The multiple-cored DEEP site has six parallel and overlapping boreholes (A to F), of which holes A and E only cover the uppermost few metres of the sediment succession. Each of the deeper holes was drilled with a diameter of 149 mm and water-based mud was used to clean the holes of cuttings and to stabilize the side walls during the coring process. Hole C was logged immediately after drilling down to 470 m b.l.f., and amongst other probes, such as a resistivity and borehole televiewer, a spectral gamma ray (SGR) and sonic probe were used. To prevent the unconsolidated sediments from caving in, the SGR probe was run through the drill pipe and a continuous record of the sediments down to 470 m b.l.f. was achieved. The SGR data were acquired using the SGR 70-slimhole tool of the Leibniz Institute for Applied Geophysics (LIAG), which records the total gamma radiation (GR), as well as the spectral components (potassium, K; thorium, Th; and uranium, U) and their contribution to the GR. The tool was run at a logging speed of 3 m min-1 and a sampling rate of 10 cm. The achievable minimum bed resolution is controlled by the size of the bismuth germanate crystal (5 × 15 cm) and the characteristics of the target formation such as the absolute value range and contrast of values between neighbouring beds (Theys, 1991). The vertical resolution is estimated to 15–20 cm. The sonic tool, which measures the seismic velocity (Vp), was applied afterwards at a speed of 4 m min-1 and a depth increment of 10 cm. To allow open-hole measurements by sonic logging, the drill pipes were successively pulled upwards until open hole sections of ca. 50 m were accessible. The uppermost 30 m b.l.f. could not be logged by the sonic tool, because some drill pipes were kept in hole to allow other probes to enter. The measuring principles are described by Rider and Kennedy (2011) and the tools are specified in Buecker et al. (2000) and Barrett et al. (2000). The data were acquired, preprocessed, and processed with the software GeoBase^® (Antares, Germany) and WellCAD^® (Advanced Logging Technology, Luxembourg).

Sedimentary cycles in lake records can be studied by cyclostratigraphic methods for a potentially orbitally driven origin (Lenz et al., 2011; Prokopenko et al., 2001; Weedon, 2003). To investigate wavelengths and amplitude of the contained signals, fast Fourier transform (Weedon, 2003) can be applied. The sliding window method (windowed Fourier transform) (Baumgarten and Wonik, 2015; Molinie and Ogg, 1990b; Torrence and Compo, 1998; Weedon, 2003) can be applied to identify the distribution of cycles within a series and their evolution over the data set: the spectral analysis is calculated for a depth interval of specific length (window size) and the resulting spectrum is allocated to the centre of the window. Subsequently, the window is moved downwards by a certain step size and the analysis is repeated at consecutive depth positions until the window border reaches the end of the data set. The results are presented in a three-dimensional spectrogram with colour coding of the relative power of the different frequency components. Generally, a small window size is favourable to maximize the length of the resulting plot. However, the contained cycle needs to be covered and cannot be determined if the window size was chosen too short, e.g. only half the signal's wavelength. The optimal window size is determined by empirical testing. Spectral analysis for identification of the characteristic periodicities (Jenkins and Watts, 1969; Priestley, 1981) was performed on normalized SGR data using fast Fourier transform within MATLAB (MathWorks^®).

The detection of cycles by SGR logging is limited by the Nyquist frequency, which is twice the sampling rate (Molinie and Ogg, 1990a). The temporal resolution can be estimated by the vertical resolution of the applied tools and the averaged sedimentation rate. For a mean sedimentation rate of, for example, 38 cm kyr-1, cycles in the range of 0.8 to 1.1 kyr are resolvable by SGR logging.

