The Use of Satellite Images in Geological Mapping, Red Sea Hills, NE Sudan

K. A. Elsayed Zeinelabdein,
master, lecturer,
Khartoum, Sudan

Introduction

Sudan is a vast country occupying (about) 2.5 million km². In the last three decades considerable efforts were made to properly map the main geological units in this country and to better understand their tectonic settings. Despite this, large areas remain inadequately mapped and little is known about their tectonic evolution let alone their potential mineral resources.

The enormous areal extent of the Red Sea Hills area together with its harsh environment made it almost inaccessible for the conventional field based type of surveys. Delineation of favourable zones for economic mineral accumulations is time consuming and at very high cost. Therefore, remote sensing techniques qualify themselves as an alternative or at least a major tool in mapping and prospecting for natural resources.

The study area is located in the central Red Sea Hills of NE Sudan. It is bound (approximately) by latitudes 19˚ 30΄ - 20˚ 00΄ N and longitudes 36˚ 30 – 37˚ 17΄ E.

In NE Sudan three major physiographic units, from east to west, can be recognized, namely the Red Sea coastal plain, the Red Sea Hills and the high plateau area, which grades gently westwards to the Nile valley.

The drainage system of the Red Sea Hills is generally structurally controlled, where most of the seasonal water courses follow fault lines and structural joints, but their tributaries form sub-dendritic patterns.

The Red Sea area of NE Sudan is part of the Arabian Nubian Shield that comprises a series of volcano-sedimentary sequences, which were intruded by igneous intrusions and metamorphosed mainly to greenschist facies. According to Kröner et al. (1987)[6], this shield evolved due to the collision of amalgamated island-arc systems with the Nile Craton during Middle to Late Proterozoic.

1. Methodology

The main materials utilized in this work were satellite data (Landsat images MSS & TM), geological maps, topographic maps, rock specimens, computer software and polarized optical microscope.

In this study, the incorporation of visual interpretation, digital image processing and ground truthing was adopted to obtain meaningful geological map. This included the preparation of base map. To this end, different color composite images were produced and the best of which was chosen as base map concurrently with the topographical maps. Also locations of different training samples, representing the various lithological units, were determined and accurately plotted in the base map.

During the field work, a total of 23 training samples were collected. These were later described and prepared for further microscopic investigations.

Also a GPS was used in navigation to determine ground control points (GCPs) which were used to establish a coordinate system to the images. Post-field laboratory work included the petrographic investigations for the collected samples and the digital image processing.

2. Results and Discussion

Visual interpretation of MSS and TM images portraying the study area was performed using the spectral and spatial characteristics of various lithological and structural units for the discrimination. The outcome of this visual interpretation was a preliminary lithological map of the study area displayed in Figure (1).

Fig. (1): Lithological map of the area based on visual interpretation of Landsat MSS and TM images.

In quest of saving computer space and processing time, one subscene was selected (Arbaat subscene) for all image-processing steps. The possible forms of digital image manipulation are literally infinite (Lillesand and Kiefer, 1994) [7]. However, the following forms were used in the current study:

3.1. Geometric Correction

The intent of geometric correction is to compensate for the distortions introduced during the imaging process so that the corrected image will have the geometric integrity of a map (Lillesand and Kiefer, 1994) [7]. These distortions were corrected by analyzing well-oriented ground control points (GCPs) occurring in an image. In the geo-referencing process, a number of 40 GCPs were located both in terms of their image coordinates (column, row numbers) on the distorted image and in terms of their ground coordinate. Then the distorted image was re-sampled.

3.2. Haze Correction

The atmospheric conditions largely affect the solar radiation reflected from the earth surface. This effect modifies the radiation, reduces image contrast and contributes an additive term, correction for this term is necessary in order to detect “only” the surface reflectance.

One of the image-based simple, but yet effective method for haze correction has been given by Crane (1971) [2]. The method, known as black pixel subtraction, is based on the assumption that somewhere in the image there is a pixel with zero illumination or reflectance, such that the radiometric contribution from this pixel represented only the additive term. It is further assumed that the atmospheric scattering is uniform throughout the scene. According to this approach, the histogram of each band is examined and shifted towards lower gray values by a constant that is supposed to represent the additive term of haze.

3.3. Contrast Stretching

Data in a single image rarely extend over the entire range of the recording and displaying device. Hence, the objective of contrast stretching is to expand the narrow range of brightness values typically present in an input image over a wider range of gray values.

Two types of contrast stretching techniques were used namely: linear stretching and histogram equalization, where feature of low contrast became distinguishable in the stretched image. It is pertinent to mention that the stretched images are never interpreted alone it must be combined with image processing procedures such as the preparation of enhanced colour composite images and the calculation of ratio images.

