ASTER Multispectral Imagery for Spectral
Unmixing based Mine Tailing cartography in the
North of Tunisia
Nouha Mezned1 , Sˆ adi Abdeljaoued1 and Mohamed Rached Boussema2
Laboratoire des Ressources Min´ rales et Environnement,
Facult´ des Sciences de Tunis,
Campus Universitaire du Belv´ d` re, 2092 Tunis, TUNISIA
Email: mezned email@example.com
ee e e ` ee
Laboratoire de T´ l´ d´ tection et Syst` me d’Information a R´ f´ rences Spatiales
Ecole Nationale d’Ing´ nieurs de Tunis,
1002 le Belv´ d` re Tunis, TUNISIA
Abstract— In the North of Tunisia, Hammam Zriba region to detect, to cartography and to follow mine tailing change
represents an important mine activity zone, it is among the and impact on the environment. Remote sensing technologies
most important in the Maghreb . The mine tailings cause the are an excellent tool for mine tailing environment change
environment degradation, so it had polluted the atmosphere, the
soils, the vegetation, the water quality and the ecosystem of the following and impact detection.
Hammam ouady (Dry River) which is discharged directly in the In this study, we are interested in the cartography of mine
Miliane River. The Miliane River represents the second important tailings around the test site and particularly to generate a de-
resource of potable water, after Medjerda River. In this study, we tailed spatially map. This work is based on the remote sensed
propose to use the ASTER VNIR and SWIR Surface Reﬂectance, data analysis and its comparison with the local (relatively to
and TIR data for mine tailing cartography. Classiﬁcation of the
SWIR product based on partial spectral unmixing is performing the considered image) terrain truth. It’s beneﬁcial for the local
using image derived endmembers identiﬁed as a mineral groups. government to supervise the mineral environmental changes
Unmixing results coincide with XRD laboratory analysis of the and be aware of the pollution status dynamically.
ﬁeld samples. Moreover, the given results show that ASTER We propose to use the multispectral Advanced Spaceborn
multispectral image is more precise for certain mineral detection Thermal Emission And Reﬂection Radiometer ASTER Level
than Landsat ETM+ image.
2B Visible Near Infra Red (VNIR) and Short Waves Infra
I. I NTRODUCTION Red (SWIR) data acquired on March 2001. This product
present respectively a 15m and 30m spatial resolution, and
In the north-Oriental of Tunisia, the Jurassic block zone a 4 bands and 6 spectral bands spectral resolution image.
represents the Tunisian ﬂuoric province (F-Ba-Zn-Pb) with Moreover, we used Thermal Infra Red (TIR) products to test
12th deposits. This province includes J. Ressas exploited for result complementarities.
lead and zinc, Hammam Zriba mine exploited for ﬂuorite and The classiﬁcation of the multispectral image was performed
Barite, etc. Particularly, Hammam Zriba region (36◦ 20’N and by the partial spectral unmixing method using the image
10◦ 13.5’E) represents an important mine activity zone, it is derived endmembers compared to mineral library spectra. This
among the most important in the Maghreb with 5 milions tons will allow us to bring more explication and interpretation
of ore with 15-35 percent CaF2 . Moreover, it is character- of mineral spatial distribution and to provide information on
ized by a touristique activity manifested by the thermal station locating tailing zones.
and an interesting agriculture productivity. Thus, it constitues
an important economic activity province . II. T EST S ITE DESCRIPTION
In this site, the abandoned Hammam Zriba mine leaving Hammam Zriba constitues an important economic activity
behind substantial quantities of ﬁnely-divided mineral waste or region by its touristique activity manifested by the thermal
tailings. The mine tailings cause the environment degradation, station, an interesting agriculture productivity and above all
so it had polluted the atmosphere, the soils, the vegetation, an important mine activity zone . The ﬂuorspar mining
the water quality and the ecosystem of the Hammam ouady industry of Hammam Zriba mine (thitonic-campanien contact)
(Dry River) which is discharged directly in the Miliane River. in the north-oriental Tunisia is among the most important in
The Miliane River represents the second important resource of the Maghreb with 5 milions tons of ore with 15-35 percent
potable water, after Medjerda River. Thus, tailings constitute a of CaF2. The mineralization is constituted by these different
threat for the environment and particularly, the human health minerals: barytine - Celestine with 40 to 45 percent, ﬂuorite
(tailings cause silicosis disease). So, it became necessary with 15 to 35 percent and quartz with 10 to 40 percent and
Fig. 1. Map showing the location of the Hammam Zriba mine 3D till
representation mapped with a ASTER VNIR imagery (60x100 pixels). The
white numerous indicate location of the mine tailings (1), thermal station (2),
El Hamam Ouady (3) and Meliane river (4).
