Along the Peruvian margin the main source for the high amounts of upwelled nutrients (30 µmol L-1 for both Si(OH)4 and NO3-; Bruland et al., 2005) is the subsurface Peru-Chile Undercurrent (PCUC), which flows southward along the continental slope and outer shelf between 4 and 14∘ S at a depth between 50 and 150 m, before it detaches from the shelf south of 15∘ S (Brink et al., 1983; Chaigneau et al., 2013; Toggweiler et al., 1991). Eastward-flowing subsurface waters of the Equatorial Undercurrent (EUC) and the Southern Subsurface Counter Current (SSCC) (see Fig. 1) feed the PCUC. These subsurface currents deliver Si(OH)4 and NO3- with mean preformed source signatures for δ30Si(OH)4 of 1.5±0.2 ‰ (Beucher et al., 2011; Ehlert et al., 2012; Grasse et al., 2013) and δ15NO3- of 7.1±0.3 ‰ (1 SD; Rafter et al., 2012; Rafter and Sigman, 2016) for the EUC. Within the SSCC preformed δ15NO3- values of 5.5±0.3 ‰ (Rafter et al., 2012) are about 1.6 ‰ lower than the EUC, resulting in an approximate average PCUC value of ∼6 ‰ (Fig. 2a; Mollier-Vogel et al., 2012).

The dissolved δ15NO3- of subsurface waters (50–150 m water depth) increases southward (Equator, EQ, to 17∘ S; Mollier-Vogel et al., 2012) as a consequence of water column denitrification, while the dissolved δ30Si(OH)4 signature remains close to the source value of 1.5 ‰ for the PCUC (Fig. 2a; Ehlert et al., 2012). This difference in the evolution of the isotopic signature from north to south is caused by the anoxic conditions off the coast of Peru only increasing the δ15NO3- signatures via denitrification in the subsurface but not affecting the δ30Si(OH)4 signatures. Accordingly, at the northern shelf between 1∘ N and 10∘ S, where subsurface O2 concentrations [O2] are >20 µmol L-1, N-loss is not observed and the δ15Nbulk values in the sediments range between 4 ‰ and 5 ‰, close to the δ15NO3- source value of 6 ‰, thus indicating a high degree of NO3- utilization (Fig. 2b; Mollier-Vogel et al., 2012). In contrast, the δ30SiBSi signatures north of 10∘ S are more variable, reflecting an overall lower degree of Si(OH)4 utilization (Doering et al., 2016; Ehlert et al., 2012). At the central Peruvian shelf (10–12∘ S), where subsurface [O2] is <20 µmol L-1 (Fig. 2a), the subsurface source value of δ15NO3- increases to 8.6 ‰ due to denitrification (Mollier-Vogel et al., 2012). The δ30SiBSi and δ15Nbulk values both increase as a consequence of higher Si(OH)4 utilization but decrease in NO3- utilization compared to the northern part of the study area (Fig. 2b), which reflects the interplay between increased upwelling intensity, high nutrient resupply and higher consumption via diatom productivity. In the southernmost part of the shelf (13–16∘ S), the highest productivity and upwelling intensity prevail today, leading to a further increase in the subsurface δ15NO3- signature of up to 12.5 ‰ at 15∘ S, whereas surface sediment mean δ30SiBSi and δ15Nbulk values further increase, reflecting moderate utilization of both Si(OH)4 and NO3- (Fig. 2a–b). The supply of dissolved Si(OH)4 strongly increases from the northern shelf to the southern shelf area (Fig. 2c), reflecting the strength of the upwelling conditions. This increase in upwelling and productivity between 10 and 15∘ S results in high accumulation rates of BSi (0.4–0.6 g cm-2 yr-1; Ehlert et al., 2012) and total nitrogen (TN, 0.026–0.035 g cm-2 yr-1; Mollier-Vogel et al., 2012) in the sediment (based on accumulation rates of Gutierréz et al., 2009). However, the NO3- supply, as indicated by subsurface (50–150 m) NO3- concentrations in the water column, slightly decreases from north to south, reflecting the loss of NO3- via denitrification.

