## Introduction

Southeast Asia (SEA) harbours an astonishing diversity of stony corals, comprising approximately 28% of the global total, located in the Maritime Continent1,2. Local ecological threats, such as coastal development, overfishing/destructive fishing and pollution, detrimentally impact 95% of the coral reefs in SEA and are amplified by large-scale global pressures2. Over the past 40 years, the Maritime Continent has undergone intense economic development and urbanization, drastically increasing human-induced pressures on the marine coastal ecosystems3,4,5,6. In Borneo, rapid urbanization of coastal zones and expansion of palm oil plantations has resulted in some of the highest levels of deforestation among all humid tropical regions of the world7,8,9,10. Approximately 26% of old-growth rainforest has been lost since the early 1970s7,8,9,10. Anthropogenic land-use change can severely impact the erosion potential of the hinterland soils, as well as the freshwater and sediment discharge into the nearshore environment, which in turn alters water quality on coral reefs11,12.

In northern Borneo, natural climatic events paired with currently unquantified contributions from climate change may exacerbate the regional anthropogenic impacts on coral reefs. The regional climate is strongly influenced by the Asian-Australian monsoon system13,14,15. The winter monsoon usually lasts from November to March, characterized by strong northeasterly winds, colder sea surface temperatures (SSTs), wetter conditions and lower salinities (Fig. 1; see Supplementary Fig. S1). The summer monsoon, usually from May to September, is distinguished by reduced southwesterly winds, highest SST, lower precipitation and higher salinity (Fig. 1; see Supplementary Fig. S1). However, precipitation in northern Borneo is more evenly distributed across the annual cycle, less monsoonal and dominated by intraseasonal and interannual variability16. Intra-monsoon rainfall occurs in April and October in any given year, with the latter often leading to wetter conditions and lower salinities in the southern SCS17. The region’s climate variability is influenced by global interannual phenomena that include the El Niño-Southern Oscillation (ENSO), the Indian Ocean Dipole (IOD), the intraseasonal Madden-Julian Oscillation (MJO) and the Borneo Vortex14,16. In SEA, drier than normal conditions are associated with El Niño, whereas wetter than normal conditions tend to be associated with La Niña14,18. Apart from this general relationship, the ENSO-precipitation connectivity in SEA shows spatial and seasonal variation. The relationship between Malaysian precipitation and ENSO strengthens from September to March of the following year19. ENSO events can generate severe floods, droughts and drought-related forest fires, which can have devastating socio-economic and ecologic consequences20,21,22. It has been suggested that the Pacific Decadal Oscillation (PDO) induces decadal precipitation and salinity anomalies in the South China Sea (SCS) on top of interannual ENSO events23. Unfortunately, the instrumental data coverage for the Miri-Sibuti Coral Reefs National Park (MSCRNP), situated off the Sarawak coast, is poor. Therefore, the lack of seasonally resolved climatic baseline data limits our understanding of the spatio-temporal change of the coral reefs’ physical environment in MSCRNP, and its regional and large-scale drivers.

Massive scleractinian corals like Porites spp. are a key archive for past climate and environmental change, including SST and hydrologic balance, and can augment scarce instrumental data24,25. The Sr/Ca ratio in coral skeletons is considered to be the most reliable proxy for SST26,27,28,29,30, though some studies have shown it to be affected by biomineralization processes, so called “vital effects”, which can influence Sr/Ca ratios independently from water temperatures31. Another method is to reconstruct SSTs from coral δ18O26. The use of δ18O for paleothermometry was pioneered by Urey and Epstein, who demonstrated a reverse relationship between δ18O and temperature values32,33,34. Weber and Woodhead35 were the first to apply it to corals in 1972, and since then δ18O from coral skeletons has been widely used as a tracer for up to monthly resolved SST36,37. However, coral δ18O varies as a function of two major variables, temperature and isotopic composition of the seawater, with the latter being affected by rainfall, evaporation, advection of water masses and freshwater runoff38,39,40,41. As such, past changes in the hydrologic balance for inshore coastal waters can be provided by using paired Sr/Ca and δ18O analysis on the same samples. This pairing quantitatively separates the effects of SST from those of the ambient seawater’s oxygen isotopic composition on the skeletal δ18O and provides reconstructed δ18Oseawater (sw), related to changes in the hydrological balance and salinity42,43,44.

