Persistent ENSO Forcing on Holocene Flooding in the Middle‐Lower Yangtze River at Millennial Timescales

El Niño‐Southern Oscillation (ENSO) is the most dominant interannual signal of climate variability and profoundly affects river flooding globally, especially in East Asia. However, ENSO also has ∼2,000 and ∼1,000‐year cycles, but due to the lack of flood records with sufficient length, little is known about the ENSO's impact on floods at these millennial timescales. Here we test this in the middle‐lower Yangtze River by reconstructing the first Holocene flood record with optically stimulated luminescence and 14C ages of flood deposits. We find the periods with high flooding probability generally correspond with intervals of weakened solar activity. Importantly, the flood record displays 2,000 and 1,000‐year cycles similar to the ENSO record, and band‐pass filter results show the two records are synchronous at these bands. Our results reveal a persistent control of ENSO on millennial‐scale hydroclimatic variability in the Yangtze basin and likely other basins.


Introduction
El Niño-Southern Oscillation (ENSO) is the most potent driver of interannual climate variability at the global scale and influences the hydroclimate over a large part of the Earth's surface (Ward et al., 2014).This is especially true in the Yangtze River (YR) basin, a region most susceptible to the ENSO dynamics.The frequent destructive floods have caused huge casualties and property losses in the past decades.Thus, a comprehensive understanding of the driving forces of floods at different timescales is of great significance to accurately predict future floods and to prepare societies against the flooding.Modern instrumental data reveal that, at interannual timescales, the ENSO is the primary driver of floods in the middle and lower Yangtze basin (e.g., Hardiman et al., 2018;Tian et al., 2016;B. Wang et al., 2017;Xiao et al., 2015).Further, flood series were extended to past few hundred years by compiling flooding information from literature (e.g., T. Jiang et al., 2006;Q. Zhang et al., 2007) and sedimentary archives (M.Wang et al., 2011;Z. Wang et al., 2017).These longer flood records also identified this interannual teleconnection and showed that floods were statistically significant during El Niño periods.However, both proxy records (Chen et al., 2016;Cobb et al., 2013) and simulation (Z.Liu et al., 2014) results revealed that the ENSO's interannual cycle (∼2-7 years) is rather variable throughout the Holocene.This signal only became significant in the late Holocene but was strongly damped or even non-existent in most of the Holocene (Chen et al., 2016;Moy et al., 2002;Rodbell et al., 1999).On the other hand, the ENSO variation at millennial-bands exhibits a more persistent pattern (Moy et al., 2002;Stott et al., 2002;Turney et al., 2004).This raises the question of whether the ENSO-flood relationship at interannual timescale holds at the millennial timescale.
The key to answer above question is to utilize flood sediments to reconstruct flood records with sufficient length and resolution that can reflect millennial flooding variation, but this is challenging because the length, quality, and completeness of flood records are strongly affected by the long-term variability in vertical (channel incision/ aggradation) and horizontal (lateral migration and later erosion) mobilities of river systems.Meta-analysis is a promising approach to circumvent these dilemmas.This analysis deploys statistical methods on synthesized or merged findings of independent studies to calculate an overall trend (Jones et al., 2015).In paleohydrology, it has advantages of incorporating a large number of flood ages from many individual locations to produce a continuous curve which enables to identify flood variation at various timescales and to infer large-scale hydroclimatic teleconnections (Macklin et al., 2012).This approach has been successfully applied in studying the centennial-scale hydroclimate changes in Europe (Hoffmann et al., 2008;Macklin et al., 2010), the Mediterranean region (Benito et al., 2015) and southwest United States (Harden et al., 2010;T. Liu et al., 2020).However, a similar analysis for the YR is still missing.Here we apply this approach to a data set containing the most detailed Holocene flood ages from the middle-lower YR to construct a first probability-based flood record, based on this we discuss the forcings of Holocene floods in the middle-lower YR basin.