The new data in this study were obtained from three short, fine-laminated trigger cores retrieved from the main upwelling region off the Peruvian margin during the German R/V Meteor cruise M77/2 in 2008 as part of the Collaborative Research Center (SFB) 754 (Fig. 1). New records of δ30SiBSi and BSi concentrations were generated for cores M77/2-024-5TC (024-5TC; 11∘05′ S, 78∘00′ W, 210 m water depth), M77/2-005-3TC (005-3TC; 12∘05′ S, 77∘40′ W, 214 m water depth) and core M77/2-003-2TC (003-2TC; 15∘06′ S, 75∘41′ W, 271 m water depth). Single centimeter slices of the sediment cores were sampled for BSi and Si isotope measurements to ensure the availability of sufficient amounts of diatoms for silicon isotope analysis (Table 1). For core 003-2TC, additional BSi concentration measurements of material extracted from individual laminations was possible (Fleury et al., 2015). As previously published δ15Nbulk values are based on samples from single laminations, these were averaged to 1 cm resolution when directly compared to the δ30Si data in the following. Core locations are shown in Fig. 1. The age models were published before in Fleury et al. (2015). The age models for all cores are given in years CE.

The amount of BSi in the sediments was measured following an automated leaching method using sodium hydroxide (DeMaster, 1981; Müller and Schneider, 1993) with a precision of 1 %–2 % (1 SD). Unfortunately, no material was left of the cores studied here to estimate dry bulk densities to calculate mass accumulation rates (MARs). Therefore, MAR values were used from nearby cores BO413 (12∘ S) and BO406 (14∘ S; Gutierréz et al., 2009), which were generally close to 0.02 (g cm-2 yr-1) during the LIA and 0.03 g cm-2 yr-1 during the CWP. The exact bulk MAR values (g cm-2 yr-1) for each time period were multiplied by the fractional concentration of BSi and TN (Fleury et al., 2015) to calculate the MAR BSi and MAR TN (Figs. 2c and 6).

For the Si isotope measurements diatoms were extracted from the sediment by chemical and physical cleaning (11 and 32 µm sieve; heavy liquid separation with a sodium polytungstate solution set at 2.15 g mL-1), as described in detail in Ehlert et al. (2012, 2013) and Doering et al. (2016). For all samples, the purity of the small diatom fraction (11–32 µm) was evaluated via light microscopy prior to dissolution and only pure (>95 %) diatom samples were treated further. All samples were dissolved in 1 mL 0.1 M NaOH and treated with 200 µL concentrated H2O2 (Suprapur). Sample solutions were diluted with 4 mL MQ water and neutralized with 0.1 mL 1 M HCl (Reynolds et al., 2008), followed by a chromatographic purification using 1 mL pre-cleaned AG50W-X8 cation exchange resin (BioRad, mesh 200–400) (de Souza et al., 2012). The Si isotopic compositions were determined in 0.6 ppm sample solutions on a NuPlasma HR MC-ICPMS at GEOMAR applying a standard-sample bracketing method (Albarède et al., 2004). Silicon isotopic compositions are reported in the δ-notation relative to the reference standard NBS28 in parts per thousand: δ30Si=((Rsample/Rstandard)-1)⋅1000, where Rsample is the 30Si/28Si ratio of the sample and Rstandard is the 30Si/28Si ratio of the NBS28. All δ30Si measurements were run at least in triplicate, with uncertainties ranging between 0.05 ‰ and 0.27 ‰ (2 SD). Repeated measurements of an in-house diatom matrix standard gave average δ30Si values of 1.03±0.21 ‰ (2 SD n=15). Long-term repeated measurements of the reference materials NBS28, IRMM018 and Big Batch gave average δ30Si values of 0.00±0.24 ‰ (2 SD), -1.40±0.21 ‰ (2 SD, n=15) and -10.60±0.24 ‰ (2 SD, n=15), respectively, in good agreement with literature values (Reynolds et al., 2007).

