Introduction

Agriculture in the Mekong River Delta (MRD) is the key to sustaining the worldwide food supply. Vietnamese rice production contributed 7.5% of the total global rice exports in 2017 CE, which was supported by its production in the MRD1,2. However, recent severe hydrological disasters with frequent floods and droughts threaten the MRD2,3. Flood events seriously impact agriculture and health hazards since the large cities and agricultural fields in the MRD are at low elevations4. Drought events, such as those in 2015 and 2019 CE, cause collapse agricultural systems by causing moisture loss from the soil and intruding saline groundwater into agricultural fields5. Over 4600 km2 of rice field was damaged, and 200,000 households suffered from safe water shortages in 2019 CE6. Understanding the dynamics of rainfall changes and the Mekong River flow rate is crucial for sustainable food supply7,8,9.

Previous studies10,11 approached the hydrological changes in the MRD using the observation and land-based proxy records of rainfall in the Indochina peninsula (e.g., tide gage, tree rings, observatory data10,11,12,13,14,15). The El Niño Southern Oscillation (ENSO) has an important role in river runoff and rice production in the MRD, as El Niño (La Niña) events would induce the short (long) wet season with a low (high) amount of rainfall2,15,16,17. The light rainy season during El Niño allows intrusions of saline water16. The heavy rainy season causes flood events16. Food sustainability requires how the hydrology event in MRD responds to the ENSO in the future warming earth2, as the ENSO behaviour has been changing in recent decades18. However, rainfall observations are spatiotemporally heterogeneous. Most of the proxy records and tidal data are far from the river mouth of the MRD13,19,20. These limitations of available records and the spatial heterogeneity of rainfall make it complex to reveal the past hydrological changes of the MRD. Here we investigate the frequency and intensity of hydrological changes during the last century using a coral proxy record. The chemical composition in a drilled coral core allows us to calculate an oxygen isotopic ratio in seawater (δ18Osw, a proxy for hydrology changes21,22,23) which enables a reliable reconstruction of paleolimnology24 and palaeoceanography25,26. Reconstruction of the interannual to decadal hydrological change using the oceanic proxy could imply mechanisms behind the flood and drought events with the anthropogenic27 and natural12,13 hydrological changes.

We drilled a coral core (CD-4, Porites. sp, 150 cm in length, Fig. S1) on 8th June 2006, from the southwestern coast of Con Dao Island (8° 33′ 25.7″ N, 106° 33′ 00.7″ E, ca. 90 km south of the MRD, Figs. 1a, b, and S2)28. Monthly resolved subsamples were obtained from sliced coral slabs for measurements of Sr/Ca (a proxy for sea surface temperature, SST29) and oxygen isotope ratio in coral skeletons (δ18Ocoral) (see Methods). The δ18Osw was calculated from the measured δ18Ocoral by subtracting the temperature component inferred from Sr/Ca using the centring method30. We took the annual minima of δ18Osw (min-δ18Osw) as an indicator for maximum freshening. The min-δ18Osw was reported as an anomaly relative to the mean during the reference period (1981–2000 CE).

Figure 1
figure 1

Calibration of coral-based records with meteorological data. (a,b) Rainfall (GPCP version 2.3) and salinity (EN4.2.145) in October from 1980 to 2005 CE. The star symbol indicates Con Dao Island. The red line shows the Mekong River. The map was generated by the GMT 4.5.1246. (c) Monthly resolution coral records of δ18Ocoral (blue), Sr/Ca ratio (red), and δ18Osw (purple) from December 1923 to June 2006 on Con Dao Island. GPCP version 2.3 rainfall from 1979 to 2020 (green) was derived from the southeastern part of the Indo-China Peninsula (7.5–15°N, 105–110°E, green box in panel a). (d) Min-δ18Osw (purple) and maximum rainfallGPCP in the MRD (green). Thin and thick lines show the annual and 3-year moving averaged data, respectively. Grey shading and dashed lines indicate the reference interval (1981–2000 CE) and thresholds used to define heavy/light rainy seasons. (e) A cross plot analysis between the min-δ18Osw and the maximum rainfallGPCP (grey: annual data; black: 3-year moving averaged data). Dashed lines indicate linear regression lines.

