The Iberian Peninsula is influenced by the major Atlantic and Mediterranean atmospheric regimes (Lionello, 2012). Today, the Atlantic climate is typically marked by relatively cool annual air temperatures and evenly distributed precipitation throughout the year. In contrast, the Mediterranean climate is characterized by pronounced seasonal contrasts with a rainy winter season and a dry and hot summer season. In general, most of the precipitation at the Iberian Peninsula occurs during the winter season (Fig. 1; Lionello, 2012). The spatial pattern of winter precipitation clearly reflects the two atmospheric regimes (Fig. 1). The Atlantic regime, which spans along the western and northern coasts, is characterized by high precipitation of more than 800 mm during winter. It is associated with the westerly wind belt at the Iberian Peninsula (Zorita et al., 1992) and to a large extent controlled by the NAO winter conditions (Hernández et al., 2015; Hurrell, 1995). During positive NAO conditions (i.e. a large pressure difference between the Azores High and the Icelandic Low), North Atlantic storm tracks and associated precipitation are directed towards northern Europe, while Iberia receives less rainfall. On the other hand, during negative NAO conditions (i.e. a small pressure difference) storm tracks are directed towards the Iberian Peninsula, which then experiences wetter winters. The pronounced difference in precipitation is also evident in the seasonal variability in the river discharges in southern Iberia. For example, the average discharge of the Guadalquivir is below about 100 hm3 month-1 from April to October and peaks during December and January with values above 400 hm3 month-1 (Fernández-Delgado et al., 2007). Nonetheless, the central parts of the Iberian Peninsula as well as the eastern and southern coasts under the influence of the Mediterranean climate remain relatively dry on average with precipitation less than 600 mm during winter (Fig. 1).

A strong seasonal contrast is also evident in the surface ocean in the Gulf of Cádiz, which receives warm waters with the eastward Azores Current (AC) that turns into a poleward surface current (Iberian Poleward Current; IPC) along the western Iberian coast during winter (Peliz et al., 2005). During summer, however, the surface current is directed southward and favours coastal upwelling along the western continental margin limiting the influence of the AC, and thus causing relatively low SSTs (Fig. 1; Haynes et al., 1993; Peliz et al., 2002). These opposite oceanographic currents limit the seasonality recognized in the SSTs in the Gulf of Cádiz. The seasonal difference at the core location of GeoB5901-2 is typically ∼6 ∘C during modern times (Locarnini et al., 2013). Atlantic surface water passes the Strait of Gibraltar and enters the Alboran Sea. Within the Alboran Sea it circulates in two anti-cyclonic gyres – the West Alboran Gyre (WAG) and the East Alboran Gyre (EAG) (Lionello, 2012). At the WAG relatively warm and fresh waters become upwelled almost continuously throughout the year (Sarhan et al., 2000). This upwelling is accompanied by an elevated MPP at the northern rim of the gyre (Minas et al., 1991). The modern seasonality in the Alboran Sea is very similar to the one in the Gulf of Cádiz (∼6 ∘C).

For this study two marine sediment cores from the Alboran Sea and the Gulf of Cádiz were analysed. Sediment core ODP-161-976A (36∘12.32′ N; 4∘18.76′ W; 1108 m water depth) was retrieved in the Alboran Sea during the JOIDES RESOLUTION cruise in 1995 (Comas et al., 1996). To achieve multi-decadal resolution, the section from 100.0 to 149.0 cm was continuously sampled at 0.5 cm distances in the IODP (International Ocean Discovery Program) core repository at MARUM (Center for Marine Environmental Sciences) in Bremen (Germany). The analysed section consists of homogenous clayey and silty sediments, which appear slightly bioturbated. No hiatus or depositional event (e.g. turbidite) has been recognized throughout the section (Comas et al., 1996). Sediment core GeoB5901-2 (36∘22.80′ N; 7∘04.28′ W; 574 m water depth) was retrieved during the METEOR cruise in 1999 in the Gulf of Cádiz (Schott et al., 2000). This sediment core, containing hemipelagic mud, was sampled at 0.5 cm resolution from 1.0 to 50.0 cm in the GeoB core repository at MARUM in Bremen (Germany). The samples used for organic geochemical analysis were freeze-dried and homogenized prior to analysis.

