1. INTRODUCTION Regional geochemical mapping makes use of various sample media which offer the general prospect either of finding a particular type of mineral deposit or, more commonly, of locating the presence of certain anthro- pogenic contaminants. Opinions on the particular advantages and drawbacks concerning different types of sample materials have been amply described (e.g. REIMANN, 1988; OTTESEN et al., 1989; MACKLIN et al., 1994; EDÉN & BJÖRKLUND, 1994; PULKKI- NEN & RISANEN, 1997; HUISMAN et al., 1997; SWENNEN & SLUYS, 1998; SWENNEN et al., 1998, and others). Yet, owing to the fact that all of these are not readily available in different parts of the globe, the general consensus has not been established hitherto, which is why the regional and local schemes of geo- chemical survey in various countries, and sometimes even in the same country, differ accordingly. In Croatia, regional geochemical mapping is based upon soil as the primary sample medium. This is main- ly due to the fact that soil is the only sample material Geochemical Comparison of Stream and Overbank Sediments: A Case Study from the Æumberak Region, Croatia Zoran PEH and Slobodan MIKO available throughout the country, particularly when the southern, almost entirely carbonate terrains are consid- ered. In the northern, Pannonian region, owing to the expansive drainage system, other sample materials, e.g. stream or overbank sediments, can be found in abun- dance. As a consequence, the question may arise whether stream sediment, or soil, or, perhaps some oth- er type of regolith material, would be the most appro- priate for sampling in order to detect the greatest varia- tion among geochemical data. This is particularly of interest in areas where two distinctly different litholo- gies, such as carbonate and non-carbonate sedimentary rocks, associate in a complex way (as in the Æumberak region). It should be emphasized that soil sampling, when compared to the sampling of stream or overbank sediments, is based on an altogether different philoso- phy. The first approach is, as it were punctuated, with no strictly defined conjoined area of underlying bedrock influence, regardless of the chosen sample design. The second approach is area-related which increases the possibility of reflecting the surface lithol- ogy of the whole area upstream from the sampling site. Besides, when observing the drainage basin as a funda- mental geomorphological unit (CHORLEY, 1969), it is certain that less bias would be inferred if one can deem it to be an area originated and embedded in the sur- rounding landscape by a set of natural processes per- taining not only to the strictly geomorphological but also to the geochemical domain. This is why many authors, especially from the northern, carbonate-free part of Europe, but also elsewhere in the world, applied, as a rule, the catchment basin analysis in their geo- chemical research (BONHAM-CARTER et al., 1987; CARRANZA & HALLE, 1997; ÓDOR et al., 1997, and others). Thus, the sampling of appropriate media in a catchment basin is rendered desirable wherever the landscape allows it, because their geochemical assem- blage may elucidate the average composition of this unit area more faithfully than any of the soil sampling designs. This may be of particular value in the case of geochemical reconnaissance both in the regions sus- pected of mineralization or for possible pollution. When reconnaissance geochemical surveys are car- ried out in low- to medium-order drainage basins, where both stream and overbank sediments are obtain- able for sampling, it may not be obvious as to which of Geologia Croatica 54/1 119 - 130 6 Figs. 4 Tabs. ZAGREB 2001 Key words: Geochemical mapping, Drainage basin approach, Stream sediment, Overbank sediment, Discriminant analysis, Æumberak, Croatia. Institute of Geology, Sachsova 2, HR-10000 Zagreb, Croatia. Abstract Geochemical comparison between the stream and overbank sediments from low- to medium-order drainage basins is grounded on the pre- supposed statistical contradistinction of their locality-paired sample correlatives. Discriminant analysis differentiated the overbank from stream material mainly on account of higher content of most of the analyzed elements in the former vis-à-vis an otherwise common geo- chemical semblance. Only the carbonate material seems to be deplet- ed in overbank sediment samples. Investigations also demonstrated that in the relatively non-contaminated area it may be more difficult to verify the supposed purity-contamination reciprocity between the investigated media, since the recent and prehistoric materials were not contrasted as regards their non-lithogenic components. 