The aim of this work was to develop a spatio-temporal reconstruction of SST, oxygen isotopic composition of seawater tracking sea surface salinity (SSS) and river runoff dynamics based on a multi-proxy analysis of coral cores from two sites in MSCRNP, Borneo (Malaysia). In September 2016 coral cores were retrieved from Eve’s Garden (EG) and Anemone’s Garden (AG). These two patch reefs lie along an inshore to offshore gradient and are therefore at different depths, distances from shore and distances from the Miri estuary and the Baram River mouths45 (Fig. 2; Table 1). Sr/Ca ratios in coral skeletons were utilized to reconstruct local SSTs. Combined Sr/Ca and oxygen isotope analysis enabled us to reconstruct the oxygen isotopic composition of the seawater related to changes in the hydrological balance and salinity, covering the past three decades. This allowed for an unprecedented comparison to instrumental data from both land and sea from 1982 to 2016. All coral proxies were validated against instrumental data products of SST, SSS, river discharge and precipitation. Our results, were further investigated for signals of large-scale climatic teleconnections in northern Borneo, related to ENSO and the PDO.

## Results

### Monthly interpolated geochemical proxy time series

The monthly interpolated time series of AG and EG coral Sr/Ca and δ18O revealed a distinct seasonality with higher values during the winter season (December to February) and lower values in the summer season (June to August) (Fig. 3). Spatial comparisons of the two reefs between 2006 and 2016 found no difference between Sr/Ca ratios at AG and EG (see Supplementary Table S1). For mean coral δ18O, AG had lower values than EG (Fig. 3; see Supplementary Table S1). Linear correlations between Sr/Ca and δ18O in both AG and EG were highly significant (p < 0.0001). The correlation between Sr/Ca and δ18O at AG (r = 0.81) was stronger than at EG (r = 0.62).

AG Sr/Ca ratios displayed an overall decrease of 0.06 mmol/mol from 1982 to 2016. Seasonal Sr/Ca amplitudes also decreased at AG from approximately 1993, although there were higher winter values in some years (e.g. 2014 to 2016; Fig. 3a). There was no trend in Sr/Ca at EG between 2006 and 2016. The AG δ18O record showed an overall decrease of 0.45‰ and had stronger interannual and decadal variability than the Sr/Ca record (Fig. 3b). In the long term, the annual mean δ18O values exhibited slight shifts on an approximately decadal time scale with a prominent shift to lower values in the mid to late 1990s (Fig. 3b). From 1982 until 2006, δ18O at AG decreased by 0.46‰. Yet from 2006 to 2016, the mean δ18O at AG increased (AG = 0.2‰, p = 0.04) while EG showed no significant trend (EG = 0.14‰, p = 0.12).

### Monthly calibration of Sr/Ca with sea surface temperature (SST) and SST variability

Ordinary least squares regressions between coral Sr/Ca and AVHRR-OISSTv246,47 were used to reconstruct absolute SST from coral Sr/Ca (Table 2). Our observed strong correlation between Sr/Ca and SST at the study sites indicates that growth rate and vital effects had an insignificant effect on skeletal Sr/Ca. The local proxy-SST relationships varied between −0.043 mmol/mol °C−1 in AG and −0.039 mmol/mol °C−1 in EG, which is in agreement with the range of published slopes26,29 that lie between −0.04 and −0.08 mmol/mol °C−1. Our regression slopes were also comparable to other coral core studies in the northern SCS48,49,50. The root mean square error (RMSE) of individual monthly interpolated reconstructed SST for the AG and EG record was 0.64 °C and 0.55 °C, respectively.

