed regions.

C-CAP spatial information on coastal change will be input to conceptual and predictive models to support coastal resource policy planning and analysis. C-CAP products will include: 1) spatially registered digital databases and images, 2) tabular summaries by state, county, and hydrologic unit, and 3) documentation. Aggregations to larger areas (representing habitats, wildlife refuges, or management districts) will be provided on a case by case basis. Ongoing C-CAP research will continue to explore techniques for remote determination of biomass, productivity, and functional status of wetlands and will evaluate new technologies (e.g. remote sensor systems, global positioning systems, image processing algorithms) as they become available. Selected hardcopy land cover change maps will be produced at local (1:24,000) to regional scales (1:500,000) for distribution. Digital land cover change data will be provided to users for the cost of reproduction.

Much of the guidance contained in this document was developed through a series of professional workshops and interagency meetings which focused on (a) coastal wetlands and uplands; (b) coastal submersed habitat including aquatic beds; (c) user needs; (d) regional issues; (e) classification schemes; (f) change detection techniques; and (g) data quality. Invited participants included technical and regional experts and representatives of key state and federal organizations. Coastal habitat managers and researchers were given an opportunity for review and comment.

This document summarizes C-CAP protocols and procedures which are to be used by scientists throughout the United States to develop consistent and reliable coastal change information for input to the C-CAP nationwide database. It also provides useful guidelines for contributors working on related projects. It is considered a working document subject to periodic review and revision.

1 Formerly known as the "CoastWatch Change Analysis Program"

Chapter 1. Introduction

The Coastal Region Management Problem

The conterminous United States lost 53 percent of its wetlands to agricultural, residential, and/or commercial land use from the 1780s to 1980s (Dahl 1990). Oil spills occurring throughout the world continue to devastate coastal wetlands (Jensen et al. 1990; Narumalani et al. 1993). Sea level has risen approximately 130 meters in the past 17,500 years. More abundant "greenhouse" gases in the atmosphere may be increasing the Earth's average temperature (Clarke and Primus 1990) and may, yet again, accelerate the global sea level rise, eventually inundating much of today's coastal wetlands (Lee et al. 1992). Unfortunately, current projections for U.S. population growth in coastal regions suggest accelerating losses of wetlands and adjacent habitats, as waste loads and competition for limited space and resources increase (U.S. Congress 1989). Coastal wetlands and submersed habitats are being destroyed by erosion, dredge and fill, impoundments, toxic pollutants, eutrophication, and (for submersed habitats) excessive turbidity and sedimentation. Most of marine finfish and shellfish depend on these coastal habitats for their survival. Salt marsh grasses, mangroves, macroalgae, and submersed grasses and forbs are essential as nourishment and as protection for spawning, raising juveniles, and hiding from predators. Continued loss of these wetlands may lead to the collapse of coastal ecosystems and associated fisheries. Documentation of the loss or gain of coastal wetlands is needed now for their conservation and to effectively manage marine fisheries (Haddad and Ekberg 1987; Kean et al. 1988; Haddad and McGarry 1989; Kiraly et al. 1990). Submersed grasses and forbs include high salinity requiring seagrasses and other species of submersed rooted vascular plants that tolerate or require low salinity water. Submersed grasses and forbs may be a crucial indicator of water quality and overall health of coastal ecosystems (Dennison et al. 1993). Submersed vegetation has the additional requirement of living at photic depths and therefore is particularly sensitive to water clarity (Kenworthy and Haunert 1991). Change (increase or decrease in areal extent, movement, consolidation or fragmentation, or qualitative change) in submersed habitat may be a sensitive integrator of overall water quality and potential for change in fisheries productivity. Submersed rooted vascular (SRV) aquatic beds define habitat critical for the support of many recreational and sport fisheries (Ferguson et al. 1980; Klemas et al. 1993; Phillips 1984; Thayer et al. 1984; Zieman 1982; Zieman and Zieman 1989). Changes in uplands, wetlands, and submersed habitats are rapid and pervasive. Hence, effective management requires frequent monitoring of coastal regions (at least twice per decade).

It has long been suspected that a crucial factor in the observed decline of fisheries in most coastal regions is the declining quantity and quality of habitat. Land cover change is a direct measure of quantitative habitat loss or gain. For many marine fisheries the habitats (ie. land covers) of greatest importance are saltmarsh and seagrass. Other fisheries, such as salmon, depend on a variety of habitats that may include upland, as well. Land cover change also is a direct measure of increases or decreases in sources of pollution, sedimentation, and other factors that determine habitat quality. Increases in developed land, for example, are accompanied by land disturbance that increases erosion and sedimentation and by hydrologic alteration that increases runoff. Similarly, cultivated land is associated with fertilizer and pesticide inputs to the land and ultimately to the marine environment. Hence, land cover change is linked to habitat quantity and quality.

The NOAA Coastal Change Analysis Program (C-CAP) Solution

For these reasons, the National Oceanic and Atmospheric Administration (NOAA) Coastal Ocean Program initiated the Coastal Change Analysis Program (C-CAP), a cooperative interagency and state/federal effort to detect coastal upland and wetland land cover and submersed vegetation and to monitor change in the coastal region of the United States (Cross and Thomas 1992; Haddad 1992). The project utilizes digital remote sensor data, in situ measurement in conjunction with global positioning system systems, and geographic information system (GIS) technology to monitor changes in coastal wetland habitats and adjacent uplands. Landsat multispectral scanner (MSS) data, Landsat Thematic Mapper (TM) data, and SPOT high resolution visible (HRV) data have been used successfully to detect major categories of wetlands (Haddad and Harris 1985; Lade et al. 1988; Jensen et al. 1993b). However, they have not been used previously to map or monitor wetlands for regional or national coverage. The use of satellite imagery for mapping wetlands provides a number of advantages over conventional aerial photographs including timeliness, synopticity, and reduced costs. While aerial photography may be appropriate for high resolution cartography, satellite imagery is better suited and less costly for rapid, repeated observations over broad regions (Bartlett 1987; Klemas and Hardisky 1987; Haddad and Harris 1985; Ferguson et al. 1993). Although the program will stress the use of satellite imagery, particularly for coastal wetlands and adjacent uplands, aerial photography or a combination of photography and satellite imagery (TM or SPOT) will be used for mapping SRV (Orth and Moore 1983) and certain other habitats as suggested by Patterson (1986) and Lade et al. (1988). A methodology to photographically observe, analyze and display spatial change in habitat defined by the presence of SRV was a prerequisite to a nationwide change detection effort (Thomas and Ferguson 1990).

The C-CAP nationally standardized database will be used to monitor land cover and habitat change in the coastal region of the United States (Thomas and Ferguson 1990; Thomas et al. 1991) and to improve understanding of coastal uplands, wetlands (e.g. salt marshes), and submersed habitats (e.g. seagrass) and their linkages with the distribution, abundance, and health of living marine resources. Coastal regions of the U.S. will be monitored every 1 to 5 years depending on the anticipated rate and magnitude of change in each region and the availability of suitable remote sensing and in situ measurements. This monitoring cycle will provide feedback to habitat managers on the success or failure of habitat management policies and programs. Frequent feedback to managers will enhance the continued integrity or recovery of coastal ecosystems and the attendant productivity and health of fish and other living marine resources at minimal cost. In addition, the geographical database will allow managers and scientists to evaluate and ultimately to predict cumulative direct and indirect effects of coastal development on wetland habitats and living marine resources. Initially, C-CAP products will document current land cover distribution and the change that has occurred in the recent past. The database, increasing with each subsequent monitoring cycle, will be an invaluable baseline resource for research, evaluation of local, state and federal wetland management strategies, and construction of predictive models. C-CAP directly supports NOAA's responsibilities in estuarine and marine science, monitoring, and management legislated in the Fish and Wildlife Coordination Act; the Magnuson Fishery Conservation and Management Act; the Coastal Zone Management Act; the Clean Water Act; the Marine Protection, Research and Sanctuaries Act; and the National Environmental Policy Act. Land cover change data are essential to the implementation of a "No Net Loss" wetlands policy.

A large community of managers, scientists, and users were involved in developing a C-CAP protocol at the national level. Guidance in this document was derived from a series of professional workshops and interagency working group meetings which focused on:

Approximately 40 scientists and environmental managers attended each major regional workshop held in the Southeast, Northeast, Pacific Coast, and Great Lakes with approximately 200 individuals participating in all workshops and special meetings. The community of users and providers of coastal habitat information were given an opportunity for review and comment. A detailed list of workshops is found in Appendix 11.4.

While C-CAP is national in scope, it is based on procedures also applicable at local and regional levels. Much of the content of this document is based on C-CAP sponsored research conducted at the regional level. For example, Klemas et al. (1993) of the College of Marine Studies at the University of Delaware developed the "C-CAP Coastal Land Cover Classification System" by investigating existing upland and wetland classification systems and then synthesizing a new system practical at the regional level. Dobson and Bright (1991, 1992, 1993) of the Oak Ridge National Laboratory (ORNL) developed a regional prototype for inventorying uplands and wetlands in the Chesapeake Bay Region. Jensen et al. (1993a) evaluated various change detection algorithms for inland and coastal wetland environments near Charleston, South Carolina. Ferguson et al. (1993) developed a regional prototype to inventory SRV in North Carolina based on protocols developed by the Beaufort Laboratory, Southeast Fisheries Science Center, National Marine Fisheries Service. Khorram et al. (1992) investigated methods of seamlessly integrating multiple region C-CAP databases.