Age models of both cores are based on existing and new AMS 14C dates, all measured at the Leibniz Laboratory at Kiel University (Table 1). Published data of cores ODP-161-976A and GeoB5901-2 were taken from Combourieu Nebout et al. (2002) and Kim et al. (2004), respectively. In addition, we measured six new AMS 14C dates on monospecific planktic foraminiferal samples of Globigerinoides ruber white + pink or Globigerina bulloides larger than 150 µm from sediment core ODP-161-976A as well as seven new dates from sediment core GeoB5901-2. A section from 116.25 to 124.75 cm in ODP-161-976A included three samples yielding the same AMS 14C age (see Table 1). Since there is no evidence for strong bioturbation, hiatuses, or turbidites, we assume a rather continuous sedimentation rate throughout the core, and thus decided to take the date at 120.25 cm as the only age control point for this depth interval since it was dated twice on two different planktic foraminifera. The dates at 116.25 and 124.75 cm were not considered for the final age model in order to avoid artificially induced calculations of extraordinary high sedimentation rates in this thin sediment increment. The final age models have been calculated using Bacon (Blaauw and Christen, 2011). During age model processing, all AMS 14C dates have been calibrated using the marine13 calibration curve including a global mean reservoir correction of 400 years (Reimer et al., 2013). Based on the AMS 14C dates, we assume a relatively high accumulation rate for ODP-161-976A. Accordingly, we set the mean accumulation parameter of the model to 50 yr cm-1 and run the model for 2.5 cm thick sections to allow for a certain variability in the accumulation rates throughout the section. The accumulation rate of GeoB5901-2 is expected to be lower in the analysed section based on AMS 14C data. Thus, the model parameters have been set to 100 yr cm-1 and 5 cm thick sections. In the case of sediment core ODP-161-976A the initial age model of Jiménez-Amat and Zahn (2015) was shifted significantly within the studied section by about 700 years. The reason for this shift are the new AMS 14C dates accomplished during this study emphasizing the need of a dense dating strategy. According to the final age model, sedimentation rates range between 9.8 and 50.0 cm kyr-1 in core ODP-161-976A and between 4.8 and 33.3 cm kyr-1 in core GeoB5901-2, respectively. With 0.5 cm sampling intervals decadal to multi-decadal resolution is achieved. The studied section of sediment core ODP-161-976A dates between approximately 5.4 to 3.0 ka cal BP exhibiting a temporal resolution between 10 to 57 years per sample with continuous high resolution between 5.4 and 4.6 ka cal BP. The analysed section of sediment core GeoB5901-2 dates between approximately 6.2 and 1.8 ka cal BP resulting in a temporal resolution between 15 and 104 years per sample. Age models and sedimentation rates for both sediment cores are shown in Figs. 2 and 3, respectively.

Lipids were extracted from the freeze-dried and finely ground sediment samples with an accelerated solvent extractor (ASE-200, Dionex) at 100 bar and 100 ∘C using a 9:1 (v/v) mixture of dichloromethane (DCM) and methanol. After extraction samples were de-sulfured by stirring for 30 min with activated copper. The de-sulfured lipids were subsequently separated by silica gel column chromatography using activated silica gel (450 ∘C for 4 h) into neutral (hexane) and polar (DCM) fractions containing n-alkanes and alkenones, respectively. The neutral fraction was further separated using silver nitrate (AgNO3) coated silica gel. Samples were then left at room temperature for approximately 24 h for homogenization.

Afterwards, n-alkanes were analysed by gas chromatography (GC) using an Agilent 6890N gas chromatograph equipped with a Restek XTI-5 capillary column (30 m × 320 µm × 0.25 µm) and a flame ionization detector (FID) at the Institute of Geosciences, Kiel University. Example chromatograms are provided in the Supplement of this paper. n-Alkanes were identified by the comparison of their retention times with an external standard containing a series of n-alkane homologues of known concentration. On this basis, n-alkanes were also quantified using the FID peak areas calibrated against the external standard. For environmental interpretation the sum of terrestrial-sourced odd n-alkane homologues C27 to C33 is used. The mean analytical error (2σ) is 7.0 ng g-1 sediment based on replicate analyses (n=62).

Various studies used ratios among individual n-alkane homologues as the environmental sensitive parameter, e.g. the Norm33 ratio (e.g. Herrmann et al., 2016). The Norm33 ratio is calculated by the following equation: 1Norm33=C33/C29+C33, where CX is the peak area of the n-alkane in the chromatogram with x being the number of carbon atoms. The mean analytical error (2σ) is 0.004 based on replicate analyses (n=62).

Alkenones were analysed by multi-dimensional double gas column chromatography (MD-GC) set up with two Agilent 6890 gas chromatographs. The compounds (C37:2 and C37:3) were quantified by calibration to an external standard. The alkenone concentration is derived from the sum of the C37:2 and C37:3 isomers with a mean analytical error (2σ) of 6.9 ng g-1 sediment. The alkenone unsaturation index (U37K′) was obtained by using the peak areas of the aforementioned compounds by applying the equation of Prahl and Wakeham (1987): 2U37K′=C37:2/C37:2+C37:3, where CX represents the respective peak area in the chromatogram. The U37K′ index was subsequently transferred into annual mean SST using the calibration of Müller et al. (1998): 3SST(∘C)=U37K′-0.044/0.033. The laboratory internal analytical error is approximately 0.12 ∘C, while the error of the calibration is 1.5 ∘C.