Reconstructed monthly interpolated SST across the two sites varied between 26.5 °C and 30.5 °C in Sr/Ca-SST and between 25.9 and 30.8 °C in AVHRR-OISSTv2 data for the 0.25° × 0.25° grid46,47. Both Sr/Ca-SST (0.85 ± 0.13 °C) and instrumental SST (0.41 ± 0.18 °C) showed a long-term warming trend (Fig. 4a). AVHRR-OISSTv2 had larger seasonal amplitudes, especially before 2005, with lower winter SSTs and higher summer SSTs (Fig. 4a). Within the reconstructed SST data, seasonal amplitudes declined after 1993, with a shift to higher winter temperatures and smaller seasonal amplitudes (see Supplementary Fig. S2). Occasionally, winter Sr/Ca-SSTs reached values similar to pre-1993, for instance between 2002–2003 and 2014–2016 (Fig. 4b). The coral-derived SST anomalies in both AG and EG relative to the 2006 to 2016 mean (years of coral core overlap) mirrored the interannual variations of instrumental SST anomalies well (Fig. 4b). Both AVHRR-OISSTv2 and our coral Sr/Ca-SST records showed no significant correlation with the Niño3.4 index, yet both the AVHRR-OISSTv2 and AG Sr/Ca-SST had significant relationships with Niño4 in Boreal winter to spring (see Supplementary Table S3).

AG Sr/Ca anomalies relative to a 1982 to 2016 mean were converted to SST using the local calibration slope (−0.04 mmol/mol °C−1; Fig. 4b), a mean slope of −0.06 mmol/mol °C−1 and a slope of −0.08 mmol/mol °C−1 to estimate an error range of the calibration26,29. The coral-derived SST anomalies followed the trends of instrumental SST anomalies reasonably well (Fig. 4b). However, the regression slopes of −0.08 mmol/mol °C−1 and −0.06 mmol/mol °C−1 resulted in SST anomalies that closely aligned with AVHRR-OISSTv2, while the locally derived −0.04 mmol/mol°C−1 slope estimate deviated the most. Between 1982 and 1994, the AG Sr/Ca-SST anomalies (relative to the 1982–2016 mean) were 0.5 °C lower than between 1994 and 2016. Sr/Ca-SST anomalies fluctuated around the 1982–2016 mean from 2002–2005, as well as from 2009–2010.

### δ18Osw reconstruction and relationship with salinity, rainfall and river discharge

Mean reconstructed δ18Osw between 2006 and 2016 was similar at EG (−0.38 ± 0.26‰) and AG (−0.36 ± 0.22‰) (Fig. 5a; see Supplementary Table S1). Mean annual and detrended δ18Osw in AG and EG were highly correlated (r = 0.93, p < 0.001). There was also a positive and significant correlation between mean annual reconstructed δ18Osw and EN4 SSS at AG (r = 0.72, p < 0.001) and EG (r = 0.68, p = 0.046). The correlation remained significant for seasonal mean values (see Supplementary Table S4). δ18Osw in AG and EG mostly mirrored the long-term and interannual to decadal variability of the EN4 SSS data between 1982 and 2016, with higher values matching higher salinity (Fig. 5a). EG δ18Osw values showed a higher variability in individual years than AG between 2006 and 2016. AG δ18Osw between 1982 and 2016 indicated a decrease of 0.27 ± 0.04‰ (p < 0.001), while EN4 SSS decreased by 0.32 ± 0.06 psu (p < 0.001). Both AG and EG δ18Osw displayed decadal variability between 2006 and 2016 in agreement with EN4 SSS (Fig. 5a). Mean annual δ18Osw at AG and EG, and EN4 SSS were negatively correlated with the local precipitation and Marudi river discharge. All correlations were statistically significant with the exception of EN4 SSS with Marudi discharge (Table 3).

Anomalies of reconstructed δ18Osw and EN4 SSS were calculated relative to the 2006 to 2016 means (Fig. 5b). With some exceptions, AG’s δ18Osw followed the interannual variability and trends observed in EN4 SSS since 1982 (Fig. 5b). The comparison of the anomalies with ENSO indices Niño3.4 and Niño451 showed good agreement between the coral-derived δ18Osw (AG and EG) and ENSO indices. For example, high δ18Osw matched the 1997/98 El Niño and was followed by a 1998/99 La Niña phase and a shift to low δ18Osw in the coral record (Fig. 5b; see Supplementary Fig. S2). Generally, both AG and EG recorded similar interannual variability with slightly larger magnitudes for EG (Fig. 5b). Both AG and EG δ18Osw also largely mirrored interannual variations in river discharge and precipitation anomalies (Fig. 5c,d). Marudi discharge indicated a significant increase of 641 ± 121 m3/s between 1989 and 2015 (p < 0.001) while Miri precipitation recorded a non-significant increase of 56 ± 38 mm month−1 (p = 0.15) between 1992 and 2016.

