Investigation of landslide susceptible terrain using landsat TM Imagery

Investigation of landslide susceptible terrain using landsat TM Imagery - Preliminary results

Scott L. Huang
Associate Professor

Robert C. Speck
Associate Professor

Been K. Chen
Graduate Student

Department of Mining and Geological Engineering
University of Alaska Fairbanks, Fairbanks,
Alaska 99775-1190, U.S.A.

Abstract
Several approaches of satellite image processing in predicting landslide susceptible terrain in subarctic are given. Integration of Geo-data with Landsat TM images provides a better result than that solely based on satellite images. From this project, it has been noted that logical manipulation of different image and geo-data files reveals landslide sites more effectively than the other processing techniques such as texture analysis, signal processing, arithmetic processing, and classification.

Introduction
Mining has played an important and traditional role in the development of the economy of the state of Alaska. Vast land in this northern region is known to contain abundant mineral resources. Extraction of minerals and fossil fuel from the delicate arctic and subarctic environment is often a controversial issue. Many important technical difficulties remain to be resolved. One of those which adversely impact mining operations is slope failure within a mine and its surroundings.

Slope failure often complicate, if not dictate, development of a surface mine. Specific information on the prevention and control of the damage is often sought by government agencies and private sectors. An accurate assessment of potential landslide problems requires a pre-mining inventory of terrain physiography, including geology, soil characteristics, and identification of potential slope failure areas. Such an inventory or data base can then permit a comprehensive evaluation of mine design and planning prior to the development.

Surface mine planning and design are very complex and demanding tasks. Even minor errors in decision - making can result in major monetary expenditures. Detailed prior knowledge of a mine site can permit more flexible and accurate designs to minimize the financial risk and unnecessary engineering difficulties associated with errors in planning. Extraction of near-surface mineral resources by surface mining is a standard method in response to the need for the commodity. Landsat imagery ad aerial photographs have been used by government agencies and mining companies for investigation and inventory of potential hazards near prospective and active mine areas (Corses and Usibelli, 1989: Anderson et al., 1976).

Aerospace remote sensing and image interpretation are powerful tools for preliminary assessment of the effects of a natural hazards on resource development. Remote sensing can play an important role in investigations of slope movement, especially in the evaluation of landslide susceptibility. Although most landslides are not directly identifiable on satellite images, the remotely sensed information, as noted by Gagnon (1975), provides regional physiography, definition of local geologic structure, and most landforms as well as land-use practices which relate to the potential of landslides.

Many of the remote sensing and image interpretation techniques can be utilized to assist in the design of a surface mine. This paper discusses the preliminary results from an on-going research effort which is investigating an inventorying pertinent parameters related to landslide susceptible terrain in the in the vicinity of surface coal mien in Healy, Alaska using satellite remote sensing techniques.

Site description
The study site, as shown in Fig.1, includes the poker Flats mining area and the entire Lignite Cree Basin. Lignite creek Flows westward into the Nenana River near the town of Healy, Alaska. The study area leis along the northern flank of the Alaska Range.

The climate in the study area indicates an average annual high of 34° F and a low of 19° F on the basis of the last 10 years of weather records. Total annual precipitation in the area ranges from 15 to 18 inches with most arriving during late summer rain storms. Spring thaw in late May also contributes much runoff.

Much of the natural ground in the Lignite Creek Basin is unstable, either presently or has been in the past. The Coal-bearing formation is prone to slope movement. Where slope spare undercut by streams, failure often result even on gentle slopes. Other slope failures are associated with solifluction and thermal degradation of underlying permafrost.

Fig.1 Image showing the Lignite Creek Basin.

Approach
Pre-mining aerial photographs of usibelli coal mine reveal substantial evidence of landslides both inside and outside of the current mine site (Corser and Usibelli, 1989). Many of the slope failure prone areas in the study site cannot be easily identified by field survey due commonly accessibility and economic reasons.

In this study, a Landsat-5 thematic Mapper (TM) image, acquired on September 22, 1984, (Scene ID Y5020520430X0) was utilized as a primary source for gathering the needed information. Landsat-5 TM sensors provide a coverage of 185 by 185km with a ground resolution of approximately 30 m in reflective spectrum. Six images in reflective bands (0,45 to 2.35mm) and one image thermal band (10.4 to 12.5 mm) were used to investigate the stud area for surface cover characteristics, such as vegetation, soils, water bodies, drainage patterns, geologic conditions, and permafrost distribution. Information related to terrain elevation, permafrost distribution, existing landslide deposits and site geology was collected and digitized. The location of existing slope failures were also digitized to serve as a reference for cross-examination of the validity of the landslide susceptibility algorithm which will be developed in the second phase of the study.

Landsat image processing was mainframe of the study, which included tasks of image rectification, enhancement and information extraction.

Image processing and digitization
The Landsat TM image was processed by a Comtal/3M image processor, with a VAX 11/750 as the host computer. A Land Analysis System (LAS) software package was utilized for image processing and Analytical Mapping System (AMS) was used for geo-data digitization.

Thirteen ground control points in the study area were selected from a topographic map to create a reference image and the TM-5 image was used as the search image. Pixel size of the search image was reformatted to 25x25m, and pixel values were resampled using the nearest neighbor (NN) method.

The same thirteen ground control points for geometric registration were also used in AMS registration to create both the reference file (Latitude and Longitude unit) and the search file (UTM unit). Landslide deposits from the US geological survey geologic map of the Healy D-4 quadrangle 22 polygons. In addition, landslides surveyed by Wilbur and Beget (1988) were digitized and converted into a file containing eight geological units. A digital elevation model of the site was obtained from topographic map.