The C-CAP protocol continues to evolve and improve. For example, projects underway in 1993 include analysis of the effects of tidal stage on remote sensing classification, change detection accuracy assessment, refined techniques for classification of forested wetlands, and advanced change detection techniques (Appendix 11.5). Research continues on functional health indicators (e.g., biomass, productivity), plant stress (e.g., mangrove freeze), new data collection instruments, and regional ecological modeling. Thus, C-CAP will continue to have a strong research and development component to improve and refine its operational techniques.

C-CAP's National Scope and Regional Implementation

No single federal or state organization will collect all the information residing in the C-CAP database. Instead, regional inventories will be completed by regional experts following C-CAP guidelines. Therefore, it is important to define the logic used to specify a C-CAP region. First, regional boundaries must coincide with NOAA/NMFS Regions, including:

Coastal regions may be further subdivided, as appropriate, on the basis of state and other administrative boundaries or ecoregions as defined, for example, by Omernick (1987).

The boundary should encompass coastal watersheds plus offshore coral reefs, algae, and seagrass beds in the photic zone. In keeping with the goals of C-CAP and anticipated funding constraints the recommended approach is to designate (1) standard coverage limits for general application, and (2) extended coverage limits for regions with special needs. Standard coverage will utilize biological and other geographical boundaries appropriate to the needs of specified C-CAP users identified through the protocol workshops. Extended coverage will be defined for each regional project in collaboration with states and other regional organizations. NOAA will make every effort to identify and accommodate research, conservation, management, and other needs that rely on wetland maps and data. Regional projects will be designed to identify special needs that may require extended coverage and to suggest sources of funds to support the additional cost of extended coverage.

The Estuarine Drainage Area (EDA), defined by NOAA/National Ocean Service (NOS) as the "land and water component of an entire watershed that most directly affects an estuary", is an appropriate standard coverage area for C-CAP. For the purposes of this program, all U. S. coasts are or will be defined as part of an Estuarine Drainage Area. The boundary of each EDA basin will be defined to be consistent with USGS Hydrologic Units and codes.

The Estuarine Drainage Boundary as defined by NOAA/NOS is considered a standard inland boundary for C-CAP regional projects. Regional analysts may employ C-CAP protocols upstream, but C-CAP funding is not intended for coverage beyond the EDA. However, C-CAP funding may be used to purchase satellite scenes that extend beyond the EDA if they are necessary to cover the coastal region. Functional definitions, such as "limits of tidal influence," may be employed in response to local situations justified by local user communities and local/regional experts on a coastal region by region or estuary by estuary basis. Regional analysts should be aware of local, state, and Federal rights and responsibilities and should seek inter-governmental and interagency cooperation. Because C-CAP interests include the effects of eutrophication due to development of uplands, information from outside the EDA may be justified in high order streams that extend beyond the coastal region. In this case, the point where the river enters the region will be defined as a point source for inputs.

The offshore boundary of each region is defined as the seaward extent of wetlands, seagrass, coral, or other submersed habitat detectable using remote sensing systems. The functional definition of "limits of detection" normally will be based on satellite and aerial sensors and will vary within and among regions. Both the "limits of detection" and the actual bathymetric range of SRV are based on light attenuation and, thus, will not be a consistent bathymetric contour even within a single region.

Overlap of regions, consistent with TM scene boundary overlap, is preferred so that analysts may calibrate results from neighboring regions. A healthy exchange between neighboring regional analysts could reconcile differences, not only in the area of overlap, but in signature identification across both regions. Each regional project team will be responsible for calibrating the relationship between remotely sensed spectral information and other information such as field measurements of biomass and photosynthetic rates. Historically, such measurements have focused on relatively few of the many species, habitats, and land cover types of significance in the coastal region. Analysts should also ensure that protocols originally developed for northern temperate latitudes are modified sufficiently to serve well in tropical areas of the Southern U.S., Caribbean, and Pacific Ocean, and in the Arctic areas of Alaska. It will be necessary, for example, to use different methods and sensors for coral reefs than for wetlands. Similarly, the identification of Arctic muskeg may require different methods and sensors from those used to identify temperate, herbaceous wetland.

Change Detection Every 1 to 5 Years

The frequency of change detection is a crucial issue. For most regions in the United States, the base year (sometimes referred to as Date 1 in the diagrams or Tb) should be the most recent year for which acceptable satellite imagery for uplands and wetlands or aerial photographs for submersed habitat can be obtained, and for which sufficient in situ information is available to conduct an error evaluation. Exceptions may occur in regions where cloud cover is a perennial problem or where other considerations favor aerial photographs over satellite imagery. The choice of the second date of imagery (Date n-1 or n+1) may be more flexible. It may be desirable to choose a date one to five years earlier than the base period to capture recent changes in coastal habitats. Plans should then be made for another change analysis no later than five years after the base time. However, plans may be altered abruptly when natural or human-induced events, such as hurricanes and oil spills, occur.

Five years is the recommended frequency of change detection for most regions, but shorter periods may be necessary in regions undergoing rapid economic development or affected by catastrophic events. Longer periods may be necessary where funds are limited or where change is exceptionally slow. Regional analysts are advised to evaluate rates of change and explicitly recommend the base year and change period as a part of each regional project proposal. Unfortunately, remotely sensed data obtained specifically for other purposes (e.g. urban analysis, forest inventory, etc.) often are not suitable for use in C-CAP. Aquatic beds, and even coastal wetlands, may not be identifiable on aerial photography obtained for other purposes.

The Need for Standardization and Guidelines

C-CAP desires to create a synoptic, digital database of coastal wetland and upland land cover by class for a base time period and to identify change between the base period and other time periods. The use of satellite remote sensing to inventory uplands and wetlands, conventional aerial photography to inventory submerged lands, and geographic information systems (GIS) to analyze the data are important elements of the C-CAP methodology. However, the goal of completing an accurate change detection product overrides any given technical consideration. Therefore, timely high-quality information whether from aerial photographs, topographic maps, field experience or other sources may be used in the preparation of the C-CAP products if appropriate guidelines are followed.

By standardizing procedures at the national level this document will benefit not only C-CAP but also coastal management research conducted by other state and federal agencies. C-CAP desires to facilitate the exchange of standardized data among programs, decrease duplication, and improve the quality and utility of decision support for wetlands policy, management, and research activities. All data accepted for inclusion and eventual distribution in the C-CAP database must adhere to the protocol described in this manual. The protocol is designed to allow for flexibility in the use of elements of the classification scheme, choice of remote sensor data, classification and change detection procedures, and other key elements that vary regionally. However, potential users must adhere to the protocol in order to maintain high quality information in the C-CAP database. Coastal land cover change databases derived independently from C-CAP will be considered for dissemination as C-CAP products if originating organizations can document compliance with C-CAP protocol and data quality standards.

General Steps Required to Conduct Regional C-CAP Projects

The general steps required to conduct regional C-CAP change detection projects using satellite remotely sensed data are summarized in Table 1. This document is organized according to these specific requirements, and in certain instances provides step-by-step instructions to be used when conducting regional projects. One of the first requirements of regional participants is to precisely identify land cover classes of interest to be monitored and eventually placed in the C-CAP change detection database. This must be performed in conjunction with an appropriate classification scheme. Unfortunately, no existing standardized classification scheme was suitable for all C-CAP requirements. Therefore, great effort went into the development of the C-CAP Coastal Land Cover Classification System which can be used to inventory uplands and wetlands using satellite remote sensor data as well as SRV using metric aerial photography.

Table 1. General steps required to conduct regional C-CAP change detection projects to extract upland and wetland information using satellite remote sensing systems.

Chapter 2. The C-CAP Coastal Land Cover Classification System

It is essential that the coastal land cover information stored in the C-CAP database be taxonomically correct and consistent with coastal wetland information derived from other agencies. The C-CAP Coastal Land Cover Classification System (Table 2) includes three Level I super classes (Klemas et al. 1993):

These super classes are subdivided into classes and subclasses at Levels II and III, respectively. While the latter two categories constitute the primary habitats of interest to NOAA, uplands are also included because they influence adjacent wetlands and water bodies. The classification system is hierarchical, reflects ecological relationships, and focuses on land cover classes that can be discriminated primarily from satellite remote sensor data. It was adapted and designed to be compatible with other nationally standardized classification systems, especially:

Dedicated workshops on the C-CAP Classification System and productive discussions and reviews with representatives from each of these major agencies resulted in a classification system which is in harmony with other major United States land cover databases. The C-CAP Coastal Land Cover Classification System includes upland, wetland, submerged land, and water in a single, comprehensive scheme. An attempt has been made to identify land cover classes that can be derived primarily through remotely sensing and that are important indicators of ecosystem change. Modifications were necessary to reconcile inconsistencies between Anderson et al. (1976) and Cowardin et al. (1979), and remove all land use categories (Dobson 1993a). C-CAP focuses on land cover and its relationship to other functional components of landscape (Dobson 1993b). Definitions of the pertinent terms are:

While all categories of the C-CAP classification system can be represented as two-dimensional features at the mapping scale of 1:24,000, some features may be mapped as lines (e.g., a Marine/Estuarine Rocky Shore) or points (e.g., unique landmarks). Most linear and point features will be obtained from non-satellite sources of information (e.g. aerial photography or in situ measurement using global positioning systems). Those classes and subclasses which are required by C-CAP and which each regional C-CAP project will include in its database are underlined in Table 1. The underlined classes, with the exception of aquatic beds, can generally be detected by satellite remote sensors, particularly when supported by surface in situ measurement.

