Environmental change in the Limfjord, Denmark (ca 7500 e 1500 cal yrs BP): a multiproxy study

approach (including sedimentary parameters, diatoms, molluscs, foraminifera, sedimentary pigments, C and O stable isotopes and plant macrofossils) has been adopted to assess environmental change over the period ca 7500 e 1500 cal yrs BP at Kilen, a coastal fjord (before AD 1856) situated in the Western Limfjord. A diatom-based salinity transfer function based on a pan-Baltic training set has been applied to the fossil diatom dataset for quantitative assessment of salinity change over the study period. This study demonstrates that large-scale shifts in salinity are a common feature of the Limfjord ’ s long-term history and are driven by the level of connection with the North Sea and the Skagerrak respectively, which in turn is likely driven by the complex interplay between climate, sea-level change, current velocity and rates of erosion/sedimentary accretion. Three shifts in state at Kilen are identi ﬁ ed over the study period: a deep, periodically strati ﬁ ed fjord with medium e high salinity (and high productivity) between ca 7500 e 5000 BP, followed by a gradual transition to a shallow benthic system with more oceanic conditions (i.e. higher salinity, lower productivity, slower sedimentary accumulation rate and poorer fossil preservation) after ca 5000 BP and no strati ﬁ cation after ca 4400 BP, and lastly, within this shallow phase, an abrupt shift to brackish conditions around 2000 BP. Environ- mental e societal interactions are discussed on the basis of the data presented in this study and current environmental hypotheses for cultural change are challenged.


Introduction
The economic and societal importance of the inner Danish coastal waters (including estuaries and fjords) over many millennia is well documented (e.g. Rasmussen, 1968;Enghoff, 1999;Andersen, 2007;Poulsen et al., 2007;Enghoff, 2011), with these habitats commonly offering a wide range of resources, easily accessible from relatively shallow waters. The Limfjord is a sound situated in the northern part of Jutland, Denmark, connected to the Kattegat in the east and the North Sea in the west (Fig. 1A). Today, the Limfjord hosts a large shellfish industry, particularly Mytilus edulis (blue mussel) and Ostrea edulis (European flat oyster), but in the past has provided important fishing and breeding grounds for several finfish species, including eel, plaice, herring and whitefish (e.g. Enghoff, 1999;Poulsen et al., 2007). Furthermore, the presence of Stone Age shell middens and other archaeological sites (Fig. 1B), dominated by marine molluscs (particularly oysters in the late Mesolithic, prior to their widespread decline at the Mesolithice Neolithic transition; Andersen, 2007), but also containing fish bones (from marine, brackish and freshwater species) along its former coastlines suggests a long history of habitation and exploitation (e.g. Madsen et al., 1900;Andersen and Johansen, 1986;Andersen, 1995a;Enghoff et al., 2007).
Despite its cultural importance, the long-term natural environmental history of the Limfjord is poorly understood. The few studies that exist for this region focus predominately on the last 2500 years, when sea level was close to its present level (e.g. Gehrels et al., 2006) and subsequently the Limfjord was a particularly dynamic environment, subject to fluctuating salinities associated with the degree of exposure to the North Sea and Skagerrak (Kristensen et al., 1995;Christensen et al., 2004;Ryves et al., 2004;Lewis, 2011;Mertens et al., 2011). However, studies extending back further than 2500 yrs BP are poorly dated, have low temporal resolution and lack any attempt to quantify changes in salinity (e.g. Petersen, 1981;Heier-Nielsen, 1992;Andersen, 1992a). This paper presents results from a high-resolution, multiproxy study, analysing changes in the marine environment at Kilen, a small semi-isolated basin of the Western Limfjord, between ca 7500e1500 cal yrs BP (hereafter "BP"). Changes in salinity, productivity, hydrography and sedimentation at this site are inferred from sedimentary parameters, diatoms, foraminifera, molluscs, plant macrofossils and sedimentary pigments. A diatom-based transfer function is employed to produce quantitative salinity estimates over the study period derived from a large, existing pan-Baltic training set (Andrén et al., 2007). Environmental hypotheses for cultural change are reviewed on the basis of the new evidence presented here, including the long-debated possible environmental events at the MesolithiceNeolithic transition (Rowley-Conwy, 1984) as seen from the perspective of the western Limfjord.
This study is particularly important as it is the first long-term, multiproxy record, with good chronological control from the Limfjord region covering all three major phases of shell midden accumulation (ca 7400e5200 BP, 4800e4400 BP and 2500e 1500 BP; Andersen, 2007). Furthermore, long-term data on natural environmental variability are urgently needed in the context of growing awareness of, and concern with, coastal and marine environmental change throughout the Baltic Sea region (Conley et al., 2009;Kabel et al., 2012;Neumann et al., 2012 and references therein).

Study area
Kilen (56 30.005 00 N, 08 34.089 00 E) is a shallow eutrophic 'lake' situated in the western Limfjord (northern Jutland) in the Danish commune of Struer (Fig. 1). It is classified as a brackish-water lake Fig. 1. A) Location map showing the position of the Limfjord and the study site Kilen. Dotted lines represent haloclines of modern surface salinity (in g L À1 ) and black arrows represent major present day current patterns in the Skagerrak, Kattegat and eastern North Sea (after Gyllencreutz and Kissel, 2006). Abbreviations: NCC ¼ Norwegian coastal current, NJC ¼ North Jutland current, SJC ¼ South Jutland current, AW ¼ Atlantic water, BC ¼ Baltic current, BW ¼ Baltic water, CNSW ¼ Central North Sea water, JB ¼ approximate location of the Jutland Bank. B) Map of the Limfjord indicating the position of the study site and Mesolithic/Neolithic shell middens (red dots; after Andersen, 1992b). Dotted line indicates proposed shoreline during the late Mesolithic (after Andersen, 1992b). Abbreviations: JB ¼ Jutland Bank, AC ¼ Agger channel. C) Map of the Kilen sedimentary basin including major fluvial inputs, basic bathymetry and key catchment characteristics (after Jensen et al., 2006). Coring location is marked with a black star. The Kilerkanal (labelled) connects the Kilen basin with the Limfjord and is responsible for maintaining brackish salinity. Abbreviations: Br. ¼ Bremdal, SB ¼ Struer Bugt. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) (w6 g L À1 total dissolved solids), following its almost complete separation from the Limfjord by the building of a road and rail embankment in AD 1856. A connection with the Limfjord is retained via a small stream in the south-east corner (Fig. 1C) and is important for maintaining brackish salinity levels. The 'lake' is approximately 5 km long, 1 km wide with a total area of 3.34 km 2 , an average water depth of 2.9 m and a maximum depth of 6.5 m (Windolf et al., 1996;Jensen et al., 2006).

Methodology
In April 2007, a 15 m sedimentary sequence (collected with a Russian peat sampler as 1 m core sections from two overlapping boreholes, correlated via physical sedimentary parameters and coring depth; Lewis, 2011) was retrieved from Kilen from a water depth of 3.9 m (Fig. 1C). The lithology of each core section was described using a modified Troels-Smith system (Troels-Smith, 1955;cf. Birks and Birks, 1980), prior to being sliced up into 1 or 2 cm intervals and subsampled for physical sedimentary analyses, micro-and macrofossil analyses plus sedimentary pigment analysis. Organic and calcium carbonate content (CaCO 3 ) was determined for each core slice (from w1 cm 3 aliquots) via loss-onignition (LOI) at 550 C and 925 C respectively (Dean, 1974;Bengtsson and Enell, 1986) following overnight evaporation of interstitial water at 105 C (also enabling calculation of dry mass and sediment water content). The remaining residue after organic and carbonate LOI forms the minerogenic component of the sediment.
