Hydro-climatic variability in the southwestern Indian Ocean between 6000 and 3000 years ago

The “4.2 ka event” is frequently described as a major global climate anomaly between 4.2 and 3.9 ka, which defines the beginning of the current Meghalayan age in the Holocene epoch. The “event” has been disproportionately reported from proxy records from the Northern Hemisphere, but its climatic manifestation remains much less clear in the Southern Hemisphere. Here, we present highly resolved and chronologically well-constrained speleothem oxygen and carbon isotopes records between ∼ 6 and 3 ka from Rodrigues Island in the southwestern subtropical Indian Ocean, located∼ 600 km east of Mauritius. Our records show that the 4.2 ka event did not manifest itself as a period of major climate change at Rodrigues Island in the context of our record’s length. Instead, we find evidence for a multicentennial drought that occurred near-continuously between 3.9 and 3.5 ka and temporally coincided with climate change throughout the Southern Hemisphere.


Introduction
The "4.2 ka event" is considered to be a widespread climate event between 4.2 and 3.9 ka (thousand years before present, where the present is 1950 CE) (e.g.Weiss et al., 1993Weiss et al., , 2016)).Many paleoclimate records from the Northern Hemisphere (NH) have characterised the event as a multi-decadal to multi-centennial period of arid and cooler conditions across the Mediterranean, Middle East, South Asia and North Africa (e.g.Finné et al., 2011;Marchant and Hooghiemstra, 2004;Migowski et al., 2006;Mayewski et al., 2004;Staubwasser et al., 2003;Arz et al., 2006;Zielhofer et al., 2017;Stanley et al., 2003;Kathayat et al., 2017).The structure of the 4.2 ka event from many proxy records, such as peat cellulose records from the eastern Tibetan Plateau (Hong et al., 2003(Hong et al., , 2018)), speleothem from northeastern India (Berkelhammer et al., 2012) and southern Italy (Drysdale et al., 2006), marine sediments from the Gulf of Oman (Cullen et al., 2000) and the northern Red Sea (Arz et al., 2006), and the dust record in the Kilimanjaro ice core (Thompson et al., 2002), typically characterised it as a single pulse-like signal in the long-term context of these records.In contrast, the structure of the 4.2 ka event in the Southern Hemisphere (SH) remains unclear.Some proxy records from the tropical and subtropical regions of Africa and Australia show a shift towards drier conditions around 4 ka (e.g.Russell et al., 2003;Marchant and Hooghiemstra, 2004;Griffiths et al., 2009;Denniston et al., 2013;Berke et al., 2012;De Boer et al., 2013, 2014, 2015;Schefuß et al., 2011;Rijsdijk et al., 2009Rijsdijk et al., , 2011)).Other records show virtually unchanged hydrological conditions (e.g.Tierney et al., 2008Tierney et al., , 2011;;Konecky et al., 2011) or two-pulsed multi-decadal length wet events (Railsback et al., 2018) during the period contemporaneous with the 4.2 ka event.(Draxler and Hess, 1998) using NCEP/NCAR reanalysis data (Kalnay et al., 1996).Bold lines indicate main cluster tracks associated with trajectories for wetter (green) and drier (brown) years.Inset shows mean monthly rainfall and temperature at Rodrigues averaged over the period of 1951 to 2015.
The goal of this study is to investigate a time period that spans the 4.2 ka event in a key region of the SH via highly resolved and precisely dated proxy records.Here, we present speleothem oxygen (δ 18 O) and carbon (δ 13 C) isotope records from La Vierge (LAVI-4) and Patate (PATA-1) caves from the Rodrigues Island (Fig. 1) in the southern subtropical Indian Ocean.The LAVI-4 and PATA-1 records span from ∼ 6 to 3 ka and from ∼ 6.1 to 3.3 ka, with an average resolution of ∼ 4 and 14 years, respectively.LAVI-4, which constitutes our primary record, has a precise age control and a subdecadal resolution, which, together with the PATA-1 record, allows us to reliably characterise the multi-decadal to centennial hydro-climate variations in the southwestern Indian Ocean during the period from 6 to 3 ka.

