From 1998 to 2001, the Natural Resource Analysis Center at West Virginia University has converted nearly 3000 historical and current paper underground and surface coal mine maps to georeferenced images and digitized coverages. Clients have included private, state and federal agencies. Concrete successes and failures have been encountered. Lessons learned from both are discussed in detail along with several innovative methods of georeferencing.
Keywords: CAD (Computer Aided Drafting), DRG’s (Digital Raster Graphics), DOQQ's (Digital Ortho Quarter Quads)
The Natural Resource Analysis Center at West Virginia University has done small digitizing projects for various agencies and organizations over the last decade but our current digitizing project has caused us to temporarily increase manpower from a low of 5 or 6 individuals to close to 40 full time employees. A contract from the West Virginia Division of Environmental Protection (DEP) provided our organization the opportunity to produce many scanned and registered digital images used to create GIS coverages. DEP maps consisted of permit area boundaries and associated drainage, valley fill, mineral removal, water and pollution monitoring, augering, deep mining, haulroad, and core hole drilling features. We captured adjacent permit areas from maps to identify permits that were not represented by the maps we received. Some of the permit maps are CAD drawings while others are drawn on top of enlarged copies of United States Geological Survey (USGS) topographic maps. Since We used USGS topographic digital raster graphics (DRG’s) almost exclusively as background data for image registration those maps were easiest to register. Other maps done with CAD were often different enough from the DRGs to make image registration difficult. Elevation contours, streams and roads can be very different. The engineering companies’ representations are not necessarily any better or worse than the drgs but the differences may make image registration extremely difficult. Of the maps that are drawn using CAD, many have landmarks that are useful in image registration while others do not.
The digitizing part of the project was very straightforward. Once images were registered digitizing was done on-screen in ArcEdit with the image as a background. Attributing was done with an AML form created with Formedit. When designing an AML for attributing we recommend the use of pull-down menus and other controls in place of textboxes thereby eliminating potential sources of error and allowing better standardization of data.
When faced with the task of registering thousands of images, any innovation or method that saves time and energy is welcomed. We created two procedures for image registration that were helpful under the right circumstances. All of our West Virginia data is in UTM zone 17 NAD27 so state plane coordinates were useful to us only after conversion to UTM meters. Grid systems representing state plane coordinates were common on DEP mining permit maps. We were able to make use of the grids using the ARC/INFO “project” command.
Registrations based on the State Plane Coordinate System (SPCS) grid were successful when the grid on the map was actually based on SPCS coordinates (which occurred most of the time). When the grid was based on some arbitrary or proprietary coordinate system, the registrations, without exception, did not work well. Registrations based on a single point worked well in some instances when measurements were well done, scale measurements were accurate and the location of the point used was accurate. This was done using coordinates from one known point and extrapolating 3 more points using the scale of the map and the direction of the north arrow. For example if the scale is 1”=500 feet, you can draw a box that is 4 inches on each side that contains the known point in one of it’s corners. It is preferable to draw the box carefully before the map is scanned so the points used to register the image can actually be seen. This method was discouraged except in special cases because it can result in unacceptable image registrations.
A test of overall registration accuracy made by comparison with background data (the DRG’s), decided whether or not we could proceed with the digitizing process. Before we instituted this double-checking process, we delivered a small number of mis-registered images (approximately 30-50 out of ~ 1200). Quality of output improved substantially after we started checking all registrations against the DRG’s regardless of the methods used to register the images. We also checked drainage and valley fill features against Digital Ortho Quarter Quads (DOQQ’s) to be sure they matched the topography from the photos. Our low point in image registration was hit when we obtained location information for points found on maps and accepted that coordinate information without checking accuracy against the background DRGs. This policy resulted in the majority of the 30-50 permits that all had to have images re-registered and coverages transformed.
We developed a Visual Basic application that writes “Arc Macro Language” programs (AML’s) that can be used to create a grid or lattice coverage that can be used for image registration (or other purposes). The coverage is built to node topology. The user selects dimensions for the grid (# of iterations in the x and y dimensions) and a starting coordinate (lower left). The user has the option of creating the grid in the input coordinate system or projecting to another coordinate system by selecting an existing projection (*.prj) file located on the users computer. If the projection routine is to be included in the AML, the original coordinates are preserved in two fields (spx, and spy) before the ArcInfo “ADDXY” command is re-issued.
This grid is very useful when registering maps that have a SPCS grid on them because the coordinates on the map can be linked to the coordinates shown by the spx and spy fields of the coverage. This method yielded a very low root mean square error (rms) when compared with our other registrations.
We are confident that our experiences with this project will help us to manage other large-scale GIS projects in the future. We hope this description of our positive and negative experiences will help others make sound decisions for their projects. If anyone would like to use the “generate file utility” that we created to assist in georeferencing images, please contact the senior author directly by e-mail at jchurchi@wvu.edu.
We wish to thank Jackie Strager for her comments on an earlier draft of this manuscript.