Sedimentary pigments from freeze-dried samples (weighed aliquots of w0.2e0.4 g from 1 to 2 cm core slices) were analysed using a modified method of Chen et al. (2001) via highperformance liquid chromatography (HPLC; using an Agilent 1200 series HPLC separation module with Quaternary pump, autosampler, ODS Hypersil column (250$4.6 mm; 5 mm particle size), and photo-diode array detector) with two standards per run. Pigments were identified based on their retention time and absorption spectra compared with published literature (e.g. Jeffrey et al., 1997) and commercial standards. All sedimentary and biological assemblages have been zoned via optimal splitting (based on an information content matrix of dissimilarity) using the package PsimPol v. 4.27 (Bennett, 2003e2009), incorporating a broken stick model to test the significance of zone splits. All stratigraphic diagrams were produced using C2 v.1.6.2 (Juggins, 1991e2009).
A total of 521 core slices 1 or 2 cm thick were wet-sieved into two fractions (500 mm and 100 mm) and hand-picked for macrofossils, both for obtaining sufficient terrestrial plant material for AMS 14 C dating and analysis of molluscs (both fractions) and foraminifera (100 mm fraction). The Kilen chronology is based on 13 AMS 14 C dates based on terrestrial plant macrofossil material in order to avoid problems associated with the marine reservoir effect Olsen et al., 2009;Philippsen et al., 2013). The final age model ( Fig. 2 and see Philippsen et al., 2013) was produced using Oxcal 4.1 (Ramsey, 2008) (k ¼ 150, A ¼ 73.3%) with the atmospheric calibration curve IntCal09 (Reimer et al., 2009) and allowances being made for changes in accumulation rate (based on CaCO 3 content; Fig. 2). Six pollen samples were analysed in order to identify the approximate location of the widespread mid-Holocene Ulmus (elm) decline (ca 5900 BP; Andersen and Rasmussen, 1993), subsequently providing an independent check on the age model (elm decline dated to ca 5850e 5800 BP at Kilen). For macrofossils (including plants, molluscs and other animals) all distinguishable remains (present from wet sieved fractions) were recorded and, where possible, identified. Mollusc nomenclature follows Petersen (2004).
Diatom samples (taken from 1 to 2 cm core slices) were prepared using standard techniques outlined in Renberg (1990) and Battarbee et al. (2001) with microspheres being added to assess diatom concentration (after Battarbee and Kneen, 1982) and the diatom flux calculated subsequently based on the age-depth model. A minimum of 300 valves were counted per sample excluding Chaetoceros resting spores and Rhizosolenia spines. Diatom preservation was assessed using a 2-stage categorization (pristine or dissolved), expressed as an F-index ratio (Ryves et al., 2006). Diatom taxonomy was harmonised with the Molten and Define project (http://craticula.ncl.ac.uk/Molten/jsp/; Andrén et al., 2007) for application of the salinity transfer function to fossil diatom data.
Although the original Molten/Define diatom training set was designed primarily as a tool to infer past nutrient (total nitrogen) concentrations in the Baltic Sea, salinity is a major control on diatom composition (see below; Andrén et al., 2007). A subset of 211 sites from the large pan-Baltic Molten/Define modern training set was selected for building a diatom-salinity model that was most appropriate for Danish coastal environments, including Danish, Swedish and Finnish samples, with one site later being removed as an outlier. In this training set, salinity was the most important variable uniquely explaining 7.8% (n ¼ 9999 Monte Carlo permutations, p ¼ 0.0001) in a single-variable analysis and exhibiting little co-variation with other significant variables under variance partitioning (dropping only slightly to 6.6% when total nitrogen, total phosphorus and depth were included as co-variables; n ¼ 9999 permutations, p ¼ 0.0001). Various diatom-salinity transfer functions were developed from this training set, their predictive power tested via bootstrapping (Â1000 cycles) and models evaluated using r 2 boot and root mean square error of prediction (RMSEP) values (Table 1). A weighted average-partial least squares (WA-PLS) component 2 model was selected as it performed best under internal validation and exhibited a similarly high predictive power to other published salinity-based transfer functions (see Table 1). Foraminifera assemblage counts (from 2 cm core slices) were made on samples using standard techniques described in Feyling-Hanssen et al. (1971) and Knudsen (1998). A minimum of 300 tests were picked, identified and counted where possible. Foraminiferal taxonomy follows Ellis and Messina (1949 and supplements). Oxygen and carbon isotopic analyses were performed on tests of Elphidium excavatum f. selseyensis from 44 samples (w5e20 tests per sample; weight 30e100 mg), using a GV

. Lithology and physical analyses
The profile is composed of near-homogeneous grey-brown marine clay gyttja containing varying amounts of silt and sand, plant matter and shell material (Fig. 3, see Lewis, 2011 and supplementary data). At the top of the profile higher organic content ( Fig. 3) is reflected by slightly darker-brown sediment (above 524 cm, ca 1900 BP). Based on the physical analyses (organic, CaCO 3 and minerogenic matter), three distinct sedimentary zones have been identified (Fig. 3), with the lowermost zone (Kil-S1; ca 7500e 5000 BP) exhibiting the highest accumulation rates over the study period (w0.33e0.51 cm yr À1 ). The general pattern of long-term sedimentation at Kilen (i.e. gradually decreasing up the profile) appears typical of the process of shallowing in lakes, where sediment focussing increases the sedimentation rate in deeper parts and declines as the water depth decreases (Hilton, 1985). However, variations in the accumulation of organic, CaCO 3 and minerogenic material are visible throughout the profile and likely reflect changing productivity and input of material into the system.
Throughout Kil-S1 organic matter accumulation is relatively high and stable, but there is an almost constant decrease in the accumulation of minerogenic material and, after 7000 BP, a minor decrease in CaCO 3 . The decrease in all three components (i.e. organic, carbonate and minerogenic matter) between ca 5500 and 4700 BP (marking the Kil-S1 e S2 transition) is probably largely an artefact of sediment focussing. A period of steady sedimentary Fig. 3. Kilen lithology and sedimentary parameters, including overall sedimentary accumulation rate (cm/yr), water content/dry mass (expressed as a percentage of the total sediment wet weight), accumulation (mg cm yr À1 ) of organic matter, calcium carbonate (CaCO 3 ), minerogenic matter, selected sedimentary pigments (pmol cm À2 yr À1 ) and associated metrics (i.e. Pheophytin-a/chlorophyll a ratio and UV-R index). Ultraviolet index (UV-R index) after Leavitt et al. (1997): UV-C/(alloxanthin þ diatoxanthin þ luteine zeaxanthin) Â 100. *Plot displayed on independent scaling. accumulation occurs after ca 4700 BP, before another substantial decline in the accumulation of both minerogenic, and to a lesser degree the organic material after ca 2700 BP. The uppermost zone, Kil-S3 (ca 1900e1500 BP), is marked by relatively slow (w0.08 cm yr À1 ), but constant accumulation of all three components.