Climatology
Rodrigues (∼ 19 • 42 S, ∼ 63 • 24 E) is a small volcanic island (∼ 120 km 2 ) situated in the southwestern Indian Ocean, ∼ 600 km east of Mauritius (Fig. 1).The island's maximum altitude is ∼ 400 m above sea level.Rodrigues' mean annual temperature is ∼ 24 • C and the mean annual rainfall is ∼ 1010 mm, of which nearly 70 % occurs during the wet season (November-April) with February being the wettest month.The seasonal distribution of rainfall is largely controlled by the seasonal migration of the ITCZ (Intertropical Convergence Zone) and the Mascarene High (Senapathi et al., 2010;Rijsdijk et al., 2011;Morioka et al., 2015) (Fig. 1).
Clim.Past, 14, 1881Past, 14, -1891Past, 14, , 2018 www.clim-past.net/14/1881/2018/Given its location at the southern fringe of the ITCZ, the austral summer rainfall at Rodrigues is very sensitive to the mean position of the southern limit of the ITCZ.This is highlighted by backward (120 h) HYSPLIT (Draxler and Hess, 1998) trajectory composites of the low-level winds (850 hPa) during the years when the total January to March (JFM) precipitation was unusually low (dry) and high (wet) than the long-term mean  at Rodrigues (Fig. 1b).It is of note that there is a major increase in the fraction of air parcel trajectories arriving from the north of Rodrigues during the wetter years, indicating an enhanced contribution of northerly moisture resulting from a more southerly position of the ITCZ (Fig. 1b).This observation is further supported by analyses of the low-level wind trajectory cluster composites of February in those years when the southern boundary of the ITCZ was anomalously north or south (Lashkari et al., 2017;Freitas et al., 2017) of its long-term mean February position (Fig. S1a, b in the Supplement).In addition to the ITCZ, ENSO (El Niño-Southern Oscillation) also modulates austral summer precipitation at Rodrigues by modulating the Hadley and Walker circulations (Senapathi et al., 2010;De Boer et al., 2014;Griffiths et al., 2016;Zinke et al., 2016).Instrumental data and our trajectory composites for selected El Niño and La Niña years suggest that an increased (decreased) summer precipitation at Rodrigues is associated with the El Niño (La Niña) events (Fig. S1c, d).

Oxygen isotopes and climatology
Modern observations of δ 18 O of precipitation (δ 18 O p ) in the study area are unavailable due to the lack of Global Network of Isotopes in Precipitation (GNIP) stations in Rodrigues.However, δ 18 O p data from the nearest GNIP station in Mauritius show a clear annual cycle in δ 18 O p with depleted values during the austral summer (Fig. S2a).Additionally, in the absence of GNIP data, we use simulated δ 18 O p data from the Experimental Climate Prediction Center's Isotopeincorporated Global Spectral Model (IsoGSM) (Yoshimura et al., 2008) to assess the large-scale dynamical processes that control δ 18 O p on interannual and decadal timescales.
Our analyses show the presence of a strong negative correlation between the δ 18 O p and rainfall amount similar to the "amount effect" (e.g.Dansgaard, 1964) (Fig. S2b, c).We therefore interpret δ 18 O p variations in the cave catchment and, consequently, in speleothems from this region as primarily reflecting variations in rainfall amount in response to both local and large-scale atmospheric circulation changes.
The relationship is such that more negative (positive) δ 18 O p values occur during times of either an anomalously southward (northward) position of the southern boundary of the ITCZ or El Niño (La Niña) conditions.(Middleton and David, 2013).The cave temperature and relative humidity at the time of sample collection (June 2015) were ∼ 25.5 • C and 95 % in La Vierge Cave and ∼ 22.5 • C and 95 % in Patate Cave.Samples LAVI-4 and PATA-1 were collected at a distance of ∼ 50 and 200 m from cave entrances, respectively.The diameters of LAVI-4 and PATA-1 are ∼ 75 and 95 mm, and their lengths are ∼ 400 and ∼ 334 mm, respectively.Both stalagmites were cut along their growth axes using a thin diamond blade and then polished.

