African Journal of Science and Technology (AJST) Science ... Vol 12 No 1.pdf · range for systems involving non polar solutes on polar stationary phases (Heberger et al. 2002). There

African Journal of Science and Technology (AJST)Science and Engineering Series

Journal Africain de Science et de Technologie

Volume 12 Number 1October 2012

ISBN No. 1607-9949.

For subscription and frurther information contact:

Pour tout renseignement complémentaire s’adresser au:

ANSTI/RAIST SecretariatUNESCO Regional Office in Nairobi

P.O. Box 30952 - 00100 Nairobi, KenyaTelephone: +254 20 7622619/20

E-mail:[email protected]

The articles appearing in this Journal express the views of their authors and not necessarily those of UNESCO (ANSTI)

The African Network of Scientific and Technological Institutions (ANSTI).

Réseau Africain d’Institutions Scientifiques et Technologiques (RAIST).

Science & Engineering Series

In pursuance of UNESCO/ANSTI’s objective to facilitate the dissemination of research results and within the framework of the organization, the African Development Bank (AfDB) has provided a grant to the African Network of Scientific and Technological Institutions (ANSTI), for the publication of the African Journal of Science and Technology (AJST).

The African Journal of Science and Technology (AJST) is an annual technical publication of the African Network of Scientific and Technological Institutions (ANSTI).

Le Journal Africain de Science et de Technologie est une revue scientifique du Réseau Africain d’Institutions Scientifiques et Technologiques (RAIST).


In pursuance of UNESCO’s objective to facilitate the dissemination ofresearch results and within the framework of the organization’s supportfor the African Network of Scientific and Technological Institutions(ANSTI), UNESCO has provided a grant to continue the publication ofthe African Journal of Science and Technology (AJST).

The African Journal of Science and Technology (AJST) is an annualtechnical publication of the African Network of Scientific andTechnical Institutions (ANSTI).

Le Journal Africain de Science et de Technologie est une revuescientifique du Réseau Africain d'Institutions Scientifiques etTechnologiques (RAIST).

For subscription and frurther information contact:Pour tout renseignement complémentaire s'adresser au:

ANSTI/RAIST SecretariatUNESCO-ROSTA - P.O. Box 30952, Nairobi, KenyaTelephone 254 2 7622619/20E-mail [email protected]

The articles appearing in this Journal express the views of their authorsand not necessarily those of UNESCO (ANSTI)

African Journal of Science and Technology (AJST) Science and Engineering Series Vol.12 No. 1

EDITOR-IN-CHIEF

Prof. Norbert Opiyo-Akech

EDITORIAL BOARD

Department of Geology E-mail: [email protected] University of Nairobi P.O.Box 30197 NAIROBI, KENYA

SUBREGIONAL/SUBJECT EDITORS

(Eastern Africa/Afrique de l’Est)/(Physics/Physique) Prof. B. Aduda, Department of Physics E-mail: [email protected] University of Nairobi P.O.Box 30197 NAIROBI, KENYA

(French speaking Africa/Afrique Francophone)/(Earth Sciences/Science de la Terre)

Dr. I.K. Njilah University of Yaunde I E-mail: [email protected] Department of Earth Sciences P.O. Box 812 Younde, CAMEROON

(Engineering/Technology/Technologie) Prof. Larry Gumbe E-mail: [email protected] Department of Environmental & Biosystems Engineering P.O.Box 30197 NAIROBI, KENYA

(Mathematics/Mathematique) Prof. Verdiana G. Masanja Civil Engineering Department E-mail:[email protected] University of Dar Es Salaam TANZANIA

(Biological and Agricultural Sciences) ( Science de Biologie et Agriculture) Prof. Ebenezer Oduro Owusu Zoology Department E-mail [email protected] University of Ghana,, Legon ACCRA, GHANA.

CONTENTS AJST, Vol. 12, No. 1: October, 2012

Page

E. Muzenda, M. Belaid, F Ntuli, and A . Arrowsmith 1 Influence of Temperature on Specific Retention Volumes of Environmentally Important Volatile Organic Compounds in Gas Liquid Chromatography

M. Messadi, A. Feroui and A. Bessaid 7 Supervised Color Image Segmentation, using LVQ Networks and K-means. Application : Cellular Image

M. L. Moropeng and A. Kolesnikov 14 Alternative Control of Nanoparticles Dispersity in High-Temperature Flow Reactors

P. K. Cheruiyot, G. M .N Ngunjiri, C. K. W. Ndiema, M.C. Chemelil, and 23 R. M. Wambua Effects of Ground Insulation and Greenhouse Microenvironment on the Rate and Quality of Biogas Production

R. T. Ranganai, 34 Euler Deconvolution and Spectral Analysis of Regional Aeromagnetic Data from the South-Central Zimbabwe Craton: Tectonic Implications

P. Sukdeo, P. Sukdeo, S. Pillay and A. Bissessur 51 An Assessment of the Presence of Heavy Metals in the Sediments of the Lower Mvoti River System

J. O. Owoseni, G. O., Adeyemi, Y. A. Asiwaju-Bello and A.Y.B.Anifowose 59 Engineering Geological Assessment of some Lateritic Soils in Ibadan, South- Western Nigeria using Bivariate and Regression Analyses

R. G. Kakai, D. Pelz and R. Palm 72 Relative Efficiency of Non-parametric Error Rate Estimators in Multi-group Linear Discriminant Analysis

F. Gbogbo, , R. Langpuur, and M. K.Billah, 80 Forage Potential, Micro-Spatial and Temporal Distribution of Ground Arthropods in the Flood Plain of a Coastal Ramsar Site in Ghana

B. Alhou, Jean-Claude. Micha and B. Goddeeris 89 Diversity of the Chironomidae (Diptera) of River Niger related to Water Pollution at Niamey (Niger)

1 AJST, Vol. 12, No.1: October, 2012

African Journal of Science and Technology (AJST)Science and Engineering Series Vol. 12, No. 1, pp. 1 - 6

INFLUENCE OF TEMPERATURE ON SPECIFIC RETENTION VOLUMESOF ENVIRONMENTALLY IMPORTANT VOLATILE ORGANIC

COMPOUNDS IN GAS LIQUID CHROMATOGRAPHY

1E Muzenda1*, 1M Belaid, 1F Ntuli, and 2A . Arrowsmith

1Department of Chemical Engineering, University of Johannesburg,Johannesburg, P O Box 17011, 2028, South Africa

2School of Chemical Engineering, University of Birmingham,Edgbaston, B15 2TT, Birmingham, United Kingdom

Email:[email protected]

ABSTRACT: Temperature dependence of specific retention volumes ogV of 13 volatile organic

compounds (VOCs) of environmental importance between the gas and liquid stationary phase(polydimethysiloxane, PDMS) are presented, determined by gas chromatographic method. Activitycoefficients at infinite dilution were calculated from these specific retention volumes and they are inagreement with those obtained from static headspace and group contribution methods by the authorsas well as literature values. The results of this work confirm that PDMS is well suited for VOCsscrubbing from waste gas streams. The measurements were carried out at temperatures (303.15,313.15, 323.15., 333.15, 353.15, 373.15, 393.15 and 423.15) K to allow transport calculations fordifferent seasons. Four PDMS polymers with average molecular weight ranging from 760 to 13 000

were used as solvents. Linear plots of log gV against T1

were obtained in all cases, thus allowing

for predictions at other temperatures not investigated in this study. Since the typical van’t Hoff plotswere nicely linear, dependable enthalpies and entropies of solute transfer from the mobile phase tothe stationary phase can also be calculated. Efforts were taken to ensure the best possible accuracyand trace the possible source of error. We devised a gas liquid chromatographic system whichsecured a simple retention mechanism and showed reproducible solute retention over a long periodof time.

Key words: Specific retention volume, Waste gas streams, Stationary phase, Gas liquid chromatography,Scrubbing, Temperature dependence

INTRODUCTION

Solvents play a critical role in key chemical operationssuch as separation processes. Mostly known organicsolvents such as polydimethylsiloxane contribute tosolving air pollution problems due to their relatively highvolatilities. For PDMS to be used effectively in thescrubbing of volatile organic compounds, it essential toknow how they interact with different solutes. Theimportant measure of this property is given by the specific

retention volume ogV or the activity coefficient at

infinite dilution i . Specific retention volumes can be

used through the reduction of infinite dilution activitycoefficients to design separation processes where the tracecomponents or impurities have to be removed. The gaschromatography (GC) is suitable for measurement of specificretention volumes because of the negligible vapourpressure of PDMS. The procedure of the measurement inparticular, the effect of flow rate of carrier gas, effect ofsample size, and liquid loading were previouslyinvestigated (Muzenda et al. 2008, 2009)

2AJST, Vol. 12, No. 1: October, 2012

E. MUZENDA

The specific retention volume is defined as the netretention volume per gram of stationary phase at 00C. Thisis very important because it allows the comparison ofretention data obtained at different temperatures withdifferent weight of stationary phase. Specific retentionvolumes were introduced by Littlewood et al. (1955). Hesuggested their use in place of partition coefficients invapour identification. Littlewood measured specificretention volumes for a series of alcohols, aromatichydrocarbons, and esters in silicone 702 – fluid and tritolylphosphate. Lichtenthaler et al. (1973) measured specificretention volumes from gas – liquid chromatography forpolydimethysiloxane – hydrocarbon at 25, 40 and 550C.The carrier gas flow rate was varied from 18 to 120 cm3 perminute and it was found to have no effect on the specificretention volumes. Summers, Tewari and Schreiber (1972)obtained specific retention volumes forpolydimethysiloxane – hydrocarbon systems in fourcolumns with different liquid loading. Experimentaltemperature and flow rate were varied from 25 to 700C and70 to 120ml/min respectively. Ashworth and co – workers(1984) reported replicate gas – liquid chromatographicbased specific retention volumes, activity coefficients andinteraction parameters of ten solutes withpolydimethylsiloxane at 303K.

Though inconclusive, the effects of temperature onretention data, in particular Kovats indices, has beeninvestigated and debated for a long time (Heberger et al.2002). Almost linear dependence of retention data for nonpolar solutes on non polar phases have been reported(Heberger et al. 2002). The Antoine type (which is nonlinear) shows better performance for wide temperaturerange for systems involving non polar solutes on polarstationary phases (Heberger et al. 2002). There have alsobeen numerous studies of temperature effects on soluteretention in reversed phase liquid chromatography (RPLC).Linear Van’t Hoff plots were observed in the typical RPLCsystems as reported and cited (Sentell et al. 1995). In thesestudies, the enthalpies of solute transfer from the mobilephase to the stationary phase were calculated from theslopes of the van’t Hoff plots. Non linear plots are oftenobserved when the temperature range is more than 450C.

Usually the temperature dependency of the specificretention volume is expressed as

QT

PVg

1ln 0(1)

RHP

s (2)

RS

MRQ

s

s

273ln (3)

Where R is the universal gas constant sM , is the molecular

mass of the stationary phase. sH , sS are,respectively, the standard molar enthalpy and entropy ofsolution for the transfer of a mole of solute, from the idealgas where its partial pressure is 1 atm, in the stationaryphase, where its mole fraction is 1x . The molecularinteractions and the environment are similar to that of aninfinitely dilute solution.

METHODOLOGY

The apparatus used has been described previously(Muzenda et al., 2000, 2002, 2008a, 2008b, 2009). The carriergas was helium. A constant sample size of 0.1µl. wasinjected into the columns at constant flow rate of 35.97ml/min. Measurements were made at 303.15, 313,15, 323.15,333.15, 353.15, 373.15, 393.15 and 423.15K. Special care wasexercised to control and determine the system temperatureaccurately.

Column Preparation

Polydimethylsiloxane (PDMS) was coated intoChromosorb P, AW - DMCS or Chromosorb W, AW –DMCS (acid washed, dimethylchlorosiloxane treated) froma solution in chloroform. Details of column preparation areshown in Table 1.

Table 1: Description of columns

Properties Column 1 Column 2 Column 3 Column 4Length, m 1 1 1 1

Chromosorb W or P,g 6.158 7.291 7.186 4.86PDMS,g 0.688 0.825 0.804 0.54Wt % PDMS 10.05 10.16 10.06 10Viscosity of PDMS, cp 5 10 50 500Mw of PDMS 760 1000 3200 13000

AJST, Vol. 12, No. 1: October, 2012

Influence of Temperature on Specific Retention Volumes of Environmentally Important VolatileOrganic Compounds in Gas Liquid Chromatography

3

MATERIALS

The materials used have been described previously(Muzenda et al., 2000, 2002, 2008a, 2008b, 2009). Thethirteen solutes used in this work were coded numericallyas in table 2 for simplicity.

Table 2: Solute Description and Code

Code no. Compound Code no. Compound1 chloroform 8 xylene2 n-pentane 9 diethylether3 n-hexane 10 butylaceta te

4 n-heptane 11Isobutyl methyl

ketone5 acetone 12 triethylamine6 toluene 13 ethyl lmethyl ketone7 cyclohexane

Calculations

The specific retention volume, corrected to 0o, is given by

cs

MR

o

i

o

i

og TW

ttF

pp

pp

V 15.273

1

1

23

2

(4)

In equation 4, F volumetric carrier gas flow rate at column

outlet temperature and pressure, ml/min; MR tt is theretention time, ie., the time difference between air and solute

peaks, min; cT is column temperature, Ko , sW is weightof polymer in the column, g; is inlet and is outlet pressure.The use of the flame ionization detector permitted theapplication of the mathematical air peak method toapproximate air peak maximum. To account for the gasholdup in the column, the retention time was taken as thedifference between the maxima of the air and solute peaks(Purnell, 1962).

RESULTS AND DISCUSSION

Variation of specific retention volumes from literaturefindings

Specific retention volumesCompound Temp (K) This work Literature % Variation Sourcechloroform 303 188.2 181.7 3.4 an-pentane 303 69.91 66.11 5.7 a

313 47.81 47.65 0.3 a333 24.94 24.76 0.7 e

n-hexane 303 191.98 179.2 7.1 b313 126.23 124.2 1.6 d333 58.8 60.91 3.5 e

n-heptane 303 508.25 482.5 5.3 b313 294.57 290.8 1.3 c333 146.75 144.3 1.7 d

toluene 303 796.45 791.1 0.7 a313 529.08 516.6 2.4 b323 300.15 269.2 11.5 b

cyclohexane 303 336.8 315.1 6.9 b333 103.4 106 2.5 e

xylene 303 2645.55 2654.6 0.3 c313 1223.61 1187.5 3 b333 516.27 536.5 3.8 d

The solute specific retention volumes reported in Table 3were calculated from corrected peak retention times usingthe well-known expression of Littlewood et al. (1955). Theretention used in the calculation of specific retentionvolumes was an average of five measurements. Individualvalues of retention times were found to vary by no morethan 1% in all cases. Specific retention volumes reportedin this work compare very well with literature findings. Thesuccessful comparison gives an indication of the GLC as arapid, simple and accurate method for studying thethermodynamics of the interaction of a volatile solute witha non volatile solvent. Therefore specific retention volumesobtained in this study can be considered to be accurateand reliable for the calculation of infinite dilution activitycoefficients.

aAshworth et al. (1984); bLichtethaler et al. (1973); cDeshpande etal. (1974); dSmidsrod and Guillet (1969)

Table 3: Variation of specific retention volumes fromliterature findings


E. MUZENDA

Temperature dependence of specific retention volumes

The specific retention volumes of chloroform, pentane,hexane, heptane, acetone, toluene, cyclohexane, xylene,diethyl ether, butyl acetate, isobutyl methyl ketone,triethylamine and ethyl methyl ketone were measured byinjecting a constant amount of sample of 0.1µl into thefour columns. Measurements were done at a constantmean flow rate of 35.97ml / min and the temperature wasvaried from 303.15 to 423.15K. Figures 1 to 4 show almost

linear dependence of log Vg on T1 with Vg decreasing

with increasing temperature. Chromatographic retentiondata from variable temperature runs may used to estimatethermodynamic properties according to the well-knownVan’t Hoff relation (Sellergren and Shea, 1995).

ln//ln ' RSRTk (5)

Figure 1: Effect T on Vg (Column 1) Figure 2: Effect of T on Vg (Column 2)

Figure 3: Effect of T on Vg (Column 3) Figure 4: Effect of T on Vg (Column4)


Influence of Temperature on Specific Retention Volumes of Environmentally Important VolatileOrganic Compounds in Gas Liquid Chromatography

5

Where is the phase volume ratio and R is gas constant.The linear portions of these plots give enthalpies

RT/ and entropies S for the transfer of onemole of solute into the stationary phase. All the Van’t Hoffplots obtained in this study were linear, and the regressioncorrelation coefficients were better than 0.999 in all cases.Typical Van’t Hoff plots are shown in Figures 1 to 4. Thetrends obtained here are in agreement with those observedby Ulrich et al. (2006), Sellergren and Shea (1995), Domanskaand Marciniak (2008), Tudor (1997), Tudor and Moldovan(1999), Lee and Cheong (1999), Makela and Pyy (1995) andSentel et al. (1995).

The choice of optimum of temperature for the absorptionprocess requires knowledge of temperature dependenceof the activity coefficient. Infinite dilution calculated fromthe specific retention volumes shows that the higher thetemperature the higher the activity coefficients. Thistendency is very favourable for the desorption, becausethe higher values at higher temperature would easeregeneration.

CONCLUSION

Specific retention volumes for 13 volatile organiccompounds in polydimethylsiloxane were measured overthe temperature range from 303.15 K to 423.15K using theGLC method. Good linear relationships were observed onthe plots of log Vg versus . This can be used to predict thespecific retention volumes and hence infinite dilutionactivity coefficients at temperatures not studied here. Themeasurements were highly reproducible with relativestandard deviation and coefficient of variation in thedetermination of specific retention volumes of 0.00013 and0.013 respectively. The control of the operating conditionsas indicated by the stability of the baseline and the shapesof the peaks indicated that the equipment was workingproperly. The close agreement of the specific retentionvolumes and infinite dilution activity coefficientscalculated from this study with those reported in literatureproved that a sound technique was used.

ACKNOWLEDGEMENTS

We are indebted to the Department of Chemical Engineeringof the University of Johannesburg for financial support.The encouragement and advice of Professor Ashton isgreatly appreciated.

REFERENCES

Muzenda, E., Arrowsmith, A., Ashton, N., “Study of theEffects of Experimental Variables in Solute RetentionVolumes by Gas Liquid Chromatogrphy (glc) in PolymerSolution Thermodynamics”, CHEMCON 2008,Chandigarh India, (2008)

Muzenda, E., Belaid, M., Ntuli, F., Arrowsmith, A.,“Absorption of Volatile Organic Compounds in Silicon:Determination of Infinite Dilution Activity Coefficientsby Dynamic Gas Liquid Chromatographic Technique”The 8th WCCE 2009, Montreal, Canada (2009)

Littlewood, A. B., Phillips, C. S. G., Price D. T., “TheChromatography of gases and vapours Part V:Partition analysis with columns of Silicone 702 andTritolylphosphate”, Journal of Chemical Society, 1480(1955)

Lichtenthaler, R. N., Newman, R. D., Prausnitz, J. M.,“Specific retention volumes from Gas – LiquidChromatography for PolydimethylsiloxaneHydrocarbon Systems”, Macro., 6, 4, 650 – 651 (1973)

Summers, W. R., Tewari, Y. N., Schreiber H. P.,“Thermodynamic interaction in Polydimethylsiloxane– Hydrocarbon Systems from Gas – LiquidChromatography”, Macro., 5, 1, 12 – 16 (1972)

Ashworth, A. J., Chien, C. F., Furio, D. L., Hooker, D. M.,Kopecni, M. M., Laub, R. J, Price, G. J., “Comparisonof static with gas-chromatographic solute infinite-dilution activity coefficients with polydimethylsiloxanesolvent”, Macro., 17, 5, 1090 – 1094 (1984)

Heberger, K., Gorgenyi, M., Kowalska, T.. “Temperaturedependence of Kovats indices in gas chromatographyrevisited”, Journal of Chromatography A, 973, 135 –142 (2002)

Sentell, K. B., Ryan, N. I., Henderson, A. N., “Temperatureand solvation effects on homologous series selectivityin reversed phase liquid chromatography”,Analy.Chimica Acta, 307, 203 – 215 (1995)

Muzenda, E., Arrowsmith, A., Ashton, N., “Infinite dilutionactivity coefficients for VOCs in polydimethylsiloxane”IChemE, Bath, United Kingdom (2000)

Muzenda, E., Arrowsmith, A., Ashton, N., “GLC as anoptimization technique”, The Fifth InternationalConference on Manufacturing Process Systems andOperations Management in Less IndustrialisedRegions held at ZITF, Zimbabwe, ISBN 0-7974-2456-3, 1 – 6 (2002)

Purnell, H., Gas Chromatography, John Willey and Sons(1962)


E. MUZENDA

Sellergren, B., Shea, J. H., “Origin of peak asymmetry andthe effect of temperature on the solute retention inenantiomer separations on imprinted chiral stationaryphases”, J. Chroma. A, 690, 29 – 39 (1995)

Bay, K., Wanko, H., Ulrich, J., “Absorption of VolatileOrganic Compounds in Biodisel: Determination ofInfinite Dilution Activity Coefficients by HeadspaceGas Chromatography”, IChemE, 84,A1, 22 – 28 (2006)

Domanska, U., Marciniak, A., “Measurements of activitycoefficients at infinite dilution of aromatic and aliphatichydrocarbons, alcohols, and water in the new ionicliquid [EMIM][SCN] using GLC”, Journal of ChemicalThermodynamics, 40, 860 – 866 (2008)

Tudor, E., “Analysis of the equations for the temperaturedependence of the retention of the retention index 1.Relation between equations”, J. Chroma. A, 858, 65 –78 (1999)

Tudor, E., Moldovan, D., “Temperature dependence of theretention index for perfumery compounds on a SE –30 glass capillary column II. The hyperbolic equation”J. Chroma. A, 848, 215 – 227 (1999)

Lee, C. S., Cheong, W. J., “Thermodynamic properties forthe solute transfer from the mobile to the stationaryphase in reversed phase liquid chromatographyobtained by squalane – impregnated C18 bondedphase”, J. Chroma. A, 848, 9 – 20 (1999)

Makela, M., Pyy, L., “Effect of temperature on the retentiontime reproducibility and on the use of programmablefluorescence detection of fifteen polycyclic aromatichydrocarbons”, J. Chroma. A, 699, 49 – 57 (1995)

Deshpande, D. D., Patterson, D., Schreiber, H. P., Su, C. S.,“Thermodynamic Interactions in Polymer Systems byGas – Liquid Chromatography. IV. Interactionsbetween Components in a Mixed Stationary Phase”,Macro., 7, 4, 530 – 535 (1974)

Smidsrod, O., Guillet, J. E., “Study of polymer interactionsby gas chromatography. Macro., 2, 272 (1969)

7 AJST, Vol. 12, No. 1: October, 2012


SUPERVISED COLOR IMAGE SEGMENTATION, USING LVQNETWORKS AND K-MEANS. APPLICATION : CELLULAR IMAGE

M. Messadi, A. Feroui and A. Bessaid

Laboratory of Biomedical Engineering, Electronic Department, Faculty of Science Engineering,University of Abou Bekr Belkaid, Tlemcen BP 119, 13000, Algeria,

Email: [email protected]; [email protected]

ABSTRACT: TThis paper proposes a new method for supervised color image classification by theKohonen map, based on LVQ algorithms. The sample of observations, constituted by image pixelswith 3 color components in the color space, is at first projected into a Kohonen map. This map isrepresented in the 3-dimensional space, from the weight vectors resulting of the learning process .Image classification by kohonen is a low-level image processing task that aims at partitioning animage into homogeneous regions. How region homogeneity is defined depends on the application.In this paper color image quantisation by clustering is discussed. A clustering scheme, based onlearning quantisation vector (LVQ), is constructed and compared to the K-means clustering algorithm.It is demonstrated that both perform equally well. However, the former performs better than the latterwith respect to the known number of although class. Both depend on their initial conditions andmay end up in local optima. Based on these findings, an LVQ scheme is constructed which is completelyindependent of initial conditions; this approach is a hybrid structure between competitive learningand splitting of the color space. For comparison, a K-means approach is applied; it is known toproduce global optimal results, but with high computational load. The clustering scheme is shownto obtain near-global optimal results with low computational load

Keywords: color image, kohonen, LVQ, classification, K-means

INTRODUCTION

In this paper the problem of color image quantisation isdiscussed. Color quantisation consists of two steps: tem-plate design, in which a reduced number of template col-ors (typically 8-256) is specified, and pixel mapping inwhich each color pixel is assigned to one of the colors inthe template. From the pattern recognition point of view,color quantisation can be regarded as a supervised classi-fication of the (2D) color space, each class being repre-sented by one color template. Since an RGB image cancontain up to (256)3 distinct colors, the classification prob-lem involves a large number of data points in a low dimen-sional space. Several techniques exist for colorquantisation. First, there is the class of splitting algorithmsthat divide the color space into disjoint regions, by con-secutive splitting up of the space. From each region acolor is chosen to represent the region in the color tem-plate. Another class of quantisation techniques performsclustering of the color space, and cluster representatives

are chosen as template colors. A frequently used cluster-ing algorithm is the K-means clustering algorithm. Here,an iterative updating of the cluster representatives and anassignment of color pixels to clusters takes place. Cluster-ing algorithms are commonly accepted as optimalquantisation approaches, but are also known as very timeconsuming. Moreover, although optimal, the above clus-tering algorithms suffer from their dependence on initialconditions. In most applications, one specific initial con-dition is chosen to present the results. However, usingother initial conditions can change the performance of thealgorithm dramatically. In this paper, the problem of localoptima in color image quantisation is studied, by applyingseveral clustering techniques. First of all, K-means is com-pared to a Competitive Learning Vector technique (LVQ).LVQ is very similar to K-means, in the sense that it mini-mizes the same objective function and the numbers of classare known. The main difference is that when using LVQ,cluster centers are updated sequentially [1].


M. MESSADI

n

jiaAimage ),(

321

232221

131211

mmm PPP

PPPPPP

Index

CLUSTERING ALGORITHMS

Principle The separation of color from a topographic map is aproblem of image color segmentation [4]. Information,which we can extract from this type of map, isrepresented with the colors:

Images that we have in entrance are coded on threeplans: red, green and blue. The color of each pixel isgiven by a vector with three components. Generally,this vector is given in the RGB space color. The threeelements of each vector corresponding to the RGBcomponents of every pixel of the image to bereprocessed are presented at the entrance of a classifier[10] with the aim of doing a color separation. An exampleof a data file with the three components, R, G, B is givenbelow (Table 1).

Table 1

R G B color60 193 111 Blue

166 182 109 Green226 48 242 Brown

0 0 0 Black12 17 16 Black

200 190 140 Brown161 24 255 Green… … … ….

The use of the color in image segmentation is arelatively recent research topic. Although one findsseveral algorithms of color segmentation, but the

Figure 1: Connection between the image and the template

literature is not rich enough than that for images in grey level[9].

The method we propose to solve the problem of the colorclassification uses the Kohonen model LVQ, the results weobtained using this approach are compared to the k-meansclassifier. The K-Means clustering Algorithm K-means algorithm [7] is a post-clustering technique that iswidely used in image coding and pattern recognition. A

sequence of iterations starts with some initial set )0(C . At

each iteration t set all data points Cc are assigned to one of

the clusters )( t

kS as defined in (2). A new centred )()(

tkC for

a cluster is computed as follows:

t

i

tkii

tj Scc

tc

1

)()1()(1

(1)

and

kk ccqCcS )(: (2)

)( t

kS : The quantisation mapping defines a set of clusters

The algorithm is known to converge to a local minimum. TheK-means algorithm was used to quantize images in [6]. For thetest images it produced smaller average errors

Iyx

yxyxICq cqcM ),(

),(),(),( )(1than the median cut

and variance-based pre-clustering algorithms. Unfortunately,


Supervised Color Image Segmentation, using LVQ Networks and K-means. Application : Cellular Image

9

the high cost of computation makes K-means impracticalfor image quantisation. The Learning Vector Quantisation (LVQ) This model corresponds to a layer of neurons and anentrance layer (Fig. 2).

Figure 2: Topological map

The entrance layer serves only for thepresentation of the entrance vectors(components R G B of the different pixels).

An adaptation layer formed by a neuron network.These neurons are some simple linear and areconnected to all components R,G,B of theentrance layer.

Every neuron j of the topological map calculates a distancebetween the x example presented to the entrance and itsweight vector Wj (entrance vectors x and weight vectorsW of the neurons of the map have the same dimension).The neuron j* (winner) is then the one that has the minimumdistance [9].

jj WxWx min* (3)

The Euclidean distance jd is very often used in the domainof the classification. It is calculated as follows:

2

0)(

N

iijij Wxd (4)

N is the dimension of the entrance vector, it is equal to 3(RGB components) in our case.

Therefore, we determine the neuron whose weight vectoris nearest to the sense of the Euclidean distance, of thepresented vector x . It then adjusts its weight in order tocome closer to the example presented again. In itstopological version, every winning neuron incites theseneighbors to also modify their weight in the same sense. Ifone notes j* the winning neuron, Vj* its neighborhoodand Wj the vector of weight of a j neuron, modificationsthat are going to take place after presentation of the vectorof x entrance to the iteration at time “t” are going to be: The LVQ1

Assume that a number of ‘codebook vectors’ W (freeparameter vectors) are placed into the input space toapproximate various domains of the input vector x by theirquantized values. Usually several codebook vectors areassigned to each class of x values, and x is then assumedto belong to the same class to which the nearest W belongs.Let

Wxc minargdefine the nearest W to x, denoted by Wc. Values for theW that approximately minimize the misclassification errorsin the above nearest-neighbor classification can be foundas asymptotic values in the following learning process.Let x(t) be a sample of input and let the W representsequences of the W in the discrete-time domain. Startingwith properly defined initial values, the following equationsdefine the basic LVQ1 process [8]:

)()()()1(

)()()()1(

kjj

kii

WXWW

WXWW

(5)

)(t is a gain term )10( )( t that decreases intime.

Algorithm: [10]

Step 1: Initialize Weights from N inputs to M output-nodesshown in Fig. 2 to small random values.

Step 2: Present New Input

Step 3: Compute Distance to all Nodes Compute distancesdj between the input and each output node j using


M. MESSADI

21

0

)()( )(

N

i

tij

tij Wxd

where )(tix is the input to node i at time t and Wij is the

weight from input node i to output node j at time t.

Step 4: Select output Node With Minimum Distance.Select node j* as that output node with minimum distancedj.

Step 5: Update Weights to Node j* and NeighborsWeights are updated for node j* and all nodes in the

neighborhood defined by )(*t

iNE . New weights arechanged by supervised update called LVQ1.

)()()()1(

)()()()1(

kjj

kii

WXWW

WXWW

Step 6: Repeat by Going to Step 2 The data file dedicated to the training includes threecolumns of the RGB (Figure 3) components and a fourthcolumn. In this fourth column the label of the colorcorrespondsto the RGB and is given (Figure 4). Note thata preliminary manual operation must be done by an operatorwhere he must select maps to study zones correspondingto different colors (brown, black, green, blue, etc.) in orderto construct a data file prototype.

25178146271961723911

BGR

x

Figure 3: Example of input image vector In RGB space

greenyellowblack

25178146271961723911

W

Figure 4: example of vector weight

APPLICATION In this section, two experiments are carried out anddiscussed to demonstrate the performance of the differentclustering algorithms LVQ and K-means. The images usedare RGB color images of 100x100 pixels. In the LVQ1, theexperiment related to the dependence of the method oninitial conditions is investigated. Several strategies arepossible to obtain an initial set of template colors forstarting a clustering algorithm. An obvious choice is arandom initial set. Applying the quantised on differentinitial sets independently, allows one to study in a statisticalway the influence of the initial conditions on the behaviourof the algorithm. A statistically representative number ofinitial sets is constructed and the algorithm is applied oneach set independently. The distribution obtained in Figure5 is shown for quantisation of color image ‘test’, quantisedto 4 colors, after 10000 independent runs, using randominitial conditions, after K-means.

In Figure 6, a few discrete local optima are clearly visible,which indicates that K-means converges to a localoptimum. However, the distance between different localoptima is large. Finally in Figure 6, a narrow distribution onthe left hand side demonstrates that the classificationapproach, which is independent of initial conditions,converges to a solution near the real global optimum. Thedistribution shows the dependence of the competitivelearning algorithm on the learning order of the presentationof the color pixels. The effect of this dependence is only afew percent. In the second experiment, the algorithms K-means is compared to LVQ1 when fixed initial conditionsare applied.



11

LVQ1 K-mean LVQ1 K-mean LVQ1 K-mean LVQ1 K-meanBrown 715 712 718 712 715 99.58% 100% Correct

total rateGreen 8145 8127 8123 8127 8123 99.78% 99.73% 99.38%Bleu 448 458 461 448 430 100% 95.98% Harm total

rate Harm

total rateMauve 692 703 698 692 670 100% 96.82% 0.21% 0.62%

Correct total rate 99.79%

A number of pixels of the

weights

A number of pixels after classification

Recognition pixels

Rate of Recognition by colors

Total rate

Figure 5: LVQ1 classification

Figure 6: Classification by K-means with 4 weights

CLASSIFICATION OF CELLULAR IMAGE This article discussed an effective algorithm for cellsegmentation and showed its integration that supportsdecision-making in clinical pathology. The nonparametricnature of the segmentation and its robustness to noiseallowed the use of a fixed resolution for the processing ofhundreds of digital specimens captured under differentconditions. The classification has been indirectly evaluatedwhich demonstrated satisfactory overall performance. Asa broader conclusion, however, this research proved thatthe segmentation, although a very difficult task in itsgeneral form.

With an aim of testing the robustness of our algorithm, weapplied it to a cellular image. This image is coded on 8 bits

(Figure 7). Each pixel of the image is represented by itsthree dimensions components RGB. The construction ofthe training file is carried out in the following way: Thefirst stage consists in specifying the number of colorpresent in the image. Then, for each weights color, oneselects, using the mouse, a small area comprising of thepixels having similar points of colors. The last stageconsists in giving labels to each color introduced into thetraining file. In order to have a good illustration of theprojection of these observations on a of Kohonen map,we used a map of size 200 neurons, and at the time of thephase of training, the observations are presentedsequentially one by one at the entry of the random network.

The images which we have to process are thus color images.On these images cells are present; these must be extracted.


M. MESSADI

We should insulate at the same time their cytoplasm andtheir core. Two bits of information must bring to recognitionthe various cellular types. Before starting the first stage ofsegmentation, we must precisely know the nature and thecontext of the images. Our images are color imagespresenting the cells coming from cytology from cereusesand colored by the international coloring standard. Thecells have a brown core and a yellow cytoplasm except forthe red blood cells. The strategy of segmentation we adoptis an ascending strategy. We extract the core and cytoplasmof the cells at the same time, and then we extract the redblood cells [11].

Segmentation is realised using colors. Classification isachieved with a supervised self-organizing LVQ1 and byusing the operators of edge detection (canny, sobel) tofinalize the work.

Figure 7: Classification of cellular image

CONCLUSION Few traditional neural network algorithms have been meantto directly operate on raw data such as pixels of an imageor samples of speech waveforms picked up from the timedomain. Most pattern recognition tasks are preceded by apre-processing transformation that extracts invariantfeatures from the raw data such as spectral components ofacoustical signals or elements of co-occurrence matricesof pixels. Selection of a proper pre-processingtransformation for a particular task usually requires carefulconsideration and no general rules can be given here. It iscautioned that if this LVQ is used for benchmarking againstother methods a proper pre-processing should always beused. In performing statistical experiments a separate dataset for training and another separate data set for testingmust be used. If the number of required learning steps is



13

bigger than the number of training samples available, thesamples must be used reiteratively in training either in acyclical or in a randomly sampled order. In this work, we proposed automatic approach ofclassification of the colored images, based on theassociation of a Kohonen map to an algorithm of supervisedtraining LVQ. This approach consists at first step torepresent the samples of observations representative ofthe pixels of an image in space 3d of the colors components.The next step is the training phase, the weights vectorscorresponding to the extracted modal areas taken asprototypes of the classes present in the image, and areused for the assignment of each pixel of the image to oneof the classes identified at the time of the phase ofclassification. This approach shows, that in a supervisedcontext, the Kohonen map allows a good automaticclassification of the color image.