Full size image

The δ18Osw was calibrated with the combined rainfall data from observations and satellite data (derived from the Global Precipitation Climatology Project dataset, GPCP version 2.331) over the MRD from 1979 to 2020 CE (Fig. 1a). The min-δ18Osw was compared with the annual maxima of rainfall derived from the GPCP dataset (maximum rainfallGPCP, reported as an anomaly relative to the reference period). The annual maximum rainfall (maximum rainfallrecon) was reconstructed following the calibration work to the min-δ18Osw and composited (maximum rainfallrecon+GPCP) with the maximum rainfallGPCP in the MRD. Based on the maximum rainfallrecon+GPCP (Fig. 2a), the occurrence rates of years with rainfall beyond one standard deviation in the reference period (heavy and light rainy seasons) were estimated using the kernel technique32. The maximum rainfallrecon was compared with the coral-based indexes of the Niño warm pool (NWP) and Niño cold tongue (NCT) as indicators of ENSO in the central Pacific (CP) and eastern Pacific (EP) Oceans18. We investigated the teleconnection of the ENSO by comparing the maximum rainfallrecon+GPCP with the thermal gradient between the western (0–10°N, 130–150°E) and central Pacific Oceans (5°S–5°N, 160–210°E)33.

Result and discussion

Calibration of the coral record from Con Dao Island

The SST-related proxies of Sr/Ca and δ18Ocoral showed 82 seasonal cycles from December 1923 to June 2006 (Fig. 1c). The range of calculated δ18Osw was from –1.11 to 0.97‰. δ18Osw during 1979–2005 CE correlated with in situ salinity on Con Dao Island and rainfall changes in the MRD over a 95% confidence level (CL) (Fig. S3; δ18Osw vs. salinity: R = 0.36; n = 291, δ18Osw vs. rainfall: R = − 0.32; n = 324). The min-δ18Osw, which ranged from − 0.84 to 0.51‰, negatively correlated with the maximum rainfallGPCP (annual data: R = − 0.37; n = 27, over 95% CL; 3-year moving average data: R = − 0.73; n = 27, slope: − 76.8 ± 14.2 mm × month−1/‰, over 99% CL) (Fig. 1d,e). The min-δ18Osw positively correlated with SST changes in the rainy season inferred from Sr/Ca through the entire coral record (annual data: R = 0.52; n = 82; over 99% CL; 3-year moving averages: R = 0.50; n = 82; over 99% CL; Fig. S4).

Our coral-based δ18Osw record captures year-to-year changes in maximum rainfall in the MRD (Fig. 1d,e). Our calibration agrees with previous works showing that the coral-based δ18Osw along the southern Vietnamese coast reflects rainfall changes34 and significantly decreases in flood years28. Since the oxygen isotope ratio in raindrops (< –5‰35,36) is much lower than δ18Osw (ca. 0‰37), river runoff and rainfall in the rainy season decrease the values of δ18Osw in the surrounding ocean. Ocean currents (e.g., upwelling) along the southern coast also may affect salinity around Con-Dao Island. Our calibration work28 (Phan et al. 2017) on the Con Dao Island showed that the influence of the ocean current is smaller than rainfall in the wet season. The correlation between min-δ18Osw and SST changes in the rainy season results from an air-sea interaction over the western Pacific Ocean38. Hence, the coral-based δ18Osw records the maximum rainfall in the MRD through the entire coral time series.