230 Th dating
Subsamples (80-130 mg) for 230 Th dating were drilled using a 0.9 mm carbide dental drill. 230Th dating was performed at Xi'an Jiaotong University, China, by using a Thermo-Finnigan Neptune Plus, which is a multi-collector inductively coupled plasma mass spectrometer (MC-ICP-MS).
The method is described in Cheng et al. (2000Cheng et al. ( , 2013)).We used standard chemistry procedures (Edwards et al., 1987) to separate U and Th.A triple-spike ( 229 T− 233 U− 236 U) isotope dilution method was used to correct instrumental fractionation and to determine U / Th isotopic ratios and concentrations (Cheng et al., 2000(Cheng et al., , 2013)).U and Th isotopes were measured on a MasCom multiplier behind the retarding potential quadrupole in the peak-jumping mode using standard procedures (Cheng et al., 2000).Uncertainties in U and Th isotopic measurements were calculated offline at the 2σ level, including corrections for blanks, multiplier dark noise, abundance sensitivity and contents of the same nuclides in the spike solution. 234U and 230 Th decay constants of Cheng et al. (2013) were used.Corrected 230 Th ages assume an initial 230 Th/ 232 Th atomic ratio of (4.4 ± 2.2) × 10 −6 and those are the values for a material at secular equilibrium with the bulk earth 232 Th/ 238 U value of 3.8.The correction for a few samples in LAVI-4 and PATA-1 is large because either U concentration is low (∼ 65 ppb) and/or the detrital 232 Th concentration is elevated (> 100 ppt) (Table S1, Fig. 2).

Stable isotope analysis
LAVI-4 and PATA-1 stable isotope (δ 18 O and δ 13 C) records were established by ∼ 962 and ∼ 190 data, respectively.The New Wave Micromill -digitally controlled tri-axial micromill equipment -was used to obtain subsamples.The subsamples (∼ 80 µg) were continuously micro-milled along the stalagmite growth axes of LAVI-4 and PATA-1 at increments of between 100 and 200 µm.The subsamples of LAVI-4 were measured using a Finnigan MAT-253 mass spectrometer coupled with an online carbonate preparation system (Kiel-IV) in the Isotope Laboratory, Xi'an Jiaotong University.The subsamples of PATA-1 were measured using an online carbonate preparation system (GasbenchII) connected to an isotope ratio mass spectrometer (Delta plus XL) in the Isotope Laboratory, Innsbruck University.The latter technique is reported in Spötl (2011) and Spötl and Vennemann (2003).
All results are reported in per mil (‰) relative to the Vienna PeeDee Belemnite (VPDB) standard.Duplicate measurements of standards show a long-term reproducibility of ∼ 0.1 ‰ (1σ ) or better (Table S2, Fig. 2).
Both schemes yielded virtually identical age models, and thus the conclusions of this study are not sensitive to the choice of the age model (Figs. 2 and S3).
The time interval from 6 to 3 ka in LAVI-4 speleothem corresponds to a sample depth of 274 to 81 mm below the top, respectively.A drip-water relocation occurred at a depth of 124 mm, which is associated with a Type L surface characterised by slow growth and narrow layers under progressively drier conditions (Railsback et al., 2013) (Figs.S3 and S4).It cannot be ruled out that there is also a hiatus at this depth (∼ 3.5 ka).If such a hiatus was indeed present, its duration would be about 100 years based on the age model (Fig. S4).The time interval from 6.1 to 3.3 ka in PATA-1 corresponds to a sample depth of 34 to 15 mm.The growth of PATA-1 ceased at ∼ 15 mm and then resumed about 630 years later, creating a hiatus (Fig. S3).The COPRA age models of PATA-1 and LAVI-4 (Figs. 2 and S3) are reported in Table S2 and used in the following discussion.

