As a component of the Revolutionary War and War of 1812 Historic Preservation Study, funded by Congress in 2000, the Cultural Resources Geographic Information Services (CRGIS) program, National Park Service (NPS) is using remotely sensed data to assess the integrity of, and threats to, surveyed battlefields. Phase one is the assessment of present integrity of the battlefield through the overlay of global positioning systems (GPS) data over classified Landsat Thematic Mapper (TM) imagery to calculate the land cover within the battlefield boundaries. In phase two, land cover adjacent to the battlefield will be identified. From this information we can then identify those sites that need further study due to possible encroachment by incompatible land use. The use of remotely sensed data to accomplish what was once a manpower and time-consuming component of historic resource study will streamline the process and allow the use of resources in a more targeted manner.
The Revolutionary War and War of 1812 Historic Preservation Act of 1996 tasked the American Battlefield Protection Program (ABPP) of the National Park Service with identifying historic sites associated with the two wars and assessing their condition, identifying threats, and making recommendations for preservation. These sites were separated into battles and military operations and associated historic properties. The Act was passed as an unfunded mandate and was not funded until the year 2000.
ABPP convened an advisory group, made up of historians and preservationists, to identify the sites and evaluate the significance of each site according to its contribution to the prosecution of the war. To accomplish the assessments required, grants were given to battlefield surveyors to research a site and perform an assessment and GPS survey of the location. The surveyors were made up mainly of historians, archeologists, preservationists and other cultural resource personnel.
CRGIS provided training, in four regional field schools, to the surveyors. Covered during the field schools were research and survey methodologies, the use of global positioning system (GPS) units, entry and manipulation of spatial data into a geographic information system (GIS) , and entry of tabular data into a digital survey form. The battlefield surveyors then researched the event, which included gathering historic descriptions of the landscape at the time of the event. Descriptions of the landscape allowed the surveyor to identify defining features on the ground to place the event spatially. Once the site was located, a field survey was completed. An important component of the survey is an estimation of land use/land cover (LU/LC) both within the boundaries of the site and in the immediate area.
Assessment of condition, identification of threats, and to a large extent, recommendations for preservation are dependent on the accurate identification of land use/land cover of the battlefield. Some of the landscapes are easy to read and lend themselves to quantification of LU/LC. Other locations are hard to read due to vegetation cover, access restrictions, or the size or discontinuity of the battlefield. In addition to often being inaccurate the in situ determination of LU/LC is time consuming and manpower intensive. In addition surveyor accuracy assessment would be too time consuming and costly because of the number of sites.
CRGIS decided to do a pilot project to determine the feasibility of using remotely sensed data and GIS to assess the integrity of, and threats to, surveyed battlefields.
Using Remote Sensing To Assess The Battlefield Landscape
Feasibility Considerations and Data Decisions
Judging the feasibility of the use of remotely sensed imagery would involve not simply the process itself, but also the possibility that the cultural resource community could apply the process; many of whom are non-spatially oriented people. Besides technical ease of use, two of the main considerations were the affordability and availability of the data. The data needed to be affordable and available while still maintaining a resolution that would allow for accurate classification of the LU/LC of a relatively small area. In addition, the entire project needed to be affordable not only in data acquisition but also in terms of manpower and computing power.
The LU/LC classification would be applied to landscapes that were local, many times measured in acres, as opposed to regional. This requirement excluded data such as AVHRR, which, while free in most cases, has too coarse a resolution. While individual battlefields themselves cover a relatively small local extent, the Revolutionary War and war of 1812 Historic Preservation Study is national in scope, covering 29 states. It was decided that Landsat Thematic Mapper (TM) data, with a resolution of 30-meters, would be appropriate for this project. In excess of 110 TM scenes would be needed to cover the study area. CRGIS has access to TM scenes from the Multi-resolution Land Characteristics (MRLC) consortium administered by the US Geological Survey (USGS). The data would be provided processed to level 1A, radiometric and geometric correction, and projected to Albers Equal Area.
battlefield ‘habitat' has been identified; it is composed
of those land uses that are compatible with the preservation
of the battlefield integrity. Those land uses are low
to no development, low impact agriculture, and, not surprisingly,
the preservation of the natural landscape. The LU/LC information
that the surveyors collect in the field is very broad,
including classes such as forest, open space (non-agriculture),
commercial, urban, etc., and does not have a 1:1 correlation
to classes easily identified in TM imagery. For the study,
generalization of LU/LC was acceptable, and, in fact,
desirable since it would simplify accuracy assessment.
The decision was made to modify the Anderson
classes (Anderson, et al 1976), using a combination
of Level 1 and Level 2, and the National Land Cover Dataset
(NLCD) Land Cover Class Definitions (ANON. 2001). CRGIS
felt the following classes best reflected the LU/LC information
being collected by the study surveyors: 1 Urban or Built-up,
2 Herbaceous Planted/Cultivated, 3 Forest and Orchard,
4 Barren, and 5 Water.
Due to the large number of scenes to be classified and
the extent of our study area we decided to use the Mapping
Zones developed for NLCD 2000 by the USGS (Homer
and Gallant, 2001) . These zones represent areas that
have similar biophysical and spectral characteristics.
The use of homogeneous zones minimizes the spectral differences
found in scene mosaics caused by disparate land cover
and topography. The zones were created at a 30-meter scale
for use with TM imagery. ERDAS
Imagine 8.4 ™ software was used for the image processing
Signature data for classes was collected for each zone. This was accomplished by using software to inspect the scenes using different band combinations, and performing a region grow to define clumps of like pixels or by describing an area of interest using a draw tool, and labeling the signature using ‘local knowledge'.
