Discriminant Function Analysis of Miocene Volcaniclastic Rocks from North-Western Croatia Based on Geochemical Data

In the north-western part of Croatia, belonging to the South Pannonian Basin, numerous small occurrences of Miocene volcanic and volcaniclastic rocks have been discovered (©IMUNI∆ et al., 1981; ANI»I∆ & JURI©A, 1985; PAMI∆, 1997 and references therein). According to the comprehensive work of PAMI∆ (1997), Neogene volcanic rocks could be divided into four groups: Egerian Eggenburgian, Karpatian, Badenian and post-Badenian.


INTRODUCTION
In the north-western part of Croatia, belonging to the South Pannonian Basin, numerous small occurrences of Miocene volcanic and volcaniclastic rocks have been discovered (©IMUNI∆ et al., 1981;ANI»I∆ & JU-RI©A, 1985;PAMI∆, 1997 and references therein).According to the comprehensive work of PAMI∆ (1997), Neogene volcanic rocks could be divided into four groups: Egerian -Eggenburgian, Karpatian, Badenian and post-Badenian.
The age of volcanic and volcaniclastic rocks could be determined by radiometric methods or on the basis of field relationships with associated sediments.Radiometric methods are still not easily available, while scarce and poor exposures, due to the vegetation cover and tectonic deformation, often make it difficult to establish the correlation between sediments and volcanic rocks.The aim of this work (which forms a part

Discriminant Function Analysis of Miocene Volcaniclastic Rocks from
North-Western Croatia Based on Geochemical Data Darko TIBLJA© 1 , Vanja LOPARI∆ 1 and Mirko BELAK 2 of a detailed investigation of volcaniclastic rocks in the area) was to see if any significant difference in geochemical characteristics of the rocks of different ages exists, i.e. to determine the geochemical fingerprint of rocks of different ages.The theoretical background for the analysis lies in the fact that the composition of volcanic rocks depends on the geotectonic settings which changed during the evolution of Pannonian basin, and therefore should be recorded in the chemical composition of the rocks.Positive results could help distinguish rocks of unknown age.
For this purpose we used statistical methods, principally the discriminant function analysis, which is a powerful technique for classifying individual cases (samples) into previously defined groups on the basis of multiple variables, and one which has a broad application in determining solutions for geoscience problematics (DAVIS, 1986).