ENSO footprint in the Mekong River Delta hydrology

A spectral analysis39 to the maximum rainfallrecon (range: from − 27.1 to 33.4 mm/month; Fig. S5) demonstrated that the major periodicity was the interannual (4.1 and 6.5 years/cycle: over 90% CL) and decadal (9.6 years/cycle: over 99% CL; 9–10 years/cycle: over 90% CL, Fig. 3a) time scales. The major periodicity of maximum rainfallrecon (Fig. 3a) agreed with that of the NWP index (6.5 and 9–10 years/cycle; over 99% CL; Figs. 2b, 3b) but not the NCT index (13.6, 8.6–9, and 6.3–5 years/cycle; over 99% CL; Fig. S6). The NWP index with a one-year lead negatively correlated with the maximum rainfallrecon (Fig. 3c: one-year leading NWP index vs. maximum rainfallrecon: R = − 0.38; n = 82; over 99% CL). The field correlation analysis showed the negative correlation between the detrended SST in the CP region with one year leading and the maximum rainfallrecon+GPCP (Fig. S7), which supported the result of the lead-lag correlation (Fig. 3c).

Figure 2
figure 2

Time series analysis of the maximum rainfall. (a) The maximum rainfallrecon+GPCP (black line) derived from the min-δ18Osw and rainfallGPCP. Grey shades indicate the 90% confidence intervals of maximum rainfall reconstruction. Cross plots indicate the year with the heavy (blue) or light (red) rainy season. Note that rainfall in the heavy rainy season (blue cross plot) shows an increasing trend. The occurrence rates of heavy (blue line) and light (red line) rainy seasons are indicated above and below panel a, respectively. Light-coloured shades indicate the 90% confidence bands of occurrence rates. (b) The NWP index18 (thin and thick lines: annual and 3-year moving averaged data). (c) A Morlet wavelet analysis for the maximum rainfallrecon+GPCP. The solid black lines indicate a 95% confidence level. White lines show the cone of influence, where edge effects reduce the variance.

Full size image
Figure 3
figure 3

Relationship between the maximum rainfallrecon and the CP ENSO. (a,b) REDFIT spectral analyses39 for the maximum rainfallrecon and the NWP index. Red lines indicate bootstrap confidence intervals for the spectral power (solid: 99%; dash: 95%; dot: 90%). The REDFIT spectral analyses were performed using PAST software version 4.0447. (c) A lead-lag correlation analysis between the NWP index and the maximum rainfallrecon. Shading areas indicate the confidence bands of correlation coefficients (light colour: 99%; thick colour: 95%). (d) A wavelet coherency between the maximum rainfallrecon and the NWP index. The solid black lines indicate a 95% confidence level. White lines show the cone of influence, where edge effects reduce the variance.

Full size image

La Niña and CP El Niño events would have led to the heavy and light rainy seasons in the MRD, respectively (Fig. 3). Anomalous development of the low-level circulation over the western Pacific Ocean during La Niña and CP El Niño events causes anomalous rainfall in the rainy season in the MRD, as discussed in meteorology and simulations16,40. The development of anomalous cyclonic circulation over the western Pacific Ocean during La Niña events intensifies moisture transportation to the Indo-China Peninsula11,17. Cold anomalies over the Indian and central Pacific Oceans enhance the Walker circulation during La Niña events, which increases the rainfall amount in the rainy season17. During CP El Niño events, the rainfall amount in the rainy season decreases due to the anomalous anticyclonic circulation over the western Pacific Ocean and the weakening of the Walker circulation. CP El Niño events have a different teleconnection pattern from EP El Niño events18,41,42,43. Since the anticyclone over the western Pacific Ocean appears further west during CP El Niño events compared to EP El Niño events37, CP El Niño events significantly influence maximum rainfall in the MRD.