Diatom analysis of cores M77/2-024-5TC, 005-3TC and 003-2TC were published previously based on three slides per sample and counting of a minimum of 300 valves for each sample (for details see Fleury et al., 2015). The diatom abundances are presented here for three groups representing different environmental conditions (Fig. 4e–g): upwelling species – Chaetoceros sp., Skeletonema costatum, Thalassionema nitzschioides var. nitzschioides; coastal planktonic species – Actinocyclus spp., Atinoptychus spp., Asteromphalus spp., and Coscinodiscus sp.; and other diatom species – Nitzschia spp., Rhizosolenia spp. and Thalassiosira spp., Cyclotella spp., and Cocconeis sp.;

The diatom assemblage abundance is compared to δ30SiBSi compositions for cores M77/2-024-5TC, 005-4TC and 003-2TC to investigate if changes in the assemblage have influenced the isotopic record. While diatom counts have been performed on bulk sediment samples δ30SiBSi was measured on the 11–32 µm size fraction. However, it was shown previously that this size range closely resembles the main assemblage, which allows for studying the influence of changes in the diatom assemblage on the δ30SiBSi record (Ehlert et al., 2012, 2013).

The degree of nutrient utilization can be described assuming either Rayleigh-type (single input followed by no additional nutrients newly supplied to a particular parcel of water followed by fractional loss as a function of production and export) or steady-state (continuous supply and partial consumption of nutrients causing a dynamic equilibrium of the dissolved nutrient concentration and the product) fractionation behavior (Mariotti et al., 1981). For simplification purposes we will only provide the values derived from steady-state fractionation, which was shown to better reflect upwelling conditions off the coast of Peru (Ehlert et al., 2012). %Si(OH)4consumed=1-δ30Si-δ30SiOH4source/30ε⋅1001%NO3consumed-=[1-δ15N-δ15NO3-/15ε]⋅100, with %Si(OH)4consumed or %NO3consumed- being the percentages of the supplied Si(OH)4 and NO3- that have been utilized. For this calculation we apply enrichment factors of -1.1 ‰ 30ε (δ30Si, (De La Rocha et al., 1997) and -5 ‰ 15ε (δ15N) and assume a constant source water signature of 1.5 ‰ for δ30Si(OH)4source (i.e., the mean δ30Si(OH)4 of the PCUC). The nutrient utilization for surface sediments calculated here is identical to the original publications (Fig. 2b; Mollier-Vogel et al., 2012; Ehlert et al., 2012). To evaluate the impact of changes in 30ε on the δ30Si signatures the potential influence of species-specific fractionation was tested based on the impact of a -2.1 ‰ enrichment factors of Chaetoceros brevis (Sutton et al., 2013). However, the estimated impact on past δ30SiBSi records, due to a change in the amount of Chaetoceros sp. present in the sediment was less than 5 % for all cores (M77/2-024-5TC, 005-3TC and 003-2TC) and thus did not alter the assumed 30ε of -1.1 ‰ substantially (based on calculations presented in Doering et al., 2016; not shown). The impact of denitrification on the δ15NO3- signatures of the past is assessed in the following section before calculating past NO3- utilization for the respective latitudes.

Based on modern observations from the water column it is known that NO3- and Si(OH)4 are incorporated in a 1:1 ratio when diatoms dominate the phytoplankton assemblage (Brzezinski, 1985; Ragueneau et al., 2000). The ratio of nutrients in the water column can, however, vary between 2:1 and 1:2 on the shallow Peruvian shelf (Grasse et al., 2016). Assuming a strict 1:1 uptake ratio of nutrients, the respective δ30SiBSi and δ15NBulk values in the underlying sediments should also reflect a 1:1 ratio (indicated by white star “1” in Fig. 3a). Based on the known δ30Si(OH)4 and NO3- source signatures of modern subsurface waters, we can calculate the actual nutrient utilization (see Sect. 2.4) and estimate the uptake ratio for NO3-:Si(OH)4 (Fig. 3b). However, it is not possible to observe a significant correlation for the entire shelf area, given that there are only a few data points for the areas along the shelf (Figs. 2, 5a). We calculate past nutrient utilization and estimate the influence of denitrification on the δ15Nbulk values based on the fact that on the shelf δ30SiBSi and δ15Nbulk values generally follow a positive linear regression (Figs. 3b, 5a). In order to estimate past changes in the δ15NO3- source values, the δ30SiBSi and δ15Nbulk values were separately plotted against each other for the time periods of the CWP, the arid LIA and the humid LIA (Fig. 5b–d). Accordingly, the δ15NO3- source value for each period was calculated based on the linear function assuming that the source δ30Si(OH)4 signature always remained stable at 1.5 ‰ over time: 2δ30Si(OH)4=a⋅δ15NO3-+b, or 3δ15NO3-=δ30SiOH4-b/a, with a indicating the slope of the line and b the intercept. For δ30Si(OH)4 we used the value of 0.4 ‰ representing near 0 % utilization (= source water δ30Si(OH)4 of 1.5 ‰ of the PCUC – 1.1 ‰ fractionation during uptake) to estimate the δ15NO3- source. Accordingly, the values estimated by Eq. (3) represent the δ15NO3- source value also assuming near 0 % utilization of NO3-.