Super Classes of the C-CAP Coastal Land Cover Classification System

Uplands

The Uplands super class consists of seven subclasses (Table 2.): Developed Land, Cultivated Land, Grassland, Woody Land, Bare Land, Tundra and Snow/Ice. Upland classes are adapted from Level I classes in the USGS Land Use/Land Cover Classification System (Anderson et al. 1976; USGS 1992; Appendix 1 - Table A1). Detailed definitions of all C-CAP classes and subclasses in Table 1 are found in Appendix 3.

Table 2. C-CAP Coastal Land Cover Classification System (Modified from Klemas et al., 1993)

1.0 Upland

1.1 Developed Land
1.11    High Intensity
1.12    Low Intensity

1.2 Cultivated Land
1.21    Orchards/Groves/Nurseries
1.22    Vines/Bushes
1.23    Cropland

1.3 Grassland
1.31    Unmanaged
1.32    Managed

1.4 Woody Land
1.41    Deciduous
1.411       Forest
1.412       Scrub/Shrub
1.42    Evergreen
1.421       Forest
1.422       Scrub/Shrub
1.43    Mixed
1.431       Forest
1.432       Scrub/Shrub

1.5 Bare Land

1.6 Tundra

1.7 Snow/Ice
1.71    Perennial Snow/Ice
1.72    Glaciers

2.0 Wetland (Excludes Bottoms, Reefs, Nonpersistent Emergent Wetlands, and
Aquatic Beds, all of which are covered under 3.0, Water and Submerged Land.)

2.1 Marine/Estuarine Rocky Shore
2.11    Bedrock
2.12    Rubble

2.2 Marine/Estuarine Unconsolidated Shore (Beach, Flat, Bar)
2.21    Cobble-gravel
2.22    Sand
2.23    Mud/Organic

2.3 Estuarine Emergent Wetland
2.31    Haline (Salt Marsh)
2.32    Mixohaline (Brackish Marsh)

2.4 Estuarine Woody Wetland
2.41    Deciduous
2.411        Forest
2.412        Scrub/Shrub
2.413        Dead
2.42    Evergreen
2.421        Forest
2.422        Scrub/Shrub
2.423        Dead
2.43    Mixed
2.431        Forest
2.432        Scrub/Shrub
2.433   Dead

2.5 Riverine Unconsolidated Shore (Beach, Flat, Bar)
2.51    Cobble-gravel
2.52    Sand
2.53    Mud/Organic

2.6 Lacustrine Unconsolidated Shore (Beach, Flat, Bar)
2.61    Cobble-gravel
2.62    Sand
2.63    Mud/Organic

2.7 Palustrine Unconsolidated Shore (Beach, Flat, Bar)
2.71    Cobble-gravel
2.72    Sand
2.73    Mud/Organic

2.8 Palustrine Emergent Wetland (Persistent)

2.9 Palustrine Woody Wetland
2.91    Deciduous
2.911        Forest
2.912        Scrub/Shrub
2.913        Dead
2.92    Evergreen
2.921        Forest
2.922        Scrub/Shrub
2.923        Dead
2.93    Mixed
2.931        Forest
2.932        Scrub/Shrub
2.933        Dead

3.0 Water and Submerged Land (Includes deepwater habitats and those wetlands
with surface water but lacking trees, shrubs, and persistent emergents)

3.1 Water (Bottoms and undetectable reefs, aquatic beds or nonpersistent emergent Wetlands)
3.11    Marine/Estuarine
3.12         Riverine
3.13         Lacustrine (Basin  > 20 acres)
3.14         Palustrine (Basin < 20 acres)

3.2 Marine/Estuarine Reef

3.3 Marine/Estuarine Aquatic Bed
3.31    Algal (e.g., kelp)
3.32    Rooted Vascular (e.g., seagrass)
3.321        High Salinity (> 5 ppt; Mesohaline, Polyhaline, Euhaline, Hyperhaline)
3.322        Low Salinity (< 5 ppt; Oligohaline, Fresh)

3.4 Riverine Aquatic Bed
3.41    Rooted Vascular/Algal/Aquatic Moss
3.42    Floating Vascular

3.5  Lacustrine Aquatic Bed (Basin > 20 acres)
3.51    Rooted Vascular/Algal/Aquatic Moss
3.52    Floating Vascular

3.6  Palustrine Aquatic Bed (Basin < 20 acres)
3.61    Rooted Vascular/Algal/Aquatic Moss
3.62    Floating Vascular

Developed Land (Derived fromthe Anderson et al. [1976] Urban or Built-up class) characterizes constructed surfaces comprised of concrete, asphalt, roofing, and other building materials with or without vegetation. This class has been divided into two subclasses based on the amount of constructed surface relative to the amount of vegetated surface present. High Intensity Developed Land contains little or no vegetation. This subclass includes heavily built-up urban centers as well as large constructed surfaces in suburban and rural areas. Large buildings (such as multiple family housing, hangars, and large barns), interstate highways, and runways typically fall into this subclass. Low Intensity Developed Land contains substantial amounts of constructed surface mixed with substantial amounts of vegetated surface. Small buildings (such as single family housing, farm outbuildings, and sheds), streets, roads, and cemeteries with associated grasses and trees typically fall into this subclass.

Cultivated Land (Agricultural Land in Anderson et al. 1976) includes herbaceous (cropland) and woody (orchards, nurseries, vineyards, etc.) cultivated lands. Seasonal spectral signatures, geometric field patterns and road network patterns may help identify this land cover type. Always associated with agricultural land use, cultivated land is used for the production of food and fiber.

Grassland differs from Rangeland in Anderson et al. (1976) by excluding shrub-brushlands. Unmanaged Grasslands are dominated by naturally occurring grasses and forbs which are not fertilized, cut, tilled or planted regularly. Managed Grasslands are maintained by human activity such as fertilization and irrigation, are distinguished by enhanced biomass productivity, and can be recognized through vegetative indices based on spectral characteristics. Examples of such areas include lawns, golf courses, forest or shrub areas converted to grassland, or areas of permanent grassland with altered species composition. This category includes managed pastures and pastures with vegetation that grows vigorously as fallow. Managed Grasslands are used for grazing or for growing and harvesting hay and straw for animal feed.

Woody Land includes non-agricultural trees and shrubs. The category alleviates the problem of separating various sizes of trees and shrubs using satellite remote sensor data but allows a height-based separation if high resolution aerial photography are available. The class may be partitioned into three subclasses: Deciduous, Evergreen, and Mixed. These three subclasses generally can be discriminated with satellite remote sensing systems.

Bare Land (derived from Barren Land in Anderson et al. 1976) is composed of bare soil, rock, sand, silt, gravel, or other earthen material with little or no vegetation. Anderson's Barren Land was defined as having limited ability to support life; C-CAP's Bare Land is defined by the absence of vegetation without regard to inherent ability to support life. Vegetation, if present, is more widely spaced and scrubby than that in the vegetated classes. Unusual conditions such as a heavy rainfall may occasionally result in growth of a short-lived, luxuriant plant cover. Wet, nonvegetated exposed lands are included in the Wetland categories. Bare Land may be bare temporarily because of human activities. The transition from Woody Land, Grassland, or Cultivated Land to Developed Land, for example, usually involves a Bare Land phase. Developed Land also may have temporary waste and tailing piles. Woody Land may be clearcut producing a temporary Bare Land phase. When it may be inferred from the data that the lack of vegetation is due to an annual cycle of cultivation (eg. plowing), the land is not included in the Bare Land class. Land temporarily without vegetative cover because of cropping or tillage, is classified as Cultivated Land, not Bare Land.

Wetlands

Wetlands are lands where saturation with water is the dominant factor determining soil development and the types of plant and animal communities living in the soil and on its surface (Cowardin et al. 1979). A characteristic feature shared by all wetlands is soil or substrate that is at least periodically saturated with or covered by water. The upland limit of wetlands is designated as (1) the boundary between land with predominantly hydrophytic cover and land with predominantly mesophytic or xerophytic cover; (2) the boundary between soil that is predominantly hydric and soil that is predominantly nonhydric; or (3) in the case of wetlands without vegetation or soil, the boundary between land that is flooded or saturated at some time during the growing season each year and land that is not (Cowardin et al. 1979). The majority of all wetlands are vegetated and are found on soil.

Wetland in the C-CAP Coastal Land Cover Classification System (Table 1) includes all areas considered wetland by Cowardin et al. (1979) except for bottoms, reefs, aquatic beds, and Nonpersistent Emergent Wetlands. The class subdivision was adopted primarily from the Cowardin system, shown in Appendix 2 (Table A2). At Level II, C-CAP incorporates certain Cowardin classes (e.g., Rocky Shore, Unconsolidated Shore, Emergent Wetland) or grouped Cowardin classes (e.g., Woody Wetland may be further divided into Scrub-Shrub and Forested categories), in combination with Cowardin systems (i.e., Marine, Estuarine, Riverine, Lacustrine, Palustrine). Thus, a typical Level II class in the C-CAP system might be Palustrine Woody Wetland.

Marine and Estuarine Rocky Shores (Cowardin et al. 1979) were combined into a single class, Marine/Estuarine Rocky Shore. The same logic was used to produce Marine/Estuarine Unconsolidated Shore.

