The B040506 (hereafter “B06”; 14∘07.90′ S, 76∘30.10′ W; 299 m water depth) and the G10-GC-01 (hereafter “G10”; 14∘22.96′ S, 076∘23.89′ W; 313 m water depth) sediment cores were retrieved from the central Peruvian continental shelf in 2004 during the Paleo2 cruise onboard the Peruvian vessel José Olaya Balandra (IMARPE) and in 2007 during the Galathea-3 cruise, respectively (Fig. 1a). We compared the age models and performed a laminae cross-correlation between the two cores in order to develop a continuous record for the last millennium (Salvatteci et al., 2014a) (Fig. S1 in the Supplement). The choice of these two cores was based on previous detailed stratigraphic investigations and available complementary multi-proxy reconstructions (Gutiérrez et al., 2006, 2009; Salvatteci et al., 2012, 2014a, b; Sifeddine et al., 2008). The box core B06 (0.75 m length) is a laminated core with a visible slump at ∼ 52 cm and three thick homogeneous deposits (1.5 to 5.0 cm thick) identified in the SCOPIX images. These intervals were not considered in our study (Fig. S1). The presence of filaments of the giant sulfur bacteria Thioploca spp. in the top of core B06 confirms the successful recovery of the sediment water interface.

According to the biogeochemical analysis in Gutiérrez et al. (2009) (i.e., palynofacies, oxygen index (Rock-Eval), total organic carbon and δ13C), B06 is characterized by a distinctive shift at ∼ 30 cm, more details are provided by Sifeddine et al. (2008) and Salvatteci et al. (2014a). The age model of B06 was inferred from five 14C-calibrated accelerator mass spectrometry (AMS) age distributions (Fig. S1), showing that this core covers the last ∼ 700 years. For the last century, which is recorded only by B06, the age model was based on downcore natural excess 210Pb and 137Cs distributions and supported by bomb-derived 241Am distributions (Fig. S2 and Gutiérrez et al., 2009). The mass accumulation rate after ca. AD 1950 was 0.036 ± 0.001 g cm-2 yr-1 and before ca. AD 1820 was 0.022 ± 0.001 g cm-2 yr-1. On the other hand, G10 is a gravity-laminated sediment core of 5.22 m presenting six units and exhibiting some minor slumping. The G10 age model was based on 31 samples of 14C-calibrated AMS age distributions, showing that the core covers the Holocene period (Salvatteci et al., 2014b, 2016). Here we used only a laminated section between ∼ 18 and 45 cm that chronologically covered part of the MCA period (from ∼ AD 1050 to 1500) and presented no slumps (Fig. S1).

The spatial regularity of the initial core sampling combined with the naturally variable sedimentation rate implied variable time rates between samples (150 samples in total). Each sample is 0.5 cm thick in B06 and usually includes 1–2 laminae. On the other hand, in core G10, each sample is 1 cm thick, including 3–4 laminae. The results considering the sedimentation rates showed that the intervals during MCA, LIA and CWP span 18, 7, and 3 years, respectively. Because of differences in the subsampling thickness between cores and variable sedimentation rates, results are binned by 20-year intervals (the lowest time resolution among samples) after linear interpolation and 20-year running mean of the original data set.

To isolate the mineral terrigenous fraction, organic matter, calcium carbonate and biogenic silica were successively removed from approximately 100 mg of bulk sediment sample using H2O2 (30 % at 50 ∘C for 3 to 4 days), HCl (10 % for 12 h) and Na2CO3 (1 M at 90 ∘C for 3 h), respectively. Between each chemical treatment, samples were repeatedly rinsed with deionized water and centrifuged at 4000 rpm until neutral pH. After pre-treatment, the grain size distribution was determined with an automated image analysis system (model FPIA3000, Malvern Instruments). This system is based on a CCD (charge-coupled device) camera that captures images of all of the particles homogeneously suspended in a dispersal solution by rotation (600 rpm) in a measurement cell. After magnification (×10), particle images are digitally processed and the equivalent spherical diameter (defined as the diameter of the spherical particle having the same surface as the measured particle) is determined. The optical magnification used (×10) allows the counting of particles with equivalent diameters between 0.5 and 200 µm. Prior to the FPIA analysis, all samples were sieved with a 200 µm mesh in order to recover coarser particles. Since particles > 200 µm were never found in any samples, the grain size distribution obtained by the FPIA method reliably represents the full particle size range in the sediment. A statistically significant number of particles (in the hundreds of thousands, up to 300 000) are automatically analyzed by FPIA, providing particle size information comparable to that obtained with a laser granulometer along with images of the individual particles. Using the images to check the efficiency of the pre-treatments, we ensured that both organic matter and biogenic silica had been completely removed from all the samples. Finally, particle counts were binned into 45 different size bins between 0.5 and 200 µm instead of the 225 set by the FPIA manufacturer in order to reduce errors related to the presence of very few particles in some of the preselected narrow bins. Grain size distributions are expressed as (%) volume distributions.