Materials and Method
By identifying flooding deposits in the field and dating the embedded organic ( 14 C) or clastic materials (optically stimulated luminescence (OSL)), one can know the ages of individual flood events.In this study, we limited our compilation of ages explicitly to those that indicate occurrences of river floods from published peer-reviewed articles.Flood ages derived from stratigraphical correlations, without direct chronological determination were not included.The locations and codes of investigated profiles/sites and ages were scrutinized to avoid collecting a flood event reported by multiple investigations.Therefore, each selected age was considered as representing a separate flood event.In total, 25 ages from the middle reaches 43 ages from the lower reaches were included in the database.All the investigated sites are shown in Figure 1a, and the chronological information, dated materials, coordinates, and sources are given in Table S1 and S2 in Supporting Information S1.
The raw 14 C ages were recalibrated using rcarbon package (Crema & Bevan, 2021) in R software with IntCal20 curve (Reimer et al., 2020).The median ages with 2σ ranges were adopted.It is noteworthy that OSL ages are usually reported as years before the year that the ages are determined, while 14 C ages are reported as years before 1950 CE.For uniformity, we converted the OSL ages to years before 1950 CE.Next, we used the rcarbon package to produce probability plots which summed all the dates entered into the analysis.Such plots visualize the temporal distribution of the dates entered into the calculation and determine periods with increased versus reduced extreme flood occurrence (cf.Hoffmann et al., 2008;Macklin et al., 2010).This approach presents the temporal distribution of a data set and has shown great promise in treating aggregated sets of flooding ages as proxies for variation in flood occurrences.This dates-as-data approach implies that increased floods leave behind a large sample of flood sediments compared to reduced flooding periods.It should be emphasized that this approach is not intended to identify and to reconstruct the magnitude of individual flood events, but to provide a probabilistic assessment of variability in the occurrence of flood events at different timescales (Macklin et al., 2011).Furthermore, we are mindful that this approach has limitations.For example, dates with a large uncertainty typically resulted in wide calibrated age ranges that can blur peaks and troughs in summed probability curves (Kennett et al., 2008).In addition, the variable gradient of the 14 C calibration curves can potentially exert an influence on the shape of probability curve.To minimize this effect, the raw probability curves were smoothed using a 100-year running mean (Pierce et al., 2004).We propose that an analysis of current database is timely, despite its limited size, as the application of this approach presents the first overview of flood variation for the middle-lower YR over the Holocene.

Results
Figures 1b-1d display flood variations for the middle, lower, and entire middle-lower reaches of the YR.Starting with the oldest part of the record in the middle reaches, the flood periods are relatively evenly distributed, without a strong bias toward a specific interval.Low flood periods are identified at 9.5-7.8 and 6.2-5.0 ka (Figure 1b).In the lower reaches, major flood periods are recognized during the middle Holocene and in the past two millennia (Figure 1c).A distinct feature is that the flooding increased considerably in the recent two millennia.The periods with high flooding probability in the middle reaches generally overlap with that in the lower reaches except at ∼2.8, 3.4, and before 9 ka (Figures 1b and 1c).When one considers the middle and lower reaches as a whole, the flood variation exhibits a more cyclic feature, with a pattern of progressively increased flooding from the early Holocene to the late Holocene (Figure 1d).

Justification of the Reliability of Flood Records
To demonstrate the robustness of our reconstruction, the sedimentary flood record for the lower YR is compared with historical flood records.Overall, the major flood periods in the lower YR correlate well with historical flood records in Yangtze delta (T.Jiang et al., 2005) and Nanjing (Bian, 2008) regions over the past two millennia (Figures 2a-2c).In addition, five large floods that occurred at 1170, 1298, 1502, 1532, and 1589 CE recorded by the floodmarks on the Mufu Mountain in Nanjing (He & Lee, 2017) correspond with peaks of reconstructed lower YR flood record (Figure 2a).
We also compared the flood variation in the middle YR with δ 13 C ASOM of acid-soluble organic matter from the Heshang cave stalagmite (HS4).The δ 13 C ASOM is considered a proxy of vegetation and soil ecosystem above the cave site, with more positive δ 13 C ASOM values during wetter conditions (Li et al., 2014).The periods of increased flooding in the middle reaches generally coincide with more positive δ 13 C ASOM , and the flood-poor periods also line up with relatively drier climates indicated by stalagmite proxies (Figures 2d and 2e).Particularly, at periods of 7.4-6.7 and 6.2-5.0 ka, the overall dry climatic phases are recorded not only by the δ 13 C ASOM record (Figure 2e), but also by a decline in humification record from the Daping swamp (Zhong et al., 2015), and reduction in water tables in the Dajiuhu peatland (H.Liu et al., 2019) in the middle Yangtze valley (Figure S2 in Supporting Information S1, see locations in Figure 1a).This implies that these flood-poor periods are most likely a result of dry climatic patterns, during which the flooding was subdued.
The middle YR flood record only partly correlated with the storm record (Zhu et al., 2017) represented by magnetite fluxes (IRM soft-flux ) from the HS4 stalagmite (Figure S1 in Supporting Information S1).The abrupt enhancement of magnetite signal is suggested to result from increased extreme precipitation events such as storms (Zhu et al., 2017).Yet, extreme rainfall events can be rather local (Hirschboek, 1988), which could have enhanced the magnetic signals in Heshang Cave, but not necessarily led to widespread flooding in a larger catchment.Therefore, it is not surprising that there is no significant correlation between the two records.We conclude that the reconstructed flood records likely captured the main feature of flood changes especially at millennial timescales, despite that some flooding events may have been missed occasionally.Therefore, we focused on the millennial timescale flooding variations in the following analysis.Jiang et al., 2005) and Nanjing region (Bian, 2008).Panels (d, e) show the comparison of middle YR flood record with δ 13 C of acid-soluble organic matter from Heshang cave stalagmite (Li et al., 2014).

