GIS for Water Resources and Watershed Management - Chapter 10

CHAPTER 10 Evaluation of the Albemarle-Pamlico Estuarine Study Area Utilizing Population and Land Use Information Robert E. Holman BACKGROUND The Albemarle-Pamlico Estuarine Study (APES) has been funding many information acquisi-tion projects over the last five years in the areas of resource critical areas, water quality, fisheries, and human environment (Steel and Scully, 1991). Most of these projects have transferred their data over to the APES’s Geographic Information System (GIS) which was created through a sub-contract with the North Carolina Center for Geographic Information System (CGIA). GIS has the ability to bring together (enter, display, edit, and manipulate) data based information with digital mapping (locational attributes). At the time of this study, the Center (CGIA) had or was creating all of the needed databases. CGIA was able to combine the data layers in various ways to analyze the relationship among different layers in a visual as well as a statistical manner. The study area encompasses approximately 23,250 square miles and includes all or portions of 37 counties in eastern North Carolina and 19 counties in Southeastern Virginia. There are six counties along the coastline, 9 counties along the sounds, and 41 cities/counties that lie in the upper drainage basin (Figure 10.1). This study also incorporates all or portions of 6 major river basins including the Chowan, Pasquotank, Lower Roanoke, Tar-Pamlico, Neuse, and White Oak (Figure 10.2). Each basin is divided into subbasins as follows: Chowan, 13; Pasquotank, 8; Lower Roanoke, 3; Tar-Pamlico, 8; Neuse, 14; and White Oak, 5. METHOD The analytical method was broken into three phases. Phase One was the creation of county land use maps from the existing Landsat classification scheme. These map products were sent to U.S. Fish and Wildlife and county officials to determine the accuracy of the defined land use classes. The land use maps were also used by the author during flights over the coastal and metropolitan areas to further clarify classification errors. Phase Two was to correct some of the errors in the ex-isting classification. This was carried out by digitizing the corrections to the map products that were returned from Fish and Wildlife and county officials. The map information was also supple-mented with other sources of information such as U.S. Fish and Wildlife Service—National Wet-land Inventory, U.S. Forest Service—Forest Inventory and Analysis, U.S. Bureau of Census—Census of Agriculture, and U.S. Soil Conservation Service—National Resources Inven-tory and Hydric Soils in North Carolina counties. Phase Three was identifying correlation between 109 © 2003 Taylor & Francis Chapters 1, 3, 5 & 6 © American Water Resources Association; Chapter 13 © Elsevier Science; Chapter 14 © American Society for Photogrammetry and Remote Sensing 110 GIS FOR WATER RESOURCES AND WATERSHED MANAGEMENT Figure 10.1. Map of study area. different data sets such as county census population and county acreage of developed land. If a strong correlation was found, then a simple linear regression model was applied in order to predict the relationship between the two parameters. These models were used to correct some of the errors in specific land use categories. Map Development There were three tasks in the development of land use and population estimates for the entire APES area: (1) defining the actual drainage area; (2) having all the basin and subbasin boundaries digitized in order to determine the land use and population; and (3) correcting for errors associated with the different land uses. First, the study area was defined as the entire drainage area of the Albemarle and Pamlico Sound system including Core and Bogue Sounds. The upper Roanoke Basin and a portion of the White Oak Basin were not included because: (1) the upper Roanoke River Basin covers approxi-mately 8,370 square miles in Virginia/North Carolina and stretches over two-thirds the length of North Carolina, and would add one-third more area to the study area; and (2) a decision was made early in APES to have Carteret County as the furthest area south. However, due to the watershed approach in defining the study area in this project, all the subbasins in the White Oak Basin were included except the furthest one southwest that starts at Camp Lejeune. There was no compatible land use data available for this subbasin. © 2003 Taylor & Francis Chapters 1, 3, 5 & 6 © American Water Resources Association; Chapter 13 © Elsevier Science; Chapter 14 © American Society for Photogrammetry and Remote Sensing EVALUATION OF THE ALBEMARLE-PAMLICO ESTUARINE STUDY AREA 111 Figure 10.2. The six APES basins and their subbasins. Second, all North Carolina basins and subbasins were digitized by the Research Triangle Insti-tute and compared closely with the U.S. Geological Survey subbasins for North Carolina. Virginia subbasin information was supplied by Information Support Systems Laboratory within Virginia Polytechnic Institute and State University