We calculated the linear regression based on all samples of the different cores from the different latitudes (11, 12, 14 and 15∘ S) during the CWP and the LIA. We also further differentiated between short-term productive phases (arid phases) and the generally prevailing humid El Niño-like conditions during the LIA (grey shadings in Fig. 4) and resolved the resulting equation based on Eq. (2) to estimate δ15NO3-. Only for the LIA (humid) phases was it not possible to directly calculate δ15NO3- values based on the linear function from Eq. (2), due to near horizontal alignment of the δ15Nbulk versus δ30SiBSi values (Fig. 5b). Therefore, for this time period the highest δ15NBulk value for each latitude was assumed to reflect ∼100 % utilization and was thus used as δ15NO3- source value. This assumption might slightly overestimate the maximum utilization, which is only ∼80 % today (Fig. 2b), and therefore might underestimate the source value slightly. For all time periods and latitudes, the linear regressions as well as correlation coefficient (r2) are given in the Supplement (Fig. S1). The results are presented in the following as the resulting δ15NO3- source values and the theoretical ratio of nutrient utilization (i.e., 1:1 or 2:1, 15:1, etc.; Fig. 5b–d) for each latitudinal range to compare the latitudinal trends between the CWP and the LIA.

Based on these calculated subsurface δ15NO3- values, we further calculated the change in nutrient utilization as well as nutrient supply for the different latitudes. Past nutrient utilization was calculated following Eq. (1). Given the estimate of nutrient demand and export productivity it is further possible to estimate changes in the supply with the relationship Δsupply=Δdemand/Δutilization by applying the equation of Horn et al. (2011) given by 4Nutrient supply=FBSi/TNsampleFBSi/TNpresent%nutrientconsumedsample/% nutrientconsumedpresent, Fopal/TN is the flux of BSi or TN and %nutrientconsumed is the percent of the Si(OH)4 or NO3- supply consumed (i.e., nutrient utilization). Given that there are no accumulation rates available for either the surface sediment samples or for any of the cores studied here to directly determine the export productivity directly, we used the BSi and TN values previously published (surface sediments; Mollier-Vogel et al., 2012; Ehlert et al., 2012) and the new BSi values presented in this study, together with mass accumulation rates (g cm-2 yr-1) for cores BO406-13 and 406-5 from Gutierréz et al., 2009, to calculate the accumulation rates of BSi and TN (MAR; g cm-2 yr-1). For the different time periods mean values for MAR BSi and MAR TN were calculated and the respective nutrient supply was calculated based on Eq. (4), indicating changes in the nutrient supply compared to modern values.

The data of the sediment cores from the shelf area between 12 and 15∘ S presented here show an increase in BSi content from mean values of 13 %–23 % during the LIA to values of 21 %–29 % during the CWP. The δ30Si records follow a similar trend of lower mean δ30SiBSi values of 0.8±0.2 ‰ (2 SD, 12∘ S), 0.8±0.1 ‰ (14∘ S) and 1±0.2 ‰ (15∘ S) during the LIA to more variable and higher mean values of 1.3±0.4 ‰ (12∘ S), 0.8±0.4 ‰ (14∘ S) and 1.5±0.2 ‰ (15∘ S) during the CWP (Fig. 3a–d; Table 1).