Salinity exhibits a horizontal gradient in coastal estuary marshes. This is evident not only through the direct measurement of salinity but in the horizontal distribution of marsh plants (Daiber 1986). Therefore, the Estuarine Emergent Wetland class is partitioned into Haline (Salt) and Mixohaline (Brackish) Marshes. For both subclasses, the C-CAP Classification System uses the Cowardin et al. (1979) definitions. Mixohaline salinity ranges from 0.5 ppt to 30 ppt, and Haline salinity is >30 ppt. Within a marsh, plant zonation is usually quite evident. Along the Atlantic coast of North America the pioneer plant on regularly flooded mudflats is saltmarsh cordgrass, Spartina alterniflora, which often appears in pure stands. In more elevated areas which are flooded less frequently saltmeadow hay, Spartina patens often dominates. The upland interfaces are bordered by marsh elder, Iva frutescens and groundsel tree, Baccharis halimifolia. Thus, salt marshes may be subdivided further into High Marsh and Low Marsh, but this distinction is not required in C-CAP regional projects.

The C-CAP Coastal Land Cover Classification System does not attempt to identify freshwater Nonpersistent Emergent Wetlands because they are invisible during much of the year and difficult to detect by remote sensors. These wetlands are classified as "Riverine Water" and "Lacustrine Water," respectively.

Water and Submerged Land

All areas of open water with < 30% cover of trees, shrubs, persistent emergent plants, emergent mosses, or lichens are assigned to Water and Submerged Land, regardless of whether the area is considered wetland or deepwater habitat under the Cowardin et al. (1979) classification.

The Water class includes Cowardin et al.'s (1979) Rock Bottom and Unconsolidated Bottom, and Nonpersistent Emergent Wetlands, as well as Reefs and Aquatic Beds that are not identified as such. Most C-CAP products will display water as a single class. However, it is recognized that the major systems (Marine/Estuarine, Riverine, Lacustrine, Palustrine) are ecologically different from one another. For this reason, the C-CAP system identifies the four systems as Level III subclasses, i.e., 3.11 Marine/Estuarine Water, 3.12 Riverine Water, 3.13 Lacustrine Water, and 3.14 Palustrine Water. While C-CAP does not require these subclasses, the option is provided to participants who may have such data available from ancillary sources. Having the water subclasses also makes the C- CAP scheme more compatible with the Cowardin et al. (1979) system. The subclass 3.11 Marine/Estuarine Water includes Bottoms and undetected Reefs and Aquatic Beds. The subclasses 3.12 Riverine Water, 3.13 Lacustrine Water, and 3.14 Palustrine Water include Bottoms and undetected Aquatic Beds as well as Nonpersistent Emergent Wetlands. Palustrine waterbodies, defined as covering < 20 acres, are smaller than Lacustrine waterbodies.

C-CAP combined Marine and Estuarine Reefs and Aquatic Beds into two classes, Marine/Estuarine Reefs and Marine/Estuarine Aquatic Beds. Marine/Estuarine Aquatic Beds includes the subclass Rooted Vascular which is subdivided into High Salinity (> 5 ppt) and Low Salinity (< 5 ppt). The >5 ppt salinity separates seagrasses from low salinity tolerating or requiring submersed grasses and forbs. Both types of plants define aquatic beds, submersed habitats, which are important to the C-CAP project. High Salinity includes mesohaline, polyhaline, euhaline, and hyperhaline salinity categories of Cowardin et al. (1979). Low Salinity includes oligohaline and fresh categories (< 5 ppt salinity).

With the noted exceptions, most of the Wetland and Water classes have definitions similar to those contained in Cowardin et al. (1979) so that data can be interchanged with other programs, such as the U.S. Fish and Wildlife Service, National Wetlands Inventory (NWI) program, which is based on the Cowardin et al. (1979) classification system. Detailed definitions of all super classes, classes and subclasses shown in Table 1 are provided in Appendix 3.

Chapter 3. Guidelines for Monitoring Uplands and Wetlands Using Satellite Remote Sensor Data

Successful remote sensing change detection of uplands and wetlands in coastal regions requires careful attention to: 1) sensor systems; 2) environmental characteristics; and 3) geodetic control. Failure to understand the impact of the various parameters on the change detection process can lead to inaccurate results. Ideally, the remotely sensed data used to perform C-CAP change detection are acquired by a remote sensor system which holds the following factors constant: temporal, spatial (and look angle), spectral, and radiometric. It is instructive to review each of these parameters and identify why they have a significant impact on the success of C-CAP remote sensing change detection projects. Table 3. summarizes the characteristics of some of the most important satellite remote sensing systems.

Table 3. Selected Satellite Remote Sensing System Characteristics 

Spectral Resolution(mm) | Spatial Resolution(m) | Temporal Resolution | Radiometric Resolution

Landsat MSS 1 -  5
 Band 1 (.50 - .60)        80 x 80                 18 days               8 bits
 Band 2 (.60 - .70)        80 x 80                 18 days               8 bits
 Band 3 (.70 - .80)        80 x 80                 18 days               8 bits
 Band 4 (.80 - 1.1)        80 x 80                 18 days               8 bits

Landsat Thematic Mapper 4 & 5
 Band 1 (.45 - .52)        30 x 30                 16 days               8 bits
 Band 2 (.52 - .60)        30 x 30                 16 dyas               8 bits
 Band 3 (.63 - .69)        30 x 30                 16 days               8 bits
 Band 4 (.76 - .90)        30 x 30                 16 days               8 bits
 Band 5 (1.55 - 1.75)      30 x 30                 16 days               8 bits
 Band 7 (2.08 - 2.35)      30 x 30                 16 days               8 bits
 Band 6 (10.4 - 12.5)      120 x 120               16 days               8 bits

SPOT HRV XS
 Band 1 (.50 - .59)        20 x 20                 pointable             8 bits
 Band 2 (.61 - .68)        20 x 20                 pointable             8 bits
 Band 3 (.79 - .89)        20 x 20                 pointable             8 bits
 Pan (.51 - .73)           10 x 10                 pointable             8 bits
1Landsat MSS 1 and 2 collected data in 7 bits.

Remote Sensing System Considerations

Temporal Resolution

There are two important temporal resolutions which should be held constant when performing coastal change detection using multiple dates of remotely sensed data. First, the data should be obtained from a sensor system which acquires data at approximately the same time of day (e.g., Landsat Thematic Mapper data are acquired before 9:45 am for most of the conterminous United States). This eliminates diurnal sun angle effects which can cause anomalous differences in the reflectance properties of the remotely sensed data. Second, whenever possible it is desirable to use remotely sensed data acquired on anniversary dates, e.g., October 1, 1988 versus October 1, 1993. Using anniversary date imagery removes seasonal sun angle differences which can make change detection difficult and unreliable (Jensen et al. 1993a). Usually precise anniversary date imagery is not available. The determination of acceptable near-anniversary dates then depends on local and regional factors such as phenological cycles and annual climatic regimes.

Spatial Resolution and Look Angle

Accurate spatial registration of at least two images is essential for digital change detection. Ideally, the remotely sensed data are acquired by a sensor system which collects data with the same instantaneous-field-of-view (IFOV) on each date. For example, Landsat Thematic Mapper data collected at 30 x 30 m spatial resolution (Table 3) on two dates are relatively easy to register to one another. Geometric rectification algorithms (Jensen 1986; Novak 1992) are used to register the images to a standard map projection (Universal Transverse Mercator - UTM, for most U.S. projects). Rectification should result in the two images having a root mean square error (RMSE) of < +0.5 pixel. RMSE > +0.5 pixel may result in the identification of spurious areas of change between the two data sets. See "Rectification of Multiple-Date Remote Sensor Data to Detect Change" for a summary of C-CAP image rectification requirements.

It is possible to perform change detection using data collected by two different sensor systems with different IFOVs, e.g. Landsat TM data (30 x 30 m) for date 1 and SPOT HRV data (20 x 20 m) for date 2. In such cases, it is necessary to decide upon a representative minimum mapping unit (e.g. 20 x 20 m) and then resample both data sets to this uniform pixel size. This does not present a significant problem as long as one remembers that the information content of the resampled data can never be greater than the IFOV of the original sensor system (i.e. even though the Landsat TM data are resampled to 20 x 20 m pixels, the information was still acquired at 30 x 30 m resolution and one should not expect to be able to extract additional spatial detail in the dataset).

Some remote sensing systems like SPOT collect data at off-nadir look angles as much as + 20ø (Table 23, i.e. the sensors obtain data of an area on the ground from an "oblique" vantage point. Two images with significantly different look angles can cause problems when used for change detection purposes. For example, consider a maple forest consisting of very large, randomly spaced trees. A SPOT image acquired at 0ø off-nadir will look directly down upon the "top" of the canopy. Conversely, a SPOT image acquired at 20ø off-nadir will record reflectance information from the "side" of the canopy. Differences in reflectance from the two datasets can cause spurious change detection results. Therefore, the data used in a remote sensing digital change detection should be acquired with approximately the same look angle whenever possible.

