Title: Iowa Vegetation Cover Mapping Project
Approvals: Iowa State University
Erwin E. Klaas, Technical Project Manager Date
Kevin Kane, QA Manager Date
U.S. Environmental Protection Agency, Region 7
Marla Downing, EPA Project Manager Date
Ernie Arnold, EPA Region 7 QA Manager Date
A2. Table of Contents
A5. PROBLEM DEFINITION/BACKGROUND
A7. QUALITY OBJECTIVES AND CRITERIA FOR MEASUREMENT DATA
B. Data Acquisition/Image Classification
B1. SAMPLE DATA ACQUISITION PROCESS DESIGN
B2. SAMPLE AND ANCILLARY DATA ACQUISITION METHODS
B4. ANALYTICAL METHODS/REQUIREMENTS
B5. QUALITY CONTROL METHODS/REQUIREMENTS
B7. INSTRUMENT/EQUIPMENT TESTING, INSPECTION AND MAINTENANCE
B8. INSTRUMENT CALIBRATION AND FREQUENCY
B9. INSPECTION/ACCEPTANCE REQUIREMENTS FOR SUPPLIES AND CONSUMABLES
D. Data Validation and Usability
D1. DATA REVIEW, VALIDATION AND VERIFICATION REQUIREMENTS
D2. VALIDATION AND VERIFICATION METHODS
Appendix: Field data Collection Packet sent to cooperating private, county, state, and federal field personnel
Figure 1. Organization Chart of Personnel Involved in Project
A3. DISTRIBUTION LIST
Iowa State University: Erwin E. Klaas, Project Manager
Kevin Kane, QA Manager
Robin McNeely, Lab Manager
Iowa DNR: Jim Giglierano, Geological Survey
Todd Bishop, Geological Survey
EPA - Region 7: Marla Downing, Project Manager
Ernie Arnold, QA Manager
Mid-America Remote Sensing Consortium:
James W. Merchant, University of Nebraska-Lincoln
David Diamond, Missouri Resource Assessment Prog.
Jonathan Jens, South Dakota State University
Larry Strong, USGS/Northern Prairie Science Center
Steve Egbert, University of Kansas-Lawrence
NRCS-Natural Resources Inventory Assessment Institute:
Dean Thompson, Director
Sara Nusser
Jay Breidt
Drake University
Tom Rosberg
The individuals directly involved with the Iowa Vegetation Cover Mapping Project and their specific responsibilities are outlined below.
The Iowa Vegetation Cover Mapping Project will be supervised by Erwin E. Klaas and Kevin Kane, co-principal investigators of the Iowa Gap Analysis Project (Iowa GAP). Iowa GAP shares a vested interest in developing the most useful and accurate land cover product possible for the state of Iowa. Co-workers and graduate students assigned to the Iowa GAP, faculty at Iowa State University, and cooperators with the Iowa Department of Natural Resources (Geological Survey Bureau, GIS Section) will provide support assistance with relevant technical issues. Co-PIs have joint responsibility for overall coordination of work, product standards and accuracy assessment. Both will oversee the preparation of the QAPP and implementation of the approved version of QAPP.
Co-PI Erwin Klaas, Leader of the Iowa Cooperative Fish & Wildlife Research Unit (USGS/BRD) and Professor of Animal Ecology at Iowa State University has more than 25 years experience in managing ecological research projects. He coordinates Iowa GAP with the National Gap Analysis Program and is the main contact between Iowa GAP and cooperators in state and local agencies assisting with field data collection in Iowa. He will also oversee the production of vertebrate distribution layers for Iowa GAP, assisted by Kathleen Anderson, graduate research assistant in Animal Ecology. He will conduct in-house audits of field operations.
Co-PI Kevin Kane is manager of the university's GIS Support Facility and is responsible for the development of database design, hardware acquisitions, software procurement, training and quality assurance for GIS- and GPS- related activities throughout the university. He is the main contact for GIS inquiries from the public as well as other local, state and federal agencies. He will supervise the production of land cover and stewardship layers for Iowa GAP, assisted by Patrick Brown, graduate research assistant in Landscape Architecture. He will conduct in-house audits of GIS operations.
Robin McNeely, ISU Research Associate and GIS Coordinator for Iowa GAP has primary responsibility for GIS classsification of vegetation, land cover mapping and production of land cover products. She will oversee data documentation and metadata development.
James Giglierano, Iowa DNR-Geological Survey Bureau, will coordinate the acquisition of satellite imagery from the EROS Data Center. He will also provide GIS technical support and produce Phase 1 land-cover classifications. Todd Bishop, Iowa DNR-Geological Survey Bureau is helping to coordinate the acquisition of field land cover data from cooperating agencies and individuals.
Tom Rosberg, Plant Ecologist, and Asst. Prof. of Biology at Drake University will be responsible for developing the vegetation classification scheme used in cover mapping. He will also assist with designing and collecting field reference data for accuracy assessment.
