CHAPTER 8

A DIGITAL PHOTOGRAMMETRIC METHOD TO MEASURE HORIZONTAL SURFICIAL MOVEMENTS ON THE SLUMGULLION LANDSLIDE, HINSDALE COUNTY, COLORADO

by Philip S. Powers and Marta Chiarle

Introduction

The traditional approach to making aerial photographic measurements uses analog or analytic photogrammetric equipment. This method usually yields accurate results (accurate to 0.5 m at 1:12,000 scale) but requires expensive instruments, takes considerable time, and requires a trained operator to set the models and acquire the results. We have developed an alternative digital method for making measurements from aerial photographs by using Geographic Information Systems (GIS) software and mostly DOS-based PC computers. This method was applied to determining the movements of visually identifiable objects, such as trees and large rocks, on the surface of the active portion of the Slumgullion landslide. The technique is based on the concept that a direct visual comparison can be made between images derived from two sets of aerial photographs taken at different times. Horizontal displacement vectors on the active slide have been measured directly from ortho-images that were produced by digital photogrammetric techniques using 1985 (1:12,000 scale) and 1990 (1:6,000 scale) aerial photographs.

For the concept to work, there must be enough movement during the period between times of the aerial photography to make measurements with accceptable accuracy. Excessive movement between aerial photographs can also be a problem, as it may be difficult to relate the two images directly. In the 5-year period between our aerial photographs, the fastest moving portion of the landslide should have moved about 25 to 30 m (Crandell and Varnes, 1961). We estimated that over most of the active slide, 15 to 20 m of displacement should occur in a 5-year period. A 2-m measurement error (~10 percent) for a direct comparison method was our minimum standard.

Methodology

Before accurate measurements can be made between two sets of aerial photographs, tilt and distortion must be removed. In our study, this could only be done by processing the photographs and transforming the images to relief-corrected ortho-images, which became the most difficult part of the work.

The process of producing tilt-free, differentially rectified, digital ortho-photos (ortho-images) from aerial photographs taken in mountainous terrain requires excellent ground control information and sophisticated software. Desktop Mapping System (DMS) with Softcopy Photo Mapper (DMS/SPM) 3.1(c), by R-Wel, Inc. 1992, was used at the GIS laboratory, USGS National Mapping Division-Rocky Mountain Mapping Center, to produce the ortho-images. DMS/SPM runs on a DOS-based 386/486 PC computer and uses the same rigorous photogrammetric techniques as those used in standard photogrammetry.

Part of the ground control necessary to produce an ortho-image with DMS/SPM is an accurate digital elevation model (DEM) of the area. Photogrammetric work had been completed by 1992 on both sets of aerial photographs, and accurate digital elevation models had been created from the digital contour data (Powers and others, 1992).

The Process of Making an Ortho-Image

Photographic color transparencies (diapositives) of the landslide aerial photographs were scanned at 400 dpi (dots-per-inch) with a Howtek scanner attached to a Macintosh computer running Adobe Photoshop. An image produced from a scanned diapositive contains more visual detail than a photographic print. The complete color diapositive was scanned, including the fiducal marks, and then converted to 256 shades of gray. Each image required approximately 14 MB of disk space. The scanned images were then transferred over a network to the DOS-based PC that runs DMS software.

Using established survey ground control coordinates (Varnes and others, 1993), we then established the relationship between the image and the survey coordinate system. The survey ground control coordinates were entered in a file, where the x, y, z values for each control point were identified by a unique identification number. The identification numbers were then related to the displayed image by clicking on their screen location with the digitizing device (mouse). This process was done for each image of a stereo pair: the two images do not need the same ground control points.

The orientation was then computed for each image, and the effect each ground control point had on the solution was analyzed. It was at this stage that a ground control point could be removed, or additional points could be added later if desired. At least five points for each image are required by the DMS/SPM program to compute the orientation.

Working with the stereo model as an anaglyph display of red and blue images on the computer screen, and using red-and-blue-lens spectacles, we measured the coordinates for survey and photogrammetric control points. This was done by positioning the floating mark (dot) on the ground surface at the point to be measured and recording the x, y, z coordinates. The coordinates were then checked against the ground control points for accuracy.

 The coverage for the ortho-photo is then selected from either the left or right scanned aerial photos. The cubic convolution method, the DEM, and the exact photogrammetric solution derived earlier with the DMS/SPM program are then used to generate the differentially rectified ortho-image. Note that this DEM must cover the entire area that is to be ortho-rectified.

The ortho-image, which was still referenced to our local coordinate system, was then checked against the ground control and photogrammetric points. The image was then converted to a Tagged Image File Format (TIFF) image, transferred to a different PC, imported into a raster editing program for processing, and then imported to CorelDraw for direct comparison with a second image.