and was based on Soil Conservation Service (SCS) in-formation. Due to the large number of subbasins identified by SCS in the Virginia portion of the Chowan and Pasquotank Basins, subbasins were combined to create areas of the same size range as subbasins identified in North Carolina. All subbasins were digitized from U.S. Geological Sur-vey’s 1:24,000 scale topographic maps. Specific subbasins were identified by a six number code that was broken into two-digit sets. The first two digits identified the regional basin; the second two digits identified the basin; and the third two digits identified the subbasin (Figure 10.2). Codes used in this report were the same ones adopted by the North Carolina Division of Environmental Management. The third task was to identify the accuracy of the land use data and to develop methods to cor-rect for the errors. Khorram and others (1992) found that with the Landsat data, the urban or built-up land use category was only 46% accurate, and the accuracy of forested wetlands was unknown. In addition, the classification of mixed pixels in the existing land use data set had to be resolved. Mixed pixels are defined as areas that could not be classified because the resolution or pixels were a mixture of many categories. The land use classification from 1987–1988 developed by Khorram will be referred to as the “Landsat” classification in this study. © 2003 Taylor & Francis Chapters 1, 3, 5 & 6 © American Water Resources Association; Chapter 13 © Elsevier Science; Chapter 14 © American Society for Photogrammetry and Remote Sensing 112 GIS FOR WATER RESOURCES AND WATERSHED MANAGEMENT Accuracy and Errors The first task was to define which level of land use to utilize. Landsat land use classification defined 18 separate classes that can be generally broken into similar U.S. Geological Survey level I and level II groupings (Anderson and others, 1976). The land use classification was based on Landsat data which Khorram interpreted mostly as land cover (the actual extent of vegetative and other cover) and some land use (interpretation of activities taking place on the land). Interpretation of land use is much more subjective than land cover and is dependent on the knowledge of the in-dividual interpreter. A level II map with 18 individual classes was provided to officials of two Fish and Wildlife Refuges within the APES area for their evaluation as to land cover accuracy. The Great Dismal Swamp National Wildlife Refuge staff reviewed the Landsat land use map of the refuge. This refuge, located on the border between North Carolina and Virginia just south of Portsmouth, Vir-ginia covers approximately 110,000 acres and is predominantly forested wetland. The staff felt there was good separation among development, agriculture, water, and forest; however, the differ-ent forest cover types had serious reliability problems. A major problem was the misclassification of wetter deciduous stands like cypress/gum and maple/gum as pine/hardwood forest. A second land cover map was sent to Mattamuskeet and Swan Quarter National Wildlife Refuges personnel for their review. These two refuges are located entirely in Hyde County, North Carolina, and Swan Quarter is adjacent to the Pamlico Sound. These refuges together cover approximately 65,700 acres and are predominantly water, wetland, and forest. The staff found quite a few areas that were referred to as mixed pixels that were actually open water and irregularly flooded brackish marshes. White Cedar stands were actually marsh impoundment areas around Lake Mattamuskeet, and pine forest was actually mixed pine/hardwood or hardwood/cypress/pine forest. In general, both refuges indicated accuracy problems with the different forest and the mixed pixel classifications. After indicating they could not evaluate all the classifications in level II land cover maps, offi-cials of counties in the Currituck Sound Basin south of Virginia Beach and adjacent to the Atlantic Ocean were sent, for comment, land cover maps which included the following attributes: USGS level I with 6 categories shown in color; U.S. Census TIGER files that displayed the road network, map scale of 1:100,000; and modified LUDA land use data for the urban or built-up category. Land Use Data Analysis (LUDA) was an early GIS effort started by U.S. Geological Survey in 1975 to define the land use for the entire United States. All the photographs were manually pho-tointerpreted. The map series consisted of 1:250,000 scale maps of North Carolina defining 37 uses based on the level II classification system. Source imagery was 1:56,000 color infrared pho-tography and 1:80,000 black and white photography dating back to 1970. Resolution was 10 acres for the urban or built-up categories and 40 acres for the remaining classifications (Kleckner, 1981). For comparison, the Khorram classification was based on 1987–1988 Landsat satellite im-agery that was semiautomatically interpreted. The county map series was at a scale of 1:100,000 with a final resolution of 1 acre. In 1991 and 1992, the author flew along the