Isotopic equilibrium tests
Conventional criteria to assess the isotopic equilibrium of stalagmites are provided by the Hendy Test (Hendy, 1971), which requires no correlation between δ 18 O and δ 13 C values measured along the growth axis as well as along the same growth lamina.The correlation between the δ 18 O and δ 13 C values in LAVI-4 and PATA-1 is 0.53 and 0.85, respectively, which suggests the possibility of isotopic disequilibrium during calcite precipitation.However, a number of studies (e.g.Dorale and Liu, 2009) pointed out that a correlation between δ 18 O and δ 13 C values does not automati-cally rule out isotopic equilibrium.Instead, the replication test (i.e. a high degree of coherence between δ 18 O profiles of individual speleothems from the same cave) is a more rigorous and reliable test of isotopic equilibrium.Particularly, the replication test is far more robust if the records used are from different caves with different kinetic/vadose-zone processes, as is the case for this study.Indeed, a high degree of visual similarity between the coeval portions of LAVI-4 and PATA-1 δ 18 O and δ 13 C records suggests that both stalagmites record primary climate signals, notwithstanding the offsets between the absolute values (Fig. 2a).The replication is further confirmed by statistically significant correlations between the LAVI-4 and PATA-1 δ 18 O (r = 0.64 at 95 % confidence level) and δ 13 C (r = 0.73 at 95 % confidence level) records calculated using the ISCAM algorithm (Fohlmeister, 2012) for their contemporary growth period between 3.4 and 6.0 ka (Fig. 2).ISCAM uses a Monte Carlo approach to find the best correlation between the proxy records by adjusting each record within its dating uncertainty.The significant levels are assessed against a red-noise background generated using artificially simulated first-order autoregressive time series (AR1).The offset in absolute δ 18 O values between LAVI-4 and PATA-1, however, remains unclear and possibly arises from processes related to the characteristics of the two karst systems, such as temperature differences as observed during our fieldwork in 2015.Therefore, in the following discussion we focus only on temporal variations of LAVI-4 δ 18 O and δ 13 C records due to their higher resolution and better-constrained chronology (Fig. 2).