MATERIALS AND METHODS
In the scope of our research we investigated Miocene volcanic and volcaniclastic rocks, as well as rocks for which they were the origin, e.g.bentonites, that crop out in north-western Croatia, along the easternmost segment of Periadriatic fault system, from the Slovenian border to the Varaaedinske Toplice.We analysed andesites from Cerovec (sample T-6-94B) and Lepoglava, dacites from TrliËno (T-5-93-17) and Donje Jesenje (T-6-93-17E), bentonites from Poljanska Luka and ©aπa, and tuffs (sensu lato) from the areas of Hromec, Jesenje -Cerje Jesensko, Lapornic a -©eprun, Jamno, Vuglovec, Podrute, Moaeenec, and Kalnik Mountain (Fig. 1).In addition to these newly collected samples we analysed samples of rocks from Moaeenec, described by KI©PATI∆ (1909) MI∆, 1993).The only published data on volcanic rocks from Moaeenec area are from KI©PATI∆ (1909), who presumed the same age for these rocks as for those from the Jesenje area.The tuffs in the Kalnik area, are, according to the unpublished fieldtrip guidebook3 , of Badenian age.Investigated samples are listed in the lower part of Table 3 on the basis of sampling locality, successively from the west to the east.
The chemical composition of the rocks was determined on an ARL 8410 X-ray fluorescence (XRF) wavelength dispersive spectrometer equipped with an Rh-tube.Glass beads, for the determination of major elements, were prepared by melting a 1:1:4 mixture of water-free sample, Li 2 B 4 O 7 and LiBO 2 .Measured intensities were corrected for matrix effect using the empirical model of LACHANCE & TRAILL (1966).H 2 O - was determined as loss of weight after 4 hr of drying at 110°C, while loss on ignition at 1025°C was taken as H 2 O + .Trace elements were analysed on pressed powder pellets.Details on the measuring conditions are given in TIBLJA© (1996).Results of the chemical analyses can be obtained from the authors on request.
The computer programme STATISTICA (STAT-SOFT, 1995) was used to carry out the discriminant function analysis of the collected data.Additional input data (element concentrations and geological age of the rocks) for the analysis were taken from PAMI∆ (1997).It was necessary to take logarithmic values of element concentrations, except in the cases of gallium and thorium, to achieve a normal distribution of the input data, which is one of the assumptions of discriminant func-tion analysis.All statistically analysed samples are listed in Table 3.The first 25 samples are from PAMI∆ (1997).They are marked with symbols of the type E7 where letter designates age, i.e.E is for Egerian -Eggenburgian, while the numbers correspond to the sample numbers in the tables, in his book, that contain chemical compositions of the rocks of the age concerned.Due to the fact that, according to their major elements, as well as immobile trace element content, the investigated volcaniclastic rocks are mainly acid and rarely neutral (TIBLJA© et al., 2000) only data for the rocks that contain >55wt.% SiO 2 were considered.Rocks of known age were assigned to three age groups ( E g e r i an -Eggenburgian, Karpatian and Badenian) and we tried to statistically distinguish these groups on the basis of rock geochemical characteristics, and finally, to assign each newly analysed sample of unknown age to one of these predefined groups.Due to the scarce available age data for the volcaniclastic rocks from ©aπa, Lapornica, ©eprun, Moaeenec and Kalnik areas their age was taken as unknown.The reason why we considered only three, instead of the four age groups described by PAMI∆ (1997), was that post-Badenian rocks are basic, while, as already mentioned, investigated volcaniclastic rocks are acid to neutral.

RESULTS
Standardised coefficients for two discriminant functions obtained by discriminant analysis of Miocene volcaniclastic rocks, based on their trace element content, are given in Table 1 together with their eigenvalues and relative and cumulative proportion of variance extracted by each function.The plot of discriminant scores given in Fig. 2 differentiates between rocks of three different ages.Function 1 explains 80.8 % variation of the data, with a high negative loading of Ce and positive of Cu where x 1 , x 2 , x 3 , ..., x m are concentrations, or their logarithms, of elements in the sample, are given in Table 2.
On the basis of classification scores S i each sample was  included in the age group for which the score was highest (Table 3).The age attributed to the samples on the basis of references mentioned in the previous section is given in the second column of the table.The following three columns contains ages ascribed on the basis of discriminant analysis, ordered into the first, second, and third choice according to the posterior classification probabilities.These classification functions could be used for the classification of samples of unknown age if their chemical composition is known.Table 4 gives a comparison between the actual and predicted number of samples in each age group.According to these results 96.5 % of samples have been classified correctly to their pre-defined groups.Misclassified samples are due to overlap in the discriminant space of Badenian samples with the Karpatian ones on one side, and Egerian -E g g e n b u r g i a n on the other (Fig. 2).