120 Geologia Croatica 54/1 these media, or perhaps both, should be sampled in order to acquire the best geochemical information from the study area. Granted that processes having control over the mass movement within the catchment basin are common in both cases, one may suppose a subsequent similitude in their overall geochemical assemblage, with only a small variation in the data. Nevertheless, for practical consideration a time dimension should always be taken into account, especially in validating the anthropogenic versus natural contribution to the chemi- cal content of alluvial materials. This is due to the fact that stream sediment represents an active material of recent origin which is temporarily suspended on a stream bed, while overbank sediment indicates alluvial regolith of earlier depositional cycle(s) produced as a result of extensive floods (OTTESEN et al., 1989). On this premise, a tacit assumption is made about the main difference in the geochemical composition between the two types of alluvial sediment. Namely, overbank sedi- ment (except in its uppermost section) should represent an unpolluted, pristine medium when deposited in a natural, pre-industrial environment, while, on the other hand, stream sediment is expected to reflect every kind of recent contamination that may arise in the investigat- ed area. Furthermore, geochemical variation resulting from potential mineralization within a catchment basin is supposedly detectable in both. Therefore, the imme- diate scope of the present study will be to explore the geochemical difference between these two sample media in an area with a well-known overall geological setting and expected, but minor, anthropogenic influ- ence. It will hopefully shed some light on the possible advantages in utilization of one or, perhaps, both of these sample media in further low- to medium-density geochemical mapping of some target areas in Croatia. 2. DESCRIPTION OF THE STUDY AREA Æumberak is a mountainous territory located to the west and in the immediate vicinity of the Croatian capital of Zagreb. To the north and west it is bordered by Slove- nia, and to the south by the Kupa river, while its east and southwest portions gently dip towards the Karlovac depression (Fig. 1). The landscape combines the fea- tures of Dinaric, highly dissected, carbonate terrains abounding in various karstic phenomena such as sink- holes, together with Pannonian, mostly non-carbonate terrains of moderate relief and with a regular drainage network. The highest point in the area is the summit of Sveta Gera (1,160 m), while the surrounding valleys of the Sava, Krka and Kupa rivers with their terraces do not exceed altitudes of about 200 m. 2.1. Geological setting The study area is geologically mapped at 1:100,000 scale and presented mostly on the sheet of Zagreb (©IKI∆ et al., 1978). Only a small portion falls within the sheets of »rnomelj (BUKOVAC et al., 1983) and Novo Mesto (PLENI»AR et al., 1976). Geotectonical- ly, it belongs to the broad boundary zone between the Dinaric carbonate platform (Dinaricum) spreading to the southwest, and adjacent Inner Dinaride area (Supra- dinaricum) lying on the northeast (HERAK & BUKO- VAC, 1988; HERAK, 1991). The geotectonic character Fig. 1 Simplified geological map showing location of the study area (after BUKOVAC et al., 1983; ©IKI∆ et al., 1978; PLE- NI»AR et al., 1976). Legend: Q2) Holocene in general; Q1) Pleistocene in general; PlQ) Plio-Quaternary: unconsolidat- ed sediments; Ng) Neogene in general: clastic rocks; K2) Upper Cretaceous: limesto- nes, dolomites and flysch; J) Jurassic in general: predomi- nantly limestones; T3) Upper Triassic: predominantly dolo- mites; T1) Lower Triassic: pre- dominantly clastic rocks; P2 , 3) Middle and Upper Permian: predominantly clastic rocks. is well reflected in the intricate tectonic and, particular- ly, lithological patterns. The oldest rocks in the area are mostly clastic sedi- mentary rocks of Middle and Upper Permian age. These consist predominantly of sandstones, more rarely of conglomerates, shales and siltites. Apart from the clas- tic rocks, limestones and dolomites occur sporadically. The Early Triassic rocks are prevalently non-carbonate, mostly sandstones in their lower part with increasing portions of carbonate component toward the upper lev- els. Sedimentation was continuous until the Middle and Upper Triassic in predominantly carbonate facies. The Upper Triassic dolomites are the most significant litho- logical member