Inconsistent trends of the hydrological events

The impacts of La Niña and CP El Niño on the frequencies and intensities of hydrological events in the MRD were not consistent throughout the past century. A wavelet analysis revealed that the major periodicity of the maximum rainfall has shifted from decadal to interannual time scales since the 1960s (Figs. 2c and S5). The periodicity shift in the maximum rainfall was observed in a wavelet coherency between the maximum rainfallrecon and the NWP index (Fig. 3d). These wavelet analyses (Figs. 2c, 3d, and S5) suggested that the frequent ENSO events in the CP region18,41 likely increased the frequency of maximum rainfall changes in the MRD in recent decades. The recent highly frequent change in the maximum rainfallrecon was confirmed by the increasing trends of occurrence rate and rainfall amount of the heavy rainy season throughout the last 100 years (rainfall amount: two-sided p < 0.01; n = 18; occurrence rate: one-sided p = 0.01; n = 18; Fig. 2a). In contrast to the heavy rainy season, these trends of occurrence rate and rainfall amount were not confirmed in the light rainy season (rainfall amount: two-sided p = 0.94; n = 20; occurrence rate: one-sided p = 0.22; n = 20; Fig. 2a).

The correlation analyses implied that the inconsistent trends between the heavy and light rainy seasons were closely related to increases of SST in the western Pacific Ocean (R = 0.36; n = 97; over 99% CL; Figs. 4a and S8) and the western Pacific thermal gradient (R = − 0.29; n = 97; over 99% CL, Fig. 4). The warming rate in the western Pacific Ocean was faster than that in the CP region and thus has intensified the thermal gradient over the western Pacific Ocean during La Niña events since the 1980s33. The intensified thermal gradient during La Niña events further strengthens the Pacific Walker circulation, which amplifies heavy rainfall in the MRD during La Niña events.

Figure 4
figure 4

Comparing the maximum rainfall with SST in the Pacific Ocean. (a) A field correlation of the maximum rainfallrecon+GPCP and SST (Extended Reconstructed Sea Surface Temperature, ERSSTv548) during 1924–2020 CE. Note the positive correlation in the western Pacific Ocean (blue box) and insignificant correlation in the CP region (red box). A green star indicates our sample site. The map was generated by the GMT 4.5.1246. Correlations were computed with the web application KNMI Climate explorer. Correlations not significant at the 10% level are masked out. (b) The upper panel indicates 3-year moving averages of the SST anomaly in the western (blue line; a blue box in panel a) and central Pacific Ocean (red line; a red box in panel a). SST data are derived from ERSSTv548. The lower panel indicates the western Pacific thermal gradient. (c) The maximum rainfallrecon+GPCP (black line). Grey shades indicate the 90% confidence interval. (d) A lead-lag correlation analysis between the western Pacific thermal gradient and the maximum rainfallrecon+GPCP. Shaded areas indicate the confidence bands of correlation coefficients (light colour: 99%; thick colour: 95%).

Full size image

Conclusion

In reference to our coral-based δ18Osw record, we revealed inconsistent trends of frequency and intensity between the heavy and light rainy seasons, which likely impact socioeconomics and human health in the MRD. The heavy rainy season recurs more frequently and has more serious impacts in recent decades, resulting from the more frequent occurrence of CP El Niño events and the intensified western Pacific thermal gradient. Hence, people dwelling in the MRD will face a higher risk of severe flood events. However, there was no clear trend in the occurrence rate and rainfall amount of the light rainy season, although catastrophic drought hit the MRD in 2015 and 2019 CE. We concluded that the hydrological changes in the MRD are sensitive to the CP ENSO events. The hydrological response to the ENSO behaviour is inconsistent throughout the last century, which will improve the prediction of the future impact of the flood and drought.

Methods

Sample preparation and proxy measurement

The coral core (CD04) was collected from Con Dao Island, located 90 km far from the MRD. The core was sliced into 5 mm thickness. X-ray images of slices were taken using a digital X-ray scanner to visualise the coral growth transaction and annual bands.

