Lake Oltina (44∘09′16′′ N, 27∘38′13′′ E, 7 m a.s.l.) is located on the floodplain of the Danube River, the lower Danube Plain, in southeastern Romania (Fig. 1). It is the largest fluvial lake on the Danube floodplain with a surface area of 33 km2 and is part of a natural reserve and a Natura 2000 site (ROSPA0056; Management Plan 2016). The lake has a major tributary, Canaraua Fetii, and is connected to two smaller lakes, Ceamurlia and Iortmac, located upstream as well as to the Danube River via an artificial dam built to prevent flooding (Telteu, 2014). The climate in the study region is wet warm-temperate continental (Köppen–Geiger class Dfa), also termed excessive, with the prevalence of harsh winters and hot summers (Posea, 2005), due to the influence of air masses from continental Asia. The mean annual temperature is ca. 11 ∘C, with a mean January temperature of ca. -1 ∘C and a mean summer temperature of 25 ∘C. Annual precipitation is about 400 mm (Adamclisi meteorological station). The geology of the catchment comprises limestone and loess deposits, whereas the soils are represented mainly by Haplic and Luvic Chernozem and Phaeozem (IUSS WRB 2006). Friable steep loess deposits are frequently eroded, delivering a clastic sedimentary influx into the lake (Romanescu et al., 2010).

The main potential vegetation types of the region include forest, forest steppe and calciphile steppe (Bohn at al., 2003). The main forest types are: thermophilous mixed deciduous broadleaf forests (subtypes G21, G22 and G34 according to Bohn et al., 2003), mesophytic deciduous broadleaf forests (subtypes F49 and F67), forest steppe (subtype L13) and steppe (subtype M5). Additionally, alluvial forests (subtypes U19 and U20), halophytic vegetation (P33) and tall-reed vegetation and sedge swamps (R1) prevail (Bohn et al., 2003). CORINE (Coordination of Information on the Environment) Land Cover data (2012) indicate that the present land cover within a 20 km radius of the lake comprises ca. 65 % arable land and orchards, 19 % steppe and semi-natural grassland, and 16 % deciduous forest (S1; Grindean et al., 2019). Important tree species in forests within this radius, and thus the most relevant for the pollen source area, are Quercus pubescens, Q. pedunculiflora, Q. robur, Q. cerris, Q. virgiliana, Carpinus orientalis, Acer tataricum, Tilia tomentosa, Fraxinus excelsior and Ulmus minor, whereas hygrophilous tree taxa, growing along the Danube River, are represented by Populus nigra, P. alba, Salix alba and S. fragilis. Shrubs occur abundantly either as understory vegetation or as thickets and are composed of Fraxinus ornus, Cotinus coggygria, Prunus mahaleb, P. spinosa, P. cerasifera, Rosa canina, Pyrus communis, Crataegus monogyna, Amorpha fruticosa, Gleditsia triacanthos, Elaeagnus angustifolia and Ailanthus altissima. Natural grassland steppic species are common on calcareous slopes and include Stipa stenophylla, S. ucrainica, S. capillata, Poa angustifolia, Festuca valesiaca, Bothriochloa ischaemum, Artemisia austriaca, Agrostis tenuis, Carex humilis, Centaurea orientalis, Astragalus ponticus and Thymus marschallianus (Sârbu et al., 2009; Grindean et al., 2019).

Sediment cores were extracted with a Livingstone piston corer (1 m long and 5 cm diameter) from the central part of the lake (1.8 m water depth) in spring 2016. The less consolidated sediment at the surface (36 cm) had previously been retrieved with a gravity corer in 2014. A lithostratigraphic description was made according to changes in texture, colour, magnetic susceptibility and the organic carbon content (loss on ignition; LOI).