REFERENCES [1] P. Scheunders. A comparison of clustering algorithms

applied to color image quantisation, Vision Lab, Dept.of Physics, RUCA University of Antwerp,

[2] Kohonen T. Self-Organizing Maps, Springer Series inInformation Sciences, 30, Springer-Verlag, New York,1995.

[3] Kotropoulos C., E. Augé and I. Pitas. Two-layerlearning vector quantizer for color image quantisation.In: Eds., Signal Processing IV: Theories andApplications, Elsevier, 1177-1180, 1992.

[4] N. Ebi, B. Lauterbach, and W. Anheier, An imageanalysis system for automatic data acquisition fromcolored scanned maps, Machine Vision andApplication, pp 148-164, 1994.

[5] Dekker A.H. Kohonen neural networks for optimalcolor quantisation. Network: Computation in NeuralSystems, 5 351-367, 1994.

[6] S. J. Wan, P Prusinkewicz, and S. K. M. Wong. Variance-based color image quantisation for frame buffer display.Color Research and Application, February 1990

[7] Y. Linde, A. Buzo, and R. M. Gray. An algorithm forvector quantizer design. IEEE Transactions onCommunication, January 1980.

[8] Teuvo Kohonen LVQ PAK The Learning VectorQuantisation Helsinki University of TechnologyLaboratory of Computer and Information ScienceRakentajanaukio 2 C, SF-02150, 1995.

[9] Abdelhafid Bessaid, Hassane Bechar, M. Karim Fellah,“Image analysis and pattern recognition as tools inmap interpretation”, Electronic Journal «TechnicalAcoustics», published ,2003.

[10] Richard P. Lippman, An Intoduction to Computingwith Neural Nets, IEEE ASSP Magazine April 1987.

[11] O. Lezoray, A. Elmoataz, H. Cardot and M. Revenu,A.R.C.T.I.C, Un système automatique de Tri Cellulairepar Analyse d’Images. Laboratoire Universitaire desSciences Appliquées de Cherbourg E.I.C, Octeville,France, 1999 ;

1 4AJST, Vol. 12, No. 1: October, 2012


ALTERNATIVE CONTROL OF NANOPARTICLES DISPERSITYIN HIGH-TEMPERATURE FLOW REACTORS

1Moropeng M. L.1 and Kolesnikov A.2

1Tshwane University of Technology, Faculty of Engineering and the Build Environment,Department of Chemical and Metallurgy Engineering,

Private Bag X 680, Pretoria 0001, South Africa

Email: [email protected]

ABSTRACT: The 1-dimentional model of aerosol process which includes a hot aerosol streamflowing through a tube with thermal gradients between the aerosol stream and the reactor cooledwalls was developed to predict the aerosol formation, growth and thermophoretic deposition inhigh-temperature reactors. The mass and energy conservation equations were solved to determinethe concentration and temperature profiles of the components. The model includes particle formationby nucleation, growth by coagulation, Brownian diffusion as well as the loss of aerosol particles bythermophoretic deposition on the cold reactor walls. The developed model results in the system ofordinary differential equations which were solved in SCILAB software.

Keywords: Thermophoretic deposition, Coagulation, Nucleation, Modeling

INTRODUCTION

Nanoparticles are at the core of nanotechnology. Theseare particles ranging in size from 1 millionth to 100millionths part of a millimeter, more than 1,000 times smallerthan the diameter of a hair. In this order of magnitude, it isnot only the chemical composition but also the size andthe shape of the particles that determine their properties.Measurements in gas-phase reactors are quite problematicas time scales are extremely small, temperatures very highand the gaseous atmosphere is often aggressive.Therefore, process simulation is a useful tool and cansignificantly improve the general understanding of particleformation and moreover can support product and processoptimization. It is crucial to understand the behavior of fine particles inorder to control them. Transport of fine particles from fluidstream to reactor surface is important in predicting therate of wall deposition and in understanding mechanismsthat lead to particle removal.

It is known that thermophoretic transport is one of themany methods which causes smaller particles to depositon the nearest surfaces.

Thermophoresis is of practical importance in many engi-neering applications such as thermal precipitators, the dis-tr ibution of soot in combustion systems andthermophoretic deposition of particulate matter onto wallsof piping systems. It is the phenomenon where very smallaerosol particles experience a net thermophoretic force whensuspended in a gas in which a temperature gradient ispresent. This force results from an imbalance in momentumtransfer associated with molecular collisions between thehot and cold sides of the particles. Therefore, this forcetends to drive the particles in the direction of negativetemperature gradient.

Thermophoresis has both negative and positive effects inapplication areas. Negative effects of thermophoresisinclude reduction of thermal conductivity of heat exchangerpipes and reduction of production yield of specialtypowders manufactured in high temperature aerosolreactors. On the other hand, the concept of thermophoresisprovides a working principle to fabricate optical fibre in amodified chemical vapor deposition (MCVD) process. Italso can be employed to remove or sample atmosphericparticles from the air in a thermal precipitator.


Alternative Control of Nanoparticles Dispersity in High-temperature Flow Reactors

15

Nomenclature

Fundamental research in to this phenomena has been hasbeen reviewed by a number of authors including Talbot,Cheng, Schefer, and Willis (1980), Bakanov (1995), Li andDavis (1995a,b), and Lee and Kim (2001). Typicalcharacteristics of processes of thermophoresis include ahot aerosol stream flowing through a tube or an annulus,and the presence of a non-negligible thermal gradientbetween the aerosol stream and the cooled walls of the

tube or of an outer tube of the annulus. Accordingly, manythermophoresis studies have targeted these geometries. Thermophoretic deposition of particles in an annular flowwas studied theoretically by Weinberg (1983) and Fiebig,Hilgenstock, and Riemann (1988). Weinberg (1983)suggested that complete collection was possible withthermophoresis and that a smaller separation distancebetween concentric cylinders resulted in higher depositionefficiency. Fiebig et al. (1988) showed that when the annuluswas oriented vertically, as a result of the buoyancy effect,the deposition efficiency tended to increase for a smallerratio of inner to outer tube radius. Chang, Ranade, andGentry (1992, 1995) carried out experiments and numericalsimulations to quantify thermophoretic deposition in anannular flow system with fixed thermal gradients betweentwo concentric cylinders. They found good agreementbetween experimental results and computational resultsusing the model of Talbot et al. (1980). Lee and Kim (2001)studied thermophoretic deposition experimentally andnumerically in an annular flow system using several modelssuggested by Derjaguin, Ravinovich, Storozhilova, andShcherbina (1976) and Talbot et al. (1980) in a cryogenictemperature range. They found that the thermophoreticmodels required modification in the cryogenic temperaturerange. A tube flow with a thermal gradient has been utilizedin many applications including heat exchanger pipes andautomobile exhaust pipes. Therefore, it is necessary to study thermophoresis in atube flow in order to understand and innovate a number ofsystems that are employed in a variety of applications.The specific details of the problem we are treating areassumed to be as follows: the gas enters with an initialparticle concentration and volume, with the maximumtemperature of the fluid, and flows through the reactorwith a wall temperature also equal to the fluid temperatureat the reactor entrance. At some distance far enoughdownstream such that the laminar incompressible flow isfully developed, the wall temperature decreases to aminimum temperature and remains there. Convection,Brownian diffusion, and thermophoresis are the mainmechanisms involved in such systems. The goal of the analysis is to develop a 1-dimensionalmathematical model of aerosol dynamics to gain insightinto the details of particle growth and formation, as well asto investigate the effect of thermophoretic wall depositionof the particle size at the outlet of the reactor.

A Total particle area concentration (cm2/cm3)c Monomer Particle velocity, (m/s)C Cooling gradient, (K/m)

Ci TiCl4 concentration (mol/cm3)Cc Slip correction factor, dimensionlessD Particle diffusivity, (m2/s)Dse Diffusion coefficient, (cm2/s)dp Particle diameter, (m)la Mean free path, (m)

I Nucleation rate (# cm3/s)k TiCl4 overall oxidation rate constant (1/s)kg TiCl4 gas phase reaction rate constant (1/s)ks TiCl4 surface reaction rate constant (cm/s)kB Boltzman constant, (gcm2s-2K-1) Kg1 Gas thermal ConductivityKng Knudsen number of the gasKp1 Particle thermal ConductivityKth Thermophoretic coefficient NAvo Avogadro’s number (1/mol)N Particle number concentration (# 1/cm3)PSD Particle size distributionRi Universal gas constant, (Kgm2s-2K-1mol-1) R Radius of reactor, (m) T Absolute Temperature, (K)Tw Wall temperature, (K)Tref Reference temperatureUth Thermophoretic velocity, (m/s)V Particle volume concentration (# 1/cm3)z Axial direction, (m)

Greek letters

Collision frequency for TiO2 particles (cm3/s)Dynamic viscosity, (Pa.s)Gas density,(g/cm3)

µg

p


M. L. MOROPENG

THEORETICAL APPROACH

Reaction model The formation of TiO2 takes place by the overall reactionof TiCl4 with O2:

2224 2ClTiOOTiCl (1)

The depletion of TiCl4 occurs by both homogeneous gasphase reaction and by the reaction at the surface of existingTiO2 particles:

isgii CAkkCk

dtdC )(

(2)

where Ci (mol/cm3) is the concentration of TiCl4, t(s) is theresidence time, A (cm 2/cm3) is the surface areaconcentration of TiO2 particles, k (1/s) is the overalloxidation rate constant of TiCl4 (Pratsinis et al., 1999):

Teks

80993exp039,4 (3)

While kg (1/s) is the gas phase reaction rate constant, T(K) is the process temperature and ks= (cm/s) is the surfacereaction rate constant (Pratsinis, & Spicer, 1998):

Teks

80993exp039,4 (4)

Monodisperse Model

The computational scheme simulates homogeneousnucleation and coagulation/ coalescence as well as aerosoltransport by diffusion and thermophoresis.Spherical

particles are assumed, 31

6

NVdp the overall process

is represented by the general dynamic equation for aerosolparticles (e.g. Friedlander, 2000). For the differential particlenumber concentration N=N (z, dp, t), where z denotes thedirection, t time and dp particle diameter, the generaldynamic equation can be written as:

COAGNUCLE

SEth

dtdN

dtdN

dtdND

dtd

vNU

dtd

vdtdN

11

(5)

Here Uth is the particle drift velocity due to thermophoresis,and DSE is the particle diffusion coefficient. The left-handside of the equation has terms that refer to temporal,convective and diffusive rates of change in the differentialparticle number concentration. The right-hand side hassource terms for aerosol dynamics due to nucleation, andcoagulation. Brownian diffusion The variation of particle number concentration with timecan be determined by solving the one-dimensional equationof diffusion, given as follows:

2

2

dzNdD

dtdN

SE For axial direction (6)

The particle diffusion coefficient is given by the StokesEinstein relation (e.g. Hinds, 1999):

dpCcTkBD

gSE

3 (7)

gKng eKnCc

1.1

4.0257.11 (8)

Thermophoretic velocity The thermophoretic particle drift velocity is modeled withTalbot equation (Talbot et al., 1980) and is given by:

TTK

Uth gth

with T in radial direction (9)

The expression for temperature as a function of time isobtained by fitting the polynomial function with thecalculated dimensionless temperature difference, fromequation (10)



17

12

3PrRe

Rxe

RrRC n

nnn (10)

In the above equation, the value of , nC , n ,

RrRn

and n are dimensionless temperature, the coefficient of theGraetz equation, the eigenvalues, eigenfunction, and thenumber of terms respectively.

wg

w

TTTzrT

),(

(11)

wwg TTTzrT ),( (11a)

According to Eq (11a), the temperature profile ),( rzTcan be calculated.

)3/1(84606.2)1( nn

nC (12)

The eigenvalues, 384 nn (13)

,........4,3,2,1n

The thermophoretic coefficient (Kth) for a spherical particleapplicable for all flow regimes from free molecular tocontinuum regimes is given by:

gp

gg

gp

g

th

Knkk

Kn

Knkk

CcK4.421*483.31

2.2

*294.2

1

1

1

1

(14)

Nucleation kinetics

AvoiNUCL

NCkIdtdN

(15)

The change of the number concentration N is proportionalto the nucleation rate I. Nucleation rate depends on therate of chemical reaction of TiCl4 oxidation.

2.2.4 Coagulation This coagulation process leads to substantial changes inparticle size distribution with time. Simple defining equationfor coagulation is given by equation (16)

2

21 N

dtdN

COAG

(16)

where is the collision frequency function of equallysized particles from free molecule to continuum particlesize regime (Hidy, G.M. (1984)):

1

242

8

pp

pp cd

Dgd

dDd (17)

with the particle diameter, velocity and diffusivity, dp, cand D, respectively, while the parameters

papapap

dIdIdId

g

2

3223

31

(18)

2

32

8518645

3 nn

nnn

pg

B

KKKKK

vTkD

(19)

with the mean free path for the particles

cDI a

8 (20)

and

pP

B

vTkc

8

(21)

The temperature along the reactor axis is a function ofdistance.

)(zfT , (22)

where )(zfT is a function approximating the axialtemperature distribution in the reactor. The temperaturedependant terms are recalculated for the variable coolinggradient, C; the temperature is given as


M. L. MOROPENG

zCTT 0 (23)

Equations (1) – (23) are solved simultaneously using thesoftware SCILAB.


The change of temperature within the reactor is shown inFig.1 (A). It can be seen that the temperature at the reactorentrance is high, and lower as it approaches the reactorexit. The change in temperature is caused by the heattransfer between the hot fluid and the cold reactor wall.Fig.1 (B) shows how the concentration of precursor varieswith the temperature. The change in the initial temperaturevalue will affect the radial temperature gradient inside thereactor. Increasing the initial temperature value will resultin more of TiCl4 converted, and lead to the increase in therate of nucleation and thus decrease the particle numberconcentration and also lower the particle size whereas theparticle surface area increases.

Typical simulation parameters ValuesT (K) 1000P ( Pa) 1.013 e+5R i 8.314N Avo (1/mol) 6.02E+23 4200 (kg/m3) P*Mw_gas/(R*T)

( kg/m3) kB Ri/6.022e+23 m (TiO 2 ) ( kg) 1.33E-25v (TiO 2 ) (m3) 3.16E-29

s (TiO 2 ) (m2 ) 1.16E-19d (TiO 2 ) (m) 3.93E-10Q rt (m3/s) 2/(1000*60) T rt (K) 273Q (m3/s) Qrt.*T/Trt

dg (O 2 ) ( m) 7.98E-09

p

g

Table 1: Simulation conditionsThe degree of thermophoretic deposition depends directlyon the level of cooling applied. Thermophoretic effect wastested by changing the cooling gradient in the axialdirection of the reactor by calculating the radial temperaturegradient while maintaining the initial temperature valueconstant. When the radial temperature gradient is greaterthan the axial temperature gradient, particles are moved tothe wall dew to higher radial gradient in comparison withaxial gradient. The axial temperature distribution isexpressed by Eq(23). The axial temperature gradient isexpressed by coefficient C in Eq(23). When axialtemperature changes, automatically the radial temperaturegradient changes through Eq (11), which representsdimensionless temperature difference between axialtemperature and wall temperature. Fig.1 - Fig.3 show theparticle behavior along the reactor axis with varying axialcooling gradient, 700, 1000, and 1200 K/m, respectively.Fig.1 (C&D) shows that the particle number concentrationat the reactor entrance is very low and becomes highdownstream, which later decreases near the reactor wall.This explains that particle number concentration initiallyrises due to chemical reaction and when the raw material(TiCl4) is depleted, coagulation causes particle numberconcentration to decrease.

In Fig.1 (D), the run without thermophoresis showed thatBrownian diffusion is dominant. The deposition rate dueto thermophoresis decreases with increase in time. Thiscan be due to the decrease in the magnitude ofthermophoresis as the particle increase. Particle losses tothe wall become much higher when the cooling gradient ofthe aerosol flow is decreased. As Fig.2 (C) and Fig.3indicate, when the cooling gradient of the aerosol flow isdecreased, the loss of ultrafine particles becomes higherdue to an increase in the thermophoretic velocity whichcauses the particles to move faster and deposit on the wallof the reactor. Fig.2 (B) shows the effect of temperature onthermophoretic velocity. Thermophoretic velocity isdirectly proportional to the inverse of absolute temperature,i.e. when the temperature increases, the negative thermalgradient increases, and so the increase in thermophoreticvelocity is favored. Fig.2 (A) shows how the system is affected at lower coolinggradients. We see that the average primary diameter ofTiO2 particles is reduced as the cooling gradient falls.



19

Axial Temperature distribution

0.0E+00

5.0E+021.0E+03

1.5E+032.0E+03

2.5E+03

0.0 0.5 1.0 1.5 2.0

z (m)

Tem

pera

ture

( K)

C= 700 K/m C = 1000 K/m C=1200 K/m

Figure 1(A): Temperature profile inside the reactor through the variation of cooling gradients.

TiCl4 Concentration

0.E+005.E-091.E-082.E-082.E-083.E-083.E-08

0.0 0.5 1.0 1.5 2.0z (m)

[TiC

l 4] m

ol/c

m3

C=700 K/m C=1000K/M C=1200 K/m

Figure 1(B): Increasing the initial temperature value will result in more of TiCl 4 converted, and lead to the increase inthe rate of nucleation with no effect of temperature gradient on the conversion of titanium chloride to form titania.

Particle Number Concentration

0.0E+00

2.0E+13

4.0E+13

6.0E+13

8.0E+13

1.0E+14

1.2E+14

0.00 0.02 0.04 0.06 0.08 0.10 0.12

Time (s)

Num

ber/c

m3

No deposition" Thermophoretic deposition

Figure 1 (C): The comparison between the particle number concentration with and without thermophoretic deposition


M. L. MOROPENG

Particle Average Diameter

0102030405060

0.0 0.5 1.0 1.5 2.0

z (m)

Parti

cle

size

(nm

)

C=700 K/m C=1000 K/m C=1200K/m

Particle Number Concentration

0.00E+00

2.00E+13

4.00E+13

6.00E+13

8.00E+13

1.00E+14

1.20E+14

0.00 0.02 0.04 0.06 0.08 0.10 0.12

Time (s)

num

ber/c

m3

C=1000K/M C=1200K/M C=700K/m

Figure 1(D): The effect of particle diameter on particle number concentration by varying the cooling gradients. Particlenumber concentration initially rises due to chemical reaction and later decreases when coagulation takes over

Figure 2 (A): The effect of axial distance on particle size by varying the cooling gradients.The large sized particles are observed on the lesser axial distance with a large cooling gradient

Thermophoretic Velocity

0.E+00

2.E-06

4.E-06

6.E-06

8.E-06

0 500 1000 1500 2000 2500

Temperature (K)

Uth

(m/s

)

C=700 K/m C=1000K/m C=1200 K/m

Figure 2 (B): The effect of temperature on thermophoretic velocity with the variation of cooling gradient



21

Thermophoretic Velocity

0.E+00

2.E-06

4.E-06

6.E-06

8.E-06

0 10 20 30 40 50 60

paericle size(nm)

Uth

(m/s

)

C=700K/m C=1000K/m C=1200K/m

Figure 2 (C): The effect of particle size on deposition velocity, when the initial temperature value is kept constant at varying cooling gradients

Deposition Flux

0.E+00

2.E+10

4.E+10

6.E+10

8.E+10

0.0 0.2 0.4 0.6 0.8 1.0

z (m)

Depo

sitio

n Fl

ux

(Kg/

cm2.

s)

C=700K/m C=1000K/m C=1200K/m

Figure 3(A): Effect of particle size on the deposition flux.

Deposition Flux

0.00E+00

2.00E+10

4.00E+10

6.00E+10

8.00E+10

0 5 10 15 20

Particle size(nm)

Flux

(Kg/

cm2.

s

C=700K/m C=1000K/m C=1200K/m

Figure 3(B): Effect of axial distance on the deposition flux


M. L. MOROPENG

The effect of particle size on deposition velocity, when theinitial temperature value is kept constant and varying thecooling gradient, is depicted in Fig.2 (C). Thermophoreticvelocity decreases as the particle size increases. With the assumption made, it can be seen from the graphsthat thermophoretic force does not play some major role inthe deposition of fine particles to the wall. However, thereal high temperature reactors will also operate in turbulentmode and therefore particle wall deposition due to turbulentflows near the wall will play a role in the process.

CONCLUSION The simulations were run using the conditions in Table 1.The simulation results obtained from the one-dimensionalmodel provides useful information of the temperature andcooling gradient effects on the average titania particlediameter, particle concentration and thermophoreticvelocity of particles at reasonable computational time. Theeffect of cooling gradient on thermophoretic velocity wasclearly studied, ranging from 700 to 1200 K/m.

Based on the final results, the following conclusions havebeen made: 1. Thermophoretic velocity increases with increase in

the cooling gradient and the distance of deposition,but decreases with increase in particle size.

2. Thermophoretic velocity affects the deposition fluxand particle number concentration directly. From theresults given, we have seen that particle numberconcentration and deposition flux does not changewith particle size when varying the cooling gradient.This results when the increase in the cooling gradientcauses the absolute temperature to increasesimultaneously and offset the process. This showsthat the thermophoretic velocity, deposition flux,particle number concentration and average particlesize are implicit functions of temperature.

3. The effect of temperature on thermophoretic velocityis independent of the cooling gradient.Thermophoretic velocity increases with increase intemperature regardless of variation of the coolinggradient.

4. Thermophoretic velocity is directly proportional tothe inverse of absolute temperature, i.e. when thetemperature increases, the negative thermal gradientalso increases, and so with thermophoretic velocity.

REFERENCES Barret J.C. & Webb N.A. (1998). A comparison of some

approximate methods for solving the aerosoldynamic equation. Journal of Aerosol Science 29, 31-39

Bird, R.B., Stewart, W.E.,& Lightfoot, E.N. (2002). Transportphenomena. 2nd edition. New York: Wiley

Chae at el, (1999). Chemical vapor deposition reactordesign using small-scale diagnostic experimentssimulation . Journal of the ElectrochemicalSociety,146,1780-1788

Chimera Technologies Inc. (2006). Fine particle modeling[online] Available from: http://www.aerosolm o d e l i n g . c o m / F r o n t / M a i n . c f m ?App=AerosolModeling & DataSource_= AerosolModeling &Co _ID_Filter=0[Accessed: 4/05/

Desilets, M, Bilodeau, J.F. & Proulx, P. (1997). Modeling ofthe reactive synthesis of ultra-fine powders in athermal plasma reactor. Journal of Physics D: AppliedPhysics, 30, 1951-1960.

Friedlander, S.K. (2000). Smoke, Dust, and Haze:Fundamentals of aerosol dynamics.2nd edition.New York. Oxford, Oxford University Press

Hidy, G..M. (1984). Aerosols: An industrial andenvironmental science. United Kingdom Ed. London,Academic Press, INC.

Kim, Y.P. & Seinfeld, J.H. (1990). Simulation ofmulticomponent aerosol dynamics by the movingsectional method. Journal of Aerosol Dynamics

Pratsinis et al (1998). Competition between gas phase andsurface oxidation of titanium chloride duringsynthesis of titania particle. Journal of ChemicalEngineering Science.

Tsantilis S. & Pratsinis, S.E. (2004). Narrowing the sizedistribution of aerosol-made titania by surfacegrowth and coagulation. Journal of Aerosol Science,35, 405-420.

Talbot et al (1980). Study on thermophoretic deposition ofaerosol particles in laminar and turbulent tube flows.Institute of Environmental Engineering, Chiao TungUniversity, Hsinchu, Taiwan.

2 3 AJST, Vol. 12, No. 1: October, 2012


EFFECTS OF GROUND INSULATION AND GREENHOUSE MICROENVIRONMENT ON THE RATE AND QUALITY OF

BIOGAS PRODUCTION

1Cheruiyot, P.K., 2Ndiema, C.K.W., 1Chemelil, M.C and 1Wambua, R. M.

1Department of Agricultural Engineering, Egerton University, Njoro, Kenya.2Department of Industrial and Energy Engineering, Egerton University, Njoro, Kenya.


ABSTRACT: A study was conducted at Egerton University, Njoro, Kenya to establish the potentialof plastic digester to produce biogas under natural and greenhouse microenvironment. The specificobjectives were to evaluate the effects of greenhouse and ground insulation on the rate and qualityof biogas generation. A greenhouse measuring 6m long, 4m wide and 2m high was constructed.Inside the greenhouse and the outside environment, three replications of thirty (30)-litre plasticbiogas digester filled to two third capacity with slurry were used. The digesters were partiallyexposed to the environment and when fully buried in the ground. Biogas yields averaged 90.3 and63.0 litres per kilogramme (l/kg) of volatile solids added for partially buried digesters undergreenhouse and natural conditions, respectively. The corresponding digester temperatures averaged27.5 and 22.2oC. The respective biogas yields averaged 312.8 and 226 litres per kilogramme volatilesolid added, while the temperatures averaged 27.9 and 24.1oC for fully buried digesters. The averagemethane content in the biogas was 61.5% and 56.4% under greenhouse and natural conditions,respectively. At the 0.05 significance level, greenhouse effect was found to enhance both the quantityand quality of biogas generation from dairy cattle dung. The effects of ground insulation had a farmuch effect on the quantity of biogas generation as compared to the effects of greenhouse conditions.Therefore ground insulation of plastic biogas digester under greenhouse conditions significantlyenhances biogas generation.

Key words: Anaerobic conditions, greenhouse, natural conditions, ground insulation, greenhouseeffect

INTRODUCTION

Most of rural populations depend on woodfuel as sourceof energy for cooking and most of these families havelittle light available at night (FARMESA, 1996). For thesereasons there has been an increasing interest in the use ofbiogas systems in rural areas. Biogas technology utilizes a wide variety of organic feed-stock such as animal wastes, night soil, agricultural resi-dues, aquatic plants and organic industrial wastes. Themajor constituents of biogas are the methane (CH4) gasand carbon dioxide (CO2) with traces of hydrogen (H2)and hydrogen sulphide (H2S). Biogas burns well whenthe relative proportions of methane to other gases are

more than 50%. It can therefore be used as a substitute forkerosene, charcoal or firewood for cooking and lighting.The digested sludge is also a good soil stabilizer to im-prove land productivity. Large scale biogas digesters notably the Chinese domeand Indian floating gasholder types have been promotedin the East African region over the years with varying de-gree of success. The main constraint to wide adoption ofthese digesters is the high cost, which makes the technol-ogy beyond reach of many smallholder farmers. In theearly 1980s a low cost biogas digester, using plastic sleeves,was developed in Colombia to meet the economic con-cerns of rural farmers (FAO, 1992). The technology waswidely adopted in Colombia and Vietnam and efforts to


P. K. CHERUIYOT

promote these systems in Tanzania, Kenya and Ugandashowed promising results (FARMESA, 1996). Seasonaland diurnal temperature variations are deleterious to meth-ane gas production and for this reason plastic digesterscannot work effectively in the highlands and other coolerareas (Lekule, 1996). Therefore this study was conductedto determine the performance of plastic biogas digesterunder natural and greenhouse conditions. The specificobjectives were to: 1) evaluate the effect of greenhouse onthe rate and quality of biogas production and; 2) evaluatethe effect of ground insulation on biogas production.

MATERIALS AND METHODS

The study was conducted with the digesters partiallyburied and when fully buried in the ground. A total of fiveexperiments were set. The first four experiments had theexperimental digesters partially buried underground whilein the fifth experiment the digesters were fully buried. Themean values of temperature and the corresponding gasyields were arranged in a two-stage ‘nested’ or hierarchicalexperimental design for analysis. Greenhouse and naturalconditions were considered as factor A the temperaturemeasurements as factor B. The gas yields formed theresponses. There were a levels of factor A, b levels offactor B nested under each level of A, and n replicates.This is a balanced two-stage nested design, since therewere equal numbers of levels of B within each level of A,and equal number of replicates. Since every level of factorB did not appear with every level factor A, there was nointeraction between A and B as demonstrated byMontgomery (1976), Ott (1988) and Montgomery et al.,(1998). A hemispherical greenhouse measuring 6m long, 4m wideand 2m high was constructed using 36m2 of transparentpolyethylene sheet, 5 pieces of galvanized steel framesand pieces of timber to reinforce the structure. Inside thegreenhouse and the outside environment three replicationsof 30-litre batch feed plastic biogas digesters were set. Sampling port was made at one end of each test digester.A probe was inserted through this port and a thermocouplewire sensor was placed in such a way that its tip rested inthe middle of the digester. The sensors were used fortaking sludge temperature readings. A hole of 1cm diameterwas made, on each digester, at approximately 10cm fromthe digester inlet end. PVC and rubber washers of 10cmdiameter with 21mm central hole were cut and fitted on theflange of the male adapters. These adapters were thenthreaded through the said hole from the inside of digestersto the outside. A second PVC washer and rubber washerwere put on the male adapter from the outside of the hole

and secured tightly with female section. A gas outlet valvewas inserted and secured into the same female section.Finally a flexible plastic hosepipe with 21mm internaldiameter, for carrying gas, was attached to the gas valve. Each experimental digester was fed through the inlet with14 kg of fresh dairy cow dung thoroughly mixed with tapwater to bring the weight of slurry to 20 kg. Mixing ofdung and water was provided by a concrete mixer rotatedby hand. The loading rate was determined using theprocedure described by FARMESA (1996). The inlets weresecurely sealed and the complete assemblies of digesterscarried and placed in trenches, which were dug undergreenhouse (GH) and natural conditions (NC) toaccommodate them. Provision for the agitation of thedigester contents during the digestion process was notmade because its effect on small-scale digesters isconsidered minimal (Barnett et al., 1978). Emptying ofsludge was done through the inlet opening after elapse ofa given hydraulic retention time.

Samples of the influent and effluent slurry from eachexperimental digester were taken for laboratorydetermination of total solids (TS), volatile solids (VS), totalnitrogen (N) and organic carbon (C). The TS and VS weredetermined by heating the sample at 105oC and 550oC,respectively. The total nitrogen was determined by thestandard micro kjeldahl method as described by APHA(1995) and total carbon by the Walkley-Black method asdescribed by Walkley and Black (1947). The daily gasyields were measured using jar displacement method andthe corresponding temperature using Delta-T loggerdevice. The methane content in the biogas was analyzedby standard GC procedures described in APHA (1995).The measurement of the TS, VS, N, and C in the cow dungslurry was limited to partially buried digesters becauseconstant trends were observed which made furthermeasurements unnecessary.

Experimental Setup

Figure 1: Experimental setup of biogas plant


Effects of ground insulation and greenhouse microenvironment on the rate and quality of biogas production

25

In a batch digester the waste is put into the plant with astarter, if available, and the gas collected as it is given off.The time in which biogas production was simply negligibleor equal for both sites was considered the hydraulicretention time. At this point the experimental digesterswere stopped and their contents discharged. New slurrywas then charged into the digesters.

Analysis of variance was run on the population means forthe variables considered during the study. This was doneusing the general linear model and Duncan’s multiple rangetest of the SAS procedure (SAS institute, 1998) at 95 percentconfidence level. An F– test was used for hypothesistesting at probability value of 5%. Scatter plots of meangas yields were plotted for the corresponding time (days)of observation and polynomial regression fitted to discernhow well the coefficient of determination (r2) explained thetrend. The correlation between sites with respect totemperature and gas yields was determined using thegeneral linear model (GLM) of the SAS procedure.

RESULTS AND DISCUSSION Influent Slurry

The percent compositions of influent slurry are shown inTable 1. Table 1: Percent composition of feedstock in the influentslurry (a) Partially buried digesters

Site Sludge TS VS N CNC Influent 10.37a 8.13a 1.84a 30.44aGH Influent 10.42a 8.11 1.84a 30.50a

For a given site, values in the vertical column followed by thesame letter are not differentstatistically ( = 0.05) according to Duncan’s multiple rangetest.

(b) Fully buried digesters

Site Sludge TS VS

NC Influent 11.25a 8.78aGH Influent 11.25a 8.78a

The influent TS content of dairy cow dung was 10.37%and 10.42% under NC and GH, respectively. The VS in theinfluent was 78% of total solids (%TS). The average

measured percent N in the influent slurry, which was takenon a dry weight basis, was 1.84 while the measured average%C in the influent slurry was 30.5 and 30.4 under GH andNC, respectively. The computed values of carbon tonitrogen (C/N) ratio in the influent slurry were 16.6 underGH and 16.5 under NC. It can be concluded from theseresults that all the respective parameters in the influentunder each of the conditions tested were not statisticallydifferent. This implies that identical concentrations of theinfluent slurries were achieved under each of the test site.

3.2 Effluent Slurry

The effluent compositions of feedstock are given inTable 2(a) and (b) for partially and fully buried digesters,respectively.

Table 2: Percent composition of feedstock in the effluentslurry

(a) Partially buried digester

Site Sludge TS VS N CNC Effluent 7.61a 5.69a 2.34a 24.61aGH Effluent 7.29b 5.52b 2.45a 24.24a

For a given site, values in the vertical column followed by thesame letter are not different statistically ( = 0.05) according to Duncan’s multiple rangetest.

(b) Fully buried digesters

Site Sludge TS VS

NC Effluent 7.23a 4.60aGH Effluent 6.71b 4.18b

The analysis of percent TS in the effluent sludge frompartially buried digesters yielded the results indicated inTable 2(a). The percent effluent TS was 7.61 and 7.29 underNC and GH conditions, respectively. The percent meaneffluent TS were statistically different for the two sites. Incomparison with the influent values shown in Table 1(a),the effluent values correspond to reductions in %TS of27% and 30% under NC and GH, respectively. Incomparison with the influent VS, the percent VS in effluentsludge were 5.69 under NC and 5.52 in GH for partiallyburied digesters. These represent reductions of 30% and32% of VS under NC and GH, respectively. As can be seenin Table 2 the mean effluent sludge under the twoconditions were significantly different. This observationcan be attributed to the difference in temperature and gasproduction rate between the sites as indicated in Tables 4and 5, respectively.


P. K. CHERUIYOT

Digesters that were fully buried as shown in Tables 1(b)and 2(b) achieved even higher percent reductions of 52%and 48% of VS under GH and NC, respectively. The mean%N in the effluent slurries for partially buried digesterswas 2.34 and 2.45 under NC and GH, respectively. Incomparison with the influent slurry as given in Table 1(a),it can be seen that there was increase in the effluentnitrogen content. The apparent increase in %N can beexplained by the reduction in TS in the effluent sludge.This could also show that the nitrogen content in thedigested slurry was retained and thus confirm the generalhypothesis that digested slurry has readily availablenitrogen and thefore good for growing crops. The studyalso shows an effective biodegradation of dairy wastematerials. However, it should be noted that the amount of%VS reduced would depend on the nature of waste,temperature, pretreatment, HRT employed and the rate ofgas production as found by Kalia et al. (2000), Singh et al.(1993) and Cho et al. (1995). Slurry pH and Density Table 3 shows the variation of slurry pH and density.

Table 3: Variation of Influent and Effluent pH and Density

(a) Influent slurry

Site Sludge pH DensityNC Influent 6.93a 1.03aGH Influent 6.93a 1.03a

For a given site, values in the vertical column followed by thesame letter are not differentstatistically ( = 0.05) according to Duncan’s multiple rangetest.

(b) Effluent slurry

Site Sludge pH Density

NC Effluent 7.42a 0.98aGH Effluent 7.41a 0.98a

Variation in pH

The measured pH value for influent and effluent are shownin Tables 3(a) and (b). From the Table, the influent pHvalues were often lower than effluent pH values. Thisapparent difference observed, could be explained by therole methane producing bacteria play in ensuring thatvolatile fatty acids (VFA), which are responsible for lowpH, are converted into primarily methane (CH4) and carbondioxide (CO2). In anaerobic digester a large quantity of

CO2 is produced during methane formation (Price andCheremisinoff, 1981). The high pH values obtained in theeffluent sludge, therefore, is normally maintained with abicarbonate buffer system, which is responsible forneutralizing the acid. The best range of pH values of slurryis widely quoted as 7 to 8.5. This means that if the value ofslurry varied too high or low, from this range the outputand quantity of gas would be affected. In the presentstudy the average pH of 7.4 attained under both GH andNC falls under the normal range. This implied that theconditions in the digesters were stable.

3.5 Slurry Density

As shown in Table 3(a), the density of influent slurry wasabout 1.03 grammes per cubic centimetres (g/cc). At theeffluent the density, as given in Table 3(b), averaged 0.98g/cc. This was 5% less than the value of the influent. Thedecrease in the density of slurry could be attributed to adecrease in the total solids content from about 10.4% inthe input to about 7.4% in the effluent sludge. From thelaboratory analysis, the density of dry matter was 50kgm-3.