Marla Downing, EPA Region 7, Project Manager has responsibility for overall coordination and implementation of the project and will review and approve QAPP and subsequent revisions in terms of project scope and objectives. She will ensure QAPP implementation.
Ernie Arnold, EPA Region 7 QA Manager has responsibility for review and approval of QAPP and subsequent revisions in terms of quality assurance aspects and will conduct assessments of field activities.
The following individuals and agencies are anticipated users of the land cover data and products. Most of these agencies will also assist with technical assistance and the collection of field data.
Project Objectives:
Background:
Environmental analyses need to include assessment of both the characteristics of land cover and land use and the spatial structure of landscapes. Extensive change has occurred in land cover and land use throughout the United States over the last two centuries. In Iowa the majority of the remaining naturally-vegetated areas are scattered and fragmented with many landscapes retaining only remnants of their original vegetation. Such altered landscapes function in radically different ways relative to continuous, non-altered systems and generally provide reduced ecological services.
In order to carry out its role as Iowa's only land-grant university and leader in the areas of education, research and out-reach, Iowa State University needs to develop databases for land cover and vegetation that can be used for landscape level environmental analyses and conservation planning.
Iowa does not currently have a detailed vegetation inventory. A map (and digital GIS database) depicting the locations and characteristics of vegetation communities is needed by the university to assist faculty and graduate students working on projects concerning conservation of natural resources, environmental protection, restoration, and analysis. This database must be constructed with a consistent and comparable classification system, and at a common spatial resolution, across the state and region. Iowa State University has the capability for mapping land cover in Iowa but lacks the necessary field data and imagery to map at the level of the vegetation community or alliance.
Iowa State University will greatly benefit from the capability to produce and maintain a detailed vegetation database for large scale analyses. This database will be designed so that the University can improve its capability for hydrologic (e.g. runoff) modelling, non-point source pollution assessment, groundwater quality evaluation, siting of solid waste disposal facilities, environmental risk assessment, natural resource conservation planning and other types of analyses that require vegetation information. These analyses and products are important for the University to conduct work relative to the provisions of the Clean Water Act and other state and federal programs.
A6. PROJECT/TASK DESCRIPTION
Study Area:
The project will include all of the state of Iowa. Much of the proposed work will consist of information synthesis in a GIS laboratory, but ground reconnaissance and accuracy assessment of site characteristics will necessitate field work statewide.
Procedures:
A7. QUALITY OBJECTIVES AND CRITERIA FOR MEASUREMENT DATA
The project data quality objective is to provide data of known and documented quality and resolution on the vegetative cover of the state of Iowa. This data will be made available to researchers and management personnel involved in resource decision-making. The data quality indicators to be measured are described in section B5 of this plan.
Goal setting guidelines for precision and accuracy of vegetation cover mapping are described in the chapters on Vegetation Cover Mapping (Stoms, 1996) and Accuracy Assessment(Crist and Dietmer, 1997) in the National Gap Analysis Handbook. These goals were developed primarily by Gap Analysis projects in western states and call for minimum mapping units as large as 100 ha. In the midwest, most natural vegetation occurs in blocks that are much smaller than 100 ha. Land cover maps with 100 ha minimum mapping units would have little value in the highly fragmented landscapes of the midwest. Thus, the Iowa Vegetation Cover Mapping Project has adopted as goals a minimum mapping unit of 2 ha and an accuracy level of 80% for each vegetation class. The minimum resolution of thematic mapper imagery is limited to approximately 30 sq. m or 0.1 ha per pixel. Thus, 2 ha is composed of approximately 20 pixels.
Drake and Faber-Langendoen's (1997) report to the National Gap Analysis Program, An Alliance Level Classification of the Vegetation of the Midwestern United States, will be used as a guide for classifying dominant canopy vegetation at the alliance level. This report follows the National Vegetation Classification Standard being developed by the Federal Geographic Data Center. Vegetation will be mapped at the alliance level to the extent that alliances can be consistently separated by spectral reflectance values on the thematic mapper imagery. When this proves to be unfeasible, alliances will be clustered at the lowest physiognomic level possible in the hierarchical classification system.
Representativeness will be addressed by collecting field data from a representative sample of vegetation formation classes (forest, woodland, shrubland, herbaceous) from each county in the state as described in this document. This will be complemented with aerial photographs and other ancillary data that is available. (See section B)
Comparability will be addressed by collecting, analyzing, and reporting the data as described in section B of this document.
A completeness goal of 100% is needed for the project. Valid data is required for each land cover and vegetation class mapped in order to complete the cover map for the entire state.
A8. SPECIAL TRAINING REQUIREMENTS/CERTIFICATION
GIS technicians will be experienced or trained in using (ESRI) ARCINFO, ARCVIEW and (ERDAS) IMAGINE software. Field personnel will be experienced or trained in the identification and classification of plants and community composition. No special certification is required.