3.4. Spatial Enhancements

In contrast to spectral image enhancements, spatial filters emphasize or de-emphasize image data of various spatial frequencies. The applications of filters make visible linear structures, joints and faults not otherwise discernible on the un-enhanced image (Alexander et al., 1973) [1]. Spatial filtering is a local operation wherein pixel values in an original image are modified in the basis of gray levels of neighboring pixels.

Both directional and non-directional high pass filters were performed. The selected directions were W, NW and NE based on the knowledge of the tectonic history of the area. Three directionally filtered images were then selected to facilitate the production of a colour composite image shown in Plate (1).

Subtle enhancements on the structural lineaments in Arbaat subscene enabled the author to produce a structural map displayed in Figure (2).

Plate (1): Spatially enhanced image of Arbaat sub-scene obtained by combining bands 7,4,1 in RGB, respectively using high pass filters.

Fig. (2): Structural map of the study area derived from satellite TM data through spatial enhancement techniques.

3.5. Colour Composite:

From the standpoint that color photographs and images have certain advantages, stemming from the human eye’s better chromatic discrimination (Drury, 1993) [3], the colour composite technique was utilized to produce enhanced coloured images for better rock discrimination.

Many colour composite images were created using various band triplets, based on the statistical method of “Optimum Index Factor” (OIF) which has been calculated for the six reflective TM bands (see Table 1).

The OIF of a three-band combination is calculated as:

Where: SDi is the standard deviation for band i and [Cci] is the absolute value of the correlation coefficient between any two of the three bands.

Table (1)

Optimum Index Factor of the six reflective TM bands

No	Band-triplet	OIF
1	7, 5, 1	39.97
2	5, 3, 1	37.57
3	7, 5, 3	37.34
4	5, 4, 1	36.42
5	7, 5, 4	35.46
6	7, 5, 2	34.48
7	5, 3, 2	33.56
8	7, 4, 1	32.68

The high values for OIF indicate bands that contain much information with little duplication (ITC, 1997) [4]. Although the band-triplet 7,5,1 has the highest OIF, the resulting image from this combination was not the best, this may be due to the high correlation between band 7 and band 5. Despite having lower OIF, the band triplet 7,4,1 was found to be the best one to produce enhanced colour composite (see Plate 2).

Plate (2): Colour composite image obtained for Arbaat sub-scene using Landsat TM bands 7,4,1 in RGB, respectively.

On this image, the post-tectonic granite appears in yellow colour, while syn- to late-tectonic granite is dark green making good lithological discrimination and the sharp contact is clear. The metasediments are violet in colour and easily discernible from other rock units. Lineaments, the other characteristic features for these rocks, were observed on this image. The metavolcanics are in light violet

Furthermore, this method was used as a complementary step after certain digital processing procedures (e.g. band ratioing, principal component analysis, spatial filtering …etc.), where the production of colour composite images have provided more improvements over the processed images themselves by adding the colour advantages to the images.

3.6. Principal Component Analysis:

Principal component transformation is a commonly used image processing technique in which a new coordinate system is calculated from the multispectral data set (Sabins, 1987) [8]. It is usually applied to a correlated set of data to produce another un-correlated multispectral data set that has certain ordered variance properties (Singh and Harrison, 1985) [9]. Also it can be used to compress the information content of the multispectral bands into just two or three principal component images (Jensen, 1996) [5]. This process yields new images that are more interpretable than the original data.

To perform principal component analysis, fundamental statistics of the original data were computed (see Table 2). Using these statistical terms, the new axes of principal component analysis were computed. This procedure has resulted in the production of six principal component images.

Table (2)

Eigen vectors of TM principal components

Eigen vector	Band 1	Band 2	Band 3	Band 4	Band 5	Band 7
PC1	0.285	0.217	0.363	0.313	0.674	0.434
PC2	0.649	0.328	0.382	0.148	-0.452	-0.314
PC3	0.573	0.021	-0.395	-0.663	0.109	0.254
PC4	0.127	-0.031	-0.178	0.003	0.564	-0.796
PC5	0.358	-0.426	-0.516	0.623	-0.105	0.118
PC6	0.159	-0.814	0.518	-0.205	0.033	-0.035

As expected, the first principal component with 95.73% variance (see Table 3) and positive loadings from the entire TM bands (see Table 2), contains lithological, topographic, linear features and drainage patterns information.

Table (3)

Eigen values and variance percentage of TM principal components

	PC 1	PC 2	PC 3	PC 4	PC 5	PC 6
Eigen alue	5131.38	189.4	26.59	7.44	3.98	1.26
Variance %	95.73	3.53	0.5	0.14	0.07	0.02
Sum	95.73	99.26	99.76	99.90	99.97	99.99

The second principal component with 3.53% variance (see Table 3) is tonally dark and displays some lithological and topographic information. The third and fourth principal components with 0.5% and 0.14% variance, respectively (see Table 3), display fair topographic information but no lithological information was encountered. While the fifth and sixth principal components contain only noise and can not be used for geological interpretation.