Sedimentary pigments
A total of 725 samples were analysed over the period ca 7500e 1500 BP, yielding eleven identifiable pigments present, together with one ultraviolet absorbing compound (UV-C). Sedimentary pigments are plotted by accumulation (pmol cm À2 yr À1 ) in Fig. 3, together with the ultraviolet radiation (UV-R) index (after Leavitt et al., 1997) and pheophytin-a/chlorophyll a ratio (to assess pigment degradation; e.g. Reuss et al., 2005). The profile has been divided into four statistically significant zones.
In KIL-SP1 (ca 7500e6200 BP) total pigment accumulation is relatively high, with diatoxanthin (from diatoms, dinoflagellates and chrysophytes), lutein (higher plants, green algae, euglenophytes), b-carotene (most algae and plants), pheophytin-a (chlorophyll a derivative) being the most abundant pigment types. A distinct drop in pigment accumulation occurs after ca 7100 BP, a pattern which is identifiable in almost all pigments present. However, this decrease is short-lived and soon after ca 6800 BP, almost all pigments increase in abundance again, with the exception of UV-C, which remains present at lower, but variable, levels.
KIL-SP2 (ca 6200e5300 BP) is characterised by high marine productivity as all carotenoid and chlorophyll pigments reach their highest accumulation rates. Lutein, diatoxanthin and b-carotene remain the most individually abundant pigments, while UV-C practically disappears from the record after ca 6150 BP. Overall productivity decreases in KIL-SP3 (ca 5500e2000 BP) as total pigment accumulation decreases substantially between ca 5500e 4700 BP. Thereafter, pigment accumulation remains relatively low, but subject to minor fluctuations throughout the remainder of the profile. In KIL-SP4 (ca 2000e1500 BP), the diatoxanthin accumulation decreases, but alloxanthin (cryptophytes) becomes more abundant. Increased accumulation of lutein, b-carotene, okenone (purple sulfur bacteria), echinenone (cyanobacteria) and the chlorophylls imply slightly elevated productivity levels after ca 2000 BP.

Macrofossils
A total of 521 samples were analysed for macrofossils and concentrations are expressed per 50 ml of wet sediment in Fig. 4. Based on the entire macrofossil assemblage (plant macrofossils and molluscs), five statistically significant zones have been identified (Fig. 4). Plant macrofossils are generally sparse throughout and with the exception of Zannichellia palustris (abundant in Kil-M5), offer little palaeoenvironmental information.
Molluscs were present in 474 samples covering the period ca 7500e1500 BP, at a concentration of up to 125 per 50 ml wet sediment (average 7.3 per 50 ml). Marine bivalves are relatively abundant throughout most of the sequence, characterised by the common inshore 'fjord' type fauna (cf. Petersen, 1981) present in other Limfjord Holocene sequences (e.g. Petersen, 1981;Kristensen et al., 1995). In Kil-M1 (ca 7500e6400 BP), Corbula gibba and Spisula subtruncata are the most common and abundant mollusc taxa in Fig. 4. Kilen plant macrofossil and molluscan records (selected species), expressed as concentration (no. per 50 ml of sediment). Molluscs ordered via class (bivalvia or gastropoda) and minimum salinity tolerance (i.e. values in brackets in g L À1 ) after Sorgenfrei (1958). *Plot displayed on independent scaling. the Kilen record, supplemented by sporadic occurrences of several other species, particularly Tellimya ferruginosa and Abra alba. A drop in mollusc concentration occurs in Kil-M2 (ca 6400e6100 BP), as S. subtruncata disappears and C. gibba decreases in abundance. A. alba is common in Kil-M2, with several occurrences of Bittium reticulatum and a peak of T. ferruginosa at ca 6400 BP.
Mollusc concentration is higher again in Kil-M3 (ca 6100e 3750 BP) with Mysella bidentata and A. alba being the most abundant mollusc species, along with regular occurrences of T. ferruginosa, Hinia reticulata, S. subtruncata, B. reticulatum and Abra spp. Another decrease in the mollusc concentration occurs in Kil-M4 (ca 3750e2000 BP) as most taxa become only sporadically present and less abundant. A. alba remains the most regularly recorded species, though B. reticulatum is relatively common between ca 3000e2400 BP. A large shift in mollusc composition occurs in the uppermost zone (Kil-M5, ca 2000e1500 BP) as Parvicardium ovale becomes more abundant along with a number of low-medium salinity species (including Hydrobia ventrosa, Hydrobia ulvae, Mytilus edulis, Cerastoderma spp. and Littorina saxatilis) as overall mollusc concentration reaches its highest Holocene levels at Kilen.

Diatom assemblages
Kilen diatom assemblages (n ¼ 80; Fig. 5) are generally well preserved for marine environments (F index ¼ 0.23e0.82; average 0.46; Ryves et al., 2004) and predominately composed of (brackish-) marine benthic and tychoplanktonic diatom taxa, but with planktonic taxa generally being well represented. In Fig. 5 the entire diatom assemblage has been divided into three statistically significant zones and is presented alongside a quantitative reconstruction of salinity. Selected diatom-based metrics are also presented in Fig. 6.
In KIL-D1 (ca 7500e4700 BP) Chaetoceros spp. (8e43%, average 23%) are generally most abundant up until ca 5400 BP, supplemented by Opephora mutabilis (the most abundant benthic taxon), Paralia sulcata, Dimeregramma minor, and Grammatophora oceanica. High-salinity demanding species such as Delphineis minutissima and Cymatosira belgica briefly appear in the record between ca 7000e6500 BP. After ca 5300 BP, P. sulcata (tychoplanktonic) increases in abundance accompanied by short-term increases in several pelagic taxa including Skeletonema costatum, Cyclotella choctawhatcheeana, Cyclotella striata and Thalassionema nitzschioides. These latter taxa generally decline again around ca 5400 BP, at roughly the same time as a substantial drop in Chaetoeceros resting spores.
P. sulcata is the most abundant species throughout KIL-D2 (ca 4700e2000 BP; 14e61%, average ¼ 35.6%) as after ca 4500 BP percentages of Chaetoceros resting spores (and other pelagic taxa) further decrease. O. mutabilis and D. minor also decrease in abundance after ca 3700 BP as G. oceanica briefly becomes the most abundant true benthic taxon (between ca 3700 and 2600 BP). In the upper section of this zone (after ca 2500 BP) C. belgica, D. minutissima and the pelagic T. nitzschioides all increase in abundance. In KIL-D3 (ca 2000e1500 BP) high-salinity diatom taxa practically disappear from the record and Chaetoceros spp., O. mutabilis and Cocconeis scutellum var. scutellum dominate the diatom assemblage. A series of brackish-water periphytic species also appear in the record early in this zone, most notably Rhoicosphenia abbreviata, Epithemia turgida and Cocconeis placentula. The lowermost foraminiferal zone (KIL-F1; ca 7400e4400 BP, Fig. 7) is generally characterised by fluctuating abundances of predominately Elphidium excavatum f. selseyensis, Elphidium incertum, Ammonia beccarii and Haynesina germanica, with smaller abundances of Elphidium magellanicum. Highest foraminifera concentrations/fluxes occur at the bottom of the profile (ca 7400e 6200 BP), along with maximum diversity and highest percentages of Haynesina depressula, E. magellanicum, Stainforthia sp., and Bulimina marginata. In KIL-F2 (ca 4400e2200 BP) E. excavatum, A. beccarii and H. germanica remain most abundant taxa, but E. incertum and E. magellanicum both decrease (after ca 4700 BP). E. excavatum and H. germanica both decrease substantially after ca 3700 BP, coinciding with an increase in A. beccarii. Slightly later (at ca 2800 BP), there is a marked increase in H. germanica, and it remains a dominant species in the upper part of this zone. Elphidium williamsoni, E. margaritaceum and Elphidium gerthi are also regularly present throughout KIL-F2 at relatively low (but fluctuating) abundances.