3.6 Effects of Greenhouse on Temperature The effects of greenhouse on sludge temperature can bededuced from Table 4 (a) and (b) and Figures 3 and 6 forpartially and fully buried digesters, respectively. Table 4: Mean sludge temperatures for digesters partiallyand fully buried across the test sites (a) Partially buried digesters

Block Site Temperature (oC)

1,2,3&4 GH 27.51aNC 22.19b

For each block, values in the vertical column followed by thesame letter are not differentstatistically ( = 0.05) according to Duncan’s multiple range

(a) Fully buried digesters

Block Site Mean Temperature (oC)

5 GH 27.92aNC 24.11b

For each block, values in the vertical column followed by thesame letter are not differentstatistically ( = 0.05) according to Duncan’s multiple rangetest.



27

As indicated in Table 4 (a) and (b) the mean sludgetemperatures for digesters that were partially buried were27.51oC and 22.19oC under greenhouse and naturalconditions, respectively. Similarly the corresponding meansludge temperature values for digesters fully buried were27.92oC and 24.11oC. These values were statisticallydifferent across the test sites at =0.05 according toDuncan’s multiple range tests (DMRT) of the SASprocedure. The optimum temperature range forfermentation of between 25oC to 400C was attained undergreenhouse conditions. During the day elevated digestertemperatures of between 40oC to 600C were recorded indigesters that were partially buried. Ordinarily temperaturewithin this range would favour or enhance gas productionbecause it is within both the mesophilic and thermophilicconditions. These conditions were, however, notsustainable or adapted throughout the day. It emergedthat at night and early morning temperatures dropped tobelow 20oC resulting in sudden high thermal variations.Methane producing bacteria are known to be sensitive tosudden temperature fluctuations (Fulford, 1988 and FAO1992). They (methane bacteria) become inactive or stopworking when such conditions develops. For optimumprocess stability, therefore, temperature should bemaintained within a narrow range of operating temperatureconditions. Although the temperatures under theconditions studied were significantly different the observedhigh temperature variation affected the rate of gasproduction in partially buried digesters.

Effect of Insulation on Biogas yield

Ground insulation involved burying the experimentaldigesters fully underground. Fully buried digesters wereused to cushion high temperature variations encounteredin digesters that were partially buried. The results showed(Table 4b) increased mean sludge temperatures and thecorresponding gas discharge (Table5b) under bothconditions. In comparison with digesters partially buried,the mean sludge temperature of digesters under naturalconditions increased by about 2oC and small increase intemperature of 0.4oC was observed under greenhouseconditions (Tables 4a and b). More importantly, however,burying the digesters underground appeared to stabilizesludge temperatures above 20oC under natural conditionsand above 25oC under greenhouse condition as indicatedin Figure 6. From Table 5(b) it can be seen that the meandaily gas yields also increased by 5.6 litres and 7.6 litresunder natural and greenhouse condition, respectivelycompared to similar digesters that were partially buriedunderground (Table 5a).

Effects of Greenhouse on Biogas yields

The mean daily gas yields in millilitres per day (ml/day)from the test digesters are shown in Tables 5 (a) and (b) forpartially and fully buried digesters, respectively. The datarepresent the means of daily samples from replicateexperimental digesters at different hydraulic retention time(HRT).

Table 5: Mean gas yields for partially and fully burieddigesters across two sites

(a) Partially buried

Block Site Mean Gas Yields (ml/day)

1 GH 2119.07a“ NC 1453.70b

2 GH 4558.30a“ NC 2337.80b3 GH 4596.70a“ NC 3128.80b4 GH 5120.50a“ NC 4680.60a

MEANS

GH 4304.00aNC 3017.80b

(b) Fully buried

Block Site Mean Gas Yields (ml/day)

5 GH 11881.90a

NC 8585.20b

The results from Table 5 (a) showed biogas yields of 4.30litres/day (l/d) and 3.02 l/d under GH and NC, respectively.These correspond to 90.3 l/kg of VS added and 63 l/kg ofVS added under greenhouse and natural conditions,respectively. The results from the test digesters that werefully buried are shown in Table 5 (b). The average gasyield under GH environment was 11.9 l/day correspondingto 312.8 l/kg of VS added. The average gas yield undernatural conditions was 8.6 l/d corresponding to 226 l/kg of

For each block, values in the vertical column followed by thesame letter are not differentstatistically ( = 0.05) according to Duncan’s multiple rangetest.


P. K. CHERUIYOT

VS added. The gas yields were statistically differentbetween the test sites. The corresponding meantemperatures were also significantly different. Underelevated temperatures, as shown in Figures 3, 4 and 6, therate of gas production appeared to be enhanced. Thisagreed with the common assertion that the higher thetemperature the higher the rate of gas yields. By buryingdigesters fully underground increased digester temperaturewas achieved. This explained the enhanced gas yieldobserved. The results from digesters that were fully buried comparereasonably well with the literature (National Academy ofScience, 1977; Moorhead and Nordstedt, 1993; Chen andHashimoto, 1980; Chen et al., 1980; Hashimoto et. al., 1980;Chen and Hashimoto, 1978 and Kalia et al., 2000). Incomparison with the results from previous studies, theresults from digesters that were partially buried appearedlow. The low values obtained did not suggest failure ofthe system. It revealed the effect of high temperaturefluctuations on gas yields. The relatively higher biogasyields in the previous studies were achieved using a mixtureof input materials (El-mashad, H. M and Zhang, R., 2007),different HRT and type of waste material. In addition theuse of pretreated input materials and the operation ofdigesters at optimum sludge temperatures could haveenhanced gas yields.

Variation of Gas with Time

Temporal plots of the mean volumetric biogas yields aredepicted graphically as shown in Figures 4 and 5, forpartially and fully buried digesters, respectively under eachof the condition tested. The plot of polynomial fits oforder two gave relatively strong responses. The gasproduction coefficients of determination (r2) for partiallyburied digesters were 0.85 and 0.88 under greenhouse andnatural conditions, respectively. In comparison the gascoefficient of determination for fully buried digesters were0.86 and 0.90 under GH and NC. The intercepts of thepolynomial fits (equations) were forced to zero because apositive intercept would imply positive gas discharge atzero HRT and therefore the polynomial fits would beunrealistic. In this case zero retention time definitely resultsin zero gas discharge. In the batch feed digester used the results indicated ageneral increase in gas discharge with increase in HRT. A

peak value was obtained followed by a general decline ingas production. There were also subsidiary peaks andsharp oscillations noticed in the daily gas productioncurves. These patterns were more pronounced in digestersunder greenhouse conditions. The optimum daily gas peaks were attained at differentHRT across two sites. Digesters under GH conditions, onaverage, attained peak values on the 17th day and 20th dayin natural conditions. It should be remarked that the highand subsidiary peaks and sharp oscillations appearedsomewhat cushioned when means of gas yields was usedto plot the resultant graphs as shown below. There was alag time before the first gas was given off. In the firstexperiment a lag time of 5 days was recorded while in thesecond, third fourth and fifth experiments a lag time of twodays each was recorded. Thus faster gas discharge wasachieved in the subsequent studies probably because thedigesters were using seed cultures for startup from theprevious experiments. Both Chowdhury (1987) andMaramba 1978) also found that biogas plants using startersludge had very short lag time. Correlation of Temperature and Gas

Table 6: Correlation of temperature and gas (a) Partially buried digesters

Site Variable Temp Gas1 Temp 1 0.15

0.00* 0.006*324 324

Gas 0.15 10.006 0324 324

2 Temp 1 0.140.00* 0.0126*324 324

Gas 0.14 10.0126* 0.00*324 324

* Probability level



29

Site Variable Temp Gas1 Temp 1 0.82

0.00* 0.0001*138 135

Gas 0.82 10.000* 0.00*135 135

2 Temp 1 0.830.00* 0.0001*138 135

Gas 0.83 10.0001* 0.00*135 135

From Table 6, the correlation coefficient betweentemperature and gas for fully buried digesters wassignificantly higher at 0.83 and 0.82 under GH and NC,respectively compared to 0.15 and 0.14 for partially burieddigesters under GH and NC. This could suggest that the

(b) Fully buried digesters effect of temperature on gas yields would be morepronounced when biogas digester were insulated. It couldbe suggested, therefore, that by insulating biogas digesterincreased temperature and gas yields are achieved.

Quality of Gas

The quality of methane gas under greenhouse conditionsvaried at 55.7% after retention period of 14 days and 68.4%after 21 days. In comparison the methane gas contentunder natural conditions was 51.6% and 61.2%,respectively. The apparent difference in the methanecontent across the test sites could be attributed to thetemperature difference. Variation of Methane with Time

The effect of increased temperature and time on methanecontent in a biogas can be discerned from Fig.2. The effectof temperature on methane content in the biogas manifesteditself after the 10th day. After this day the methane contentunder greenhouse condition was higher than in naturalcondition. The difference between methane content underthe conditions studied could be attributed to observedtemperature difference.

0

2 0

4 0

6 0

8 0

1 0 0

0 1 0 2 0 3 0 4 0 5 0T I M E ( D A Y S )

% M

ETH

AN

E

% C H 4 ( G H ) % C H 4 ( N C )

Figure 2: Daily variation of methane in biogas under GH and NC


P. K. CHERUIYOT

0

1000

2000

3000

4000

5000

6000

7000

0 10 20 30 40Time(Day)

Gas

Yie

lds

(ml/d

ay)

0

5

10

15

20

25

30

35

Tem

pera

ture

oC

Temp(GH)Temp(NC)Gas(GH)Gas(NC)

Figure 3: Temporal volumetric gas yields and digester temperature undergreenhouse (GH) and natural condition (NC)

y = -16.169x2 + 542.43xR2 = 0.875

y = -21.588x2 + 704.33xR2 = 0.8469

0

1000

2000

3000

4000

5000

6000

7000

0 10 20 30 40Time (days)

Gas

Yie

lds

(ml/d

ay)

Gas(GH) Gas(NC) Poly. (Gas(NC)) Poly. (Gas(GH))

Figure 4: Plots of gas yields and polynomial equations for the corresponding timeunder GH and conditions NC



31

y = -25.4 x2 + 11 38 .1xR 2 = 0 .8 96 9

y = -34.4 79x 2 + 15 51 .4 xR 2 = 0 .86 42

0

2000

4000

6000

8000

10000

12000

14000

16000

18000

20000

0 10 20 30 40 50T IM E (D AY S )

GA

S YI

ELD

S (M

L/D

AY)

G as(G H 5 ) G a s(N C 5 ) Po ly . (G as(N C 5)) P oly. (G a s(G H 5))

Figure 5: Plots of gas yields and polynomial equations for the corresponding time under GH and NC

0

2000

4000

6000

8000

10000

12000

14000

16000

18000

20000

1 6 11 16 21 26 31 36 41 46Time (days)

Gas

Yie

lds

(ml/d

ay)

0

5

10

15

20

25

30

35

Tem

p D

eg. C

Gas(GH5)

Gas(NC5)

Temp(GH5)

Temp(NC5)

Figure 6: Plots of gas yields and temperature for the corresponding time of under GH and NC


P. K. CHERUIYOT

Digester Size

Based on the results of study, the size of biogas digestersuitable for the output of four cows and operated undertemperature of 28oC for twenty days would be 3.2m3. Thedaily amount of gas generated from the digester would be2.4 m3. This corresponds to the energy production of 31MJ per day and is further equivalent to 9 kWh per day or360 W. The reduced size of digester could provide adequateenergy for cooking and lighting for a typical Kenyan familyof six people.

CONCLUSION

Since identical influent loading concentrations wereachieved for each of the test digesters, the high gas yieldsobserved under greenhouse was attributed to elevatedtemperature. The methane content in the biogas wasrelatively higher under greenhouse compared to naturalconditions. These findings testify that the methane contentin a biogas is a function of temperature. For a batch feeddigester the trend of daily gas discharge can be stronglypredicted using polynomial regression equations. Apartfrom overflow of the digested slurry and low gas pressure,a plastic biogas digester is easy to operate and requiresvirtually no maintenance problems. Insulating biogasdigesters cushioned the adverse effect of high thermalfluctuations. Heating and insulating the digester undergreenhouse enhanced the digestion process, reducedretention time and digester size made smaller than for thebiogas unit under natural conditions. The energy generatedfrom the reduced sized digester is sufficient for cookingand lighting for a family of six persons. Thereforegreenhouse effects reduces the size of digester and hencethe cost of the digester. Ground insulation of biogas plantsunder greenhouse conditions is recommended.

REFERENCES

APHA (1995). American Public Health Association:Standard Methods for the Examination of Water andWastewater, 19th edition. American Public HealthAssociation. Washington, D.C.

Barnet, A, Pyle, L. and Subramanian, S.K. (1978). BiogasTechnology in the Third World: A multi-disciplinaryReview. Intermediate Development Research CentrePublication. Ottawa.

Brown, N. (1987). Biogas Systems in Development:Appropriate technology, vol. 4, No.3. P. 5-7.

CBS (2000). Central Bureau of Statistics. Ministry ofPlanning and National Development. GOK. Nairobi,Kenya.

Chen, Y.R. and Hashimoto, A.G. (1978). Kinetics of methanefermentation. Biotechnology and BioengineeringSymposium. No.8 pp 269-282.

Chen, Y.R. and Hashimoto, A.G. (1980). Energy requirementsfor anaerobic fermentation of livestock wastes.Proceeding of the 4th international symposium onlivestock wastes. Amarillo, TX. April 17-17th.

Chen, Y.R., Varel, V.H., Vincent, H., and Hashimoto, A.G.(1980). Methane production from AgriculturalResidues. A short Review Information and Economicproduct research and development. Dec 1980, pp 471-477.

Cho, J.K., Park, S.C. and Chang, H.N. (1995). BiochemicalMethane Potential and Solid waste AnaerobicDigestion of Korean food wastes. BioresourceTechnology 52 (19950 245-253.

Chowdhury, R. (1987). Kinetics Studies of AnaerobicDigestion: Comparing the performance of Batch andSemi-continuous Systems. M.Phil. Thesis, Universityof Reading, England.

El-mashad, H. M and Zhang, R (2007). Co-digestion offood waste and dairy manure for biogas production.Transactions of the ASABE 50 (5): 1815-1821

FARMESA (1996). Farm-level Applied Research Methodsfor East and Southern Africa. Tubular Plastic Bio-digester. Farm-level Applied Research Methods forEast and Southern Africa Publication. Harare.

FAO (1992). Biogas Processes for SustainableDevelopment. FAO Agricultural Services Bulletin No.95. Rome.

Fulford, D. (1988). Running Biogas Programme: AHandbook Intermediate Technology Publication.London.

Hashimoto, A.G., Chen, Y.R., Varel, V.H. and Prior, R.L. (1980).Anaerobic fermentation of Agricultural Residues. IN:Utilization and Recycle of Agricultural waste andResidues. Edition M. Shuler. CRC, Baton Raton, FL.,pp135-196.

Hill, D.T. and Bolte, J.P. (2000). Methane production fromlow-solid concentration liquid swine waste usingconventional anaerobic fermentation. BioresourceTechnology 74(2000) 241-247.

Kalia, V.C., Sonakya, V. and Raizada, N. (2000). Anaerobicdigestion of banana stem waste. BioresourceTechnology 73 (2000, 191-193.

Kristoferson, L.A. and Bokalders, V. (1991). RenewableEnergy Technologies: IntermediateTechnology Publication. London.\

Lekule, F.P. (1996). Technologies for Improving the WellBeing of Rural Women in Tanzania. Final ReportSubmitted to FAO/SIDA, Farming System Programme.Harare.



33

Maramba, F.D. (1978). Biogas and Waste Recycling: ThePhilippines Experience; Liberty flourmills Inc., MetroManila, Philippines.

Montgomery, D.C. (1976). Design and Analysis ofExperiments. John Wiley & sons Inc. Toronto.

Montgomery, D.C., Runger, G.C. and Hubele, N.F. (1998).Engineering Statistics. John Wiley and Sons.New York.

NAS (1977). Methane Generation from Human, Animaland Agricultural Wastes. National Academy ofScience. Washington, D.C.

Ott, L. (1988). An Introduction to Statistical Methods andAnalysis. PWS-KENT Publishing Company.Boston.

Price, E.C. and Chereminisoff, P.N. (1981). BiogasProduction and Utilization. Ann Arbor S c i e n c ePublishers Inc. Michigan.

SAS Institute, Inc. (1998). SAS users guide; SAS Institute,Inc., Cary, NC. USA, pp 434-448.

Singh, L., Maurya, M.S., Bai Ram, M. and Alam, S.I. (1993).Biogas production from Night soil- Effects ofLoading and Temperature. Bioresource Technology45(1993) 59-61.

Twidel, J.W. and Weir, A.D. (1987). Renewable EnergyResources. E & F.N. Spon Ltd. London.

Walkley, A. and Black, I.A. (1947). A Critical Examinationof Rapid Method for Determining Organic Carbon inSoils- effect of Variation in Digestion Conditions andof Inorganic soil constituents. Soil Science 251-64.



EULER DECONVOLUTION AND SPECTRAL ANALYSIS OF REGIONALAEROMAGNETIC DATA FROM THE SOUTH-CENTRAL ZIMBABWE

CRATON: TECTONIC IMPLICATIONS

1Ranganai, R. T.

1Department of Physics, University of Botswana, P.Bag UB0704, Gaborone, Botswana.


ABSTRACT: Existing regional aeromagnetic data from the south-central Zimbabwe craton hasbeen analysed using 3D Euler deconvolution and spectral analysis to obtain quantitative informationon the geological units and structures for depth constraints on the geotectonic interpretation of theregion. The Euler solution maps confirm and extend the structural pattern previously identifiedusing shaded relief imaging and derivative techniques: ENE, NNE, NNW, NW and WNW; thusconfirming the geological significance of the qualitative interpretation. In places, Euler solutionsalso show additional patterns typical of sill-edges, thus mapping previously unrecognised mafic/ultramafic intrusions. Most structures identified are predominantly of shallow origin, with Eulerdepths solutions 1.0 km, and cut across the greenstone belts. A number of isolated deep Eulersolutions are associated with ultramafic complexes, the Great Dyke and the Umvimeeela dyke; andthese points could represent the original magma chambers and/or feeder points for these units. Alinear cluster of solutions with depths around 2.0 km marks the Zimbabwe craton-Limpopo Beltboundary remarkably well. Spectral analysis results suggest the magnetic basement at about 8 kmdepth, and this probably corresponds to a crustal boundary deduced from gravity and seismic datato occur at 7-9 km depth.The geostructural framework of the area is compatible with the postulatedlate Archaean collision involving the Zimbabwe and Kaapvaal cratons and the Limpopo Belt, andlater crustal extension during the break-up of Gondwana. The geological-tectonic correlationsuggests that the interpreted regional trends are mainly 2.6 Ga and younger, and relate to tectonicevents including the reactivation of the Limpopo Belt at 2.0 Ga and the major regional igneous/dyking events at 1.8-2.0 Ga (Mashonaland), 1.1 Ga (Umkondo) and 180 Ma (Karoo). The greenstonebelts were an integral part of the lithosphere before much of the upper crustal (brittle) deformationoccurred.

Key words: Aeromagnetic data, Euler deconvolution, Spectral analysis, Tectonic interpretation,Zimbabwe craton

INTRODUCTION

The south-central region of the Archaean Zimbabwe craton(ZC) encompasses several geological features: basementgneisses and tonalites of the ~3.5 Ga Tokwe Segment,greenstone belts, the Great Dyke and its satellites, theNorth Marginal Zone (NMZ) of the Limpopo belt, post-volcanic granite plutons, and mafic dyke swarms as shownin Fig. 1 [Wilson, 1981; 1990; Wilson et al., 1987; 1995;Bickle and Nisbet, 1993]. The ~3.5 Ga Tokwe Segment (TS,index map in Fig. 1) is a unique terrain considered to be anucleus, from where the craton grew westwards and

northwards by crustal accretion (Wilson, 1990; Wilson etal., 1995; Kusky, 1998; Horstwood et al., 1999; Jelsma andDirks, 2002). The study area is therefore considered ofcrucial importance to the understanding of the tectonicevolution of the craton as a whole, the Limpopo belt, andthe Archaean crust in general (e.g., Nisbet, 1987; Wilson,1990; Bickle and Nisbet, 1993; Fedo et al., 1995; Kusky,1998; Horstwood et al., 1999; Jelsma and Dirks, 2002).Consequently, many geological studies have been carriedout in this regard and these have been complemented bygeophysical investigations at various stages (e.g., Podmoreand Wilson, 1987; Gwavava, et al., 1992; Jones et al., 1995;


Euler Deconvolution and Spectral Analysis of Regional Aeromagnetic Data fromthe South-Central Zimbabwe Craton: Tectonic Implications

35

Mushayandebvu, 1995; Ranganai, 1995; Ranganai et al.,1995, 2008). The geophysical studies involve gravity andaeromagnetic data, and a number of greenstone belt gravitymodels have been derived (Ranganai et al., 2008). However,the aeromagentic studies have so far only provided aqualitative interpretation with an indication of wherechanges in magnetic susceptibility occur, but not what

produces the anomaly nor its depth (e.g., Ranganai et al.,1995; Ranganai and Ebinger, 2008). Typically, the mostcommon parameter sought for in aeromagneticinterpretation is the location of the magnetic source bodiesand their depths (Bournas et al., 2003 and referencestherein).

Figure 1. Simplified geological map of the study area. Greenstone belts are named after respective towns (By =Bulawayo). Other units are: TS (see insert) = ~3.5 Ga Tokwe Segment (north-eastern area between Zvishavane and

Mashava), ED = East dyke, GT = Gurumba Tumba serpentinite; mcd = Mashava-Chivi dykes, SPD = Sebanga-Poort dyke,UD = Umvimeela dyke, ILSZ = Irisvale-Lancaster Shear Zone, JF = Jenya fault, MF = Mchingwe fault, Sh = Snake-head

section (Mberengwa belt), MwF = Mwenezi fault, FRD = Fort Rixon dykes.


R. T. RANGANAI

In this paper, existing regional aeromagnetic data are usedto estimate the location and depths to susceptibilitydiscontinuities through standard 3D Euler deconvolutionand spectral analysis techniques (e.g., Spector and Grant,1970, 1975; Hahn et al., 1976; Thompson, 1982;Ruotoistenmaki, 1983, 1987; Reid et al., 1990). The paperextends the preliminary invetsigations of Ranganai (1999).The main objectives are to provide depth constraints onprevious qualitative structural interpretation using shadedrelief imaging and derivative techniques. These data-enhancement techniques present maps in a form whichassists human comprehension through recognition ofspecial patterns while the present study providesquantitative information for geological interpretation. Theapplication of Euler’s homogeneity relation through theprocess of deconvolution has been demonstrated to be aneffective method for delineation of potential fieldboundaries and the estimation of depth to their upper edges(e.g., Reid et al., 1990; McDonald et al., 1992; Paterson etal., 1992; Mushayandebvu et al., 2001, 2004; Bournas etal., 2003). Spectral analysis has been commonly appliedfor the determination of depths of magnetic assemblageswithin the upper crust, depth to magnetic basement and/or crustal thickness variations (e.g., Hinze, 1985; Cowanand Cowan, 1993; Poudjom-Djomani et al., 1995; Allek andHamoudi, 2008). Both techniques are therefore useful forinvestigating structures, magnetic zones, and crustaldomains previously identified in the area (Ranganai, 1995;Ranganai et al., 1995, 2008). The computed depthsconstitute important quantitative constraints on thegeological interpretation of the area, and the elucidationof its tectonic evolution; they help discriminategeotectonic models.

GEOLOGICAL SETTING The study area is the south-central part of the ArchaeanZimbabwe craton and lies between latitudes 19.9ºS and21.1ºS and longitudes 28.9ºE and 30.5ºE, with mining townslocated throughout the area (Fig. 1). Principal geologicalunits in the study area include the Mberengwa (Belingwe),Buhwa, Filabusi, Fort Rixon and Gwanda greenstone belts,‘young’ K-rich (post-volcanic) granites, and sections ofthe Great Dyke and its satellites (Umvimeela and Eastdykes), set within ancient tonalitic gneisses (Fig. 1). The‘young’ granite plutons (2.7-2.6) intrude and deform boththe older gneisses and the greenstone belts (Wilson et al.,1995; Jelsma et al., 1996; Horstwood et al., 1999), but theseare in turn cut by the ~2.5 Ga NNE-striking Great Dyke andits nearly parallel satellite dykes and features. The GreatDyke is a linear mass of mafic-ultramafic rocks while itssatellites are true gabbroic dykes (Wilson and Prendergast,1988; Oberthür et al., 2002); together they form the first

major igneous event after cratonization (e.g., Wilson, 1990;Hortswood et al., 1999; Jelsma and Dirks, 2002; Schoenberget al., 2003). Many workers argue for a close relationshipbetween the emplacement of the Great Dyke and satellites,the intrusion of the late plutons and tectonic events in theLimpopo belt (e.g., Mukasa et al., 1998; Frei et al., 199;Oberthür et al., 2002). Several layered ultramafic intrusionsof the ~2.8 Ga Mashava Ultramafic Suite (Wilson, 1979,1981, 1990; Prendergast and Wingate, 2007) and maficdykes of various ages are scattered throughout the area(Wilson et al., 1987). They both appear to be intimatelyrelated to the tectonic processes that produced the mainArchaean granite-greenstone terrain (Wilson, 1981, 1990;Wilson et al., 1995). Greenstone stratigraphy includes the ~2.9 Ga LowerGreenstones, the dominant/widespread 2.7 Ga UpperGreenstones and minor ~2.7 Ga Shamvaian type sediments(Wilson, 1981, 1990; Taylor et al., 1991; Bickle and Nisbet,1993; Wilson et al., 1995; Jelsma et al., 1996; Horstwood etal., 1999). Generally, all the greenstone belts show acharacteristic sequence of ultramafic, mafic, felsic andvolcanic-sedimentary assemblages, mainly at greenschistfacies metamorphism rising to amphibolite facies at theirmargins, close to batholith contacts. The Mberengwa(Belingwe) greenstone belt (Fig. 1) contains the mostcomplete greenstone sequences in the craton, and its wellpreserved and exposed stratigraphy has been correlatedwith units across much of the craton (Wilson, 1979, 1981;Bickle and Nisbet, 1993; Wilson et al., 1995; Jelsma andDirks, 2002; Prendergast, 2004). The greenstone sequencesand configuration may reflect rifted or overplatedsequences related to emplacement of mantle plumes, withdeformation being attributed to vertical tectonic processes,or remnant oceanic crust or island-arc material that wasamalgamated with continental fragments during some formof subduction-accretion (Jelsma and Dirks, 2002; Ranganaiet al., 2008). Detailed revisions of the greenstonestratigraphy can be found in Wilson et al. (1995) andHortswood et al. (1999) while comprehensive summariesof the craton are given by Blenkinsop et al. (1997) andJelsma and Dirks (2002). Campbell et al. (1992) provide aprovisional tectonic map and tectonic evolution of thecountry. The study area is bounded on its south-eastern edge bythe Northern Marginal Zone (NMZ) of the ArchaeanLimpopo orogenic belt (LB), in thrust contact with cratonicgranitoids (Rollinson and Blenkinsop, 1995; Mkweli et al.,1995; Fedo et al., 1995; Frei et al., 1999). The NMZ mayconsist mainly of reworked granitoid-greenstone rocks ofthe craton at amphibolite facies metamorphism (Hickman,1978; Van Reenen et al., 1992), with several inclusions of



37

mafic dykes, ultramafics and banded iron formation(Rollinson and Blenkinsop, 1995). The ZC-NMZ boundaryis traditionally taken as the orthopyroxene isograd butrecent geological mapping points to a tectonic break(Mkweli et al., 1995; Rollinson and Blenkinsop, 1995;Blenkinsop et al., 1995). It is hoped that this study willcontribute in-depth information that will help elucidate thenature of the contact previosuly defined by aeromagneticsignatures alone (Ranganai, 1995; Ranganai et al., 1995).

AEROMAGNETIC DATA AND CORRELATION OFANOMALIES WITH GEOLOGY

The aeromagnetic data used in this study were obtainedfrom the Zimbabwe Geological Survey (ZGS) and are basedon 1 km spaced flight lines with 305 m constant mean terrainclearance. Two surveys were conducted in 1983 and 1988using Geometrics proton precession (0.1 nT resolution)and Scintrex cesium vapour (0.001 nT resolution)magnetometers, respectively. Flight directions were E-Wand/or N-S, approximately perpendicular to the dominant

geological trends, the greenstone belts. Tie-lines were flown14 km apart and the data were diurnally corrected and flightline-levelled using a combined computer-manual method.Data from the two surveys were combined following theprocedure discussed by Barritt (1993). The levelled flightline data were first gridded in the UTM co-ordinate systemat 250 m cell size using a bidirectional algorithm (e.g., Smithand Wessel, 1990), and then reduced to the pole. Reductionto the pole (RTP) assumes induced magnetisation and shiftsthe anomalies to lie directly over the sources (e.g., Blakely,1995), thus producing anomaly maps that can be morereadily correlated to the surface geology. RTP is also arequirement for the Euler deconvolution and spectralanalysis algorithms used in this study (Spector and Grant,1970; Geosoft, 2004; see Appendix). The Geosoft algorithmsused to calculate the RTP caters for both high and lowmagnetic latitudes (-60 in the study area). The RTP datawere then contoured at 50 nT interval (Fig. 2) and sun-shaded (Fig. 3) for geological and structural mapping beforethe deconvolution and spectral analysis.

In general, the pole-reduced aeromagnetic data display a

Figure 2. RTP aeromagnetic contour map with grey-scale grid. Basic contour interval is 50 nT. Dark tones represent lowvalues while white tones are high values. Prominent magnetic units are labelled: GD= Great Dyke, B=Buhwagreenstone belt (BIF quartzite); BKD= Botswana Karoo Dyke (swarm); Ma= Mashava (ultramafic) Suite; HX=

Interpreted ultramafic body; SR= Shamba (ultramafic) Range; NMZ = North Marginal Zone (Limpopo belt); other labelsas in Figure 1


R. T. RANGANAI

typical granite-greenstone signature, and anomaliescorrelate well with geological units: the shapes are clearlyoutlined and broad lithological boundaries are discernable(cf. Figs. 1 and 2, 3). In all greenstone belts, the extensiveUpper Greenstone basalts are characterised by flat,homogeneous relief as the volcanic lavas contain littlemagnetite and also probably due to the low metamorphicgrade. The only exception is the Buhwa greenstone belt(B, Fig. 2) with high magnetic anomalies due to themagnetite, quartzite and haematite which dominate thelithologies (Fedo et al., 1995). Magnetic highs occur overmafic dykes and ultramafic intrusions (e.g., GD, Ma, SR) aswell as komatites and banded iron formation horizons withingreenstone belts (white tones in Figure 2). Three or fourdifferent magnetic zones can be identified on the RTPcontour map (Fig. 2) based on anomaly textures, definedby parameters like linearity, relief, and background level,and features such as anomaly shapes and wavelengths.

The northern half of the study area is generallycharacterised by relatively high magnetic signatures(~31300 nT) and appears to be a separate terrain. Thecentral and south-central parts have intermediateanomalies (~31000 nT) while the southwest and southeastareas have low (30000 nT) and bipolar (very high, ~32500nT and low, ~30000 nT) signatures, respectively (Fig. 2).

The latter occur over the Buhwa greenstone belt and theNMZ, reflecting the high metamophic grade of the area(e.g., Grant, 1985), and possibly the effect of mafic-ultramafic-BIF inclusions in the gneisses. The northernedge of this broad high marks the ZC-LB boundaryremarkably well. The four different magnetic zones probablyrepresent crustal domains or magneto-tectonic provinces,and a more detailed qualitative interpretation anddiscussion is underway.

Figure 3. Shaded relief RTP magnetic map of study area. ‘Sun’ illumination angle is 30 and declination angles are 60,115. Note the use of two declination angles in order to display the magnetic data which reflect structures at many

orientations. Structural features labelled are discussed in text: D1 = dyke; ILSZ = (Irisvale-Lancaster) Shear Zone; SRe= Shamba (ultramafic) range extension; Gw = Gwanda greenstone belt; Mb = Mberengwa (Belingwe) greenstone belt;

other labels as in Figs. 1 and 2). Note the dominance of NNW (FRD dyke) and NNE (Great Dyke) dyke trends and NW toWNW fault directions



39

Overall, the aeromagnetic data display a considerable rangeof wavelengths and amplitude variations but are dominatedby high amplitude, short wavelength anomalies fromshallow sources (Fig. 2). The latter are clearer on shadowand derivative maps (e.g., Fig. 3; Ranganai and Ebinger,2008) and are expected to be clearly isolated by the gradientbased 3D Euler deconvolution technique (e.g.,Mushayandebvu et al., 2001; see below and Appendix).Several new features and/or extensions of known units areobserved, such as NNW-striking dykes (FRD, D1), ESE-trending dykes (BKD), Shamba Range (ultramafic)extension (SRe), and a possible ultramafic body (anomalyHX) (cf. Figs. 1 and 3). The latter magnetic body (HX) hasan associated Bouguer gravity high (Ranganai et al., 2008),thus pointing to a probable ultramafic composition/originfor the anomaly source. Euler deconvolution is appliedparticularly to quantitatively identify the boundaries ofsuch magnetic entities, and other features. 3D Euler Deconvolution and Structural Mapping In order to investigate the source of the magneticlineaments seen on shadow and derivative maps (e.g., Fig.3), and to confirm the position, structure and/or geologicalassociation in previous qualitative interpretation (e.g.,Ranganai et al., 2008), Euler deconvolution was applied toprovide additional 3D information. The technique is lesssubjective than shaded relief maps commonly used inlocating low gradient anomalies, and it also assists in thedelineation of crustal blocks with different magneticparameters and, therefore, tectonic interpretation (e.g.,Bournas et al., 2003). It is particularly useful where thereare interfering sources such as in the study area, as itinvolves the analysis of gradients. However, it should benoted that the depth estimates provided by this methodare inherently less well determined than the positionalestimates (e.g., McDonald et al., 1992; Mushayandebvu etal., 2001). Further, although it has been claimed that nogeological model is assumed (Thompson, 1982; Reid et al.,1990), the optimal use of the algorithm depends to a largepart on the user’s a priori knowledge of the geology in anarea. Its successful application also depends on the qualityof the data, as well as the selection of the processingparameters, namely: the structural index, N and the gridwindow size, W (see Appendix). N is a measure of the fall-off rate of the anomaly with distance, and is closely relatedto the geometry of the causative body with simple bodieshaving prescribed values, between 0 and 3, as discussedby Reid et al. (1990). We note that the conventionaltechnique used in this study (Reid et al., 1990) assumesthat the observed field in each Euler window is due to a 3Dsource and this leads to generally poorly constrainedsolutions where the source is in fact 2D (Mushayandebvu

et al., 2004). The various improved versions (e.g., Fairheadet al., 1994; Barbosa et al., 1999; Mushayandebvu et al.,2001, 2004) were not available for this study and thereforethe results are not fully ‘cleaned out’. Based on spatial observations from the RTP and derivativemaps, a number of structural indices (SI’s) and windowsizes were applied. In general the number of solutions fromthe RTP magnetic grid increased as the structural indexwas increased from SI = 0 to SI = 3, and window size wasalso increased from 4 (1x1 km) to 12 (3x3 km). When the SIused was between 0 (contact of considerable depth extent)and 1 (dyke or sill edge), the solutions obtained tended tobe of shallow depth. At higher values of SI (SI = 1 to 2), thesolutions exhibited a clear focus in the location and depthof the solutions, and results for these indices tend to besimilar (e.g., Figs 4 and 5). This emphasises the need toapply several structural indices, particularly in geologicallycomplex areas such as are under study. This was also notedby Reid et al. (1990) who suggested that gross structuraltrends could still be outlined even with a poor choice of N(albeit with inaccurate depth solutions). It should also benoted that due to the gridding process which often resultsin ‘strings of pearls’ for dykes (e.g., BKD, Figs. 2 to 5), thedyke anomalies could resemble a line of dipoles, whichhas a structural index of 2 (Reid et al., 1990; Paterson et al.,1992; Mushayandebvu et al., 2004).