A9. DOCUMENTATION AND RECORDS
Digital files of land cover data for each county in Iowa will be produced in ARCINFO (vector) format and stored on CD-ROM disks. Multi-color hard copy maps of land cover can be produced at various geographic scales from these digital files. The Iowa Gap Analysis Project plans to produce hard copy land cover maps for all 99 Iowa counties. Other products will be produced as required by the National Gap Analysis Program, cooperators and other data users.
Metadata documentation will be developed according to guidelines set forth by the National Gap Analysis program and the U.S. Geological Survey, Biological Resources Division (Cogan and Edwards, 1996). Metadata will document data sources, processing techniques, accuracy assessment, and other pertinent information.
Appendix A represents the field data collection packet used for this project. Other records and documentation to be developed for this project include the following: digital files of spatial data, field data, and scanned photographs.
Records of field data forms, original aerial photos, and digital files used for classifying vegetation and accuracy assessment will be maintained and archived. The final retention and disposition of records will be determined by the Iowa State University Records Retention Schedule (rev. 1991).
B1. SAMPLE DATA ACQUISITION PROCESS DESIGN
The production of a land cover map is an iterative process based on data from satellite imagery, aerial photography, existing vegetation maps and field reconnaissance. Satellite imagery consisting of spring, summer, and fall dates for 12 scenes in 1991 and 1992 covering the entire state has been purchased. The availability of recent aerial photography is limited to a few counties and sites. Detailed vegetation maps are available for nine small state parks. Thus, ground reference data must be collected to train the computer software to recognize the spectral reflectance of various land cover categories represented in the Landsat TM imagery. Since ground reference data generally cannot be collected for the entire project area, representative samples will be used.
The National Vegetation Classification Standard will be used as the basis for assigning vegetation categories. The smallest land cover category used to represent existing vegetation for Iowa Gap is the alliance. A vegetation alliance is defined as a "physiognomically uniform group of plant associations sharing one or more dominant species, which as a rule are found in the uppermost strata of vegetation" (Drake and Faber-Langendoen 1997).
The Iowa GAP and DNR staff will coordinate a network of cooperating private, county, state, and federal field personnel who will acquire actual ground locations of typical vegetation alliances for use in mapping land cover of Iowa. These locations will be used to conduct supervised classifications of remote sensing data from satellite imagery.
Field data on land cover will be collected in two sets. The first set will be used to classify vegetation on the land cover map; another independent set will be collected for accuracy assessment. The first set will consist of all available and appropriate ancillary data including aerial photography, data from other surveys and research projects, and on-the-ground data collected specifically for Iowa GAP. Field data for the first set will be collected by Iowa GAP staff and a network of private, county, state and federal field personnel who have agreed to provide supplemental field data. This network includes biologists, foresters, and land managers with County Conservation Boards, the Iowa DNR, and federal agencies.
The second set of data for accuracy assessment will be collected with more stringent requirements for quality control and statistical design (See Section B5).
Field data will be collected according to a protocol that requires the completion of a Vegetation Rapid Assessment form (Appendix A). This form will serve as backup documentation of the type of vegetation surveyed by field personnel.The GIS lab manager, with guidance from the project's consulting plant ecologist, will review field data forms and the Vegetation Rapid Assessment form and assign appropriate classification prior to digitizing the data for GIS analysis. Descriptions of the vegetation that cannot be assigned a class corresponding to the the scheme used in labeling classes on the land cover map will be rejected.
Types and numbers of samples required: We will acquire representative ground locations for each land cover class and vegetation alliance labeled on the land cover map. The number of locations will depend on the extent of the land cover and vegetation class. A large number of samples will not be possible for vegetation alliances that are relatively rare. Other common alliances will require more samples, but some alliances, such as agriculture cropland and certain urban classes, will need fewer sample points due to the low spectral variability within these alliances.
Sampling Locations and frequencies: For the first set of field data Iowa GAP has a goal of 500 field sites statewide with a minimum of 10 sites for each vegetation alliance or unvegetated cover class. Data are being requested from all 99 counties to provide a representative sample.
B2. SAMPLE AND ANCILLARY DATA ACQUISITION METHODS
Phase 1 Acquisition:
Ancillary data will be used to classify the TM images into six information classes. The Iowa DNR is using existing aerial photos, topo maps and field data from the Natural Resource Conservation Service (NRCS) collected in 1990 and 1992 as sources to define general vegetation polygons. The geographic location of the polygons is known and is matched to the same location on the imagery.
Phase 2 Acquisition:
Vegetation alliances are identified in the field by an observer who is knowledgeable about plant identification and the National Vegetation Classification Standard. Observed alliances are recorded on data forms provided by the Iowa Gap Analysis Project (Appendix A). No specialized equipment is used to collect the sample data. Plant identification is standardized by use of a dichotomous key (Appendix A).
Ancillary data will be used to supplement the sample data gathered by the network of field personnel. These sources include tree species information on private lands gathered by State District Foresters; tree species information gathered for a research project in Northeast Iowa by Bill Norris of Iowa State University; vegetation cover for selected state parks gathered by the Iowa DNR during the state park ecosystem planning process; color infrared, black and white and color aerial photography of the same time period as the imagery; National Wetlands Inventory data and other sources that become available during the classification process.