### Regional and large-scale teleconnection of reconstructed δ18Osw with climate parameters

Coral δ18Osw at AG and EG, precipitation from the local weather station, Marudi river discharge and EN4 SSS were correlated with ENSO SST indices Niño3.4 and Niño451 and the PDO52 (Table 4). All correlations with ENSO indices, exhibited statistical significance averaged over a 3-month period (Table 4). All data, except for precipitation and river discharge (negative correlation), were significantly positively correlated with Niño3.4 and Niño4 (Table 4). Miri precipitation and AG δ18Osw had the highest correlations in the February to April (FMA) season, while EN4 SSS had it between March to May (MAM) (Table 4). EG δ18Osw showed the highest correlations with Niño3.4 and Niño4 between November to January and December to February, respectively (Table 4). Miri precipitation and Marudi river discharge indicated the highest, statistically significant (p < 0.05) correlations between February and April and August to October (Table 4).

Seasonal δ18Osw at AG (April to June) and EG (February to April) were positively correlated with the PDO52 between 1982 and 2016 and 2006 to 2016, respectively (Table 4). The same holds for mean annual δ18Osw at AG (r = 0.56, p = 0.002) and EG (r = 0.75, p = 0.02). Marudi discharge (1989–2015) and Miri station precipitation (1992–2016) were significantly negatively correlated with the PDO on seasonal time scales (February to April; Table 4). The correlation was positive and lower between EN4 SSS and the PDO, yet significant (p < 0.05) for both seasonal (Table 4) and mean annual time scales (r = 0.38, p = 0.038).

Spatial correlations of AG and EG δ18Osw and local precipitation against OISSTv2 displayed the typical ENSO “horseshoe” SST pattern at both seasonal and mean annual time scales (Fig. 6a–d). AG and EG δ18Osw were positively correlated with the central and eastern Pacific SST and negatively correlated with the northwestern, central and southwestern Pacific SST. Further there was a positive correlation with the northeastern and central Indian Ocean SST and negative correlation with the southeastern Indian Ocean SST (Fig. 6a–d).

Correlations of local rainfall (Fig. 6e,f) and river discharge (Fig. 6g,h) indicated the same pattern in the Pacific, except with an inverse relationship between precipitation, river discharge and SST. There was significant negative correlation between Miri precipitation and river discharge with northern and central Indian Ocean SST for both seasonal (FMA) and annual means (Fig. 6e,f). Additionally, seasonal means of local precipitation and river discharge displayed positive correlation with southeast Indian Ocean SST. The spatial correlations of Miri station precipitation with SST also revealed the “horseshoe” ENSO SST pattern (Fig. 6g,h).

## Discussion

Coral-derived δ18Osw from AG and EG exhibited significant correlation with ocean SSS (offshore grid), local precipitation and river discharge suggesting that the coral record can provide reliable information on changes in the hydrological balance in reef waters. Regional monitoring of oxygen isotopes in rainfall between 2004 and 2011 upstream and downstream in the Baram catchment (Mulu and Lambir Hills National Park) indicated a clear inverse relationship with the amount of rainfall62,63. Therefore, δ18O of seawater around Miri is influenced by the balance between precipitation and evaporation (the P-E balance), and freshwater runoff. Thus, sea surface salinity, which is also largely modified by P-E balance and riverine freshwater runoff, is linearly related to δ18Osw39,64,65. In the southern part of the SCS precipitation exceeds evaporation57. EN4 SSS66 for the grid closest to our coral reefs showed significant correlations with precipitation at Miri (r = −0.53, p < 0.009) yet not with river discharge at Marudi (r = −0.29, p = 0.166). The modest correlations of EN4 SSS66 with precipitation and river discharge could be explained by salinity being averaged over a larger offshore oceanic grid (3.5–4.5°N, 112.5–113.5°E) that is some distance from the rainfall station and rivers. However, correlations of coral δ18Osw from AG and EG with the local precipitation (AG: r = −0.68, EG: r = −0.83, p < 0.003) indicated a significant inverse relationship between the reefs’ seawater δ18O and the seasonal cycle of local rainfall, with lowest δ18Osw during the winter monsoon. The same holds for correlations with river discharge at both EG and AG. Mean δ18Osw in EG and AG did not differ, despite EG’s closer proximity to the river mouths. Exceptionally lower peaks in EG than AG, for example in the winter months of 2008/9 and 2010/11, could be attributed to heavy rainfall and high runoff events, which are typically more pronounced closer to the river source. The AG δ18Osw record for 1982 to 2016 was also in good agreement with trends and interannual to decadal variability in oceanic salinity data with few exceptions. The same holds for correlations between AG δ18Osw and river discharge between 1989 and 2016. As such, both AG and EG reliably recorded the regional P-E balance and river discharge. Yet, isotope analysis and salinity measurements of water samples on a transect from the river mouth to the reefs would provide a better understanding of riverine impact in future studies.