In this study, downhole logging data, in conjunction with stratigraphically aligned tephra layers in the cores, is used to construct an age–depth model. Therefore, matching of core and logging depth is required. To provide age control by distinct tephra layers, the layers need to be identified in the cores by visual description or by their physical properties, such as susceptibility from core logging, in contrast to the background sediments. Artifacts in the coring process, such as incorrect depth allocation of coring tools or gas extension of sediments due to pressure release, produce erroneous depth. Furthermore, depth shifts between core and logging depth are generated because the downhole data originate from one hole (C; down to 470 m b.l.f.) and the core composite record (see Francke et al.; 2015) is composed of four different holes which are tens of metres apart. The depth of a distinct sediment layer may differ up to 3 to 4 m between these holes. The matching of borehole logging data and sediment core is described in detail in Francke et al. (2015) and based on a correlation of K contents from SGR with K intensities from X-ray fluorescence (XRF) scanning, and using magnetic susceptibility from downhole logging and multi-sensor core logging (MSCL) on sediment cores. Trends and patterns were compared and matched; the larger features were preferred over correlation of small-scale features in the data. Cross correlation was used to prevent systematic depth shifts of these data sets and for quality control.

To perform cyclostratigraphic studies and to estimate an age–depth relationship and sedimentation rates, compaction and associated reduction of sediment thickness due to overburden pressure must be considered. The original (decompacted) thickness of the sediments can be calculated if the initial (surface) porosity and the compaction coefficient (Brunet, 1998) can be determined. The amount of porosity decrease with greater depth depends on sediment properties, such as grain size and sorting (Serra and Serra, 2003), and can be expressed as Φ(z)=Φ0⋅e(-cz), where the porosity (Φ) at a specific depth (z) is to be estimated; Φ0 is the initial porosity and c is the compaction coefficient (Athy, 1930; Brunet, 1998).

Porosity can be measured directly on the sediment cores, e.g. by Archimedean weighing. However, the physical properties, in particular from (unconsolidated) sediment cores, are typically disturbed due to drilling, release of pressure, and core handling. Therefore, measurements by downhole logging are more suitable; in situ porosity can be gained by neutron porosity logging or derived, e.g. from bulk density. These tools operate with radioactive methods and the import procedure into foreign countries is usually extremely complicated and seldom successful. Therefore, the radioactive tools from the LIAG could not be used at Lake Ohrid. However, porosity was derived by an empirical relationship from sonic data (Vp) (Erickson and Jarrard, 1998), which were recorded continuously from below 30 m b.l.f. The software 2DMove^® (Midland Valley Exploration Ltd.) was used to decompact the sediments and calculate the original sediment thickness.

The output curves from SGR were compared to estimate the contribution of the spectral components to the total gamma ray. GR is mainly controlled by K and Th, which develop uniformly (R > 0.9). GR and K were used for further investigations. K was chosen over Th, because it is also available from XRF core scanning and the interpretation can be reviewed easily.

The downhole logging data (GR and K) from 0 and 240 m b.l.f. was compared to the global benthic isotope stack LR04 (Lisiecki and Raymo, 2005). In order to select an appropriate temporal window, we considered the current age estimates from stratigraphical aligned tephra layers (eight age–depth points; Table 1). After the anchor points from tephra deposits were defined, significant variations in the data were correlated; a very similar cyclicity with a positive correlation between GR and K from downhole logging data and δ18O data was observed (Fig. 2a). The onset and terminations of several marine isotope stages can be easily distinguished in the downhole logging data, whereas warm and/or humid periods (decreased δ18O values) correlate with low GR values. Thirty additional tie points (Table 1) were set due to the strong similarities between the curves characteristics. After matching these tie points, high correlation of both data sets (R= 0.75) was observed. The tephra ages and tie points from correlation between GR, K, and LR04 were used to assign a (preliminary) timescale to the data (Fig. 2b). Within the data, the timescale between the tie points was generated by linear interpolation. According to the established age model, the 240 m long sediment succession covers a time period between 630 ka (including MIS 15) and the present.