For foraminiferal analysis sediment samples of approximately 10 cm3 were washed over 63 µm sieves, dried at 40 ∘C, and subsequently dry-sieved for larger fractions. Planktic foraminifera assemblages were analysed in the size fractions >150 µm enabling the application of commonly used transfer techniques for SST reconstructions based on relative abundances of 26 taxonomic categories within the assemblage, following the concept by Pflaumann et al. (1996). For reliable assemblage counts samples were dry-split into aliquots of at least 300 specimens with a Kiel dry sample splitter. For SST estimates we used the SIMMAX non-distance-weighted modern analogue technique, using a similarity index of >0.963 and based on 10 closest analogues (Pflaumann et al., 2003). Because the study sites are influenced by Atlantic and Mediterranean ocean circulation, we combined an updated North Atlantic core-top database (Kucera et al., 2005a, b; Salgueiro et al., 2014) and the Mediterranean database (Hayes et al., 2005) with modern temperature at 10 m water depth taken from the World Ocean Atlas 1998 (Salgueiro et al., 2014). Seasonal temperatures are averaged for Northern Hemisphere summer (July to September) and winter (December to February). This method, applied on 1212 core-top samples from the Atlantic and the Mediterranean, yields a root mean square error predicted accuracy of ±1.3 ∘C for summer and winter seasons.

Terrestrial proxies. The terrestrial n-alkane concentration varies between 67 and 714 ng g-1 sediment exhibiting no long-term trend across the covered time period from 5.4 to 3.0 ka cal BP (Fig. 3). Three distinct concentration minima below 200 ng g-1 sediment can be observed at 5.4 ka cal BP, from 5.0 to 4.9 ka cal BP, and from about 4.8 to 4.7 ka cal BP. Less distinct minima are observed between 4.4 and 4.3 ka cal BP and at about 3.7 ka cal BP. A sharp increase towards high concentrations of up to 617 ng g-1 sediment is evident at about 4.2 ka cal BP. n-Alkane concentrations appear to remain generally above 450 ng g-1 sediment until 3.8 ka cal BP, when concentrations sharply decrease towards 209 ng g-1 sediment at 3.7 ka cal BP. Afterwards, n-alkane concentrations reveal a slightly increasing trend towards 3.0 ka cal BP. The Norm33 ratio exhibits no trends and varies between 0.29 and 0.49 (Fig. 3). Four sharp increases towards values greater than 0.42 can be recognized at about 5.4 ka cal BP, from 5.0 to 4.9 ka cal BP, from 4.8 to 4.7 ka cal BP, and at about 4.6 ka cal BP. Apparently, the Norm33 maxima parallel the n-alkane concentration minima between 5.4 and 4.7 ka cal BP.

Marine proxies. The alkenone derived annual mean SST remains quite stable between 18.9 and 20.0 ∘C (Fig. 3). No trends are visible in the analysed time period. Only weak maxima in the annual mean SST of 0.5 to 1.0 ∘C in amplitude are apparent at about 5.4, 3.3, and 3.0 ka cal BP as well as between 5.1 and 4.6 ka cal BP at about 100-year intervals (Fig. 4). All foraminifera-derived seasonal SST reconstructions suggest similar stable temperature conditions without any obvious trends between 5.4 and 3.0 ka cal BP. Summer SSTs vary around 22.8±0.1 ∘C with several cooling episodes of up to 1 ∘C at about 5.0, 4.6, 4.3, 3.6, 3.3, and 3.2 ka cal BP. Winter SSTs generally vary around 15.0±0.1 ∘C with weak warming episodes of up to 1 ∘C at about 5.0, 4.6, 4.3, 3.6, 3.3, and 3.2 ka cal BP. Thus, the winter SST maxima and summer SST minima are contemporaneous and result in a decreasing seasonal SST difference during these periods. Notably, the seasonal SST maxima or minima, respectively, do not parallel the SST maxima observed in the annual mean reconstruction. The alkenone concentration varies between about 224 and 440 ng g-1 sediment and shows an increasing trend from 5.4 to 4.0 ka cal BP followed by a decreasing trend towards 3.0 ka cal BP (Fig. 4). These trends are superimposed by several minima from 5.4 to 5.3 ka cal BP; around 5.2, 5.0, and 4.8 ka cal BP; from 4.7 to 4.6 ka cal BP; at 3.8 ka cal BP; and from 3.1 to 3.0 ka cal BP. The alkenone concentration is to some degree correlated (r=0.61) to the annual mean SST across the studied time period (see Supplement) with warmer annual mean SSTs paralleled by decreased alkenones content at about 5.4, 5.2, 4.9, 4.7, and 4.6 ka cal BP, and from 3.1 to 3.0 ka cal BP (Fig. 4).