Making the Direct Comparison

We were able to directly compare two ortho-photos by registering each to ground control and mapping the junction of trees and their shadows in the open areas of the active slide. This was done by converting a 1985 gray-scale image to a monochrome image with a passive white (transparent) and a black color, and mapping from the 1985 monochrome image to the 1990 gray-scale image through the transparent zones of the 1985 image (fig. 1). This process was continued until the stereo pairs for both 1985 and 1990 (fig. 2) were combined in one 32.5-MB file and all the displacement vectors representing movements of points on the landslide were drawn (fig. 3).

Processing the Displacement Data

The first step in processing the displacement data was to separate the vectors from the remainder of the CorelDraw file. Because the vectors were in units used in the plotting program, ARCINFO and a spreadsheet program were used to convert the vectors to proper units of magnitude and direction (meters and degrees). The processed spreadsheet file was then used as an input into Golden Software's SURFER program, and the magnitudes were contoured using a minimum-curvature algorithm. This contour file was then imported into CorelDraw and combined with the vectors to produce a final plot.

Accuracy

The photogrammetric work done by Smith (1993) provided numerous ground control check points to verify the accuracy of our results. Smith's photogrammetric work was reported to have a standard deviation of 0.44 m in horizontal position.

Accuracies were measured at three different stages. The first stage was during viewing of the DMS stereo model. The next series of accuracies were recorded while viewing the geocoded ortho-image. The last and final set of accuracies was recorded in CorelDraw prior to drawing the displacement vectors. Table 1 is a summary of the simple averaged differences.

Table 1.-- Summary of the accuracies, where x is east-west, y is north-south, and z is vertical.

	Average errors in x (meters)	Average errors in y (meters)	Average errors in xy (meters)	Average errors in z (meters)
Stereo models 1990	0.9	1.4	1.7	3.3
Stereo models 1985	1.3	2.0	2.4	5.2
Ortho-images 1990	1.2	1.1	1.6
Ortho-images 1985	1.6	1.3	2.1
CorelDraw 1990	1.1	1.1	1.6
CorelDraw 1985	1.7	1.0	2.0

The average errors of the ortho-images and the CorelDraw images are almost the same. This suggests that there is no loss of accuracy during the conversion. The stereo model errors are slightly larger, probably due to the difficulty of viewing portions of the image in three dimensions: good three-dimensional perspective is necessary to make accurate measurements. The slightly higher average errors in the table for the 1985 data are caused by the difference in scale between the original aerial photographs (1:12,000 for 1985, and 1:6,000 for 1990).

Conclusions

Direct visual comparisons can be made between ortho-images produced from two sequences of aerial photographs taken at different scales. Horizontal motion vectors can be measured accurately between ortho-images by direct comparison. This approach provides a method to measure motion over more of the slide surface in a shorter period of time than the standard photogrammetric approach.

More than 800 motion vectors have been drawn on the active slide surface, and a horizontal vector displacement map with displacement contours has been produced showing the 5-year movement between 1985 and 1990 (fig. 3). The map is in AutoCad DXF format and is available for additional studies.

We learned from this study that distribution of ground control is most important and must be taken into account prior to the taking of any aerial photographs. Ground control must be evenly distributed along both sides of the area of interest, with several points within the area of interest. We found that the accuracy of the ortho-images greatly improved after adding additional control points. These points, from the photogrammetric work of Smith (1993), were added along the sides and within the slide (fig. 2).

The newly developed digital technique proved to be a viable tool for the analysis of the Slumgullion aerial photographs, especially when comparing two different sets of aerial photographs. It has been applied to the measurement of surficial movements on an active slide, but could be useful in the study of aerial photographs of any kind of feature that changes position with time. The method is not a definitive one: the choice among topographic survey, analog (or analytic) photogrammetry, and digital photogrammetry must take in account the economic, human, and time resources available and the requirements of accuracy.

Acknowledgments

The authors wish to thank Robert Desawal and Michael Crane of the USGS National Mapping Division-Rocky Mountain Mapping Center.

References Cited

Crandell, D.R., and Varnes, D.J., 1961, Movement of the Slumgullion earthflow near Lake City, Colorado, in Geological Survey Research, 1961: U.S. Geological Survey Professional Paper 424-B, art. 57, p. 136-139.

Powers, P.S., Varnes, D.J., and Savage, W.Z., 1992, Digital elevation models for Slumgullion landslide, Hinsdale County, Colorado based on 1985 and 1990 aerial photography: U.S. Geological Survey Open-File Report 92-535, 5p.

Smith, W.K., 1993, Photogrammetric determination of movement on the Slumgullion slide, Hinsdale County, Colorado 1985-1990: U.S. Geological Survey Open-File Report 93-597, 17p.

Varnes, D.J., Smith, W.K., Savage, W.Z., and Varnes, K.L., 1993, Control and deformation surveys at the Slumgullion slide, Hinsdale County, Colorado--a progress report: U.S. Geological Survey Open-File Report 93-577, 15p., 1pl.