All ancillary data that arrives in the GIS lab in non-digital form will be inspected for accuracy and appropriateness by the GIS coordinator and then digitized for use in the GIS classification process. Where applicable, aerial photos will be scanned on a flatbed scanner and the resultant images will be geo-rectified using a specialized software program. The rectified image will be viewed in a GIS application and the necessary data from the image will be traced. Attribute information will be attached to each traced polygon and saved as a file. Where data can be matched visually from a paper copy to a digital source it will be digitized directly on the screen without scanning.
B3. DATA HANDLING/CUSTODY
Field data forms provided by Iowa GAP are mailed back to the GIS Lab via business reply envelopes. All ancillary data sources are filed by county in the GIS Lab. When hardcopy data is digitized or otherwise entered into the computer, backups of the digital files to removable media will be made to ensure no loss of data due to machine failure.
B4. ANALYTICAL METHODS/REQUIREMENTS
Phase 1 Classification (coarse-grained analysis):
The Iowa DNR is using spring, summer, and fall dates of Landsat Thematic Mapper imagery to conduct a general land cover inventory of the state. Six of the twelve scenes covering the state had hyperspectral clusters prepared by the EROS Data Center for the Multi-Resolution Landscape Characteristics Consortium (MRLC). We found these to be of good quality and labeled the 240 spectral classes into six information categories (row crop cover, grass cover, tree cover, artificial cover, water and barren) using whatever ground information was available, including aerial photos, topo maps and data from the NRCS collected in 1990 and 1992. Ground truth sample polygons were initially divided into two randomly selected groups, one for image labeling and the other for classification accuracy testing. The labeling phase is in progress, with accuracy testing to come later. The other six original MRLC scenes had only one date of imagery and hence the included clustered data was of little value. Additional imagery from the USGS was made available to the DNR by the Iowa GAP Project. The DNR is conducting an unsupervised classification using the best dates for each remaining scene using EASI/PACE software from PCI. After running the unsupervised classification, the same ground truth information previously mentioned is used to label the spectral clusters into the six information categories. This phase is expected to take approximately one year and is expected to be completed by July, 1998.
Phase 2 Classification (fine-grained analysis):
Phase 2 classification will be done by the Iowa Gap Analysis Project at Iowa State University. The satellite image processing software, IMAGINE (ERDAS, Inc), will be used to classify images in Phase 2. Classification will be done using the geographic extents of one scene. The product of the Phase 1 classification will be used as input to the supervised classification process. One category, for example tree cover, will be selected as the focus of a classification operation. Appropriate ground samples and ancillary polygons containing alliance data, located and labeled by cooperating personnel, will be matched with corresponding areas on the original satellite images and the image polygons will be classified using on- screen interpretive techniques.
The process will be repeated for the herbaceous cover category using field samples of herbaceous vegetation and other ancillary data. The crop category will be further classified by the NRCS Natural Resource Inventory program using their own field-collected data and that information will be used for the agriculture alliances. The artificial and barren categories from Phase 1 will be overlaid with field sample and ancillary data to distinguish alliances, which may include urban and sparse vegetation alliances. National Wetlands Inventory data, combined with field sample and ancillary data, will be used to define wetland alliances. Standing and flowing water will be classified using ancillary data. A detailed account of data processing techniques will be documented according to the standards established in the "Metadata" chapter of the National GAP Handbook. NBII MetaMaker software will be used to record the metadata for this project. The software is based on and is in full compliance with the Content Standard for Digital Geospatial Metadata. The software is a metadata data entry program produced by the Environmental Management Technical Center (BRD, USGS). This phase is expected to take approximately one year and will be completed by March, 1999.
B5. QUALITY CONTROL METHODS/REQUIREMENTS
The data forms returned by field personnel containing located alliance polygons are reviewed prior to entry and alliances are verified for accuracy against the rapid assessment data submitted on the form. Each alliance polygon will be assigned a level of quality based on the known expertise of field personnel collecting the data. Level 1 are plant ecologists with graduate training in plant identification and community composition. Level 2 are natural resource managers with an intermediate level of knowledge in plant identification. Level 3 are other persons with minimal training and experience in plant identification. It is understood that the data collected is highly variable due to the variation in training of field personnel. The quality levels are a way for us to rank the input data and weigh the influence each polygon will have on the final vegetation map.
B6. ACCURACY ASSESSMENT
Assessing the accuracy of land cover mapping products is an elusive and challenging problem that calls for continuing research and development within GIS and remote sensing technology. Crist and Deitner (1997) provide a comprehensive overview of the "state-of-the-art" of accuracy assessment as it pertains to remote sensing and land cover mapping applications in the National Gap Analysis Program. They state that the criteria for accuracy assessment reflect the need to balance the requirements for rigor and defensibility with practical limitations of cost and time. The assessment methods must be scientifically sound, economically feasible, and nationally applicable. States should strive to use similar assessment methods and reporting, regardless of the land cover mapping procedure used, to support regional, multistate analyses. Iowa GAP will coordinate their land cover mapping with GAP projects in neighboring states to ensure uniformity, consistency and accuracy in edge-mapping.