of the series. Jurassic sedimentary rocks are almost entirely carbonate with limestones as a dominant member, while the overlying discordant Upper Cretaceous rocks consist of a thick flysch-type series which was deposited during the Cenomanian- Senonian period. The latter include predominantly cal- careous and clayey marls and calcarenites which, together with the former, frame the margin of the Dinaric carbonate platform. Their contact with the sur- rounding Upper Triassic dolomites (Supradinaricum) is clearly tectonic, the nappe front frequently masked by vertical neotectonic faults (HERAK & BUKOVAC, 1988). Frequent tectonic activity with periodic changes of depositional environment during the Tertiary result- ed in greater diversity of the clastic sedimentary facies. From Palaeocene to Pliocene, a variety of clastic sedi- mentary rocks were formed, mostly sandstones and marls. Carbonate clastic rocks predominate only through the Badenian. The transition into the Quater- nary was marked by the onset of freshwater sedimenta- tion with an extensive and thick series of lithofacially differentiated sediments - from gravel to clays. Plio- Quaternary sediments occupy a considerable part of the study area, particularly its lowered southern rim border - ing the Karlovac depression. Quaternary deposits are represented almost entirely by the Holocene alluvial sediments of the local streams. 2.2. Mineral occurrences Mineralization in the investigated area is related mostly to the layers or veins hosted in the fine-grained clastic Permian rocks. Apart from a number of scattered Fe, Pb, Zn, Cu, Au and Hg occurrences together with gyp- sum and barite, there is a small-scale siderite-haema- tite-sulphide ore deposit in the valley of Rudarska Gradna. Until the middle of the last century it had been mined extensively for iron for a few hundred years. The iron ore typically occurs in the form of a siderite layer of submarine sedimentary origin interstratified between the Palaeozoic sandstone beds (©INKOVEC, 1971; ©IFTAR, 1989). The main ore body is accompanied by haematite lenses as well as with sulphide veins contain- ing chalcopyrite and a barite-galena paragenesis. The latter occur invariably in the underlying sandstone stra- ta, while the overlying sandstone series contains in its uppermost parts the thick (45 m) gypsum-anhydrite bed (©INKOVEC, 1971) which marks the border with the Triassic. Small mineral occurrences of the same origin are widely disseminated through the Permian outcrops of the nearby valleys such as LipovaËka Gradna and Lud- viÊ, as well as OkiÊnica. A different type of mineraliza- tion can be found in the western part of the investigated area. It appears as small remnants of sedimentary limonite deposits covering the Middle and Upper Trias- sic dolomite palaeosurface. A few such occurrences are strewn over the area of the Slapnica valley, north of KraπiÊ. 2.3. Anthropogenic influence The area of investigation is free from the immediate impact of great industrial or other sources of contami- nation. The nearest industrial center, the capital city of Zagreb, is more than 20 kilometres away to the east. Small cities such as Samobor, scattered on the perime- ter of the investigated mountainous area, do not employ industries of great scale or pollution capacity. There are, however, two known sources of human influence that may be observed in the local catchment areas. One can be ascribed to the bygone mining activi- ties that may have left traces of increased concentra- tions of heavy metals such as Pb, Zn and Fe both in the stream and overbank sediments within some of the low- order catchments. The other can result from recent agri- cultural activity, generally viniculture, with anticipated increases in Cu (from bluestone) and P (from fertiliz- ers). The former is restricted to the inner, mountainous part of the study area, while the latter can be found scattered over the wider zone of the southern slopes of Samoborsko gorje, particularly in the surroundings of KraπiÊ and Jastrebarsko. 3. MATERIALS 3.1. Sampling More than forty low- to medium-order drainage basins ranging in size from 0.65 to 122.94 km2 were sampled, in a close-spaced sampling design covering the territory of approximately 600 km2 (Fig. 2). For a closer inspec- tion into the sources of geochemical variance between the stream and overbank sediments the paired samples of both media were regularly collected from the same sample site (within a few metres). This procedure nec- essarily excluded a number of smaller, dominantly mountainous drainage basins (mostly of the fourth order) from the analysis, where only stream samples were available for sampling. A total of 40 sites with both stream and overbank material have been sampled over the entire area (Table 1), a sampling density of approximately one sample per 15 square kilometers being thus defined. 