Monthly coral skeletal powdered subsamples for geochemical analyses (for Sr/Ca and δ18Ocoral) were drilled at ca. 1 mm intervals along the maximum axis of coral growth. δ18Ocoral of powdered subsample, 20–40 μg each, was analysed using a stable isotope ratio mass spectrometer (Finnigan MAT253) and an automated carbonate device (Kiel IV) at Hokkaido University. The coral skeletal powdered subsamples, 90–110 μg, were dissolved in 25% nitric acid and diluted to a Ca concentration of 7 ppm with ultrapure water for Sr/Ca determination48 on an inductively coupled plasma optical emission spectrometer, Thermo Scientific iCAP 6200 at Hokkaido University. The one-sigma precisions for δ18Ocoral and Sr/Ca data were ± 0.05‰VPDB and ± 0.07% RSD, respectively.

We calculated the δ18Osw value (relative to Vienna Standard Mean Ocean Water, VSMOW) based on paired measurements of Sr/Ca and δ18Ocoral using the centring method30. Centred values of Sr/Ca and δ18Ocoral (values relative to the mean of the entire records) were used for the δ18Osw estimation to avoid the large uncertainty source for this estimation26. We used the slope of our Sr/Ca-SST dependency on Con Dao Island (− 0.0504 ± 0.004 mmol × mol−1/°C)28 and the slope value of the δ18Ocoral-SST relationship (− 0.18‰VPDB/ºC30,50). The one-sigma uncertainty of calculated δ18Osw was ± 0.06‰. The age model for the coral records was developed using the maxima and minima of the Sr/Ca values in any annual cycle as anchoring points. The anchoring points were tied to the maxima and minima of SST on Con Dao Island28. We used linear interpolation to obtain monthly resolution time series using AnalySeries software51, version 2.0.8.

Statistical analyses

We estimated the mean SST in the rainy season and obtained the annual minima of δ18Osw (min-δ18Osw). Min-δ18Osw was reported as the anomaly relative to the reference period (1981–2000 CE). The reference period was defined as the 20-year period in which the min-δ18Osw and rainfallGPCP overlapped. Our coral record was compared with rainfallGPCP (7.5–15°N, 105–110°E, green box in Fig. 1a), the NWP index (index for the CP ENSO)18, and the NCT index (index for the EP ENSO)18. We estimated the annual rainfall maxima (maximum rainfallrecon, values relative to 1981–2000 CE) from the 3-year moving average of the min-δ18Osw based on the slope between the 3-year moving average datasets of min-δ18Osw and the maximum rainfallGPCP (− 76.8 ± 14.2 mm × month−1/‰, Fig. 1e). The uncertainty of maximum rainfallrecon+GPCP was estimated using the Monte Carlo approach. We considered an error for the slope between the min-δ18Osw versus maximum rainfallGPCP and the differences between maximum rainfallGPCP versus maximum rainfallrecon. We repeatedly estimated maximum rainfall by adding random numbers known from the slope errors in the 20,000 times loop.

We estimated the periodicity using REDFIT spectral analysis39 in PAST software version 4.0447. The correlation coefficient and confidence level were estimated using the bootstrap technique and the BCa method in the “wBoot” package in R software52. The occurrence rates of the heavy/light rainy season were estimated using the kernel technique in Caliza software32. The heavy/light rainy season was defined as a year with maximum rainfall beyond the threshold (± 15.4 mm/month). The threshold value was estimated as the average value plus/minus one standard deviation in maximum rainfallrecon+GPCP during the reference period (1981–2000 CE). The bandwidth for the estimation of the occurrence rate was set to 11 years to detect decadal-scale variations. The confidence interval of the occurrence rate was determined using a bootstrap technique and a percentile-t method. Events were simulated with replacement and used for occurrence rate calculations, which were repeated 5000 times. Wavelet analyses were conducted using the “biwavelet” package in R software52. We surveyed the significance of trends in the occurrence rate and rainfall amount of the heavy/light rainy season based on the Cox-Lewis test in Caliza software32 and the Mann–Kendall trend test in R software52, respectively.