As different particle transport/deposition processes are known to influence the grain size distribution of the lithic fraction of sediment (e.g., Holz et al., 2007; Pichevin et al., 2005; Prins et al., 2007; Stuut et al., 2005, 2002; Sun et al., 2002; Weltje and Prins, 2003, 2007; Weltje, 1997), identifying the individual components of the polymodal grain size distribution is decisive for paleoenvironmental reconstructions. The numerical characteristics (i.e., amplitude (A), geometric mean diameter (Gmd), and geometric standard deviation (Gsd) of the individual grain size populations whose combination forms the overall grain size distribution) were determined for all samples using the iterative least-squares method of Gomes et al. (1990). This fitting method aims to minimize the squared difference between the measured volume grain size distribution and the one computed from a mathematical expression based on lognormal function. The number of individual grain size populations to be used is determined by the operator, and all statistical parameters (A, Gmd and Gsd) are allowed to change from one sample to another. This process presents a strong advantage compared to end-member modeling (e.g., Weltje, 1997), in which the individual grain size distributions are maintained constant over the whole time series, the only fitting parameter being the relative amplitude, A. Indeed, it is unlikely that the parameters that govern both transport and deposition of lithogenic material, and therefore grain size of particles, remain constant over time. In turn, variations in these parameters are expected to induce changes in the grain size distribution parameters such as Gmd and Gsd.

Both sediment cores (B06 and G10) exhibit a roughly bimodal grain size distribution presenting significant variation in amplitude and width. These modes correspond to fine-grain-size classes from ∼ 3 to 15 µm and coarser grain size classes between ∼ 50 and 120 µm (Fig. S3). A principal component analysis (PCA) based on the Wentworth (1922) grain size classification identifies four modes that could explain the total variance of the data set (Fig. S4). The measured and computed grain size distributions show high correlations ranging from R2= 0.75 to 0.90, demonstrating that the use of four grain size modes is well adapted to our sediment samples and that the computed ones may be reliable for further interpretation (Fig. 2). Lower correlations only occurred for six samples that are characterized by small proportions of terrigenous material compared to biogenic silica, organic matter and carbonates. In these cases, the number of lithic particles remaining after chemical treatments was small, which increased the associated relative error. However, these samples have been included in the data set since they all presented a high contribution of coarser particles.

Grain size parameters are presented in Table 1. The first mode (M1), with a Gmd of approximately 3 ± 1 µm, and the second one (M2), with a Gmd of 10 ± 2 µm, are characterized by large Gsd (∼ 2σ), indicating a low degree of sorting. Such a low degree of sorting suggests a slow and continuous depositional process as occurs in other environments (Sun et al., 2002). The coarsest modes, M3 and M4, display mean Gmd values of 54 ± 12 and 91 ± 13 µm, respectively. These modes present Gsd values close to 1σ. The Gmd values of the two coarsest modes are consistent with the optimal grain size transported under conditions favorable to soil erosion (lack of vegetation, low threshold friction velocity, surface roughness and low soil moisture) and low wind friction velocity (Iversen and White, 1982; Kok et al., 2012; Marticorena and Bergametti, 1995; Marticorena, 2014; Shao and Lu, 2000). Such conditions prevail in the studied area because central coastal Peru consists of a sand desert area characterized by the absence of rain, a lack of vegetation and persistent wind (Gay, 2005; Haney and Grolier, 1991).

