Mayotte coral reveals hydrological changes in the western Indian Ocean between 1881 and 1994



[1] We reconstruct the hydrologic history of the tropical western Indian Ocean by calculating the δ18Oseawater from coupled coral Sr/Ca and δ18O measurements in a massive Porites coral from Mayotte (Comoros) between 1881 and 1994. We found that the precipitation-evaporation balance varies naturally on time scales of 5–6 years and 18–25 years. High (low) SSTs are associated with positive (negative) δ18Oseawater implying that atmospheric variability is linked with remote climate modes in the Indian Ocean and the tropical/extratropical Pacific Ocean. Warm El Niño-Southern Oscillation events are associated with a negative freshwater balance at Mayotte. This case study demonstrates that a much denser network of δ18Oseawater reconstructions is crucial for understanding the spatial patterns of hydrological conditions.

1. Introduction

[2] Future global warming will trigger significant changes in surface temperature and the hydrological cycle, with both combined leading to uncertain societal and environmental consequences. For the Indian Ocean and the adjacent continents the IPCC's latest AR4 projections show a large spread [Annamalai et al., 2007]. Precipitation over the oceans has a complex spatial structure and often differs from trends in underlying sea surface temperature (SST) [Kumar et al., 2004]. Over the western Indian Ocean, instrumental data is too sparse to investigate the rainfall/temperature response over the ocean on interannual to multidecadal time scales [Annamalai et al., 1999; Deser et al., 2004]. Satellite-borne rainfall measurements, which provide high spatial coverage over the oceans, are not available before 1979, a period too short to reduce the uncertainty in the trend estimates.

[3] However, the stability of the ocean-atmosphere coupling in a changing climate is important for future climate projections. Long proxy records with sufficient temporal resolution are the only alternative to assess the stability of such connections.

[4] The close link between SST and the El Niño-Southern Oscillation (ENSO) in the western Indian Ocean north of 10°S is confirmed by a number of coral records [Charles et al., 1997; Cole et al., 2000; Timm et al., 2005; Zinke et al., 2005; Pfeiffer et al., 2006], while the response of the hydrological cycle is not well understood. Pfeiffer et al. [2006] find positive rainfall excursions during El Nino years north of 10°S. Rainfall data from southern Africa, however, suggests that El Niño events may lead to pronounced droughts south of 10°S [Richard et al., 2000].

[5] In this study we investigate long-term hydrological changes over the western Indian Ocean south of 10°S using a 113 year long, bimonthly resolved record of δ18O and Sr/Ca from a Porites solida coral drilled at Mayotte (12°39′S, 45°06′E; Comoro Archipelago; for methodology see auxiliary materials). The combination of Sr/Ca as a temperature proxy with δ18O as an independent proxy for SST and δ18Oseawater can provide simultaneous information on SST and hydrological changes [Cahyarini et al., 2008, and references therein]. Our coral δ18Oseawater series is the first of its kind from the western Indian Ocean. We will use a range of hydrological data (δ18Oseawater and salinity measurements, rainfall and precipitation-evaporation data) collected at and around Mayotte to demonstrate that our coral proxy data can expand the current understanding of the hydrological cycle.

2. Climatic Setting of the Study Area

[6] Air Temperature (AT) and rainfall for Mayotte are available between 1949 to present and 1933 to present, respectively (NOAA GHCN monthly Weather Station at Pamandzi/Dzaoudzi, 12.80°S, 45.28°E). An inferred local SST record (hereafter SSTi) back until 1949 is derived from linear regression of in-situ lagoonal SST with local air temperature for the calibration period 1999–2002 (equation: SST = 0.8535*AT + 4.7733; r2 = 0.85). Regional precipitation minus evaporation (P-E) and outgoing long-wave radiation (OLR) data are taken from NCEP/NCAR reanalysis [Kalnay et al., 1996] averaged over 10–15°S, 40–45°E. Monthly gridded sea level pressure (SLP) data are taken from the Met Office of the Hadley Centre of the UK [Allan and Ansell, 2007].

[7] SST and rainfall show an unimodal distribution with highest temperatures and rainfall between December to April (Figure S1). We investigated the relationship between (i) annual mean regional rainfall in the region of Mayotte (averaged between 30–50°E, 10–20°S) and the global SLP field (Figure 1a), and (ii) seasonal mean (December–March, hereafter DJFM) AT/SSTi from Mayotte with the global SST field between 1951 and 1993 (Figure S2). The correlation maps indicate the typical global ENSO SLP (Figure 1a) and SST (Figure S2) pattern. At Mayotte, negative rainfall and positive SST anomalies occur during El Niño events (Table S1).

Figure 1.