Volume magnetic susceptibility (κvol) was measured with a Bartington Instruments Ltd MS2 meter and E sensor (Bartington Instruments, 2008). For LOI, samples were dried overnight at 105 ∘C and combusted for 5 h at 550 ∘C and then for 2 h at 900 ∘C, and LOI is expressed as percentage loss of the dry weight (Heiri et al., 2001). A composite sedimentary core record totalling 964 cm was constructed using the uppermost 10 cm from the gravity core and 10 overlapping Livingstone core sections. To determine the erosion and anoxic conditions in the lake, elemental geochemical concentration was measured on 190 sediment samples (extracted at ∼ 5 cm intervals along the core) and subsequently dried and homogenized using a non-destructive Niton XL3t 900 x-ray fluorescence analyser (FPXRF). An NCS DC73308 was employed as a certified reference material (CRM). Measurement follows the procedure described by Hutchinson et al. (2015). We selected the detrital element Zr as a proxy for erosion (Kylander et al., 2011) and employed the Fe : Mn ratio to reconstruct anoxic conditions in the lake (Nacher et al., 2013).

The chronology was established based on 18 AMS (accelerator mass spectrometry) 14C measurements (Table 1). Attempts to constrain the chronology of the top core via 210Pb and 137Cs gamma assay measurements failed to produce any meaningful results, probably due to surface sediment mixing. An attempt to establish a tephra-based chronology also failed due to a poor match to any known volcanic eruptions. The radiocarbon age estimates were converted into calendar years BP via the Bacon software (Blaauw and Christen, 2011) using the IntCal13 data set of Reimer et al. (2013). An age–depth curve was derived based on a smoothing spline model (Fig. 2; for details on age–depth construction, see Fig. A1 in the Appendix).

To determine past disturbance by fire, macroscopic charcoal particles were counted in samples of 2 cm3 extracted at 1 cm contiguous intervals. Samples were bleached, wet-sieved through a 150 µm mesh and identified under a stereomicroscope following the methodology described in Feurdean et al. (2017a). Here we report only the results for total macro-charcoal particles. We calculated the macro-charcoal accumulation rate (CHAR, particles cm-2 yr-1) by dividing the total macro-charcoal concentration by sediment deposition time (yr cm-1). To determine past grazing activity, coprophilous fungi (Sporormiella, Sordaria and Podospora) were tallied during routine pollen counting (van Geel et al., 1980; Baker et al., 2013). Percentages of coprophilous fungi were determined by adding their own sum to the total terrestrial pollen sum.

The lithology of the core at Lake Oltina showed little variability throughout the profile and comprises clay, gyttja clay and sandy clay. The age–depth model indicates a rather constant sediment accumulation rate with a mean of 5 yr cm-1 and no evidence of hiatuses (Figs. 2; A1). Organic matter (OM) as determined from LOI at 550 ∘C varied between 3 % and 10 % with slightly higher values between 5000 and 3500 cal yr BP and over the past 2000 years (Fig. 3). Our selected geochemical detrital element, Zr, a proxy for erosion, showed the lowest values between 5500 and 3500 cal yr BP (up to 150 ppm) and greater, highly fluctuating values at the beginning of the record and over the 2500 years (Fig. 3). The Fe:Mn ratio, a proxy for anoxic conditions in the lake, generally displayed large variability throughout the profile, although values were higher between 4000 and 3500 cal yr BP, around 3000 cal yr BP, and between 2000 and 1500 cal yr BP (Fig. 3).

The cluster analysis applied on the pollen record indicated three major periods of change in land cover and vegetation openness over the last 6000 years: open temperate and xerothermic deciduous broadleaf forest between 6000 and 4200 cal yr BP, the maximum extent of broadleaf tree cover between 4200 and 2500 cal yr BP, and the expansion of grassland between 2500 cal yr BP and the present (Fig. 4). Results from the REVEALS model suggest that landscape openness was ca. 10 %–15 % greater than the estimates derived from the raw pollen data. Overall, REVEALS indicate a greater proportion of Carpinus orientalis, Tilia, Acer, Rosaceae, Cerealia and Asteraceae and a lower proportion of Corylus avellana, Betula, Ulmus, Alnus, Fraxinus, Salix, Artemisia and Chenopodiaceae than the raw pollen data (Fig. 4). Quercus, Plantago lanceolata and Poaceae show a largely similar abundance in both the raw data and REVEALS estimates.