The results of standard Euler deconvolution are as shownin Figures 4 and 5. For a given N and W, the techniquecalculates from the magnetic gradients in the x, y, and zdirection the boundary of a magnetic unit and the depth tothe boundary. The located boundary point is plotted on amap and represented by a circle, whose size is scaledaccording to the depth units. The Euler solution mapspresented (Figs. 4 and 5) indicate several structures, witha direct coincidence of linear clustering solutions withknown features such as the Umvimeela dyke (UD), theGreat Dyke (GD), and the Grumba Tumba serpentinite (GT),and these form obvious features on all maps (cf. Fig. 1).The latter is an outstanding semi-circular (arc-shaped)feature at the centre of the maps, that partly follows theMwenezi fault (Mw) in the west, cutting across the GreatDyke to terminate against the Mchingwe fault (MF) to theeast (Figs. 4 and 5). The widths of these known featuresare also represented well, particularly at small SI where, forexample, both edges of dykes are clear (cf. Figs. 1 and 4).Some linear solutions are traceable for tens of kilometresto just over 100 kilometres (e.g. ED, UD, FRD, BKD; Figs 4and 5), but others are broken up into segments. Faults canbe interpreted at these breaks, but the longer breaks mayrepresent zones of constant susceptibility. Other anomaliesare much shorter but the various segments form part of


R. T. RANGANAI

Figure 4. Euler solution map for RTP magnetic grid; N=1, W= 8 (2 x 2 km). Acceptance level set at 70%. Features and/ortrends discussed in text are labelled

Figure 5. Euler solution map for RTP magnetic grid; N=2, W= 8 (2 x 2 km). Acceptance level set at 60%. Features and/ortrends discussed in text are labelled. Note the general similarity of solution patterns with Figure 4 (N = 1)



41

more continuous features (e.g., MF, Mw; Figs. 4 and 5).These are best viewed at certain directions with the map inhand, allowing their identification as continuous trendsand/or significant structures of considerable strike. Otherscan only be interpreted in conjunction with the magneticderivative and shaded relief maps (e.g., Fig. 3).

Several other observations can be made from the maps.For example, a number of solutions in the northern areaexhibit patterns typical of sill-edges (e.g., HX and HY onFigs. 4 and 5). The Euler deconvolution indicates that thebody situated at HY extends to 2.0 km and coincides witha palaeomagnetic interpreted magma source for theUmvimeela dyke (Bates and Mushayandebvu, 1995).Newly found NNE- to NE-trending structures are indicated,particularly in the Mberengwa greenstone belt area (e.g.trends marked AA and PP on Figs. 4 and 5). The Mwenezifault (Mw F, Figs. 1 and 3) can be extended in bothdirections from the mapped exposure to cut across theentire study area and into the Limpopo Belt in the south-east (Mw-Mw, Figs. 4 and 5). The ZC-NMZ boundary is

characterised by a distinct linear clustering of solutionson all maps (feature Tz, Figs. 4 and 5). This confirms recentgeological interpretations that the boundary is a tectonicbreak/contact (e.g. Mkweli et al., 1995; Blenkinsop et al.,1995; Rollinson and Blenkinsop, 1995; Ranganai, 1999),rather than the orthopyroxene isograd previously used.Unfortunately, the present results cannot distinguishbetween a vertical and a thrusted contact, but recentadvances in the technique may allow the estimation of dipof the magnetic body (e.g., Mushayandebvu et al., 2001,2004). Several other linear solutions, corresponding tomafic, ultramafic and iron formation inclusions can also beidentified within the NMZ. Surprisingly, magnetic zones(briefly discussed above) previously identified on the RTPand apparent susceptibility maps (Ranganai, 1995;Ranganai and Ebinger, 2008) are not represented in anyrecognisable pattern on the solution maps. This partlyconfirms the interpretation that the zones reflect relativelydeep crustal blocks, although the effect of window sizecannot be ruled out.

Figure 6. Geological and Structural Interpretation map of the study area (see Figure 1 and Table 1 for comparison). BKD= Botswana Karoo dykes; ED = East dyke, FRD = Fort Rixon dykes, HX = Interpreted ultramafic complex; SRe = ShambaRange extension; UD = Umvimeela dyke, Mchin Fault = Mchingwe fault, Mw F = Mwenezi fault, NLTZ = North Limpopo

Thrust Zone. Other geological unit labels (FR, Ma and NMZ) for reference purposes only (cf. Fig. 3)


R. T. RANGANAI

The final interpretation (Fig. 6) was guided by printed colourmaps at various scales and ‘on screen’ displays with higherresolution than figures presented. It is a compilation of (a)known structures, (b) anomalies calibrated by surfacegeology, and (c) structures interpreted by analogy to (b).Generally, the western half of the study area is characterisedby NNW-trending structures, in places cut by NW-to-WNW-trending faults whereas the east is dominated byNNE-trending structures, in places cut by NW-to-WNW-trending faults and NNW-trending dykes (e.g. Figs. 2 to6). The southern part shows EW-to-ESE-trending dykes inthe west and ENE-trending structures in the east.Significantly, the observed magnetic trends haverepresentatives craton-wide, which implies that ourinterpretation and inference can be applied to the rest ofthe craton with some degree of confidence. Relative agesof the structures can be inferred from the details of theintersection relationships and other geochronologicalinformation (e.g., Taylor et al. 1991; Wilson et al., 1995;Horstwood et al., 1999; Jelsma and Dirks, 2002; Oberthüret al., 2002; see discussion below).

Euler Depth Solutions and their Significance In general, the minimum depths returned in Eulerdeconvolution are of the order of the grid interval, whilethe maximum depths are about twice the window size (Reidet al., 1990; Geosoft, 2004). In this study, however,comparable solution depths were obtained from differentwindow sizes and structural indices, probably implyingthe ‘true’ depths of features. At low values of N (SI = 0 to1), the solutions tended to be shallower than for highervalues of N (SI = 1 to 2), confirming the out- and sub-cropping nature of several linear features represented bythe low SIs. To allow depth investigations for someindividual units, solutions have been plotted as circlesscaled for the depth magnitudes and subdivided into threelevels represented by different sizes (Fig. 7). More thanseventy five percent of the depth solutions are less thanor equal to 1.0 km, and very few are over 2 km. This meansthat most of the features mapped are shallow, more so if wehave to consider the fact that solution depths using a lowSI cluster near the mid-point of a steep feature (McDonaldet al., 1992). These shallow structures (particularly 150m) are considered important for regional groundwaterexploration, and therefore the technique could be used toassess such structures and features in the region (e.g.,Ranganai and Ebinger, 2008). The intermediate (1-2 km)

and deeper solutions occur mainly in the southern part ofthe area: at the ZC-NMZ boundary in the south-east, andover the Gwanda (Gw) greenstone belt area in the south-west (cf. Figs. 1 and 7). A few also occur over ultramaficand mafic intrusions, where there are well-definedanomalies (e.g., Ph, S, D, HX in Fig. 7). The two isolateddeepest solutions are associated with ultramafic complexes(Ph and S, Fig. 7), and these points could represent theoriginal magma chambers.

Using the various Euler solution maps, the magneticsources in the northern parts of the area generally appearshallower than in the southern parts. This suggests thateither the sources were emplaced at shallow levels or thatthe north probably experienced more uplift and highererosion levels than the south. The latter is supported bythe fact that the northern part of the Mberengwa (Belingwe)greenstone belt (Sh, Fig. 1) is considered to be a deeperlevel crustal section than the main belt to the south (Martin,1978; Bickle and Nisbet, 1993). This is reported to be athigher grade metamorphism (amphibolite facies) than themain belt (greenschist facies). Based on geological andgeophysical evidence, Ranganai et al. (1995) infer that thearea underwent at least one major period of heating anduplift, followed by erosion. Magnetic modelling of profilesin several places across the Umvimeela and East dykeswithin the study area show a progressive increase in depthto top from north to the south (Mushayandebvu, 1995).Both dykes have shallow dips south of latitude 20.5S.This, together with palaeomagentic data, was interpretedto suggest a tilting of the craton adjacent to the Limpopobelt, with the affected block being limited by the cross-cutting Mchingwe fault (Mushayandebvu, 1995).

However, the Great Dyke and its satellites are seen to haveisolated areas having slightly deeper solutions of 1.5 to2.0 km within the northern parts of the area. For the GreatDyke, the area of deep solutions (A, Fig. 7) approximatelycoincides with the boundary of the Wedza and Selukwecomplexes (Wilson and Prendergast, 1988), but it is not yetpossible to place any significance to this. A similar area(D) occurs on the Umvimeela dyke (Fig. 7). On this dyke(UD), another area of deep solutions just north of theMchingwe fault (F, Fig. 7) correlates with a point interpretedas its possible feeder point, identified through magneticfabric analysis (Bates and Mushayandebvu, 1995). Notethat D also encompasses solution patterns typical of silledges like HX, and could therefore be a concealed ultramaficbody.



43

Figure 7. Euler depth solution map for RTP magnetic grid; N=2, W=8 (2 x 2 km). Solution depths: green = 0-1 km, red =1-2 km, and blue = 2.0-4.0 km. Acceptance level set at 60%. Features and/or trends discussed in text are labelled. (For

interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article)

Source Depth Estimates from Magnetic Spectra Since the aeromagnetic anomaly patterns are shaped verylargely by the depths and volumes of the sources (Spectorand Grant, 1970), and of course, their magnetisation, meandepths to the sources can be determined by spectralanalysis (Ruotoistenmaki, 1983, 1987). The algorithm usedfor RTP produces a 2D radially averaged power spectrumas a function of wavenumber (cycles/km). [The amplitudespectrum is a 2D function of the amplitude relative towavenumber and direction, while the radially averagedpower spectrum is a function of wavenumber alone.] Weuse the method of Spector and Grant (1970) where theearth is modelled as an ensemble of rectangular, vertical-sided parallelepipeds of varying depth, width, thickness,and magnetisation. For such a model, the average ensembledepth h is simply half the gradient on the log radial spectra(Appendix). For wavenumbers in cycles/km h is thencalculated from the relation (Spector and Grant, 1970):

h4slope (1)

Best-fit straight lines are drawn on the spectra. In general,there are limitations and errors on the depth estimates dueto the window size, the grid spacing and the quality of thelinear fit on the plots (e.g., Spector and Grant, 1970;Ruotoisenmaki, 1983, 1987; Poudjom Djomani et al., 1995;Allek and Hamoudi, 2008). For example, the data lengthmust be six times the maximum source depth, and theshallowest depth estimate must be not less than 40% ofthe grid spacing (see Poudjom Djomani et al., 1995 andreferences therein). Figure 8 shows the 2D radially averaged power spectrumof the RTP aeromagnetic data of the study area. The plotshows a small degree of curvature, rather than purelystraight line segments. There are several possible reasonsfor this curvature. It could be due to leakage back into theNyquist interval, but this is difficult to ascertain, and wesuggest that this indicates increasing values of mean depthof ensembles. Ruotoisenmaki (1983, 1987) attributes suchproblems to the size effect (Appendix) in the method thatis used here (Spector and Grant, 1970), particularly if thereare areas with consistently varying depths. This leads to


R. T. RANGANAI

interpreted depths that are larger than the source depths(Spector and Parker, 1979) and, therefore, the computeddepths should be considered as maximum depths. It shouldalso be noted that the depths estimated from theaeromagnetic spectra are relative to the flight altitude (305m in this case). It is possible to distinguish the residual and regionalcomponents in Fig. 8, although the former shows a rangeof possible depths. Two straight line segments are identifiedand their slopes used to calculate depths to thecorresponding magnetic susceptibility contrasts. For eachlinear segment, we define three or at least two possiblelines, estimate the corresponding depths, and define the

error from these computed depths (h1, h2, hr in Fig. 8; seePoudjum Djomani et al., 1995). Mean depths of 0.77 ± 0.15km and 2.52 ± 0.50 km for the shallow and deep sources,respectively, are determined. The latter predominate atwavenumbers less than 0.2 cycles/km, corresponding towavelengths longer than 5.0 km. Wavelengths smaller thanabout 0.6 km represent noise, and possibly someoutcropping features. Alternatively, a second straight linesegment can be drawn for the shallow sources (hs1 andhs2, Fig. 8), with a depth estimate of 0.55 ± 0.10 km. This islarger than the flight height and gives an ensemble depth(from ground surface) of about 0.15 km for these sources,which is >40% of the grid spacing, and thereforeacceptable.

Figure 8. Radially averaged 2D spectra of aeromagnetic data for the study area (hb = magnetic basement depth; hr =depth for regional magnetic sources; hs = depth for shallow sources). Dashed lines show slopes used to estimate error

quoted in text following the method of Poudjom-Djomani et al. (1995)



45

This shallowest layer may be associated with shortwavelength features such as the iron formations,ultramafics and mafic dykes that outcrop in various placeswithin the study area. Comparable results have beenobtained in similar granite-greenstone terrains in Australiaand Canada (e.g. Hinze, 1985; Cowan and Cowan, 1993).The deepest features (hb, Fig. 8) yield a depth of 8.0 0.5 kmbelow surface, most probably representing the magneticbasement in the area. Significantly, a crustal ‘boundary’ ata depth of about 7-9 km was estimated from gravity(Ranganai, 1995; Ranganai et al., 2008) and seismic (R. Clark,pers. comm., 1995) data from this area. However, it shouldbe noted that the various techniques ‘look’ at differentparts of the anomaly source and the differences in depthsare therefore not unusual. Also, the power spectrum of anarea is a statistical estimate such that all spectral estimatesare averages. Analysis of spectra covering severalmagneto-tectonic terrains may produce averages whichare not found in either province (Cowan and Cowan, 1993). For the latter reason, and because the area contains anumber of crustal domains where variations in crustal

structure and thickness are expected (Ranganai, 1995;Ranganai et al., 1995), the region has been subdivided intotwo smaller blocks (north and south of latitude 20.5° S).Figure 9 shows the spectrum for the northern block,corresponding to the northern zone of high magneticsignatures (i.e., the separate terrain discussed earlier).Again, the deep and shallow sources are easily separablebut the spectra from the latter show several possibledepths. Reliable depths estimates of 4.1 ± 0.50 km areobtained for the deep ensemble (hb) and 0.90 ± 0.20 km forthe ‘regional’ sources (hr). A shallowest layer (hs) at anaverage depth of 0.4 ± 0.10 km can be determined. Withinthe defined error limits, this gives depths equal to andgreater than the flight height, corresponding tooutcropping features in the area. Again, the 4.0 km depthsuggests that the basement in the northern area is shallowerthan in the southern part since the average depth from theoverall grid is 8 km. The southern part (south of latitude20.5° S) could not be evaluated alone due to the interferenceeffects of the NMZ regional anomaly.

Figure 9. Radially averaged 2D spectra of aeromagnetic data for the northern part of the study area (hb = magneticbasement depth; hr = depth for regional magnetic sources; hs = depth for shallow sources). Dashed lines show slopesused to estimate error quoted in text following the method of Poudjom-Djomani et al. (1995)


R. T. RANGANAI

SUMMARY AND CONCLUSIONS Regional aeromagnetic data from the south-centralZimbabwe craton were processed for geological andstructural mapping to elucidate the tectonic evolution ofthe region. Depths to susceptibility contrasts have beendetermined by spectral analysis and 3D Eulerdeconvolution, with the latter technique also used forstructural interpretation. The well-defined Euler solutionshave confirmed the location of both pre-existing and thenewly interpreted linear geological features, and gaveestimates of their depths, thus confirming the geologicalsignificance of the qualitative interpretation. The NNE,NNW, NW, ENE and WNW structures seen on shadowand derivative maps are spatially well-represented by theEuler solutions. The Umvemeela and East dykes appear toextend beyond their mapped exposures into the LimpopoBelt (UD and ED, Figs. 4 and 5). This observation alsoapplies to the Mwenezi fault (Mw) and the interpreted ENE-trending BKD dykes (Figs. 3 to 7). The intersection patternsof all these features provide relative age constraints onthe time of crustal extension, dyke intrusion, and theLimpopo orogeny (Table 1). Some new NNE to NE-trendingfeatures are defined, particularly in the central-eastern partof the study area. Additional solution patterns suggestthe presence of ultramafic sills, which also supportsprevious qualitative magnetic interpretation and gravitystudies (Ranganai, 1995; Ranganai et al., 1995, Ranganaiand Ebinger, 2008). Structural and lithologic trends havetherefore been established with greater confidence thanwould be possible by magnetic anomaly- geologycorrelation alone. For example, a linear cluster of solutions(depth ~2.0 km) mark the ZC-NMZ boundary, confirminggeological interpretation of a tectonic contact. Overall, five major structural trends (regional lineaments)can be identified and associated with the various geologicalfeatures and craton tectonic events as summarised in Table1 (cf. Fig. 6), based on previous studies and cross-cuttingstructures (Ranganai et al., 1995; Ranganai and Ebinger,2008). The geological-tectonic correlation suggests thatthe interpreted regional trends are mainly 2.6 Ga andyounger, and relate to events including the formation ofthe Limpopo belt and its subsequent tectonic reactivation

at 2.0 Ga (Kamber et al., 1995; Bumby et al., 2001) and themajor regional igneous/ dyking events at 1.8-2.0 Ga(Mashonaland; Wilson et al., 1987; Wilson, 1990), 1.1 Ga(Umkondo; Wilson, 1990; Hanson et al., 1998; Wingate,2001) and 180 Ma (Karoo; Reeves, 2002; Jones et al., 2001;Le Gall et al., 2002). For example, EW to ESE dykes in thesouth-western corner form the eastern extension of the>1000 km long Botswana late Karoo Dyke Swarm that hasbeen mapped across northern Botswana (cf. Reeves, 1978,2000; Wilson et al. 1987; Le Gall et al., 2002). These mayconstitute a failed third arm of a rift triple junctionassociated with the break- up of Gondwana, with the Sabiand Lebombo monoclines forming the other two arms(Reeves, 1978, 2000). Ranganai et al. (1995) conclude thatthe observed structures are due to inter- and intra-cratoniccollisions and block movements involving the Zimbabweand Kaapvaal cratons and the Limpopo Belt, and latercrustal extension during the break-up of Gondwana. Themovements produced structures, or reactivated olderfractures, that were exploited by late Archaean and earlyProterozoic mafic intrusions. Most structures identified are predominantly of shalloworigin, with Euler depths solutions of 2.0 km. A number ofisolated deep Euler solutions are associated with ultramaficcomplexes, the Great Dyke and the Umvimeeela dyke; andthese points could represent the original magma chambersand/or feeder points for these units. There is a generalincrease in solution depths from north to south whichsuggests a tilt of the basement in sympathy with the surfaceterrain (cf. Mushayandebvu, 1995). This probablycorresponds to, or reflects, variable uplift and erosion levelsbetween the two halves of the area, separated by theMchingwe fault. Alternatively, the southern parts couldhave been affected (tilted?) by loading of the area byLimpopo Belt rocks thrust onto the southern edge of thecraton. There are indications of this situation in the spectralanalysis results as well, where the northern area yields ashallower depth to magnetic basement than the overallgrid. Considering the statistical averaging effect of thetechnique, the results mean that the magnetic basement inthe southern parts should be much deeper than the overallgrid result suggests.



47

Trend/Direction Geological Features/Craton Tectonic Events and TimingE-W to ESE/WNW Botswana Karoo Dyke swarm, BKD; Plumtree dyke swarm

(Gondwana break-up: Reeves, 1978, 2000; Wilson et al, 1987; LeGall et a l., 2002) 170-200 Ma

ENE/WSW Sabi-Limpopo dyke swarm (Karoo Igneous event; Wilson et al.1987; Reeves, 2000; Le Gall et al., 2002) 170-200 Ma

Dolerite sills (Not in study area) Umkondo Igneous event (Wilson, 1979, 1990; Wilson et a l., 1987;Hanson et al., 1998). 1100 Ma

Sebanga Poort Dyke set, including Filabusi and Fort Rixon dykes,FRD. (Mashonaland Igneous Event: Wilson, 1979, 1981; Wilson etal. 1987). 1800-2000 Ma

NNW/SSE

NW/SE Mchingwe/Jenya/Mwenezi faults plus others (Dextral shear coupleacting on craton-Wilson, 1990; Campbellet a l., 1992). ~2000 Ma

NNE/SSW Great Dyke, East and Umvimeela dyke, plus Popoteke fault set(Grea t Dyke Fracture System: Wilson, 1979, 1990). 2500 Ma

ENE/WSW NMZ, Chivi/Razi granites, (LB overthrust onto ZC: Van Reenen etal. 1992; Mkweli et al., 1995; Frei et a l., 1999) ~2600 Ma

Table 1. Major Aeromagnetic Structural Trends and their Geological Association.(After Ranganai et al., 1995; Ranganai and Ebinger, 2008) (See Figs. 2 to 6 for comparison)

Spectral analysis of the magnetic data shows that thereare generally three statistical populations of magneticsources: features at average depths of 0.5 to 1.0 km, ‘deep’sources between 2.0 and 4.0 km, and regional features at8.0 km. Except for the last depth which most probably mapsthe magnetic basement in the area, these results arecomparable to estimates from Euler deconvolution, whereaverage depths of 0.5, 1.0 and 2.5 km are obtained. Theshallowest depths can be easily associated with the knowngeological units: the out-cropping and near surface (i.e.sub-cropping) features such as mafic dykes, iron formationhorizons in greenstone belts, and ultramafic intrusions.Significantly, magnetic modelling of several profiles acrossthe Umvimeela and East dykes have yielded depthsgenerally ranging from 100 m to 300 m (Mushayandebvu,1995). This lends support to our results for the shallowestfeatures, particularly the magnetic dykes as well as the BIFand ultramafic horizons within the greenstone belts. Themagnetic basement depth of 8 km corresponds to the crustalboundary estimated at about 7-9 km from gravity andseismic data. An important implication is that the

greenstone belts were an integral part of the lithospherebefore much of the upper crustal (brittle) deformationoccurred. The geostructural framework of the area iscompatible with the postulated late Archaean collisioninvolving the Zimbabwe and Kaapvaal cratons and theLimpopo Belt, and later crustal extension during the break-up of Gondwana.

ACKNOWLEDGEMENTS Edgar Stettler, Oswald Gwavava and Sue Webb are thankedfor their constructive comments on an earlier version ofthe paper. Advice and encouragement given by Dai Jonesand Alan Reid at various stages of the study is greatlyappreciated. The Zimbabwe Geological Survey providedthe aeromagnetic data and gave permission for the data tobe published. The study benefited from the British CouncilLink Scheme between the Departments of Earth Sciences(University of Leeds) and Physics (University ofZimbabwe).


R. T. RANGANAI

REFERENCES

Allek, K., Mohamed Hamoudi, M., 2008. Regional-scaleaeromagnetic survey of the south-west of Algeria: Atool for area selection for diamond exploration. Journalof African Earth Sciences 50, 67–78

Barbosa, V. C. F., Silva, J. B. C., and Medeiros, W. E., 1999.Stability analysis and improvement of structural indexestimation in Euler deconvolution: Geophysics 64, 48–60.

Barritt, S.D., 1993. The African Magnetic Mapping Project,ITC Journal 1993-2, 122-131.

Bates, M. P. and Mushayandebvu, M. F., 1995. Magneticfabric in the Umvimeela Dyke, satellite of the GreatDyke, Zimbabwe. Tectonophysics 242, 241-254.

Bickle, M. J. and Nisbet, E. G. (Eds.), 1993. The geology ofthe Belingwe greenstone belt, Zimbabwe: A study ofthe evolution of Archaean continental crust.Geological Society Zimbabwe Special Publication 2,239p. A. A. Balkema, Rotterdam.

Blakely, R. J., 1995. Potential Theory in Gravity andMagnetic Applications. 441p. Cambridge.

Blenkinsop, T.G., Martin, A., Jelsma, H.A., Vinyu, M.L., 1997.The Zimbabwe Craton. In: de Wit, M.J., Ashwal, L.D.(Eds.), Greenstone Belts. Oxford monograph ongeology and geophysics, Oxford University Press,Oxford, p. 567-580.

Bournas, N., Galdeano, A., Hamoudi, M., Baker, H., 2003.Interpretation of the aeromagnetic map of EasternHoggar (Algeria) using the Euler deconvolution,analytic signal and local wavenumber methods.Journal of African Earth Sciences 37, 191–205

Bumby, A.J., Eriksson, P.G., Van Der Merwe, R., Brümmer,J.J., 2001. Shear-zone controlled basins in the Bloubergarea, Northern Province, South Africa: syn- and post-tectonic sedimentation relating to ca. 2.0 Gareactivation of the Limpopo Belt. Journal of AfricanEarth Sciences 33, 445-461

Campbell, S.D.G, Oesterlen, P.M., Blenkinsop, T.G., Pitfield,P.E.J. and Munyanyiwa, H. 1992. A Provisional 1:2 500000 scale Tectonic map and the tectonic evolution ofZimbabwe. Annals of the Zimbabwe GeologicalSurvey, XVI (1991), 31-50.

Cowan, D. R. and Cowan, S., 1993. Separation filteringapplied to aeromagnetic data. Exploration Geophysics24, 429-436.

Fairhead, D.J., Bennett, K., Gordon, D.H. and Huang, D.,1994. EULER: beyond the black-box. 64th AnnualInternational Meeting, Society of ExplorationGeophysicists, Expanded Abstracts, 422-424.

Fedo, C.M., Eriksson, K. and Blenkinsop, T.G. 1995. Geologichistory of the Archean Buhwa Greenstone Belt andsurrounding granite-gneiss terrane, Zimbabwe, withimplications for the evolution of the Limpopo Belt.Canadian Journal of Earth Sciences, 32, 1977-1990.

Frei, R., Blenkinsop, T.G., and Schönberg, R., 1999,Geochronology of the late Archaean Razi andChilimanzi suites of granites in Zimbabwe: implicationsfor the late Archaean tectonics of the Limpopo beltand Zimbabwe craton: South African Journal ofGeology 102, 55-63.

Geosoft, 2004. Oasis Montaj (V5.1.8) and Euler 3DDeconvolution System (V5.1.5) manuals. Geosoft Inc.,Toronto. Canada.

Grant, F. S., 1985. Aeromagnetics, geology and oreenvironments, I. Magnetite in igneous, sedimentaryand metamorphic rocks: an overview. Geoexploration23, 303-333.

Gwavava, O., Swain, C.J., Podmore, F. and Fairhead, I.D.,1992. Evidence of crustal thinning beneath theLimpopo Belt and Lebombo monocline of southernAfrica based on regional gravity studies andimplications for the reconstruction of Gondwana.Tectonophysics 212, 1-20.

Hahn, A., Kind, E.G., and Mishra, D.C., 1976. Depthestimation of magnetic sources by means of Fourieramplitude spectra. Geophysical Prospecting 24, 287-308.

Hanson, R.E., Martin, M.W., Bowring, S.A., Munyanyiwa,H., 1998. U–Pb zircon age for the Umkondo dolerites,eastern Zimbabwe: 1.1 Ga large igneous province insouthern Africa–east Antarctica and possible Rodiniacorrelations. Geology 12, 1143– 1146.

Hickman, M. H., 1978. Isotopic evidence for crustalreworking in the Rhodesian Archaean craton,southern Africa. Geology 6, 214-216.

Hinze, W. J. (Ed.), 1985. The utility of regional gravity andmagnetic anomaly maps . Society ExploratinGeophysicists. Tulsa. 454p.

Horstwood, M.S.A., Nesbitt, R.W., Noble, S.R. and Wilson,J.F., 1999. U-Pb zircon evidence for an extensive earlyArchean craton in Zimbabwe: a reassessment of thetiming of craton formation, stabilization and growth.Geology 27, 707-710.

Jelsma, H.A., Dirks, P.H.G.M., 2002. Neoarchaean tectonicevolution of the Zimbabwe Craton. In Fowler, C.M.R.,Ebinger, C., Hawkesworth, C.J. (Editors), The EarlyEarth: Physical, Chemical and Biological Development.Geological Society of London, Special Publications,199, 183-211.



49

Jelsma, H.A., Vinyu, M.L., Valbracht, P.J., Davies, G.R.,Wijbrans, J.R., Verdurmen, E.A.T., 1996. Constraintson Archaean crustal evolution of the Zimbabwecraton: U-Pb zircon, Sm-Nd and Pb-Pb whole-rockisotope study. Contribution Mineralogy andPetrology 124, 55-70.

Jones, D.L., Bates, M.P., Podmore, F. and Mushayandebvu,M.F., 1995. The Great Dyke of Zimbabwe and ItsSatellites: Recent Geophysical Results and TheirImplications. In: Srivastava, R.K. and Chandra, R. (Eds),Magmatism in Relation to Diverse Tectonic Settings,Oxford and IBH Publishing Co. Pvt. Ltd, p209-222.

Jones, D.L., Duncan, R.A., Briden, J.C., Randall, D.E., Mac-Niocaill, C., 2001. Age of the Batoka basalts, northernZimbabwe, and the duration of Karoo large igneousprovince magmatism, G3 Geochemistry, Geophysics,Geosystems 2, 1 –15.

Kamber, B.S., Kramers, J.D., Napier, R., Cliff, R.A. andRollinson, H.R. 1995. The Triangle Shearzone,Zimbabwe, revisited: new data document an importantevent at 2.0 Ga. in the Limpopo Belt. PrecambrianResearch 70, 191-213.

Kusky, T.M., 1998. Tectonic setting and terrane accretionof the Archean Zimbabwe craton. Geology 26, 163-166.

Le Gall, B., Tshoso, G., Jourdan, F., Fe´raud, G., Bertrand,H., Tiercelin, J.J., Kampunzu, A.B., Modisi, M.P.,Dyment, J., Maia, M., 2002. 40Ar/39Ar geochronologyand structural data from the giant Okavango andrelated mafic dyke swarms, Karoo igneous province,Botswana. Earth and Planetary Science Letters 202,595– 606.

McDonald, A. J. W., Fletcher, C. J. N., Carruthers, R. M.,Wilson, D. and Evans, R. B., 1992. Interpretation ofthe regional gravity and magnetic surveys of Wales,using shaded relief and Euler deconvolutiontechniques. Geological Magazine 129, 523-531.

Mkweli, S., Kamber, B. and Berger, M. 1995. Westwardcontinuation of the craton-Limpopo Belt tectonic breakin Zimbabwe and new age constraints on the timingof the thrusting. Journal of the Geological Society,London 152, 77-83.

Mukasa, S.B., Wilson, A.H. and Carlson, R.W. 1998. Amultielement geochronologic study of the Great Dyke,Zimbabwe: significance of the robust and reset ages.Earth and Planetary Science Letters 164 (1/2), 353-369.

Mushayandebvu, M. F., 1995. Magnetic modelling of theUmvimeela and East dykes: Evidence for regionaltilting of the Zimbabwe craton adjacent to the LimpopoBelt. Journal of Applied Science in Southern Africa1, 47-58.

Mushayandebvu, M. F., van Driel, P., Reid, A.B., andFairhead, J.D., 2001. Magnetic source parameters oftwo dimensional structures using extended Eulerdeconvolution: Geophysics 66, 814-823.

Mushayandebvu, M.F., Lesurz, V., Reid, A.B., Fairhead,D.J., 2004. Grid Euler deconvolution with constraintsfor 2D structures. Geophysics 69(2), 489–496

Nisbet, E.G. 1987. The Young Earth: an introduction toArchaean geology. Boston, Allen and Unwin, 402p.

Oberthür, T.; Davis, D. W.; Blenkinsop, T. G.; and Höhndorf,A. 2002. Precise U-Pb mineral ages, Rb-Sr and Sm-Ndsystematics of the Great Dyke, Zimbabwe: constraintson late Archean events in the Zimbabwe Craton andLimpopo Belt. Precambrian Research 113: 293–305.

Paterson, N. R., Kwan, K. C. H. and Reford, S. W., 1992.Use of Euler deconvolution in recognizing magneticanomalies of pipelike bodies. 62nd AnnualInternational Meeting, Society of ExplorationGeophysicists, SEG Expanded Abstracts, Volome 1,642-645.

Podmore, F. and Wilson, A. H., 1987. A reappraisal of thestructure and geology of the Great Dyke, Zimbabwe.In: Halls, H. C. and Fahrig, W. F. (Eds.), Mafic DykeSwarms. Geological Association Canada, Special Paper33, 317-330.

Poudjom-Djomani, Y. H., Ebinger, C. E. Diament, M.,Fairhead, J. D., 1995. Effective elastic thicknessvariations in West-Central Africa inferred from gravitydata. Journal of Geophysical Research 100(B11),22047-22070.

Prendergast, M.D., 2004. The Bulawayan Supergroup: alate Archaean passive margin-related large igneousprovince in the Zimbabwe craton. Journal of theGeological Society 161(3): 431 - 445.

Prendergast, M.D., Wingate, M.T.D., 2007. Zircongeochronology and partial structural re-interpretationof the late Archaean Mashaba Igneous Complex,south-central Zimbabwe. South African Journal ofGeology 110(4), 585-596.

Ranganai, R.T., 1995. Geophysical Investigations of theGranite-Greenstone Terrain in the South-CentralZimbabwe Archaean Craton. Ph D Thesis 288p,University of Leeds, Leeds, UK.

Ranganai, R.T. 1999. Structure and Depth Mapping in thesouth-central Zimbabwe Craton using 3D EulerDeconvolution and Spectral Analysis of RegionalAeromagnetiv Data. Journal of African Earth Sciences28(4A), 67-68.

Ranganai, R.T. and Ebinger, C.J., 2008. Aeromagnetic andLANDSAT TM Structural Interpretation forIdentifying Regional Groundwater Exploration Targets,South-Central Zimbabwe Craton. Journal of AppliedGeophysics (in press).


R. T. RANGANAI

Ranganai, R.T., Whaler, K.A. and Ebinger, C.E. 2008. Gravityanomaly patterns in the south-central Zimbabwe(Archaean) craton and their geological interpretation.Journal of African Earth Sciences 51 (In press). http://dx.doi.org/10.1016/j.jafrearsci.2008.01.011

Ranganai, R.T., Whaler, K.A., Ebinger, C.E., Stuart, G.W.,1995. Crustal structure of the south-central ZimbabweArchaean craton from Gravity and Aeromagnetic data:Implications for tectonic evolution. EAEG ExtendedAbstracts, Paper D28, Glasgow.

Reeves, C.V., 1978. A failed Gondwana spreading axis insouthern Africa. Nature 273, 222-223.

Reeves, C.V. 2000. The geophysical mapping of Mesozoicdyke swarms in southern Africa and their origin in thedisruption of Gondwana. Journal of African EarthSciences 30, 499-513.

Reid, A. B., Allsop, J. M., Granser, H., Millet, A. J. andSomerton, I. W., 1990. Magnetic interpretation in threedimensions using Euler deconvolution. Geophysics55, 80-91.

Rollinson, H.R. and Blenkinsop, T.G. 1995. The magmatic,metamorphic and tectonic evolution of the NorthernMarginal Zone of the Limpopo Belt in Zimbabwe.Journal of the Geological Society of London 152,65-75.

Ruotoistenmaki, T., 1983. Depth estimation from potentialfield data using the Fourier amplitude spectra.Geoexploration 21, 191-201.

Ruotoistenmaki, T., 1987. Estimation of depth to potentialfield sources using the Fourier amplitude spectrum.Geological Survey Finland, Bulletin 340, 84p.

Schoenberg, R., Nägler, T.F., Gnos, E., Kramers, J.D.,Kamber, B.S., 2003. The source of the Great Dyke,Zimbabwe, and its tectonic significance: evidence fromRe-Os isotopes. Journal of Geology 111, 565–578.

Spector, A. and Grant, F. S., 1970. Statistical models forinterpreting aeromagnetic data. Geophysics 35, 293-302.

Spector, A. and Grant, F. S., 1975. Comments on ‘Two-dimensional Power Spectral Analysis of AeromagneticFields’. Geophyscal Prospecting 23, 391.

Spector, A. and Parker, W., 1979. Computer compilationand interpretation of geophysical data. GeologicalSurvey Canada, Economic Geolology Report 31, 527-544.

Smith, W.H.F., Wessel, P., 1990. Gridding with continuouscurvature splines in tension. Geophysics 55, 293-305.

Taylor, P.N., Kramers, D.J., Moorbath, S., Wilson, J.F.,Orpen, J.L. and Martin, A. 1991. Pb/Pb, Sm-Nd andRb-Sr geochronology in the Archaean craton ofZimbabwe. Chemical Geology (Isotope Geosciences)87, 175-196.

Thompson, D. T., 1982. EULDPH: a new technique forcomputer-assisted depth estimates from magneticdata. Geophysics 47, 31-37.

Van Reenen, D. D., Roering, C., Ashwal, L. D. and de Wit,M. J. (Eds.), 1992. The Archaean Limpopo granulitebelt: Tectonics and deep crustal processes.Precambrian Research 55, 1-587.