Spectral Resolution

A fundamental assumption of digital change detection is that there should exist a difference in the spectral response of a pixel on two dates if the biophysical materials within the IFOV have changed between dates. Ideally, the spectral resolution of the remote sensor system is sufficient to record reflected radiant flux in spectral regions that best capture the most descriptive spectral attributes of the object. Unfortunately, different sensor systems do not record energy in exactly the same portions of the electromagnetic spectrum, i.e. bandwidths (Table 3). For example, Landsat MSS records energy in four relatively broad bands, SPOT HRV sensors record in three relatively coarse multispectral bands and one panchromatic band, and TM in six relatively narrow optical bands and one broad thermal band (Table 3). Ideally, the same sensor system is used to acquire imagery on multiple dates. When this is not possible, the analyst should select bands which approximate one another. For example, SPOT bands 1 (green), 2 (red), and 3 (near-infrared) can be used successfully with TM bands 2 (green), 3 (red), and 4 (near-infrared) or MSS bands 1 (green), 2 (red), and 4 (near-infrared). Many of the change detection algorithms to be discussed do not function well when bands from one sensor system do not match those of another sensor system, e.g. utilizing TM band 1 (blue) with either SPOT or MSS data is not wise.

Radiometric Resolution

An analog to digital conversion of the satellite remote sensor data usually results in 8-bit brightness values with values ranging from 0 to 255 (Table 3). Ideally, the sensor systems collect the data at the same radiometric precision on both dates. When the radiometric resolution of data acquired by one system (e.g., MSS 1 with 7-bit data) are compared with data acquired by a higher radiometric resolution instrument (e.g., TM with 8-bit data) then the lower resolution data (e.g., 7-bit) should be "decompressed" to 8-bits for change detection purposes. However, the precision of decompressed brightness values can never be better than the original, uncompressed data.

The Preferred C-CAP Satellite Sensor System

TM is currently the primary sensor recommended for C-CAP image acquisition and change analysis for all land cover except aquatic beds. A TM image, although its spatial resolution is not as good as that of a SPOT satellite or aircraft MSS image, is generally less expensive to acquire and process for large-area coverage. Compared to SPOT imagery, TM has better spectral resolution and specific spectral bands that are more applicable to wetlands delineation (bands 5 and 7). In addition, TM is preferred over SPOT because TM has collected data for a longer time (since 1982 as opposed to SPOT since 1986) and because many TM scenes of the United States coastal regions were systematically collected on a routine basis.

There are advantages and disadvantages to using other sensors. Aircraft multispectral scanners are more expensive and complex to utilize over large regions (Jensen et al. 1987). However, good algorithms are now available for georeferencing, and in certain cases (e.g., when higher spectral or spatial resolution is needed and when unfavorable climactic conditions for satellite sensors exist) aircraft sensors may be optimum. The SPOT sensor has a greater temporal coverage because the satellite can collect data off-nadir. However, if off-nadir SPOT imagery is used for C-CAP change analyses, the data must be normalized to compensate for different look angles that may preclude pixel-to- pixel spectral-change analysis. Nevertheless, SPOT imagery may be a reasonable alternative in certain areas due to cloud cover or other impediments to TM data availability.

C-CAP remains flexible in order to take advantage of new sensors and other technologies that become operational during the lifetime of the program. Regional participants should work with the C-CAP program coordinators to ensure that the sensor selection meets the following C-CAP requirements.

Environmental Characteristics of Importance When Performing Change Detection

Failure to understand the impact of various environmental characteristics on the remote sensing change detection process can also lead to inaccurate C-CAP results. When performing change detection it is desirable to hold environmental variables as constant as possible. Specific environmental variables and their potential impacts are described below.

Atmospheric Conditions

There should be no clouds, haze, or extreme humidity on the days remote sensing data are collected. Even a thin layer of haze can alter spectral signatures in satellite images enough to create the false impression of spectral change between two dates. Obviously, O% cloud cover is preferred for satellite imagery and aerial photography. At the upper limit, cloud cover > 20% is usually unacceptable. It should also be remembered that clouds not only obscure terrain but the cloud shadow also causes major image classification problems. Any area obscured by clouds or affected by cloud shadow will filter through the entire change detection process, severely limiting the utility of the final change detection product. Therefore, regional analysts must use good professional judgment in evaluating such factors as the criticality of the specific locations affected by cloud cover and shadow, and the availability of timely surrogate data for those areas obscured (e.g. perhaps substituting aerial photography interpretation for a critical area). Even when the stated cloud cover is 0%, it is advisable to "browse" the proposed image on microfiche at the National Cartographic Information Center in each state to confirm that the cloud cover estimate is correct.

Assuming no cloud cover, the use of anniversary dates helps to ensure general, seasonal agreement between the atmospheric conditions on the two dates. However, if dramatic differences exist in the atmospheric conditions present on the n dates of imagery to be used in the change detection process, it may be necessary to remove the atmospheric attenuation in the imagery. Two alternatives are available. First, sophisticated atmospheric transmission models can be used to correct the remote sensor data if substantial in situ data are available on the day of the overflights. Second, an alternative empirical method may be used to remove atmospheric effects. A detailed description of one empirical method of image to image normalization is found in "Radiometric Normalization of Multi-date Images to Detect Change".

Soil Moisture Conditions

Ideally, the soil moisture conditions should be identical for the n dates of imagery used in a change detection project. Extremely wet or dry conditions on one of the dates can cause serious change detection problems. Therefore, when selecting the remotely sensed data to be used for change detection it is very important not only to look for anniversary dates, but also to review precipitation records to determine how much rain or snow fell in the days and weeks prior to remote sensing data collection. When soil moisture differences between dates are significant for only certain parts of the study area (perhaps due to a local thunderstorm), it may be necessary to stratify (cut-out) those affected areas and perform a separate analysis which can be added back in the final stages of the project.

Vegetation Phenological Cycle Characteristics

Vegetation grows according to seasonal and annual phenological cycles. Obtaining near-anniversary images greatly minimizes the effects of wetland seasonal phenological differences which may cause spurious change to be detected in the imagery. One must also be careful about two other factors when dealing with man-made upland seasonal agricultural crops. First, many monoculture crops (e.g. corn) normally are planted at approximately the same time of year. A month lag in planting date between fields having the same crop can cause serious change detection error. Second, many monoculture crops are comprised of different species (or strains) of the same crop which can cause the crop to reflect energy differently on multiple dates of anniversary imagery. These observations suggest that the analyst must know the biophysical characteristics of the vegetation as well as the cultural land-tenure practices in the study area so that imagery which meets most of these characteristics can be selected for change detection.

The choice of image date is best determined by mutual agreement among remote sensing specialists, biologists, ecologists, and local experts. The selection of the acceptable window of acquisition will be made independently by participants in each region. No single season will serve for all areas because of substantial latitudinal variation extending from temperate to tropical regions. For example, coastal marshes in the Mid Atlantic Region are best inventoried from June through October while submersed habitats in southern Florida may be inventoried best in November. Even within regions, some cover types will be more easily distinguished in different seasons. For example, in the Caribbean, estuarine seagrasses can be detected best in early January, yet marine seagrasses can be detected best in May or June. Technically, these vegetation patterns should be monitored at optimal times throughout the year, but cost limitations usually limit the analyst to a single date.

Effects of Tidal Stage on Image Classification

Tide stage is a crucial factor in satellite image scene selection and the timing of aerial surveys. Ideally, tides should be constant between time periods, but this would rule out synoptic satellite sensors since tide stages are not synchronized within a region or even within a single image. Alternatively, analysts should avoid selecting the highest tides and should take into account the tide stages occurring throughout each scene. Tidal effect varies greatly among regions. In the Northwest, for example, when all of the temporal, atmospheric, and tidal criteria are taken into account the number of acceptable scenes may be quite small. In some regions it may be necessary to seek alternative data such as SPOT satellite data, aerial photographs, or other land cover databases. For most regions, mean low tide (MLT) or lower will be preferred, one or two feet above MLT will be acceptable, and three feet or more will be unacceptable (Jensen et al. 1993a). Ideally, tides for aerial photographic surveys of submersed habitat should approach low tide as predicted in NOAA, National Ocean Service (NOS) tide tables, but optimal visualization of the subtidal bottom depends on water clarity as well as depth. Two of the 1993 C-CAP protocol development projects focus on improving the C-CAP protocol for tidal effects (See Appendix 5).

Image Processing of Satellite Remote Sensing Data to Inventory Upland and Wetland Change

With the Classification scheme developed and the appropriate remote sensor data selected, it is possible to process the data to extract upland and wetland change information. This involves geometric and radiometric correction, selection of an appropriate change detection algorithm, classification if necessary, creation of change detection products, and error evaluation (Figure 1). A separate section (Chapter 4) describes the extraction of information on SRV because aerial photography and significantly different photogrammetric techniques must be utilized.

Rectification of Multiple-Date Remote Sensor Data to Detect Change

Georeferencing (spatial registration of a remotely sensed image to a standard map projection) is a necessary step in digital change detection and cartographic representation. The following C-CAP recommendations should be followed when rectifying the base image to a standard basemap:

Rectification of an earlier date (Tb-1) or later date (Tb+1) to the base image (Tb) can be accomplished in several ways. The primary concern is to accomplish the most exact co-registration of pixels from each time period and thus reduce a potentially significant source of error in change analysis (Lunetta et al. 1991). The following are minimum recommendations and requirements:

Radiometric Normalization of Multi-Date Images to Detect Change

The use of remotely sensed data to classify coastal and upland land cover on individual dates is contingent upon there being a robust relationship between remotely sensing brightness values (BVs) and actual surface conditions. However, factors such as sun angle, Earth/sun distance, detector calibration differences between the various sensor systems, atmospheric condition, and sun/target/sensor geometry (phase angle) will also affect pixel brightness value. Differences in direct beam solar radiation due to variation in sun angle and Earth/sun distance can be calculated accurately, as can variation in pixel Bvs due to detector calibration differences between sensor systems. Removal of atmospheric and phase angle effects require information about the gaseous and aerosol composition of the atmosphere and the bi-directional reflectance characteristics of elements within the scene. However, atmospheric and bi-directional reflectance information are rarely available for historical remotely sensed data. Also, some analysts may not have the necessary expertise to perform a theoretically based atmospheric path radiance correction on remotely sensed data. Hence, it is suggested that a relatively straightforward "empirical scene normalization" be employed to match the detector calibration, astronomic, atmospheric, and phase angle conditions present in a reference scene.