The (C27–C29) / (C31–C33) ratio in Lake Oltina varied between 0.63 and 5.99 and showed the greatest values (2.84) between 4200 and 2000 cal yr BP and the lowest values between 5500 and 4200 cal yr BP (1.76) as well as over the past 2000 years (1.83; Fig. 3). The ACL varied between 27 and 30 and shows an opposite pattern to the (C27–C29) / (C31–C33) ratio. Paq varied between 2 and 12 and showed the greatest values between 6000 and 5500, 4000 and 2500, and 1200 and 500 cal yr BP.

The pollen-based quantitative land cover reconstruction shows an average tree cover of 40 % between 6000 and 4200 cal yr BP and a tree cover maximum of 50 % between 4200 and 2500 cal yr BP in the surroundings of Lake Oltina (Figs. 4 and 5). In a pollen-based biome reconstruction, such a proportion of trees is likely to be indicative of a forest steppe or open-woodland type (Marinova et al., 2018). This woodland consists of tree taxa of xerothermic character including Quercus (likely Q. cerris and Q. pubescens) and Carpinus orientalis, along with temperate trees such as Carpinus betulus, Acer, Tilia, Ulmus and Fraxinus. Shrub (Rosaceae, Rosa canina, Sambucus, Prunus and Cornus), grass and forb communities were abundant and composed of a diverse mixture of mesophytic, xerothermic and halophilous taxa (Fig. A2). Coeval with the maximum extent in tree cover, the n-alkanes were dominated by shorter chain lengths (ACL) and a higher n(C27 + C29) / (C33 / C31) ratio, indicative of an increased contribution of tree-derived n-alkanes (Meyers, 2003; Aichner et al., 2010). However, the concentration of individual n-alkanes > C27 (not shown) varied with that of detrital element Zr. It was high between 6000 and 4200 cal yr BP and declined between 4200 and 2500 cal yr BP, suggesting a reduction in terrestrial plant delivery into the lake during the highest tree cover. The maximum extent of tree cover parallels a slight Fe : Mn ratio increase more evident between 4000–3500 and 3000–2500 cal yr BP. This may indicate the establishment of more anoxic conditions (Naeher et al., 2013), possibly associated with a higher lake level due to the intensification of Danube water and sediment discharge into the lake, a higher lake trophic status or less turbulent conditions. Increased anoxia linked to a higher lake trophic status and the decomposition of organic matter is supported by the slight increase of submerged aquatic macrophyte (Paq; Fig. 3). The presence of more aquatic plants in the lake at the time of the rise in the Fe : Mn ratio is also reflected in the increased abundance of aquatic and wetland taxa (Potamogeton, Myriophyllum and Typha/Sparganim; Fig. 4). On the other hand, low values of the lithogenic element Zr between 4200 and 2500 cal yr BP indicate more stable slope conditions with lower catchment run off, which might support the hypothesis that increased anoxia may have resulted from less turbulent conditions in the lake.

On a regional scale, the 6000–4200 cal yr BP interval was characterized by contrasting climate conditions north and south of 45∘ N latitude, due to differences in the dynamics of the storm tracks carrying moisture from the North Atlantic Ocean and Mediterranean Sea (Persoiu et al., 2017). Whilst southern Europe lake levels were low prior to 4500 cal yr BP, those in central Europe showed high stands during the same period (Magny et al., 2013). An opposite pattern is, however, visible after 4500 cal yr BP, when southern Europe lake levels increased, whereas those in central Europe declined. The lake level increase after 4500 cal yr BP also paralleled an intensification of fluvial activity in several rivers in southern Romania and the southern part of the Danube River at this time (Bozilova and Tonkov, 1998; Filipova-Marinova et al., 2007; Howard et al., 2004). Coastal lakes along the Black Sea were also at higher water levels at this time, whereas the Black Sea's level fluctuated strongly (Lamy et al., 2006). Lake Oltina is situated at 45∘ N and thus at the transition between these contrasting southern and northern Europe changes in hydro-climatic conditions. Our evidence of the maximum expansion of the tree cover and possibly an association with wetter conditions after 4500 cal yr BP may indicate the response of tree cover to increased moisture availability in the region (Fig. 3). A greater-than-present extent of tree cover at 6000 cal yr BP in response to higher precipitation has also been simulated for this region (Patriche et al., 2020).

