Digital Orthophotos and DEMs Combat Closed Landfills
DEM extraction and digital orthophotos are combined for routine monitoring and preventative maintenance of closed landfill sites.
By Chris Stohr, Robert Darmody, Susanne Aref and Philip Cheng

Because regulations, operating methods and types of refuse have evolved over time, a landfill is typically not a single unit structure. A landfill is more likely to be a complex group of distinctly constructed and differently covered cells, sometimes constructed on top of each other.
      Settlement of a landfill cover caused by decomposition and consolidation of wastes, can form depressions in the cover. These depressions can collect and pond rain and runoff that can infiltrate into the wastes. If water mixes with the wastes, the wastes decompose to produce methane (CH4), CO2 and reduced volume of refuse. The production and release of methane into the atmosphere and of leachate into groundwater is detrimental to the surrounding ecosystem (Figure 1). Regulatory agencies require filling of depressions on landfills to avoid ponding of water.
      Prevailing practices for evaluating effectiveness of old landfills in isolating wastes include: monitoring wells, point measurements, and field reconnaissance. All have their advantages and disadvantages. Table 1 shows how traditional field methods can cause errors in finding sources of infiltration and depressions. An automated method is needed to find areas where water is ponded on a landfill cover before infiltration generates leachates from the landfill to pollute groundwater and air.
      Early recognition of potential problems could bring about important cost savings and reduction in environmental degradation because repairs could be made before the water infiltration problem requires extensive (and expensive) remedial measures. Since a closed landfill no longer generates income, the potential economies of a remote sensing method of postclosure monitoring of cover performance are attractive even if it has some limitations and uncertainty.
      Figures 2 and 3 show a landfill covered with 1m of compacted fill in 1988. A field transit survey using a Sokkia Set 3 with third-order methods was made on two adjoining (left and right) sites in 1993. Aerial photography was flown over the site in May 1995 at a scale of 1:3000 using a 6-inch lens to extract a DEM.
      A DEM is a three dimensional array of positions and elevations which can be used to derive drainage and topographic maps, and rectify and correct relief displacements of scanned aerial photos to create orthophotography. This automatic extraction of a DEM used to be very time consuming, expensive, and sometimes inaccurate. With the improvement in speed and affordability of computers, the lower cost and higher resolution of scanning aerial photos, and the development in DEM extraction algorithm, this task can be achieved efficiently and economically.