121Peh & Miko: Geochemical Comparison of Stream and Overbank Sediments... 122 Geologia Croatica 54/1 The sample sites were selected at the basin outlets, sufficiently upstream from the confluence with higher or same order streams in order to avoid sampling the sediment that may result from mixing of material from the two channels during the flood flow. An active stream sediment, which represents the composite of the recently deposited bed load material, was collected from several spots (ordinarily 5-10 as recommended by SALMINEN et al., 1998) over a short channel stretch upstream of the selected site. Simultaneously, a single overbank sediment sample was taken from approxi- mately the same point at the exposed area of either bank of a channel. The latter is composite material tak- en from the bank section ranging in height from 25 cm beneath the surface down to the water level (usually 0.5 to 2 m thick), while the first 25 cm of upper, near sur- face, horizon was avoided because of possible anthro- pogenic disturbance and pedogenesis. In both cases a quantity of about 3 kg of sediment was collected to yield enough representative material for sieving and analysis. 3.2. Sample preparation Collected samples were air-dried (at <40°C) for approximately three months. After drying, the samples were disaggregated in a porcelain mortar, homoge- nized, and finally dry-sieved through stainless-steel screens to the fraction of <125 µm. This fraction was preferred because the highest concentration of most of elements, especially trace elements, occur in the fine- grained, usually from 63 to 125 µm size fraction (e.g. RHOADS & CAHILL, 1998). Also, different studies show that the <125 µm size fraction makes up more the 95% of the particles in most samples (SWENNEN et al., 1998). 3.3. Analytical methods Analytical work was performed at the ACME Analyti- cal Laboratories in Vancouver, Canada, where samples were subjected to multi-acid digestion ICP analysis, and geochemical Hg analysis by flameless AA. A total of 36 elements were thus analyzed with Au, Be, Bi, Mo, U and W invariably, and Ag, Sb, Sn and Cd mostly having concentrations below the detection limit. Ele- ments such as Th, Y, Nb and Sc were measured slightly above the threshold, so that all of these were omitted at the outset from further considerations. 4. STATISTICAL ANALYSIS 4.1. Univariate statistics and data transformation A set of 22 elements was used in statistical analysis. Eight major and 14 minor and trace elements were selected as predictor variables in the process of discrim- ination. Table 2 displays the summary statistics of the analytical data including the skewness coefficient as a measure of normality. Due to the fact that a number of variables in both groups show highly positively skewed Fig. 2 Sample map. 123Peh & Miko: Geochemical Comparison of Stream and Overbank Sediments... frequency distributions, transformation must have been carried out for most of the minor and trace elements such as Hg, Pb, Cu, Sr, Ba, As, Zn and Cr, but also for some major elements such as Ti and P. The process of conventional log- and ln-transformation was applied separately for each group, but in some cases (Ni for example) the results were poorer than original distribu- tions. In such instances the variables were left natural (the total data set can be requested from the authors). 4.2. Basic principles of discriminant analysis Geochemical variation between the two investigated sample media, already “known” to be geochemically separable, can be thoroughly investigated by the use of multivariate discriminant analysis. Thus, the two-group (K=2) problem is introduced which presents the sim- plest case with a solitary discriminant function as a basis for separation. In analysis these groups are labelled STREAM or OVERBANK, respectively. The discrimination procedure revolves generally around how to compute a linear combination of original (pre- dictor) variables that will best distinguish between the groups. This is achieved by both maximizing the ratio of between-group in comparison to within-group vari- ability and generating the smallest misclassification errors (DILLON & GOLDSTEIN, 1984; DAVIS, 1986; ROCK, 1988). The latter is also enhanced by limiting Case Sample Drainage basin Order Catchment Area (km 