The diatom assemblages (Fig. 4e–g; based on Fleury et al., 2015) show an association of the amount of upwelling species and δ30SiBSi signatures, with decreases of up to 20 % in upwelling species often accompanied by a reduction of δ30SiBSi by about 0.5 ‰–1 ‰. However, not every decrease in δ30SiBSi is mirrored by a change in the diatom assemblage and vice versa (e.g., Fig. 4f at 1650 CE). Overall, the diatom assemblage data indicate little change in the mean conditions and a slight reduction of upwelling strength at 12 and 15∘ S during the LIA in comparison to the CWP (Fig. 6). The most distinct shift of lower abundances of upwelling species (∼50 %) to higher values during the CWP (∼70 %) is found at 15∘ S (003-2TC) corresponding to the strongest changes in BSi and δ30SiBSi at this location.

The sedimentary BSi concentrations and δ30SiBSi signatures at 12∘ S (005-3TC) and 15∘ S (003-2TC) were lowest during the LIA (Fig. 4c, e), in agreement with previously published records from 11∘ S (M77/1-470; Fig. 4a) and 14∘ S (Ehlert et al., 2015; Fig. 4d). An exception is core 024-5TC (Fig. 4a) from 11∘ S, where δ30Si mean values of the LIA (1.3±0.4 ‰) are similar to CWP mean values (1.4±0.1 ‰). Furthermore, both the BSi concentrations and δ30SiBSi signatures of core 024-5TC were significantly higher during the LIA than at nearby core M77/1-470 (Fig. 4a; Ehlert et al., 2015). However, comparison with the cumulative diatom assemblage indicates overall little difference in the amount of upwelling and coastal planktonic diatom species between the LIA and the CWP at 11∘ S (Fig. 4f), with intervals of reduced abundances of upwelling species generally lasting less than 50 years, much shorter than the 100 to 150 years intervals observed at 12 and 15∘ S. Furthermore, the finely laminated sediment layers do indicate short periods of higher productivity during the LIA in a phase with more arid conditions (Fig. 4, grey shading; for details see Fleury et al., 2015). Accordingly, the high mean BSi and δ30SiBSi values obtained from core 024-5TC may be an artifact of low sampling resolution with only two δ30SiBSi samples representing the time period between 1700 and 1800 CE and δ30SiBSi analyses not evenly covering all the short events (∼50 years) of reductions in the abundance of upwelling diatom species (Fig. 4f). Alternatively, the increase in Si(OH)4 utilization, decoupled from an increase in diatom abundance (Fleury et al., 2015; not shown here) may indicate stronger silicification of the diatom frustules, as often observed under Fe-deficient conditions and associated with an increase in the Si(OH)4:NO3- incorporated by the diatoms (De La Rocha et al., 2000; Takeda, 1998; Wilken et al., 2011).

As previously shown, the δ15Nbulk values of the three cores (M77/2-024-5TC, 005-3TC and 003-2TC) presented in this study were on average 0.8 ‰ lower during the LIA than during the CWP (Fleury et al., 2015). The δ15Nbulk values reported for core 005-3TC (12∘ S) are close to values of nearby core B0406-13 (Gutiérrez et al., 2009). Similarly, the δ15Nbulk values of core 003-2TC (15∘ S) agree well with previously published δ15Nbulk record of core B0405-6 (14∘ S, Fig. 4j, k; Gutiérrez et al., 2009).

During the humid phases of the LIA the calculated δ15NO3- source values were lower, reaching values of 6 ‰ between 11 and 12∘ S and 7.5 ‰ between 14 and 15∘ S (Figs. 5b, 6a). The calculated NO3- utilization was higher during this time, reaching values between 70 % and 90 %, while Si(OH)4 utilization ranged between 6 % and 60 %. The MAR TN was lowest (<0.02 g cm-2 yr-1) during the LIA, however, with little difference between humid and arid phases (Fig. 6a, c, right side). The MAR BSi values were similar to today during the LIA (humid) ranging between 0.2 and 0.5 g cm-2 yr-1 (Figs. 2, 6a right side). The calculated NO3- supply was lowest during the LIA (humid) ranging between 0.3 and 0.7, with little change over latitude in accordance with the prevalence of more oxygenated waters, whereas the Si(OH)4 supply strongly increased from 0.5 to 3.8 at 12∘ S (Fig. 6a).