blend, galena and pyrite with 5 to 15 percent and calcite with
2 to 10 percent . (c)
The mine is located near El Hamam Ouady (36◦ 20’N Fig. 2. XRD analysis results. The d-spacing value: 3.406 indicates the
10◦ 12’E) (Figure 1) which discharge directly in the Meliane presence of Barite, 3.034 indicates the presence of Calcite, 3.15 indicates
the presence of Fluorite, 2.96 indicates the presence of Galena, 6.59 indicates
river, the second important river in Tunisia after Medjerda the presence of Hemimorphite, 3.46 indicates the presence of Mizzonite, 3.34
river. The resulting mine tailings , in particularly the residues indicates the presence of Quartz and 2.74 indicates the presence of Smithsonite
of lead, zinc and above all ﬂuor and barium cause an impact
which is more visible on the human (cause silicosis disease),
faune and ﬂore health. The exceeding of ﬂuorite and barite B. ASTER data preprocessing
due to several years of ﬂuorite barite-melting activities threat
human life and let vegetation and ecosystem suffering. The Japanese Advanced Spaceborne Thermal Emission and
Reﬂection Radiometer ASTER, one of the principal instru-
ments on Terra, was launched in December 1999 to estabilish
III. S PECTRAL DATA AND F IELD DATA C OLLECTION a spaceborne capability for high spatial, multispectral visible-
shortwave infrared and thermal infrared remote sensing data
A. Field data collection and analysis
mapping of the Earth environment . It covers the approxi-
Tailings samples were collected at deposit location, con- mately 0.55 to 11.3 µm range, 4 VNIR bands, 6 SWIR bands,
sidered as the pollution source, through out the study site and 5 (TIR) bands with respectively 15m, 30m, and 90m
in February 2007. This choice of the campagne date was spatial resolution.
fundamental for the study to transmet and to preserve the We have used on-demand product ASTER Surface Re-
tailing state since the data acquisition period. ﬂectance which contains surface reﬂectance for each of the
The samples cover different tailings areas within deposits to nine VNIR, SWIR and TIR bands. The Level 2B surface
include tailings of various mineral compositions and oxidation reﬂectance data set contains surface reﬂectance after apply-
stages (6 radials for each dyke). Each sample was gathered ing the atmospheric corrections to observed radiances. The
from the top of the tailing surface (ﬁrst 3-4cm). We try by this accurate atmospheric correction removes effects of changes
restrict sampling to have an homogenous and a signiﬁcative in satellite-sun geometry and atmospheric conditions and
sample for each dyke. improves surface type classiﬁcation. Accurate calibration of
The analysis using X Ray Diffraction XRD method (Figure ASTER band 9 however is considered problematic because
2) was conducted to reveal the mineral composition of these of the uncertainty of water vapor estimates for atmospheric
samples as listed below: Barite, Calcite, Fluorite, Galena, correction. So, ASTER Band 9 reﬂectance can be too high
Hemimorphite, Mizzonite and Smithsonite with different rel- that some mineral maybe overestimated .
ative concentration. This result conﬁrms the band ratio com- The complete processing of the ASTER Surface Reﬂectance
bination analysis. data is illustrated by the ﬂowchart on ﬁgure 3. It consists on
Fig. 3. Flowchart illustrated the complete processing of the ASTER Surface Reﬂectance data.
three main steps. The ﬁrst step is the geometrical registration, respectively B13/B14 and B12/B13 Ratio image . Carbon-
so as to have the same geodetic reference. The second one con- ates have a unique spectral feature. They have low emissivity
cerns the image re-enhancement and the image transformation in ASTER band 14 spectral region, and high emissivity in
by the band combination for easier image interpretation. In the ASTER band 10 to 13 spectral region. The SiO2 minerals
last step, we make a principal component analysis (PCA) for have another unique spectral property in TIR. They have lower
optimal endmembers selection needed for the partial spectral emissivity in ASTER band 10 and band 12 spectral regions
unmixing method. than in band 11. . The resulting index maps are shown
The registration to the relief map is performed using the in ﬁgure 5. Figure (a) show the repartition of the carbonate
Digital Elevation Model (DEM), so that the precision of the minerals, particularly calcite mineral as proved by the XRD
registration is less than half-pixel. The band combination analysis. This repartition indicates that this mineral existing
has been also tested to provide best image interpretation within deposits but not very highly signiﬁcant (with a relative
using false-color composite image. The ASTER SWIR band important concentration). Whilst, quartz mineral (Figure (b))
combination were successfully applied for geologic  and is less important than calcite within mine tailings. This is not
lithologic  interpretation. A general picture of the test site a surprising result since the principal minerals are Barite and
is shown in Figure 4, combining SWIR bands 4,6 and 8 in ﬂuorite.