The conspicuous similarity of the data sets allows calculation of a synthetic δ18O curve from GR by a regression of both data sets. Therefore, normally distributed data are required and thus the prominent tephra layers at 20 and 68 m b.l.f. (Fig. 2c) were considered to be outliers and removed. Best results were achieved by a linear solution (R2= 0.60) as follows: δ18Ocalc(‰)=3.19+0.03⋅GR.(gAPI) Whereas cycles and trends are similar in both data sets, the amplitudes between the LR04 stack and the synthetic δ18O (derived from GR; δ18Ocalc; Fig. 2c) do not completely match. The δ18Ocalc values from 630 to 430 ka are lower than compared to LR04. From 430 to 185 ka the amplitude of δ18Ocalc is higher and during the past 185 kyr, δ18Ocalc is mostly decreased compared to LR04. These three zones (A to C) are indicated in Fig. 2c.

After visual comparison of GR and K with LR04 documenting a strong correlation of periods with low GR and K with warm and/or humid periods, the application of spectral analysis by sliding window analysis objectively identifies the possible cycles and their temporal distribution. The spectral analysis was calculated with a window of 90 m length and a step size of 1 m. Thus, the stepwise calculation for the depth section from 0 to 240 m b.l.f. and the resulting three-dimensional spectral plot (Fig. 3a) is composed of 150 spectra. The plot ranges from 45 to 240 m b.l.f., because the first spectrum is allocated to the window centre (Sect. 3.2) and therefore half of the window length is not displayed. Two prominent spectral peaks are evident in the data set, as indicated by colour with wavelengths of 30 and 45 m. The distribution of the cycles is non-uniform along the data set and a break in the spectral characteristics occurs at about 110 m b.l.f. Based on the reduced relative power of the 30 m signal and the subsequent increased power of the 45 m frequency at 110 m b.l.f., the plot can be split into a lower interval I and an upper interval II (Fig. 3a). In addition, two single spectra from depths of 170 and 50 m b.l.f. (Fig. 3b, c) are displayed.

The similar cyclicity in the LR04 stack and GR (Fig. 2a, b) suggests that the 100 kyr cycle, known as dominant periodicity in sedimentary archives for the past ca. 900 kyr and clearly documented in the δ18O data, has the strongest effect on the cyclic characteristics of the GR data. The highest amplitudes were therefore linked to the 100 kyr cycle. Averaged sedimentation rates can be calculated using this link (45 m ≡ 100 kyr cycle) for 110 to 0 m b.l.f. as follows: 45m/100kyr=45cmkyr-1.

Furthermore, the time of deposition can be estimated by using this sedimentation rate and the length of the interval II (length of 110 m): 110m/45cmkyr-1=244ka.

The sedimentation rate for interval I (length of 130 m) can be calculated as 30 cm kyr-1 (30 m ≡ 100 kyr cycle). Thus, for interval I the duration of deposition of 433 ka is calculated, which gives an overall time of deposition (sum of interval I and II) of 677 ka.

The sedimentation rates from sliding window analysis show a distinct shift from 30 to 45 cm kyr-1 at 110 m b.l.f. (Fig. 4). However, the sedimentation rates from visual tying to LR04 are more variable and range from 22 to 71 cm kyr-1. Exceptionally high sedimentation rates occurred during MIS 6 and lowest sedimentation rates occurred during MIS 11 and 13. The sedimentation rates from LR04 tie points were averaged over the length of the intervals and show mean values of 35 cm kyr-1 for interval I and 48 cm kyr-1 for interval II (Fig. 4).

The effect of decreased sediment thicknesses due to compaction over time was determined to investigate its impact on the estimates of sedimentation rates. The Vp data from sonic logging were used to derive porosity after Erickson and Jarrard (1998). The porosity values were averaged for layers of 100 m thickness (Fig. 5a). The initial porosity (Φ0) was determined at 80 % and the compaction coefficient was estimated at 0.39 km-1. These parameters were used as input data for modelling decompaction of the sediments, and the 2-D model was calculated for layers of 50 m thickness. The modelling process starts with the removal of the top layer and subtraction of its overburden pressure. The new thicknesses of the lowermost layers are thereafter calculated and these steps were repeated again downwards. The resulting thicknesses of the sediment layers after decompaction show a quasi-linear increase with greater depth for these small depths (Fig. 5b). Decompaction of the sediment layers ranges from 10 to 30 % downwards and the cumulative thickness of the sediment sequence (present thickness of 250 m) is increased by 35 m (to 285 m).