Terrestrial proxies. The terrestrial n-alkane concentration varies between 9 and 535 ng g-1 sediment and has an increasing trend towards younger ages between 5.4 and 3.0 ka cal BP (Fig. 3). During the studied time period at least three periods of decreasing n-alkane concentration below 100 ng g-1 sediment are found from 5.1 to 4.7 ka cal BP, from 4.4 to 4.3 ka cal BP, and at about 3.7 ka cal BP. The oldest concentration minimum reveals a W-shaped pattern with increasing concentrations of up to 110 ng g-1 sediment at about 4.9 ka cal BP. An additional concentration minimum (46 ng g-1 sediment) at 5.2 ka cal BP is only corroborated by a single data point, and thus not robust. The Norm33 varies without any trends between 0.31 and 0.51 showing high amplitude maxima with values above 0.42 at about 5.0, 4.8, 4.3, and 3.7 ka cal BP (Fig. 3). Notably, the Norm33 reveals maxima during periods of low terrestrial n-alkane concentration.

Marine proxies. Alkenone-derived annual mean SSTs vary without any trends between 5.5 and 3.0 ka cal BP and reveal SSTs around 20.4±0.3 ∘C with the exception of a period from about 4.3 to 3.9 ka cal BP, where SST vary around 21.6±0.6 ∘C with maximum SST being 22.7 ∘C (Fig. 5). Foraminifera-derived summer SSTs reveal a slight decreasing trend towards the present between 6.2 and 1.8 ka cal BP. Notably, between about 6.2 and 4.0 ka cal BP the variability in summer SST is quite high with a mean of 22.4±0.7 ∘C superimposed by several warm events of up to 23.9 ∘C that occur at 6.0, 5.8, 5.5, 4.7, and 4.1 ka cal BP. Also, the variability and the amplitudes of the warm events decrease towards the present. After 4.0 ka cal BP summer SSTs vary around 21.7±0.2 ∘C. Winter SSTs reveal a very similar pattern compared to the summer SSTs with high variability (16.0±0.7 ∘C) until 4.0 ka cal BP and less (16.8±0.2 ∘C) afterwards. Overall, there is a warming trend from about 16 to 17 ∘C towards younger ages with major cooling episodes at 6.0, 5.8, 5.5, 4.7, and 4.1 ka cal BP. Winter SSTs decrease below 15.0 ∘C during most of these periods. The observed SST events in the seasonal reconstructions (summer and winter) are contemporaneous and result in an increasing seasonal difference of up to 9.3 ∘C during these events (Fig. 5). Overall, the seasonal difference is decreasing towards younger ages from 6.4 to 4.7 ∘C. SST events within the seasonal reconstructions do not parallel any events in the alkenone-derived annual mean SST. The warm period observed in the annual mean data between 4.3 and 3.9 ka cal BP is not visible in the seasonal data. Also, the annual mean SST is very close to summer conditions, even exceeding those around 4.2 ka cal BP.

In general, a good agreement between the n-alkane concentration in both sediment cores is apparent (Fig. 3). This is corroborated by the high correlation coefficient (r=0.82) for the period between 5.5 and 3.6 ka cal BP. Since the correlation calculation is based on 100-year time slices (see Supplement), there is not such a robust correlation in the younger part between 3.6 and 3.0 ka cal BP (r=0.04). Nonetheless, visual inspection implies a good agreement between both records when only the more pronounced minima and maxima are considered for this younger interval. The n-alkane concentration minima observed at about 4.3 and 3.7 ka cal BP appear to be contemporaneous in both regions. Also, the two events from about 5.0 to 4.9 ka cal BP and from 4.8 to 4.7 ka cal BP in ODP-161-976A as well as the W-shaped event between 5.1 and 4.7 ka cal BP in GeoB5901-2 correspond to a large extent. Only the drop towards low n-alkane concentrations between 5.1 and 5.0 ka cal BP is offset by about 100 years (Fig. 3). Bearing in mind the chronological uncertainties around 5.0 ka cal BP of ±110 years and ±160 years in the age model of GeoB5901-2 and ODP-161-976A, respectively, this offset is likely chronological bias. The Norm33 maxima of both sediment cores correspond well between 5.4 and 4.7 ka cal BP (Fig. 3).

With respect to absolute SST values and trends in the temperature reconstructions, differences between both cores are larger than for the terrestrial biomarker records. The alkenone-based SSTs as well as the foraminifera-derived winter SSTs were about 1 ∘C warmer in the Gulf of Cádiz, while summer SSTs were about 1 ∘C warmer in the Alboran Sea. The most apparent difference between both regions is noticeable in the seasonal warm/cold events, which do not agree in their timing, as is the same for the much more moderate events observed in the alkenone-based SST records. Also, the overall trend in the difference in seasonal SST estimates is not similar in both regions. While in the Gulf of Cádiz (GeoB5901-2) the seasonal difference increases (i.e. summer warming and winter cooling), the seasonal difference in the Alboran Sea record (ODP-161-976A) does not change significantly.