DISCUSSION
The results of performed discriminant function analysis, presented in Tables 3 & 4 and Fig. 2, showed that Miocene volcanic rocks of different ages (Egerian -E g g e nburgian, Karpatian and Badenian) could be confidently distinguished on the basis of their geochemical data.Almost all (55) of the 57 samples of known age, that were used as input data for the analysis, were classified correctly.This analysis showed that Karpatian samples could be successfully differentiated from Egerian -Eggenburgian ones.A slight overlap of Badenian samples with Karpatian and Egerian -Eggenburgian ones is in agreement with the conclusion of PAMI∆ (1997) that E g e r i an -Eggenburgian and Badenian rocks have similar geochemical characteristics.But in spite of that their discrimination could be performed quite successfully.
The high negative loading of Ce on discriminant function 1, that successfully discriminates the Karpatian samples is in accordance with the elevated Ce content of Karpatian volcanics (PAMI∆, 1997).PAMI∆ (op.cit.) also noticed that Karpatian rocks have higher contents of Zr, Nb and La, and lower Sc and Y than other Tertiary rocks, but this is not obvious from standardised coefficients listed in Table 1.These coefficients indicate a higher Cu content in the Egerian -E g g e n b u r g i a n samples than for the other two groups.Badenian samples are discriminated by function 2, in which Cr and V carry the main discriminatory potential.
As shown in Table 3, most of the volcaniclastic rocks from the Moaeenec and Kalnik areas, of previously unknown age, are of Egerian -Eggenburgian age according to discriminant function analysis based on geochemical data.The only exception is the sample from the Kalnik area (VHK 500) which is Karpatian.Such results for the samples from the Moaeenec area support the conclusions of KI©PATI∆ (1909), based on petrographic investigations, about the contemporaneous age of the rocks from the Moaeenec and Jesenje areas.Equivocal results for volcaniclastic rocks from Kalnik could indicate volcanic activity in both Egerian -Eggenburgian and Karpatian times, but such results are probably a consequence of problems encountered during investigations of the chemical composition of any volcaniclastic rock.The original chemical composition of volcaniclastic rocks is a consequence of magma composition, which depends mostly on geotectonic environment, and could be strongly modified by: (1) separation during transport of pyroclastic material from the volcanic vent to place of deposition, (2) alteration and (3) admixing of epiclastic or any other detrital material.Because of (3) we excluded from the analysis all of the samples that were an obvious mixture of volcanic and other detrital material.A Karpatian age obtained for the sample VHK 500 from the Kalnik area is probably the result of separation during transport.This is a sample of crystal tuff (sensu lato), and probably, its modal and, consequently, chemical compositions were strongly modified by processes of separation.The obtained chemical composition is thus probably nonrepresentative for the magma and the result for this sample should be treated cautiously.The Egerian -Eggenburgian age for samples from Kalnik is not in accordance with available data from ©IMUNI∆ (1981) 3 .We also encountered the problem of modified chemical composition due to alteration, or rather, due to the product of alteration of volcanic glass.All of the samples with authigenic zeolites (clinoptilolite and/or mordenite) have relatively elevated contents of barium and strontium.These higher contents are the result of zeolite selectivity for these elements.Therefore we have to exclude these elements from the discriminant model, and the conclusion of PAMI∆ (1997) that Rb/Ba and Rb/Sr contents could differentiate Egerian -E g g e n b u rgian and Badenian rocks could not be confirmed.
It is obvious that, for distinguishing volcaniclastic rocks of different age on the basis of geochemistry, it is necessary to know their chemical composition but also their mineral and petrographic composition.
The statistical analysis was carried out on a limited number of samples and therefore their results should be taken cautiously and should be tested by further investigation.In addition, the available chemical analyses were occasionally incomplete.Also, comparison of the literature and our own data, mainly because of different experimental methods was problematical and should be acknowledged.

CONCLUSIONS
The discriminant function analysis of geochemical data, which was carried out on a relatively limited number of samples of Miocene volcanic rocks from the Sava-Drava interfluve, enabled the following preliminary conclusions to be made.

Sample (locality)
Age according to Age according to discriminant function analysis literature data Probability 1 2 3  1) It is possible to obtain an indication about the age of Miocene volcanic and volcaniclastic rocks on the basis of their chemical composition.
2) Most of the investigated volcaniclastic rocks from Moaeenec and Kalnik areas, whose age was unknown, are, according to discriminant function analysis of Egerian -Eggenburgian age.
It must be stressed that such conclusions are only preliminary and should be confirmed or refuted by further investigations on more samples.

Fig. 1
Fig. 1 Location map of investigated area.Circles represent sampling locations of volcaniclastic rocks.