In the vicinity of desert areas, where wind-blown transport prevails, particles with grain size as high as ∼ 100 µm can accumulate in marine sediments (e.g., Flores-Aqueveque et al., 2015; Stuut et al., 2007) or even in lacustrine sediments (An et al., 2012). Indeed, Stuut et al. (2007) reported the presence of distributions typical of wind-blown particles with ∼ 80 µm grain size (∼ 29∘ S North Chile), which is consistent with our results. In the studied area, the emission and transport of mineral particles are related to the strong wind events called Paracas. Paracas dust emission is a local seasonal phenomenon that preferentially occurs in winter (July–September) and is due to an intensification of the local surface winds (Escobar Baccaro, 1993; Haney and Grolier, 1991; Schweigger, 1984). The pressure gradient of sea level between 15 and 20∘ S, 75∘ W is the controlling factor of Paracas winds (Quijano, 2013), along with local topography (Gay, 2005). Coarse particles found in continental sediments off Pisco cannot have a fluvial origin because substantial hydrodynamic energy is necessary to mobilize particles of this size (50–100 µm), and this region is devoid of large rivers (Reinhardt et al., 2002; Scheidegger and Krissek, 1982; Suess et al., 1987).

Therefore, the coarsest modes (M3 and M4) can be interpreted as markers of aeolian transport resulting from surface winds and emission processes (Flores-Aqueveque et al., 2015; Hesse and McTainsh, 1999; Marticorena and Bergametti, 1995; McTainsh et al., 1997; Sun et al., 2002) and indicate a local and proximal aeolian source (i.e., Paracas winds). This interpretation is in contrast to the Atacama Desert source suggested by Ehlert et al. (2015) and Molina-Cruz (1977). Ehlert et al. (2015), who used the same sediment core (B06), and also indicated difficulties in the interpretation of the detritical Sr isotopic signatures as an indicator of the terrigenous sources. These difficulties can be associated with the variability in the 87Sr / 86Sr due to grain size (Meyer et al., 2011). The finest M1 component (∼ 3 µm) may be linked to both aeolian and fluvial transport mechanisms. Thus, because its origin is difficult to determine, and because its trend appears to be relatively independent of the other components, we do not use it further.

The M2 component (∼ 10 µm) is interpreted as an indicator of fluvial transport (Koopmann, 1981; McCave et al., 1995; Stuut and Lamy, 2004; Stuut et al., 2002, 2007). Indeed, this is consistent with the report by Stuut et al. (2007) for the fluvial mud (∼ 8 µm) in the south of Chile (> 37∘ S), where the terrigenous input is dominated by fluvial origins. A fluvial origin of this M2 component is also supported by showing the same trend in the geochemical proxies, such as radiogenic isotope compositions of detrital components (Ehlert et al., 2015), mineral fluxes (Sifeddine et al., 2008) or %Ti (Salvatteci et al., 2014b), indicating more terrigenous transport during the LIA, when humid conditions were dominant. The M2 component is interpreted as being linked to river material discharge, mostly from the north Peruvian coast, and redistribution by the PCUC and bottom currents (Montes et al., 2010; Rein et al., 2004; Scheidegger and Krissek, 1982; Unkel et al., 2007).

Grain size component (M2, M3 and M4; Table 1) variations in the composite records (B06 and G10) express changes in wind stress and fluvial runoff at multidecadal to centennial scales during the last millennium. The sediments deposited during the MCA exhibit two contrasting patterns of grain size distributions. A first sequence dated from AD 1050 to 1170 has low values of D50 (i.e., median grain size) that vary around 16 ± 6 µm and are explained by 50 ± 14 M2, 16 ± 8 M3, 21 ± 5 M4 and 13 ± 5 % M1 contributions. A second sequence, dated from AD 1170 to 1450, was marked by high values of D50 in the range of 34 ± 18 µm, with average contributions of 36 ± 8 for M2, 21 ± 10 for M3, 29 ± 15 for M4 and 14 ± 6 % for M1 (Table 2). These results indicate high variability in transport of particles during the MCA, with more fluvial sediment discharge from 1050 to 1170, followed by an aeolian transport increase between AD 1170 and 1450 (Fig. 3).

During the LIA (AD 1450–1800), the deposited particles were dominated by fine grain sizes with a D50 varying around an average of 15 µm, explained by 53 ± 15 % M2 contribution. In contrast, the contribution of M3 averaged 19 ± 9 % and ranged from 4 to 45 %, whereas M4 showed an average contribution of 14 ± 11 % and varied from 0 to 44 % during the same period. The dominant contribution of the finest-sized particles of M2 suggests a high fluvial terrigenous input to the Peruvian continental shelf. It is important to note that M2 contributions increased from the beginning to the end of the LIA at ∼ AD 1800, suggesting a gradual increase in fluvial sediment discharge input related to the enhancement of the continental precipitation (Fig. 3c). Indeed, during the LIA, our results confirm previous interpretations of wet conditions along the Peruvian coast (Gutiérrez et al., 2009; Salvatteci et al., 2014b; Sifeddine et al., 2008). These results also imply that this period was characterized by weak surface winds and hence a weaker coastal upwelling.