Image normalization reduces pixel BV variation caused by non-surface factors so variations in pixel BVs between dates can be related to actual changes in surface conditions. Normalization enables the use of image analysis logic developed for a base year scene to be applied to the other scenes. This can be accomplished using techniques pioneered by personnel of the U.S. Bureau of Land Management (Eckhardt et al. 1990). Image normalization is achieved by developing simple regression equations between the brightness values of "normalization targets" present in Tb and the scene to be normalized (e.g. Tb-1 or Tb+1). Normalization targets are assumed to be constant reflectors, therefore any changes in their brightness values are attributed to detector calibration, astronomic, atmospheric, and phase angle differences. Once these variations are removed, changes in BV may be related to changes in surface conditions.

Acceptance criteria for potential "normalization targets" are (Eckhardt et al 1990):

The mean BVs of the Tb targets are regressed against the mean BVs of the Tb-1 or Tb+1 targets for the n bands used in the classification of the remote sensor data (e.g. TM bands 2, 3, and 4). The slope and y-intercept of the n equations are then used to normalize the Time 2 Landsat TM data to the Time 1 Landsat TM data. Each regression model contains an additive component (y- intercept) that corrects for the difference in atmospheric path radiance between dates, and a multiplicative term (slope) that corrects for the difference in detector calibration, sun angle, Earth/sun distance, atmospheric attenuation, and phase angle between dates.

It is customary to first normalize the remote sensor data and then perform image rectification (using nearest-neighbor resampling if image classification is to take place). These data are then ready for individual date classification or the application of various multi-image change detection algorithms. Most studies that attempt to monitor biophysical properties such as vegetation biomass, chlorophyll absorption, health, and other biophysical properties require atmospheric correction.

Selecting the Appropriate Change Detection Algorithm

C-CAP is the first federal program to state as a primary goal the monitoring of coastal habitat change using satellite technology (Cross and Thomas 1992). The implementation and continuing evolution of the program is based on the fact that improved cartographic, digital image processing, and photointerpretation methods must be developed for a program of this geographic coverage, spatial resolution, and temporal frequency (nationwide, 30 x 30 m pixel, every one to five years). Initial implementation of C-CAP will require a blend of traditional and innovative approaches to change analysis. The fact that the program has adopted a digital format, with the TM as a primary sensor, means that new techniques in processing can be easily incorporated into future iterations.

The selection of an appropriate change detection algorithm is very important (Jensen 1986; Dobson and Bright 1991, 1992, 1993; Jensen et al. 1993a). First, it will have a direct impact on the type of image classification to be performed (if any). Second, it will dictate whether important "from-to" information can be extracted from the imagery. C-CAP requires that the "from-to" information be readily available in digital form suitable for geographic analysis and for producing maps and tabular summaries. At least seven change detection algorithms are commonly used by the remote sensing community, including:

It is instructive to review these alternatives, identify those acceptable to C- CAP, and provide specific examples where appropriate.

Change Detection Using Write Function Memory Insertion

It is possible to insert individual bands of remotely sensed data into specific write function memory banks (red, green, and/or blue) in the digital image processing system (Figures 1. & 2.) to visually identify change in the imagery (Jensen et al. 1993b). For example, consider two Landsat Thematic Mapper scenes of the Fort Moultrie quadrangle near Charleston, SC obtained on November 11, 1982 and December 19, 1988. Band 1 of the 1982 image was placed in the green image plane and band 1 of the 1988 image in the red image plane and no image in the blue image plane (Figure 3.). All areas which did not change between the two dates are depicted in shades of yellow (i.e. in additive color theory, equal intensities of green and red make yellow). The graphic depicts numerous changes, including:

[Figure 1.]

[Figure 2.]

[Figure 3.]

Advantages of this technique include the possibility of looking at two and even three dates of remotely sensed imagery at one time as demonstrated by Jensen et al (1993b). Unfortunately, the technique does not produce a classified land cover database for either date and, thus, does not provide quantitative information on the amount of area changing from one land cover category to another. Nevertheless, it is an excellent analog method for quickly and qualitatively assessing the amount of change in a region which might help with the selection of one of the more rigorous change detection techniques to be discussed.

Multi-date Composite Image Change Detection

Numerous researchers have rectified multiple dates of remotely sensed imagery (e.g. selected bands of two Thematic Mapper scenes of the same region) and placed them in a single dataset (Figure 4.). This composite dataset can then be analyzed in a number of ways to extract change information. First, a traditional classification using all n bands (6 in the example in Figure 4.) may be performed. Unsupervised classification techniques will result in the creation of 'change' and 'no-change' clusters. The analyst must then label the clusters accordingly.

[Figure 4.]

Other researchers have used principle component analysis (PCA) to detect change (Jensen 1986). Again, the method involves registering two (or more) dates of remotely sensed data to the same planimetric basemap as described earlier and then placing them in the same dataset. A PCA based on variance-covariance matrices or a standardized PCA based on analysis of correlation matrices is then performed (Fung and LeDrew 1987, 1988; Eastman and Fulk 1993). This results in the computation of eigenvalues and factor loadings used to produce a new, uncorrelated PCA image dataset. Usually, several of the new bands of information are directly related to change. The difficulty arises when trying to interpret and label each component image. Nevertheless, the method is of value and is used frequently.

The advantage of this techniques is that only a single classification is required. Unfortunately, it is often difficult to label the change classes and no "from-to" change class information is available.

Image Algebra Change Detection

It is possible to simply identify the amount of change between two images by band ratioing or image differencing the same band in two images which have previously been rectified to a common basemap. Image differencing involves subtracting the imagery of one date from that of another (Figure 5.). The subtraction results in positive and negative values in areas of radiance change and zero values in areas of no-change in a new "change image". In an 8-bit (28) analysis with pixel values ranging from 0 to 255, the potential range of difference values is -255 to 255. The results are normally transformed into positive values by adding a constant, c (usually 255). The operation is expressed mathematically as:

Dijk = BVijk(1) - BVijk(2) + c

where Dijk = change pixel value
BVijk(1) = brightness value at Tb
BVijk(2) = brightness value at Tb-1 or Tb+1
	  c = a constant (e.g., 255).
	  i = line number
	  j = column number
	  k = a single band (e.g. TM band 4).

[Figure 5.]

The "change image" produced using image differencing usually yields a BV distribution approximately Gaussian in nature, where pixels of no BV change are distributed around the mean and pixels of change are found in the tails of the distribution. Band ratioing involves exactly the same logic except a ratio is computed between Tb and Tb+1 and the pixels which did not change have a value of "1" in the change image.

A critical element of both image differencing and band ratioing change detection is deciding where to place the threshold boundaries between "change" and "no-change" pixels displayed in the histogram of the change image (Jensen 1986). Often, a standard deviation from the mean is selected and tested empirically. Conversely, most analysts prefer to experiment empirically, placing the threshold at various locations in the tails of the distribution until a realistic amount of change is encountered. Thus, the amount of change selected and eventually "recoded" for display is often subjective and must be based on familiarity with the study area. There are also analytical methods which can be used to select the most appropriate thresholds. Unfortunately, image differencing simply identifies those areas which may have changed and provides no information on the nature of the change, i.e. no "from-to" information. Nevertheless, the technique is valuable when used in conjunction with other techniques such as the multiple-date change detection using a binary change mask to be discussed in "Multi-Date Change Detection Using a Binary Change Mask Applied to Tb-1 or Tb+1".

Post-Classification Comparison Change Detection

This is the most commonly used quantitative method of change detection (Jensen 1986; Jensen et al. 1993a) and may be used in regional C-CAP projects under certain conditions. It requires rectification and classification of each of the remotely sensed images (Figure 6.). These two maps are then compared on a pixel by pixel basis using a "change detection matrix" to be discussed. Unfortunately, every error in the individual date classification map will also be present in the final change detection map (Rutchey and Velcheck 1993). Therefore, it is imperative that the individual classification maps used in the post-classification change detection method be extremely accurate (Augenstein et al. 1991; Price et al. 1992).

[Figure 6.]