Spatial correlation of mean annual averages (July to June) of (a) rainfall data averaged between 30–50°E and 10–20°S including Mayotte (black circle) and (b) of δ18Oseawater (δ18Osw) with gridded sea level pressure (HadSLP) [Allan and Ansell, 2007]. Shading represents confidence of 90% and greater. Red shading indicates positive correlations and blue negative correlations. (c) Linear regression between mean annual and December to March averaged Sr/Ca ratios with local sea surface temperatures, (d) same as in Figure 1c but for δ18Oseawater (δ18Osw) with outgoing longwave radiation (OLR) [Kalnay et al., 1996] for the period 1958 to 1993 and (e) same as in Figure 1d but for local sea surface temperatures. Correlation coefficients are indicated. Correlations are significant at the 5% level.

3. Local Hydrology and Seawater Oxygen Isotopic Variations

[8] In the Mayotte region δ18Oseawater is modified by (i) rainout of isotopically depleted rainfall from warm air masses over deep convective regions (the so-called amount effect [Dansgaard, 1964]) during the intense rainy season, (ii) isotopic fractionation during evaporation and (iii) freshwater surface runoff.

[9] A detailed hydrological study at Mayotte for 1993 (dry year) versus 1994 (normal year) revealed that evaporation contributed 45–60% to the annual freshwater balance and was 15% higher under drier conditions in 1993 (Figure S3) [see also Lapègue, 1999]. For the given example years, evaporation contributes significantly to δ18Oseawater. The NCEP/NCAR reanalysis data for Mayotte further indicates that evaporation intra/interannual variability is of the same order as the precipitation variability.

[10] We monitored δ18Oseawater for several months during the dry and wet season in 2002 and between September 2004 and August 2005 (Figure S3). Our δ18Oseawater shows a maximum seasonal amplitude of 1.13‰, ranging between 0.26‰ and 1.38‰. The cause of the high values in September and October 2004 is unknown since we lack salinity data. The maximum δ18Oseawater amplitude is reduced to 0.51‰ when September and October 2004 are not considered. The negative δ18Oseawater values observed between August and October 2002 are related to low salinities in the lagoon (Figure S3). River runoff shows considerable interannual variability and this runoff may potentially influence δ18Oseawater in the Mayotte lagoon (Figure S1). For 2002, minimum salinity in the lagoon lags maximum river flow by 1 month (Figure S3). This implies that the integral effect of P-E over the length of the intense rainy season and freshwater runoff into the lagoon of Mayotte resulted in a strong signal in δ18Oseawater that persists into austral winter of 2002.

4. Results and Discussion

4.1. Calibration, Calculation of δ18Oseawater and Error Estimates

[11] Bimonthly coral Sr/Ca ratios primarily reflect SST; linear regression yields a slope value of −0.058 mmol/mol per 1°C in agreement with published relationships [Corrége, 2006]. Mean annual Sr/Ca ratios show a statistically significant correlation of r = −0.47 (p < 0.003) with SSTi for the period 1950–1994 (Figure 1c). To assess the uncertainty of the SST datasets, we correlated SSTi with various SST products (ERSST, HADISST, COADS) for the grid cell including Mayotte (annual means): r ranges between 0.5–0.6. This gives us a crude estimate of the uncertainty of the SST data: approximately 70% of the SST variability is not reproducible and may reflect noise (local climatic effects, measurement errors etc.). Given these uncertainties of SSTi (and other SST products), the correlation between SSTi and coral Sr/Ca on annual mean scales indicates that Sr/Ca is a useful temperature proxy (Figure S4).

[12] We calculated δ18Oseawater by subtracting the thermal component from coral δ18O based on (i) the relative changes in Sr/Ca-SST and (ii) the relative changes of SSTi (following Cahyarini et al. [2008]). Our aim is to test and evaluate our δ18Oseawater reconstruction by using SSTi as an alternative temperature record to calculate δ18Oseawater (Figure 2). We use a δ18Ocoral-SST relationship of −0.2‰/1°C [Juillet-Leclerc and Schmidt, 2001] and our Sr/Ca-SST relationship of −0.058 mmol/mol per 1°C. Our δ18Oseawater reconstruction using Sr/Ca-SST (SSTi) results in a maximum seasonal amplitude of 0.69‰ (0.44‰) and a 1σ standard deviation of 0.17‰ (0.13‰; Figure 2). This agrees well with observed δ18Oseawater variations in the monitoring experiment.

Figure 2.