The methods for accuracy assessment of Iowa land cover products will follow general guidelines established by Crist and Deitner (1997). The basic unit of the Iowa land cover mapping process is a polygon of 2 hectares or larger that represents a vegetative or other land cover class with a relatively homogenous composition. An accuracy assessment will be conducted by selecting a sample of locations (e.g., centroids of mapped polygons) from the final version of the land cover map and determining the true land cover classification at these locations. These data are frequently called the reference data set (Stehman and Czaplewski 1997). Properly executing an accuracy assessment involves knowing the nature of the created map, identifying the field methods for obtaining the reference data, designing a sound method for selecting reference data, actually collecting the data, conducting statistical analyses, and reporting the results (Crist and Dietner 1997).
Several states are using aerial videography to assess accuracy of land cover maps for their Gap Analysis projects. Aerial videography allows for checking a large number of locations at relatively low cost. It has proven to be most successful in states having extensive forest cover. We have used aerial videography in Iowa on other projects and encountered great difficulty in accurately delineating herbaceous cover classes, a major component of the land cover of Iowa and other states of the Great Plains. Thus, aerial videography will not be used in Iowa because of the limitations of resolution in landscapes with a high proportion of herbaceous vegetation (grasslands and wetlands).
Digital orthophoto quads are currently available for only a small number of counties in Iowa and thus have limited use for accuracy assessment. Current aerial photography at the level of resolution needed for accuracy assessment is also not available statewide.
This project has a goal of mapping land cover with 80% accuracy. We will attempt to measure thematic accuracy as a percentage of the land cover map classified correctly overall and by cover type with a standard error no greater than 8%.
Summary of steps and standards used in Accuracy Assessment:
Step 1: A final version of a land cover map will be produced as described in section B4. We anticipate having a total of 30-35 cover classes (including agriculture classes) that can be delineated on the satellite imagery. Because classification will be done in phases, one scene at a time, it will not be necessary to wait until the mapping is completed for the entire state to begin accuracy assessment. In fact, it will be desirable to test the methodology on a smaller area before applying the assessment statewide. Knowledge of the characteristics of the map to be assessed is important in determining the sampling frame (number, size, and classification of polygons). The methodology used to collect the reference data will match the classification system of the cover map.
Step 2: We plan to use field collected data as the primary source of reference data to assess the quality of the final cover map. Ground-truthing involves physically visiting the site in question to determine its true land cover type and will require substantial cooperator support and coordination. The plant ecologist responsible for developing the land cover classification scheme will be consulted to develop a field sampling plan that will guarantee consistency between reference data and the needs of the assessment project and future remapping, i. e., the method of collecting the field data will enable the land cover to be identified at the same level of detail as the land cover map.
We plan to follow the recommendation of Crist and Deitner (1997) that the design of the assessment study be stratified by, and only by, land cover types present in the final land cover map. The protocol for selecting field sampling sites will be based on the final number of land cover classes, the number of polygons within each class, and the number of samples needed to accomplish statistical precision.
With a minimum mapping unit of 2 hectares (the stated goal of the Iowa land cover mapping project), we anticipate that the occurrence of other unmapped cover types (inclusions) within a polygon will cause few problems in collecting field data. Nevertheless, the project's plant ecologist will be consulted to develop field protocols to ensure that each mapped cover type can be correctly identified in the field. The characteristics of land cover types that may affect these protocols are: polygon sizes (small, medium, large), polygon shapes (linear or non-linear), and heterogeneity of the plant communities (degree of patchiness and size of inclusion patches).
An individual measurement will result in a decision as to whether or not the field reference point agrees with the land cover map's label of that polygon. Accuracy is the statistical reduction of many samples into a statement of percent agreement.
Step 3: Sample unit identification has been a large topic of debate within remote sensing circles (Crist and Deitner, 1997).
Sampling units are defined here as all areas within the project area geographically contiguous and of homogenous primary attribute, that is, vector polygons or contiguous raster clusters of the same primary land cover type code (Crist and Deitner 1997). Land cover maps are based on algorithmic clustering of TM pixels with the resultant categories being spectrally similar. Therefore, pixels are probably not independent of each other. Although polygon boundaries are not precise, they are believed to represent real patterns on the the ground and the polygon is the defined feature that should be assessed. Therefore, the sampling unit is defined as a mapped polygon. The sample frame is the list of all polygons that comprise the final land cover map.