A drop in foraminiferal concentration occurs at the onset of KIL-F3 (ca 2200e1900 BP). A. beccarii percentages decrease drastically in this zone. E. excavatum, H. germanica and E. williamsoni are the most abundant taxa throughout KIL-F3, though foraminifera remain at low concentrations until ca 1900 BP, after which A. beccarii increases and E. margaritaceum, E. incertum, E. gerthi and H. depressula become present again at low frequencies. After ca 2000 BP (Zone KIL-F4; ca 1900e1500 BP), foraminifera are again scarce with regularly less than 30 specimens per sample (and absent between ca 1600e1500 BP). Where present, Elphidium albiumbilicatum is dominant and E. williamsoni and E. excavatum are also frequent.
Foraminiferal d 18 O (Fig. 7) decreases steadily from þ1.7 to þ0.4& between ca 6600 BP and 4700 BP before increasing abruptly at ca 4700 BP and remaining relatively constant up until ca 3700 BP. After ca 2800 BP the limited data suggests d 18 O isotope values decrease again. d 13 C fluctuates between À1.1 and À3.0& over the period ca 6600e4600 before increasing in the upper section of the profile (À1.2 to À0.4& between ca 4600 and 2500 BP).

Discussion
5.1. Marine environmental change at Kilen (ca 7500e1500 BP): regional context In Fig. 8 a summary diagram is provided detailing key multiproxy findings from this study alongside selected regional environmental and climatic parameters. Whilst direct alignment of proxy records must be treated with caution, particularly due to possible dating uncertainties between different sites, there appears to be a close connection between palaeoceanographic changes in the North Sea . Kilen diatom associated metrics including diatom concentration (per g dw; dry weight) and flux (no. per cm À2 yr À1 ), Hills N 2 diversity index, diatom dissolution (F index: Ryves et al., 2009), pelagic: benthic: tychopelagic ratios diatom-inferred salinity (g L À1 ) and goodness of fit between the fossil data and the DI-salinity model at each level (see text for details). P1-3 ¼ phases of decline in pelagic diatom abundance referred to in the text. and Skagerrak with marine environmental change in the western Limfjord. This is exemplified by three apparent shifts in "state" of the Limfjord over the study period ranging from a "state" with stratified waters with high-salinity bottom waters and brackishmarine surface conditions in the lower part of the profile, gradually shifting to a "state" with greater mixing of the water column and thus more oceanic sea-surface conditions (high salinity) after ca 4400 BP and a change to brackish conditions throughout the water column in the time interval 2000e1500 BP. These changes appear to be driven by the degree of direct connection between the Limfjord and the North Sea and Skagerrak, for which DI-salinity appears to be a very good proxy, representing relative inputs of fully marine water from the north and west openings.
It is argued that the (diatom, foraminifera and mollusc-based) qualitative interpretation of salinity and quantitative DI-salinity reconstruction presented here are relatively accurate, agreeing well with other published studies in the region in the more recent past (i.e. last, 2500 years; Kristensen et al., 1995;Christensen et al., 2004;discussed below) and with ecological knowledge of the species present. The importance of salinity in driving trends in the fossil dataset were apparent using detrended correspondence analysis (DCA) using Canoco v. 4.5 (Ter Braak and Smilauer, 2002), with fresh/brackish water species falling at opposite ends of DCA axis 1 (explaining 15.7% of the total variation) to marine species (Lewis, 2011). Goodness-of-fit of the DI-salinity model was assessed for each fossil sample after deleting Chaetoceros (resting spores and vegetative cells) by (1) the proportion of fossil diatom data included in the diatom-salinity model and (2) the minimum dissimilarity coefficient between each fossil sample and the training set (Fig. 6). Average coverage of fossil data within the diatom-salinity model was 60% (range 37e84%), while analogues scores ranged from 70 to 113, with a mean of 93, below the threshold range of 100e150 for good analogues suggested by Jones and Juggins (1995);Juggins (2001). Reduced coverage of fossil data within the training set is almost entirely explained by Chaetoceros taxa, which generally increase down core. However, no independent test for spatial autocorrelation (Telford and Birks, 2005) was employed in this study and therefore it is acknowledged that performance statistics of the DI-salinity model (Table 1) might be slightly exaggerated due to an element of spatial structuring within the training set.
5.1.1. Kilen phase I, ca 7500e4400 BP: deep, productive and stratifying estuary 5.1.1.1. Environmental change from ca 7500 to 6200 BP. Diatom-inferred salinities (average 25 g L À1 , range w20e31 g L À1 ) for the late Mesolithic and most of the Neolithic period (ca 7500e 4400 BP; Fig. 8) generally suggest similar or slightly lower values than at present (ca 24e27 g L À1 at present for the Struer region; Burman and Schmitz, 2005;Hofmeister et al., 2006), despite evidence for higher relative sea levels in the Limfjord (1.5e2 m at Struer; Mertz, 1924) and deeper waters (perhaps 15e20 m deeper at Kilen as suggested by multiproxy evidence in this study) existing during the mid-Holocene. It is therefore argued here that the Limfjord was less open to the west than has been previously suggested (e.g. in palaeo-shoreline estimates; cf. Fig. 1B) as any substantially greater opening to the North Sea in the west than at present (w1 km; Fig. 1B) is likely to have produced far higher surface salinity (and more oceanic conditions) than suggested by the Kilen diatom record (Fig. 5). More localised factors for lower salinity such as increased freshwater input or development of beach ridges/sediment banks (reducing the connection of Kilen with the Limfjord) are deemed unlikely due to the foraminiferal evidence (Fig. 7) for high bottom-water salinity existing in Kilen at this time (see below).
The main opening to the Limfjord most likely existed in the north (or northewest), opening into the Skagerrak, as suggested by high reconstructed salinities in the northern Limfjord during the mid-Holocene (Petersen, 1981;Burman and Schmitz, 2005). Obstructive sand banks such as the emergent Jutland Bank (Leth, 1996) might have largely blocked off the western entrance to the North Sea prior to its drowning sometime around ca 6200 BP (Leth, 1996;Gyllencreutz, 2005) and impeded the flow, transport and erosion capacity of the North Jutland current (Fig. 1A) thereby enhancing sediment accretion along the western margin of the Limfjord and subsequently reducing water exchange between the Limfjord and the North Sea. A strengthening of the Jutland current and a shift towards coarser sediment in the Skagerrak has been shown to broadly coincide with the drowning of the Jutland Bank (Leth, 1996;Gyllencreutz and Kissel, 2006).