2) 1 41 LudviÊ 4 Sava 4.41 2 194 Orejovec 4 Kupa 5.25 3 196 Piroπki potok (SLO) 4 Krka 6.03 4 197 Skradnja (SLO) 4 Krka 5.07 5 214 FuËanski jarak 4 Sava 2.81 6 221 Velika draga 4 Sava 2.99 7 349 Jaπevnica 4 Kupa 12.02 8 351 Ponornica 351 4 Krka? 1.01 9 354 Ponornica 354 4 Kupa? 2.86 10 366 VorbaπËica 4 Kupa 4.56 11 371 potok 371 4 Kupa 0.65 12 19 ©krobotnik 5 Sava 8.79 13 21 Breganica 5 Sava 11.21 14 42 LipovaËka g. 5 Sava 26.00 15 43 Rudarska g. 5 Sava 15.48 16 178 Reka 5 Kupa 9.19 17 187 OkiÊnica 5 Kupa 19.11 18 190 Potok 190 5 Kupa 3.15 19 198 Suπica (SLO) 5 Krka 9.79 20 275 Bregana 5 Sava 14.52 21 276 Rakovac 5 Sava 8.39 22 328 ÆumberaËka reka 5 Kupa 15.96 23 331 Suπica 5 Krka 8.51 24 333 Suvaja 5 Kupa 22.91 25 334 Potok 5 Kupa 5.44 26 335 Svilnica 5 Kupa 3.63 27 337 Ponikva 5 Kupa 6.46 28 338 Slapnica 5 Kupa 16.25 29 339 Puπkarov jarak 5 Kupa 9.77 30 340 Brebrovac 5 Kupa 5.69 31 342 Stiska 5 Kupa 11.43 32 343 Malunja 5 Kupa 6.87 33 344 Gonjeva 5 Kupa 8.11 34 345 Kamenica 5 Kupa 17.23 35 346 Bukovica 5 Kupa 12.25 36 347 Slatinek 5 Kupa 4.33 37 368 Stiper 5 Kupa 5.72 38 370 Selna 5 Kupa 5.32 39 18 Bregana 6 Sava 57.40 40 72 KupËina 6 Kupa 122.94 Table 1 General data describing the sample sites. 124 Geologia Croatica 54/1 the size of both groups at an approximately equal num- ber of objects so that the optimal cutting score is placed exactly between their centroids. Graphically, the calcu- lated discriminant function portrays a new axis along which the groups are maximally set apart. Finally, according to variables with the most prominent discrim- inant loadings, geological meaning could be attached to the axis. Thus it can be labelled with regard to a specif- ic gechemical process which is liable for separation between the groups. Other measures for determining the individual contribution of predictor variables, such as the standardized coefficients, may be often fairly misleading as they are more subject to instability caused by intercorrelations between discriminatory variables (DILLON & GOLDSTEIN, 1984). 5. RESULTS The results of the two-group discriminant analysis are briefly summarized in Table 3. Owing to data standard- ization the optimal cutting score for the two groups of equal abundance (40) has a zero value, with the group centroids placed at equal distance from the cutting point along either side of the discriminant axis. On the value of the test statistics the difference in the separation of the STREAM and OVERBANK centroids can be judged as statistically significant, although a few sam- ples in both groups appear to be more loosely scattered about their means. Accuracy of the discrimination pro- cedure can be inspected from the classification matrix (Table 4). 69 samples out of total of 80 in the data set are correctly classified on the basis of their geochemi- cal composition, which makes 86 percent for the com- bined population of both groups. As can be seen from the Table 4, the unequal classification efficiency of the two groups shows that these are asymmetrically dis- criminated. Better results are achieved with the STREAM group having only three misclassified sam- ples (7.5 percent incorrect), while OVERBANK shows much more asymmetry with 20 percent of inaccurately classified samples. This asymmetry, however, does not diminish the efficacy of discrimination between the two sample media, particularly considering the problems of multivariate normality (ROCK, 1988). Despite the rec- ommended normalization procedures of observed data, some variables still tend to be distributed differently in both groups, which results in unequal dispersion of pre- dicted groups. When the significance of a particular subset of ele- ments in the general discrimination scheme is consid- ered, it is evident that in spite of the apparently bipolar nature of the discriminant function, the accent is heavi- ly placed on the positive pole (Fig. 4), where the bulk of the analyzed elements are loaded. The axis weighs V, Al, Fe, Zr and La against essentially a single element - Ca, which is obviously a reflection of the inverse rela- tionship between the aluminosilicate and carbonate component in the two sample media. Owing to their low discriminant loadings, other elements add little to STREAM OVERBANK Mean St.D. Skew. Mean St.D. Skew. Fe (%) 1.55 0.67 0.17 2.25 0.86 0.83 Ca (%) 9.56 4.90 -0.20 7.10 4.16 0.11 Mg (%) 3.28 2.50 0.45 2.67 2.12 0.89 Ti (%) 0.22 0.13 2.48 0.25 0.07 -0.17 Al (%) 2.94 1.20 0.00 4.18 1.22 0.19 Na (%) 0.36 0.15 0.72 0.41 0.11 0.51 K (%) 0.80 0.35 0.33 1.08 0.41 1.40 P (%) 0.05 0.02 1.76 0.04 0.01 0.89 Cu (ppm) 21.63 