Both the gridded SSS data and the coral-derived δ18Osw displayed an interdecadal shift in the mid-1990s superimposed on the interannual ENSO variability (see Supplementary Fig. S3). Several studies in the SCS attributed these interdecadal shifts to the PDO. Zeng et al.67 observed a salinification trend in the SCS from approximately 1973 to 1993, a freshening from 1993 to 2012 and again salinification from 2012 to the present. These trends were also observed in the coral-derived δ18Osw (salinification 1982 to 1994, freshening 1995 to 2013, salinification 2014 to 2016) and gridded SSS data in MSCRNP (Fig. 5; see Supplementary Fig. S3). Both AG and EG δ18Osw, as well as precipitation and river discharge, indicated robust relationships with the PDO, while EN4 SSS showed weaker, yet statistically significant relationships (Table 4). It has been suggested that the positive PDO phase induces dry sinking air along with reduced rainfall and vice versa in the negative phase in the SCS23. In addition, salinification during positive PDO is further enhanced due to a directional switch in the horizontal current through the Luzon Strait, leading to the advection of more saline waters into the SCS67. In line with these findings, Deng et al.68 showed significant correlation between δ13C and δ18O from a coral proxy record in the northern SCS and the PDO. They suggest that PDO is causing precipitation anomalies on interdecadal scales, influencing the hydrological balance in the region due to changing rainfall and river runoff dynamics. It is therefore likely that Pacific decadal variability has influenced long-term salinity trends in the SCS. Therefore, our coral δ18Osw records should be extended in future studies to cover the past two centuries to assess the stability of the ENSO teleconnection and decadal shifts related to the PDO identified in our study.

The correlations of precipitation and the coral δ18Osw in Miri with ENSO indices demonstrated a robust relationship between the ENSO phenomenon and hydroclimate in northern Borneo. The spatial correlations resembled the ENSO “horseshoe” SST pattern in agreement with previous studies19,58 (Fig. 6). The El Niño events in 1983, 1988, 1992, 2010 and 2014–2016 all showed consistent positive δ18Osw and salinity anomalies indicating reduced runoff or P-E testifying to the paramount impact of ENSO on SEA drought. The Marudi river discharge and Miri precipitation records confirmed the reduction in freshwater runoff from the Baram River since 1989 and precipitation since 1992 during all El Niño years. The comparison of coral-derived δ18Osw anomalies with the SST anomalies in the Niño 3.4 and Niño 4 regions also displayed good agreement, with a particularly striking shift in δ18Osw anomalies matching the 1997/98 El Niño followed by 1998/99 La Niña. The latter is not surprising since winter precipitation in Borneo was reduced by more than 50% during the 1997/98 El Niño event69. In addition, 2005 appeared to show an El Niño-like response in our coral δ18Osw records, SSS, precipitation and river discharge. 2005 was classified as a weak El Niño year in the Niño 3.4 and Niño 4, yet as El Niño Modoki59 or central Pacific El Niño, respectively. The response in 2005 could be an indicator of different local signals of Borneo’s hydroclimate to varying spatial expressions of ENSO SST anomalies as suggested by Tan et al.59. In contrast, during La Niñas of 1996, 1999–2001, 2008 and 2011 negative δ18Osw were observed in agreement with lower SSS and higher river discharge. Several other paleoclimate studies in the SCS on corals from the northern SCS48, a speleothem record from northwestern Borneo70 and the δ18O of rainfall interannual anomalies at Gunung Mulu and Lambir Hills National Parks in northern Borneo have also shown clear ENSO impacts62,63. The proxy records together with the rainfall isotope data support the observed pattern of anomalously low precipitation and droughts in the Maritime Continent during El Niño71,72. This drying across the Maritime Continent is associated with warm SST anomalies during the El Niño mature phase causing an eastward shift in convection72. During La Niña convection strengthens in the Maritime Continent, causing anomalously high precipitation and floods. In northern Borneo the impact of ENSO starts in Boreal autumn and persists through Boreal spring with strongest responses in the winter season19,58,73,74, which is in accordance with the results from the coral records. The peak drying response to El Niño in northern Borneo in winter and spring is associated with the northeastward shift in intraseasonal convective activity into the northern hemisphere and the development of an anticyclone in the West North Pacific19,58. Tangang and Juneng19 suggested that the correlation between northern Borneo precipitation and ENSO in December to February is not a direct response to the eastern Pacific pole of canonical ENSO, but rather related to a dipole in SST strengthening the western Pacific arm of the ‘horseshoe’ SST pattern generated by ENSO. This interpretation is in agreement with our results showing highest correlations of coral δ18Osw (also precipitation, Marudi discharge and Miri precipitation) in winter/early spring (February to April) with northwestern Pacific SST east of the Philippines (Fig. 6). Thus, coral δ18Osw proves to be a reliable tracer of ENSO impacts on precipitation and river discharge in the region. In addition, our results provided clear evidence for ENSO-driven synchrony of climate impacts on both terrestrial and marine environments in northern Borneo. Furthermore, our findings provided invaluable data to better understand ENSO and PDO impacts on the southern SCS, but also for reef monitoring and protection, since both thermal stress and freshwater flux and related sediment transport may put coral reefs increasingly at risk along the northern Borneo coast in the coming decades. Future studies will expand the proxy reconstructions for this region both temporally and spatially to test for long-term stability of ENSO and PDO relationships with the Asian-Australian monsoon and its potential modification by recent global warming.