Terrestrial plants synthesize long-chain n-alkanes in their leaves as coatings for protection against water loss (Eglinton and Hamilton, 1967). These long-chain n-alkanes are either eroded directly from the leaves by wind or deposited in the soils in autumn. Afterwards, n-alkanes are removed from the atmosphere or eroded from the soils by rain and are further transported into the marine realm via aeolian and riverine transport (e.g. Bird et al., 1995; Conte and Weber, 2002; Schreuder et al., 2018). The fraction of the riverine input in coastal areas is usually much higher compared to the aeolian input. Therefore, terrestrial n-alkane concentrations have already been successfully applied to study the riverine input in marine settings in the Mediterranean (Abrantes et al., 2017; Cortina et al., 2016; Jalali et al., 2016, 2017). More specifically, Rodrigo-Gámiz et al. (2015) have shown from radiogenic isotopes that in the Alboran Sea the riverine dominates the aeolian input during the studied time period. Because of the dominant riverine transport mechanism, it is assumed that both proxies integrate spatially over the river catchment areas shown in Fig. 1, and thus indicate climate conditions for the entire southernmost Iberian Peninsula. The dominant catchment for sediment core GeoB5901-2 is the one of the Guadalquivir river draining into the Gulf of Cádiz. The smaller mountainous rivers draining the southern Sierra Nevada area are considered more relevant for sediment core ODP-161-976A. A delivery of material from the Guadalquivir with the inflow of the Atlantic water through the Strait of Gibraltar cannot be excluded though. Since the river discharge and also the plant growing season is strongly coupled to precipitation and the rainy season at the Iberian Peninsula in winter (Fig. 1; Lionello, 2012), the n-alkane proxies are probably biased towards the winter season.

Consequently, intervals of very low n-alkane concentration in the studied cores are interpreted as periods of dry winters, which occurred in southern Iberia at 5.4±0.3 ka cal BP, from 5.1 to 4.9±0.1 ka cal BP, from 4.8 to 4.7±0.1 ka cal BP, from 4.4 to 4.3±0.1 ka cal BP, and at 3.7±0.1 ka cal BP (Fig. 3). Drier conditions at about 5.5 ka cal BP are well in phase with the end of the African Humid Period (deMenocal et al., 2000). This transition appears to be marked by an aridity event as evidenced by speleothem data from El Refugio Cave and Grotte de Piste (Walczak et al., 2015; Wassenburg et al., 2016) and high charcoal concentration in Cabo de Gata between about 5.4 and 5.3 ka cal BP. Also, a contemporaneous remarkable decrease in Mediterranean forest was noticed by Ramos-Román et al. (2018b) from 5.5 to 5.4 ka cal BP. Schröder et al. (2018) describe a drought event centred at about 5.3 ka cal BP in Lake Medina (SW Iberia). The dry phases between about 5.1 and 4.9 and from 4.8 to 4.7 ka cal BP observed in this study are corroborated by regional speleothem records (Moreno et al., 2017; Wassenburg et al., 2016). A dramatic forest decline occurred between 5.0 and 4.5 ka cal BP in SE Iberia (Pantaléon-Cano et al., 2003) along with high charcoal concentrations at Cabo de Gata indicating more frequent wild fires (Burjachs and Expósito, 2015). Moderate forest declines are found in pollen records from the Alboran Sea and Elx sequence (Fig. 6; Burjachs and Expósito, 2015; Fletcher and Sánchez Goñi, 2008). The following period between 4.4 and 3.8 ka cal BP comprising the 4.2 ka event is discussed in more detail in Sect. 4.2. The youngest dry phase found in this study at about 3.7 ka cal BP has also been inferred from pollen data at Elx and Villaverde (Burjachs and Expósito, 2015; Carrión et al., 2016), while pollen data from the Alboran Sea indicate a moderate forest decline (Fletcher and Sánchez Goñi, 2008). Additional evidence for a severe dry phase around 3.7 ka cal BP again stems from speleothem data (Moreno et al., 2017; Walczak et al., 2015).

Notably, in the GeoB5901-2 data all dry episodes are paralleled by Norm33 maxima, while in the ODP-161-976A record the dry phases at about 5.4, 4.9, and 4.7 ka cal BP coincide with Norm33 maxima (Fig. 3). A tendency towards higher n-alkane chain lengths, as revealed by increasing Norm33 values, is an indication for warmer and/or drier environmental conditions (Bush and McInerney, 2013; Leider et al., 2013; Rommerskirchen et al., 2006). A local study from southern Iberia further shows that increasing chain lengths is a function of decreasing water availability (García-Alix et al., 2017). Consequently, the Norm33 maxima observed in this study indicate high water-stress for the plants in relation to the dry climate episodes. This is supported by Sierra Nevada bog sediments Borreguil de la Virgen and Borreguil de la Caldera (Fig. 6; García-Alix et al., 2018), which also show Norm33 maxima during such dry episodes. It has further been shown that Mediterranean forests declined while scrubs expanded in southern Iberia during the respective dry phases (Fletcher et al., 2007; Ramos-Román et al., 2018a, b). Accordingly, the dry episodes caused a noticeable response in the vegetation, which suffered from decreasing water availability. This further implies that the vegetation in southern Iberia at that time was very sensitive to winter precipitation changes.