Wilson, A. H. and Prendergast, M. D., 1988. The GreatDyke of Zimbabwe-I: tectonic setting, stratigraphy,petrology, structure, emplacement and crystallization.In: Prendergast, M. D. and Jones, M. J. (Eds.),Magmatic Sulphides- the Zimbabwe Volume, pp1-20.Institute of Mining and Metallurgy.

Wilson, J. F. 1979. A preliminary reappraisal of theRhodesian Basement Complex. Geological Society ofSouth Africa Special Publication 5, 1-23.

Wilson, J. F., 1981. The granite-gneiss greenstone shield,Zimbabwe. In: Hunter, D. R. (Ed.), Precambrian of thesouthern hemisphere, Elsevier, 454-488.

Wilson, J. F., 1990. A craton and its cracks: some of thebehaviour of the Zimbabwe block from the LateArchaean to the Mesozoic in response to horizontalmovements, and the significance of some of its maficdyke fracture patterns. Journal African Earth Sciences10, 483-501.

Wilson, J. F., Jones, D. L. and Kramers, J. D., 1987. Maficdyke swarms in Zimbabwe. In: Halls, H.C. and Fahrig,W.F. (Eds.), Mafic Dyke Swarms. GeologicalAssociation Canada, Special Paper 33, 433-444.

Wilson, J.F., Nesbitt, R.W. and Fanning, C.M. 1995. Zircongeochronology of Archaean felsic sequences in theZimbabwe craton: a revision of greenstonestratigraphy and a model for crustal growth. In:Coward, M.P. & Ries, A.C. (Eds.), Early PrecambrianProcesses, Geological Society Special Publication 95,109-126.

Wingate, M.T.D., 2001. SHRIMP baddeleyite and zirconages for an Umkondo dolerite sill, Nyanga Mountains,Eastern Zimbabwe. South African Journal of Geology104, 13-22.



AN ASSESSMENT OF THE PRESENCE OF HEAVY METALS INTHE SEDIMENTS OF THE LOWER MVOTI RIVER SYSTEM

Prisha Sukdeo1, Prisha Sukdeo, Srinivasan Pillay1 and Ajay Bissessur .2

1School of Environmental Sciences, University of KwaZulu Natal,Durban, 4000, South Africa

2School of Chemistry, University of KwaZulu Natal, Durban, 4000, South Africa


ABSTRACT:- Excessive levels of heavy metals present in aquatic systems are often a result ofanthropogenic activities. Sediment analysis for this type of contamination is often preferred over thedynamic water column. Due to accumulation of these elements over time, sediment analysis canprovide a pollution-history for the site. Heavy metals at elevated levels are potentially toxic toaquatic life, and, because they bio-accumulate in food webs, are also potentially detrimental tohuman life. This study assesses the presence of heavy metals in the lower Mvoti River and Estuary.Levels of aluminium, arsenic, chromium, copper, iron, lead, magnesium, manganese, nickel, titanium,vanadium and zinc were determined using Inductively Coupled Plasma Optical Emission Spectroscopy(ICP-OES). The results show that the riverine and estuarine sites closest to industrial effluentdischarge sites and informal settlements displayed the highest levels of heavy metal contamination.The results of the estuarine analysis were compared to current levels of heavy metals present in twoother South African estuaries : the St. Lucia Estuary, also located on the north of KwaZulu Natal andthe Swartkops Estuary in the Eastern Cape, as well as two international estuaries. Even though thelower Mvoti River and Estuary does experience some heavy metal sediment contamination, theabove-mentioned comparisons illustrate the level of contamination is relatively low in comparisonto other ecologically significant South African estuaries, and selected international estuaries. Withrespect to heavy metals, these results bode well for the Mvoti, a system historically reported to be inserious ecological degradation from other pollution sources.

INTRODUCTION

Coastal seas inevitably receive much of the effluent ofthe world. Rivers are responsible for transporting a rangeof both dissolved and particulate matter from land intothe sea (Klavins et al., 2000). These contaminants mostoften derived directly from human activities, aresometimes harboured in estuaries and other coastalembayments before being flushed out to sea.Contaminants like potentially toxic metals are introducedinto the environment either naturally or anthropogenically, or both (Harikumar & Jisha, 2010). Despite being naturalcomponents of the earth’s crust, the severity ofcontaminants like heavy metals in the environment hasdrastically increased, primarily due to anthropogenicactivities (Harikumar & Jisha, 2010; Chen & Kandasamy,2008).

Due to its variable physical and chemical properties, thesediment component of aquatic systems is often the largeraccumulator, and acts as potential sinks of contaminants,(Sundarajen & Natesan, 2010). The presence of high levelsof heavy metals in the sediments of a system is a possibleindication of human-induced pollution, derived fromanthropogenic activities, as opposed to natural processeslike weathering and erosion (Binning & Baird, 2001; Klavinset al., 2000). According to Nriagu & Pacyna (1988) in Chen& Kandasamy (2008), human induced inputs of metals likearsenic, nickel and zinc into the environment are oftenmore than twice the input of the same metals from naturalsources. The presence of these elements wheninvestigating heavy metal contamination, can be detectedby analysing water, sediments or biota. However, analysisof sediments are often more reliable and advantageousthan the water column, as sediments assimilate these


S. PILLAY

pollutants over time, while the water column on the otherhand, is fairly dynamic and experiences significantchanges over space and time (Landajo et al., 2004; Binning& Baird, 2001).

The absence or very low levels of contaminants in thewater column may be due to the accumulation of heavymetals by the sediment component of the system overtime. As a result of this sediment accumulation, someindication of the history of pollution in a system may beunderstood (Chatterjee et al., 2006; Landajo et al., 2004;Binning & Baird, 2001). In aquatic systems the mostcommon sources of heavy metals are usually industrialeffluent discharge (Topalián et al., 1999). As noted byMartin & Whitfield (1983), in Binning & Baird (2001), inaddition, around 90% of the particulate material carriedby rivers settles in estuaries and coastal areas.

Another important factor to consider when investigatingheavy metal contamination of aquatic sediments is therelationship between pollutant contamination andsediment texture. The finer components of sediment, likeclays and silts are more likely to carry and accumulateheavy metal and organic contaminants in aquatic systemsas opposed to the larger components of the sediments(Palanques et al., 2008; Binning & Baird, 2001). This isdue to their relatively high adsorption ability (Chatterjeeet al., 2006). This adsorption is dependant upon physico-chemical factors of the system like pH, dissolved oxygen,and oxidation-reduction potential (Ghrefat & Yusuf, 2006).However, contaminants are not necessarily fixed tosediments and are often re-mobilised via various chemical,physical and biological processes (Landajo et al., 2004).Sometimes, they are released into the water column andbecome available to living organisms (Landajo et al.,2004). There is also potential for the bio-accumulation ofthese contaminants in food webs (Chatterjee et al., 2006;Jordao et al., 2007) resulting in possible detrimental effectsto biota and humans (Harikumar & Jisha, 2010).

This study focuses on the heavy metal characterization ofthe lower reaches of the Mvoti River, located approximately70 km north of Durban on the KwaZulu- Natal coastal zone.The lower Mvoti River flows through a highly modifiedregion of northern KwaZulu Natal. Much of the catchmentis under agriculture, and sugarcane farming is extensive inthe lower regions. Despite being one of South Africa’srelatively smaller systems, the Mvoti River is an importantresource for a number of towns, settlements and industrialdevelopments along its approximately 197km long course(Wepener, 2007). Hence the river is subjected to a range ofeffects and influences associated with human activities.

In the upper and middle catchment of the river, agricultural,rural and informal users dominate, whilst in the lowercatchment a distillery, sugar mill, pulp and paper mill aswell as the sewerage works associated with the coastaltown of KwaDukuza (formerly known as Stanger) are theprincipal users of the Mvoti. Two large tributaries, theNtshaweni and Mbozambo rivers enter the Mvoti in thelower reaches (Malherbe et al., 2010). Prior to its confluencewith the Mvoti River, the Ntshaweni River receives effluentdischarge and return flow from milling processes locatedin the area , and the Mbozambo River is the recipient oftreated sewage and waste water from the town of Stanger.A short distance downstream, the river- dominated MvotiEstuary receives this water before it eventually drains intothe Indian Ocean. Since estuarine systems receive andfrequently accumulate these catchment-derived pollutants,and considering the relatively poor water quality the lowerMvoti River system experiences (Sukdeo et al., 2010), theMvoti estuary is a potential contaminant sink for heavymetals with possible adverse biotic effects. Although there has been significant research conductedregarding heavy metal contamination within estuarinesediments, and despite the vast anthropogenic influenceswithin the study region, a significant assessment of suchcontamination within the Mvoti Estuary has not previouslybeen attempted. Considering these factors, an evaluationof the abundance of heavy metals in this river system wouldbe beneficial for future management of the system. Hencethe purpose of this study is to provide an assessment ofthe heavy metals present in the sediments of the lowerMvoti River and Estuary, and to compare these resultswith levels of heavy metals present in the St. Lucia estuary- another ecologically significant system also on thenorthern coast of KwaZulu Natal, but one that is regardedto be in fairly pristine condition.

MATERIALSAND METHODS

Sampling Sites

A total of eight sample sites in the lower Mvoti Riversystem were selected, of which five were located withinthe estuary itself. The remaining three sites were withinthe lower reaches of the river, but upstream of the estuarinereaches.

The extraction of samples was carried out using aPolyvinylchloride (PVC) Pipe Sediment Extractor. At eachsite, samples were collected at three different points: onemidstream, and others midway between the midstream pointand either bank. As this study aimed to assess the most


An Assessment of the Presence of Heavy Metals in the Sediments of the Lower Mvoti River System

53

Figure 1. Location of the Mvoti Estuary in relation to the St. Lucia and Swartkops Estuaries along the SouthAfrican coastline

Figure 2. Map of the lower Mvoti River and estuarine system indicating the sampling sites selected for the study


S. PILLAY

recent heavy metal pollution due to increasedanthropogenic activity, the upper 30 cm of the river andestuarine bed sediments were specifically sampled.Following collection, samples were sealed in polyethylenejars, stored at low temperatures, and sent for chemicalanalysis within 24 hours.

Sediment particle size analysis

Textural analyses to determine the particle size distributionof the samples was conducted in the soil sciencelaboratory by using a standard dry sieving technique(Abed, 2006). Equal proportions of the three samplesweighing approximately 500 g for each site werehomogenised and oven-dried at 110ºC for 48 hours.Samples were then disaggregated using a pestle and mortar,and split using a riffle box sample splitter. One portion ofthe split samples were used to determine particle size usinga Retsch sieve shaker, and the remaining portion wasreserved for chemical analysis.

Chemical analysis to determine the heavy metalconcentrations was carried out using Inductively CoupledPlasma Atomic Emission Spectroscopy (ICP-OES)

(Stoeppler, 1992 in Abed, 2006). According to Walsh (1997),in Abed (2006), ICP-OES measures atomic spectra of theelements being determined. Analysis by ICP offers a greateradvantage in terms of sensitivity and freedom frominterference. For the purpose of this study the followingmetals, namely, aluminium (Al), arsenic (As), chromium (Cr),copper (Cu), iron (Fe), magnesium (Mg), manganese (Mn),nickel (Ni), lead (Pb), titanium (Ti), vanadium (V) and zinc(Zn) were measured by the ICP- OES method. Statistical analysis The T-test was used to ascertain any differences betweenheavy metal concentrations in the sediments of the Mvotiand St. Lucia estuaries. Data for the St. Lucia Estuary froma prior study (Ajee, et al., 2010) was made available to thisstudy for comparative purposes. All statistical analyseswere completed using SPSS version 15.0 for Windows.


The mean concentrations of heavy metals determined ateach site are illustrated Table 1.

Table 1. Heavy metal concentrations (ppm) and mean (SE) at each sample site

Sample Site Heavy metal concentrations (ppm) in the sediments of the estuarine section Al As Cr Cu Fe Mg Mn Ni Pb Ti V Zn

Site E1 29.1 0.03 0.5 0 65.66 31.89 0.98 0 0.11 2.47 0.02 0Site E2 50.6 0.37 0.11 0 64.99 40.23 1.27 0 0 5.8 0.1 0.07Site E3 67.53 0 0.19 0.07 74.33 41.28 1.32 0.08 0.76 9.87 0.24 0Site E4 58.69 0.17 0 0.22 123.96 42.05 0.98 0.11 0.97 16.14 0.39 0.1Site E5 75.76 0.2 0.43 1.23 181.5 51.99 2.25 0.16 1.13 21.17 0.55 0.97

Mean (SEₓ) 56.34 0.15 0.16 0.3 102.09 41.49 1.36 0.07 0.59 11.09 0.26 0.23

Sample Site Mean (SEₓ) heavy metal concentrations (ppm) in the sediments of the riverine section of the study area

Al As Cr Cu Fe Mg Mn Ni Pb Ti V ZnSite R1 365.7 0.9 1.27 0.21 603.6 54.24 7.6 1.14 4.23 48.48 1.91 1.65Site R2 57.01 0 0.92 0.03 211.14 22.44 1.34 0.06 0.13 26.79 0.44 0.54Site R3 67.95 0 0.17 0 117.8 11.8 10.45 0 0.05 10.17 0.26 0.03

Mean (SEₓ) 163.55 0.3 0.79 0.08 310.85 29.49 6.46 0.4 1.47 28.48 0.87 0.74



55

The upper reaches of the estuary, site E5 (Fig.1), displayedthe highest levels of all the metals tested for within theestuarine section of the study area, with the exception ofarsenic. This is the closest estuarine site where the MvotiRiver experiences industrial discharge and utilization ofthe river for domestic purposes by informal settlements.At this point and at site E4, the estuary widens anddecreases in depth due to excessive sedimentation, andbecomes slow-flowing. The lagoon-like characteristics ofE4 and E5 facilitate the settling of contaminants in thisregion. There is a general decrease in the concentrationsof the heavy metals (with the exception of aluminium,which has a higher concentration at E3 than at E4) for theremaining estuary sites, approaching the Mvoti mouth.This decrease is signifcantly evident at site E1, closest tothe Mvoti mouth, where levels of contaminants are at theirlowest. The highest concentrations of heavy metals tested for,with the exception of manganese, was experienced atsample site R1across both the riverine and estuarine sites.Site R1 was strategically chosen as it is directly under anN2 (national road) bridge and downstream at a point wherethe Mbozambo and Ntshaweni Rivers join the Mvoti.Effluent according to Malherbe ( 2006), from the pulp andpaper mill, the sugar mill, and the Stanger sewage worksenter into the Mvoti River slightly upstream of this site. Inaddition, the site is significantly influenced by informalcommunities, adjacent to the river, who use the river mainlyfor domestic purposes. In the vicinity of site R2 (locatedwhere the Mvoti River intersects the town of Groutville)the river is used extensively for domestic purposes as wellas subsistence agriculture by the local community. This ispossibly why the majority of heavy metal concentrationsat this site were relatively higher in comparison to site R3.The lowest concentrations of heavy metals was recordedat site R3 (adjacent the Glendale Distillery). In contrast thehighest concentration of manganese was recorded at this .The main use of the river at this point is the abstraction ofwater by the distillery, for use in its cooling processesthererafter in the irrigation of the surrounding sugarcanefields. According to Ram et al. (2009), the grain size effect hassignificant bearing on the concentration of contaminantspresent within the sediment, therefore to compensate forthis, the results were normalized for textural variations insediment. The dependency of contaminant accumulation

on sediment grain size is well documented (Palanques etal., 2008; Binning & Baird, 2001). This is clearly observedin both the estuarine and riverine sections of the studyarea where increases in contaminant concentrations aremore likely associated with finer sediments. Estuarine site5 has a higher concentration of heavy metals than theremaining estuarine sites which show decreases inconcentration toward the river mouth. In addition, site 5possesses higher amounts of fine sand and fines (veryfine sand, silts and clays) compared to the other estuarinesites. Higher proportions of fine sediments located at sitesaway from the river mouth imply a proportional relationshipbetween finer sediment and contaminant accumulation. Thesediment composition of the Mvoti system is not typicalof documented estuarine sediments, where the highestproportion of fines is found within the estuary. This ispossibly due to the fact that the Mvoti is a perched, river-dominated system, with little or no significant marineinfluences. Hence there is a low accumulation of mud andfine sediments due to reduction in flocculation affected bylow salinity. As a consequence of regular fluvial outflows,scouring of fine material is a common feature. The site R1 with the highest percentage of fine sedimentwithin the riverine section contains a high concentrationof heavy metals. This unusually high proportion of finersediment at the riverine sites (R1 in particular), most likelyis a contribution from the Mbozambo and Ntshaweni rivers.Excessive siltation of the Mvoti River system according toSukdeo et al., (2010) is due to extensive and uncontrolledsandmining, riparian zone disturbances and agriculturealong the river.A similar study on the properties of sediments in the St.Lucia Estuary was conducted by Agjee et al. (2010). Thissystem, located also on the north coast of KwaZulu Natal,is one of the most ecologically significant systems in SouthAfrica due to its rich biodiversity. St.Lucia, a naturalheritage site, is regarded as a fairly pristine environmentbut has recently been under threat of that status due topressures from increased tourism and catchment activities.This is evident from the fact that five of the eight heavymetals analysed in a study by Agjee et al. (2010)correspond to those assessed in this study. Sampletechniques and chemical methods of analysis are similar inboth studies, thus enabling a comparison between the twoestuaries. A comparison of heavy metals between the StLucia by Agjee et al. (2010) and Mvoti estuaries arepresented in Table 3.)


S. PILLAY

Estuary Heavy metal concentrations (ppm)Cr Cu Ni Pb Zn

St. Lucia 123.89 47 63.44 16 46.89Mvoti 0.16 0.3 0.07 0.59 0.28Mean 62.025 23.65 127.02 8.295 23.59Std. Deviation 0.168 0.525 0.586 0.511 0.417

P- values <0.05 <0.05 <0.05 <0.05 <0.05*p-Value refers to the significant value obtained for the statistical procedure. Where p < 0.005, results are deemed significant

Table 3. Heavy metals (ppm) within the sediments of the St. Lucia Estuary (Agjee et al,. 2010) and the Mvoti Estuary

The results of the statistical analysis show that significantdifferences exist between the two data sets illustrated inTable 3. The high concentrations of chromium and nickelaccording Agjee et al., (2010), may be attribute tocatchment geology. It has been suggested that theseelements were mobilised and transferred into the systemvia leaching of minerals from the catchment area.Tributaries have also been known to be a potential sourceof these elements, and a protracted drought in the areahas accelerated the accumulation of metals in the system(Agjee et al., 2010). The concentrations of heavy metalsare significantly higher within the sediments of the St.Lucia estuary, in comparison to the Mvoti Estuary.

The Mvoti Estuary has, by comparison, relatively lowerlevels of heavy metals (contaminants) than the othersystems as shown in table 4. Shown also in table 4 is themean concentrations of selected heavy metals present intwo international estuaries, namely Galvesto Bay in Texas

(Morse et al.,1993. In Binning & Baird, 2001), and Hudson-Raritan Estuary in New York Hence (Wolfe et al., 1996. InBinning & Baird, 2001) , despite its reported polluted waterquality status and the numerous other environmentalproblems which the estuary experiences (Sukdeo et al.,2010), the sediments of the Mvoti estuary stilldemonstrates fair ly low levels of heavy metalcontamination in comparison to other important SouthAfrican, as well as international systems.

Despite the data from these studies have been collectedmore than a decade ago it has been used as a comparatorto show the superior status of current South Africanestuaries over its American rivals. This may be possiblyattributed to relatively low metal contributions fromcatchment geology and less point anthropogenic sourcesof contamination. In addition, the Mvoti estuary itself asa perched estuary possess strong fluvial outflow thatallows for continual flushing of sediments especially atthe estuary mouth.

Table 4. Comparative amounts of heavy metals (ppm) within the sediments of two international and two South Africanestuaries

Heavy metal concentrations (ppm)Heavy metal

Galvesto Bay (Texas)

Hudson-Raritan Estuary (New York)

Swartkops Estuary (South Africa)

Mvoti Estuary (South Africa)

Cr 37 122 20.3 0.16Cu 8 142 6.8 0.3Mn 605 No data available 115 1.36

Pb 25 160 33 0.59Ti No data available 21.7 99 11.9Zn 55 299 36 0.23*Galvesto Bay in Texas (Adapted from Morse et al.,1993. In Binning & Baird, 2001) *Hudson-Raritan Estuary in New York Hence (Adapted from Wolfe et al., 1996. In Binning & Baird, 2001)*Swartkops Estuary in South Africa (After Binning & Baird, 2001)



57

CONCLUSION

In this study it was found that that even though heavymetal contamination is present in the Mvoti system, it isrelatively lower than those experienced by other selectedestuaries. The distribution of heavy metals within thesystem is a possibility of primarily industrial effluentdischarge into the system and domestic utilization of theriver. However, further, in-depth analysis is required toconfirm that these results are either due to domestic andindustrial use in isolation, or the cumulative effect ofimproper catchment practices and management. Theincreased contamination and accumulation of heavymetals in riverine and estuarine sediments is a major causefor concern, as these elements are often remobilised intothe water column and accumulate in food webs withdetrimental end-results. It is therefore imperative to monitorthe Mvoti sediments, thus preventing increasing levels ofpotentially toxic contaminants. This will also assist inimproving the overall health condition of the system thatis currently in an extremely poor condition.

REFERENCES

Abed, R. (2006). A study of the heavy metal content ofsediments in selected KwaZulu Natal estuaries.Unpublished Honours thesis. University of KwaZuluNatal, Durban, South Africa.

Agjee, N., Pillay, K., & Pillay, S. (2010). An assessment ofsediment heavy metal concentrations in the lower St.Lucia Estuary, KwaZulu Natal, South Africa.Manuscript submitted for publication.

Binning, K. & Baird, D. (2001). Survey of heavy metals inthe sediments of the Swartkops River Estuary, PortElizabeth, South Africa. Water SA. Vol 27. No 4. 461-466 pp.

Chatterjee, M., Silva Filho, E. V., Sarkar, S. K., Sella, S. M.,Bhattacharya, A., Satpathy, K. K., Prasad, M. V. R.,Chakraborty, S. & Bhattacharya B. D. (2006).Distribution and possible source of trace elements inthe sediment cores of a tropical macrotidal estuaryand their ecotoxilogical significance.

Chen, T. C & Kandasamy S. (2008). Evaluation of elementalenrichments in surface sediments off southwesternTaiwan. Environmental Geology.54: 1333-1346.

Ghrefat, H. & Yusuf, N. (2006). Assessing Mn, Fe, Cu, Zn,and Cd pollution in bottom sediments of Wadi Al-Arab Dam, Jordan. Chemosphere.

Harikumar, P. S., & Jisha, T. S. (2010). Distribution patternof trace metal pollutants in the sediments of an urbanwetland in the southwest coast of India. InternationalJournal of Engineering Science and Technology. Vol.2(5). 840-850.

Jordao, C. P., Pereira, J. L., & Jham, G. N. (1997). Chromiumcontamination in sediment, vegetation and fish causedby tanneries in the State of Minas Gerais, Brazil. TheScience of the Total Environment. 207. 1-11.

Klavins, M., Briede, A., Rodinov, V., Kokorite, I., Parele, E.,& Klavina, I. (2000). Heavy metals in the rivers of Latvia.The Science of the Total Environment. 262: 175-183.

Landajo, A., Arana, G., de Diego, A., Etxebarria, N., Zuloaga,O., & Amouroux, D. (2004). Analysis of heavy metaldistribution in superficial estuarine sediments (estuaryof Bilbao, Basque County) by open-focusedmicrowave-assisted extraction and ICP-OES.Chemosphere. 56: 1033-1041.

Malherbe, C. W. (2006). The current ecological state ofthe lower Mvoti River, KwaZulu Natal. UnpublishedMSc thesis. University of Johannesburg, South Africa.

Malherbe, W., Wepener, V., & van Vuren, J. H. J. (2010).Anthropogenic spatial and temporal changes in theaquatic macroinvertebrate assemblages of the lowerMvoti River, KwaZulu Natal, South Africa. AfricanJournal of Aquatic Science. 35(1). 13-20.

Martin, J. M., & Whitfield, M. (1983). The significance ofriver input of chemical elements to the ocean. InBinning, K. & Baird, D. (2001). Survey of heavy metalsin the sediments of the Swartkops River Estuary, PortElizabeth, South Africa. Water SA. Vol 27. No 4. 461-466 pp.

Morse, J. W., Presley, B. J., Taylor, R. J., Benoit, G. &Santschi, P. (1993). Trace metal chemistry of GalvestonBay: Water, sediments and biota. In Binning, K. &Baird, D. (2001). Survey of heavy metals in thesediments of the Swartkops River Estuary, PortElizabeth, South Africa. Water SA. Vol 27. No 4. 461-466 pp.

Nriagu, J. O., & Pacyna, J. M. (1988). Quantitativeassessment of worldwide contamination of air, waterand soils by trace metals. In Chen, T. C & KandasamyS. (2008). Evaluation of elemental enrichments insurface sediments off southwestern Taiwan.Environmental Geology.54: 1333-1346.

Palanques, A., Masque, P., Puig, P., Sanchez-Cabeza, J. A.,Frignani, M., & Alvisi, F. (2008). Anthropogenic tracemetals in the sedimentary record of the Llobregatcontinental shelf and adjacent Foix Submarine Canyon(northwest Mediterranean).Marine Geology. 248. 213-227.

Ram, A., Rokade, M. A., & Zingde, M. D. (2009). Mercuryenrichments in the sediments of the Amba Estuary.Indian Journal of Marine Sciences. Vol. 38(1). 89-96.

Sukdeo, P., Pillay, S., & Bissessur, A. (2010). A proposal forthe restoration and management of the lower MvotiRiver and Estuary, KwaZulu Natal, South Africa.Manuscript submitted for publication.


S. PILLAY

Sundarajen, M., & Natesan, U. (2010). Geochemistry ofelements in core sediments near Point Claimere, thesoutheast coast of India.

Stoeppler, M. (1992). Analytical methods andinstrumentation – a summarizing overview. In Abed,R. (2006). A study of the heavy metal content ofsediments in selected KwaZulu Natal estuaries.Unpublished Honours thesis. University of KwaZuluNatal, Durban, South Africa.

Topalian, M. L., Castane, P. M., Rovedatti, M. G., & Salibian,A. (1999). Principal component analysis of dissolvedheavy metals in water of the Reconquista River(Buenos Aires, Argentina). Bulletin of EnvironmentalContamination and Toxicology. 63: 484-490.

Walsh, J. N. (1997). Inductively coupled plasma-atomicemission spectrometry (ICP-AES). In Abed, R. (2006).A study of the heavy metal content of sediments inselected KwaZulu Natal estuaries. UnpublishedHonours thesis. University of KwaZulu Natal, Durban,South Africa.

Wepener, V. (2007). Carbon, nitrogen and phosphorousfluxes in four sub-tropical estuaries of northernKwaZulu Natal: Case studies in the application of amass balance approach. Water SA. 3(2). 203-214.

Wolfe, D. A., Long, E. R., & Thursby, G. B. (1996). Sedimenttoxicity in the Hudson-Raritan Estuary: Distributionsand correlations with chemical contamination. InBinning, K. & Baird, D. (2001). Survey of heavy metalsin the sediments of the Swartkops River Estuary, PortElizabeth, South Africa. Water SA. Vol 27. No 4. 461-466 pp.



ENGINEERING GEOLOGICAL ASSESSMENT OF SOME LATERITICSOILS IN IBADAN, SOUTH-WESTERN NIGERIA USING

BIVARIATE AND REGRESSION ANALYSES

1Owoseni, J. O., 2Adeyemi, G. O., 1Asiwaju-Bello, Y. A., and 1Anifowose, A.Y.B.

Email: [email protected]; [email protected]

2Department of Geology, University of Ibadan, Ibadan.

E-mail: [email protected]

ABSTRACT: Bivariate correlation and regression techniques were employed to evaluate therelationship between pairs of geotechnical variables for residual lateritic soils derived from threegenetic crystalline rocks in Ibadan metropolis, south-western Nigeria. The significance of meangroup differences (parent-rock and level of compactive effort) at 5% level of significance wasdetermined using paired t-test analysis. This is with a view to ascertaining the influence of thepedogenic factor of parent rock, percentage fines, and energy of compaction on engineering indexproperties of the lateritic soils. The clay-size contents had positive correlations with both OptimumMoisture Content (OMC) and plasticity index, and a negative correlation with the Maximum DryDensity (MDD). The MDD and OMC had significant negative and positive correlations respectivelywith the amount of fines. The amount of fines and Unconfined Compressive Strength (UCS) hadsignificant negative and positive correlations respectively with the California Bearing Ratio (CBR).The study shows significant parent-rock group differences in most engineering properties. Thebandedgneiss-derived soils were found to be better engineering soils than the migmatite-gneiss- andquartzite/quartz- schist-derived soils. The modified AASHTO level of compactive effort which producedbetter compacted soils than the West African level is recommended for the soils.

Key words: Regression analysis, engineering properties, lateritic soils, pedogenic factors.

INTRODUCTION

The suitability of some lateritic soils for engineeringpurposes had been tied to naturally stable gradingscoupled with a suitable proportion of clayey materialsacting as binder (Ackroyd, 1960). The percentage of silt-and clay-size contents of a soil is a strong determinant ofits sensitivity to moisture. Careful soil investigation istherefore indispensable prior to utilization of soils for anyengineering purpose. Where necessary, appropriatemodification of the properties of the soil is made so thatits engineering performance is improved. The influence ofthe pedogenic factor of parent rock, percentage of silt-and clay-size contents, and energy of compaction onengineering index properties of lateritic soils can not beoveremphasized in lateritic soil engineering.

This paper reports on the studies carried out on lateriticsoils derived from quartzite/quartz-schist, banded gneiss,

and migmatite gneiss in parts of Ibadan, south-westernNigeria (Fig. 1). The quantitative influence of amount offines (%), clay-size contents (%), level of compactive effortand pedogenic factor of parent rock were investigated.Engineering geological variables determined weresubjected to statistical treatments such as paired(correlated) t-test- a parametric test used to compare themeans of two sets of observations from pairs of lateriticsoils derived from different genetic rock typeswhich aresignificantly different from one another.

THE STUDY AREA

The study area lies within latitudes 7o 25l N and 7o 27l N andlongitudes 3o 53l E and 3o 56l E (Fig. 1). The climate is of theWest African monsoonal type, characterized by distinctwet and dry seasons typical of West African regions.Average temperatures reach a peak of about 32o C aroundFebruary and a threshold of about 21o C around August.


J. O. OWOSENI

Figure 1: Location Map of the Study Area


Engineering Geological Assessment of some Lateritic Soils in Ibadan,South-Western Nigeria using Bivariate and Regression Analyses

61

Relative humidity ranges from about 70% around Januaryto about 90% in July. The climate is considered the mostimportant primary factor of lateritization. Maignien (1966)suggested that contemporary lateritic soils have developedat mean annual temperatures of around 25o C and lateriticsoils are believed to always correspond to climates in whichthe wet period is warm. GEOLOGICAL SETTING OF THE STUDY AREA

The area under investigation lies within the PrecambrianBasement Complex of Nigeria (Fig. 2a). Researchers havevariedly classified the assemblages into four lithologicalgroups, viz; minor intrusives, Older granites, schist beltsand migmatite-gneiss-quartzite complex. The study area is

underlain by the migmatite-gneiss-quartzite complex(Odeyemi, 1981; Rahaman, 1988; Elueze, 2002; Adekoya etal., 2003).

Three rock types within the migmatite-gneiss-quartzitecomplex were encountered in the study area (Fig. 2b). Themassive Quartzite component in association with quartz-schist) have a pseudo-conglomeratic and schistose nature.The quartz-schist outcrops strike between 160o and 170o

with dips ranging between 46oE and 58o E. They stand astopographic highs above the surrounding terrain. Thebanded gneisses exhibit some quartzo-feldspathic veins.There is a generally north-south strike direction andaverage dip angles of 36o – 47o E.

Figure 2a: Geological map of Nigeria showing the Basement Complex (Rahaman, 1988)


J. O. OWOSENI

METHODOLOGY

Nine bulk soil samples were collected, three samples fromeach parent rock type. Various laboratory techniques wereapplied to test the soil samples for engineering indexproperties such as grain size distribution, liquid limit, plasticlimit, linear shrinkage, Maximum Dry Density (MDD),Optimum Moisture Content (OMC), California BearingRatio (CBR), and Unconfined Compressive Strength (UCS).The tests were carried out in accordance with theprocedures outlined in the British Standards (BS) 1377(1990), with necessary modifications. The results weresubjected to statistical analysis at 5% level of significance(i.e. a=0.05) with a view to assessing the influence of thepedogenic factor of parent rock, percentage of silt- andclay-size contents, and compactive efforts on engineeringproperties such as amount of fines, OMC, MDD, UCS andunsoaked CBR. Again, to deduce the degree of correlationbetween the amount of fines and soaked CBR, and betweenthe UCS and unsoaked CBR. Grapher®, Microsoft Excel®

and SPSS® software packages were employed in dataanalyses and presentation.

RESULTS AND DISCUSSION The grain-size distribution characteristics (Fig. 3) showthat the soil samples are generally well-graded. Table 1shows the results of the engineering tests carried out on

Figure 2b: Geological map of the study area (Modified after Jones and Hockey, 1964)

the soil samples. From the particle-size distributioncharacteristics, soils derived from banded gneiss (with anaverage value of 67% gravel and sand-size fractions) arethe most sandy, followed by soils from quartzite/quartz-schist (with an average value of 65% gravel and sand-sizefractions) while soils from migmatite-gneiss are the leastsandy (with an average value of 41% gravel and sand-sizefractions). They are generally well-graded, revealing adecreasing degree of both leaching and weathering. Thiscan also be related to the textural characteristics of theparent-rock types. The specific gravity values show thatthe banded gneiss-derived soils (with an average value of2.75) has the highest degree of laterization, followed bythe quartzite/quartz-schist (with an average value of 2.72),while the soils derived from migmatite-gneiss-derived soils(with an average value of 2.71) is the least lateritized. Thebanded gneiss-derived soils which have the least linearshrinkage (with an average value of 12.8%) values will bethe best for highway construction, followed by themigmatite-gneiss (with an average value of 13%). Thequartzite/quartz-schist-derived soils which have thehighest linear shrinkage values (with an average value of14%) would be the worst for highway construction.