The first three principal component images (PC1, PC2 and PC3) were combined to create a colour composite by assigning the first principal component to Red, the second one to Green and the third to Blue filter (see Plate 3).

Plate (3): Principal component analysis composite image (PC1, PC2, PC3 in RGB, respectively) obtained using the six reflective Landsat TM bands.

Although, there is some reduction in the colour variations in this image, some spectral enhancements were achieved, which permit most of the lithological units to be clearly mapped. Example of which the boundary between the metasediments and the granite. On the other hand, the younger granite, which is clear in any of the standard Landsat images, can not be easily discriminated here due to this reduction in the colour variations.

It is pertinent to mention that this principal component composite image enabled a clear and distinct discrimination between the metavolcanosedimentary sequence and the granitic rocks in the one hand and between the metavolcanics and the metasediments on the other hand.

Fig. (3): Integrated geological map of the study area based on visual interpretation of Landsat MSS, TM, various image processing techniques, published and unpublished geological information.

Conclusions

Identification of lithological units, structural features and delineation of potential target zones for natural resources (i.e. water and mineral resources) can easily be portrayed through digital image processing techniques over large areas. That is simply because in the digital domain the investigator is able to emphasize or de-emphasize the bands spectrally and spatially in quest of producing new images using different band combinations.

The use of conventional hard copy remote sensing data usually provides the geologist with individual scenes that may enable him to differentiate some rock types in the area under investigation, but many others will remain undetected and would be classified mistakenly under other rock units. However, applying digital image processing techniques paved the way to overcome such problems by producing several different images from the same data set, which facilitated the task to discriminate more efficiently the previously indiscriminative units. The performed digital image processing techniques within the context of this study has demonstrated the viability of the above statement.

The importance of the spatial enhancement stems from the ability to enhance edges whereby linear features, drainage system and geological boundaries are emphasized.

Colour composite technique serves to increase the contrast between the various land cover types by adding the colour advantage to the image. However, the use of contrast stretch with this technique renders more interpretable images. Nevertheless, not all the units are differentiated. So far, it is clear that there is no single colour composite image capable to discriminate between all rock units in the area. Thus for a given area, the more the number of images produced, the most reliable results are obtained.

The obtained images through Principal Component Analysis have improved the efficiency of discriminating between some lithological units, which could not have been otherwise differentiated. This is clearly illustrated by the image in Plate (3), wherein the volcanosedimentary sequences have been vividly separated from the metasediments and the granitic rocks.

The obtained geological map may be considered a new contribution to the geology of the Red Sea Hills since it contains new information about the lithological units in the area under study and some corrections to the previous geological maps. In addition, the map was produced in a relatively short time utilizing the advantages of digital image processing techniques and GIS.

Digital image processing techniques should be a prerequisite precursor to any geological investigations directed to geological mapping and/or prospecting for natural resources.

References

1. ALEXANDER, F. H. and BILLINGSLEY, G. F. C. (1973): Digital image enhancement techniques used in some ERTS application problems. Contribution to geology, vol. 12, No. 2, pp. 7 – 21.

2. CRANE, R. B. (1971): Processing techniques to reduce atmospheric and sensor variability in multispectral scanner data. –Proceed. 7^th int. Symp. Remote Sens. Environ., 2, 1345-1355, Ann. Arbor, MI.

3. DRURY, S. A. (1993): Image interpretation in geology. 2nd ed., -271 pp., (Chapman and Hall), London.

4. ITC (1997): Ilwis 2.2 application guide. ITC, Enschede.

5. JENSEN, J. R. (1996): Introductory digital image processing: A remote sensing perspective. 2^nd ed., -316 pp., (Prentice Hall), New York.

6. KRÖNER, A.; GREILING, R.; REISCHMANN, T.; HUSSEIN, I. M.; STERN, J.; DURR, S.; KRUGER, J. and ZIMMER, M. (1987): Pan African crustal evolution in northeast Africa. In: Kröner, A. (ed.) Proterozoic lithospheric evolution. American Geophysical Union, Geodynamic series, 17, pp. 235-257.

7. LILLESAND, T. M. and KIEFER, R. W. (1994): Remote sensing and image interpretation. 3^rd ed., - 750 pp., New York, NY (Wiley).

8. SABINS, F. F. Jr. (1987): Remote sensing principles and interpretation. 2^nd ed., -449 pp., San Francisco (Freeman).

9. SINGH, A. and HARRISON, A. (1985): Standardized principal components. Int. J. Remote Sens., 6 (6), pp. 883 – 896, (Taylor and Francis), London.