## Methods

### Coral sampling and core treatment

In September 2016, coral cores were retrieved from two reef sites in MSCRNP, at Sarawak’s northern coast (Fig. 1). Fieldwork was approved by the Sarawak Forestry Commission (permit no (61)/JHS/NCCD/600-7/2/107) and methods were executed in accordance with the approved guidelines and regulations. Coral cores were imported under CITES licence number 002259. The two reefs, Eve’s Garden (EG) and Anemone’s Garden (AG), lie along an inshore to offshore gradient and at different depth, distance from shore and distance from river mouths (Fig. 1; Table 1)45. Coral cores were drilled using a SCUBA tank driven pneumatic drill (Silverline Air Drill Reversible), with a diamond-coated drilling head. Core segments of approximately 30 cm in length and 4 cm in diameter were obtained along the central growth axis of massive, dome-shaped Porites sp. colonies. Cores were then sectioned into 7–8 mm thick slabs longitudinally, along the axis of growth and cleaned following the chemical treatment method developed by Nagtegaal et al.75. Slabs were submerged for 24 hours in a bath of sodium hypochlorite solution NaOCl (with 6–14% active chlorine) diluted to a 1:1 ratio with distilled water. Thereafter, slabs were rinsed three times in an ultrasonic bath filled with distilled water for 10 min, with water exchanged after every run. In between turns coral segments were blown with compressed air to remove stray particles. Finally, slabs were dried in a drying cabinet at 50 °C for 24 to 48 h. X-ray radiography was used to visualize annual density bands and to determine the sampling path along the major growth axis (see Supplementary Fig. S4). Samples for geochemical analysis were drilled every 1.1 mm, using a diamond coated dental drill. The number of samples was n = 271 for Anemone’s Garden and n = 103 for Eve’s Garden. The sample powder was then split for stable isotope and Sr/Ca measurements.