Fig.2 Digital geological map showing coal-bearing unit.

In order to extract information required for creation of an algorithm which predicts landslide susceptible terrain, the seven TM images were manipulated. Among the processing routines used, spatial analyses and signal processing were found to be ineffective in predicting the occurrence of landslide deposits. The was due to the fact that the brightness values of the existing 22 landslides deposits varied widely in the images.

Principal component analysis of the six TM images (Excluding TM-6) was performed. The first two transformed images contained total data variations of 91.13% and 7.88%, respectively. Using these principal component images unsupervised classification (KMEANS) was later carried out. Through the statistical classification, 8 classes of landcover types were identified. Unsupervised classification was also performed on the original six TM images, and 20 classes were created. As compared to the digitized landslide deposits, it was noted that no single landcover class from either approach could contain all landslides areas, unless several classes were regrouped into larger a class. The main reason for this less accurate result was probably due to the reason as that for spatial and signal processes.

Table 1: Statistics of logical algorithms

Equations	Area predicted By equation	Area overlapped With Existing Deposits
1	10.10%	0.61%
2	13.40%	0.81%
3	10.15%	0.77%

Several logical processes performed byte-wise logical operation among the images. For TM images (TM-1, -3, -4 and -5) were combined using and "OR" operator with the ranges of brightness values as shown in equation 1. The logically processed image was later combined by an operator "AND" with the geological map file containing coal-bearing formation to further reduce the area covered by, the algorithm.

{ [ ( 54 < TM -1 < 56) OR (26 < TM - J < 28) OR (34 < 5M - 4 < 48) OR (20 < RM - 5 < 32 ) ]
AND (Geological unit = coal - bearing formation)}---------------------------(1)

Preliminary analysis indicated that equation 1 estimated 10.10% of the entire Lignite Creek Basin was landslide-prone. As compared to the published landslides deposits map, which includes 2.06% of the basin areas, the areas estimated by Equation 1 was relatively large. Further evaluation of accuracy of the algorithm was undertaken by overlapping the landslide-prone area predicted by equation 1 with the digitized landslide deposit polygons. The results indicated 0.61% of the basin area fell into this overlapping zone.

Fig.3 Landslide susceptible areas predicted by Equation 1.

A similar approach was used for the other two logical processing of images. Figure 4 and 5 and Table 1 summarize the outputs from equations 2 and 3 shown below:

{ (1.9 < TM-1 / TM-3 < 2.2) OR [ ( 9 < TM - 7 < 11) AND (157 < PC - 2 < 159 ) ]
AND (geological unit = coal - bearing formation)} ---------------------------(2)

and

{ [ ( % slope = 0 ) AND ( 650 m < contour interval < 850 m)
AND ( Slope aspect = horizontal ) } OR ( 54 < TM -1 < 56)
AND ( 2.6 < ( TM-1/TM-2 ) <2.9)
AND (geological unit = coal-bearing formation)}---------------------------(3)

It is desirable to have to model that not only will predict all the existing landslides deposits with 100% accuracy, but also will not cover any larger area than necessary. For example, a model can estimate the entire drainage basin is landslide prone which, of course, will include all the landslide deposit areas. Therefore, the decision rule used here in judging a good model is that the area predicted by the model should be as close to the areas outlined by the existing landslide deposits as possible (i.e. 2.06% of the entire Lignite Creek Basin) and the areas predicted by the model should also coincide as much with the existing sliding areas as possible.

Preliminary analysis of this study indicates that Equation 3, which involved TM images and geo-data manipulation, gave a better result (although not as good as expected) than models involving solely TM images. From this experiment, geo-data such as drainage basin characteristics, permafrost distribution, groundwater accurrances, etc. will be gradually incorporated into the development of model during the next phase of study.

Fig.4 Landslide susceptible areas predicted by Equation 2.

Fig.5 Landslide susceptible areas predicted by Equation 3.

Summary
Several approaches of satellite image processing in the applications of predicting landslides susceptible terrains have been tried. With a mixture of Landsat TM images and geo-data, it is possible to develop a model which predicts the landslide areas with an acceptable accuracy. These are, however, difficulties which must be overcome before the model can be put in use. These difficulties require creation of a set of geo-data in the study area. Since the logical process proves a higher accuracy, the experiment described here has helped authors to focus their efforts on logical integration of geodata with images in the development of a model predicting landslide susceptible terrain.

Acknowledgement
The authors wish to thank Ms. Nancy Mighells for the typing of this paper and to the U.S. Bureau of Mines, Generic Ground Control Center for the financial support. This research would not be possible without the assistance of Mr. Harold Garbeil of ADVAL Laboratory at the University of Alaska a Fairbanks.

References

Anderson, A.T., Schultz, D. and M. Nock, 1976, Satellite data for subsurface-mine inventory, report X-923-76-199, Goddard Spce Flight Venter, Green Belt, Maryland, 13 p.
Corser, P. and M. Usibelli, 1989, Operational and Geotechnical constraints to coal mining in Alaskan's Int Mining Engineering, January 1989, pp.21-23.
Gagon, H., 1975, Remote Sensing of landslide hazards on quick clays of eastern Canada, Proc. 10th Int'l Symp. Of Remote Sensing of Environment, pp. 803-810, October G-10, 1975, ERI?M, Ann Arbor, Michigan.
Wilbur, S.C. and J.E. Beget, 1988, Landslide motion in discontinuous permafrost, Proc. 5th Int'l conf. on Perma.