Millennial Periodicities of the Flood Record and the Climatic Forcing
Overall, the pattern of increased flooding throughout the Holocene in the middle-lower YR agrees with an progressively increased ENSO variance (Moy et al., 2002) and transient coupled ENSO simulations (Z.Liu et al., 2014) (Figures 3a-3c).However, the major flood periods do not always coincide with peaks of strengthened ENSO variances (Figures 3a-3c).Although El Niño events can greatly increase the probability of above-average rainfall in the middle-lower YR, the relationship between the two is non-linear (Shankman & Lai, 2022;B. Wang et al., 2017).Analysis of El Niño events occurred from 1957 to 2016 shows that the weak El Niño can also enhance the post-El Niño summer rainfall over the middle-lower YR catchment (B.Wang et al., 2017).Interestingly, the periods with high flood probability generally correspond with reduced total solar irradiance (TSI) variation (Steinhilber et al., 2012) except at ∼1.4, ∼1.1, and ∼0.1 ka (Figure 3d).The inconsistency might be attributed to the anthropogenic perturbations in the late Holocene (Z.Wang et al., 2010).But it should be noted that low TSI periods do not always correspond to a flood-rich stage, suggesting other flood-forcing mechanisms beyond the influence of TSI, and future studies with more ages of flood events at finer timescales and mechanismbased sophisticated models should be able to explain this.
The evidence of climatic control on flooding in the Yangtze catchment at millennial timescales can be gathered through the analysis of periodicity and phase relationship, rather than looking at the occurrence of individual events.Spectral analyses reveal common periodicities of ∼2,000 and ∼1,000 years between the ENSO variability and the flood record, and the 1,000-year cycle (Eddy cycle) also exists in the TSI record (Figures 4a-4c).Importantly, band-pass filter and cross-wavelet transform results show that the flood variation is nearly in-phase with ENSO record at 2,000 and 1,000-year bands (Figures 4d, 4e, and 4g), and it also displays a consistent antiphase relationship with TSI at the 1,000-year band (Figures 4f and 4h).Hence, the floods in the middle-lower YR were broadly associated with weak solar irradiation as well as strong ENSO activities at millennial timescales in the Holocene.Indeed, this millennial-scale teleconnection in hydroclimate has also been observed in the Δδ 13 C series of Luoshui cave stalagmite (Z.Wang et al., 2022) and a water table record (H.Liu et al., 2019) of the Dajiuhu Peatland (see locations in Figure 1a) in the middle Yangtze valley.These records also exhibit 1,000 and/or 2,000-year periodicities in the Holocene, and band-pass filter results of two records generally correspond with that of the reconstructed Yangtze flood record (Figure S3 in Supporting Information S1), lending support to our results.(Moy et al., 2002).(c) Modeled El Niño-Southern Oscillation amplitude in 100-year windows (standard deviation of El Niño 3.4 interannual (1.5-7 years) SST variability) (Z.Liu et al., 2014).(d) Loess-smoothed total solar irradiance (Steinhilber et al., 2012).Gray bands show intervals of increased flooding probability.
The phase relationships presented here are remarkable as it implies a persistent influence of ENSO on flooding at millennial timescales.Our current knowledge of ENSO modulation over the Yangtze flooding at interannual and centennial timescales could help explain this.At the interannual timescale, the reduced solar forcing could weaken the Pacific Walker circulation (Gleisner & Thejll, 2003;Meehl et al., 2008;van Loon et al., 2007) and favor a more El Niño-like condition.In East Asia, this could further weaken the strength of summer monsoon (Y.Wang et al., 2005;P. Zhang et al., 2008) and result in a westward shift of western Pacific subtropical high, leading to the long-lasting hovering of Mei-yu (a precipitation season from middle June to middle July in East Asia) belt over the middle-lower YR and bringing excessive rainfall to this region Jiang et al., 2015;Zhou et al., 2009).(Moy et al., 2002), and (c) the total solar irradiance (TSI) (Steinhilber et al., 2012).Panels (d, e) show 2,000 and 1,000-year band pass filter results for the flood record and ENSO variance, and (f) shows 1,000-year band pass filter results for the flood record and the TSI.Panels (g, h) display cross-wavelet transform power of flood record with ENSO and TSI, respectively.The phase relationship is shown by arrows (nearly in-phase, pointing to the right; nearly anti-phase, pointing to the left).The white dashed boxes indicate 2,000 and 1,000-year bands.
with lower solar irradiance inducing warmer eastern Pacific sea surface temperatures (El Niño-like).At the millennial timescale, indeed, Steinhilber et al. (2012) found that the lower TSI corresponded to the weaker summer monsoon.This may imply that at the 1,000-year timescale, solar forcing can also modulate the flooding regime in the middle-lower YR through ENSO system.However, whether the 2,000-year cycle of the ENSO mean state is a result of internal dynamics operating independently or is driven by external forcing is uncertain (Moy et al., 2002).Although this warrants future study, we propose the in-phase relationship between the mean of ENSO and flooding at the 2,000-year band can be explained by the similar mechanism in which the ENSO impact the flooding at interannual to centennial-timescales.