Macro-charcoal-based reconstruction of biomass burning and thus disturbance by fire was low between 4200 and 2500 cal yr BP at the time of greater forest cover and wetter climate conditions (Fig. 3). This fire–climate–vegetation relationship is typical for a temperate environment but contrasts with the pattern found in environments with low vegetation productivity, where increased moisture tends to enhance vegetation productivity and therefore fuel availability (Pausas and Ribeiro 2013; Feurdean et al., 2020). Disturbance by herbivores, as inferred from the abundance of coprophilous spores, showed moderate values at the beginning of the record but declined during the interval of tree cover increase (Fig. 3). This may point to some impact by herbivores on the degree of forest openness, i.e. increased tree cover with a decline in grazing activity. However, given the large size of the study lake, the distance from the lakeshore to the coring point might have limited the transportation of these spores and have an impact on how representative their presence might be (Baker et al., 2013).

Tree cover dropped from 50 % at 2500 cal yr BP to ∼ 20 % at 2200 cal yr BP (Fig. 5). Tree species composition retained a xerothermic character, although Carpinus orientalis, typical of hot and dry climatic and soil conditions (Sikkema and Caudullo, 2016), declined most strongly from 40 % to 10 % (Fig. 4). The pollen-based reduction in tree cover was concurrent with a decreasing n(C27 + C29) / (C33 / C31) ratio, characteristic of an increased contribution of grasses, further suggesting a tree loss (Fig. 4). An average of 20 % tree cover at 2200 cal yr BP is close to the current tree cover in the surroundings of Lake Oltina (15 %), which suggests the opening up of the forest steppe to a similar extent as today. Anthropogenic conversion of broadleaf forests to agricultural land from 2500 cal yr BP onwards is suggested by the rise in the pollen of Cerealia (Secale cereale, Triticum, Zea and Hordeum), grassland (Poaceae), and pastoral and ruderal indicators (P. lanceolata, Urtica, Rumex, Chenopodiaceae, Artemisia, Asteraceae and Apiaceae; Figs. 4, 5 and A2). The levels of pollen of grazing indicators and nutrient-enriched soils (Plantago lanceolata, Urtica and Rumex), as well as coprophilous fungi, are particularly elevated after 2200 cal yr BP and may reflect more intensive animal husbandry. A higher representation of Cerealia in the REVEALS estimate (40 %) than the raw pollen record (5 %) is not surprising given the poor productivity and dispersal of most of the Cerealia pollen types. However, the magnitude of this difference is greater than generally reported in the literature. Nevertheless, the good match between the proportion of arable cover in the REVEALS and current land cover maps from CORINE for recent times suggest that our REVEALS reconstruction best reflects that of cultivated land. Uncertainties in pollen-based land cover reconstruction are common and connected to the availability and accuracy of the productivity estimates (PPEs), changes in cropland and grassland management, and the general assumptions of the REVEALS model (Sugita, 2017; Hellman et al., 2009; Trondman et al., 2015; Feurdean et al., 2017b). Our Cerealia pollen includes Triticum, Zea, Hordeum and Secale cereale, for which we have used productivity estimates derived from the calibration of local surface pollen samples with a vegetation inventory (Grindean et al., 2019). These productivity estimates are considerably lower than the average for Europe (0.22 vs. 1.85; Mazier et al., 2012) and are therefore the main cause of the high proportion of Cerealia cover reconstructed by REVEALS and the disparity between the outcome of this study and others elsewhere in Europe. Furthermore, the crop species included in Cerealia also vary regionally and with time, which may also introduce further variation when applying their PPEs for landscape reconstruction over an extended period of time. Lastly, the occurrence of wild-grass species with pollen that fall in the Cerealia pollen type (all Poaceae grains larger than 40 µm) may have led to an overestimation of the proportion of Cerealia at certain times in the past. Biomass burning increased concurrently with the spread of pasture and grassland communities, which may indicate the use of fire for land use management (Fig. 3). However, fire activity fell to its lowest level in the profile in the last 1000 years, probably due to a decline in biomass availability associated with intensive land use and landscape fragmentation typical of lowlands (Marlon et al., 2016; Feurdean et al., 2013, 2020).