During the CWP the calculated δ15NO3- source signatures based on Eqs. (2) and (3) result in values of 7.6 ‰ at 11∘ S, 8.6 ‰ at 12∘ S and 10.4 ‰ between 14 and 15∘ S during the CWP (Figs. 5c, 6b, S1 in the Supplement (Please confirm.)), which reflects a southward increase in δ15NO3- source signatures as observed today (Fig. 2a). Based on these δ15NO3- values the nutrient utilizations estimated based on Eq. (1) range between 30 % and 90 % for NO3- and 40 %–100 % for Si(OH)4 (Fig. 6b). During the LIA (arid) similar values are calculated with a δ15NO3- mean value of 8 ‰ between 11 and 12∘ S, increasing to a value of 9 ‰ between 14 and 15∘ S (Fig. 6c). The respective nutrient utilization ranges between 2 % to 70 % for NO3- and 20 % to 85 % for Si(OH)4. The MAR TN were lower during the CWP by about 0.02 (g cm-2 yr-1) than today and MAR BSi values were generally higher by about 0.1–0.35 (g cm-2 yr-1). The calculated Si(OH)4 supply indicates a slight increase compared to today (as indicated by positive values) but has remained rather stable around 1 over all latitudes, while NO3- supply also indicates values of ∼1 at 11 and 15∘ S but the supply increased to 2 at 12∘ S (Fig. 6b, right side). During the LIA (arid) MARs of TN and BSi were both lower in comparison to the CWP, ranging between 0.014 and 0.017 (g cm-2 yr-1) and 0.25–0.7 (g cm-2 yr-1), respectively. The Si(OH)4 supply was similarly stable, as observed during the CWP, but slightly higher ranging from 0.6 to 1.55, while the NO3- supply was lower and decreased from north to south from 1 to 0.3.

Recent evidence shows that a cool EEP plays a key role in climate change due to its linkage to a slowdown in global warming (England et al., 2014; Kosaka and Xie, 2013), highlighting the importance of understanding Pacific climate variability in the past (Rustic et al., 2015). The last millennium has been divided into warmer global conditions over the Medieval Warm Period (MWP), colder temperatures over the LIA and rising temperatures since the beginning of the CWP (Mann et al., 1999). The transition between the MWP and the LIA (∼1150 to 1500 CE) has been associated with an anomalous strong zonal SST gradient and with transitional Northern Hemisphere (NH) cooling into the LIA, as evidenced by cooler SSTs at Galápagos (Rustic et al., 2015). After ∼1500 CE, the EEP cooling trend ended and local SSTs began to increase until around 1600 CE, when an anomalous weak zonal gradient was established when the EEP temperatures reversed from cooling to warming. This reversal occurred when the NH descended into the coldest part of the LIA and persisted throughout most of the LIA resulting in an extended El Niño-like mean state (Mann et al., 2009). Evidence links the ITCZ to hemispheric warming and cooling cycles, implying southward ITCZ displacements during NH cold periods (Chiang and Bitz, 2005; Schneider et al., 2014). Accordingly, a southward shift of the ITCZ during the MWP-LIA transition has been proposed for the Atlantic and Pacific (Haug et al., 2001; Peterson and Haug, 2006; Sachs et al., 2009). The El Niño-like conditions during the LIA have been associated to a gradual intensification of the fluvial input of sediments to the continental shelf, as reflected by an increase in the terrigenous sediment flux (Briceño-Zuluaga et al., 2016; Gutiérrez et al., 2009; Sifeddine et al., 2008) and changes in the radiogenic isotopic composition of the terrigenous fraction due to changes in the provenance and material transport (Ehlert et al., 2015), as well as better oxygenation and a lower productivity in the Peruvian upwelling area (Gutiérrez et al., 2009; Salvatteci et al., 2014b; Sifeddine et al., 2008). Accordingly, most of the LIA, i.e., the humid phases, have been characterized by low productivity and weak denitrification intensity between 10 and 15∘ S (Díaz-Ochoa et al., 2009; Salvatteci et al., 2014b; Sifeddine et al., 2008), which is supported by the absence of a significant southward increase in the source value of δ15NO3- reconstructed from our records (Figs. 5b, 6a). Correspondingly, high δ15Nbulk and little change in reconstructed NO3- supply indicate more complete NO3- utilization during the LIA (humid), while δ30SiBSi signatures and utilization remained low and Si supply high (Fig. 6a). This indicates a shift towards a dominance of non-siliceous phytoplankton productivity causing NO3- limitation and low uptake of Si. This is in agreement with modern conditions during El Nino events for which physical-biogeochemical models, together with in situ and satellite observations (1958–2008), have shown that the temperatures and sea level increase, the thermocline and nutricline deepens, and the phytoplankton (mainly diatoms) and nutrient concentration decrease along the Peruvian coast (Espinoza-Morriberón et al., 2017). Coastal trapped waves propagating along the coast can seasonally increase the depth of thermocline and nutricline, decreasing the NO3- vertical flux into the surface layer. The NO3- and Fe content of the upwelling source waters may also strongly decline (Espinoza-Morriberón et al., 2017). Our calculations show that phases of lower productivity during so-called El Niño-like conditions during the LIA (humid) have a δ15NO3- source delivered to the Peru upwelling area similar to today (Fig. 6a), but due to less or no denitrification δ15NO3- source does not increase southward (Figs. 5b, 6a). Instead δ15NBulk values are mainly affected by the variability of NO3- concentrations, which seem to be the limiting factor for primary productivity (PP) similar to that observed during El Niño events today (Espinoza-Morriberón et al., 2017). The stronger stratification due to deepening of the nutricline potentially does not allow for similarly efficient N remineralization (or N is transported offshore due to eddy activity, Espinoza-Morriberón et al., 2017) and may result in NO3- to be utilized more strongly than Si(OH)4 (Fig. 7a). Accordingly, the NO3- supply was diminished while Si(OH)4 was still available.