Red Green Blue RGB and processed to increase the color
IV. U NMIXING BASED M ULTISPECTRAL ASTER
saturation. In deed, Red-pink areas mark mostly the presence
of kaolinite and/or alunite. As we can see, these minerals are
absent in the tailing deposit. A. Mineral spectral library
Moreover, the spectral indices resulting from the orthogonal Based on the XRD analysis, different spectra was selected
transformation of the ASTER (SWIR) bands are used for from USGS  or JPL Spectral library  and have been
discrimination and mapping of surface rock types particularly, resampled to demonstrate major spectral signatures of minerals
clays (such as Alunite, Kaolinite, Montmorillonite etc.). The in SWIR. The mineral spectra which shown in ﬁgure 6 was
ASTER TIR bands can also used for the spectral indices used to the image derived endmembers identiﬁcation. In deed,
generation. The resulting spectral index images are useful for a ﬁrst three endmembers were chosen from the ASTER SWIR
lithologic mapping and were easy to interpret geologically. image window (60x100 pixels) using the scatter plots in the
We proposed to use ASTER TIR bands to mapping quartz ﬁrst three principal components , which contain 99.64
and carbonate minerals. percent of the variability in the data. The mine tailing map
The carbonate and the SiO2 index can be achieved using resulting from the classiﬁcation based on the three image
Fig. 4. SWIR bands 4,6 and 8 in Red Green Blue RGB color composite. The given white arrow indicates the mine tailing deposit location.
Fig. 5. Images showing the spectral indices resulting from B13/B14 and B12/B13 band Ratio combination for respectively the carbonate and the quartz
minerals analysis. The given red arrow and red line indicates the mine tailing deposit location.
derived endmembers was used for masking undesired zones Filtering to ﬁnd the abundances of endmembers. For this
and highlight only the mine tailing interesting area. Thus, a technique, not all of the endmembers in the image need to
second three endmembers were chosen from the tailing area be known. It maximizes the response of the known endmem-
masked image and illustrated in ﬁgure 7. Compared to the ber and suppresses the response of the composite unknown
resampled library mineral spectra, only Calcite, smithsonite, background, thus matching the known signature. It provides a
galena and quartz minerals are identiﬁed and characterized by rapid means of detecting speciﬁc materials based on matches
an absorption feature in SWIR bands near 2.336 µm for calcite to library or image endmember spectra and does not require
and smithsonite, 2.167 µm and 2.262 µm for galena and quartz knowledge of all the endmembers within an image scene .
respectively. These minerals can be classiﬁed as a groups. So, results can be obtained in an area without any detailed
The class 1 groups quartz, calcite, smithsonite and galena. prior knowledge of the geology of the area. In other sides, this
The second class groups calcite, smithsonite and galena with technique may ﬁnd some false positives for rare materials.
the gross shape curve in the mean ASTER image spectra The results of the matched ﬁltering as shown by the ﬁgure
correspond to the signiﬁcant spectral absorption of calcite appear as a series of gray-scale images, one for each selected
and smithsonite in band 8 centered in 2.334 µm . This same endmember. It provide a means of estimating relative degree
mineral composition belong also to Class 3 but the gross shape of match to the reference spectrum and approximate sub-pixel
curve in the mean ASTER image spectra correspond to the abundance. The ﬁgure 7 shows three maps for the different
absorption of galena in band 5 centered in 2.167 µm. It will be classes: class1, class 2 and class 3. The maximum of the three
not possible to map particular minerals using ASTER imagery. classes is detectable in the tailing deposits area (indicates by
However, it is possible to differentiate between mentioned a white color) with different relative abundance. Black color
mineral groups. This is a signiﬁcant improvement over Landsat indicates the absence of materials.
Enhancement Thematic Mapper plus (ETM+) data .