Conclusions
In this study, we chose the middle and lower YR basin to understand the influence of ENSO on flooding at millennial timescales.We reconstructed the Holocene flood record by constructing a data set including the most detailed flood ages over the middle and lower YR.The record shows that the periods with high flood probability periods generally correspond with intervals of weakened solar activity.Importantly, spectral power analyses revealed the 2,000 and 1,000-year cycles of the flood record.The band-pass filter and cross-wavelet transform results further show an in-phase relationship of the flood variation with ENSO variance at 2,000 and 1,000-year bands, and anti-phase relationship with TSI at the 1,000-year band.Based on the current understanding of ENSO modulation over the Yangtze flooding at interannual and centennial timescales, we propose this mechanism may also hold at the millennial timescale.Our results for the middle-lower Yangtze fit observations around the globe and demonstrate the persistent impact of ENSO, in association with variations in solar activity, on Holocene flooding regimes.

Figure 1 .
Figure 1.(a) Digital elevation map of East-Asia showing the Yangtze River (YR) and sites (cross marks) included in this study.The upper, middle, and lower reaches of the Yangtze are demarcated by Yichang and Hukou.Panels (b, c) show flood variation for the middle and the lower reaches of the YR, respectively.Panel (d) shows the entire middle-lower Yangtze basin.The dashed line in each panel represents the average value.

Figure 2 .
Figure 2. Comparisons of reconstructed flood records in the middle and lower Yangtze River (YR) with local historical flood records and regional paleohydrological reconstructions.(a-c) Correlations of flood record in the lower YR with historic flood records in the Yangtze Delta (T.Jiang et al., 2005) and Nanjing region(Bian, 2008).Panels (d, e) show the comparison of middle YR flood record with δ 13 C of acid-soluble organic matter from Heshang cave stalagmite(Li et al., 2014).

Figure 4 .
Figure 4. Spectral power analyses for the (a) middle-lower Yangtze flood record, (b) the El Niño-Southern Oscillation (ENSO) variance(Moy et al., 2002), and (c) the total solar irradiance (TSI)(Steinhilber et al., 2012).Panels (d, e) show 2,000 and 1,000-year band pass filter results for the flood record and ENSO variance, and (f) shows 1,000-year band pass filter results for the flood record and the TSI.Panels (g, h) display cross-wavelet transform power of flood record with ENSO and TSI, respectively.The phase relationship is shown by arrows (nearly in-phase, pointing to the right; nearly anti-phase, pointing to the left).The white dashed boxes indicate 2,000 and 1,000-year bands.