Whilst the diatom assemblage (Fig. 5) is largely reflecting salinity in the surface waters and littoral areas, the benthic foraminifera assemblage (Fig. 7) suggests that highest bottom-water salinities occurred between ca 7400 and 5000 BP (maxima ca 7400e6900 BP). This is marked by maximum abundances of E. incertum, which requires at least 20e24 g L À1 and is commonly found in a narrow zone just below the halocline (Lutze, 1965(Lutze, , 1974, Fig. 8. Comparison of the Kilen multiproxy data with regional hydrographic and climate proxies. A) Stratification indicators: high-salinity foraminifera ¼ summed abundance of Elphidium incertum, Elphidium magellanicum, Elphidium margaritaceum, H. depressula, Stainforthia sp. and B. marginata. P1-3 ¼ pelagic declines (see text). B) Diatom-inferred salinity, with aeolian events "A2eA4" identified in Clemmensen et al. (2009) and water column state (inferred from this study). C) Ultraviolet radiation (UV-R) index. D) Sedimentary pigment accumulation: smoothed curves (0.1 span) for total (left axis), okenone and echinenone (right axis). E) Total mollusc concentration. F) Sortable silt (core MD99-2286) and estimated Holocene bottom current strength evolution in the Skagerrak (from Gyllencreutz and Kissel, 2006) and timing of the drowning of the Jutland Bank (JB; Leth, 1996). G) Pollen-inferred mean annual air temperature ( C) at Lake Trehörningen, Sweden (Antonsson and Seppä, 2007). Grey shaded area covers the brackish period, when the Limfjord was isolated from the North Sea and Skagerrak. Cultural divisions after Fischer and Kristiansen (2002) H. depressula (requiring a minimum of 24 g L À1 ; Haake, 1962;Alve and Murray, 1999) and presence of E. magellanicum and B. marginata, which both require relatively high and stable salinity conditions (Murray, 1991;Conradsen et al., 1994). This might imply periodic stratification of the water column, as seen today in areas of the Limfjord (Grooss et al., 1996) and in some of the deeper Danish fjords (e.g. Mariager Fjord), whereby high-salinity oceanic water is overlain by fresher water input at the surface (with salinity differences of up to 6 g L À1 recorded between surface and bottom waters in the Limfjord today; Grooss et al., 1996).
It is highly likely that the Kilen basin was deep enough to stratify during the mid-Holocene, with sediment likely accumulating at w10e15 m below the current lake bed between ca 7400 and 5400 and with higher relative sea level than present day (Mertz, 1924). This might have enabled the higher salinity demanding foraminifera and mollusc species (e.g. T. ferruginosa, S. subtruncata and A. alba; Fig. 4) to live at depth beneath the halocline, and taxa able to tolerate lower salinities (e.g. C. gibba, H. ulvae) to occupy the surface waters or shallower, lower salinity (i.e. more regularly mixed) areas of the fjord. Deep water with regular stratification is further supported by the high percentage of pelagic diatom taxa (particularly Chaetoceros resting spores; Fig. 5), presence of deep-water favouring foraminifera (such as B. marginata, Fig. 7; cf. Murray, 1991) and the relatively high abundance of okenone, produced by planktonic purple sulfur bacteria that live in anoxic conditions (Fig. 3).
d 18 O values are also interpreted as a proxy for bottom-water salinity up until ca 4600 BP (Fig. 7), and in agreement with the foraminifera data they suggest relatively high salinity between ca 6600 and 4600 BP (high d 18 O), though characterised by a gradual freshening of the bottom water (decreasing d 18 O), likely due to decreasing regularity of stratification as the fjord becomes progressively shallower (resulting from high sedimentation rates (Fig. 3) and sea-level decline (after ca 6000 BP) due to isostatic rebound). Low d 13 C (below 'typical' marine values of w0&; Sharp, 2007) between ca 6600 and 4600 BP (Fig. 7) probably indicate foraminifera utilising greater respired/mineralised carbon dioxide (CO 2 ) under stratified conditions. Sea-level change in the western Limfjord is extremely complicated and poorly understood, particularly due to variable rates of isostatic rebound across the Limfjord (e.g. Petersen, 1981;Gehrels et al., 2006), with the study site being very close to the isostatic 'hinge point' between the uplifting and subsiding regions, which itself has migrated through time (Gehrels et al., 2006). The only available sea-level curves are from southwest Denmark (e.g. Gehrels et al., 2006;Pedersen et al., 2009) within the subsidence zone, where relative sea level has been rising over the study period and to the north, where uplift has caused relative sea-level decline since about 5900 BP (e.g. Petersen, 1981), neither of which can be directly applied to SW Limfjord.
It is likely that the sea level in the Kilen area reached its maximum height ofw1.5e2 m above present (Mertz, 1924) sometime in the Late Mesolithic (cf. Petersen, 1981;Berglund et al., 2005) and has been largely declining since then, though whether any of the transgressive/regressive phases seen across Denmark and the Baltic Sea (e.g. Berglund et al., 2005) are identifiable in the south western Limfjord remains uncertain. In terms of salinity (and productivity), sea level (and subsequent exposure to the North Sea and Skagerrak) is clearly an important driver, though the level of connection of the Limfjord with the open seas is further complicated by changing coastal geomorphology (i.e. building up of coastal spits and sand barriers) and climate change, particularly storminess associated with atmospheric pressure systems (e.g. North Atlantic Oscillation (NAO); see below).
High pigment concentrations and organic matter content between ca 7400 and 7000 BP (Fig. 3) suggest that the productivity was relatively high at the bottom of the profile, coinciding with the highest Holocene temperatures in Scandinavia (Holocene Thermal Maximum (HTM); Snowball et al., 2004;Antonsson and Seppä, 2007;Seppä et al., 2009;Brown et al., 2012;Fig. 8). There is, however, a brief decrease in pigment accumulation rates between ca 7000e6800 BP, which coincides with increased DI-salinities (>28 g L À1 ; Fig. 5) and the first major reduction in pelagic diatom taxa (P1; Fig. 6). This might be the result of increased input of saline water from the North Sea/Skagerrak, perhaps during stormier conditions (e.g. Yu, 2003;Yu and Berglund, 2007 and see below) or alternatively might be related to a short-term reduction in freshwater input to Kilen, supplying nutrients as well, and possibly causing a breakdown in stratification.
The high abundance of UV-C between ca 7400 and 6200 BP (Fig. 3), suggests increased depth penetration of UV-radiation (Leavitt et al., 1997). UVR pigments are primarily produced by benthic sediment dwelling organisms (e.g. cyanobacteria) as photoprotectants when subjected to increased UV-exposure (Leavitt et al., 1997;McGowan, 2007). Dissolved organic carbon (DOC) is important in aquatic systems for shading out UV-radiation (e.g. Scully and Lean, 1994;Leavitt et al., 1997) which might suggest that DOC levels were lower before ca 6200 BP, and protective pigmentation was needed despite the likely greater light shading from a deep, turbid water column with high minerogenic content and significant algal biomass (Fig. 3). Plenty of light penetration and low DOC is further supported by the presence of purple sulfur bacteria (indicated by okenone in the pigment record; Fig. 3), which exist under anoxic conditions, often in highly productive systems, but when light can still reach the deeper water layers (i.e. clear water). Okenone is therefore often used to infer stratification in limnic and coastal systems (Smittenberg et al., 2004;McGowan, 2007). 5.1.1.2. Environmental change from ca 6200 to 4400 BP. A high productivity phase occurs between ca 6200 and 4900 BP, marked by a high organic matter content and increased accumulation of sedimentary pigments (Fig. 3) and diatoms (including an increased presence of pelagic diatom taxa; Fig. 6). There is also a striking disappearance of UV-C after ca 6200 BP, most likely due to internal loading of DOC from phytoplankton blooming (Mannino and Harvey, 2000) and direct shading by phytoplankton, reducing the UV-penetration within the fjord, at least to the benthic areas. The increase in okenone after ca 6300 BP (Fig. 3) suggests that light still penetrated into the stratified anoxic layers. Progressive shallowing of the basin (especially after ca 5400 BP) might also have increased turbidity.