The GR data were stretched (Fig. 5c) and subsequently spectrally analysed by sliding window method (Fig. 5d) to determine the effect of decompaction on the spectral analysis. The spectral analysis shows two spectral peaks (Fig. 5d; wavelength of 36 m from 276 to 118 m and wavelength of 48 m from 118 to 0 m). Linking these spectral peaks to the 100 kyr cycle provided sedimentation rates of 36 cm kyr-1 for interval I and 48 cm kyr-1 for interval II. Times of deposition remain constant and are 433 ka for interval I and 244 ka for interval II.

The strong correlation of GR and K with LR04 (low GR and K during warm and/or humid periods, high in cold and/or drier periods) suggests a response of the sedimentary system to the temperature and hydrological changes related to the global ice-volume fluctuations during the glacial–interglacial cycles. The fluctuations are likely to be controlled by the input and deposition of clastics (K and Th sources), which are suggested to have increased during the past ca. 136 kyr, when glacial conditions prevailed (except MIS 1 and 5e) at Lake Ohrid (Vogel et al., 2010a). In particular, the reduced input of organic matter and calcium carbonate during cold and/or drier periods seems to amplify the enhanced input of clastic material. During warm and/or humid periods, carbonate production and preservation is increased (Vogel et al., 2010a). In combination with higher organic matter flux, the clastic content of the sediments is reduced and the GR and K data are lower. Either the total content of clastics is lower during warm-humid periods or the amount is decreased relative to carbonate and organic matter (diluted). However, as discussed by Vogel et al. (2010a), less vegetation cover during cold and drier periods is likely and also suggests increased erosion in the catchment and subsequent higher input of clastic material. Based on our interpretation, these cyclic changes from carbonate-rich to clastic-rich sedimentation were constant at least for the past 630 kyr.

Three zones were observed, based on amplitude differences between the synthetic δ18O curve from GR (δ18Ocalc) and the referenced record: A, 630 to 430 ka (MIS 15 to MIS 12); B, 430 to 185 ka (MIS 11 to MIS 7); C, 185 to 0 ka (MIS 6 to MIS 1). Comparison of the synthetic δ18Ocalc to LR04 revealed systematically lower values in zones A and C and higher values in zone B.

We suggest the following reasons of the GR signature:

Climate-dependent supply and deposition of K-rich clastics (feldspars, clays). They are increased during cold, dry periods (high GR) and decreased during warm, wet periods (high carbonate deposition; low GR) in the sediments.

Therefore, the GR is most likely controlled by several factors and cannot be described in total by linear solution. However, the main component seems to be a linear response of K-rich sediments to the global climate trend (1). Further, we consider the attempt useful to compare both data sets and to show the very unusual similarity in cyclic characteristic between LR04 and GR.

Matching of GR and K data with the global climate reference record (LR04) equates the timing of the climate dynamics recorded in the oceans to Lake Ohrid. This means that if the downhole logging data are tied to this timescale, differences between the response times of Lake Ohrid and the global climate trend (e.g. lead or lag effects of the onset of terminations) are lost. However, even if the comparable small system was likely to be subject to a faster climate response compared to records from the marine realm, correlation with the global signal and the resulting age–depth model is verified in large parts (down to 206 m b.l.f.) by tephrochronology.

The relative power of the spectral peaks from sliding window analysis seems to be higher in interval I (wavelength of 30 m) in comparison to interval II (wavelength of 45 m). The strength of the signal depends on the number of cycles that are detected by spectral analysis. The 45 m long cycle can be contained up to 2.4 times in the 110 m long interval II, whereas the cycle of wavelength of 30 m might be recorded more frequently (4.3 times) in the 130 m long interval I. This can contribute to the higher intensity of the 30 m amplitude and thus we cannot interpret a stronger cyclicity for the lower part.