The sampling protocol for accuracy assessment will be designed to meet the statistical precision needed to accomplish the stated objectives for accuracy and standard error. Field sites will be selected through a stratified, two-stage probability sample. Accuracy assessment field data will be recorded on forms and returned to the GIS lab for analysis (see Appendix A). Probability sampling, as opposed to purposive selection of "representative" elements or haphazard selection of convenient elements, is now a standard scientific tool since it guards against selection biases and it leads to objective statistical inferences. Stratification will ensure good geographic spread of the sample across the state and will provide a representative sample of alliances.
Two stages of sampling will be employed. In the first stage, large tracts of land (e.g. counties, Landsat scenes, or some other convenient unit) will be selected in a stratified sample. In the second stage, sampling points within the large tracts will be selected. The reason for sampling in two stages, as opposed to sampling sites directly, is that direct sampling of sites would lead to a widely-scattered sample with high logistical costs.
Because cost of collecting field data could be limiting, consideration will be given to stratifying according to the relative cost or effort required to measure the sampling site, as was done in Utah and Arkansas by Gap Analysis personnel (Edwards, in press; Dzur et al. 1996). In these two cases, distance from the closest road was used as an indicator of cost of observation. The relative effort between types of polygons was adjusted to minimize cost for a given overall precision. Until a portion of the Iowa land cover map is completed, we cannot determine whether this level of stratification will be necessary.
Step 4. GIS methods will be used to select sampling units from the sampling frame which consists of all the polygons in a vector map.
Field surveys will use methods similar to those used to collect data for classification purposes (Appendix A). However, reference data will be collected by 2 or 3 well-trained field observers who have no knowledge of the primary attribute given by the land cover map for the sampling unit. This will involve providing each observer with coordinates and a map showing the polygon to be sampled but without the associated land cover type label. The field maps will typically have base information such as roads, streams, and locational grids such as UTM coordinates or Township/Range/Section lines. Field observers will also be provided ownership information so that permission can be obtained to enter private lands. It is the policy of the National Gap Analysis Program and USGS/BRD not to enter private lands without owner permission.
Observers will be trained and field tested in the typical techniques used for biological inventories. They will also be given training in the classification scheme employed in the land cover mapping process. They will be provided written guidelines, identification keys and other materials to assure that consistent, repeatable results are obtained (Appendix A).
The field data for each sampling unit will be assigned a pointer that identifies its location on the land cover map. Reference data will be compiled as a GIS coverage containing both the locations of samples and their attributes. Metadata will include a description of the method used by the analyst to determine agreement between the map and reference data and a measure of observer reliability in order to replicate the published analysis. Field forms will be archived and GIS data managed in accordance with procedures outlined in this document. If it becomes necessary to gather any data under a privacy agreement for nondistribution, the data will be treated the same as endangered species point locations from Natural Heritage Programs which can require a range of conditions including destruction of the data after its primary application, generalization, or password protection for limited use.
Step 5. Measurements from field sampling units will be compared with labeled polygons on the land cover map. As a first step in statistical analysis, agreements, or lack thereof, will be tabulated in a matrix whose rows represent mapped categories and columns represent observed cover types. The resulting error matrix is a contingency table which represents the probabilities of every possible correct or incorrect classification.
Statistical analyses of the measurements from the assessment sample need to recognize that the data arise from a complex sample. It is not valid to analyze these data as if they are independent and identically distributed, a standard assumption in many "textbook" statistical analyses. Analyzing data from a stratified two-stage sample as if they were independent and identically distributed will typically lead to confidence intervals which are unrealistically narrow and hypothesis tests which reject too easily. That is, the precision of the analysis is overstated (F. Jay Breidt, ISU Statistics Dept., pers. comm.). Proper methods for dealing with data from stratified two-stage samples will be employed in this study.
Iowa GAP does not intend to use "fuzzy set analysis" (Crist and Deitner 1997) to describe the degree of acceptability of misclassification. However, the final report will include a discussion of the nature of the possible sources of errors in classification and the range of reliability in accuracy assessment.
The report of results from the accuracy assessment study will follow the format of a scientific article and will include enough information for users to duplicate the assessment and derive their own accuracy measures based on their own specific purposes. The report will include the method of attributing observations, sample design, and sample and strata classes (number of polygons observed and total number of polygons in the land cover map). The results will provide users with an accuracy measure and standard error estimate for each land cover type and an error matrix at each level.
Limitations and Constraints:
In planning accuracy assessments, three general constraints (technology, logistics, and cost) must be considered because of the limitations they place on our ability to obtain ideal data sets (Crist and Deitner 1997).
Technological constraints: This category of constraints includes measurement errors relating to aquiring field observations. Error in determining the true location of the sampling unit in the field should not be a major problem in Iowa because the terrain is moderate and bisected by an elaborate system of roads and highways. Sampling units will be outlined in advance on topographic maps, county road maps, and aerial photos (if available) and provided to field observers.
The vegetation of Iowa is highly fragmented and nearly all sampling units will be small. Field observers will usually be able to survey entire sampling units, thereby reducing error caused by inadequate integration of all attributes of a unit.