The fact that UV-C is not found in the Kilen sequence even after this productivity maximum (ending w4900 BP; Fig. 3), by which time Kilen was shallower and less productive (and therefore probably experienced greater benthic UV exposure), would accord with a deep-water benthic habitat for the organism(s) responsible for these pigments, although sediment re-suspension and turbidity would also increase with lower water level, perhaps also limiting purple sulfur bacteria (i.e. low okenone accumulation after ca 5000 BP). Increased input of terrestrially sourced DOC, perhaps associated with catchment changes such as forest restructuring following the Ulmus decline and the introduction of agriculture (ca 5900 BP; Andersen and Rasmussen, 1993), might also have helped to block out UV-radiation.
Accordingly, high productivity between ca 6200 and 5800 BP (Fig. 3) is most likely the result of changes in the marine environment, rather than changing inputs from terrestrial sources. Stratification is still prevailing, as indicated by the foraminifera (Fig. 7), and might even have intensified (i.e. become more regular as exemplified by both high pigment fluxes and increased diatom plankton) following the drowning of the Jutland Bank (ca 6200 BP; Leth, 1996;Gyllencreutz and Kissel, 2006) and likely opening (or widening) of the western Limfjord connection to the North Sea. Subsequently, nutrient recycling might have become more intense under increasingly more stratified conditions, resulting in higher productivity and enabling nutrient-demanding planktonic organisms (e.g. diatoms; Chaetoceros resting spores, C. choctawhatcheeana, C. striata, S. costatum; Fig. 5) and bacterioplankton to bloom. This might in turn induce greater production of DOC and reduce UV-light penetration to the benthic areas. A prolonged period of high productivity in the Limfjord might have been sustainable due to regular input of highly saline, oxygenated water through the northern and western openings (encouraging stratification of water masses), perhaps associated with increased storminess as evident in the GISP2 ice core record between ca 6000 and 5000 BP (Mayewski et al., 1997) and linked to sea-level events in the Baltic Sea (Yu, 2003;Yu and Berglund, 2007) and subsequently increased erosion along the Limfjord coastline.
A secondary increase in sedimentary pigment concentrations occurs ca 5800 BP (Fig. 3), which might suggest a later role for terrestrial input of nutrients following human activities (e.g. induced by early forest clearance) as evident in numerous terrestrial pollen records (e.g. Iversen, 1941;Aaby, 1986;Odgaard, 1994;Andersen, 1995b;Rasmussen, 2005). In the palynological record, elevated abundances of Poaceae pollen at ca 5800 BP, and shortly after the appearance of Plantago lanceolata (ribwort plantain), marks the first signs of major agriculture-based land-use changes in the Kilen area (P. Rasmussen, unpublished data). This interpretation must, however, be treated with caution due to the low number of pollen analyses at Kilen, but would be in good agreement with other Danish palynological studies (Aaby, 1986;Odgaard, 1989Odgaard, , 1994Rasmussen, 2005).
After a brief drop, DI-salinity begins to rise again after ca 5200 BP, reaching near fully marine sea-surface conditions (>30 g L À1 ) shortly after ca 4200 BP (Fig. 5). The foraminifera suggest that the seasonal stratification of the water column ceased after ca 4400 BP and that the water column was mixed during the remaining part of the record. Whilst the d 18 O suggest an increase in salinity after ca 4600 BP (marked by a w0.8& increase in d 18 O; Fig. 7), again probably reflecting the shift to mixed water conditions (i.e. absent or rare stratification), after which d 18 O become a proxy of overall water column salinity, rather than bottom-water salinity only.
The reduction in organic matter, flux of sedimentary pigments (Fig. 3) and decline in pelagic diatom abundance (i.e. second major reduction in pelagic diatom taxa, P2; Fig. 6, particularly Chaetoceros resting spores; Fig. 5) suggest a decline in marine primary productivity after ca 5400 BP. This might signify reduced stratification (supported by low okenone abundance after ca 5000 BP) due to continuous shallowing of the Kilen basin caused by both sediment accumulation and isostatic land uplift. There is little change in the pheophytin-a/chlorophyll a ratio, which suggests that poorer preservation is not responsible for this drop in pigment concentrations (Fig. 3). After ca 4800 BP, the sedimentary pigment concentration and accumulation rates generally remain relatively low. Increased energy in the system (due to stronger currents, increased exposure and shallower depth) as marine conditions intensified at Kilen (after ca 5000 BP), might also have caused greater transportation, re-working and re-deposition of sediments out of the Kilen basin and poorer preservation (e.g. lower plant macrofossil and mollusc concentration (Fig. 4), lower diatom F index; (Fig. 6)).

Phase II, ca 4400e2000 BP: non stratified, high salinity oceanic estuary
Diatom-inferred salinities gradually rise throughout the Bronze Age and the early Iron Age perhaps due to gradual widening of the western Limfjord opening, though the overall assemblage indicates that Kilen became a shallow, benthic dominated system from ca 4400 BP, completing a process that began perhaps a thousand years before, although the final stage was rapid. This third major drop in pelagic diatom taxa (P3, Fig. 6), and an increase in foraminifera indicative of shallow intertidal conditions, particularly E. williamsoni ( Fig. 7; cf. Alve and Murray, 1999) likely marks the end of any stratification, as the Kilen basin becomes too shallow to prevent mixing of water masses.
The foraminiferal assemblage suggests a transitional phase between ca 4400 and 3600 BP following the termination of stratified conditions within the basin and subsequent mixing of the water column, as indicated by the remaining relatively high contents of E. excavatum (Fig. 7). Gradually more unstable conditions prevail between ca 3700 and 2800 BP, although with relatively high salinity being inferred by the presence of foraminiferal taxa such as E. magellanicum and E. margaritaceum. Both the diatom and foraminiferal assemblages document a salinity decline around 2800 BP, where the subtidal euryhaline species H. germanica (cf. Murray, 1991) becomes dominant and the intertidal E. williamsoni appears. This might be related to a globally synchronous, solar-forced climatic event occurring ca 2800 years ago, which manifested itself as a cooler and wetter phase in northwest European sediment records (Martin-Puertas et al., 2012), or a sea-level decline reducing the input of marine water from the North Sea, though no reliable record of sea-level change exists for the SW Limfjord during this period. d 18 O values remain relatively stable after ca 4600 BP (Fig. 7), with the exception of a decline at ca 2500 BP, which might represent a rapid freshwater input event, though as this is not identifiable in any other proxies might be a response to some other variable (e.g. test dissolution), or re-working. The d 13 C increase by w0.5e1& after ca 4600 BP (Fig. 7), is likely due to greater mixing with open marine water following reduced stratification and greater input of normal marine salinity waters from the North Sea (as marked by increasing DI-salinity). Therefore, overall the rather limited isotope data set (particularly after ca 4500 BP) support the foraminiferal assemblage counts and other proxy data for changes in stratification state.