Based on sliding window analysis, the sedimentation rates are constant (30 cm kyr-1) for a long period of time (433 kyr), apart from a shift to increased rates at about 110 m b.l.f. to 45 cm kyr-1. However, small-scale variations cannot be resolved due to averaging and the window of 90 m length used. More variable and realistic results are indicated by tying of GR and K to LR04. The lowest rates (minimum of 22 cm kyr-1) occur during MIS 13, 11, and 9. Strongly increased sedimentation rates (up to 71 cm kyr-1) occur during MIS 8 and in particular during MIS 6. Therefore, decreased sedimentation rates during interglacials and increased rates during glacial periods can be derived. Even if these fluctuations do not correlate with all MIS, they seem to be largely coupled to glacial–interglacial dynamics. This trend suggests higher accumulation of clastic-rich glacial deposits compared to calcium carbonate-rich deposition during interglacials. The overall trend from lower mean rates in the bottom part (> 110 m b.l.f., sliding window; > 130 m b.l.f.; LR04 tie points) to increased values towards the top is comparable.

The cumulative time of deposition based on these sedimentation rates and corresponding sediment thicknesses range from 630 ka (LR04 tie points) to 677 ka (sliding window method) and are overall in a similar range (Fig. 6). We consider the estimated time of deposition from the tuning to LR04 to be of higher accuracy compared to the averaged estimates from sliding window analysis. The agreement with age estimates from cores, based on wiggle matching of (bio)geochemistry data, such as XRF scanning data, to LR04 and local insolation patterns (Francke et al., 2015), supports the age–depth model. Nonetheless, the sliding window method is useful to estimate averaged sedimentation rates and to provide complementary cyclic characteristics of the data. Whereas age control from tephrochronology is only available down to 206 m b.l.f., our interpretation was extended down to 240 m b.l.f. based on the constant cyclic characteristics.

The required initial porosity for modelling of the effect of compaction was derived from sonic logging (Φ0= 80 %) and is commonly lower (at ca. 65 %) for sedimentary basins. Studies of physical properties of marine sediment cores have shown comparable values for seafloor deposits (Φ= 80 %, Vp = 1540 m s-1; Kim and Kim, 2001) and even higher porosity values (> 80 %; Kominz et al., 2011) for unconsolidated sediments. The compaction coefficient c was estimated at 0.39 km-1 and lies in the range between c used for sands (0.20 km-1) and carbonates (0.50 km-1) and thus is considered as reasonable. The overall trend of a quasi-linear evolution of decompaction with increasing depth was observed, although Eq. (1) describes an exponential curve, it can be approximated to a line over short lengths (< 500 m). This is in accordance with the average linear increase in Vp with greater depth. The tephra layers, with thicknesses of only a few centimetres, are considered to have little impact on the compaction of these sediments.

Spectral analysis of GR data on decompacted depth scale (Fig. 5c) revealed a very similar spectral characteristic compared to the sliding window plot of the compacted data. However, the two spectral peaks (wavelengths of 30 and 45 m) are shifted to higher wavelengths and increased to wavelengths of 36 m for interval I and 48 m for interval II. Therefore, the sedimentation rates are increased accordingly to 36 cm kyr-1 and would be underestimated by 7 to 20 %, if they are not corrected for the effect of compaction. Underestimation of sedimentation rates from sliding window analysis is supported by comparison of results from LR04 tie points; the latter indicate generally higher sedimentation rates (35 cm kyr-1, interval I; 48 cm kyr-1, interval II) and show very good agreement with the estimates of the effect of compaction.

To determine the time of deposition, two input quantities were used: (1) sedimentation rates and (2) the thickness of the sediment layer for which they apply. Due to the stretching of the data set, the resulting sedimentation rates and the length of the intervals are increased and thus the duration of deposition remains unchanged. Therefore, the effect of compaction on this calculation is negligible.