(a) Inferrerd SST = SSTi (dashed line; σTi = 0.38°), grid-SST (thin line [Reynolds et al., 2002]; σERSST = 0.25°C) and Sr/Ca-SST (thick grey line; σSr/Ca-SST = 0.8°C). (b) Annual mean reconstructed of δ18Oseawater (δ18Osw) from relative changes in SSTi (stippled line) and Sr/Ca-SST (thick bold line) compared to c) OLR (dash-dotted line) and (d) precipitation minus evaporation (P-E; thick grey line) taken from NCEP/NCAR reanalysis [Kalnay et al., 1996]. The 1-sigma errors for δ18Oseawater based on the analytical error (0.037‰) and the Sr/Ca-SST mean annual regression (r = −0.47; 0.13‰) are indicated in Figure 2b. SST, OLR and P-E time series were normalized.

[13] To provide error estimates for our δ18Oseawater reconstruction, we combine two different approaches. (1) The error of δ18Oseawater is estimated based on the analytical error of coral δ18O and Sr/Ca [Cahyarini et al., 2008] (Figure 2). This approach is based on the assumption that with the exception of the analytical errors, Sr/Ca is a perfect record of local SST, while δ18O is a perfect record of local SST plus δ18Oseawater. For bimonthly δ18Oseawater values, the error (1σ) is 0.09‰, for annual mean δ18Oseawater values the error reduces to 0.036‰ (see auxiliary material). (2) We estimate the statistical error resulting from the regression of δ18Oseawater with annual mean Sr/Ca-SSTi. This approach is based on the assumption that SSTi is a perfect record of local SST at the coral site, while Sr/Ca explains only 21% of the SST variance (based on r = −0.47 with SSTi). We obtain an 1σ error of 0.13‰, for mean annual δ18Oseawater. Note, however, that this error is based on the Sr/Ca-SSTi regression only, and we cannot repeat this exercise for coral δ18O due to the lack of long δ18Oseawater records. A combined error is obtained from the two error estimates above by using the square root of the sum of squares which results in an error of 0.16 ‰ (1σ, auxiliary material).

[14] Despite potentially large errors, our δ18Oseawater estimates based on Sr/Ca and SSTi show consistent variations. Averaging over decadal time scales and replication of Sr/Ca and δ18O measurements from multiple cores could help to reduce the uncertainty significantly in the near future.

4.2. Local Climate Interpretation of δ18Oseawater

[15] Our coral Sr/Ca and δ18O records show a trend towards more negative values between 1881 and 1994, indicating a long-term increase in SST (Figures 3a and 3b). Reconstructed δ18Oseawater shows pronounced decadal and interannual variability but no long-term trend (Figure 3c). Spectral analysis reveals periodicities ranging between 5–6 years and 18–25 years (Figure S6). Linear regression analysis of DJFM and mean annual values (Figure 1d) reveals that δ18Oseawater shows a positive (negative) correlation with OLR (rainfall and P-E; Figure S5). For the peak rainfall season our reconstruction varies coherently with the local rainfall record between 1933 and 1993 (Figure 3d). This provides confidence that our reconstruction is robust on long time scales.

Figure 3.

Coral proxy data for the period 1881 to 1994, shown as anomalies relative to their mean. (a) Sr/Ca, (b) oxygen isotopes and (c) reconstructed δ18Oseawater (δ18Osw; bottom). Sr/Ca has been scaled so that −0.058 mmol/mol corresponds to 1°C (see 4.1. for discussion). Bold lines indicate 11-year running mean. (d) Comparison of reconstructed δ18Oseawater (δ18Osw; in red) for the main rainfall season (December to March) with local rainfall from the Mayotte weather station (in blue). Both time series were detrended and normalized.

[16] The temporal changes in P-E and OLR are anticorrelated with the underlying SST (Figures 1e and S5), i.e., cooler mean SSTs are associated with higher rainfall. This is particularly clear in the 1960–1980 interval, where cooler mean SSTs are accompanied by positive (negative) P-E (OLR) anomalies (Figure 2).

[17] Consistent with this, coral Sr/Ca tends to be anti-correlated with δ18Oseawater indicating that periods of higher SST are associated with a negative freshwater balance (Figures 3a and 3c). This is particularly clear during the early phase of the pronounced wet spell that occurred between the 1960s and 1970s (Figure 2; note that the magnitude of these excursions is larger than our error estimates). Between 1961 and 1974, mean SSTs during the rainy season (NDJF) were only 27.72°C (compared to 28.16°C between 1951–1961, and 28.06°C between 1975 and 1998). Assuming a local thermodynamic relationship between SST and atmospheric convection over Mayotte, we would expect mean SSTs > 28°C to be associated with increased convection and a more positive freshwater balance [Timm et al., 2005, and references therein]. However, the reverse is the case. This implies that remotely forced, large-scale atmospheric variability influences the hydrological balance at Mayotte (see 4.3.).