Logistical constraints: Most sampling units in Iowa will be located within a mile or two of a road and can be visited without great expense. Few locations will be inaccessible due to dangerous terrain. Probably the primary constraint will be from landowners who deny access to private land. If sampling measurements cannot be made at a site, then these sites may need to be dropped from the sampling scheme and replaced with more accessible ones or reported as nonresponses in the assessment.
Financial constraints: This constraint is theoretically the least binding because more funding can be supplied if the political will decides it is necessary. However, we accept that increased costs of sampling a large number of sites to narrow the confidence interval has practical limits. We will attempt to conduct an accuracy assessment that is a reasonable balance between available funding and scientific soundness.
B7. INSTRUMENT/EQUIPMENT TESTING, INSPECTION AND MAINTENANCE
No instruments or equipment requiring testing, inspection or maintenance will be used for this project.
B8. INSTRUMENT CALIBRATION AND FREQUENCY
No instruments or equipment requiring calibration will be used for this project.
B9. INSPECTION/ACCEPTANCE REQUIREMENTS FOR SUPPLIES AND CONSUMABLES
Supplies used in the GIS Lab will be inspected upon receipt by the GIS Lab manager for visible signs of damage. All data will be backed up on removable storage media so that failure of primary storage media will not result in data loss. Supplies will be purchased from reputable vendors to ensure quality.
B10. DATA ACQUISITION REQUIREMENTS
Landsat Thematic Mapper (TM) digital satellite imagery will be the primary data source for constructing base maps of vegetative distribution. Ancillary information will be drawn from other imagery where applicable.
Full state coverage (12 scenes) of Landsat TM imagery from 1991/1992, primarily from the month of May, has been obtained from the Multi-Resolution Land Characteristics Consortium (MRLC) and from the Iowa Department of Natural Resources. It is desirable to base land cover mapping upon multi-season data (spring, summer, fall) since multiple seasons allow the use of information about vegetation phenology in the classification process. The Iowa Gap Project has purchased additional imagery dates to complete the multi-season coverage for the state. Further information concerning the choice of Landsat data for this project can be found in the "Imagery" chapter of the National GAP Handbook.
The Landsat TM data are received in digital format on a CD. The images have been pre-processed to correct missing information and other problems inherent in satellite-gathered imagery. The images obtained from the MRLC were also geo-referenced to real-world coordinates and clustered into 240 spectrally distinct classes prior to our receipt.
B11. DATA MANAGEMENT
Landsat TM imagery arrives in the GIS lab on CD. It is stored on CD until processing time when it is copied to the hard drive of a DEC Alpha workstation. Data forms with field information arrive via the US mail and are stored in raw form in the lab. Data from the forms are digitized and stored on the hard drive of a computer in the lab. Backup copies of all digital data are made to removable media. All data forms are checked prior to digitizing for accuracy and then after digitizing to assure correspondence to the original form. All necessary data from ancillary sources are digitized or copied to the hard drive of a computer in the GIS lab and then backup copies are made of the digital data. Where ancillary data have been digitized, the GIS lab manager checks that the original data correspond correctly to the digitized data.
The data processing equipment are computers. A combination of IBM-clone personal computers using the Windows 95 operating system and DEC Alpha workstations using the UNIX operating system will be used to process the data. An effort was made to purchase machines with the most memory, largest hard drives and fastest processing speeds that were available at the time. Additional hard drive space and random access memory will be purchased as project needs require. A suite of software will be used to process the data. All software packages are industry standard and represent the best application available for each processing function.
C1. ASSESSMENTS AND RESPONSE ACTIONS
Co-Principal Investigators will conduct monthly in-house audits of data quality and staff performance to assure that work is being performed according to EPA and National Gap Analysis standards. Monthly audits will be documented in a written laboratory journal and initialed by one or both of the co-PIs. If audits show that the work is not being performed according to standards immediate corrective action will be implemented and documented in the laboratory journal. A list of vegetation alliances following the National Vegetation Classification Standard will be prepared for the state of Iowa and submitted for peer review to a panel of plant ecologists.
The EPA QA manager (or designee) may conduct an audit of the field activities for this project as requested by the EPA project manager according to R7 ENSV SOP 2152.2A, "Conducting On-site Reviews of Field Sampling Activities." The QA manager may also select this project for an audit during a Management System Review of their program. The EPA project manager will have the responsibility for initiating and implementing response actions associated with findings identified during the on-site audit. Once the response actions have been implemented, the EPA QA manager (or designee) may perform a follow-up audit to verify and document that the response actions were implemented effectively.
C2. REPORTS TO MANAGEMENT
The Iowa GAP staff meets on a regular weekly or bi-monthly schedule depending on what is sufficient to provide adequate guidance and structure for the staff to execute the project tasks. These meetings will be organized using an agenda structure to track the activities throughout the term of the project.
Reports will be submitted to the EPA Project Manager at regular semi-annual meetings of the Mid America Remote Sensing Consortium.