To demonstrate the post-classification comparison change detection method, consider the Kittredge (40 river miles inland from Charleston, S. C.) and Fort Moultrie, S. C. study areas (Figure 7. abcd) (Jensen et al. 1993a). Nine (9) classes of land cover were inventoried on each date (Figure 8. abcd). The 1982 and 1988 classification maps were then compared on a pixel by pixel basis using an n by n GIS "matrix" algorithm whose logic is shown in Figure 9.. This resulted in the creation of "change images maps" consisting of brightness values from 1 to 81. The analyst then selected specific "from-to" classes for emphasis. Only a select number of the 72 (n2-n) possible off-diagonal "from - to" land cover change classes summarized in the change matrix (Figure 9.) were selected to produce the change detection maps (Figure 10. ab). For example, all pixels which changed from any land cover in 1982 to "Developed Land" in 1988 were color coded red (RGB = 255, 0, 0) by selecting the appropriate "from - to" cells in the change detection matrix (10,19,28,37,46, 55,64, and 73). Note that the change classes are "draped" over a TM band 4 image of the study area to facilitate orientation. Similarly, all pixels in 1982 which changed to "Estuarine Unconsolidated Shore" by December 19, 1988 (cells 9,18,27,36,45,54,63, and 72) were depicted in yellow (RGB = 255, 255, 0). If desired, the analyst could highlight very specific changes such as all pixels which changed from "Developed Land" to "Estuarine Emergent Wetland" (cell "5" in the matrix) by assigning a unique color look-up table value (not shown). A color coded version of the "Change Detection Matrix" can be used as an effective "from-to" change detection map legend (Jensen and Narumalani 1992).

[Figure 7.]

[Figure 8.]

[Figure 9.]

[Figure 10.]

Post-classification comparison change detection is widely used and easy to understand. When conducted by skilled image analysts it represents a viable technique for the creation of C-CAP change detection products. Advantages include the detailed 'from-to' information and the classification map for each year. Unfortunately, the accuracy of the change detection is heavily dependent on the accuracy of the two separate classifications. The post-classification comparison is not recommended for C-CAP regional projects except under special circumstances, such as when different sensors are involved or when two separate organizations are classifying the same region at different times.

Multi-Date Change Detection Using a Binary Change Mask Applied to Tb-1 or Tb+1

This method of change detection is highly recommended for C-CAP regional projects. First, the analyst selects the base image, Tb. Date 2 may be an earlier image (Tb-1) or a later image (Tb+1). A traditional classification of Tb is performed using rectified remote sensor data. Next, one of the bands (e.g. band 3 in Figure 11.) from both dates of imagery are placed in a new dataset. The two band dataset is then analyzed using various image algebra functions (e.g. band ratio, image differencing, principal components) which produces a new image file. The analyst usually selects a "threshold" value to identify spectral "change" and "no-change" pixels in the new image as discussed in "Image Algebra Change Detection". The spectral change image is then recoded into a binary mask file, consisting of pixels with spectral change between the two dates, and these are viewed as candidate pixels for categorical change. Great care must be exercised when creating the "change/no-change" binary mask (Dobson and Bright 1993; Jensen et al. 1993a). The change mask is then overlaid onto Tb-1 or Tb+1 and only those pixels which were detected as having changed are classified in Tb-1 or Tb+1. A traditional post-classification comparison (previous section) can then be applied to yield "from-to" change information. Hence, many pixels with sufficient change to be included in the mask of candidate pixels may not qualify as categorical land cover change.

[Figure 10.]

Dobson and Bright (1991, 1992, 1993) utilized this change detection methodology to inventory change in the area surrounding the Chesapeake Bay using Thematic Mapper imagery obtained on September 9, 1984 and November 3, 1988 (the region centered on Metomkin Inlet is shown in Figures 11. and 12.). The 1988 base image was classified using traditional supervised classification techniques (Figure 13.). A "change/no-change" mask was derived by performing image arithmetic on bands 3, 4, and 5 of the two date dataset. All change pixels were combined into a single change mask (Figure 14.) . The "change/no-change" mask was then overlaid onto the earlier date of imagery and only those pixels which were detected as having changed were classified in the earlier image. A "from-to" matrix similar to the one shown in Figure 9. was then used to produce a change map of the region (Figure 15.). Summary statistics for the region are found in Table 3. This process may be repeated with a later scene to determine successive change.

[Figure 12.]

[Figure 13.]

[Figure 14.]

[Figure 15.]

This method may reduce change detection errors (omission and commission) and provides detailed "from-to" change class information. The technique reduces effort by allowing analysts to focus on the small amount of area that has changed between dates. In most regional projects, the amount of actual change over a one to five-year period is probably no greater than 10% of the total area. The method is complex, requiring a number of steps and the final outcome is dependent on the quality of the "change/no-change" binary mask used in the analysis. A conservative threshold may exclude real change while a liberal threshold may create problems similar to those of the post classification comparison technique (See Post-Classification Comparison Change Detection).

Multi-date Change Detection Using Ancillary Data Source as Tb

Sometimes there exists a land cover data source which may be used in place of a traditional remote sensing image in the change detection process. For example, the NWI is in the process of inventorying all of the wetlands in the United States at the 1:24,000 scale. Some of these data have been digitized. Instead of using a remotely sensed image as Tb in the analysis, it is possible to substitute the digital NWI map of the region (Figure 16.). In this case, the NWI map would be "recoded" to be compatible with the C-CAP Coastal Land Cover Classification System (Table 2.). This should not be difficult since the two systems are highly compatible. Next, Tb-1 or Tb+1 is classified and then compared on a pixel by pixel basis with the Tb information. Traditional "from-to" information can then be derived. As with any other post classification comparison, the accuracy of the change database is dependent on the accuracy of both input databases (C-CAP and NWI).

[Figure 16.]

Advantages of the method include the use of a well-known, trusted data source (NWI) and the possible reduction of errors of omission and commission. Detailed "from-to" information may be obtained using this method. Also, only a single classification of the Tb-1 or Tb+1 image is required. It may also be possible to up-date the NWI map (Tb) with more current wetland information (this would be done using a GIS 'dominate' function and the new wetland information found in the Tb-1 or Tb+1 classification). The disadvantage is that the NWI data must be digitized, generalized to be compatible with the C-CAP Coastal Land Cover Classification System, then converted from vector to raster format to be compatible with the raster remote sensor data. Any manual digitization and subsequent conversion introduces error into the database which may not be acceptable (Lunetta et al. 1991).

Manual On-screen Digitization of Change

Considerable amounts of high resolution remote sensor data are now available (e.g. SPOT 10 x 10 m, the aircraft mounted CASI - Calibrated Airborne Spectrographic Imager, NAPP- National Aerial Photography Program). These data can be rectified and used as planimetric basemaps or orthophotomaps. Often aerial photographs are scanned (digitized) at high resolutions into digital image files (Light 1993). These photographic datasets can then be registered to a common basemap and compared to identify change. Digitized high resolution aerial photographs displayed on a CRT screen can be interpreted easily using standard photo interpretation techniques based on size, shape, shadow, texture, etc. (Ryerson 1989). Therefore, it is becoming increasingly common for analysts to visually interpret both dates of aerial photographs (or other type of remote sensor data) on the screen, annotate the important features using heads-up "on- screen" digitizing, and compare the various images to detect change (Wang et al., 1992; Lacy 1992; Cowen et al. 1991; Westmoreland and Stow 1992; Cheng et al. 1992). The process is especially easy when a) both digitized photographs (or images) are displayed on the CRT at the same time, side by side, and b) they are topologically "linked" through object-oriented programming so that a polygon drawn around a feature on one photograph will have the same polygon drawn around the same object on the other photograph. Scanning aerial photographs unavoidably will reduce the spatial and spectral resolution of the source data. This loss may be significant in photographs of submerged features, which are subject to interferences from aquatic as well as atmospheric sources. As with other new technologies, demonstration of the appropriateness of interpretation of scanned photographs will be a critical step in expanding the C-CAP Protocol (Also see "Accuracy Assessment for Individual Date Classification of Water and Submersed Habitat Data", page #). The manual on-screen approach is recommended as a useful adjunct to other change detection methods. Its principle drawback is the time required to cover large regions in such a labor-intensive fashion.

Selecting Appropriate Classification Algorithms

C-CAP requires that the classification procedures used as part of the change detection process be approved and documented. Selection of the classification algorithms used in each region will be based on the capabilities and needs of the regional participants. C-CAP assumes that the regional participants are experienced in image processing and mapping. If not, C-CAP will attempt to provide fundamental technical assistance on a case by case basis.

The previous section indicated that three of the seven most commonly used change detection algorithms are acceptable for C-CAP regional projects:

Each of these requires a complete pixel by pixel classification of one date of imagery and, at least, a partial classification of an additional date. Hence, it is instructive to review the C-CAP approved image classification logic which may be used in the regional projects.

Supervised and Unsupervised Image Classification Logic

The primary reason for employing digital image classification algorithms is to reduce human labor and improve consistency. It is expected that regional analysts will have sufficient expertise to assess the advantages of alternative classification algorithms and to recognize when human pattern recognition and other types of intervention are necessary. In practice, it may be necessary to employ a suite of algorithms including both supervised and unsupervised statistical pattern recognition approaches. Currently maximum-likelihood classifiers often serve as a good first step, but new statistical approaches are being developed and implemented on a routine basis (Jensen et al. 1987; Hodgson and Plews 1989; Foody et al. 1992). It is important for analysts to remain flexible with regard to procedures and algorithms.

Standard supervised and unsupervised classification techniques have been available for more than 20 years and are well documented in texts by Jensen (1986) and Campbell (1987). In a supervised classification, the analyst "trains" the classifier by extracting mean and co-variance statistics for known phenomena in a single date of remotely sensed data (Gong and Howarth 1990). These statistical patterns are then passed to a minimum-distance-to-means algorithm where unknown pixels are assigned to the class nearest in n-dimensional feature space, or to a maximum-likelihood classification algorithm which assigns an unknown pixel to the class in which it has the highest probability of being a member. Great care must be exercised when selecting training samples (Mausel et al. 1990).