From the Unconfined Compressive Strength and CaliforniaBearing Ratio (soaked and unsoaked) values at bothmodified AASHTO and West African levels of compactiveefforts, the banded gneiss-derived soils (with the highest



63

0 . 0 0 1 0 . 0 1 0 .1 1 1 0

P a rt i cl e s iz e (mm )

0

2 0

4 0

6 0

8 0

1 0 0

Perc

enta

ge p

assin

g (%

)

Quart zi te/Quart z- schist (A-001)Quart zi te/Quart z- schist (A-002)Quart zi te/Quart z- schist (003)

0 .0 0 1 0 . 0 1 0 . 1 1 1 0 1 00P a rt i cl e si z e (m m )

0

20

40

60

80

1 00

Per

cent

age

pass

ing

(%)

Banded Gneiss (B-001)Banded Gneiss (B-002)Banded Gneiss (B-003)

0 . 00 1 0. 0 1 0 . 1 1 1 0P a r t ic l e s i ze (m m )

20

40

60

80

1 00

Perc

enta

ge p

ass

ing (

%)

Mi gmati te Gnei ss (C -001)Mi gmati te Gnei ss (C -002)Mi gmati te Gnei ss (C -003)

(a) Uncompacted natural soi ls

0. 0 0 1 0 .0 1 0 . 1 1 1 0 1 00

P a r t i cl e si z e (m m )

2 0

4 0

6 0

8 0

10 0

Perc

enta

ge

pass

ing

(%)

Quart zit e/Quartz-Schist ( A-001)

Quart zit e/Quartz-schist (A-002)

Quart zit e/Quartz-schist (A-003)

0 . 00 1 0 . 0 1 0 . 1 1 1 0 1 0 0P a rt i cl e s iz e (mm )

20

40

60

80

1 00

Perc

ent

age

pass

ing

(%)

Banded Gneiss (B-001)



0 .0 0 1 0 . 0 1 0. 1 1 10P art i cl e S i ze (mm )

4 0

6 0

8 0

1 0 0

Perce

ntag

e p

assin

g (%

)

Migmatit e Gnei ss (C-001)



(b) Compacted soil s at West African level

0 .0 0 1 0 .0 1 0 .1 1 10 1 00

P a rt ic l e s iz e (mm )

20

40

60

80

1 00

Perc

enta

ge p

assin

g (%

)

Quartzit e/ Quartz- schist (A-001)Quartzit e/ Quartz- schist (A-002)Quartzit e/ Quartz- schist (A-003)

0 . 0 0 1 0 . 0 1 0 . 1 1 10P a rt ic l e s i ze (mm )

2 0

4 0

6 0

8 0

10 0

Perc

enta

ge p

ass

ing

(%)

Banded Gneiss (B-001)Banded Gneiss (B-002)Banded Gneiss (B-003)

0 . 0 0 1 0 . 0 1 0. 1 1 10P a rt i cl e si z e ( mm )

2 0

4 0

6 0

8 0

10 0

Per

cent

age p

ass

ing (

%)

Migmati te Gneiss (C-001)

Migmati te Gneiss (C-002)

Migmati te Gneiss (C-003) (c) Compacted soils at Modified AASHTO level

pe

rcenta

ge pa

ssing

(%)

pe

rcenta

ge pa

ssing

(%)

pe

rcenta

ge pa

ssing

(%)

Figure 3: Grain-size distribution curves


J. O. OWOSENI

Unsoaked Soaked Unsoaked Soaked

Quartzite/ A-001 A-6 60 40 2.73 25 11 14 15.2 1770 1940 21 16 279 388 24 15 33 22

Quartz Schist A-002 A-2-7 70 30 2.73 60 34 26 14.1 1780 1`970 17 14 301 420 30 20 40 29

A-003 A-7-6 64 36 2.7 46 22 24 12.6 1760 1910 21 17 270 354 28 17 36 24

B-001 A-7-6 64 36 2.73 43 25 18 11.2 1780 1950 19 15 296 345 33 22 43 30

B-002 A-2-7 67 33 2.75 52 23 29 15.6 1800 1980 17 14 341 573 27 18 38 27

B-003 A-2-7 69 31 2.77 41 24 17 11.7 1780 1960 20 16 225 456 25 14 34 26

C-001 A-6 46 54 2.73 37 16 21 13.7 1680 1860 23 18 250 318 22 11 27 18

C-002 A-7-6 44 56 2.7 55 23 32 11.1 1660 1820 24 19 287 400 20 10 26 16

C-003 A-7-6 33 67 2.7 51 26 25 14.1 1690 1880 22 17 246 315 18 10 23 14

Banded Gneiss

Migmatite Gneiss

UCS (KN/m3)

CBR (%)

WA MA WA MA WA MA WA MA

LL PL PI SL MDD (Kg/m3) OMC (%)Parent Rock Sample No.

AASHTO Classification

Gravel + Sand (%)

Silt + Clay (%)

Gs

Table 1. Engineering Properties of the Soil Samples

average values) have the highest strength and stability,followed by the quartzite/quartz-schist derived soils whilethe migmatite-gneiss-derived soils (with the lowest averagevalues) have the least strength and stability. Thecompaction characteristics indicate that the banded gneiss-derived soils (with the highest average MDD and lowestOMC values) exhibit the highest strength followed byquartzite/quartz-schist-derived soils while the soils derivedfrom migmatite-gneiss (with the least average MDD andhighest OMC values) exhibit the least strength.

Paired t-test

Significance of parent-rock group difference Paired t-test analysis of results of engineering geologicalvariables (such as MDD, OMC, CBR and amount of fines)reveals a strong influence of parent-rock factor. Geneticallydifferent soils exhibited varied compaction characteristicsand degrees of degradation under compaction at bothlevels of compactive efforts used (Tables 2 - 4). The studyalso revealed some significant influence of parent-rockgroup difference on soaked CBR at OMC of samplescompacted at both West African and Modified AASHTOlevels(Tables 5).

Significance of energy of compaction group difference Paired t-test analysis of values of Unconfined CompressiveStrength (UCS) of the soils reveals a significant difference

LL=Liquid Limit, PL=Plastic Limit, PI=Plasticity Index, Gs=Specific Gravity, SL=Shrinkage Limit,MDD=Maximum Dry Density, CBR=California Bearing Ratio OMC=Optimum Moisture Content, UCS=Unconfined CompressiveStrength, WA=West African level, MA=Modified AASHTO level

between UCS values for West African level and modifiedAASHTO level of compaction for each of the three geneticsoils (Table 6). An analysis of grain-size distributioncharacteristics also shows some significant difference inpercent increase in amount of fines for West African leveland Modified level of compaction (Table 7). Upon compaction, there was a particularly excessivebreakdown of soil grains in samples of quartzite/quartz-schist-derived soils (Figures 3b and 3c). The migmatitegneiss-derived soils had proved the most mechanicallystable among the studied soils, being the least susceptibleto degradation under dynamic load. The observed increasein the amount of fines upon compaction, as the case iswith soils developed on Quartzite/quart schist and bandedgneiss, can be attributed to the binding effect ofsesquioxide on smaller particles - a factor of high degree ofleaching and lateritization.

Correlation and regression analysis

Regression analysis of determined parameters gave someuseful regression models (Equations 1-8). Pearsoncorrelation coefficient (r) revealed significant correlationsbetween some pairs of engineering geological variables at5% level of significance (i.e. a = 0.05).



65

Table 2: Significance of parent-rock group difference in the values of MDD

Table 3: Significance of parent-rock group difference in the values of OMC

Parent rock

(Group X)

Parent rock

(Group Y)t–in table

West African level Modified AASHTO level

Computed tSignificance of

parent rock group difference



Quartzite/Banded Gneiss 2.132 0.41 Not Significant 0.64 Not Significant

Quartz-SchistBanded Gneiss

Migmatite-Gneiss

2.1324.16

Strongly Significant 3.7

Strongly Significant

Migmatite-Gneiss

Quartzite/ 2.132

2.31Fairly Significant 2.24 Fairly Significant

Quartz-schist

Parent rock

(Group X)

Parent rock

(Group Y)t–in table






Quartzite/ Quartz-Schist

Banded Gneiss

2.132 1.65 Not Significant 1.21 Not Significant

Banded Gneiss

Migmatite-Gneiss

2.132 2.62 Fairly Significant

5.62 Strongly Significant

Migmatite-Gneiss

Quartzite/ Quartz-schist

2.132 4.13 Strongly Significant

3.52 Strongly Significant

Table 4: Significance of parent-rock group difference in the values of amount of fines

Parent rock (Group X)

Parent rock (Group Y) t–in table Computed t

Significance of parent rock group

differenceQuartzite/

Banded Gneiss 2.132 0.62 Not SignificantQuartz-Schist

Banded GneissMigmatite-Gneiss

2.1326.01 Strongly Significant

Migmatite-Gneiss

Quartzite/ 2.132

4.79 Strongly SignificantQuartz-schist


J. O. OWOSENI

Parent rock (Group X)

Parent rock (Group Y) t–in table




Computed t

Significance of parent

rock group difference

Quartzite/ Banded Gneiss 2.132 0.22 Not Significant 0.61

Not SignificantQuartz-Schist

Banded GneissMigmatite-Gneiss

2.1323.142

Strongly Significant 4.58


Migmatite-Gneiss

Quartzite/ 2.132

3.51Strongly Significant 4.7

Strongly SignificantQuartz-schist

Table 5: Significance of parent-rock group difference in the values of CBR

Parent Rock

Unconfined compressive

strength (Group X)Unconfined compressive

strength (Group Y)

t – in table Computed-t

Significance of energy of

compaction group difference

Quartzite/ Quartz schist

West African levelModified AASHTO

2.1324.95


Banded GneissWest African level Modified

AASHTO2.132

2.33Fairly Significant

Migmatite Gneiss

West African level Modified AASHTO

2.1322.73

Fairly Significant

Table 6: Significance of level of compactive effort in values of UCS

Parent Rock

Natural soil West African level of compaction

Modified AASHTO level of compaction

% fines% fines

% Increase in fines

% fines % Increase in fines

Quartzite/ Quartz schist

40 50 25 60 5030 47 56.67 50 83.3336 50 38.89 60 66.67

Banded Gneiss

36 43 19.44 46 27.7833 36 9.09 42 27.2731 36 16.13 38 22.58

Migmatite-Gneiss

54 63 16.67 68 25.9356 59 5.36 49 -67 79 17.91 83 23.88

Table 7: Influence of energy of compaction on amount of fines



67

Relationship between amount of fines (%) and compactioncharacteristics

The correlations established between the amount of finesand compaction characteristics are in agreement withGidigasu (1972) on some Ghanaian soils. The relationshipbetween the amount of fines (%) and Maximum DryDensity (MDD) at West African and modified AASHTOlevels of compaction gave regression equations (1a) and(1b) with significant negative correlations (R2 being 0.689and 0.231 ) respectively.

yMDD = -3.158xfines + 1906 ……...(1a)

yMDD = -1.889xfines + 2024 ……...(1b)

On the relationship between the amount of fines andOptimum Moisture Content (OMC), regression equations(2a) and (2b) with significant positive correlations (R2 being0.467 and 0.150) were obtained respectively.

yOMC = 0.121xfines + 14.18 ………(2a)

yOMC = 0.047xfines + 13.58 ………(2b)

Association between amount of fines and soaked CBRgave regression models (3a) and (3b) with significantnegative correlations R2 being 0.485 and 0.461)respectively.

yCBR = -0.221xfines+ 26.6 ……….(3a)

yCBR = -0.279xfines+ 38.46 ……….(3b)

Relationship between amount of fines and unsoaked CBRresulted in regression models (4a) and (4b) with significantnegative correlations (R2 being 0.2966 and 0.3895)respectively.

yCBR = -0.187xfines+ 35.632 ……….(4a)

yCBR = -0.3018xfines+ 50.132 ……….(4b)

Relationship between amount of clay-size content in thefines and compaction characteristics

This study revealed some correlations between the clay-size content and compaction characteristics. Theassociation between clay-size content and Maximum DryDensity (MDD) at West African and modified AASHTO

levels of compaction gave regression equations (5a) and(5b) with significant negative correlations (R2 being 0.6841and 0.1295) respectively.

yMDD = -4.4422xclay-size + 1926.1 ….…….(5a)

yMDD = -1.3432xclay-size + 1975.2 ……….. (5b)

Relationship between the clay-size content and OptimumMoisture Content (OMC) gave regression models (6a) and(6b) with significant positive correlations (R2 being 0.5146and 0.077) respectively.

yOMC = 0.1398xclay-size + 15.006 ………. (6a)

yOMC = 0.0323xclay-size + 14.871 ……….. (6b) Relationship between clay-size content and consistencylimits

This study revealed some correlations between the clay-size content and consistency limits. The associationbetween clay-size content and liquid limit and plastic limitgave regression equations (7a) and (7b) with negativecorrelations (R2 being 0.0024 and 0.0916 respectively) whilePlasticity index gave regression equation (7c) withsignificant positive correlation (R2 being 0.0581).

yliquid-limit = -0.0492xclay-size + 46.972 ………(7a)

yplastic-limit = 0.1836xclay-size + 27.95 ………(7b)

yplasticity index = 0.1343xclay-size + 19.023 ……….(7c) Relationship between Unconfined Compressive Strength(UCS) and California Bearing Ratio (CBR)

Association between Unconfined Compressive Strength(UCS) and un-soaked CBR at West African and modifiedAASHTO levels of compaction gave regression equations(8a) and (8b) with significant positive correlations (R2 being0.219 and 0.213) respectively. This agrees with the findingsof Adeyemi (1992).

yUCS = 3.365xCBR + 192.3 ………(8a)

yUCS = 6.491xCBR+ 247.9 ………(8b)


J. O. OWOSENI

Figure 7: Correlation between amount of fines and Compaction characteristics

West Africa n leve l

y = -0.187x + 35.632R2 = 0.2966

0

5

10

15

20

25

30

35

0 20 40 60 80 100 120 140Percentage fi nes (%)

Unso

aked

CB

R (%

)

Modified AASHTO level

y = -0.3018x + 50.132R2 = 0.3895

0

10

20

30

40

50

0 20 40 60 80 100 120 140

Percentage fines (%)

Unso

aked

CBR

(%)

West African level

y = -0.2213x + 26.608R2 = 0.4855

0

5

10

152025

30

0 20 40 60 80 100


Soak

ed C

BR (%

)

Modified AA SHTO leve l

y = -0.2798x + 38.462R2 = 0.4616

0

5

10

15

20

25

30

35

0 50 100 150

Percentage fi nes (%)

Soak

ed C

BR (%

)

Figure 8: Correlation between amount of fines and bearing capacity

West African level

y = -3.9701x + 1957.6R2 = 0.6555

0

500

1000

1500

2000

2500

0 20 40 60 80 100


Max

imum

Dry

Den

sity

(K

g/m

)


y = -1.889x + 2024R2 = 0.2312

1800

1850

1900

1950

2000

2050

0 20 40 60 80 100


Max

imum

Dry

Den

sity

(Kg/

m

)

West African level

y = 0.1217x + 14.185R2 = 0.4674

0

5

10

15

20

25

30

0 20 40 60 80 100


Opt

imum

Moi

stur

e C

onte

nt (%

)


y = 0.0474x + 13.582R2 = 0.1504

0

5

10

15

20

0 20 40 60 80 100


Opt

imum

Moi

stur

e C

onte

nt

(%)



69

We st African Level

y = 0.1398x + 15.006R2 = 0.5146

05

1015

202530

0 20 40 60 80 100Clay-size content (%)

Opt

imum

Moi

stur

e Co

nte

nt

(%)

M odi fied AASHTO Le ve l

y = 0.0323x + 14.871R2 = 0.077

0

5

10

15

20

0 20 40 60 80 100Clay-s ize co ntent (%)

Op

timum

Moi

stu

re C

ont

ent

(%)

West African leve l

y = -4.4422x + 1926.1R2 = 0.6841

0

500

1000

1500

2000

2500

0 20 40 60 80 100Clay-size content (%)

Maxi

mum

Dry

Den

sity

Kg

m

M od ifie d AASHTO level

y = -1.3432x + 1975.2R2 = 0.1295

1 800

1 850

1 900

1 950

2 000

0 20 40 60 80 100

Clay s ize co nten t ( )

Max

imum

Dry

Den

sity

Kg

m

Figure 9: Correlation between Clay-size content in fines and compaction characteristics

y = -0.049 2x + 46.972

R2 = 0.0024

0

10

20

30

40

50

60

70

0 10 20 30 40 50 6 0 70

Clay-size conte nt in fine s(%)

Liqu

id lim

it (%

)

y = -0.1836x + 27.95R2 = 0.0916

0

5

10

15

20

25

30

35

40

0 10 20 30 40 50 60 70

Clay-size cont ent in f ines (%)

Plast

ic lim

it (%

)

y = 0.1343x + 19.023

R2 = 0.0581

0

5

10

15

20

25

30

35

0 10 20 30 40 50 60 70

Clay-size content (%)

Plas

ticity

inde

x (%)

Figure 10: Correlation between Clay-size content in fines and Consistency limits


J. O. OWOSENI

West African level

y = 3.3659x + 192.33R2 = 0.2194

0

50

100150

200

250

300

350400

0 10 20 30 40 50 60

Unsoaked CBR (%)

Unco

nfine

d Com

pres

sive

Stre

ngth

(KN/

m )


y = 4.6884x + 240.28R2 = 0.1538

0

100

200

300

400

500

600

700

0 10 20 30 40 50 60 70 80

Unsoaked CBR (%)

Unco

nfin

ed C

ompr

essiv

e St

reng

th (K

N/m

)

Figure 11: Correlation between UCS and unsoaked CBR

CONCLUSIONS

The grading characteristics suggest that the banded-gneiss- and quartzite/quartz-schist-derived soils may beadjudged poor to fair sub-base and subgrade materials.The banded-gneiss-derived soils which exhibited the leastclay-size content, least linear shrinkage and highestMaximum Dry Density can be adjudged the bestengineering soil group among the three groups. Themodified AASHTO level of compactive effort producedbetter compacted soils (as evident in the lower values ofOMC, and higher values of MDD, CBR and UCS) than theWest African level. Based on Madedor’s (1983) maximumvalue of 8% for linear shrinkage, the soils could pose onlylittle compaction problem.

The study revealed that the pedogenic factor of parentrock significantly influenced the engineering indexproperties of the lateritic soils under investigation. Apartfrom the parent rock factor, both the level of compactiveeffort and the amount of fines (especially the clay-sizefractions) have been found to play significant roles in thestrength characteristics of the soils. The particle-sizedistribution characteristics of the soils shows that the soilsderived from banded gneiss are the most sandy, followedby soils from quartzite/quartz-schist while soils frommigmatitie-gneiss are the least sandy, indicating adecreasing degree of leaching and weathering on the onehand and a reflection of the textural characteristics of theparent-rock types on the other hand. The specific gravityvalues indicate that the banded gneiss-derived soils hadthe highest degree of lateritization, followed by thequartzite/quartz-schist, while the soils derived frommigmatite-gneiss were the least lateritized.

From the Unconfined Compressive Strength and CaliforniaBearing Ratio (soaked and unsoaked) values at bothmodified AASHTO and West African levels of compactivewefforts used, the banded gneiss-derived soils would exhibitthe highest strength and stability, followed by the quartzite/quartz-schist-derived soils while the migmatite-gneiss-derived soils would exhibit the least strength and stability.From the compaction characteristics, the banded gneiss-derived soils (with the highest average MDD and lowestOMC values) exhibit the highest strength, followed byquartzite/quartz-schist-derived soils while the migmatite-gneiss-derived soils exhibit the least strength. The bandedgneiss-derived soils with the least linear shrinkage valuescould be adjudged the best highway construction material,followed by the migmatite-gneiss-derived soils. Thequartzite/quartz-schist-derived soils with the highest linearshrinkage values could be the worst.

REFERENCES

Ackroyd, I. W., 1960. Notes on crushing strength of someWestern Nigerian concretionary gravels and theirselection for use as building material. Min. Transp.,Ibadan. W. Nigeria, Tech. Paper., 6.

Adekoya, J. A., Kehinde-Philips, O. O. and Odukoya, A.M., 2003. Geological distribution of mineral resourcesin southwestern Nigeria. In: A. A. Elueze (ed).Prospects for investment in mineral resources ofsouthwestern Nigeria, pp. 1 – 13.

Adeyemi, G.O., 1992. Highway geotechnical properties oflaterised residual soils in the Ajebo-Ishara geologicaltransition zone of South-western Nigeria. UnpublishedPh.D. (Geology) thesis, Department of Geology,Obafemi Awolowo University, Ile-Ife, Nigeria, 86pp.



71

British Standards.1377., 1990. Methods of test for soils forcivil engineering purposes. General requirements andsample preparation. BS 1377. British StandardsInstitution, 38pp.

Elueze, A. A. 2002. Compositional character: Veritable toolin the appraisal of Geomaterials. An inaugural lecture,University of Ibadan. 43pp.

Gidigasu, M. D. 1972. Mode of formation and geotechnicalcharacteristics of laterite materials of Ghana in relationto soil-forming factors. Engineering Geology 6, pp.79-150.

Jones, R. A. and Hockey, R. D. 1964. The Geology of partsof South-western Nigeria, Explanation of 1:250,000sheets Nos. 59 and 68. Geological survey of NigeriaBulletin 11, 101pp.

Maignien, R. 1966. Review of Research on Laterites. Nat.Resours. Res. IV. UNESCO, Paris, 148 pp.

Madedor, A. O. 1983. Pavement design guidelines andpractice for different geological areas in Nigeria. In: S.A. Ola (ed). Tropical soils of Nigeria in Engineeringpractice, A. A. Balkema, Rotterdam, The Netherlands,pp. 291-297.

Odeyemi, I. B. 1981. A review of the Orogenic events in thePrecambrian Basement in Nigeria, West Africa. Geol.,und.,Vol. 70, pp. 897-909.Rahaman, M. A. 1988.Different Advances in the Basement Complex ofNigeria. In: Precambrian Geology of Nigeria, apublication of the Geological Survey of Nigeria, pp.11-14.



RELATIVE EFFICIENCY OF NON-PARAMETRIC ERRORRATE ESTIMATORS IN MULTI-GROUP LINEAR

DISCRIMINANT ANALYSIS

1R. Glele KaKai, D. Pelz2 and R. Palm3

1Faculty of Agronomic Sciences, University of Abomey-Calavi, 04 BP 1525, Cotonou, Benin.2Department of Forest Biometry, University of Freiburg,

Tennenbacherstr. 4, D-79085, Freiburg (Germany)3Gembloux Agricultural University, Avenue de la Faculté d’Agronomie 8, B-5030,


ABSTRACT: A Monte Carlo study was achieved to assess the relative efficiency of ten non-parametricerror rate estimators in 2-, 3- and 5-group linear discriminant analysis. The simulation design tookinto account the number of variables (4, 6, 10, 18) together with the size sample n so that: n/p = 1.5,2.5 and 5. Three values of the overlap, e of the populations were considered (e = 0.05, e = 0.1, e =0.15) and their common distribution was Normal, Chi-square with 12, 8, and 4 df; theheteroscedasticity degree, was measured by the value of the power function, 1-β of the

homoscedasticity test related to (1-β = 0.05, 1-β = 0.4, 1-β = 0.6, 1-β = 0.8). For each combinationof these factors, the actual error rate was empirically computed as well as the ten estimators. Theefficiency parameter of the estimators was their relative error, bias and efficiency with regard to theactual error rate, empirically computed. The results showed the overall best performance e632estimator. On the contrary, e0, epp, eppCV and eA recorded the lowest performance in terms of meanrelative error and mean relative bias. The ranks of the estimators were not influenced by the numberof groups but for high values of the later, the mean relative bias of the estimators tend to zero.

Keywords: Error rate; Estimation; Efficiency; Multi-group; Linear rule; Simulation.

INTRODUCTION

Discriminant analysis is a statistical method of allocationof an unknown individual to one group, from at least twoforeknown groups, by using a classification rulepreviously established on well-known individuals. Anumber of classification rules are available and the mostused are linear, quadratic and logistic methods. Many classification rules have been proposed in literatureand the most common is the linear classification rule(Fisher, 1936). Let us suppose g p-variate populations,

),...,1( gkGk , with mean vectors, ),...,1( gkk μand common covariance matrices, Σ . The linear rule ( LR)

is a Normal-based classification rule for which

),N( Σμ kF (McLachlan, 1992):

);,...1,(

);()(5.0)/ln(

),N(,LR1'

lkglk

pp lklkilk

ki

μμΣμμ

Σμ

x

x

(1.1)

The unknown observation vector ix is assigned to kG if:

klglki ;,...,1 0),N(,LR Σμx

In the case of data samples, LR can be established by

replacing in (1.1) the parameters, ),...,1( gkk μ and


Relative Efficiency of Non-parametric Error Rate Estimators in Multi-group Linear Discriminant Analysis

73

Σ by their estimates, kμ̂ (k=1,...,g) and kΣ̂ ; Σ̂ is

considered in (1.1) as the estimated pooled covariancematrix of the k populations.

Whatever the rule established is, it is subject to aprobability of misclassifications. Then, an actual error rateis associated with any classification rule established ondata samples in order to evaluate its efficiency. In practice,it is impossible to precisely determine the actual error rate,because it is only computed on the actual parameters ofthe populations, which are usually unknown. To solve thisproblem, some parametric and non parametric estimatorsof the actual error rate were established (McLachlan, 1992).Parametric estimators were established for two normalhomoscedastic groups and the actual error rates estimated,using some parameters related to the considered samplessuch as the estimated Mahalanobis distance between thetwo groups. On the contrary, non-parametric error rateestimators do not depend on any hypothesis of use andare based on resampling methods. For two-groupdiscriminant analysis, many comparative studies of errorrate estimators have been done in linear discriminantanalysis, in order to deduce the ones that have the lowesterrors compared with the theoretical actual error rate. Athorough review of these studies is provided by Schiavoand Hand (2000). However, in real world problems, morethan two groups are often considered in discriminantanalysis. This paper evaluates and compares, by simulationtechnique, the efficiency of ten non parametric error rateestimators for 2-, 3- and 5-groups submitted to lineardiscriminant analysis.

ACTUAL ERROR RATE The actual error rate can be defined as the theoreticalproportion of misclassified observations, obtained byvalidating a classification rule established on data samplesto any other observation taken from the same populations.This error rate is useful in practice because it gives theexpected misclassification rate when a previouslyestablished rule is used. Let us assume two samples, E1 and E2 with p variables andcommon size n. The mean vectors and the pooled

covariance matrix are 21 , xx and S , respectively. Letus also suppose also that these samples are taken from

normal populations, P1 and P2, with mean vector ( kμ =1,2).

The actual error rate specific to the group keck , ( 2,1k )

and the overall actual error rate are given by McLachlan(1975):

2

1

211

21

211

1

and

)()'(

)()(21

)1(

kkk

kk

ecpec

ecxxSSxx

xxSxxk

1

2μ

(2.1)

where kp and are, respectively, the prior probabilityrelated to the group and the cumulative function of theNormal distribution.

The relations (2.1) can only be used in two-groupdiscriminant analysis when the linear rule is establishedon two normal homoscedastic populations. In the othercases, the actual error rate associated with a classificationrule can be empirically computed, for two groups, bydetermining the proportion of misclassified observationswhen the rule is established on the samples E1 and E2 andvalidated on a couple of large samples, of size 10,000 forexample.

ESTIMATION OF THE ACTUAL ERROR RATE For more than two groups submitted to discriminantanalysis, only non-parametric estimators can be used toassess the actual error rate associated with an establishedrule; parametric estimators were only conceived for two-group discriminant analysis. Ten non-parametric error rateestimators were considered in the study and presentedbelow. Resubstitution estimator, eA (Smith, 1947): i.e., proportionof misclassified observations when the rule wasestablished and validated on the same samples. Cross validation estimator, eCV (Lachenbruch, 1967): i.e.,proportion of misclassified observations when gndiscriminant analyses are done on gn -1 observations byremoving, at each step, one observation and by allocatingthe removed observation to one of the considered groupson the basis of the rule established on the gn-1observations.


R. GLELE KAKAÏ

1eS and 2eS Estimators (Hand, 1986):

eCVn

neSeCVn

neS32

2 and 12

221

(3.1)

epp Estimator (Fukunaga and Kessell, 1972):

gn

iigign

epp1

1 )(τ),...,(τ̂max11 xx (3.2)

The symbols gkik ,...,1)(τ̂ x represent the

posterior probability that an individual i of observations

vector ix belongs to population Gk and is defined as:

g

lilikik ff

1)(ˆ/)(ˆ)(τ̂ xxx

where )(ˆikf x is the value of the estimated density

function at ix for population kG .

eppcv Estimator (Fukunaga and Kessell, 1972): i.e.,computed by using the relation (3.2) in which the posterior

probabilities, gkik ,...,1 )(τ̂ x of the observations

vector ix was determined, using the classification rule

established on gn -1 observations, the vector ix , beingremoved.

Jackknife estimator, (Quenouille, 1949): i.e., computedby realising discriminant analyses on gn -1 observations.For each sample of gn -1 observations, the observation

being removed, the resubstitution estimator, )(ikeA ,

specific to gkGk ,...,1 , was computed. By assuming,

kAe

n

iikk eA

nAe

1)(

1 (3.3)

the Jackknife estimator is computed as:

g

kkkk AeeAneA

geJc

1))(1(1

(3.4)

where keA is the resubstitution estimator specific to andcomputed from the overall sample. Ordinary bootstrap estimator, (Efron, 1983): i.e., computedon 100 bootstrap samples, a sample of size n being takenwith replacement in each initial sample of size n. For eachbootstrap sample, the classification rule is established and

the resubstitution estimator, *kjeA 100,...,1;,...,1 jgk

specific to kG was computed. The same rule is also used

to compute the proportions, *kr of misclassified

observations, the rule being validated on the initial sample.

The bias, ),...,1( gkbk of *kjeA is computed as

follows:

100

1

** )(100

1j

kkjk reAb (3.5)

The overall bootstrap estimator is computed as:

g

kkk beA

geboot

1

1 (3.6)

keA being the resubstitution estimator specific to kG

when the rule is established on the gn initial observations.

0e Estimator (Chatterjee and Chatterjee, 1983): i.e.,

computed on 100 bootstrap samples )100,...,1(t * ii ,

taken from the initial sample . For each bootstrap sample, aclassification rule is established and the proportion ofmisclassified observations of t, which do not belong to

*t i , was computed. The 0e estimator is the mean of the100 proportions.

632e Estimator (Efron, 1983): i.e., computed as follows:

0632.0368.0632 eeAe (3.7)

SIMULATION DESIGN

Discriminant model We consider the case of 2-, 3- and 5-groups submitted tolinear discriminant analysis and characterized by theirmeans and covariances matrices. In the case of 2 groups,



75

the mean vector, 2 ,1 kkm is:

IRmm ;)'0,...,0,( 0; 21 mm .

The covariance matrix, 2 ,1kkΣ , is a diagonal matrix

with 2 ,1kkv , the vector of diagonal elements givenby

)1(1 vv ; λ2 vv where IRλ and )'1,...,1 ,( λ v .

In the case of 3- and 5-groups, the mean vectors, andcovariance matrices, are given below:

For 3-groups:

m1= 0; m2 = )'0,...,0,(m m3 = )'0...,0,,0( m ; 1v = )1(v ;

λ32 vvv .

For 5-groups

m1= 0; m2 = )'0,...,0,(m m3 = )'0...,0,,0( m ; m4 =

)'0,...0,( m ; m5 = )'0,...,0,,0( m

)1(1 vv ; 2v = 3v = 4v = 5v = λv .

It is known that the linear rule is invariant under a non-singular linear transformation (McLachlan, 1992). So,appropriate linear transformations applied to the simplemodels proposed above, will help to extend the results ofthe study to a large variety of real world problems.

To assess the heteroscedasticity degree of the populations,a heteroscedasticity parameter is defined for gpopulations submitted to discriminant analysis as:

= ΣΣk

g

k

1

ln , (4.1)

with kΣ and Σ , being the covariance matrix of kG andthe pooled covariance matrix of the g populationsrespectively. For data samples, an estimated ̂ can be

computed by replacing kΣ and Σ , respectively with

kΣ̂ and Σ̂ .

By considering the discriminant model proposed above, it

can analytically be shown that the parameters g (g = 2, 3

and 5) and λ (defined in Section 4) are linked by thefollowing relations:

45

5

5

23

3

32

2

2

λ5λ)41(

ln)λ(

;λ3λ)21(

ln)λ(;λ2λ)1(

ln)λ(

(4.2)

The inverse of these functions helped in choosing the

appropriate values of g according to λ .

Population features and comparison criteria The factors considered in the assessment of the efficiencyof the non-parametric error rate estimators were the numberg of groups (g = 2, 3 and 5), the common distribution of thevariables of the p-variate populations that is Normal(named N), Chi-square with 12, 8 and 4 degrees of freedom,named C(12), C(8) and C(4), respectively. The number p ofvariables was 4, 6, 10, 18; three values of the common sizesample, n were considered for each value of pnp /: =

1.5; pn / = 2.5 and pn / = 5. For each number g ofgroups, four values of the heteroscedasticity degree,

5 and 3 2, kk of the populations were chosen

from established empirical power function, 1-β of the

homoscedasticity test related to k under normality case

(1-β = 0.05: homoscedasticity; 1- β = 0.4: lowheteroscedasticity; 1- =0.6: average heteroscedasticity; 1-

β =0.8: high heteroscedasticity. Table 1 presents for eachnumber of groups, the mean values related to each of thefour values of 1-β . Three values of the overlap, e of thepopulations were considered: e = 0.05 (low overlap); e =0.1 (average overlap) and e = 0.15 (high overlap). Thegroup-prior probabilities were considered equal and theoverlap was thus equal to the optimal error rate. For eachof the combination of population features described above,the values of the parameter m (defined in section 4) wereiteratively computed to obtain each of the three values ofthe overlap (or optimal error rate) of the populations.However, the expression (2.3) for the computation of theoverlap e was difficult to manipulate for g > 2 so we usedan empirical approach to compute the overlap, e .


R. GLELE KAKAÏ

We present below (without loss of generality), thecomputational method of e for three p-variate populations,

1P , 2P and 3P , of theoretical density functions 1f , 2f

and 3f . In the discriminant model considered in section 4the differences between the means vectors were onlycarried by the first two variables of the populations. Insuch cases, the other variables did not influence theoverlap, of the populations. So, it can be deduced fromequation (2.3) that, for equal group-prior probabilities:

)(31

321 eeee with:

Table 1. Values of k according to the 4 values of 1-β

non-parametric error rate estimators were computed. Theactual error rate ec was also empirically computed foreach sample by validating the established linear rule on alarge sample of size 10,000g and used to calculate theRelative Error (RE), the Relative Bias (RB ) and the RelativeEfficiency (REff ) of each estimator:

min(RE)or)RE(estimatRE

;)(estimator100RB

;estimator

100RE

ff

ecec

ecec

(4.4)

In equation (4.4), the symbol min(RE) represents the relativeerror of the best estimator for the considered sample. TheMean Relative Error (MRE), the Mean Relative Bias (MRB)and the Mean Relative Efficiency (MREff) related to eachestimator were computed for each of the 1728 combinationsof the factors.

RESULTS The MRE of the non-parametric estimators for eachcombination of the factors were replaced by ranks. For agiven combination of the factors, the ranks of the errorrate estimators were computed, the estimator of the lowestrelative error having the rank 1. The median ranks of theestimators were calculated for each factor level as well astheir median rank for all the 1728 combinations of factorsand placed in Table 2. It can be noticed that 632e is theoverall best estimator; the other estimators of goodperformance were 2eS and 1eS . On the contrary, , 0e ,

epp and eArecorded the lowest relative efficiencies. Theranks of the ten estimators for each level of populationfeatures did not globally depend on the number g ofgroups, except 2eS estimator whose relative performanceslightly decreased with increased number of groups. Thepopulation features seemed not to have influenced theranks of the estimators. However, eboot and eppCVimproved their ranks for increased values of the ratio pn /whereas an opposite trend was observed, not only in thecase of eJc , but also 1eS and 2eS , especially for 5-

groups. Moreover, the relative efficiency of eppCV and632e became low with the increased overlap of the

populations. The median rank of the estimators for thelevels of population features did not help in analysing thequantitative difference between their performances.

3321233121

2231323121

1231313212

P)()(and)()(3

P)()(and)()(33

P)()(and)()(2

P)()(and)()(22

P)()(and)()(1

P)()(and)()(11

)()(

)()(

)()(

xxfxfxfxfxxfxfff

xxfxfffxxfxfff

xxxxfxfxxfxfxfxf

dxdyxfdxdyxfe

dxdyxfdxdyxfe

dxdyxfdxdyxfe

xx

xxxx

ff

(4.3)

In equation (4.3), 1e , 2e and 3e represented the group-conditional error rates of the Bayes rule. The used empiricalapproach considered these conditional error rates as thevolume of solids constituted of successive elementary

volumes of width, dx ( dx = ii xx 1 ), length, dy ( dy

= ii yy 1 ) and height, the value of the bivariate

probability density function at )(dx,dydx . The samemethod was used in the case of 2 and 5 groups.

A total of 1728 combinations of the factors were consideredand for each of them, 100 samples of size gn weregenerated from the g populations. For each of them, the 10

g = 2 g = 3 g = 51 - β = 0.05 0 0 01 - β = 0.4 1.2686 1.6331 2.14461 - β = 0.6 1.7009 2.1901 2.86441 - β = 0.8 2.1851 2.7979 3.6571



77

Boxplots of the mean relative efficiencies (MREff) of theerror rate estimators were presented in Figure 1.

This figure confirms the best performance of 632e , as well

as 2eS , 1eS , eJc , eboot and eCV with however, a lossof efficiency of about 28 % of the latter compared to 632e ,which is equivalent to a mean relative error of 12.8 % forthese estimators for 10 % of relative error for 632e . Exceptthe resubstitution estimator, eA , that presented a loss of

efficiency of more than 100 % compared to 632e , the otherestimators presented losses of efficiency that vary from 28% to 70 % compared to 632e . As far as the dispersion ofthe MREff of the estimators was concerned, Figure 1 showsthe very low variability of 632e , which maintains its bestperformance over the various populations featuresconsidered in this study. Estimators eCV , 0e , epp and

eA that present the lowest performance are also the lessstable.