Proxy interpretation
The temporal resolution of the LAVI-4 δ 18 O record between 6 and 3 ka varies from 1.2 to 16.4 years with an average resolution of ∼ 3.2 years.The δ 18 O temporal variability is large (∼ 3.5 ‰) and, as noted earlier, we interpret the δ 18 O variations to dominantly reflect changes in the precipitation amount.This line of reasoning is justified given the island's isolated setting far removed from large-sized landmasses and its low topographic relief, which minimises isotopic variability stemming from processes such as the continentality and altitude effects as well as the mixing of distant water vapour sources with significantly different isotopic compositions.This interpretation is additionally supported by moderate to strong covariance between the LAVI-4 δ 18 O and δ 13 C profiles.Although the process of stalagmite precipitation may be affected by evaporation and/or degassing (Treble et al., 2017;Cuthbert et al., 2014;Markowska et al., 2016;McDermott, 2004;Lachniet, 2009), the temporal variations in the latter can stem from changes in vegetation type and density, soil microbial productivity, prior calcite precipitation (PCP), and groundwater infiltration rates (e.g.Baker et al., 1997;Genty et al., 2003), all of which may drive δ 18 O and δ 13 C values in the same fashion (e.g.Brook et al., 1990;Dorale et al., 1992;Bar-Matthews et al., 1997).The significant covariance between the δ 13 C and δ 18 O records could therefore indicate that both proxies reflect a common response to changes in rainfall amount at Rodrigues or a rainfall limit on the extent of vegetation and other related processes in the epikarst as mentioned above.
Overall, the climate variations recorded at Rodrigues from 4.2 to 3.9 ka are characterised by high-frequency (decadal to multi-decadal) fluctuations, including the major arid/wet events mentioned above.Notably, however, the mean hydroclimatic state of this time interval inferred from both δ 18 O and δ 13 C data is indistinguishable from the average state between 6 and 4.2 ka (Fig. 3).In this regard, the climatic events or anomalies between 4.2 and 3.9 ka are not distinctly larger in amplitude nor longer in duration in comparison to similar anomalies between 6 and 4.2 ka (Figs. 3 and 4).Consistently, in the context of the long-term climate variance between 6 and 3 ka, there is no evidence for an unusual climate anomaly between 4.2 and 3.9 ka.
The most prominent feature of our record is a switch from an interval characterised by high-frequency δ 18 O variance (i.e. from 6 to 3.9 ka) to a multi-centennial excursion with progressively higher δ 18 O and δ 13 C values: a prolonged mega-drought at Rodrigues.Starting at ∼ 3.9 ka, this megadrought became progressively more severe leading to a diminished growth rate or a ∼ 100 yr long hiatus around 3.5 ka in LAVI-4.Growth rate picked up subsequently, followed by abrupt (∼ 100 yr long) and large decreases in both δ 18 O (∼ 2 ‰) and δ 13 C (∼ 5 ‰) to their average values of the entire records between 6 and 3 ka.As such, the structure of the mega-drought event shows a saw-tooth pattern with a multicentennial drying trend followed by a ∼ 100 yr long return to the mean state (Fig. 3).The multi-century mega-drought recorded by our stalagmites between 3.9 and 3.5 ka is also evident in Sahiya Cave, north India (Kathayat et al., 2017), and from Lake Edward (Russell et al., 2003), Lake Victoria (Berke et al., 2012), the Zambezi Delta (Schefuß et al., 2011) and the Tatos Basin (De Boer et al., 2014) (Fig. 5).In the eastern sector of the southern Indian Ocean, speleothem records from Tangga (Wurtzel et al., 2018), KNI-51 (Denniston et al., 2013) and Liang Luar (Griffiths et al., 2009) caves also show a shift to drier condition at approximately 4 ka (Fig. 4).
The LAVI-4 δ 13 C record shows a pattern broadly similar to the δ 18 O record and clearly delineates three major droughts between 6 and 3 ka, centred at 5.43, 4.62 and 3.54 ka, respectively.These three drought events share a distinct saw-tooth pattern characterised by a long-term gradual positive excursion (drying) followed by an abrupt return to the mean values (Fig. 3).
To sum up, our Rodrigues records show evidence of multidecadal-decadal hydro-climate fluctuations around the mean state between 6 and 3 ka.After 3.9 ka, the hydro-climate was characterised by a multi-centennial trend toward much drier conditions, which ended with a return at ∼ 3.5 ka within ∼ 100 years to the mean hydro-climate state.This pattern is different from the "pulse-like" event between 4.2 and 3.9 ka as documented in many other proxy records mainly from the NH.Additionally, the mega-drought between 3.9 and 3.5 ka is clearly a later event unrelated to the 4.2 ka event.