Subsequently, D50 variations show multidecadal variability during the last ∼ 200 years that is divided into three distinctive periods. The first one, from ∼ AD 1800 to 1850, shows dominance of coarse particles around 50 µm, explained by the high contribution of M3 and M4 (up to 45 and 50 %, respectively) during this period. These results suggest a period of drier climate and very strong wind conditions. The second one, from AD 1850 to 1900, displays values around ∼ 20 µm explained by ∼ 40 of M2, ∼ 20 of M3 and ∼ 20 % of M4, suggesting that fluvial sediment discharge was the dominant transport mechanism, although not as significant as during the LIA. The third period spans from AD 1900 to the final part of record and covers the CWP. Our results reveal a dominance of coarse particles during the most of this period (D50 up to of 80 µm) that arise from high contributions of M3 and M4 (∼ 40 and ∼ 50 %, respectively). However, a clear decrease in the D50 is displayed at the end of this period that is explained by a decrease in contributions of the aeolian component M4 (∼ 20 %), although the contribution of M3 and M2 remain relatively stable (∼ 25 and 30 %, respectively). These conditions display no clear dominance of a given transport mode during this time. In addition, markedly coarser particles in the M4 component were very common during this time (the last 200 years), indicating a strong probability of extreme wind stress events (Fig. 3f).

Our findings suggest a combination of regional and local atmospheric circulation mechanism changes that controlled the pattern of sedimentation in the study region. Our record is located under the contemporary seasonal Paracas dust storm path, but it also records discharged fluvial muds that are supplied by the rivers along the Peruvian coast. Hence, this record is particularly well suited for a reconstruction of continental runoff/wind intensity in the central Peruvian continental shelf during the last millennium. The interpretation of the changes in the single records of the components (M2, M3 and M4) and their associations (e.g., ratios) can reflect paleoclimatic variations in response to changes in atmospheric conditions. Here, we used the ratio between the aeolian components, defined as the contribution of the stronger winds over total wind variability: M4 / (M3 + M4). We consider this ratio to be a proxy of the local wind surface intensity and thus of the SPSH atmospheric circulation (Fig. 4a). Previous studies have similarly and successfully used grain size fraction ratios as paleoclimate proxies of atmospheric conditions and circulation to explain other sediment records (Holz et al., 2007; Huang et al., 2011; Prins, 1999; Shao et al., 2011; Stuut et al., 2002; Sun et al., 2002; Weltje and Prins, 2003).

As explained above, the MCA was characterized by a sine-like peak structure that depicts two different climate stages. During the first stage, spanning from ∼ AD 1050 to 1170, the fluvial input show a peak centered at AD 1120 that was linked to a precipitation increase accompanied by a decrease in wind intensity. Those results suggest a southward ITCZ displacement (Fig. 4e) as a response to more El Niño-like conditions as suggested by Rustic et al. (2015) (Fig. 4g and h). In contrast, during the second stage the surface winds had their greatest intensity with a peak centered at ∼ AD 1200 as a consequence of displacement of the ITCZ–SPSH system. The displacement of the SPSH core towards the eastern South American coast intensified alongshore winds as a regional response to stronger Walker circulation. These features are in agreement with the ocean thermostat mechanism proposed by Clement et al. (1996). This mechanism produces a shallow thermocline in the eastern Pacific (Fig. 4g and h) and consequently more intense upwelling conditions and a stronger OMZ offshore of Pisco recorded in low values of the Re / Mo ratio (Fig. 4d). These two patterns (i.e., enhanced fluvial transport/enhanced wind intensity) might have been triggered by the expression of Pacific variability at multidecadal timescales with the combined action of the Atlantic Multidecadal Oscillation (AMO). Indeed, other works provide evidence during the MCA for low South American monsoon system (SAMS) activity at multidecadal timescales driven by the AMO (Fig. 4f) (Apaéstegui et al., 2014; Bird et al., 2011; Reuter et al., 2009). Thus, besides the displacement of the ITCZ, the AMO could have modulated Walker circulation at a multidecadal variability during the MCA through mechanisms such as those described by McGregor et al. (2014) and Timmermann et al. (2007).