Figure 1: Boxplots of the MREff of the estimators

G 2 3 5 2 3 5 2 3 5 2 3 5 2 3 5 2 3 5 2 3 5 2 3 5 2 3 5 2 3 5

Global 1 1 1 2 3 4 4 4 4 5 5 5 5 5.5 6 6 6 5 7 7 6 7 7 8 8 9 8 10 10 10N 1 1 1 2 2 3 4 3.5 4 4.5 5 5 5 6 6 5 5 5 7.5 8 7 7 7 8 8 9 9 10 10 10C(12) 1 1 1 2 3 4 4 4 4 4 5 5 5 5 6 6 5.5 5 7 7 6 7 7 8 8 9 9 10 10 10C(8) 1 1 1 2 3 4 4 4 4 5 5 5 5 5 6 6 6 5 7 7 6 7 7 8 8 9 8 10 10 10

C(4) 1 1 1 2 3 4 4 4 4 5 5 5 5 5 6 6 6 5 7 7 7 7 8 8 8 9 8 10 10 10e =0.05 1 1 1 2 3 3 4 5 4 5 5 5 5 4 6 6 6 6 6 6 5 8 8 8 8 9 9 10 10 10e =0.10 1 1 1 2 3 3.5 4 4 4 4 5 5 5 6 6 5 5 5 7 7 6 7 7 8 9 9 8 10 10 10

e =0.151-β=0.05 1 1 1 2.5 3 3 4 4 4 5 5 5 6 5 6 6 6 6 7 7 6 7 7.5 8 9 9 9 10 10 101-β=0.4 1 1 1 2 3 3.5 4 4 4 5 5 5 5 6 6 6 6 6 7 7 6 7 7 8 8 9 8 10 10 10

1-β=0.6 1 1 1 2 3 4 4 4 4 4 5 5 5 6 6 6 5 5 7 7 6 8 7 8 8 9 9 10 10 101-β=0.8 2 1 1 2 3 4 4 4 3 4.5 5 5 5 5 6 5 6 5 7 7 7 7 7 8 8 9 8 10 10 10n /p=1.5 1 1 1 2 2 3 4 4 3 4 4 4 7 6 6 6 5 5 8 7 7 7 7 7 9 9 9 10 10 10n /p=2.5 1 1 1 2 3 3 4 4 4 5 5 5 5 4 5 6 6 5 8 8 7 7 7 8 9 9 9 10 10 10n /p=5 1 1 1 2 4 5 4 4 5 5 6 6 3 4 5 5 6 6 6 4 2 8 8 9 7 9 7 10 10 10

8 9 8.5 10 10 108 8 7 6 7 76 6 6 6 5 54 3 3 5 5 44.5 1 1 2 3 5

632e2eS 1eS eJc eboot eCV eppCV 0e epp eA

Table 2: Median ranks of the estimators according to the populations features


R. GLELE KAKAÏ

The Mean Relative Bias (MRB) helps in appreciating thedirection of the deviation of the estimators’ performancefor 2-, 3- and 5-groups. Table 3 shows that almost all thenon parametric estimators performed well when the numberof groups became more important. For 2- and 3-groupdiscriminant analyses, 1eS and eJc present the lowestabsolute MRB (2.5 % for 2-groups and 0.1 for 3-groups)whereas for 5-groups, 632e became the best with 0.2 % ofabsolute MRB. The resubstitution estimator, eA , presents the most

optimistic bias whereas 0e presented the most pessimisticone.

Table 3. Mean and standard deviation of the MRB of theerror rate estimators

Estimation-14.9 9 -0.5 5.1 -0.2 3.7

-3.1 5 -5.6 4.9 -6.7 4.7

2.4 5.6 -0.2 4 -1.4 2.8

2.5 5.3 0.1 3.9 -0.6 2.8

31.8 15.1 23.4 8.4 16.3 5.8

5.6 7 2.8 4.8 1.5 3.3

39.6 14.8 -26.3 14.1 -14.6 11.8

26.7 14.9 24.2 13 20.7 10.3

-38.4 21.1 -36.9 17.4 -26.4 16.1

-86.3 11.2 -43.1 16.6 -35.9 14.1

g =2 g =3 g =5m σ m σ m σ

632e

2eS1eS

eJcebooteCVeppCV0e

eppeA

DISCUSSION AND CONCLUSION

The estimation of the actual error rate for practical use isone of the relevant topics in discriminant studies and asynthesis of the various estimators of the actual error ratewas provided in McLachlan (1992). Most studies havebeen done to compare in two group-discriminant analysesthe performance of the error rate estimators, especiallyassociated with the linear classification rule, and asynthesis of them was done by Schiavo and Hand (2000).The originality of our study is that the relative efficiencyof non-parametric error rate estimators can be analysed inmulti-group discriminant analysis. The obtained resultshelp to point out the overall best efficiency of 632eirrespective of the number of the considered groups. Fortwo-group linear discriminant analysis, many studies cometo almost the same conclusions (Wehberg and Schumacher,2004; Glèlè Kakaï et al., 2003). Other studies pointed outthe efficiency of this estimator for non-linear classificationrules. Jain et al. (1987), using multivariate normaldistributions in nearest neighbour discriminant analysis,found that 632e out-performed all the other estimators

( eCV , eboot and 0e ). However, we noticed from thepresent study that for high overlap of the populations inthe case of two groups, the performance of this estimatordecreased. Fitzmaurice et al. (1991), using two-groupdiscriminant analysis concluded that 632e became lessreliable as the true actual error rate increased above 0.35,but more reliable as the true error rate decreased. Otherestimators that performed well in the present study were

1eS , 2eS and eJc . On the contrary, , 0e , epp , eppCV

and eA recorded the lowest performance, in most of thecases considered in the study. The ranks of estimators were less influenced by thepopulations’ features, probably due to the fact that theywere all based on resampling methods that do not replicateconditions of use. However, the number of groups had ahigh impact on the performance of the estimators. Thelatter became more efficient as the number of groupsincreased. The highest positive relative bias was obtained by eAwhereas 0e had the highest and negative relative bias.These results have already been obtained by Wehbergand Schumacher (2004), Chatterjee and Chatterjee (1983)and Chernick and Murthy (1985) who qualified eA and

0e as the optimist and pessimist estimators, respectively..

REFERENCES Chatterjee S., Chatterjee S. (1983). Estimation of

misclassification probabilities by bootstrap methods.Communication in Statistics – Simulation andComputation, 12, 645-656.

Chernick M. R., Murthy V. K. (1985). Properties of bootstrapsamples. American Journal of Mathematical andManagement Sciences, 5, 161-170.

Efron B. (1983). Estimating the error of a prediction rule:improvement on cross-validation. Journal of theAmerican Statistical Association, 78, 316-331.

Fitzmaurice G. M., Krzanowski W. J., Hand D. J. (1991). AMonte Carlo study of the 632 bootstrap estimator oferror rate. Journal of Classification, 8, 239-250.

Fukunaga K., Kessell D. L. (1972). Application of optimumerror reject function. IEEE Transactions onInformation Theory, 18, 814-817.

Glèlè Kakaï R., Palm. R. (2006). Methodological contributionto control heteroscedasticity in discriminant analysisstudies. Global Journal of Pure and AppliedSciences, 12 (1), 107-110.



79

Glèlè Kakaï R., Piraux F., Fonton, N. Palm R. (2003).Comparaison empirique des estimateurs de tauxd’erreur en analyse discriminante. Revue deStatistique Appliquée, 51(2), 91-104.

Hand D. J. (1986). Recent advances in error rate estimations,Pattern Recognition Letters, 4, 335-346.

Jain A. K., Dubes R. C., Chen C. (1987). Bootstrap techniquesfor error rate estimation. IEEE Transactions on PatternAnalysis and Machine Intelligence, 9, 628-633.

Lachenbruch P. A. (1967). An almost unbiased method ofobtaining confidence intervals for the probability ofmisclassification in discriminant analysis. Biometrics,23, 639-645.

McLachlan G. J. (1975). Confidence intervals for theconditional probability of misallocation in discriminantanalysis. Biometrics, 31, 161-167.

McLachlan G. J. (1992). Discriminant analysis andstatistical pattern recognition, Wiley, New York.

Quenouille M. H. (1949). Approximate tests of correlationin time series. Journal of the Royal Statistical Society(Serie B), 11, 18-84.

Schiavo R. A., Hand D. J. (2000). Ten more years of errorrate research. International statistical review, 68, 295-310.

Smith C.A.B. (1947). Some examples of discrimination.Annals of Eugenics, 13, 272-282.

Wehberg S., Schumacher M. (2004). A comparison ofNonparametric Error Rate Estimation Methods inClassification Problems. Biometrical Journal, 46, 35-47.

ACKNOWLEDGEMENTS

This work was supported by the Alexander-Von-HumboldtFoundation (AvH) and the Third World Academy ofSciences (TWAS).



FORAGE POTENTIAL, MICRO-SPATIAL AND TEMPORALDISTRIBUTION OF GROUND ARTHROPODS IN THE FLOOD

PLAIN OF A COASTAL RAMSAR SITE IN GHANA

Gbogbo, F., Langpuur, R. and Billah, M. K.

Department of Animal Biology and Conservation Science, University of Ghana, Legon


ABSTRACT: Despite the critical roles played by arthropods in ecosystem functioning and nutrientcycling, a general lack of information about the ecology of many arthropods in West African coastalwetlands persists. An investigation into the abundance, distribution and forage potential of groundarthropods to waterbirds inaWest African Coastal Ramsar site, indicated that the distribution andabundance of the arthropods were similar along both the latitudinal and longitudinal axes of thelagoon’s flood plain. Agelenidae (house spiders), Formicidae (ants) and Gryllidae (True crickets)respectively constituting 52.68%, 36.58% and 5.85% of the total arthropod abundance, dominatedthe 23 families of arthropods. On the basis of percentage biomass and per capita biomass compositions,Gryllidae and Agelenidaewere of the most important to waterbird foraging. Although Formicidaeoccurred in large numbers, the small-size nature of the individuals indicated that they wereof littleimportance to waterbird foraging. Ocypodidae (Ghost and Fiddler crabs) (0.3%) and Acrididae(short- horned grasshoppers) (0.3%) constituted a negligible fraction of the arthropod abundancebuthad the highest per capita biomass and would be the most profitable forage.The low abundanceof Ocypodidae and Acrididae were attributed to marginalisation of the sampling method employedin the study.

INTRODUCTION

The importance of West African coastal wetlands in thesupport of Palearctic migrant waterbirds of the African-Eurasian flyways has been noted (Reneerkens et al., 2009;Ajonina et al,. 2007; Gbogbo 2007; Blomert et al., 1990;Ntiamoa-Baidu and Grieve, 1987, Ntiamoa–Baidu andHepburn, 1987). Over the past few decades, the flyway’spopulations of waterbirds have faced steady declines(Abdourahamane 2010, Underhill et al. 2000, Tripet &Yesou 1998, Zwarts et al. 1998) and several investigationshave since commenced. In Ghana, wetland studies havefocused largely on flora (Oteng-Yeboah, 1999), fisheries(Gbogbo et al., 2008; Ahulu et al., 2006; Entsua–Mensah,2000), waterbirds (Gbogbo and Attuquayefio, 2010;Ntiamoa-Baidu et al.,1998; Ntiamoa–Baidu,1991) andbenthic invertebrates consisting largely of annelids andmolluscs (Gordon, 2000). Arthropods are by far the most important herbivores inmany ecosystems and are valuable food sources for many

species of animals (Siemann et al., 1998; Schmidta et al.2005). Despite these roles played by arthropods inecosystem functioning and nutrient cycling, a general lackof information about arthropods in West African coastalwetlands persists. Waterbird population declines in coastalWest Africa have been linked to several factors, includingcompetition between birds and humans for fisheries(Gbogbo et al. 2008, Van der Winden et al., 2000).Nevertheless, arthropods may be an important food sourcefor waterbirds in coastal West Africa and may supplementthe nutritional demands of waterbirds, particularly undersuch competitive conditions. Besides, wetlands aresensitive ecological areas, and data on the abundance anddistribution of arthropods on wetlands can serve as asource of reference in assessing changes in wetlandecological status resulting from human use, climate changeand pollution. This paper examines the forage potential,micro-spatial and temporal distribution of groundarthropods in the flood plains of the coastal Ramsar site inGhana as a step to identifying their contribution to waterbirdsupport in coastal West Africa.


Forage Potential, Micro-Spatial and Temporal Distribution of Ground Arthropods in theFlood Plain of a Coastal Ramsar Site in Ghana

81

MATERIALS AND METHODS

Study area

The study was carried out at the coastal wetland of SakumoII Lagoon (generallycalled Sakumo Lagoon). Sakumo IILagoon (Fig. 1), located in the Tema Metropolitan Area inGhana, is about 15km East of the capital of Ghana –Accra.Itis one of the five coastal Ramsar sites (Wetland ofInternational Importance) in Ghana, with a totalconservation area of 13.4km2. About 7km2of the Sakumo IILagoon is made up ofalluvial plain and this surroundsthebrackish water lagoon of 3.5km2 (Wetland International,1998). The area of the brackish water lagoon is howeverreduced to about 1 km2 during the dry season (Sep./Oct. toMar./Apr.) (Pauly, 1975).The Lagoon is linked to the seaby a sluice which allows exchange of water with the sea

depending on the tides and rains. The surrounding floraincludes low-lying grasses such as Cyperus sp. andPaspalumthat invade most of the estuary bed,andAvicennia sp. (white mangrove) which hasbasically beenlost due to commercial and household activities that resortto it as a source of fuel wood (Oteng-Yeboah, 1999).Manywaterbird species forage in the flood plains and marginalwaters of the lagoon with few species making use of theopen water. According to BirdLife International (2012),Sakumo II Lagoon serves as a habitat for about 70 speciesof waterbirds, with an estimated maximum number of 30,000.Flocks are usually dominated by Black-wing Stilts(Himantopus himantopus), Ringed Plovers (Charadriushiaticula), Curlew Sandpipers (Calidris ferruginea),Greenshanks (Tringa nebularia), Common Tern (Sternahirundo), Black Terns (Chlidonias niger) and Little Egrets(Egretta garzetta).

Study Locale

Figure 1: Map of Study Area


F. GBOGBO

Demarcation and trap layout

The flood plain area in thewestern bank of the lagoon,constituting the core area used by foraging waterbirds,was divided into three (3) zones based on proximity to thesea (Study Locale, Fig 1). These included the Southernzone(5°36’57.26"N, 0° 2’6.64"W -closest to the sea), theNorthern zone (5°37’41.04"N, 0° 2’27.76"W - thefarthestfrom the sea) - and the Middle zone (5°37’19.16"N,0° 2’12.49"W – lying between the two). Each zone wasfurther divided into three areasbased on proximity to themain body of the lagoon. These included (i) the marshyshoreline area with pockets of sedges and rushes (closestto the lagoon waterfront), (ii) the extensive intertidal areaadjoining the marshy shoreline area (which was either abaremudflat or covered with pockets of Sesuviumportulacastru), and (iii) the dry grassy semi-terrestrial areamarking the maximum water edge of the lagoon (farthestfrom the lagoon waterfront). The Southernzonehowever,hadseveral scattered pools of water and itsdemarcation into shoreline, intertidal and dry grassy areawas not pronounced.

In each of the months of October and December 2011 andFebruary 2012, twenty-eight (28) pitfall traps were set ineach zone in four rows of seven traps. Rows werepositioned parallel to the long axis of the lagoon. TheSouthern zone was however, not sampled in Februarybecause of difficulties with accessibility. As a result ofdifferences in the size of the sampling areas, two of therows of traps were set in the intertidal area adjoining themarshy shoreline of each zone while a row each was set inthe marshy shoreline and dry grassy semi-terrestrial areas.Traps consisted of shallow plastic buckets with 13 cm rimdiameter and 750ml volume. To minimize depletion effectsthat can occur with pitfall trapping (Digweed et al.1995),traps were set at 10m apart but in the intertidal area adjoiningthe marshy shoreline, the two rows of traps were 15m apart.Each pit-fall trap was buried to its rim and the space aroundit filled and smoothened with the dug-out soil. Traps wereabout one-third filled (250-300 ml) with water and a littledetergent added to reduce the surface tension of the waterso that capture organisms would sink and drown in thetraps. The top lid coverings of the traps were raised withsticks to serve as shelter over the traps to prevent excessivedesiccation or flooding from rains. Traps were checked forcatches at 3-7 day intervals, depending on climaticconditions (either too much rain that would flood thecontents away or high temperatures that would dry thecontents). Catches from individual traps were sievedthrough 0.5mm mesh and stored in 70% alcohol in labelledplastic vials for further processing in the laboratory.

Laboratory processing and identification

Specimens were sorted into their respective families,counted and identified using literature and identificationkeys (Katson, 1978; Roth,1993; Jackman, 1997; Ubick etal, 2005), under a Leica EZ4 D microscope. Individualfamilies of organisms were placed in Petri dishes and driedin a laboratory at 55oC for 3 days. The dry weights of thespecimens were then taken with a sensitive electronic scalefor analysis.

Data analysis

Abundance, biomass and distribution data were analysedon the basis ofthe lagoon’s source-to-mouth andshoreline-to-land dimensions, as well as temporal factors. Becauseof the several scattered pools of water in the Southernzone that limited its demarcation into shoreline, intertidaland dry grassy areas, data from the Southern zone wasdiscountedin the shoreline-to-land distributional analysis.Relative abundance (RA) of individual families ofarthropods was calculated as follows:

RA = Number of individuals in a family/ Number of trap nights (Vodzogbe et al., 2005) where onetrap night was defined as one trap setsuccessfully for one night.

In calculating the RA, trap nights were only based on trapsthat successfully passed all nights in the trapping session.Traps that were overturned, removed, destroyed byanimalsor wetland users, or which got completelysubmerged in the lagoon water were not considered to besuccessful.

The Shannon-Wiener index was used to determine thewithin-habitat diversity, whereas the Pielou’s index Emeasured the within-habitat evenness (Magurran, 1988).Sorensen’s index (Cs) was used to measure the multivariatefamily overlaps among sample areas (Magurran, 1988).

RESULTS

General occurrence and biomass of ground arthropods

A total of 3,037 individual arthropods belonging to three(3) classes, eleven (11) orders and 23 familieswere capturedby the pitfall traps (Table 1). Arachnida constituted 52.68%of the total catch, compared to Insecta (47.29%) andCrustacea (0.03%). All the captured Arachnida belongedto the family Agelenidae in contrast to 21 families of Insecta



83

Class Order FamilyFamily Common

Name

Total Occurren

ce (%)

Total Biomass

(%)

Per capita Biomass

(g)

Arachnida Araneae Agelenidae House spiders 52.68 42.42 0.01

Insecta Hymenoptera Formicidae Ants 36.58 4.67 <0.00

Apidae Common Bees 0.49 0.24 <0.00

SphecidaeThread-Waisted Wasps 0.3 0.4 0.01

Pompilidae Spider Wasps 0.07 0.12 0.02

Orthoptera Gryllidae True Crickets 5.85 11.49 0.02

TetrigidaePygmy Grasshoppers 0.72 0.79 0.01

AcrididaeShort-horned grasshoppers 0.3 10.65 0.22

TettigonidaeLong-horned Grasshoppers 0.36 0.59 0.01

Coleoptera Gyrinidae Whirligig Beetles 0.3 4.04 0.11Coccinnelidae Ladybird Beetles 0.03 0.08 <0.00Carabidae Ground Beetles 0.16 0.12 0.04Belidae Primitive Weevils 0.1 0.28 0.02Staphylinidae Rove Beetles 0.03 0.28 0.07

Blattaria Blattidae Cockroaches 0.07 0.55 0.07Diptera Sciomyzidae Marsh Flies 0.26 2.06 0.07

SepsidaeBlack Scavenger Flies 0.2 0.67 0.03

Mydidae Mydas Flies 0.07 0.24 0.03

Hemiptera NaucoridaeCreeping Water Bugs 0.73 5.82 0.07

Isoptera Termitidae Termites 0.23 <0.00 <0.00Lepidoptera Cossidae Miller Moths 0.07 0.16 0.02

Trichoptera LeptoceridaeLong-horned Caddis Flies 0.03 0.04 0.01

Crustaceae Decapoda OcypodidaeGhost and Fiddler Crabs 0.03 13.7 3.46

Table 1: Occurrence and biomass of ground arthropod taxa in Sakumo II Lagoon


F. GBOGBO

and one family of Crustacea. The most dominant familiesof insects were the ants, Formicidae, and the True crickets,Gryllidae - constituting 36.58% and 5.85% respectively ofthe arthropod catches.

Total biomass of the captured arthropods was 25.25g, ofwhich Arachnida constituted 42.42% compared to 43.88and 13.7% respectively, for Insecta and Crustacea. Amongthe insects, Orthoptera contributed 23.52% of the totalarthropod biomass, followed by Hemiptera(5.82%),Hymenoptera (5.43%) and Coleoptera (4.8%). Theremaining five orders of insects constituted only 9.11% ofthe total arthropod mass. Data on the per capita massindicated Ocypodidaeas arthropods with the highest massper individual, followed by Acrididae. Although Formicidaeoccurred in high numbers, its per capita biomass was verysmall. Source-to-Mouth distribution

The relative abundance of the individual families acrossthe three zones is presented in Table 2. No significantdifferences existed among the relative abundance of thearthropods in the Northern, Middle and Southern zones(ANOVA, F = 0.001, F Critical 2,51 = 3.187, p > 0.05), a ShannonDiversity Indexvalues(Table 3) indicating particularly lowfamily diversity in the Southern zone while Pielou indexvalues (Table 3) suggestedpronounced species dominancein the Southern zone. Indeed, Table 2shows the relativeabundance of Agelenidaeto beparticularly higher than thatof the remaining families in the Southern zone and thusconfirming its exceptional dominance in the Southern zone.

The most dominant families in the Northern zone wereFormicidae (1.99), followed by Agelenidae (1.5). These twospecies also dominated the Middle zone, but in a reversedorder (Table 2). Sorensen’s Index value of 0.79 was obtainedfor the Northern and Middle zones, 0.56 for the Northernand Southern zones and 0.28 for the Middle and southernzones. Thus, arthropod family composition of the Northernand Middle zones were more similar (79%)than that of theNorthern and Southern zones (56%) andthe Middle and

Southern zones (28%), indicating that the arthropodcommunity composition changed with distance from themouth of the lagoon. Table 2: Source-to-mouth distribution of ground arthropodin the Sakumo II Lagoon

SHORELINE-TO-LAND DISTRIBUTION

The total relative abundance of arthropods in the marshyshoreline habitat was 4.63 compared to 3.28 and 4.34 in the

Relative Abundance

FAMILY Northern Zone

Middle Zone

Southern Zone

Agelenidae 1.5 2.31 1.21

Formicidae 1.99 1.06 0.01

Apidae 0.02 0.01 0.02

Sphecidae 0.02 0.01 -

Pompilidae 0.01 - 0.01

Gryllidae 0.17 0.26 0.12

Tetrigidae 0.02 0.04 -

Acrididae 0.02 0.01 0.01

Tettigonidae 0.02 < 0.01 0.01

Gyrinidae 0.01 0.01 -

Coccinnelidae 0.02 0.01 -

Carabidae 0.01 0.01 -

Belidae - 0.01 -

Staphylinidae < 0.01 < 0.01 -

Blattidae - 0.01 -

Sciomyzidae 0.01 0.01 0.01

Sepsidae 0.01 0.01 -

Mydidae - 0.01 -

Naucoridae 0.01 0.05 -

Termitidae - 0.02 -

Cossidae 0.01 - -

Leptoceridae - < 0.01 -

Ocypodidae - < 0.01 -

TOTAL 3.87 3.83 1.39



85

Table 3: Ecological diversity values on spatial and temporal scale

intertidal area and dry grassy – semi terrestrial arearespectively (Table 4).These values are statistically thesame (ANOVA, F =0.104, F Critical 2, 45 = 3.211, P > 0.05).Relative abundance of the individual families indicated thedominance ofAgelenidae and Formicidae in each of thedemarcations. Shannon Weiner diversity and Pielou’s Indexvalues for the Marshy shoreline, Intertidal area and thedry grassy semi-terrestrial area, are shown in Table 3and further confirms family dominance. All the threedemarcations had similar community composition ofground arthropods with the Marshy shoreline and theIntertidal areas scoring a Sorensen’s index of 0.63. Similarlythe marshy shoreline and the dry grassy semi-terrestrialarea scored Sorensen’s index value of 0.76, while theintertidal area and the dry grassy semi terrestrial area had0.69.

TEMPORAL DISTRIBUTION

Agelenidae and Formicidae dominated the arthropodsthroughout the period (Table 5). This observation issupportedby the Pielou’s Index values(Table 3). ShannonDiversity values also remained fairly stable over time (Table3). Similar to the spatial distribution of the arthropods, therelative abundance data again indicated the dominanceofAgelenidae and Formicidaethroughout the period. Theabundance of the arthropods (Table 5) over the studyperiod however remained statistically similar (ANOVA, F =0.396), F critical 2,45 = 0.3211). Sorensen’s Index values showedsimilar community composition of arthropod familiesbetween October and December (0.67), December andFebruary (0.62) and October and February (0.60).

Table4: Shoreline-to-land distribution of ground arthropodinthe Sakumo II Lagoon

Ecologicaldiversity

Source-to-Mouth distribution Shoreline-to-land distribution Temporal Distribution

Northern Zone

Middle zone

Sourthern zone

Marshy Shoreline

Intertidal area

Dry grassy semi-terrestrial area

October December February

Shannon-Weiner 1.1 1.15 0.42 1.12 1.04 1.17 0.1 1.12 1.35Pielou 0.4 0.38 0.19 0.44 0.37 0.39 0.35 0.44 0.45

Relative Abundance

Family Marshy Shoreline

Intertidal area

Dry grassy semi-terrestrial area

Agelenidae 2.06 2.01 1.59Formicidae 1.93 0.99 2.19Apidae 0.05 0.01 0.01Sphecidae 0.02 0.01 0.01Pompilidae < 0.01 - -Gryllidae 0.31 0.1 0.36Tetrigidae 0.02 0.03 0.04Acrididae 0.05 0.02 0.03Tettigonidae 0.02 - 0.03Gyrinidae 0.04 0.02 -Coccinnelidae 0.01 0.02 0.01Carabidae 0.01 0.01 -Belidae 0.02 - -Staphylinidae 0.01 - -Blattidae 0.01 - -Sciomyzidae < 0.01 - 0.02Sepsidae 0.02 - 0.02Mydidae < 0.01 - 0.01Naucoridae 0.02 0.05 0.01Termitidae 0.02 0.01 -Cossidae < 0.01 - 0.01Leptoceridae < 0.01 - -Ocypodidae 0.01 - -TOTAL 4.61 3.29 4.32


F. GBOGBO

FamilyRelative Abundance

October December FebruaryAgelenidae 2.01 1.78 1.14Formicidae 1.71 1.71 0.11Apidae 0.02 0.02 0.01Sphecidae - 0.02 0.01Pompilidae - - 0.01Gryllidae 0.13 0.15 0.47Tetrigidae <0.01 0.02 0.08Acrididae 0.02 0.03 0.06Tettigonidae 0.01 0.01 0.02Gyrinidae 0.02 - 0.01Coccinnelidae 0.01 0.02 -Carabidae < 0.01 0.01 0.01Belidae 0.01 - -Staphylinidae - < 0.01 -Blattidae < 0.01 < 0.01 -Sciomyzidae < 0.01 - 0.04Sepsidae - 0.02 -Mydidae < 0.01 - -Naucoridae 0.04 - 0.01Termitidae < 0.01 - 0.02Cossidae - - 0.01Leptoceridae < 0.01 - -Ocypodidae - - 0.01Total 4 3.79 2.07

Table5: Temporal distribution of ground arthropod in theSakumo II Lagoon

DISCUSSION

This study established Agelenidae,Formicidae and Gryllidaeas the most dominant ground arthropod families in SakumoII Lagoon. However, on the basis of percentage biomasscomposition, Agelenidae, Ocyporidae, Gryllidae, andAcrididae were the highest contributors. AlthoughFormicidae constituted 36.58% of the total arthropodabundance, its per capita biomass was negligible andcomparable to that of Apidae, Coccinnelidae and Termitidaeeach of which constituted lees than 0.5% of the totalarthropod abundance. The Optimum Foraging Theorypredicts foraging organisms to reject small prey items ifthey are less profitable (Charnov, 1976; Yahnke, 2006). Thisimplies that although Formicidae occurred in large numbersin Sakumo II Lagoon it might not be an important foodsource for waterbirds because of their small sizes.

Ocypodidae might be the most profitable prey to waterbirdsamong the arthropods. However, pitfall traps are not the

best for sampling decapods and thus the sampling methodemployed in this study would largely under-represent theoccurrence of Ocypodidae. Much as this family potentiallyappears to be the most profitable prey to waterbirds in thisstudy, it should be borne in mind that predators usuallyhave some critical maximum size of food items above whichhandling time and energy expended becomes unprofitable.So the suitability of the Ocypodidae as the most profitablefamily will also depend on the range and sizes of waterbirdsfound at the study site. The abundance and importance ofdecapods to waterbirds’ foraging activities have earlierbeen noted using specialised traps (Gbogbo et al., 2008).Other families of arthropods that might be under-represented in this study as a result of the employedsampling methods include Acrididae, Sciomyzidae,Mydidae, Sepsidae, Coccinnelidae, Cossidae, Tettigonidae,Tetrigidae, Pompilidae, Sphecidae and Apidae. However,the fact that these families occurred in the pitfall trapssuggest that they might abound in the study area and thebiomass data indicates that Acrididae in particular, mayconstitute a significant food source for the waterbirds.Among the typical ground arthropods however, Gryllidaeand Agelenidae appeared to have higher potential as foodsources for waterbirds based on the product of theiroccurrence and per capita biomass.

In structuring waterbird foraging habits and diets in coastalGhana, species belonging to guilds 2, 3 and 4 have beennoted as the only waterbirds that feed on invertebrates(Ntiamoa-Baidu et al., 1998). Foraging microhabitats usedby theses pecies ranged from the dry grassy boundary ofthe flood plain to water depth of 14cm but with differentspecies exhibiting different limits and preferences (Ntiamoa-Baidu et al.,1998). The fact that the abundance of groundarthropods was similar among the shoreline, intertidal areaand the dry grassy terrestrial areas, as well as among thelagoon’s source, middle and mouth, are indications thatthe entire flood plain of the lagoon is of the same foragequality to waterbirds. Thus, abundance and/or distributionof ground arthropods did not appear to be the major factorsdetermining the choice of foraging area by the waterbirds.

The low similarity between the arthropod communitycomposition between the Southern and Middle zones,aswell as the Northern and southern zones is however anindication that the arthropod community compositionindeed changes with distance from the mouth of the lagoon.Hypersaline conditions in lagoons and estuaries oftencharacterise lower reaches of many estuaries and lagoonsand are known to impact negatively on invertebratepopulations (Gordon, 2000). Piersma and Ntiamoa-Baidu(1995) conceptualised the population of benthicinvertebrate to fall drastically under hypersaline conditions,



87

and that many such populations are restored with the influxof rains (Gordon, 2000). The mouth of the lagoon may becharacterised with similar hypersaline condition since thisstudy was carried out during the dry season. There istherefore the need to investigate seasonal changes inground arthropod communities in West African coastallagoons to ascertain possible influences of salinity.

In relation to temporal distribution, many Palearctic migrantwaterbirds arrive in coastal West Africa by September /October and depart by March/ April (Ntiamoa-Baidu, 1991).The similarity in the community composition and relativeabundance of the arthropods throughout the months ofthe study however indicated that the presence of the birdsneither affected the community composition nor depletedthe abundance of the arthropods. In recognition of the diverse ground arthropod family andtheir abundance in the study area, there is the need toinvestigate their utilisation by waterbird species on thecoast. Several species of egrets and herons described asexclusive fish-eating in coastal Ghana (Ntiamoa-Baidu etal., 1998), were reported to feed on morediverse organisms,including arthropods in some other part of the world(Liordos, 2010). This work is however part of a more diversestudy aimed at investigating the importance of groundarthropods and aerial insects to foraging waterbird incoastal Ghana. Further studies should be designed toidentify selectivity of the arthropod family and theirutilisations as forage, particularly by waterbird species thatfeed on both invertebrates and fish.

CONCLUSIONS Distribution and abundance of ground arthropods weresimilar along both the latitudinal and longitudinal axis ofthe flood plains of Sakumo II Lagoon. The occurrence ofground arthropods in the flood plains of the lagoon wasdominated by Agelenidae and Formicidae, but,the smallsizes of the Formicidae indicated that they were of littleimportance to waterbird foraging. Ground arthropodfamilies in the lagoon that may be of importance to waterbirdforaging were Gryllidae and Agelenidae. Acrididae andOcypodidae also had forage potential to waterbirds, butrecorded low abundance values which were attributed tomarginalisation by the sampling method employed.

ACKNOWLEDGEMENTS

The authors acknowledge the contributions of Mr. HenryDavis of the African Regional Postgraduate Program inInsect Science (ARPPIS), and S.K.B Boni of the Departmentof Animal Biology and Conservation Science (DABCS),University of Ghana. The support of Mr. DanielOsaeYeboah, Adaworomah Bernard Bobson, Ernest Asanteand Grace Rechelle Brown-Engmann, during the field workis noted and appreciated.

REFERENCES

Abdourahamane, I. S. (2010). Waterbird as bioindicatorsof wetland quality: case study of the Muni-PomadzeRamsar site, Ghana. MPhil Thesis, University of Ghana,Legon. 178pp

Aheto, D. W., Asare, C. Mensah, E. A and Aggrey-Fynn, J.(2011). Rapid Assessment of anthropogenic impactson exposed sandy beaches in Ghana using GhostCrabs (Ocypod spp.) as Ecological Indicators .Momona Ethiopian Journal of Science 3: 93-103,

Ahulu, A. M., Nunoo, F. K. E. and Owusu, E. H. (2006).Food preferences of the Common Tern, Sterna hirundo(Linnaeus, 1758) at the Densu Floodplains, Accra.West African Journal of Applied Ecology 9: 1-7

BirdLife International, (2012) Important Bird Areasfactsheet: Sakumo Lagoon Ramsar Site. http://www.birdlife.org on 15/05/2012

Charnov, E. L. (1976). Optimal foraging, the Marginal ValueTheorem. Theoretical Population Biology 9:129-136

Digweed, C.S, Currie, C. R, Carcamo, H.a and Spence J. R.(1995). Digging out “the digging” in effect: influencesof depletion and disturbance on catches of groundbeettle (Coleoptera, Carabidae). Pedobiologia 39:561-

Entsua –Mensah, M., Ofori-Danson, P.K and KorantengK.A. (2000). Management issues for the sustainableuse of Lagoon fish resources. In: Abban, E.K., Cosal,C.M.V., Falk, T.K and Pullin R.S.V (Eds) Biodiversityand sustainable use of fish in the coastal zone,ICLARM Conference Proceedings , 24-27

Gbogbo, F. (2007). The importance of unmanaged wetlandsto waterbirds at coastal Ghana. African Journal ofEcology 45: 599-606

Gbogbo, F. and Attuquayefio D.K. (2010). Issues arisingfrom changes in waterbird population estimates incoastal Ghana. Bird populations 10: 79-87

Gbogbo, F., Oduro, W. and Oppong, S. (2008). Nature and


F. GBOGBO

patterns of lagoon fisheries resource utilisation andits implications to waterbird management in coastalGhana. African Journal of Aquatic Science 33:211-222

Gordon, C. (2000) Hypersaline lagoons as conservationhabitats: macro-invertebrates at Muni Lagoon, Ghana.Biodiversity and Conservation 9: 465 – 478

Jackman. J. A. (1997). A Field Guide to Spiders andScorpions of Texas. Taylor Trade Publishing 202pp\

Liordos, V. (2010). Foraging guilds of waterbirds winteringin a Mediterranean coastal wetland, Zoological Studies49: 311-323

Kaston, B.J. (1978). How to Know the Spiders, third ed.Wm C. Brown Company Publishers. Dubuque, Iowa.272 pp.