The manifestation of the 4.2 ka event across the Mediterranean is currently intensively debated with a particular focus on the western part. In the eastern and central Mediterranean many studies suggest a dry phase in the time window between 4.4 and 3.8 ka cal BP (e.g. Calò et al., 2012; Cheng et al., 2015; Finné et al., 2017; Zanchetta et al., 2016). In the western Mediterranean evidence for a dry phase associated with the 4.2 ka event comes from northern Algeria (Ruan et al., 2016) and pollen data from southern Iberia (Ramos-Román et al., 2018a). Furthermore, a drastic forest opening in SE Iberia is indicated from pollen sequences at Cabo de Gata around 4.4 ka cal BP and at Elx at about 4.3 ka cal BP (Fig. 6; Burjachs and Expósito, 2015). Additional indications for a dry phase coinciding with the 4.2 ka event come from lithological analyses in SE Iberia (Navarro-Hervás et al., 2014), a hiatus in pollen data from SW Iberia (Schröder et al., 2018), and speleothem data (Moreno et al., 2017; Walczak et al., 2015; Wassenburg et al., 2016).

Compared to the prolonged dry phase between about 4.4 and 3.8 ka cal BP recorded in speleothems from the Italian Peninsula and Algeria (Ruan et al., 2016; Zanchetta et al., 2016), our data indicate a more complex environmental pattern around the 4.2 ka event. The terrestrial n-alkane concentrations suggest a brief dry period from about 4.4 to 4.3 ka cal BP followed by a rapid return to relatively moister conditions after about 4.2 ka cal BP, which lasted until approximately 3.8 ka cal BP. The Norm33 data, further, shows a peak in the GeoB5901-2 at about 4.3 ka cal BP indicating that plants in the Guadalquivir basin were suffering from water stress. Contrastingly, the intermediate drop in n-alkane concentration in ODP-161-976A from the Alboran Sea is not paralleled by a Norm33 maximum. This might imply that the dry episode between about 4.4 and 4.3 ka cal BP was more severely manifested in the lowlands of the Guadalquivir basin compared to the high altitudes of the Sierra Nevada. Altogether, our reconstructions show that the onset of the dry phase at about 4.4 ka cal BP is in line with proxy data from the central Mediterranean, while the rapid return to relatively moister conditions at about 4.2 ka cal BP lasting until about 3.8 ka cal BP highlights regional differences in the duration of the 4.2 ka event compared to the prolonged dry phase associated with the 4.2 ka event in Italian and Algerian speleothems (Ruan et al., 2016; Zanchetta et al., 2016). This interpretation is corroborated by dry conditions at 4.4 ka cal BP and wet conditions at 4.2 ka cal BP found in ostracod shells from northern Morocco (Zielhofer et al., 2017, 2019).

In contrast to the terrestrial climate variability our marine reconstructions suggest fairly stable conditions between 5.4 and 3.0 ka cal BP in the Alboran Sea and between 6.2 and 1.8 ka cal BP the Gulf of Cádiz. The maximum range in the annual mean values based on alkenones is about 1 ∘C, and thus within the calibration uncertainty. The low temporal resolution of previously studied cores from the area (below about 115 and 250 years for annual mean and seasonal data, respectively) further hampers the comparison with observed SST events. Therefore, a careful interpretation of the alkenone-based SST results is required.

Alkenone-derived SSTs from sediment core ODP-161-976A suggest annual mean temperatures between 18.9 and 20.0 ∘C (Fig. 4), which exceed the modern values of about 18.0 ∘C (Locarnini et al., 2013). An overall cooling trend over the course of the mid- to late-Holocene in response to decreasing insolation (Fig. 5) is known on the regional as well as on the global scale (e.g. Cacho et al., 2001; NGRIP-Members, 2004), thus corroborating warmer SSTs during the mid-Holocene compared to modern SSTs. Annual mean SST values around 19 ∘C at that time were also reconstructed by other studies from the Alboran Sea (Ausín et al., 2015; Cacho et al., 2001; Martrat et al., 2014; Pérez-Folgado et al., 2003; Rodrigo-Gámiz et al., 2014). Moreover, Cacho et al. (2001) described a cold SST event of about 1.5 ∘C in the Alboran Sea between 5.94 and 4.75 ka cal BP. Our data from the Alboran Sea shows no cooling event of similar magnitude between 5.4 and 4.7 ka cal BP. Bearing in mind the error of the calibration there might be a cooling at about 5.3 ka cal BP in the order of just 0.5 ∘C, but surface water conditions between 5.4 and 4.7 ka cal BP appear to have even been warmer compared to the period afterwards. In the Gulf of Cádiz alkenone-based SST values vary between 20.0 to 22.7 ∘C and are warmer compared to recent annual mean temperatures (18.7 ∘C; Locarnini et al., 2013). Higher SSTs during the mid-Holocene might be partly a consequence of higher insolation during this period. However, previous alkenone analyses of Kim et al. (2004) on the same sediment core yielded SST values between 19 and 20 ∘C using a different SST calibration of Prahl et al. (1988), which would result in slightly cooler SSTs for our data as well. Additionally, annual mean SSTs around 20 ∘C in the Gulf of Cádiz were also found by Cacho et al. (2001).

