Solar Forcing of ENSO on Century Timescales

Understanding how El Niño‐Southern Oscillation (ENSO) responds to natural variability is of key importance for future climate projections under a warming climate. However, there is no clear consensus on what drives ENSO's variability on centennial timescales. Here, we find that the epikarst in southeastern Alaska is effective at filtering ENSO and solar irradiance signals from the Aleutian Low regional climate, which are subsequently recorded in speleothem proxy data. By applying a correlation test, we find that ENSO was significantly influenced by solar irradiance over the past ∼3,500 years. This relationship dissolved after ∼1970 CE, with ENSO now being dominated by anthropogenic forcing. This implies a new ENSO mean state that will need to be incorporated into future climate projections.

Regions of significant correlation are highlighted by color bands (Pearson's correlation [90% CI]).Black box indicates NINO3.4 region, while the speleothem sample location of this study is marked by the green star.Refer to Figure S2 in Supporting Information S1 for a more detailed map of the study area.This plot was generated using Climate Reanalyzer (http://cci-reanalyzer.org), Climate Change Institute, University of Maine, USA.(Alexander et al., 2002;Bjerknes, 1966Bjerknes, , 1969)).Sea-surface temperatures in the equatorial Pacific, in turn, may be influenced by solar irradiance (Clement et al., 1996;Emile-Geay et al., 2007), which would impact the mean-state of ENSO (Clement et al., 1996;Emile-Geay et al., 2007), and consequently affect the strength of the Aleutian Low via the atmospheric bridge.It has been shown that stronger solar irradiance is generally associated with a weaker Aleutian Low and vice versa (Osterberg et al., 2014).Therefore, there is a known link between both solar irradiance and ENSO and the strength of the Aleutian Low, typically on decadal timescales.
Based on the robust teleconnection between the equatorial Pacific and the North Pacific, we hypothesize that solar irradiance forces ENSO mean state changes which, in turn, force the strength of the Aleutian Low via the atmospheric bridge.In other words, decreased solar irradiance should correspond to an increased frequency of El Niño events and result in an overall strengthened Aleutian Low.Conversely, increased solar irradiance should correspond to an increased frequency of La Niña events and result in an overall weakening of the Aleutian Low.This hypothesis is in agreement with the ocean thermostat mechanism (Clement et al., 1996;Emile-Geay et al., 2007).
Speleothems, which provide long-term, high-resolution reconstructions of climate in the region (Wilcox et al., 2019), offer an untested approach to record the response of changes in the strength of the Aleutian Low, and hence ENSO/solar irradiance, responses in the North Pacific.Here, we utilize speleothems from southeastern Alaska to generate a high-resolution and precisely dated record spanning continuously the past ∼3,500 years.Our data demonstrate that the local epikarst in southeastern Alaska is effective at filtering ENSO and solar irradiance signals from the Aleutian Low regional climate.From this, we find that solar forcing has been the primary driver of ENSO variability on centennial timescales, and that this relationship dissolved at ∼1970, likely due to anthropogenic forcing.

Site Location and Samples
Our data sets were developed from two stalagmites retrieved in spring/summer 2021 in two caves on Prince of Wales Island, located in the temperate rainforest of the southern Alexander Archipelago in Alaska.Klawock, the nearest village to the caves (Figure S2 in Supporting Information S1), has a mean annual air temperature of 7.4°C and receives ∼2000 mm of precipitation annually.Speleothem WB-21-5-A is 536 mm in length and was found 50 m inside Wishbone Cave (55.776°N, 133.195°W; 350 m a.s.l.), and WA-21-6-A is 181.5 mm in length and was found 100 m inside Walkabout Cave (55.774 N, −133.191W; 420 m a.s.l.) (Figure S3 in Supporting Information S1)."Hendy" tests (Hendy, 1971) were performed at four different locations at 2.3, 6.5, 9.9, and 15 cm in WA-21-6-A, and five different locations at 4, 13.8, 25.6, 37.1, and 48 cm in WB-21-5-A to test for isotope equilibrium fractionation (Figure S4 in Supporting Information S1).The fabric texture was identified visually, and shows columnar fabric throughout speleothems, with no evidence of hiatuses.
Interior cave temperatures are a constant 5.6°C for Walkabout Cave, and vary between 2.5 and 8.6°C for Wishbone Cave (Figure S5 in Supporting Information S1).Both speleothems were actively dripping during recovery, suggesting that the speleothem tops are modern.There are no visually detectable hiatuses, and growth rates are constant for both speleothems (Figure S6 in Supporting Information S1).Speleothems WB-21-5-A and WA-21-6-A were sampled for δ 18 O and δ 13 C at 0.5 and 0.25 mm resolution, respectively, producing a temporal resolution of ∼2-5 years for both speleothems.Additionally, speleothem WB-21-5-A was sampled for fluid inclusion δD every 0.5 cm, producing a temperature record with a resolution of ∼40 years (Figure S7 in Supporting Information S1).