Finely laminated sediment from the LIA from the oxygen minimum zones (OMZs), which have also been used in this study, resolve multidecadal variations in precipitation over the continent, and of variations in detrital and biogenic fluxes in relation to precipitation and upwelling intensity (Briceño-Zuluaga et al., 2016, Díaz-Ochoa et al., 2009; Fleury et al., 2015, 2016; Salvatteci et al., 2014a). Stable oxygen isotope compositions of individual planktic foraminifera point to greater ENSO activity (high frequency between alternating La Niña and El Niño conditions) in the EEP based on records from the Galápagos (Rustic et al., 2015). It was shown by coupled models that such multi-decadal variation in ENSO amplitude can arise from episodic strengthening and weakening of the thermocline feedback (Borlace et al., 2013). The difference we observed in the isotopic evolution of nutrients (δ15N and δ30Si) between the arid and humid phases during the LIA support the development of multidecadal phases of prevailing La Niña- or El Niño-like conditions. Similar interannual variance has been observed based on organic carbon and carbonate proxies during the LIA within the California current system, which the authors related to large ENSO events in contrast with an apparent reduction in such variability during the CWP (Abella-Gutiérrez and Herguera, 2016).

This El Niño-like mean state appears to have ended at the beginning of the CWP (Rustic et al., 2015). Evidence for increasing precipitation off the coast of Panama after 1700 CE likely reflects the northward shift of the ITCZ (Linsley et al., 1994) from its more southerly LIA position. During the CWP the OMZ intensified and marine productivity increased together with surface temperature cooling and an increase in terrigenous material input (Briceño-Zuluaga et al., 2016; Gutiérrez et al., 2011). We find that the CWP and the LIA (arid) are characterized by high upwelling intensity, productivity, and N-loss processes (Fleury et al., 2015; Salvatteci et al., 2014b; Sifeddine et al., 2008) and are associated with southward increasing δ15NO3- source signatures caused by denitrification, reflecting moderate NO3- utilization and moderate to high Si(OH)4 utilization (Fig. 7b). The highest δ30SiBSi and utilization values at 15∘ S are potentially caused by progressive Fe limitation during diatom blooms, causing a NO3-:Si(OH)4 ratio of up to 1:2. Southward increasing δ15Nbulk values and calculated δ15NO3- demonstrate the consistent incorporation of higher isotopic compositions due to subsurface denitrification under anoxic subsurface conditions, in agreement with decreasing NO3- supply illustrating the N-loss process.