Summer temperatures. Reconstructed summer SSTs based on foraminiferal assemblage variations vary around 22.8 ∘C in the Alboran Sea and around 22.4 and 21.7 ∘C before and after approximately 4.0 ka cal BP, respectively, in the Gulf of Cádiz. Alboran Sea summer SSTs are exceeding modern ones (21.4 ∘C; Locarnini et al., 2013) likely due to higher insolation forcing during the mid-Holocene. This is even better demonstrated by the summer SST record in the Gulf of Cádiz. Here, summer SST estimates are progressively cooling towards mean summer SSTs around 21.7 ∘C, which perfectly agree with modern summer temperatures in the Gulf of Cádiz (21.7 ∘C; Locarnini et al., 2013). Reconstructed Alboran Sea summer SSTs are supported by Rodrigo-Gámiz et al. (2014), who estimated summer SSTs varying around 22 ∘C between 5.5 and 3.0 ka cal BP. Pérez-Folgado et al. (2003) found even warmer summer SSTs between 24 and 25 ∘C in the Alboran Sea. This difference may be a result of the Pérez-Folgado et al. (2003) data calibrated against August temperatures only, while our method used the mean of the summer season (July–September). These authors also found a 0.5 ∘C cooling between about 5.0 and 4.5 ka cal BP, when our data implies two cold events of approximately 1 ∘C at about 5.0 and 4.6 ka cal BP. In the Gulf of Cádiz, in sediment core MD99-2339 (Salgueiro et al., 2014), summer SSTs vary between approximately 22 and 24 ∘C, exhibiting a cooling trend towards the present, which is in harmony with our summer SST reconstructions deduced from GeoB5901-2.

Winter temperatures. Winter SSTs in the Alboran Sea vary around ca. 15.0 ∘C between 5.4 and 3.0 ka cal BP and agree with modern conditions (15.4 ∘C; Locarnini et al., 2013). Moreover, winter SSTs (February) around 14.5 ∘C during the mid- to late-Holocene reported by Pérez-Folgado et al. (2003) support our results. Winter SSTs between 16 and 17 ∘C between 6.2 and 1.8 ka cal BP in the Gulf of Cádiz are also in good agreement with the modern winter SSTs of 16.0 ∘C (Locarnini et al., 2013). Slightly colder winter SSTs between 15 and 16 ∘C were reconstructed from dinocyst assemblages in the neighbouring core MD99-2339 (Penaud et al., 2016).

Under modern conditions the NAO is responsible for much of the variability in winter precipitation at the Iberian Peninsula (Hurrell, 1995; Zorita et al., 1992). Many paleo-climatic studies provide evidence for NAO-like climate variability since the mid-Holocene (Abrantes et al., 2017; Deininger et al., 2017; Olsen et al., 2012). We thus compare the observed sequence of dry climate episodes to a reconstruction of NAO-like variability from Lake SS1220 in Greenland (Fig. 4; Olsen et al., 2012). All dry phases observed in this study can be associated with positive excursions in the NAO reconstruction. Moreover, it is evident that a wetter period between about 4.2 and 3.8 ka cal BP coincides with a period of more stable, less positive NAO-like conditions. Accordingly, we conclude that the dry phases observed between 5.4 and 3.0 ka cal BP in southern Iberia were caused by more dominant positive NAO-like conditions. Yet, other atmospheric circulation patterns like the Scandinavian and the east Atlantic patterns also influence precipitation variability at the Iberian Peninsula (Abrantes et al., 2017; Hernández et al., 2015). This may explain why our records do not reveal dry episodes between 4.7 and 4.5 ka cal BP and at about 3.1 ka cal BP despite the NAO reconstruction suggesting a strong positive circulation mode for these times. Interestingly, at around 5.2, 4.6, and 3.0 ka cal BP we found contemporaneous minima in the MPP and maxima in annual mean SSTs in the Alboran Sea (Fig. 6). This linkage is also visible in the dry phases at about 5.4 ka cal BP and between 5.0 and 4.8 ka cal BP, while during those at about 4.3 and 3.7 ka cal BP the warming of the annual mean SSTs was more pronounced. This linkage was previously described by Ausín et al. (2015) since 7.7 ka cal BP in the Alboran Sea and can also be related to NAO forcing. Currently, the inflow of the Atlantic Waters through the Strait of Gibraltar is strong under negative NAO conditions, and thereby its flow within the Alboran Sea is shifted southwards. This promotes intensified upwelling of cold and nutrient-rich waters, which also result in higher MPP (Sarhan et al., 2000). During positive NAO conditions, on the other hand, the inflow of the Atlantic Waters is weakened resulting in a more stable water column with limited upwelling, and thus low MPP (Sarhan et al., 2000). Despite the calibration uncertainty in the annual mean SST reconstructions, a similar mechanism is also indicated by our data during the mid- to late-Holocene. Accordingly, we conclude that a NAO-like variability was coupled to the terrestrial variability in southern Iberia as well as for the surface water variability in the Alboran Sea between 5.4 and 3.0 ka cal BP.