U-Th Ages
A total of 10 powdered calcite samples were manually drilled for U-Th dating under a laminar flow hood; 5 from WA-21-6-A, and 5 from WB-21-5-A (Figure S3 in Supporting Information S1) (Wilcox et al., 2022).U-Th samples were processed at the University of Minnesota Trace Metal Isotope Geochemistry Lab and analyzed using a ThermoFisher Neptune Plus multi-collector inductively coupled plasma mass spectrometer equipped with an Aridus desolvation nebulizer, following the method of Shen et al. (2012).Ages are reported with 2σ errors in BCE/CE.A time-depth model was created in OxCal 4.4 using the Bayesian approach (Bronk Ramsey, 2008, 2009;Bronk Ramsey & Lee, 2013).

Stable Isotopes
A total of 1800 stable isotope locations were drilled using a Merchantek micromill.In WA-21-6-A, samples were drilled every 0.25 mm, yielding a temporal resolution of ∼5 years.Samples in WB-21-5-A were drilled every 0.5 mm, yielding a temporal resolution of ∼2-5 years.Stable isotope samples were analyzed at the University of Innsbruck using a ThermoFisher Delta V isotope ratio mass spectrometer equipped with a Gasbench II (Spötl, 2011).Stable isotopes are reported in per mil relative to Vienna Peedee Belemnite (VPDB).Long-term analytical precision is less than or equal to 0.08‰ for both δ 13 C and δ 18 O (1σ).

Fluid Inclusions
Speleothem fluid inclusion water isotopes were analyzed at the University of Innsbruck using a continuous-flow technique via high-temperature reduction on glassy carbon (Dublyansky & Spötl, 2009).δD fi isotope ratios are given in per mil (‰) using the standard delta notation and are reported relative to the Vienna Standard Mean Ocean Water.We extracted 95 calcite blocks from 87 different depths, weighing between 1 and 1.5 g, from the central growth axis of stalagmite WB-21-5-A.Replicates were produced at 0.5, 10.5, 20.5, 30.5, and 40.5 cm depth.See (Dublyansky & Spötl, 2009) for details on the crushing procedure.The precision of replicate measurements of our in-house calcite standard is typically 1.5‰ for δD fi for water amounts between 0.1 and 1 μl.Because crushing of our calcite samples released up to 1 μl of water (mean 0.40 μl), the precision of 1.5‰ for δD fi was found to be adequate for this study.Temporal resolution is ∼40 years.
The paleotemperature record of stalagmite WB-21-5-A was reconstructed based on the modern-day regional water isotope-temperature relationship (Rozanski et al., 1992).Only δD fi values were used for calculating paleotemperatures for the following reasons: post-depositional processes can alter the original δ 18 O fi in fluid inclusion water and thus limit the use of δ 18 O fi for paleotemperature calculations (McDermott, 2004).In addition, δD fi is not affected by isotopic fractionation during calcite precipitation and remains unaltered as there is no hydrogen source once the water is entrapped in the calcite matrix.We used the global meteoric water line (δD = 8*δ 18 O + 10‰) to convert δD fi to δ 18 O calculated .Modern-day drip-water yielded a δ 18 O value of −10 ‰ which was used as the modern-day δ 18 O anchor point.Fluid inclusion δ 18 O calculated values were subtracted from this modern-day δ 18 O anchor point to obtain δ 18 O difference .Next, a temperature-δ 18 O transfer function (TF) was used to convert δ 18 O difference into temperature.Because it is unclear which TF is appropriate, we evaluated a range of possible values, between 0.26 and 0.36 ‰/°C, which represents the error range of the south-central Alaska temperature-δ 18 O slope of 0.31 ‰/°C (Bailey et al., 2019).Because there is a minor 0.1°C difference in temperatures calculated from the range of TF values, we report temperatures based on the TF of 0.31 ‰/°C.Finally, we subtracted the mean annual temperature of a nearby weather Station in Klawock (55.555°N, 133.096°W; 24 m a.s.l.-Figure S2 in Supporting Information S1) of 7.4°C (Western Regional Climate Center) to obtain paleotemperature anomaly values: A regional lapse rate likely causes cooler mean annual temperatures at the cave sites ∼400 m higher in elevation, probably closer to the interior cave temperatures of Walkabout Cave (∼5.6°C) (Figure S5 in Supporting Information S1).In lieu of using unavailable long-term site-specific temperature data, we report temperature data as anomalies versus weather station Klawock (Figure 3; Figure S7 in Supporting Information S1).
Uncertainties reflect isotope measurement errors, and one standard deviation of repeated measurements.The uncertainties are applied through all steps of the paleotemperature calculation.Further, uncertainties are propagated between sampling locations.