V. C ONCLUSION
B. Mine tailing cartography
In this paper, a mine tailing cartography in the Hammam
Several recent results have shown that spectral analysis of Zriba mine was achieved using the ASTER VNIR and SWIR
airborn imaging spectrometer and satellite data can provide reﬂectance surface and TIR data. False-color composite image
useful mineralogical and geochemical information for geologi- was used to get an overview about the area mineral compo-
cal mapping and exploration. In addition, the image processing sition. Band rationing is also applied in an attempt to map
techniques developed to facilitate surface compositional map- distribution of SiO2 and carbonates minerals. Moreover, the
ping using spectral unmixing techniques are now more robust classiﬁcation was performed with partially spectral unmixing
. Our investigation builds on the existing knowledge of method based on image derived endmembers. The unmixing
surface mineral identiﬁcation techniques, particularly partial results has shown that the ASTER multispectral data is more
spectral unmixing method with reference to spectral library. precise than Landsat ETM+ data on mine tailing cartography.
In deed, we used as spectral mapping technique the Matched Furthermore, multispectral ASTER image cartography results
(a) (a’) (b) (b’)
(c) (c’) (d) (d’)
(e) (e’) (f) (f’)
(g) (g’) (h) (h’)
Fig. 6. Mineral library spectra and their correspondant resampled spectra to ASTER SWIR bandpasses: barite (a) and (a’), calcite (b) and (b’), ﬂuorite (c)
and (c’), galena (d) and (d’), hemimorphite (e) and (e’), mizzonite (f) and (f’), quartz (g) and (g’) and smithsonite (h) and (h’).
show a complementary with laboratory analysis. However, it  Ninomiya, Yoshiki, Mapping quartz, carbonate minerals, and maﬁc-
could be important to make comparison of two different mine ultramaﬁc rocks using remotely sensed multispectral thermal infrared
ASTER data, SPIE, SPIE Vol. 4710, pp. 191-202, 2002.
type results.  USGS, Digital Spectral library 05 Website adress:
VI. ACKNOWLEDGMENT  C. I. Grove, S. J. Hook and E. D. Paylor II, Laboratory Reﬂectance
Spectra of 160 Minerals, 0.4 to 2.5 Micrometers JPL Publication 92-2,
The authors would like to thank the NASA Land Processes 1992.
Distributed Active Archive Center and the User Services  P. Junek, Geological mapping in the Cheleken Peninsula, Turkministan
area using Advanced Spaceborne Thermal Emission and Reﬂection Ra-
USGS Earth Resources Observation and Science (EROS) for diometer (ASTER) data, ISPRS, 2004.
providing numerous remotely sensed data.  M. Abrams and S.J. Hook, Simulated Aster Data for Geologic Studies,
IEEE TGARS, Vol. 33, No. 3, 1995.
R EFERENCES e
 A. Plaza, P. Martinez, R. P´ rez and J. Plazza, Spatial/Spectral Endmem-
ber Extraction by Multidimensional Morphological Operations, IEEE
 S. Bouhlel and J. P. Fortun´ , Contribution a l’´ tude min´ ralogique et
e ` e e Trans. Geosci. Remote Sensing, vol. 40, N◦ 9, 2000.
g´ ochimique des min´ raux de Pb-Zn-Cu de la province ﬂuor´ e tunisi-
e e e  N. Mezned, S. Abdeljaoued and M. R. Boussema, Mine tailings
enne., Notes Serv. G´ ol. Tunis, n◦ 51, p. 155-176. mapping using Landsat multispectral imagery of the versant basin amont
 ANPE, Etude de rehabilitation du site de l’ancienne mine de Hammam of Medjerda river in the north of Tunisia, IGARSS’06, 2006.
Zriba, Plan d’action et variante de r´ habilitation du site de l’ancienne
e  J. W. Boardman and F. A. Kruse, Automated Spectral Analysis: A
mine de ZRIBA, 2002. Geological Example Using AVIRIS Data, North Grapevine Mountains,
 Y. Yamaguchi, A. B. Kahle, H. Tsu, T. Kawakami and M. Pniel, Overview Nevada, Proceedings of the Tenth ERIM Applied Geologic Remote
of advanced spaceborne thermal emission reﬂectance radiometer, IEEE Sensing Conference, San Antonio, Texas, p 407-418, 1994.
TGARS, vol. 36, p. 1062-1071, 1998.
 Y. Yamaguchi and C. Naito, Spectral indices for lithologic discrimination
and mapping by using the ASTER SWIR bands, International Journal of
Remote Sensing, Vol. 24, p. 4311-4323, 2003.
Fig. 7. Class fraction maps derived from the partial spectral unmixing of ASTER SWIR image using the image derived endmembers spectra. Bright tones
in the fraction maps represent high fractions and dark tones represent low fractions. The given red arrows indicate the mine tailing deposit location.