Outer Banks and inland around the estuarine por-tion of the study area and over portions of Wake, Durham, and Orange Counties to verify the prob-lems with the categories of urban or built-up, wetland, and mixed pixels. The urban or built-up class was underestimated on the 1987–1988 land use maps mainly due to forest crown cover that obscured the true land use on the ground. High spectral reflectance of bare agricultural fields was also a problem because these fields were being classified as developed areas. The problem with fields being identified as developed areas was especially evident in the Landsat scene furthest west that included the Raleigh metropolitan area. This category was a very small percentage (3.3 to © 2003 Taylor & Francis Chapters 1, 3, 5 & 6 © American Water Resources Association; Chapter 13 © Elsevier Science; Chapter 14 © American Society for Photogrammetry and Remote Sensing EVALUATION OF THE ALBEMARLE-PAMLICO ESTUARINE STUDY AREA 113 7.9%) of the overall land use of the study area but is critical in that urban or built-up land use can cause the greatest impact on natural resources of the APES area. Problems associated with the wetland class were found to be interference from forest crown cover. Open marsh and pocosin wetlands were usually accurately defined by the 1987–1988 land use maps but closed forest canopy prevented standing water below the forest to be seen. Therefore these true wetland types were usually defined as forest. The category of mixed pixels is a grouping the classification scheme could not identify. Flights over the coastal and metropolitan areas verified that in most cases they were a mixture of standing water and wetland vegetation. The only exception to this observation was in Pasquotank County where poorly drained agricultural land was defined as mixed pixels or wetland on the county land use map. These land use classification problems and others were identified in a workshop the author at-tended to verify remotely sensed land cover data for the Coastwatch Change Analysis Program of the National Oceanic and Atmospheric Administration (Burgess and others, 1992). The problems fell into four categories: classification error, cover versus land use, categorical resolution, and change de-tection. Classification errors included “salt and pepper” effect of individual pixels, shadows and bare ground as urban areas, and problems with the degree of wetness during image acquisition. Cover ver-sus land use had the inherent problem with distinguishing land uses and required ancillary data. Cat-egorical resolution was related to the spatial resolution and improper classification. The change detection problem involved the ability to detect a change but not always the nature of the change. From the author’s own observations and the results of this workshop, methods were developed to overcome some of the problems associated with remotely sensed land cover data. Land use information used in this study was analyzed according to the Khorram and others (1992) classification system but condensed from 18 to 7 categories. Certain corrections were in-corporated into some classes depending on the observed and documented error associated with each class. The LUDA data set was used to determine “developed land” because the information appeared to be closer to the actual extent and location than the original 1987–1988 Landsat data set. Corrected Landsat built-up areas on the maps returned by county officials were found to have a high degree of correlation (R2 of 0.9) with the LUDA “developed” category. A linear regression model was used to predict built-up land from the LUDA data. U.S. Fish and Wildlife’s National Wetland Inventory (NWI) data were used as a reliable source of wetland acreage for the coastal plains of North Carolina (Wilen, 1990 and Burgess and others, 1992). Wetland acres for twelve of the coastal counties was provided by Kevin Morehead of the Savannah River Ecology Laboratory. The same procedure used to correct built-up areas was used to reconcile the Landsat wetland cat-egory with the NWI acreage. A high correlation resulted with a R-squared of 0.9 and a simple lin-ear regression model was used to predict wetland acres from the NWI values. The mixed pixel unidentified category was determined by the U.S. Fish and Wildlife personnel and two overflights of the APES area to be predominantly wetland in nature. The mixed pixel figures were incorpo-rated into the wetland classification. RESULTS Land Use The entire APES area land use classification is based on a modified U.S. Geological Survey’s level I classification scheme. One fact to keep in mind is that the water class is not a true land use but is a very important classification. There were seven classes with the following percentages: “urban” 4.8%, “agriculture” 28.2%, “forest” 28.4%, “water” 14.6%, “wetland” 20.5%, “shrub © 2003 Taylor & Francis Chapters 1, 3, 5 & 6 © American Water Resources Association; Chapter 13 © Elsevier Science; Chapter 14 © American Society for Photogrammetry and Remote Sensing ... - tailieumienphi.vn