18.10 3.89 41.55 101.33 6.19 Pb (ppm) 42.43 149.09 6.29 25.7 20.11 3.41 Zn (ppm) 47.23 21.17 1.70 56.28 24.92 2.94 Ni (ppm) 27.93 17.08 1.20 40.23 19.78 1.25 Co (ppm) 9.25 5.14 0.60 13.00 5.81 0.30 Mn (ppm) 757.13 509.10 0.73 847.65 590.75 1.12 As (ppm) 8.08 3.75 1.77 9.76 7.24 2.95 Sr (ppm) 118.83 100.56 2.68 128.20 127.61 3.06 V (ppm) 49.25 18.37 0.32 70.10 19.26 0.15 La (ppm) 19.15 8.51 0.05 25.60 7.78 -0.38 Cr (ppm) 47.68 24.66 1.53 58.80 19.70 0.46 Ba (ppm) 195.03 118.46 2.63 337.93 589.20 6.09 Zr (ppm) 24.6 11.73 0.26 33.68 11.63 0.10 Hg (ppb) 62.63 139.00 6.10 363.48 1831.75 6.31 Table 2 Univariate statistics (me- an, standard deviation and skewness) of major, minor and trace elements in the stream and overbank sediment sam- ples of the Æumberak region. 125Peh & Miko: Geochemical Comparison of Stream and Overbank Sediments... the geochemical distinction between STREAM and OVERBANK. This is particularly evident with some major and minor elements such as Mg, Na, Ti, P, but also Mn and Sr. 6. DISCUSSION Comparison between the classification results repre- sented by the plot of all samples (Fig. 3) against the plot of variable contributions (Fig. 4) along the discrim- inant axis offers insight into the geochemical relation- ship between the groups. There is an immediate impact of stream and overbank sediments differing primarily in the relative abundance of a bulk of analyzed elements, as only Ca (and disregarding Mg and P) can be found increased in the stream samples. This can be interpreted in two ways: either as an indication of the predominant- ly carbonate character of the active stream sediment in the investigated catchment basins, or (not excluding the first), that material composing the overbank samples contains greater concentrations of elements with a high- er natural variability and considerably higher metal contents (particularly V, Al and Fe). The latter can be of interest in further investigations for possible anom- alies concerning mineralization or contamination in the study area. The main reason for the shift to the increase of Ca in the stream sediments can be sought from two sources. One is probably due to the dominant carbonate lithology in most of the sampled drainage basins - over two-thirds of their area is underlain by dolomite and limestone. However, having in mind that we are dealing here with the present-day stream bed material, it can- not, naturally, represent the whole catchment area but is limited only to a portion which is exposed to recent flu- vial erosion (OTTESEN et al., 1989). Most obviously, the central part of the Æumberak, undergoing vigorous tectonic uplift (PRELOGOVI∆, 1969), provides an ample source of carbonate material which is eroded from the bedrock in the deeply cut valleys (such as Bre- gana, for example) and feeds the sample sites at the basin outlets. Regional tectonic influence is of particu- lar importance because it represents a clear-cut example that a tacit assumption of equal erosion through the whole drainage basin may not be true (ROSE et al., 1979). The lower parts of the sampled drainage basins, especially in the southeast part of the studied area (the catchment of the Kupa river) are distinguished by the processes of aggradation or, at least, by the greatly reduced capacity for erosion due to the low channel gradients. Of no lesser significance is that their stream channels, particularly those of higher order (fifth and sixth), run mostly through non-carbonate (Neogene clastic) rocks, highlighting the contrast between “alien”, more carbonate, recent material in the stream bed and adjacent overbank. The other reason for the higher content of Ca in the stream samples as contrasted with its overbank counter- part is of a geochemical nature and still more empha- sizes their recentness. During the short depositional his- tory of the stream sediment, the friable Ca and Mg min- erals from dolomite and limestone were subjected most- ly to physical weathering which resulted primarily in the finer grain size of clastic particles downstream. Chemical weathering contributes little to the loss of carbonate material via running water and out of the sys- tem (which also includes the riverplain), especially in the Æumberak streams which are weakly alkaline (pH = 7-8). On the other hand, most of the dissolved material Number of variables in model (p) 22 Number of groups (K) 2 Number of functions (K-1) 1 Number of cases (n1, n2) 40, 40 approximate F ratio (degrees of freedom) 4.04 (22, 57) p-level 0.000008 Canonical R 0.78 Average R value for STREAM (centroid), R1 -1.23 Average R value for OVERBANK (centroid), R2 1.23 Mahalanobis distance D2 6.23 Table 3 Compositional differences between STREAM and OVERBANK sediments. PREDICTED GROUP OBSERVED STREAM OVERBANK % Total GROUP (p=0.5) (p=0.5) correct STREAM 37 3 92.50 40 OVERBANK 8 32 80.00 40 Total 45 35 86.25 80 Table 4 Classification matrix. 