Cave Monitoring
HOBO Pro v2 temperature loggers (accuracy: ±0.21°C) were placed near the entrance of both Walkabout and Wishbone Caves, and near the extracted stalagmites.Temperature was measured at 1-hr intervals for 1 year, starting when the stalagmites were extracted (Figure S5 in Supporting Information S1).
A Stalagmate drip logger was placed directly where stalagmite WA-21-6-A was extracted, and recorded drip counts at 1-hr intervals for 1 year, starting when the stalagmite was extracted (Figure S5 in Supporting Information S1).

Statistical Analyses
For correlation estimation on two time series that are not observed on the same timescale, the "binned correlation coefficient" was used (Mudelsee, 2014).This measure equals Pearson's correlation coefficient calculated on the respective averages within time bins of the two-time series.The optimal binwidth was obtained using the software TAUEST (Mudelsee, 2002), the temporal spacings of the two series, and the binwidth formula after (Mudelsee, 2014).The uncertainty of the estimated binned correlation coefficients was determined by calibration of 90% Student's t confidence intervals using the software PearsonT3 (Ólafsdóttir & Mudelsee, 2014).It should be noted that a confidence interval (for an estimation) is an uncertainty measure that is superior to a p-value (for a hypothesis test) because it carries more quantitative information.That means a confidence interval is for a statement about the strength of an association, not merely whether there is one (Efron & Tibshirani, 1993;Yates, 1951).Furthermore, the suitability of this correlation method to climate data and its validity have been demonstrated by means of observed and artificial series (Mudelsee, 2014).
The spectra for the records were estimated on the detrended time series obtained with a Gasser-Müller nonparametric trend (Mudelsee, 2014) calculated with a bandwidth of 500 years in order to exclude distortions from long-term variations.The spectral power was determined by means of the Lomb-Scargle Fourier transform combined with Welch's overlapped segment averaging procedure (Mudelsee, 2014), which is implemented in REDFIT software (Schulz & Mudelsee, 2002).To determine a frequency-dependent correct factor for estimation bias stemming from the uneven spacings, we conducted 10,000 Monte Carlo simulations of an AR(1) red-noise process