DEM Extraction Software and Algorithm
Film negatives of two aerial photographs were scanned at a resolution of 600 dpi on a Crosfield drum scanner in order to preserve geometry. In order to extract the DEM from the stereo pairs, an airphoto orthorectification and DEM extraction software package developed at PCI was used. The algorithm was based on the photogrammetric method of space resection by collinearity. Studies have shown the space resection by collinearity is the best method to orthorectify aerial photos. The package includes ground control collection, mathematical modeling, resampling, and manual and automatic DEM extraction. One advantage of the software package is that it can be executed on most workstations and personal computers, and hence provides more choices of computer platforms for the users. To create orthophotos, the user can either provide the DEM of the area to be orthorectified or extract the DEM from the overlap area of stereo aerial photos.
      To extract a DEM, the first step is to calculate the mathematical model from the ground controls for each of the stereo photos. The next step is to resample the right image into an epipolar image. This ensures that the left and the epipolar images are offset only in the horizontal direction. This is important for the next step, the extraction of the DEM.
      A neighborhood matching method is used to match pixels in the left image and the epipolar image, using statistics calculated in defined windows. Matching is performed by considering the neighborhood surrounding a given pixel in the left image (i.e., a template) and moving the template within a search area in the epipolar image until a position is found which gives the best match. The difference in location between the center of the template neighborhood and the best match neighborhood is the disparity. This value is input to the mathematical models to compute the elevation at the center of the template. The advantage of the epipolar projection is that the pixel search is effectively limited to the horizontal direction, greatly improving the algorithms efficiency and accuracy. A suite of DEM editing tools including interpolation, filtering, and smoothing functions completes the process.
      A 1:3000, 80 percent overlap stereopair of the landfill site was used for DEM extraction. The photos were scanned at 600 dots per inch. The ground resolution of the scanned aerial photo is about 0.14 m per pixel. Fifteen ground control points (GCPs) and 10 independent check points (ICPs) were collected using a Leica System 200 GPS system. The horizontal and vertical accuracy of the GPS unit are both within 2 cm. ICPs are points which were collected inside the area bounded by the ground controls and were not used in calculating the mathematical model. Table 2 shows a summary of the RMS residuals of the GCPs and RMS errors of the ICPs of the left and right image using the photogrammetric method. We can see that the errors are within 5 pixels (0.14 m per pixel). This is mainly due to the errors in identifying corresponding positions on the uncorrected photo which were not well defined on the image. To demonstrate the capability of the photogrammetric method, Table 2 also shows the results of the second order polynomial mathematical model. The RMSE GCP residuals can be up to 30 pixels and the RMS ICP errors are up to 54 pixels.
      The DEM was extracted at half the resolution of the original image, i.e., 0.28 m . The computer time required was approximately three hours on a SUN SPARC 20 workstation and four hours on a Pentium 90MHz running Windows 3.1. Over 97 percent of the overlap area was extracted successfully. Figure 3 shows the extracted DEM overlain with field-surveyed 1-m contours and outlines of depressions digitized by DGPS. Two points from the GCPs could not be compared because the points were outside the overlap area. A summary of the elevation extraction results comparing the input GCPs and ICPs, showed that the comparison of the RMS residual to the input GCPs is 0.28 m and the RMS error compared to the input ICPs is 0.17 m .

Comparison of Depression Outlines
When results were compared with field surveyed contours generated from ARC/INFO TIN model, visual comparison and statistics showed a very high degree of agreement. The 15- and 30-cm depths of five depressions were correctly located and computed by the DEM. All results fall within a contour interval of 0.15 m. Figure 4 shows a comparison of the extracted DEM with contours generated from ARC/INFO TINCONTOUR. Local relief over the landfill study area is slightly more than 15 meters (222 to 237 m). Minimum relief for four depressions is 0.15, minimum relief for the fifth depression (center of the image) is 0.3 m.
      Depressions which ponded water only 15-cm deep were correctly located and computed by the DEM indicating that a high degree of confidence could be extended to areas of no or low vegetation where there were no field transit surveys. Statistical comparison of the DEM showed a very high degree of agreement with selected contours from the field surveys.
      Surface drainage of depressions were digitized in the field using a resource-grade, Trimble Pro XL processed to obtain DGPS outlines of five depressions. The digitized outlines compared so well with the field transit and DEM that it is difficult to tell that the sources are different.
      The outlines of depressions using field-surveyed topography, field digitizing using DGPS and DEM extraction from a stereopair show that all of the methods are useful for delineation. DEM extraction and digital orthophotography would likely be a more rapid, unbiased method lending itself to routine monitoring and preventative maintenance for reducing infiltration through earthen landfill covers.
      Automatic DEM extraction using stereo photos is not limited to closed landfill sites. It can be applied to other applications such as monitoring the usage and elevation of open landfill sites in order to determine fill and void volumes and to provide topographic data needed for stability analysis.

ACKNOWLEDGMENTS
The authors would like to thank P. Jahn, R. Rice, and the Illinois State Geological Survey, and M. Sievers, Illinois State Water Survey who helped collect and process GPS data. Dave Randall, Technical Support Services, PCI Inc., contributed to this article. Scanning of the negatives was performed by Scantech Inc., Champaign, Ill. Mention of trade names is for information and not endorsement.

About the Authors:
Chris Stohr and Robert Darmody are graduate student and professor respectively, in the Department of Natural Resources and Environmental Sciences, University of Illinois at Urbana-Champaign. Susanne Aref is a professor of Crop Sciences, University of Illinois at Urbana-Champaign. Philip Cheng is a software engineer with PCI Inc.

Back