Our results, combined with other paleo-reconstructions, suggest that the LIA was accompanied by a weakening of the regional atmospheric circulation and of the upwelling favorable winds. During the LIA, the mean climate state was controlled by a gradual intensification of the fluvial input of sediments to the continental shelf, thus indicating more El Niño-like conditions (Fig. 4b). These features are confirmed by an increase in the terrigenous sediment flux, as described by Sifeddine et al. (2008) (Fig. 4c) and Gutiérrez et al. (2009) and by changes of the radiogenic isotopic composition of the terrigenous fraction (Ehlert et al., 2015). These wet conditions are also marked by an intensification of the SAMS and the southern meridional displacement of the ITCZ, as evidenced by paleo-precipitation records in the Andes and in the Cariaco Trench (Apaéstegui et al., 2014; Haug et al., 2001; Peterson and Haug, 2006) (Fig. 4e). At the same time, a prevalence of weak surface winds (Fig. 4a) and an increase in subsurface oxygenation driving sub-oxic conditions in the surface sediment are recorded (Fig. 4d). These characteristics also support the hypothesis of the ITCZ–SPSH southern meridional displacement and are consistent with a weakening of the Walker circulation (Fig. 4g).

The transition period between the LIA and CWP appears as an abrupt event showing a progressive positive anomaly in the wind intensity synchronous with a rapid decrease in fluvial input to the continental shelf (Fig. 4a and b). This transition suggests a rapid change of meridional (ITCZ–SPSH) and zonal (Walker) circulation interconnection, which controls the input of terrigenous material (fluvial/aeolian). Gutiérrez et al. (2009) found evidence of a large reorganization in the tropical Pacific climate with immediate effects on ocean biogeochemical cycling and ecosystem structure at the transition between the LIA and CWP. The increase in the regional wind circulation that favors aeolian erosive processes simultaneously leads to an increase in the OMZ intensity related to upwelling intensification.

Finally, during the CWP (∼ AD 1900 to present), a trend to steadying of low fluvial input (Fig. 4b) was combined with an increase in wind intensity (Fig. 4a) that was coupled to a strong OMZ. This setting suggests the northernmost ITCZ–SPSH system position. This hypothesis is supported by other studies on the continental shelf of Peru (Salvatteci et al., 2014b) and also in the eastern Andes, where a decrease in rainfall of between ∼ 10 and 20 % relative to the LIA was reported for the last century (Reuter et al., 2009). Enhancement of wind intensity is also consistent with the multidecadal coastal cooling and increase in upwelling productivity since the late nineteenth century (Gutiérrez et al., 2011; Salvatteci et al., 2014b; Sifeddine et al., 2008) and confirms the relations between the intensification of the upwelling activity induced by the variability in the regional wind intensity from SPSH displacement.

The increase in the wind intensity over the past two centuries likely represents a result of the modern positioning of the ITCZ–SPSH system and the associated intensification of the local and regional winds (Fig. 4a). The contributions of aeolian deposition material (Fig. 3e and f) and, as a consequence, the wind intensity and its variability during the last 100 years are stronger than during the second sequence of the MCA (Fig. 4a) under similar conditions (i.e., position of the ITCZ–SPSH system). This variability implies a forcing mechanism in addition to the enhancement of the wind intensity, one that may be related to the current climate change conditions (Bakun, 1990; England et al., 2014; Sydeman et al., 2014). Moreover, during the CWP, the wind intensity showed a direct relation with OMZ strength (Fig. 4a and d) that suggests an increase in the zonal gradient and thus in the Walker circulation on a multidecadal scale.

Our record shows that on a centennial scale, the fluvial input changes are driven by the meridional ITCZ position and a weak gradient of the Walker circulation, consistent with El Niño-like conditions. In contrast, variations in the surface wind intensity are linked to the position of the SPSH modulated by both the meridional variation in the ITCZ and the intensification of the zonal gradient temperature related to the Walker circulation and expressing La Niña-like conditions. A clear relation between the zonal circulation and wind intensity at a centennial timescale is displayed. All these features modulate the biogeochemical behavior of the Peruvian upwelling system.