A. beccarii is a dominant species accounting for >40% of the fauna between ca 3600 and 2300 BP (Fig. 7). This species is found today both in brackish waters (Walton and Sloan, 1990;Murray, 2006) and in fully marine waters (Rouvillois, 1970;Alve and Murray, 1999) and in the Kilen sequence, it appears to become dominant during the transitional phase of mixed water, corresponding to its present-day distribution in Danish marginal marine environments (Alve and Murray, 1999) The high abundances of P. sulcata and G. oceanica (Fig. 5) and regular presence of high-salinity demanding mollusc taxa (i.e. B. reticulatum, A. alba; Fig. 4) suggests that near fully marine conditions existed at Kilen at this time. It is argued here that in this section of the profile (i.e. between ca 4400 and 2300 BP) the diatoms and molluscs are more accurately reflecting the salinity pattern (i.e. increasing salinity reaching near fully marine conditions), though it is acknowledged that the DI-salinity values are over-estimated in some samples between ca 2300 and 2000 BP (see below), whilst the foraminifera must be responding to some other variable, possibly decreasing water depth as the benthic habitat becomes more restricted (e.g. decreased living space, increased competition for resources), or are being heavily affected by test dissolution, apparent in samples after ca 4400 BP.
After ca 2800 BP, DI-salinity increases again reaching maximum Holocene levels between ca 2500 and 2000 BP, at which point the assemblage is dominated by high-salinity demanding species such as P. sulcata, C. belgica and D. minutissima (Fig. 5). A slight increase in E. excavatum (indicative of open water conditions), together with E. incertum and E. magellanicum in the foraminiferal record between ca 2100 and 2000 BP support increased salinity in the late Pre-Roman period, whilst maximum abundances of E. williamsoni indicate a further decrease in water depth (Fig. 7). This salinity maximum is likely to mark the maximum connection of the western Limfjord with the North Sea and presumably also implies that a connection still existed with the Skagerrak to the north.
Higher sea levels might explain this period of higher salinity, perhaps due a period of increased storminess causing piling up (and subsequently greater input) of high-salinity sea water in the eastern North Sea and Baltic Sea, similar to the prolonged NAO þ type cycles identified by Yu and Berglund (2007), though the long-term history of the NAO remains uncertain and only tentative links can be made here. Nevertheless, high salinity is documented at the more northerly Bjørnsholm Bay site around 2200e2800 BP (Christensen et al., 2004;Mertens et al., 2011), suggesting that a connection between the Limfjord and the Skagerrak existed at this time, despite long-term sea-level decline occurring in the northern region due to isostatic uplift (Petersen, 1981). Higher summer and winter temperatures, inferred by regional pollen reconstructions (Antonsson and Seppä, 2007;Seppä et al., 2009;Brown et al., 2012, Fig. 8) and drier conditions, inferred from several Danish and Swedish peat and lake records (e.g. Barber et al., 2004;Olsen et al., 2010), possibly indicate that greater evaporation and reduced freshwater input from the catchment during the Pre-Roman Iron Age might also be important for explaining near-fully marine salinities between ca 2500 and 2000 BP (Fig. 8).
In several of the more poorly preserved samples (F < 0.4, Fig. 6), exceptionally high salinities (35e43 g L À1 ) are inferred, even exceeding present day average salinities for fully marine conditions (i.e.>35 g L À1 ). However, even during maximum exposure to the North Sea, the salinity of the Limfjord is unlikely to exceed 35 g L À1 and therefore we conclude salinity has been over-estimated by the model in this section of the record ca 2300e2000 BP. Poor preservation has been linked elsewhere in the Limfjord and in inland saline lakes, to overestimation of DI-salinity (Ryves et al., 2004(Ryves et al., , 2006. Nevertheless, (as indicated above) almost fully marine salinities in the late Pre-Roman Iron Age are consistent with foraminiferal, dinoflagellate and mollusc records from the Bjørnsholm Bay region (Kristensen et al., 1995;Christensen et al., 2004;Mertens et al., 2011) and also with high salinity inferred from diatoms and molluscs at Horsens Fjord, East Jutland and Tempelkrog in Isefjord, Zealand (D.B. Ryves et al., unpublished data). Despite this, overall the application of the diatom-salinity model is considered to have produced a reliable reconstruction of salinity change over the study period (further supported by C/N and d 13 C analyses on bulk sediment from Kilen; Philippsen et al., 2013) that can also be qualitatively confirmed by ecological knowledge. Existing palaeosalinity inferences for the Limfjord are only available for the last ca 2800 years, over which period the Kilen DI-salinity record is in relatively good agreement with other published records of salinity change (e.g. Christensen et al., 2004;Mertens et al., 2011).

Kilen phase III, ca 2000e1500 BP: shallow brackish lagoon
A marked shift to brackish conditions occurs around 2000 BP, with DI-salinity falling by over 10 g L À1 (w19.9 g L À1 at ca 1900 BP; Fig. 8). The high-salinity diatoms such as C. belgica and D. minutissima practically disappear from the record, and P. sulcata abundance drops substantially, as brackish-marine species such as O. mutabilis and C. scutellum become more abundant in the record (Fig. 5). This is further supported by an increase in the abundance of brackish water molluscan taxa (e.g. M. edulis, H. ventrosa, H. ulvae, Cerastoderma spp.; Fig. 4) and the regular presence of Z. palustris (horned pondweed; Fig. 4) after ca 2000 BP, which only grow in brackish-water with an upper salinity tolerance of 12e15 g L À1 (Moeslund et al., 1990).
After 2000 BP, the foraminiferal assemblage is seriously affected by dissolution of tests, but when present, dominance of E. albiumbilicatum supports brackish-water conditions at this time (Fig. 7). This species tolerates salinity as low as 3 g L À1 , but the presence of additional species such as E. excavatum, E. williamsoni and H. germanica suggests that salinity did not fall below 15 g L À1 . The cause of this salinity decline is likely to be a closing, or at least severe narrowing, of both the northern (seen also in the central Limfjord; Kristensen et al., 1995;Christensen et al., 2004) and western openings of the Limfjord, probably due to greater accretion of sediments (building up spits and sand barriers), driven by climate change (Clemmensen et al., 2009) and/or sea-level decline due to isostatic uplift (Petersen, 1981).
In addition to long-term deposition of sediments transported from the west by the Jutland Current and re-deposited along the Jutland-Skagerrak coastline (Jiang et al., 1997;Gyllencreutz and Kissel, 2006), this closure also broadly coincides with a period of increased aeolian activity and subsequent dune building along the northwestern and northern coasts of Jutland (Clemmensen et al., 2009). The initiation of these aeolian events (indicated on Fig. 8) has been linked to wet/cool summers (in Swedish peat bog records) and are believed to represent more frequent passage of cyclones across Denmark in the summer season (Clemmensen et al., 2009). Increased freshwater input to the Limfjord might have lowered salinity, whilst increased storminess and movement of aeolian material probably contributed to the closure of the Limfjord's entrances (via sedimentary accretion) to the North Sea and Skagerrak.
