Semi-automatic structural mapping in arid terrain from remotely sensed images

Synopsis

Remotely sensed digital imagery of many arid terrains contains abundant structural geological information. This information is typically photointerpreted from topographic shading clues about bedding plane attitudes, rock competence based on local relief and fault offsets. Geological structural mapping can be performed from pairs of images using stereophotometric techniques. However, this is still relatively expensive and difficult to attempt from spaceborne sensor data. Systems such as JPL’s IGIS system assume the prior availability of digital elevation models for their operation. We report here the development of an approach to structural mapping of geological information using single images with no ancillary DEM. Using techniques more commonly used in computer vision we perform the following tasks: extraction of the image subscene that contains most of the structural information using textural segmentation, mapping topographic “edges” that correspond to lithological boundaries, searching for valley/bedding plane “V” criteria and mapping dip directions there, quantifying bedding plane orientations using photoclinometry We have tested our ideas on an image of LANDSAT Thematic Mapper data from a part of the alpine Atlas mountains of south central Tunisia.

The mapping of geology from aerial photographs at regional and large scales is long-established¹. In particular, the interpretation of geological structure is enhanced by the ability to view the surface in 3D using stereoscopic properties of aerial photographs and extract simple geometrical relationships such as the dip and strike of a bedding plane from three points on the plane. This concept has been developed by the assistance of computers such that mathematical structural models can be created and used interactively with photogrammetric equipment for deriving quantitative measurements from air photo pairs². The advent of digital remotely sensed images should have increased the ability of geologists to extract structural information using computer techniques. However, progress has been slow mainly because the photointerpretation techniques of the human are difficult to simulate by computer and the ancillary requirement for digital topographic data (DEM) is an obstacle to ease of implementation.

Stereo pairs of images from the SPOT satellite can be used to create DEMs³ but the need for two images and intensive computer processing is inhibitory. DEMs can be created from photogrammetric data or from topographic maps but there are still large areas of the world for which the quality or availability of DEMs or the data to produce them is poor or lacking. There have been a number of attempts to utilise co-registered remotely sensed data and a DEM for the purpose of extracting structural information or recognising geomorphological elements of the scene. The IGIS system⁴ developed at JPL uses image processing and computer graphics techniques to derive strike/dip measurements, cross section construction and stratigraphic section measurements from co-registered Thematic Mapper and DEM data. The user of this system must map lithological boundaries using a trackball cursor, there is no automatic element to this basic photointerpretation task. It is only when this is done that a relationship between the geometrical information of the DEM can be related to the geology of the scene. Chorowicz et al.⁵ used DEM data in a more fundamental way to identify geomorphological features. Firstly, slope classes are identified for each pixel and then pattern recognition rules are applied to strings of these classes to derive geomorphological features such as strike-ridges. Interaction with remotely sensed spectral data is by superimposition after this classification process using the DEM.

The work we report attempts a similar task to that of the IGIS system but without the use of an ancillary DEM. We wish to extract useful quantitative information on geological structure from a single digital remotely sensed image using shading information from the structurally-controlled topography. We use the term semi-automatic for our approach to indicate that it is a combination of human interactive photointerpretational skills and automatic computer vision/image processing operations. We do not deny the value of the combined DEM/remotely sensed approach which because of its greater information content must be able to achieve superior results. Rather we ask the question — what can be achieved using the remotely sensed data alone? Given the current relative difficulty in obtaining co-registered remotely sensed and DEM data for exploration areas, is this approach a useful, practical alternative? We do not provide complete answers to these questions here, work at the time of writing is continuing and a full discussion of these ideas lies in the future.