Signatures were also collected in the field using GPS. A GPS data dictionary was developed listing land use/land cover types and points or polygons were recorded to define the location. Multiple signatures were collected for each LU/LC to account for differing spectral values for the same class. We are finding that around 50 signatures per scene give us the spectral coverage to accurately classify the image. This translates into 50 classes, which are then combined into the five target classes.
A supervised classification was then run using the signature file that was developed. Because it was found that, in the pilot scenes, the histograms for the bands of data had a normal distribution a maximum likelihood parametric decision rule was used. ERDAS considers this the most accurate of the classifiers because of the number of variable used. During the classification process a distance file is generated, this is a one-band raster image that graphically illustrates the ‘distance' a pixel is from the mean of the pixel's class.
After classification, the image is evaluated by using the distance file to generate a threshold histogram to identify pixels that are likely to be misclassified. Additional signatures may be collected and further classification completed if needed. The image is then re-coded to reduce the number of redundant classes and an accuracy assessment is completed. Ground truth and other ancillary data is used in the accuracy assessment process, with 20-30 reference samples collected per class.
A goal has been set of 80% accuracy, which has not been reached at this point, but CRGIS is confident of achieving this goal as we gain more experience in image processing. In addition, the 80% accuracy must be evaluated in the context of the accuracy that surveyors at ground level are capable of achieving when dealing with multiple LU/LC types that may be spread over a wide area. The areas of greatest error are those that contain urban or built-up land. Unfortunately these are the areas, as cultural resource professionals, in which we are the most interested in looking. Specifically, within the scope of this study, we need to accurately identify areas that are being developed to quantify the integrity of the battlefield and any threats to the landscape.
Analysis of Battlefield Integrity
The study surveyors, using their research and fieldwork, generated ESRI ArcView shapefiles for three boundaries for each battlefield. The Study Area encompasses the entire area in which the event occurred, including troop movements, encampments, and staging areas as well as the location of the actual battle. The Core Area is the actual location of the fighting, often referred to as ‘bloody ground', and will wholly fall within the Study Area. Finally, the Potential National Register (PotNR) boundary is that area within the Study Area that retains integrity. We used these boundaries to subset the classified imagery and calculate the percent of each LU/LC type. The sums were charted and compared with the land use totals the surveyors calculated in the field. In most cases, after ground truth, we have found that the surveyor estimated totals were inaccurate when compared to the imagery derived calculations. The condition of the battlefield was evaluated on the basis of percentage of land use that is compatible with battlefield preservation. This battlefield ‘habitat' has been the subject of much discussion and we still do not have a clear quantitative definition of what constitutes a good, fair, or poor condition. This is a preservation argument and the data now exists to plug into any of the three definitions. In addition, the integrity is assessed using the same sub-set of imagery and calculation of LU/LC percentages. Integrity differs from condition in that it not only lends to preservation but it also take into account the amount of the present LU/LC that reflects the use at the time of the event and allows the event to be interpreted in its context.
Analysis of Threats
Threats are land uses that degrade the integrity of the battlefield landscape. Evaluation of threats takes into consideration present and historic land use and local zoning, or lack of zoning, on and in the vicinity of the battlefield. The classified imagery provides the present LU/LC information and can, along with other sources, provide information on historic uses. The types of LU/LC are reviewed in the context of local, state and regional zoning and planning when making threat statements.
The assessment of condition is used to evaluate the threats to the battlefield inside the boundaries. Designation of special status, such as state or national park, may make any external land use a non-issue as a threat. However, with many locations, the external, outside the study area boundary, land uses present threats to the landscape. The Study area boundary is buffered, again taking into consideration local zoning, and used to subset the classified image. That subset is re-coded to reflect ‘compatible' and ‘incompatible' LU/LC. A temporal study to establish trends in usage would allow a better understanding of the processes at work in the area. Use of ancillary data, such as census data, would be useful in the threat assessment. A threshold of ‘incompatible' land use would lead to a designation of threatened, this designation would inform preservationists in any land acquisition.
Analysis of Acquisition Priorities
The classified subset of land use within the battlefield boundary can be re-coded to reflect land use ‘compatible' and ‘incompatible' with battlefield preservation. The ‘compatible' class would be given a code of 1 and the ‘incompatible' class is given a code of 0, the ‘compatible, LU/LC would then be clumped to show contiguity. The ‘compatible' clumps are then weighted according to size and what percentage lies within different parts of the battlefield. Those clumps that lie completely or partially within the Core Area, ‘bloody ground', would receive a greater weight then those outside. Planners could then take this data into consideration when making decisions on what pieces of land to target for acquisition and preservation. Land acquisition and preservation decisions for battlefields are often made absent spatial data. The classification of LU/LC within a battlefield boundary and the designation of ‘compatible' and ‘incompatible' uses give preservationists and planners additional data when making land acquisitions.
The on-ground estimation of land use/land cover by surveyors is time consuming, manpower intensive, and prone to errors from numerous sources. It is hoped that the use of remotely sensed data to accomplish this component of historic resource study will streamline the process, generate more accurate data, and allow the use of resources in a more targeted manner
Report prepared by James Stein, CRGIS Ninth Biennial Forest Service Remote Sensing Applications Conference