### Sr/Ca measurements and SST reconstruction

Sr/Ca ratios were measured at the University of Kiel with a simultaneous inductively coupled plasma optical emission spectrometer (ICP-OES, Spectro Ciros CCD SOP), following a combination of the techniques described by Schrag76 and de Velliers77. Sr and Ca were measured at their 421 and 317 nm emission lines, respectively. 175 ± 25 µg of coral powder was dissolved in 1 ml nitric acid (HNO3 2%). Prior to analysis, this solution was further diluted with 4 ml HNO3 2% to a final concentration of approximately 8 ppm. An analogously prepared in house standard (Mayotte coral) was measured after each sample batch of 6 samples to correct for drift effects. The international reference material JCp-1 (coral powder) was analysed at the beginning and end of every measurement run. Internal analytical precision based on replicate Sr/Ca measurements was 0.008 mmol/mol (1σ) or 0.08%. Average Sr/Ca value of the JCp-1 standard from multiple measurements on the same day and on consecutive days was 8.831 mmol/mol with 0.085% relative standard deviation (RSD). The comparison to the certified Sr/Ca value of 8.838 mmol/mol78 with an expanded uncertainty of 0.089 mmol/mol indicates a high external precision of <0.08%.

SST reconstructions in this study are based on calibrations of the coral Sr/Ca ratios with the OISSTv2 data. The relationship between the skeletal Sr/Ca and the instrumental data was quantified by (OLS) regression79.

$$Sr/C{a}_{coral}=m\times SS{T}_{recon.}+b$$
(1)

The error estimate for absolute SST reconstructions was calculated with the mean squares of the residuals of the inverted Eq. 141.

The calculation of Sr/Ca-SST anomalies is based on relative changes in SST in comparison to a climatological mean. This method was used to eliminate errors associated with absolute SST reconstructions, since SST anomalies are independent from the intercept of the calibration. Sr/Ca anomalies for the AG record were calculated relative to the 1983 to 2015 mean, and converted to SST based on a range of published slopes to indicate uncertainties of the estimate26,29. Anomalies for the shorter records starting in 2006 of AG and EG were calculated relative to the average of 2006 to 2015. Calculation of anomalies was carried out with the KNMI Climate Explorer80. Based on empirical studies by Corrège26 and Gagan et al.29, the slopes of −0.06 mmol/mol °C−1 and −0.084 mmol/mol °C−1 were used for the conversion of Sr/Ca anomalies to SST next to local calibration slopes.

### Oxygen isotope measurements

Stable oxygen and carbon isotope measurments were performed at the Museum für Naturkunde, Leibniz-Institut für Evolutions- und Biodiversitätsforschung, Berlin (Germany). 200 ± 50 µg of coral powder from each sample was put into clean 10 ml exetainers. After being sealed with a septum cap, the remaining air was flushed out of the vessel with helium gas for 6 min at a flow of 100 ml/min. This process was followed by an injection of 30 µl of anhydrous phosphoric acid (H3PO4) through the septum. Each sample had a reaction time of approximately 90 min at 50 °C before measurement. Oxygen and carbon isotopic composition of the reaction product (CO2) was measured using a Thermo Finnigan GASBENCH II coupled with a Thermo Finnigan DELTA V isotope-ratio mass spectrometer (IRMS). Pure CO2 was used as reference gas for the measurements. The reference gas was calibrated against the VPDB (Vienna Pee Dee Belemnite) standard by using the IAEA (International Atomic Energy Agency) standards NBS18 and NBS19. Isotope values are shown in the conventional delta notation (δ18O) in per mil (‰) versus VPDB. The internal analytical precision based on replicate measurements of the reference gas was <0.05‰. The reproducibility of replicated lab standards (pulverized limestone) was generally <0.1‰ (1σ) for both carbon and oxygen isotopes.

### Age model

Age-depth models for each core were established using the software AnalySeries2.081 and the NOAA AVHRR-OISSTv2 High Resolution Dataset for a 0.25° grid46,47. The chronology was generated based on the pronounced seasonal cycle of Sr/Ca ratios. The highest Sr/Ca values of each cycle were assigned to the coldest month from the instrumental data for each year. In cases of clear low peaks in the Sr/Ca ratios, these were assigned to the warmest months. The starting year was assigned to 2016, since the cores were drilled from living colonies. First, age assessments for all samples between the anchor points were obtained by linear interpolation in AnalySeries2.081. In a second step, the sample data were interpolated to 12 equidistant points per year, creating a monthly time series. The time scale error in any given year is 1 to 2 months, due to interannual differences in the exact timing of peak SST. As such, AG covers the period 1982 to 2016 while EG covers 2006 to 2016. Application of lagged correlations to monthly/seasonal climate indices (Fig. 6) revealed the highest correlations at zero lag, suggesting that our linearly interpolated age model did not obfuscate seasonal trends in geochemical proxy data, despite the 1–2-month uncertainty in identifying annual anchor points.