Preliminary versions of land cover maps will be made available for inspection by the EPA Project Manager as they become available. Classifications of one or more scenes from Path 25 should be available for review by mid-April 1998. Subsequent scenes from Path 26, 27 and 28 will follow at approximately 4 month intervals. Accuracy assessment data for Path 25 will begin in Summer 1998 but most of the second set of field data will be collected in the summer of 1999.
Once the project is complete and the resulting data obtained, the co-principal investigators will submit a final report to the EPA project manager. The final report will consist of a GIS land cover map of Iowa, metadata, and a report of accuracy assessment activities as outlined in section B. A summary of the activities performed during the project and the resulting data (along with any statements about problems concerning data quality). The report will be forwarded to the EPA project supervisor.
D. Data Validation and Usability
D1. DATA REVIEW, VALIDATION AND VERIFICATION REQUIREMENTS
In summary, the Iowa Vegetation Cover Mapping Projec will use spring, summer, and fall scenes of Landsat Thematic Mapper imagery to conduct a general land cover inventory of the state. Ancillary data consisting of field surveys, available photography and existing vegetation maps will be used to classify vegetation and label distinct spectrally clustered polygons on the imagery. Vegetation classification will follow the National Vegetation Classification Standard for the Midwestern United States (Drake and Faber-Langendoen 1997). The project will follow methods and quality control standards outlined in this QAPP and in the National Gap Analysis Handbook. The project has a goal of achieving 80 percent accuracy in the overall classification of vegetation. The coverage will include the entire state of Iowa with a minimum mapping unit of two hectares. An independent set of ground reconnaissance data will be obtained to conduce the accuracy assessment analysis. Because of inherent technological, logistical, and financial constraints (Section B6), it is possible that the accuracy goal may not be achieved for all vegetation classes. However, accuracy assessment will be essential for validating the final vegetation cover map and providing the user with a measure of reliability.
D2. VALIDATION AND VERIFICATION METHODS
The chain of custody for data throughout the project will be:
Issues will be resolved through mutual consultation between the Co-Principal Investigators and the GIS specialists.
The final versions of the land cover maps and the accuracy assessment report will be peer reviewed prior to its release to agency cooperators and the public. Prior to release, the Principal Investigators have responsibility for reviewing all data and verifying that final products achieved QAPP-defined goals for accuracy, completeness and acceptance criteria.
The final version of the land cover map will be conveyed to users as digital GIS files in ARC/INFO format on CD-ROM disks. Hard copy maps will be provided free to cooperators. The general public may be charged a fee to cover the cost of printing.
D3. RECONCILIATION WITH USER REQUIREMENTS
Once the final version of the Iowa Land Cover Map is produced, the EPA Project Manager will review the product and the accuracy assessment report to determine if they fall within the acceptance limits as defined in this QAPP. Completeness will also be evaluated to determine if the completeness goal for this project has been met. If data quality indicators do not meet the project's requirements as outlined in this QAPP the data may be returned for revisions.
D4. LITERATURE CITED
Cogan, C. and T. C. Edwards. 1996. Metadata Standards for Gap Analysis. National Gap Analysis Handbook. USGS/BRD. Idaho Cooperative Fish and Wildlife Research Unit, University of Idaho, Moscow.
Crist, P. and R. Deitner. 1997. Assessing Land Cover Map Accuracy. National Gap Analysis Handbook. USGS/BRD, Iowa Cooperative Fish & Wildlife Research Unit, University of Idaho, Moscow.
Drake, J. and D. Faber-Langendoen. 1997. An alliance level classification of the vegetation of the midwestern United States. A report to the University of Idaho, Cooperative Fish and Wildlife Reserch Unit and the National Gap Analysis Program from The Nature Conservancy, Midwest Conservation Science Department, Minneapolis, MN.
Dzur, R. S., M. E. Garner, K. G. Smith, W. F. Limp, D. G. Catanzaro, and R. L. Thompson. 1996. Cooperative accuracy assessment strategies for sampling a natural land cover map of Arkansas. In: Spatial Accuracy Assessment in Natural Resources and Environmental Sciences: Second International Symposium, edited by H. Todd Mowrer, Raymond L. Czaplewski, and R. H. Hamre. May 21-23, 1996, Fort Collins, CO, General Technical Report RM-GTR-277.
Edwards, T. C., G. G. Moisen, and D. R. Cutler. [1998] Assessing map accuracy in a remotely-sensed, ecoregion-scale cover-map. Remote Sensing of Environment: In press.
Stehman, S. V. and R. L. Czaplewski. 1997. Basic structures of a statistically rigorous thematic accuracy assessment. In: Proceedings of the Annual Meetings of the American Society of Photogrammetry and Remote Sensing, Seattle, WA, April 6-10. American Society of Photogrammetry and Remote Sensing, Bethesda, MD, Vol. 3, pp. 543-553.
Stoms, D.M. 1996. Actual Vegetation Layer. National Gap Analysis Handbook. USGS/BRD, Idaho Cooperative Fish & Wildlife Research Unit, University of Idaho, Moscow.