Controls on Speleothem δ 18 O
A pseudo amount effect in combination with changing source regions are likely the dominant controls on δ 18 O in precipitation at the cave sites, as indicated by a regional comparison of modern precipitation (Figure S9 in Supporting Information S1).Essentially, a strengthened Aleutian Low will draw more moisture meridionally from southern regions (Liu & Alexander, 2007), which has relatively low δ 18 O values (Bailey et al., 2019).A strengthened Aleutian Low will also contribute to higher rainfall amounts at the cave sites, resulting in more depleted δ 18 O values, and hence, a pseudo amount effect.On the other hand, a weakened Aleutian Low will draw an increased fraction of moisture from zonal regions, which has relatively high δ 18 O values (Bailey et al., 2019).A weaker Aleutian Low will also cause lower rainfall amounts at the cave sites, resulting in more enriched δ 18 O values.We can reasonably exclude topographic barriers as dominant controls on δ 18 O in precipitation as there are no topographic barriers between the cave sites and the Pacific Ocean to cause isotopic depletion.Further, there is a weak relationship between changes in air temperature and δ 18 O of precipitation in southcentral Alaska, with surface air temperatures explaining only ∼30% of variability in the δ 18 O precipitation data (Bailey et al., 2019).Although there are no networks monitoring isotopes in southeastern Alaska, the region as a whole is strongly influenced by the Aleutian Low (Bailey et al., 2019) and likely has a similar weak relationship between air temperature and δ 18 O of precipitation.Therefore, we argue that δ 18 O of calcite is dominantly controlled by the amount of precipitation in combination with the source region, with depleted isotope values indicating more precipitation and a southerly derived source region, and vice versa.This is corroborated by drip rate data from the site of speleothem WA-21-6-A, which closely mirrors local precipitation amounts (Figure S5 in Supporting Information S1).
A pseudo amount effect and changing source regions controlling δ 18 O in precipitation at the cave sites implies that speleothem oxygen isotopes are a reliable proxy to determine the strength of the Aleutian Low.Depleted speleothem δ 18 O indicates a strengthened Aleutian Low, and vice versa, consistent with regional lake proxy data (Anderson et al., 2005) and ice core data (Osterberg et al., 2014).To test if ENSO and solar irradiance signals can be extracted from speleothem δ 18 O, we applied the binned correlation coefficient (r) to find reliable statistical linkages among pairs of proxy data on unequal timescales (Mudelsee, 2014) (see methods for details on the statistics).First, we examine correlations between speleothem data, and then apply the correlation test with solar irradiance and ENSO proxy data.

Speleothem Proxy Correlations
For speleothem WB-21-5-A, we find that the fluid inclusion temperature reconstruction correlates significantly with δ 18 O at r = 0.59 with a 90% calibrated bootstrap confidence interval of [0.43; 0.71] (Figure S10 in Supporting Information S1).The δ 18 O series from speleothems WA-21-6-A and WB-21-5-A are significantly correlative, r = 0.25 [0.01; 0.45] (Figure S10 in Supporting Information S1), highlighting regional consistencies that would be expected between neighboring speleothem δ 18 O data that precipitated close to isotopic equilibrium with the drip water (Figure S4 in Supporting Information S1).We argue that deviations between speleothem δ 18 O are a result of spectral frequency differences between sites (Figure S8 in Supporting Information S1), with speleothem WB-21-5-A recording a high-frequency δ 18 O signal and WA-21-6-A recording a low-frequency δ 18 O signal (Figure S11 in Supporting Information S1).This is likely the result of the large difference in growth rate (Figure S6 in Supporting Information S1), and hence drip rate, between the two sites.Since drip water transports the δ 18 O signal from the soil zone to the speleothem site, different drip rates will lead to different δ 18 O frequencies at each speleothem site.We hypothesize that the local epikarst acts as a low-pass filter, resulting in a modulation of the Aleutian Low regional climate signal (Figure 2).In speleothem WA-21-6-A, we observe a climate signal that is potentially linked with total solar irradiance (TSI), while in speleothem WB-21-5-A, we observe a climate signal that is potentially linked with ENSO.We then explore the statistical significance of these potential climate linkages to determine if they are robust.