### δ18Osw reconstruction

For the calculation of δ18Osw the method of Ren et al.44 was followed, with the assumption that coral Sr/Ca is solely a function of SST and that coral δ18O is a function of both SST and oxygen isotopic composition of the seawater. The method uses instantaneous changes of both variables instead of looking at the absolute values. Effects of seawater isotopy on coral δ18O are separated from thermal effects by breaking the instantaneous changes of coral δ18O into separate contributions by instantaneous SST and δ18Osw changes, respectively. This is possible due to paired measurements of Sr/Ca and δ18O on the coral. In these calculations, we used a slope of −0.2‰ per 1 °C82 for δ18Ocoral - SST, and for Sr/Ca-SST a slope of −0.06 mmol/mol °C−126. Error estimation was performed by neglecting the error caused by non-climatic factors that may influence the proxies. The error σδ18Osw in this study is 0.103‰. It was calculated following the approach of Cahyarini et al.38 (Eq. 2), with σδc being the error of measured δ18Ocoral, σSr/Ca being the error of measured Sr/Cacoral, and γ1 and β1 being the slopes of the linear regression of δ18O versus SST and Sr/Ca versus SST, respectively.

$${\sigma }_{\delta sw}^{2}={\sigma }_{\delta c}^{2}+{(\frac{{\gamma }_{1}}{{\beta }_{1}})}^{2}\times {\sigma }_{Sr/Ca}^{2}$$
(2)

Anomalies of δ18Osw from all coral records and gridded SSS65 for the study area were calculated relative to the 2006 to 2015 mean using the KNMI Climate Explorer80.

### Historical climate data and climate indices

Local mean air temperature data from the Miri airport station (WMO station No. 964490, 4.4°N, 114.0°E, elevation: 51 m a.s.l.) were obtained from the Global Historical Climatology Network-Monthly (GHCN-M version 3) temperature quality-controlled dataset83. Data were provided by the National Centers for Environmental Information (NCEI) of the U.S. National Oceanic and Atmospheric Administration (NOAA), downloaded from KNMI Climate Explorer80. Local monthly precipitation data from the Miri airport station (WMO station No. 964490, 4.4°N, 114.0 °E, elevation: 51 m a.s.l) were obtained from the Global Historical Climatology Network Monthly (GHCN-M version 2) quality-controlled dataset, which contains temperature, precipitation, and pressure data84. Data were provided by NOAA’s NCEI, downloaded from KNMI Climate Explorer80.

River discharge data for the Baram River catchment were provided by the Department of Irrigation and Drainage (DID) in Malaysia15. The nearest station to the Baram outflow north of Miri was Marudi station (4.10°N, 114.18°E), which provided continuous data coverage between 1989 and 2015.

The NOAA 0.25° daily Optimum Interpolation Sea Surface Temperature version 2 (OISSTv2) dataset was used46,47. The dataset combines observations from ships, buoys and satellites (infrared satellite data from the Advanced Very High Resolution Radiometer AVHRR), with data gaps filled by interpolation. Daily SST data was available at a 0.25° grid resolution. The grid used in this study was 4.24–4.5°N, 113.75–114.00°E. Data were provided by NOAA’s NCEI, downloaded from KNMI Climate Explorer80.

EN4.2.0 SSS was obtained from the EN4 quality controlled subsurface ocean temperature and salinity compiled dataset, which is based on data of all instruments capable of profiling the water column66. Observations were interpolated and available in a grid resolution of 1°. The grid used in this study was 3.5–4.5°N, 113.5–114.5°E, but only surface salinity data were used. Data were provided by Met Office Hadley Centre, downloaded from KNMI Climate Explorer80.

ENSO indices Niño3.4 and Niño4 used in this study were based on OISSTv246. Niño3.4 includes SST anomalies in the 5°N–5°S and 170°W–120°W region, Niño4 in the 5°N–5°S and 160° E–150°W region51. For both indices, anomalies were calculated relative to a 1981–2010 base period. The Niño 3.4 region represents the core region of ENSO. ENSO events were defined when 5-month running means of SST anomalies in the Niño 3.4 region exceeded a threshold of ±0.4 °C for a period of six months or more85.