The timing of forest loss coincides with the peak number of archaeological finds in the Iron Age but is pre-dated by the abundant archaeological finds of the Bronze Age, i.e. 4000 cal yr BP (Fig. 5). Interestingly, the typology of the houses in this area (i.e. small houses three-quarters buried in the ground with one-quarter above ground comprising mud brick walls, roofed with straw or reeds typical during the past 3000 years) also reflects the limited availability of timber for building (Ailincăi, 2009). Historical records show that, due to its smaller size, Carpinus orientalis was managed in coppice stands for household items and fuel production as firewood or charcoal (Goldstein et al., 1995; San-Miguel-Ayanz et al., 2016), and this may have also been the cause of its marked reduction from 2500 cal yr BP at Lake Oltina. Quercus decline on the other hand was more modest except for the last 1000 years. Historically, the aftermaths of the Slavic (first millennium CE) and Mongol (1224 CE) invasions could be a reason for a strong Quercus decline around this time (Matei, 1984; Epure, 2004). Oak was preferentially used for fortification in southwestern Romania (Sava et al., 2019; Gumnior et al., 2020). Nevertheless, tree cover and Quercus, in particular, recorded episodic rises with the strongest being visible between 1700 and 1300 cal yr BP. This reforestation phase appears to coincide with the post-Roman decline in rural settlement and the subsequent re-growth of secondary forests (Roberts et al., 2018). However, selective oak preservation, due to its economic significance, may have also led to this pattern of Quercus increases, as documented over large areas in Europe (Gardner et al., 2002; Feurdean et al., 2011; Jamrichová et al., 2017; Gumnior et al., 2020). Models of deforestation rates, using a scenario that accounts for population history and technological advances, suggest that the extent of deforestation in the lower Danube basin increased continuously from 4000 cal yr BP and rapidly doubled after 1000 CE (Kaplan et al., 2009; Giosan et al., 2012), thus much later than our pollen-based land cover reconstruction. Assuming no age correction for the hard-water effect would result in an age 1000 years older than that calculated by taking into account this dating limitation, which pushes the timing of major deforestation back to 3500 cal yr BP. This corroborates with the timing of deforestation found in Transylvania, central Romania, based on the REVEALS model (Feurdean et al., 2015) and with the onset of a local increase in archaeological finds of the Bronze Age (Fig. 5).

In-lake and catchment changes are also apparent in the Lake Oltina around the onset of forest loss. A slight increase in the Fe : Mn ratio, along with the sharp rise in Zr concentrations between 2500 and 1500 cal yr BP, may reflect soil erosion and detrital input into the lake associated with a diminished tree cover (Fig. 3), an inference also supported by an abrupt rise in charcoal and n-alkane > C27 concentrations. Intensification in fluvial activity in several rivers in southern Romania (Howard et al., 2004; Persoiu and Radoane, 2017), as well as on a wider scale in central Europe (Wirth et al., 2013), has been reconstructed after 3000 cal yr BP. The Danube's water and sediment discharge could therefore be another source of detrital and terrestrial plant delivery into the lake. In contrast, the Black Sea level lowered, which also suggests drier conditions in the eastern Mediterranean and over the Black Sea around the time of deforestation (Lamy et al., 2006), in line with the drier conditions in southwestern Romania found between 2000 and 1000 cal yr BP (Dragusin et al., 2014). Southern European lake levels remained high at the time of deforestation, while those from central Europe (Magny et al., 2011, 2013) and central Romania declined (Feurdean et al., 2013), though others in Romania increased (Magyari et al., 2009).