Solar Irradiance and ENSO Correlations
To determine if the δ 18 O series from speleothem WA-21-6-A is statistically linked with TSI, we compared the δ 18 O proxy to the most up-to-date physics-based reconstruction of TSI currently available (Wu et al., 2018) (Figure 3).We find that TSI correlates significantly with speleothem WA-21-6-A δ 18 O after ∼0 CE at r = 0.52 10.1029/2023GL105201 7 of 11 [0.32; 0.68] (Figure S10 in Supporting Information S1), with decreased solar irradiance correlating with a strengthened Aleutian Low in speleothem WA-21-6-A, and vice versa.However, we find no significant correlation prior to ∼0 CE.The lack of correlation for the earlier part could be due to proxy bias, with all previous TSI reconstructions dependent on 14 C and 10 Be data.Additionally, we performed a spectral analysis, which shows solar cycles (Moussas et al., 2005) at all 3 spectral peaks of 16-, 19-, and 28-year periods (Figure S8 in Supporting Information S1), providing enhanced confidence that speleothem WA-21-6-A reliably records TSI variability.However, due to the lack of correlation prior to ∼0 CE, we only provide interpretations for the past ∼2000 years.
To determine if the δ 18 O series from speleothem WB-21-5-A is statistically linked with ENSO, we compared the δ 18 O proxy to an ENSO reconstruction produced from a coral δ 18 O record in the central tropical Pacific spanning the past ∼750 years (23) (Figure 3).We find that the coral δ 18 O (Dee et al., 2020) is significantly correlated with speleothem WB-21-5-A at r = 0.54 [0.10; 0.80] (Figure S10 in Supporting Information S1), with increased frequency of El Niño events correlating with a strengthened Aleutian Low in speleothem WB-21-5-A, and vice versa for La Niña events (Figure 3).Additionally, we performed a spectral analysis, which shows the "classical" ENSO power (Allen, 2000) at the 7.2-year period (Figure S8 in Supporting Information S1), which is added confidence that speleothem WB-21-5-A reliably records ENSO variability.
The correlation tests confirm that the strength of the Aleutian Low correlates significantly with either solar irradiance or ENSO.And, since speleothem WA-21-6-A δ 18 O (solar irradiance signal) and WB-21-5-A δ 18 O (ENSO signal) are significantly correlated, this provides compelling evidence that solar irradiance can influence ENSO mean state changes.More specifically, it agrees with our hypothesis that solar irradiance forces ENSO mean state changes which, in turn, force the strength of the Aleutian Low via the atmospheric bridge (Figure 2).Our data thus supports the ocean thermostat mechanism (Clement et al., 1996;Emile-Geay et al., 2007) before 1970 CE.After 1970 CE, ENSO deviates from natural variability based on the divergent trends of temperature and precipitation (Figure 3).