It has further been suggested that NAO-like variability in the past was responsible for oceanic changes in the subtropical Atlantic such as the position of the Azores Front (Goslin et al., 2018; Repschläger et al., 2017) as well as for increased upwelling and related SST changes along the western Iberian margin (Abrantes et al., 2017). In more detail, Abrantes et al. (2017) suggest that warm winters and cool summers during the Medieval Climate Anomaly were related to positive NAO-like conditions. For the time interval between 6.2 and 1.8 ka cal BP we cannot confirm a general link between the NAO and the hydrological conditions in the Gulf of Cádiz. Interestingly, we observe two notable differences between the hydrological conditions in the Gulf of Cádiz and the Alboran Sea: (1) the summer/winter SST variability in the Gulf of Cádiz is opposite, resulting in a greater seasonality during seasonal SST events, while seasonality appears to decrease during summer/winter events in the Alboran Sea within methodological error, and (2) the observed seasonal warming/cooling events in both adjacent basins are not contemporaneous. This suggests that different mechanisms are driving the hydrological variability in the Gulf of Cádiz and the Alboran Sea. The pronounced seasonal SST variability in the Gulf of Cádiz is in general agreement with the stacked ice-rafted debris (IRD) records for the North Atlantic (Bond et al., 2001; Fig. 5). This implies that larger seasonal SST contrasts in the Gulf of Cádiz relate to periods of enhanced iceberg discharge in the North Atlantic during Bond events 3 and 4. During Bond events, cold polar water masses have been advected as far south as the latitude of Britain (Bond et al., 1997). Our data suggests that this regime resulted in colder winter temperatures in the Gulf of Cádiz as well. Moreover, the Bond events likely weakened the deep water formation in the North Atlantic by limiting the oceanic heat transport towards the north (Bond et al., 2001; Wanner et al., 2011). Repschläger et al. (2017) further proposed that early-Holocene freshwater forcing from the Laurentide ice sheet may have resulted in a weak subsurface heat transport from the subtropical gyre (STG) towards the North Atlantic, and subsequently in a subsurface heat storage within the STG. Following these mechanisms, we hypothesize that during Bond events 3 and 4 the northward heat transport was blocked due to freshwater forcing in the North Atlantic, which in turn results in a heat storage within the STG. During summer, when the Intertropical Convergence Zone (ITCZ) moved northwards, these heated water masses probably reached the Gulf of Cádiz via the Azores Current resulting in warmer summer temperatures. These warmer water masses could also reach the Alboran Sea through the Strait of Gibraltar, but are not recorded within our data because summer temperatures in the Alboran Sea were generally warmer by about 1 ∘C during the studied period. Interestingly, it seems that annual mean SSTs in the Gulf of Cádiz warmed by up to 2 ∘C during Bond event 3 (Fig. 5). This might rather imply a growing season shift of the alkenones producing coccolithophores from spring to summer rather than a warming of the annual mean SSTs. Such a reaction of the coccolithophores might be also indicated to a lesser extent during Bond event 2 when annual mean SSTs warmed by approximately 1 ∘C at about 3.2 ka cal BP. Our data does not cover the whole interval though. Also, Bond event 4 is not entirely covered so that this interpretation remains hypothetical. But all in all, the influence of cold water from the North Atlantic during winter along with the influence of warmer subtropical water masses during summer likely resulted in a pronounced seasonality during Bond events 3 and 4 (Fig. 5). Notably, the amplitude of the seasonal events and their accordingly high seasonality as well as the overall seasonality in the Gulf of Cádiz were decreasing towards the present. It is, moreover, interesting to note that Bond event 2 is not visible in our seasonal SST records at all (Fig. 5). The general decrease in seasonality in the Gulf of Cádiz can be attributed to decreasing summer insolation. We hypothesize that this is also true for the decreasing amplitude of the seasonal SST events. During Bond event 4, when summer insolation was high, the more northward position of the ITCZ during summer allowed for a stronger inflow of warmer subtropical water masses into the Gulf of Cádiz. On the other hand, during winter the ITCZ was much further south compared to its present position, allowing enhanced southward flow of the colder water masses from the north. During Bond event 3 the summer and winter positions of the ITCZ were already less extreme, thus weakening the influence of either warm and cold water masses during summer or winter, respectively. Later, during Bond event 2, seasonal movements of the ITCZ were even more limited such that no influence of either cold water masses during winter or warm subtropical water masses during summer is recognizable by a pronounced seasonal temperature difference in the Gulf of Cádiz.