Possible mechanisms
A close examination of our Rodrigues δ 18 O and δ 13 C records shows that a persistent multi-centennial drying trend began effectively at ∼ 4.1 ka and ended at ∼ 3.5 ka, suggesting a prolonged northward shift of the mean position of the ITCZ (Figs. 3 and S1a).This inference, if correct, is partially in contrast with the southward shift of the ITCZ, which is often invoked to explain the weakening of the Asian monsoon since ∼ 4.2 ka (e.g.Wang et al., 2005;Kathayat et al., 2017).Thus, the observed drying trends on both the northern and southern fringes of the ITCZ in both hemispheres argue against the model of a southward shift in the mean position of the ITCZ as a viable cause of the 4.2 ka event.A more likely explanation involves an overall contraction in the north-south range of the migrating ITCZ belt in the region (e.g.Yan et al., 2015;Denniston et al., 2016;Scroxton et al., 2017).This mechanism is broadly consistent with the spatial pattern of hydro-climate changes observed in both hemispheres around and after the 4.2 ka event.As mentioned above, the wet period between ∼ 4.1 and 4.0 ka recorded at the northern fringe of the ITCZ (Kathayat et al., 2017(Kathayat et al., , 2018) ) coincided with a wet period on the southern limit of the ITCZ as recorded in Dante Cave (Railsback et al., 2018), the Zambezi Delta (Schefuß et al., 2011), Tatos Basin (De Boer et al., 2014), La Vierge Cave (this study) and KNI-51 Cave (Denniston et al., 2013) (Figs. 4 and 5).The subsequent arid period between ∼ 3.9 and 3.5 ka was also basin-wide and affected both the northern and southern limits of the ITCZ over the Indian Ocean and adjacent regions (Figs. 4 and 5).
In parallel with drier condition along the southern limit of the austral summer ITCZ, proxy records from Lake Edward (Russell et al., 2003), Lake Victoria (Berke et al., 2012) and Tangga Cave (Wurtzel et al., 2018), which are located near the northern limit of the contemporary austral summer ITCZ, also exhibit drier conditions.In contrast, records within the core location of the austral summer ITCZ, such as Lake Challa (Tierney et al., 2011), Lake Tanganyika (Tierney et al., 2008), Lake Malawi (Konecky et al., 2011) and Makassar Strait (Tierney et al., 2012), show either slightly wetter or virtually unchanged hydro-climatic conditions (Figs. 4  and 5).Based on the observed spatial patterns, we suggest that the contraction of the ITCZ both in terms of a northsouth meridional shift as well as with respect to its overall width may have played an important role in modulating the hydro-climate in our study area during and after the 4.2 ka event.Kathayat et al., 2017), Mg concentration of endogenic calcite from Lake Edward (Russell et al., 2003), δD leaf wax records from Lake Victoria (Berke et al., 2012), Lake Challa (Tierney et al., 2011), Lake Tanganyika (Tierney et al., 2008), Lake Malawi (Konecky et al., 2011), δD of n-C 29 alkanes (dark blue) and n-C 31 alkanes (orange) from the Zambezi delta (Schefuß et al., 2011), δ 18 O record (black) and ln (Latania/Eugenia) records (brown) from Tatos Basin (Mauritius;De Boer et al., 2014), and the LAVI-4 δ 18 O and δ 13 C record from La Vierge Cave (this study).The shaded vertical bar marks the mega-drought from ∼ 3.9 to 3.5 ka.Grey and green dashed arrows mark the drying and wet trend inferred from east Africa lake records, respectively.All y axes are inverted to show drier conditions downwards.

Figure 1 .
Figure 1.Proxy locations and climatology.(a) Mean January to March (JFM) precipitation from the Tropical Rainfall Measuring Mission (TRMM) (https://trmm.gsfc.nasa.gov/,last access: 2 December 2018) averaged over the period from 1998 to 2014.The shaded area bounded by solid brown lines (3 mm day −1 isohyet) depicts the mean position of the ITCZ.The dashed line shows the mean position of JFM 850 hPa geopotential height marking the location of the Mascarene High.The locations of Rodrigues Island (yellow star, this study) and other proxy sites (green dots) discussed in the text are also shown.(b)Four times daily, low-level (∼ 850 hPa) JFM air parcel back-trajectory (120 h) composites for anomalously wet (green) and dry (brown) years.Trajectories were computed using the NOAA HYSPLIT model(Draxler and Hess, 1998) using NCEP/NCAR reanalysis data(Kalnay et al., 1996).Bold lines indicate main cluster tracks associated with trajectories for wetter (green) and drier (brown) years.Inset shows mean monthly rainfall and temperature at Rodrigues averaged over the period of 1951 to 2015.

Figure 3 .
Figure 3. Inferred hydro-climatic variability at Rodrigues from 6 to 3 ka.LAVI-4 δ 18 O and δ 13 C record are z-score transformed.Inferred droughts (z-score > 1) and pluvial episodes (z-score < −1) are shaded (increasing saturation index indicates increasing intensity).Dashed lines indicate 1 standard deviation.The blue bar marks the classical 4.2 ka event interval and the yellow bar marks the "post-event" mega-drought, inferred from LAVI-4.