Discussion
Our data shows that solar forcing can have an influence on ENSO mean state changes on centennial timescales over the past 2,000 years.Notably, this confirms the ocean thermostat mechanism whereby solar irradiance induces changes in the east-west temperature gradient of the tropical Pacific and, hence, ENSO activity (Clement et al., 1996;Emile-Geay et al., 2007).In general, periods of increased solar irradiance correspond to an increased frequency of La Niña events, while decreased solar irradiance corresponds to an increased frequency of El Niño events (Figure 3).Only the El Niño mean state change at ∼500 CE does not fully agree with solar irradiance (Figure 3), and may be associated with the strongest volcanic eruption in the past 2,500 years (Sigl et al., 2015).This event is in conjunction with one of the most extensive regional glacial advances in the past 2000 years (Sigl et al., 2015), and may correspond to the Late Antique Little Ice Age, whereby unprecedented summer cooling is recorded by tree rings in Europe (Büntgen et al., 2016).Therefore, we find that ENSO mean state changes are mostly insensitive to volcanic eruptions, consistent with coral reconstructions (Dee et al., 2020), except for exceedingly rare super-eruptions such as the ∼500 CE eruption.
In southeastern Alaska on centennial timescales, warm/dry conditions correspond to La Niña mean states and cool/wet conditions correspond to El Niño mean states.While we recognize that modern instrumental data in the region (Figure S1 in Supporting Information S1) indicates warm/wet conditions during individual El Niño events and cool/dry conditions for individual La Niña events on interannual timescales, we suggest that this does not take into account long term trends.For example, the expected increase in cloudy conditions, which have a net cooling effect (Peirrehumbert, 2010), during an El Niño mean state would probably lower the regional temperature on centennial timescales, and vice versa for La Niña events.This is corroborated by regional glacial retreats/ advances in the region, which follow the long term trend of El Niño during glacial advances and the long term trend of La Niña during glacial retreats (Figure S12 in Supporting Information S1) (Wiles et al., 2008).
Atmospheric CO 2 strongly disconnects from the natural variability of the speleothem record after ∼1850 CE (Figure S13 in Supporting Information S1), in agreement with the rise of CO 2 above pre-industrial levels (Ahn et al., 2012).This is notable, given that the equatorial Pacific represents the largest CO 2 source globally (Takahashi et al., 2009), and is clearly at odds with natural forcings indicated by the speleothem record.Given the insignificant increase in solar irradiance to such a remarkable rise in atmospheric CO 2 , it is obvious that anthropogenic greenhouse gases are responsible for the deviation from natural variability.We suggest that this resulted in the breakdown of the ocean thermostat mechanism in the 1970s, in conjunction with a well-documented ENSO and North Pacific regime shifts in the late 1970s (Diaz et al., 2001;Giamalaki et al., 2018;Graham, 1994;Hare & Mantua, 2000;Mayo & March 1990).This has resulted in increasingly warmer/wetter conditions in southeast Alaska unseen during the previous ∼3,500 years.The warm/wet combination is inconsistent with previous regional El Niño or La Niña mean-state responses, and indicates a significant change in ENSO properties (Freund et al., 2019;Graham, 1994;Wang et al., 2019).
While the regime shifts in the late 1970s could also be attributed to a disruption in the atmospheric bridge linking the equatorial Pacific to the North Pacific, we argue that this is unlikely given the significant correlation between modern-day instrumental records of ENSO and regional southeastern Alaska precipitation/temperature (Figure S1 in Supporting Information S1).This implies that the atmospheric bridge is still strong, with ENSO continuing to force the strength of the Aleutian Low even under increased greenhouse gases.
We suggest that a switch to the weaker Walker mechanism in the 1970s, whereby zonal tropical sea-surface temperatures are reduced (Vecchi et al., 2008), resulted in the mean state shift of ENSO.Based on our speleothem data, the switch to a weaker Walker mechanism in the 1970s would not be possible if driven by natural forcings alone, and required the input of anthropogenic greenhouse gases.Reduced zonal tropical sea-surface temperatures have led to an El Niño mean-state and a strengthened Aleutian Low, bringing increased precipitation.However, the regional atmospheric warming in conjunction with the increased precipitation is atypical of an El Niño mean-state on centennial timescales.This implies that the wet/cool conditions typical of El Niño mean-state and associated with glacial advances during past centuries is being overwhelmed by anthropogenic warming.This significant change in ENSO may suggest that a climate change tipping point may have been crossed in the 1970s.We recommend that climate models fix biases that will result in both a diminished ocean thermostat response and an increased weaker Walker response at ∼1970 CE for improved climate projections of ENSO.

Figure 1 .
Figure 1.Spatial correlation of ERA5 Reanalysis 2 m temperature (a) and precipitable water (b) with the Niño 3.4 index.Regions of significant correlation are highlighted by color bands (Pearson's correlation [90% CI]).Black box indicates NINO3.4 region, while the speleothem sample location of this study is marked by the green star.Refer to Figure S2 in Supporting Information S1 for a more detailed map of the study area.This plot was generated using Climate Reanalyzer (http://cci-reanalyzer.org), Climate Change Institute, University of Maine, USA.

Figure 2 .
Figure 2. Schematic illustrating how the Aleutian Low is forced by both solar irradiance and El Niño-Southern Oscillation (ENSO) through the atmospheric bridge.The epikarst is effective at filtering the regional Aleutian Low climate into either the solar irradiance or ENSO signals.