There is some evidence for increased marine productivity in the Roman period with both mollusc and plant macrofossil concentrations increasing after ca 2000 BP. Total pigment accumulation also increases slightly after ca 1900 BP (Fig. 3), driven by specific sedimentary pigments, most notably alloxanthin (from cryptophytes), lutein (from green algae, euglenophytes, higher plants) and canthaxanthin (from colonial cyanobacteria). In contrast to this, diatoxanthin (from diatoms, dinoflagellates and chrysophytes) and the diatom concentration/flux (Fig. 6) decrease around the onset of the Roman period (ca 2000e1500 BP) and remain low thereafter. This decrease might be explained by a decline in silica supply, increased competition for resources from other algal groups or greater predation of diatoxanthin-producing communities by molluscs, ostracods (both of which increase in abundance in the brackish Roman period) and other higher organisms.

Interactions between environment and society
Shell middens are relatively common archaeological sites across coastal Denmark, but somewhat surprisingly are absent from the western Limfjord (see Fig. 1B) throughout the entire Holocene (Andersen, 1992a(Andersen, , 1995a(Andersen, , 2000b(Andersen, , 2007. For the late Mesolithic/ early Neolithic period, lower surface salinity in the western Limfjord might explain this absence, with regular stratification likely preventing high-salinity water from reaching the innermost, shallow areas of fjords and estuaries, thereby restricting the heavily exploited, high-salinity demanding oysters (abundant in the Mesolithic layers of shell middens throughout Denmark; Andersen, 2007) to the deeper, inaccessible waters. High sedimentary accumulation rates during the Ertebølle period (e.g. 0.36e 0.51 cm À2 yr À1 ; 7400e5800 BP) might also be important for preventing oyster colonisation and establishment of beds. O. edulis struggles with higher volumes of fine sediment, being less efficient at ejecting continuously accumulating sediment from its mantle cavity than species such as Cerastoderma edule (Yonge, 1960;Bailey and Milner, 2008).
The widely observed oyster decline in shell middens across Denmark at the MesolithiceNeolithic transition (ca 5900 BP) has encouraged the development of a geographically widespread model to explain this shift, with a decline in salinity being the most commonly cited, single causal factor (Rowley-Conwy, 1984;Andersen, 2007;cf. Schulting, 2010). Archaeologically, this model has already been challenged by inconsistencies in other shell middens, such as Krabbesholm and Visborg, both of which contain Neolithic layers in which oysters remain abundant (Andersen, 2000a(Andersen, , 2005Nielsen, 2008). Here, it can be challenged by environmental data from the western Limfjord (Figs. 4e7), which directly contradicts the hypothesis that lower salinity existed across the Limfjord during the early Neolithic period. Similarly, a decline in salinity is also unable to explain the oyster decline in the Norsminde Fjord shell midden (east Jutland) and subsequently a sedimentary hypothesis is proposed for this site (Lewis, 2011). It is likely that more localised factors, dependent upon catchment characteristics (cf. Nielsen, 2008;Lewis, 2011) might be more important than previously thought. In future, propagation of regional environmental changes into individual fjord and estuarine systems from common forcing factors such as sea-level and climate change must be considered more closely and critically tested by local palaeodata (Lewis, 2011).
In the Bronze Age (3600e2400 BP), marine resources were still exploited (e.g. occasional shell layers and fishing tools and artefacts; Rasmussen, 1992;Andersen, 1998;Ringtved, 1998) but appear to be much less important and coastal shell middens are absent from the entire Limfjord (Andersen, 2007). This study suggests that higher salinity conditions, and generally lower marine productivity (i.e. lower pigment fluxes) existed in the Bronze Age (Fig. 8), likely accompanied by greater energy and currents and coarser sediment, which might have limited or confined natural shell beds to specific areas of the Limfjord.
Shoreline shell middens are also absent from the entire Limfjord throughout the Iron Age (2400e900 BP), though this is somewhat aberrant and difficult to explain (large shell middens predominately composed of the blue mussel (M. edulis) are abundant along the east coast of Jutland, concentrated between ca 2300 and 1700 BP; Poulsen, 1978;Andersen, 2007). Regional molluscan records document that M. edulis was present in the Limfjord during the Roman period (Kristensen et al., 1995; this study) ruling out absence of this resource. The high abundance and widespread spatial distribution of mussels during the Iron Age might have meant that site selection was more heavily determined by additional resources available to these cultures. Alternatively mussels might have been collected more sporadically (perhaps as a dietary supplement), from different localities, preventing significant accumulation of shell middens: a number of sites have been found several kilometres inland, containing heaps of marine molluscs or smaller shell middens (Mikkelsen, 1994;Andersen, 2007), while some existing data indicate small numbers of shells were transported to more inland settlement sites prior to shelling (Poulsen, 1978;Ringtved, 1992).

Conclusions
This high-resolution study of Kilen has provided new and detailed information about the environmental history of the western Limfjord, much of which can be extrapolated to the Limfjord proper, with three major salinity and productivity shifts being documented between ca 7500 and 1500 BP. Kilen in phase I (ca 7500e4400 BP) was a brackish-marine estuary with high productivity and regular stratification of water masses. There was a gradual transition to phase II, which was characterised by near fully marine salinities, low productivity and shallow water (probably encouraging greater mixing of water masses and ending stratification of the basin), and then a sudden switch to brackish conditions with medium-low productivity in phase III (between ca 2000e1500 BP), likely in response to the previously documented Iron Age closure of the Limfjord. These shifts can be (broadly) synchronously linked to palaeoceanographic events registered in the North Sea and Skagerrak (e.g. drowning of the Jutland Bank) and to climate change in northern Europe. Of particular importance is the relative exposure of the Limfjord to these two seas (North Sea and Skagerrak), which appears to be driven by a complex interplay between climate, sea level, ocean currents, erosion and sedimentary accretion along the outer margin of the Limfjord.
Overall, the data here indicate that the Limfjord is a fragile system, subject to large scale changes in ecosystem structure and physical conditions over a range of timescales. Importantly, this study has provided a high-resolution, well dated multiproxy record of environmental change from an important archaeological region, enabling future debate concerning links between sea and society to be placed within a proper environmental context.

Acknowledgements
Beth Stavngaard, Ole Bennike and Bent Odgaard are thanked for field and laboratory support. Molten and Define members are also thanked for transfer function production, support and helpful comments and discussion. Funding from Loughborough University Development Fund (Ph.D. funding for JPL), Quaternary Research Association (QRA) New Researchers' Award (for sedimentary pigments to JPL) and NERC Isotope Geosciences Facilities Steering Committee (NIGFSC; award no. IP/1080/1108, for isotope analysis to DBR and JPL) and the Geological Survey of Denmark and Greenland (for funding several 14 C-datings) are gratefully acknowledged. Two anonymous reviewers are thanked for thorough and thoughtful reviews that have improved the paper. This manuscript is dedicated to Kaj Strand Petersen, who sadly passed away during the course of this project and will be greatly missed by family, friends and colleagues.