In an unsupervised classification, the computer is allowed to query the multispectral properties of the scene using user specified criteria and to identify x mutually exclusive clusters in n-dimensional feature space (Chuvieco and Congalton 1988). The analyst must then convert (label) the x spectral clusters into information classes such as those found in the C-CAP Coastal Land Cover Classification System. Training sites visited in the field and identifiable in the digital imagery are also indispensable when labeling clusters in an unsupervised classification. The following sections discuss C-CAP guidelines for collecting training and verification samples.

Selection of Training and Verification Samples for Supervised and Unsupervised Classification

Only training sites which were actually visited on the ground by experienced professionals should be selected for extracting the multispectral statistical "signature" of a specific class when performing a supervised or unsupervised classification. It is suggested that a minimum of five training sites per land cover class be collected. This creates a representative training set when performing supervised classification and makes labeling clusters much easier in an unsupervised classification. In addition to the image analysts, the field team should contain specialists in ecology, biology, forestry, geography, statistics, and other pertinent fields such as agronomy. Field samples should be stratified by land cover type and by various physical factors such as slope, elevation, vegetation density, species mix, season, and latitude. The polygonal boundary of all field sites should be measured using global positioning systems whenever possible, and the locational, temporal, and categorical information archived.

The collection of field training sites often requires multiple visits to the field. Some of the field sites may be used to train a classifier or label a cluster while a certain proportion of the field sample sites should be held back to be used for classification error assessment to be discussed.

The following materials are indispensable to a successful field exercise:

It is advisable to perform, at least, a cursory classification before initiating fieldwork. In this case, both raw and classified data should be taken to the field. The primary function of the cursory classification is to guide field workers in targeting the covers and signatures that are most difficult and confusing. Keep in mind that the vast majority of all cover will be easy to identify on the ground and on the imagery. Efficient use of field time provides for quick verification of easy cover types and maximum attention to difficult, unusual, and ecologically critical cover types.

Field investigators should anticipate the need to know, not only the geodetic coordinates of training sites, but also the layout of the road network that will provide access. It is advisable to imbed roadway information into the raw imagery. This can be done using the Bureau of the Census's Topologically Integrated Geographic Encoding and Referencing (TIGER) files. Imbedding is preferred rather than transparent overlay techniques which are cumbersome and difficult to use under field conditions.

C-CAP investigators have assembled and tested a field station based on a color laptop computer with commercial software. At present the software supports visualization of raster imagery (eg. satellite data, digital orthophotographs, scanned aerial photographs) and vector databases (eg. Tiger road networks, NWI wetlands). A version of the software soon to be available from commercial vendors will allow for realtime input of GPS coordinates. It will then be possible to follow field movements directly on the image and map data. The software also allows for completion of field forms on screen in the field. Preliminary tests are encouraging, but the field station is not fully operational at this time. One shortcoming, for example, is the poor performance of active matrix color screens in sunlight.

The Use of Collateral Data in Image Classification

The overriding goal is to produce accurate individual date classifications and change detection databases. Any information or operation that enhances data quality is generally encouraged. C-CAP does not endorse the notion that the use of collateral data in a remote sensing project is "hedging" or "fudging". Instead, the objective is to innovatively use collateral data to improve the accuracy of the C-CAP database.

There are many potential sources of collateral data including soil maps, NOAA coastlines (T-sheets), timber surveys, USGS digital line graphs, and digital elevation models (for elevation, slope, and aspect). These can be incorporated by masking, filtering, probability weighting, or inclusion in the signature file (Ryerson 1989; Baker et al. 1991). Depending on the importance of each category, analysts may use certain categories to overrule others (Jensen et al. 1993a).

The NWI is an especially valuable collateral database that may be of value when classifying wetlands. Regional analysts should incorporate NWI data to the maximum extent possible. NWI data are recognized as the most authoritative and complete source of wetlands land cover data (Wilen 1990). However, NWI maps are not temporally synchronized in each region and are not in a digital format for many regions. An approach based on complementary use of NWI and imagery will be an asset to both C-CAP and NWI. At a minimum, NWI maps and/or digital data should be used to define training samples, to check intermediate results, and to aid in the final verification of the wetlands portion of the C-CAP maps. NWI digital data may be used as a probability filter in the classification process. In this approach, C-CAP recommends an "innocent until proven guilty" attitude toward the NWI data. In other words, the NWI category is considered correct for a given pixel area for each time period, unless spectral signatures or collateral data suggest that the NWI category is incorrect or a land cover change has occurred. Even if the NWI data were 100% correct at the time of NWI mapping, overriding by spectral data would be necessary to detect change over time. Ultimately, in turn, the C-CAP change detection database can assist NWI managers in determining the need for NWI updates.

Cartographic Portrayal of Individual Classification Maps and Change Detection Maps

C-CAP products must meet stringent cartographic standards. The following sections discuss the minimum measurement unit and its proper use when aggregating change information. Formats of classification maps and change maps must satisfy C-CAP criteria whenever hardcopy maps are produced.

The Concept of the Minimum Measurement Unit

The minimum measurement unit is a measure of both precision and accuracy of the input data. For most C-CAP regional projects the input data will be 30 x 30 m pixel data recorded by a Landsat TM sensor. The minimum measurement unit, however, combines the ability (e.g., sensor limitations) and effort (e.g., field verification) required to measure a category with the spatial precision and accuracy necessary to accomplish the intended use of the data. Each land cover category could potentially have a different minimum measurement unit based on the size of individual parcels and the distinctiveness of the signature. Thus, the minimum measurement unit differs from a traditional minimum mapping unit which by definition imposes a predetermined polygon (or pixel) size for all land cover categories (for example, a rule that all parcels of one hectare or larger will be mapped). This traditional approach is acceptable for manual mapping using analog aerial photographs but is difficult to apply to raster imagery. Regional analysts will be responsible for defining minimum measurement units, generally larger than a single pixel but no larger than three pixel dimensions on the short axis.

Regardless of the minimum measurement unit, change analysis will be conducted pixel by pixel. C-CAP protocol requires that the inherent resolution of the raw data must be retained throughout the classification and change analysis processes. Aggregation and filtering of pixels should occur only in regard to cartographic presentation of the completed change detection database.

Regardless of the techniques employed, the final database should be capable of representing land cover by class for the base time, land cover by class for each earlier or later time, and land cover change by class for each change period. The final database should contain the full change matrix (all "from" and "to" categories) for each change period.

Analog (Hardcopy) Cartographic Products

Hard copy maps of the final database are not specifically required by C- CAP, but they are certain to be of use when presenting the results. Often it is useful to produce a smaller scale regional map (usually requiring some pixel aggregation) to give an impression of the scope of the effort and several larger scale maps at full resolution to demonstrate the level of detail and to highlight notable findings. All maps should come directly from the final database complying with C-CAP protocols, but overlay or imbedding of ancillary data, such as DLG and Bureau of the Census TIGER data, is encouraged with proper notation.

If the statistical summary of changes is present on a map, C-CAP recommends that the numbers included in it always be calculated for the area shown on the map. It is not acceptable to associate the summary of changes for one area (larger or smaller) with a map of another. The statistical summaries of the change detection matrix must always be calculated from the database at full resolution, rather than from the aggregated data of the plot file. It is not advisable to allow the numerical count of class area to float with the level of cartographic aggregation. Unless all counts are based on the full resolution database, some classes composed of small features may disappear at higher levels of aggregation. Map readers will become confused if matrix numbers change with aggregation for the same territory.

Technically, the minimum cartographic presentation is 1) a map for the base time, 2) a map showing gains by class, and 3) a map showing losses by class. A full classification for the earlier or later (non-base) time may be useful, but it is not essential to present the matrix of possible changes. Examples of some of these products are found in Figures 7-14. (Table 4.).

Table 4. Statistical summary of areal change (ha) by land cover class for the Metomkin Inlet area shown in Figures 12-16.. Read across each row to find which categories the 1988 totals came from. Read down each column to find which categories the 1984 totals changed to. Bold numbers along the diagonal indicate the area that did not change from 1985 to 1988.

   Dev.  | Grass | Forest | Scrub | Pal. For. | Est. Em. | Pal. Em. | Water | Bare | Total
_________________________________________________________________________________________________
Dev.     | 1158 |    85 |     8  |   0   |     0     |      4   |       0  |     1 |   0  |  1256
Grass    |    0 | 21341 |   562  |   0   |     0     |      2   |       0  |     0 |  17  | 21922
Forest   |    0 |   165 | 18915  |   0   |     0     |      1   |       0  |     0 |   0  | 19081
Scrub    |    0 |   240 |   562  | 854   |     0     |      2   |       0  |     0 |   0  |  1658
Pal. For.|    0 |    20 |     9  |   0   |   787     |      0   |       0  |     0 |   0  |   816
Est. Em. |    0 |    26 |     9  |   0   |     0     |  11587   |       0  |    13 |   8  | 11643
Pal. Em. |    0 |     0 |     0  |   0   |     0     |      0   |       0  |     0 |   0  |     0
Water    |    0 |     4 |     0  |   0   |     0     |      2   |       0  | 37172 | 144  | 37322
Bare     |    0 |    19 |     0  |   0   |     0     |     23   |       0  |   124 | 507  |   673
___________________________________________