4.3. Large-Scale Climate Teleconnections of δ18Oseawater

[18] The instrumental SST and Sr/Ca records of Mayotte are strongly related to ENSO (Figure S7). Here, we use our δ18Oseawater reconstruction as a proxy for changes in the hydrological balance to test for ENSO influences (Table S1 and Figure S2). Correlation maps of seasonal and mean annual values for the period 1950 to 1993 between δ18Oseawater and global SLP show similar, but weaker correlation patterns as with regional-scale rainfall data from the western Indian Ocean south of 10°S (Figures 1a, 1b, and S2) resembling the ENSO SLP pattern. We observe an enrichment of δ18Oseawater in response to ENSO warm events on both seasonal and mean annual scales (Figure S7). This is in contrast to the findings from central Indian Ocean corals (Chagos Archipelago) where δ18O indicates a positive freshwater balance in response to ENSO warm events over the last 30 years [Pfeiffer et al., 2006]. However, our results agree with earlier studies, in which the western Indian Ocean south of 10°S shows an opposite ENSO response compared to the equatorial Indian Ocean/East Africa [Hastenrath et al., 1993; Richard et al., 2000; Zinke et al., 2004]. The western pole of maximum rainfall anomalies during warm ENSO events is located between 10°S–10°N, 50–70°E which would favor moisture convergence equatorwards of 10°S and a possibly dryer atmosphere to the south [Neelin et al., 2003]. The correlation with the Nino3.4 Index is significantly higher in the period 1975–1994 (r = 0.61, p < 0.006) than between 1951–1974 (r = 0.29, p < 0.19; Figure S7). This is consistent with results from the Chagos Archipelago where the ENSO correlation increased after 1975 [Pfeiffer et al., 2006]. However, the El Nino response at Mayotte (drier) and Chagos (wetter) oppose each other. This agrees with the expected rainfall pattern across the Indian Ocean from remote ENSO forcing. Our findings add to the growing evidence of a strengthening of the ENSO signal in Indian Ocean rainfall since 1975 [Richard et al., 2000; Pfeiffer et al., 2006; Zubair and Ropelewski, 2006].

[19] The correlation map also indicates high correlations of the instrumental and δ18Oseawater record with the extratropical North and South Pacific SST and SLP resembling the Pacific Decadal Oscillation (PDO) pattern (Figures 1a, 1b, and S2). The PDO is the dominant mode of interdecadal variability in Indo-Pacific SST/SLP [Deser et al., 2004, and references therein]. Mayotte is located in the region of most significant correlations with the PDO [Deser et al., 2004; Crueger et al., 2008]. Linear least squares regression indicates a strong relationship between our δ18Oseawater record and the PDO between 1950–1994 on mean annual and seasonal time scales (r = 0.6, p < 0.003: Figure S8). The observed periodicity in the δ18Oseawater record ranges between 18–25 years, adding further evidence for an atmospheric bridge between the North Pacific and the Indian Ocean.

5. Conclusions

[20] We developed a 113 year long, bimonthly resolved coral record of δ18Oseawater from Mayotte (south-western Indian Ocean). The δ18Oseawater record indicates that the precipitation-evaporation balance varies on time scales of 5–6 years and 18–25 years. We observe a warm/dry (cool/wet) relationship between SST and δ18Oseawater, suggesting that atmospheric variability is remotely forced by large-scale climate anomalies in the Indian Ocean and the tropical/extratropical Pacific Ocean. The observed interdecadal variability of δ18Oseawater and its spatial signature indicate a possible link to the Pacific Decadal Oscillation, while interannual variability is clearly related to the El Niño-Southern Oscillation. Warm ENSO events are associated with a negative freshwater balance. The ENSO correlation increases after 1975. This agrees with coral data from the Chagos Archipelago (central Indian Ocean) where the ENSO correlation also increased after 1975. However, the El Niño response at Mayotte (drier) and Chagos (wetter) oppose each other. This agrees with the expected rainfall pattern across the Indian Ocean from remote ENSO forcing. Our case study demonstrates that a much denser network of δ18Oseawater reconstructions is crucial for the understanding of spatial patterns in hydrological conditions.


[21] The coral was drilled by the TESTREEF-group in 1994. Seawater monitoring was carried out by DAF/SPEM Mayotte and F. Seguin. B. Nicet provided the SSTi and SST logger data. Oxygen isotope analysis were carried out at IFM-GEOMAR and the University of Erlangen. B. Schnetger (University of Oldenburg) provided measurement capacity at the ICP-AES. This research was supported by the German Science Foundation and by the German government (KIHZ). We acknowledge the Nederlandse Organisatie voor Wetenschappelijk Onderzoek (NWO-ALW grant 854.00.034 to Jens Zinke). OT was supported by the Japan Agency for Marine-Earth Science and Technology (JAMSTEC) through its sponsorship of the International Pacific Research Center. MP was sponsored by the DFG (grant PF 676/1). CD was supported by the DFG (Leibniz award). This is IPRC publication 552 and SOEST publication 7565.