Magurran, A.E. (1988). Ecological Diversity and itsMeasurements. Princeton University Press, NewJersey.179 pp

Ntiamoa-Baidu Y., Piersma T., Wiersma P., Poot M., BattleyP. and Gordon C. (1998). Habitat selection, dailyforaging routines and diet of waterbirds in Coastallagoons in Ghana. Ibis 140:89-103

Ntiamoa–Baidu, Y. (1991). Seasonal changes in theimportance of coastal wetlands in Ghana for wadingbirds. Biological Conservation 57:139-158

Oteng-Yeboah, A. A. (1999). Biodiversity studies in threecoastal wetlands in Ghana West Africa. Journal of theGhana Science Association 3:147-149

Pauly, D. (1975). On the ecology of a small West AfricanLagoon. Berichte derDeutschenWissenschaftlichenKommission furMeeresforschung 24: 46-62

Piersma, T. and Ntiamoa – Baidu, Y. (1995). Waterbirdecology and the management of coastal wetlands inGhana, Netherlands, NIOZ, 105pp

Roth, V. D.( 1993). Spider genera of North America. Thirdedition. American Arachnological Society, Gainesville,Florida.

Tripet, P. and Yesou, P. (1998). Mid-winter counts in theSenegal Delta, West Africa, 1993–1997. Wader StudyGroup Bulletin 85:83–87.\

Ubick, D., Paquin, P., Cushing, P.E, Roth, V. (Eds) (2005).TheSpider Genera of North America: An IdentificationManual. The American Arachnological Society, 377pp.

Underhill, L. G., Whittington, P. A. and Calf, K. M. (2000).Shoreline birds in Robben Island, Western Cape,South Africa. Wader Study Group Bulletin 96, 37–38.

Van der Winden, J., Nyame, S. K. , Ntiamoa-Baidu, Y. andGordon, C. (2000). Black Terns in Ghana, October 2000.Bureau Waardenburgbv, Netherlands. 61 pp.

Vodzogbe, F., Attuquayefio, D.K., and Gbogbo, F. (2005).The flora and mammals of the moist semi-deciduousforest zone in the Sefwi –Wiawso District of theWestern Region, Ghana. West African Journal ofApplied Ecology 8:49 – 64

Wetland International, (1998). Coastal WetlandManagement Project, Wildlife Department.www.wetlands.org/reports/r is/1GH004en.pdf(Accessed: 29/10/09).

Yahnke, C. J. (2006). Optimal foraging theory: using birdpredation on goldenrod galls. The American BiologyTeacher 68: 471- 475

Zwarts, L., Kamp, J., Overdijk, O., Spanje, T. M., Veldkamp,R., West, R. and Wright, M. (1998). Wader count ofthe Banc d’ Aguine, Mauritania in January February1997. Wader Study Group Bulletin 86: 53-69

8 9 AJST, Vol. 12, No.1: October, 2012


DIVERSITY OF THE CHIRONOMIDAE (DIPTERA) OF RIVER NIGERRELATED TO WATER POLLUTION AT NIAMEY (NIGER)

1*Bassirou Alhou, 2Jean-Claude. Micha and 3Boudewijn Goddeeris

1University Abdou Moumouni of Niamey, Faculty of Sciences,Department of Biology, B.P: 10,662 Niamey (Niger)

2FUNDP, Department of Biology, Unit of Research in Biology of Organisms,

61, Rue de Bruxelles B-5000 Namur (Belgium)

3Royal Belgian Institute of Natural Sciences, RBINS,Freshwater Biology, Rue Vautier 29, B-1000 Brussels

ABSTRACT: This paper presents the first results on the water quality of the River Niger at Niameybased on the Chironomidae. Artificial substrata of stones covered with galvanized wire netting wereused for collecting the chironomid larvae. Water samples were taken for physicochemical analysis.Twenty taxa of chironomidae were collected. Among these, Chironomus gr. plumosus represented51%and Microchironomus sp. 26% of all larvae collected. The distribution of Chironomidae throughoutthe river showed differences between the sampling stations. Organic pollution and nutrients loadingwere the main factors explaining the differences in spatial distribution. Some taxa, e.g. Chironomusgr. plumosus and Microchironomus sp. are positively correlated to those factors, while others asTanytarsini, Nilodosis sp., Micropelopinae and Procladius sp. are negatively correlated. Accordingto these results, Chironomidae appear an excellent tool for the assessment of the biological qualityof western African rivers.

Key words: Chironomidae, River Niger, Niamey, water quality

.

INTRODUCTION

Chironomidae constitute the most diverse group ofaquatic insects: their larvae are aquatic, but the adultsterrestrial (some species have terrestrial larvae). Up tonow, 4 147 species are known, of which 406 are fromtropical Africa (Ferrington 2008). New species arecontinuously discovered; the African species especially,are poorly known (Eggermont and Verschuren 2003).Dejoux (1984) enumerates in West Africa 96 species fromTogo and Benin and 31 species from Niger.

Because of their diversity and their specific sensitivity toenvironmental changes, chironomids are widely used inecological investigations (Rosenberg and Resh 1993).Therefore, they are also used in monitoring water qualityof lakes and rivers (Sharley et al. 2004; Callisto et al.2002; Evrard 1996).

This study is a first attempt to use chironomid diversity inevaluating the water quality of a main African river, i.e.River Niger, with special attention to the pollution aspectsof Niamey, the capital of Niger. This survey is also acontribution to the knowledge of the diversity ofChironomidae of a West African river.

MATERIAL AND METHODS

Study area

Chironomidae are sampled along the River Niger at Niamey,the capital of Niger in West Africa (218 m asl, 3°31' Nand2°26' E) (Fig. 1).

The climate is ar id: rainfall, temperature andevapotranspiration averages during 1995 - 2005 wererespectively 517.90 mm, 29.8°C and 2 802.5 mm. River Niger

90AJST, Vol. 12, No.1: October, 2012

B. ALHOU

1 Tondibia (TON)

2 National Hospital of Niamey (HNN)

3 University of Niamey (UAM)

4 Big Hotel (BH)

5 Tannery (TAN)

6 MESS (ME)

7 BRANIGER (BRA)

8 ENITEX (ENI)

9 Saga (SA)

1 0

Figure 1: Map of the localization of the sampling stations in the River Niger near Niamey (RHNL = National Hospitalof Lamordé sewage; RHNN = National Hospital of Niamey sewage; RUAM = University Abdou Moumouni sewage;RGOU = Goutiyena domestic sewage; RBH = Big Hotel domestic sewage; RMES = Mess domestic sewage; RSLA =

Slaughterhouse sewage; RBRA = Brewery sewage; RENI = Nigerian Enterprise of Textile sewage).

AJST, Vol. 12, No.1: October, 2012

Diversity of the Chironomidae (Diptera) of River Niger related to Water Pollution at Niamey (Niger)

91

is the unique main river of the country. Its water regime atNiamey greatly depends on the upstream rainfall seasonsof its very large basin and is characterized by high waterslevels from November to February, low waters from Marchto June, and a small ascent of waters bound at the localrain season in July - September. Sampling methods Ten sampling stations were selected according to thewastewater discharges of Niamey city. One station (TON)is located upstream of the discharges, eight stations inthe discharge area of the city (HNN, UAM, BH, TAN, ME,SLA, BRA, ENI) and one station (SA) downstream. Water samples were taken, to 30 cm of depth and 3 m ofthe banks, in the morning between 9 to 10 hours at eachstation in order to respect the flux of wastewaterdischarges. The sampling was done from March to July2004, from January to July 2005 and from November 2005to January 2006. Three samples were taken monthly (tendays interval) and conserved in polyethylene bottles at4°C for laboratory analysis. The following variables wereanalysed: pH, conductivity, dissolved oxygen,orthophosphates, total phosphorus, nitrates, nitrites,ammonium nitrogen, chemical oxygen demand andtemperature. Conductivity, pH, temperature and dissolvedoxygen were measured in situ respectively byconductivity WTW LF 318/SET; pH WTW 330i/SET andoxygen WTW Multiline P3 pH/Oxi-SET. At the laboratory,nitrate was analyzed according to the cadmium reductionmethod, nitrite by diazotisation method, ammoniumnitrogen by Nessler method, chemical oxygen demand byreactor digestion method, orthophosphates by phosVer 3methods and total phosphorus by acid persulfatedigestion process, using a DR/2000 spectrophotometeraccording to Hach manual.Two sampling methods were used to collect Chironomidae.Chironomidae were sampled during low water (< 50 m3 s-1)in May-June 2004 and April-May 2005 by artificialsubstrates made of stones in galvanized wire netting (48cm length, 38 cm wide and 10 cm high). Four artificialsubstrata were used in 2004 and eight in 2005 once peryear and per station. The artificial substrata stay 6 weeksin water. Organisms were sorted (die) out at the laboratory througha column of sieves (5 mm, 1 mm and 0.4 mm mesh size). All the collected specimens were preserved in formalin10% before identification.

Specimens identification Chironomid head capsules were cut and mounted ineuparal ventral face upwards on microscopal slides (thebody was mounted laterally together with the headcapsule). Identifications were done with a microscopeReichert Zetopan at 100-400x magnification by referringto Eggermont and Verschuren (2004a, 2004b, and 2003),Eggermont et al. (2005), Eggermont (2004), Wiederholm(1983), Moller Pillot (1984) and Durand & Lévêque (1981).The majority of the specimens (63%) have been identifieddirectly without preparation under a binocular WILD M10 Leica. Identification of larvae was based on ventral tubuli, andon head capsule features such as mentum, eye patches(form, position, number), mandibules, ventromental platesand antennal characteristics. Statistical analysis The characterization of the stations is done on a matrix ofphysicochemical parameters using data starting twomonths before chironomid sampling. This matrix isconsidered as representative of the mean water quality ofeach sampling station to which the observed chironomidcommunity was submitted and able to react (Younes-Baraille et al. 2005; Ndaruga et al. 2004; Usseglio-Polateraand Beisel 2002). The physicochemical data had to be normalized andstandardized according to Legendre and Legendre (1998).The abundance of taxa per artificial substrata have beenconsidered in the biological matrix. Chironomid abundancedata were log (x + 1) transformed before statistical analysisin order to normalize and stabilize the variance. The stations and species have been ordinated accordingto the environmental parameters by using canonicalcorrespondence analysis (CCA, Miserendino and Pizzolon2003; Miserendino 2001). Monte Carlo test allows selectingthose environmental factors that explain significantly thedistribution of taxa along the River Niger at Niamey. Thismethod eliminates all environmental variables presentingan inflation factor superior to 10 (Ter Braak and Smilauer1999). These factors were correlated by those variablesso that they were less contributed to explain thedistribution of chironomids. The option down-weightingof rares species was applied. CANOCO for Windows for the ACC is the software usedhere (Ter Braak and Smilauer 1999).


B. ALHOU

The analysis of community structure is based on the taxarichness, Shannon-Weaver diversity and Shannonequitability index (Spellerberg and Fedor 2003).

The analysis of community structure is based on the taxarichness, Shannon-Weaver diversity and Shannonequitability index (Spellerberg and Fedor 2003).

RESULTS

Structure of the communities

51% were of Chironomus gr. plumosus Linnaeus, 1758and 26% of Microchironomus sp. Kieffer, 1918. Theproportions of the other species vary from 0.01 to 10%.

The total number of sampled larvae varies significantlyfrom upstream to downstream of each wastewaterdischarge station: i.e. 220 larvae at TON, 436 at HNN, 158at UAM, 2 002 at BH, 98 at ME, 210 at BRA, 1 022 at ENIand 989 at SA. The number of larvae was very low at TANand SLA, with 4 and 11 individuals respectively. Sixteen taxa were collected, all sampling stations lumped.The highest score, i.e. 11 taxa, was recorded at UAM andthe lowest, i.e. 1 taxon, was obtained at TAN and BRA(Tab. 1). The results of the Shannon-Weaver index (Legendre andLegendre, 1979) show a difference between samplingstations (Tab. 2).

Taxa\Stations TON HNN UAM BH TAN ME SLA BRA ENI SAChironominaeChironomini sp. 1 39 37 14 0 0 0 0 0 0 433Chironomini sp. 2 0 2 9 44 0 2 0 0 0 0Chironomus gr.plumosus Linnaeus, 1758

0 286 14 1045 4 51 9 210 981 35

Cryptochironomini sp. 0 0 0 0 0 0 0 0 0 0Cryptochironomus sp. 1 Kruseman, 1933

0 0 23 0 0 0 0 0 0 0

Cryptochironomus sp.2 Kruseman, 1933

0 0 27 0 0 0 0 0 0 44

Dicrotendipes sp. Kieffer, 1913

0 2 5 0 0 0 0 0 0 0

Glyptotendipes sp. Kieffer, 1913

0 0 0 59 0 3 0 0 0 0

Microchironomus sp. Kieffer, 1918

80 57 14 845 0 41 0 0 41 256

Nilodosis sp. 0 0 5 0 0 0 0 0 0 0Parachironomus sp. Lenz, 1921

0 10 0 0 0 0 0 0 0 0

Polypedilum spp. 1 Wulp

18 39 18 0 0 0 0 0 0 159

Polypedilum spp.2 Wulp

25 0 0 7 0 0 2 0 0 0

Tanytarsini spp. 16 0 0 0 0 0 0 0 0 53Xenochironomus sp. 0 0 0 0 0 0 0 0 0 0TanypodinaeAblabesmya sp. 23 2 14 0 0 0 0 0 0 9Cf Procladius sp. 0 0 0 0 0 - 0 0 0 0Clinotanypus sp. 0 0 0 0 0 - 0 0 0 0Micropelopiinae sp. 18 0 18 0 0 - 0 0 0 0OrthocladiinaeOrthocladiinae sp. 1 - 0 0 0 - 0 0 0 0Number of taxa 8 8 11 5 1 5 2 1 2 7

Table 1: Taxonomic list of Chironomidae sampled in different stations along River Niger near Niamey (+ = present; -= absent). Genus determination according to Wiederholm (1983) and Eggermont (2004)



93

Table 2: Characteristics of the Chironomidae fauna of the River Niger near Niamey

Stations Substrats Abundance Taxonomic richness

Index of Shannon

Shannon Equitability (%)

TON Sand + clay+ stones 220 8 2.54 91HNN Clay 436 8 1.65 55UAM Sand 158 11 3.3 95BH Clay 2 002 5 1.32 57TAN Clay + plant remnants 4 1 0 -ME Clay 98 5 1.32 57SLA Clay + stomach content 11 2 0.91 35BRA Clay + sand 210 1 0 -ENI Clay 1 022 2 0.24 24SA Clay+ stones 989 7 2.14 71

(AS = Artificial substrata; HN = Hand net; TON = Tondibia ; HNN = National Hospital of Niamey; UAM = University ABDOUMoumouni ; BH = Big Hotel; TAN = Tannery ; ME = Mess ; SLA = Slaughterhouse ; BRA = Brewery; ENI = Nigerian Enterprise ofTextile; SA = Saga)

The Shannon-Weaver index reached a score 2 at TON,UAM and SA, was between 1 and 2 at BH, HNN and MEand inferior to 1 at SLA, ENI, BRA and TAN.

The most elevated equitability index is recorded to UAM(95%). It is from 91% at TON and 71 at SA. It was on theother hand very low to ENI (24%) and SLA (35%). HNN,BH and ME presented an intermediate equitability indexranged between 50 and 60%.

Distribution of chironomids according to environmentalvariables

The distribution of taxa according to the environmentalparameters (P < 0.05) is shown in Figure 2.

The total amount of variance of the environmental factors(Fig. 2A) explained by axis 1 and 2 of the CCA was 42%

(31% explained by the first axis and 11% by the secondaxis).

Tree significant variables (conductivity, nitrites,ammonium,) explained 47% of the variance (0.613 of totalinertia) of taxa distribution.

In ordination of the environmental variables the first axissupported information relating to organic matter andnutrients. The second axis was correlated to conductivityand nitrites (Fig. 2A).

The axes 1 and 2 of the CCA for sampling stations (Fig.2B) clearly differentiate “TON, UAM, SA” and “BRA,SLA, BH, ENI, TAN, ME, HNN”. The first axis mainlyexplains this difference suggested by a gradient in organicmatter and nutrient concentrations (cf. mean value in Table3).


B. ALHOU

Figure 2: Ordination of environmental factors (A), taxa (B) and sampling (C) in the two first axes of the CCA(artificial substrata method, a = May-June 2004, b = April-May 2005).

A

-0,8 -0,6 -0,4 -0,2 0,0 0,2 0,4 0,6 0,8 1,0 1,2-0,8

-0,6

-0,4

-0,2

0,0

0,2

0,4

0,6

0,8

1,0

Temp

pH

DCO

NO3

PO4

Ptot

O2Pr

Cond

NH4

NO2

B

Axe 1 (31 %)

-1,5 -1,0 -0,5 0,0 0,5 1,0 1,5

Axe 2 (11 %)

-1,2

-1,0

-0,8

-0,6

-0,4

-0,2

0,0

0,2

0,4

0,6

0,8

1,0

Chip

Mchi

Pchi

Cryp 1

Cryp 2

Chir 1

Chir 2

Glyp

Pol1

Pol2

Dicr

Nil

Abla

Mpel

Tan

C

Axe 1 (31 %)

-2,0 -1,5 -1,0 -0,5 0,0 0,5 1,0 1,5 2,0

Axe 2 (11 %)

-2

-1

0

1

2

3

SLAbBRAa

BRAb

ENIaENIb

BHa BHb

HNNa

HNNb

MEaMEb

SA a

SA b

TANa

TONa

TONb

UAMa

UAMb



95

Stations /Parameters TON HNN UAM BH TAN ME SLA BRA ENI SA

T (° C)

M 25.5 30.1 29.7 26.4 25.9 26.2 26.4 26.2 21.6 25.9

SD 3.01 0.75 0.21 3.41 3.13 3.37 3.31 3.18 3.21 3.05

Min 20 29.6 29.6 20.5 20.5 20.5 21.2 21 21 20.9

Max 29 30.7 29.9 30.5 29.8 30.1 30.5 29.9 30.3 29.5

EC (µS cm-1 )

M 60.6 73.5 68.7 92.7 70.2 66.6 63.8 118.2 112.8 62.9

SD 8.82 5.68 4.47 29.18 15.1 13.18 11.03 43.38 41.76 9.98

Min 46.1 69.4 65.5 47.7 48.9 46.4 46.4 48.6 46.1 46.2

Max 71.9 77.5 71.8 134.2 101.8 98.2 78 169 163.6 80.3

pH

M 6.8 7.1 7.2 7.2 6.9 7 6.6 6.7 7.8 7

SD 0.37 0.02 0.13 0.57 0.38 0.37 0.19 0.28 0.93 0.43

Min 6.1 7.1 7.1 6.1 6.2 7.8 6.4 6.2 6.1

Max 7.3 7.2 7.3 8 7.4 7.4 7 7.2 9.5 7.5

DCO (mg l-1 )

M 9 19.4 25.2 28.1 22.1 22.3 57.1 129.9 35.2 13.8

SD 3.13 7.42 11.31 13.94 11.19 11.32 61.58 96.18 31.82 5.74

Min 5 14.2 17.2 7.2 6.7 7.8 7.8 6.7 5.2 4.2

Max 16.2 24.7 33.2 52.2 49.7 52.5 248.3 363.3 129.7 22.8

NH4 (mg l-1 )

M 0.3 1.1 0.7 0.9 0.6 0.6 0.6 0.7 0.4 0.5

SD 0.43 0.78 0.9 0.64 0.43 0.42 0.37 0.51 0.43 0.53

Min 0.1 0.6 0.1 0.3 0.3 0.3 0.3 0.2 0.1 0.1

Max 1.7 1.7 1.4 2.9 2 2 1.7 2.2 1.8 2.2

NO3 (mg l-1 )

M 0.3 0.4 0.4 0.4 0.4 0.4 0.3 0.3 0.3 0.3

SD 0.15 0.23 0.34 0.17 0.15 0.16 0.17 0.15 0.15 0.14

Min 0.1 0.2 0.1 0.1 0.1 0.1 0.1 0 0.1 0.1

Max 0.6 0.5 0.6 0.6 0.6 0.6 0.6 0.5 0.5 0.5

NO2 (mg l-1 )

M 0.002 0.001 0.002 0.005 0.004 0.002 0.002 0.002 0.002 0.002

SD 0.001 0.001 0.001 0.002 0.002 0 0.001 0.001 0 0

Min 0.001 0.001 0.001 0.001 0.001 0.002 0.001 0.001 0.002 0.002

Max 0.003 0.002 0.002 0.007 0.011 0.003 0.003 0.003 0.003 0.003

PO4 (mg l-1 )

M 0.1 0.1 0.1 0.2 0.2 0.2 0.2 0.2 0.1 0.1

SD 0.04 0.06 0.02 0.09 0.09 0.08 0.05 0.07 0.03 0.03

Min 0.1 0.1 0.1 0.1 0 0.1 0.1 0.1 0.1 0

Max 0.2 0.2 0.2 0.4 0.4 0.3 0.3 0.4 0.2 0.2

Ptot (mg l-1 )

M 0.2 0.3 0.3 0.4 0.2 0.2 0.6 0.5 0.4 0.2

SD 0.13 0.08 0.21 0.39 0.12 0.22 0.84 0.51 0.49 0.22

Min 0 0.2 0.1 0.1 0.1 0.1 0.1 0 0 0

Max 0.6 0.3 0.4 1.7 0.6 0.9 3.2 2.1 1.9 0.9

O2 (mg l-1 )

M 7.4 6.8 7.1 7.2 6.9 7 6.6 5.6 6.5 7

SD 0.29 0.26 0.17 1.09 0.6 0.37 0.72 0.49 0.64 0.51

Min 7.1 6.7 6.9 3.8 5.3 6.2 4.7 4.1 4.7 6.1

Max 8.2 7 7.2 8.2 7.6 7.8 7.5 6 7.4 7.9

Pr (cm)

M 90.4 78.5 83 189 182.4 136.2 131.1 110.3 89.4 83.6

SD 37.16 21.58 22.33 63.73 74.95 56.28 47.48 50.43 49.32 42.13

Min 41.8 63.3 67.2 70.3 57.2 38.8 42.2 34 33.2 33.3

Max 174 93.8 98.8 309.6 298.8 234.4 205.5 212.7 174.1 160.5

Table 3: Physicochemical characteristics of water in the sampling stations of the River Niger near N iamey N = 7(T=temperature; EC=conductivity; DCO=oxygen chemical demand; Ptot=total phosphorus; O2 =dissolve oxygen;Pr=depth; M=mean; SD=standard deviation)


B. ALHOU

The sampling stations TON, UAM and SA were negativelycorrelated to axis 1 while BRA, SLA, BH, ENI, TAN, MEand HNN were positively correlated to this axis.

A clear separation of taxa is illustrated in Fig. 2B. The firstbiological canonical variable was positively correlated withChironomus gr. plumosus and Glyptotendipes sp. andnegatively correlated with Cryptochironomus sp. 2,Ablasbesmyia sp., Dicrotendipes sp., Tanytarsini sp.,Chironomini sp. 1 and Polypedilum spp. 1 (Fig. 2B).

DISCUSSION

In this study, we notified a taxonomic compositiondifference between stations located upstream anddownstream the wastewater discharges due to a clearpresence of species like Cryptochironomus sp.,Chironomini sp. 1, Polypedilum spp.1, Micropelopiinaesp., Nilodosus sp. and Tanytarsini sp. in the upstreamstation TON, but mainly absent in the downstream stationSA. Moreover, Chironomus genus is not present in TON,but present in SA. Furthermore, the structure of the communities ofChironomidae shows weak similarity between sewagedischarge stations. Low taxonomic diversity wasobserved in some stations located at the wastewaterdischarges, i.e. BH, TAN, SLA, BRA and ENI (Fig. 3).

Mineralization of organic compounds of wastewater linkedto oxygen consumption by bacteria, especially at thebottom, could explain the lower taxonomic diversityobserved in these stations and the absence of some taxa.Moreover, the sampling stations, especially those in frontof Niamey, were characterized by quite different substrates(BH, TAN, ME, SLA, BRA and ENI). According to severalstudies (Henriques-Oliveira et al. 2003; Beisel et al. 1998),the types of substrates play an important role in thedistribution of most aquatic insects, and particularly ofchironomids. Unstable substrates can explain poorabundance of taxa (Olive et al. 1988). The sediments atthe sewage stations in River Niger at Niamey appear to bevery unstable, as was observed during sampling. The dominance of Chironomus gr. plumosus at BH, SLA,BRA, ENI and at TAN indicated that it was the onlychironomid taxon found that could be an importantindicator to explain the anthropogenic impact onchironomids communities in those stations. Chironomusspp. is well adapted to low oxygen concentrations: theirhaemolymph possesses haemoglobin and the 7th and 8thabdominal segment possesses respiration tubuli. Severalstudies correlated the increase in the abundance of larvaeof the Chironomus genus in aquatic ecosystem to organicenrichment and its consequence on water quality (Callistoet al. 2002; Marques et al. 1999; Petrucio and Furtado1998). Other species are only found at the upstreamunpolluted station TON, e.g. the species Xenochironomus

Figure 3: Diversity index



97

sp., Nilodosis sp. and Cryptochironomus sp. seem to beprimarily related to the sandy substrate with presence ofblocks of stones, a habitat with a mostly favourable oxygencondition at the water-sediment limit. In any case, 16chironomid taxa at TON were by far the highest numberfound, confirming the better habitat conditions at thatupstream station. The Tanytarsini and Cryptochironomini sp. in this studyare the only taxa present in both no sewage stations TONand SA, but completely absent in the sewage stations.Tanytarsini are known to be more sensitive to pollution(in lake typology, a distinction is made between eutrophicChironomus lakes and oligotrophic Tanytarsus lakes).Perhaps their presence at SA may be interpreted as anindication of the first signs of a fast recovery of the originalNiger fauna, once the sewage discharge of Niamey is over.This seems to be confirmed by the drastic drop of thedominance of the pollution indicator Chironomus gr.plumosus at SA station. The few larvae found may beemanating from drifting. A special case is Dicrotendipes sp., which is negativelycorrelated to chemical oxygen demand, nitrites andorthophosphates. Nevertheless, this species has neverbeen found either in TON or in SA. Maybe this speciesrequires organic matter in combination with good oxygenconditions. In any case, it had a restricted distribution,only present in two stations, HNN and UAM. Artificialsubstrates which are well-exposed to the water currentmay be expected to have higher oxygen concentrationsthan in the upper sediment. The wastewater discharge ofNiamey on the water quality of River Niger indicates adrastic lowering of the chironomid species diversity. Somespecies appear to be more or less tolerant to pollution.There are indications that the water quality rapidlyrecovers once the Niamey discharge is over.

ACKNOWLEDGEMENT

The authors thank Pierre Dumont for his help in preparinglarvae and in taking pictures of all taxa identified, andDodo Abdelkader for his advice. Ousmane MamaneSalissou and all participants in the manuscript-writingwork at Cotonou (Benin) organized by InternationalFoundation for Science (June 2007) are appreciated fortheir contribution correcting the English grammar. Thisresearch was supported by the Belgium TechnicalCooperation, the International Foundation for Science,Stockholm, Sweden and Organisation of Islamic

Conference Standing Committee on Scientific andTechnological Cooperation (COMSTECH), Islamabad,Pakistan.

REFERENCES

Angelier E. 2000. Ecologie des eaux courantes. Edit.Technique et Documentation, 199 p.

Armitage P., Cranston P. S. and Pinder L. C. V. 1995. TheChironomidae. The biology and ecology of non-bitingmidges. In: �Ustao lu M. R., Balik S. and

Beisel J. N., Usseglio-Polatera P., Thomas S. andMoreteau J. S. 1998. Stream community structure inrelation to spatial variation: the influence ofmicrohabitat characteristics. Hydrobiologia 389: 73-88.

Botts P. S. 1997. Spatial pattern, patch dynamics andsuccessional change: chironomid assemblages in aLake Erie coastal wetland. Freshwater Biologie 37: 277-286.

Callisto, M., Moreno, P., Gonçalves, J. F. Jr., Leal, J. J. F.& Esteves F. A. 2002. Diversity and biomass ofChironomidae (Diptera) larvae in an impacted coastallagoon in Rio De Janeiro, Brazil. Brazilian Journal ofBiology 62: 77-84.

Dejoux C. 1984. Contribution à la connaissance desChironomides de l’Afrique de l’Ouest (Diptères-Nématocères). 3e note. Revue Hydrobiologie Tropicale17: 65-76.

Durand, J.R. & Lévêque, C. 1981. Flore et fauneaquatiques de l’Afrique sahelo-soudanienne. TomeII. ORSTOM. Paris, 847 p.

Eggermont H et Verschuren D. 2004a. Sub-fossilChironomidae from East Africa. 1. Tanypodinae andOrthocladiinae. Journal of Paleolimnology 32: 383-412.

Eggermont, H & Verschuren, D. 2003. Sub-fossilChironomidae from Lake Tanganyika, East Africa. 1.Tanypodinae and Orthocladiinae. Journal ofPaleolimnology 29: 31-48.

Eggermont, H & Verschuren, D. 2004b. Sub-fossilChironomidae from East Africa. 2. Chironominae(Chironomini and Tanytarsini). Journal ofPaleolimnology 32: 413-455.

Eggermont, H, Verschuren, D. & Dumont, H. 2005.Taxonomic diversity and biogeography ofChironomidae (Insecta: Diptera) in lakes of tropicalWest Africa using subfossil remains extracted fromsurface sediments. Journal of Biogeography 32: 1063-1083.


B. ALHOU

Eggermont, H. 2004. Fossil Chironomidae (Insecta, Diptera)as biological indicators for past salinity variation inAfrican Lakes. Taxonomy, quantitative inferencemodels, and assessment of model performance inspace and time. Ph D, Univ. Gent, 501 p.

Evrard, M. 1996. Utilisation des exuvies nymphales deChironomidae (Diptera) en tant qu’indicateursbiologiques de la qualité des eaux de surfacewallonnes. Thèse de Doctorat, FUNDP, PressesUniversitaires de Namur (Belgique), 205 p.

Fend S. V. and Carter J. L. 1995. The relationship of habitatcharacteristics to the distribution of Chironomidae(Diptera) as measured by pulpa exuviae collections ina large river system. Journal of Freshwater Ecology10: 343-359.

Ferrington JR. L. C. 2008. Global diversity of non-bitingmidges (Chironomidae; Insecta-Diptera) in freshwater.Hydrobiologia 595: 447-455.

�Frouz J., Mat na J. and Ali A. 2003. Survival strategiesof chironomids (Diptera: Chironomidae) living intemporary habitats: a review. European Journal ofEntomology 100: 459-465.

Gonçalves J. F. Jr., Esteves F. A. et Callisto M. 2000.Succession and diversity of Chironomidae in detritusof Typha domingensis in a coastal lagoon (ParqueNacional da Restinga de Jurubratiba, State of Rio deJaneiro, Brazil). Verh. Internat. Verein. Limnology 27:2374-2377.

Henriques-Oliveira A. L., Dorvillé L. F. M. and NessimianJ. L. 2003. Distribution of Chironomidae larvae fauna(Insecta: Diptera) on different substrates in a streamat Floresta da Tijuca, RJ, Brazil. Acta LimnologyBrasilia 15: 69-84.

Legendre, L. & Legendre, P. 1979. Ecologie numérique 1: le traitement multiple des données écologiques ;Masson, Paris, New York, Barcelone, Milan, 171 p.

Legendre, P. & Legendre, L. 1998. Numerical Ecology.Second English edition. ELSEVIER Science B.V.,Amsterdam, 853 p.

Marques M. M. G. S. M., Barbosa F. A. R. and Callisto M.1999. Distribution and abundance of Chironomidae(Diptera, Insecta) in an impacted watershed in South-East Brazil. Revue of Brasilia Biology 59: 553-561.

Miserendino M. L. 2001. Macroinvertebrates assemblagesin Andean Patagonian rivers and streams:environmental relationships. Hydrobiologia 444: 147-158.

Miserendino M. L. and Pizzolon L. A. 2003. Distributionof macroinvertebrates assemblages in the Azul-Quemquemtreu river basin, Patagonia, Argentina. NewZealand Journal of Marine and Freshwater Research37: 525-539.

Moller Pillot, H. K.M. 1984. De larven Der NederlandseChironomidae (Diptera). Inleiding, Tanypodinae andChironomini 1A. Derde druk, 278 p.

Ndaruga A. M., Ndiritu G. G., Gichuki N. N. and WamichaW. N. 2004. Impact of water quality onmacroinvertebrate assemblages along a tropical streamin Kenya. African Journal of Ecology 42: 208-216.

Olive J. H., Jackson J. L., Bass J., Holland L. and SaviskyT. 1988. Benthic macroinvertebrates as indexes ofwater quality in the Upper Cuyahoga River. OhioJournal of Sciences 88: 91-98.

Petrucio, M. M. & Furtado, A. L. S. 1998. Concentraçõesde nitrogênio e fòsforo da coluna d’àgua da lagoaImboassica. In: Callisto M., Moreno P., Gonçalves J.F. Jr., Leal J. J. F. and Esteves F. A. 2002. Diversity andbiomass of Chironomidae (Diptera) larvae in animpacted coastal lagoon in Rio De Janeiro, Brazil.Brazilian Journal of Biology 62: 77-84.

Rosenberg, D. M. & Resh, V. H. 1993. Freshwaterbiomonitoring and benthic macroinvertebrates. In:Eggermont H, Verschuren D. et Dumont H. 2005.Taxonomic diversity and biogeography ofChironomidae (Insecta: Diptera) in lakes of tropicalWest Africa using subfossil remains extracted fromsurface sediments. Journal of Biogeography 32: 1063-1083.

Shannon C. E. and Weaver W. 1963. The mathematicaltheory of communication. University Illinois Press,Urbana. In: Beisel J. - N., Usseglio-Polatera P.,Bachmann V. and Moreteau J.-C. 2003. A comparativeanalysis of evenness index sensitivity. Internat. RevueHydrobiologie 88: 3-15.

Sharley D. J., Pettigrove V. and Parsons Y. M. 2004.Molecular identification of Chironomus spp. (Diptera)for biomonitoring of aquatic ecosystems. AustralianJournal of Entomology 43: 359-365.

Smock L. A. and Stoneburner D. L. 1980. The response ofmacroinvertebrates to aquatic macrophytedecomposition. Oikos 35: 397-403.

Spellerberg I. F. and Fedor P. J., 2003. A tribute to ClaudeShannon (1916-2001) and a plea for more rigorous useof species richness, species diversity and the‘Shannon-Wiener ’ index. Global Ecology &Biogeography 12: 177-179.

Tate C. M. and Heiny S. J. 1995. The ordination of benthicinvertebrate communities in the South Platte Basin inrelation to environmental factors. Freshwater Biology33: 439-454.

Ter Braak C. J. F. and Smilauer P. 1999. CANOCO forWindows (version 4.02)- a FORTRAN program forcanonical community ordination. Centre for biometryWageningen. Wageningen. The Netherlands.



99

Usseglio-Polatera P. and Beisel J.-N. 2002. Longitudinalchanges in macroinvertebrate assemblages in theMeuse river: anthropogenic effects versus naturalchange. River Resources Application 18: 197-211.

Vårdal H., Bjørlo A. and Sæther O. A. 2002. AfrotropicalPolypedilum subgenus Tripodura, with a review ofthe subgenus (Diptera: Chironomidae). ZoologicaScripta 31: 331-402.

Wiederholm, T. 1983. Chironomidae of the Holarcticregion. Keys and diagnoses. Part 1. Larvae.Entomologica Scandinavica. Supplement, 457 p.

Woodcock T., Longcore J., Mcauley D., Mingo T., ReidBennatti C. and Stromborg K. 2005. The role of pH instructuring communities of marine wetlandmacrophytes and chironomid larvae (Diptera).Wetlands 25: 306-316.

Younes-Baraille, Y., Garcia, X.-F. & Gagneur, J. 2005.Impact of the longitudinal and seasonal changes ofthe water quality on the benthic macroinvertebrateassemblages of the Andorran Streams. Compte RenduBiologies 328: 963-976.


Journal Africain de Science et de Technologie

Volume 12 Number 1October 2012

ISBN No. 1607-9949.

For subscription and frurther information contact:

Pour tout renseignement complémentaire s’adresser au:

ANSTI/RAIST SecretariatUNESCO Regional Office in Nairobi

P.O. Box 30952 - 00100 Nairobi, KenyaTelephone: +254 20 7622619/20

E-mail:[email protected]

The articles appearing in this Journal express the views of their authors and not necessarily those of UNESCO (ANSTI)

The African Network of Scientific and Technological Institutions (ANSTI).

Réseau Africain d’Institutions Scientifiques et Technologiques (RAIST).

Science & Engineering Series

Documents

African Journal of Science and Technology (AJST) Science ... Vol 12 No 1.pdf · range for systems involving non polar solutes on polar stationary phases (Heberger et al. 2002). There