126 Geologia Croatica 54/1 in the streams comes from solution by groundwater (EASTERBROOK, 1969). This is the process, perhaps, that accounts for most of the lack of Ca and Mg in overbank sediment over the study area, namely, decal- cification of the previously deposited alluvium due to fluctuating hydrological conditions in the area. The dis- solution of carbonate minerals in the overbank sediment is induced by the raising of the groundwater level which creates periodic waterlogged conditions with a significant decrease in pH in the upper part of the over- bank profile (BERG & LOCH, 1998). In the two-group discrimination model it is reflected the other way round, that is, through the increasing carbonate compo- nent in the STREAM group. The second indication distinguishing the two sample media, as seen in Fig. 4, is related to the greater defi- ciency of analyzed elements in the stream sediment. Lower overall concentrations in this material have been generally observed earlier (e.g. REIMANN, 1987; SWENNEN et al., 1998), and can be accounted for the fact that much of the element contents (V, Al and Fe in particular in this case) are dispersed in the finer fraction (silt-clay) which is winnowed out leaving the stream sediment coarser and “depleted”. Apart from being accordingly attenuated, this material is also fairly non- homogeneous as can be seen from the asymetrically shaped histogram of discriminant scores (DS) for STREAM (Fig. 5a). In contradistinction to its stream counterpart, the overbank sediment appears to have a higher “natural” background, much on account of its longer depositional history which supplied it with a mixture of geochemical Fig. 3 Plot of discriminant scores (DS) (projec- tion of samples onto discriminant function line). Fig. 4 Plot of discriminant loadings (projection of variables onto discri- minant function line). 127Peh & Miko: Geochemical Comparison of Stream and Overbank Sediments... and mineralogical characteristics that embrace the whole drainage area upstream from the sample site. This is particularly important with respect to the distrib- ution of heavy metals and trace elements in both sample media (originating either from ore mineralization or man-made pollution) because, as can be seen from Fig. 4, there is no preferential distribution pattern. Apart from the relative abundance of non-carbonate components in the overbank sediment, a much greater geochemical homogeneity of precipitated material (ana- lyzed as a composite sample) can also be expected. This is clearly shown by the normally distributed dis- criminant scores for OVERBANK (Fig. 5b). The more uniform dispersion of chemical elements through over- bank samples affects the boundary between OVER- BANK and STREAM groups making it a little unilater- ally diffuse. As a result, more overbank samples with lower element concentrations are lost to the other group showing more affinity with the stream sediment. The cohesion of the STREAM group can be deemed consid- erable as only three samples bear more similarity with adjoiningly sampled overbank. This is obvious from closer inspection into the areal distribution of discrimi- nant scores for both groups (Figs. 6a and b). Excluding the eastern borders of the studied area, where the over- bank composition clearly conforms with entirely non- carbonate bedrock, it may be hard to locate an uninter- rupted section with a greater agglomeration of drainage basins containing highly discriminated overbank sam- ples (DS>1), despite attempting to relate them with the underlying bedrock. Conversely, the stream sediment characteristics are much more perceptible on the terrain as the pertinent basins are scattered almost evenly all over the study area (DS<-1) with only a minor cluster- ing along its southern and northern fringes: stream sedi - ment samples in the southwest are almost indistiguish- Fig. 5 a) Histogram of dis- criminant scores for ST- REAM; b) Histogram of discriminant scores for OVERBANK. a b 128 Geologia Croatica 54/1 b a Fig. 6 a) Map displaying the areal dispersion of discriminant scores for STREAM. Legend: 1) misclassified samples (DS>0); 2) weakly sepa- rated samples (-12). 129Peh & Miko: Geochemical Comparison of Stream and Overbank Sediments... able from the adjacent overbank (-1