Our quantitative record of vegetation cover indicates a higher-than-present tree cover across the landscape of the eastern lower Danube Plain between 6000 and 4200 cal yr BP with an absolute maximum of 50 % (60 % raw pollen percentages) between 4200 and 2500 cal yr BP (Fig. 6). The composition and structure of Mid to Late Holocene vegetation of the eastern lower Danube Plain resembled, to a large degree, that of other European forest steppe areas, although particularities also exist. Whilst in central-eastern European forest steppe Quercus and Carpinus betulus were the dominant tree species, with a lower occurrence of Tilia, Ulmus, Corylus and Pinus (Magyari et al., 2010; Feurdean et al., 2015; Kuneš et al., 2015), on the Eastern European Plain Tilia and Quercus and, in some places, Pinus were the dominant tree taxa (Kremenetski et al., 1995; Novenko et al., 2016, 2018; Shumilovskikh et al., 2018, 2019). Forests in the Black Sea region also included thermophilus taxa, i.e. Quercus cerris and C. orientalis (this study; Marinova and Atanassova, 2006; Tonkov et al., 2014). Remarkably, Carpinus orientalis was found to be significantly more abundant around the Black Sea coast, i.e. Romania (20 %) and Bulgaria (> 5 %) with only scattered occurrences in Ukraine, whilst it was absent from central European forest steppe. Our pollen–vegetation calibration model shows that the adjusted abundance of C. orientalis is twice that recorded by its raw pollen percentages (Fig. 4).

Mid Holocene landscape openness near Lake Oltina (ca. 45 % raw pollen percentages) appears to fall in between that of the steppe in southeastern Bulgaria (60 %–80 %; Tonkov et al., 2014), the forest steppe of Ukraine (25 %; Kremenetski et al., 1995, 1999) and that of the Eastern European Plain (20 %–50 %; Shumilovskikh et al., 2018). However, landscape openness was greater than in other forest steppe sites from central-eastern Europe, i.e. Romania (Feurdean et al., 2015; Tantau et al., 2006), Hungary (Willis et al., 1997; Magyari et al., 2010), the Czech Republic and Slovakia (Pokorny et al., 2015; Hajkova et al., 2013; Kuneš et al., 2015), where it varied between 10 % and 35 % (Fig. 6). The composition of herbaceous plant cover included grasses (Poaceae) and a diversity of forbs thriving on a wide variety of habitats ranging from dry and saline soils (Artemisia, Chenopodiaceae, Asteraceae Compositae and Asteraceae Tubuliflorae) to dry and wet meadows (Filipendula, Galium, Anthemis, Aster, Caryophyllaceae, Euphorbia, Helianthemum, Hypericum, Fabaceae, Plantago lanceolata, P. major/P. media, Teucrium, Thalictrum and Verbascum; Figs. 4; A2). Notably, however, the proportion of steppe and saline elements (Artemisia and Chenopodiaceae) was greater at sites located in the Black Sea region, in agreement with greater temperature seasonality, lower precipitation and the occurrence of saline soils in this region.

The comparison of pollen records from the European forest steppe shows a west-to-east gradient in the timing and magnitude of deforestation (Fig. 6). For example, the timing of major anthropogenic ecosystem transformation in the lower Danube Plain from about 2500 cal yr BP falls in between that of other records in lowland areas in central-eastern and southeastern Europe, where it generally occurred after 3000 cal yr BP (Fig. 6). However, this is earlier than on the Eastern European Plain, where it mostly occurred after 2000 cal yr BP. On the Thracian Plain, southeastern Bulgaria, anthropogenic deforestation was, however, noted already from 4000 cal yr BP (Connor et al., 2013). The anthropogenically driven opening up of the forest steppe soon reached a similar extent as today in most regions, which then remained open until the present day, although climate conditions could have allowed for the recovery of tree cover. Notably, however, the study region is increasingly confronted by desertification (European Environmental Agency, 2016). Given its dry character, the conversion of forests to cropland may have acted as a positive feedback to the warm and dry climate, enhancing evaporation and altering the moisture balance, further contributing to the tendency towards the aridization of Lower Danube landscapes. Ongoing climate change (warmer temperatures and a decline in precipitation), coupled with agricultural intensification, will probably exacerbate the process of desertification.