Prehistoric shifts in tropical cyclone season in the South China sea: evidence from daily resolution records of giant clam shells

Abstract

The impact of climate change on tropical cyclones (TCs) remains a critical concern, but understanding is limited by short instrumental records and low-resolution prehistoric data. Here, we present the first daily-resolution proxy data for investigating prehistoric TC activity, using a 12-year growth rate record of a fossil Tridacna shell (circa 3 ka BP) from the northern South China Sea (NSCS). By analyzing temporal patterns in the proxy data, we derived monthly TC frequency estimates. While modern TC frequency in the NSCS peaks in autumn (September–October), our results showed that TCs at 3 ka BP occurred more frequently in summer (June–July–August), with approximately 15% higher frequency than present. Combined with paleoclimate records, we suggested that this seasonal shift and increased frequency were likely linked to the relatively northward migration of the Intertropical Convergence Zone, which provided favourable conditions for TC formation and development. Our findings imply that future warming would contribute to earlier TC seasons and increased TC frequency in the NSCS.

Introduction

Tropical cyclones (TCs) are one of the most destructive weather events, usually accompanied by strong winds, heavy rains, huge waves and other meteorological disasters. They can also cause secondary disasters such as landslides and mudslides, posing a huge threat to the ecosystem destruction, human survival and social economy^1,2. Currently, there is significant interest of public and scientific circles in whether TC activity has already changed and how it may change in the context of global warming^3,4,5,6. These have become focal issues in climate change research in recent decades.

Our understanding of climate controls on TCs is limited by the short length of observational records, typically less than 100 years, which cover only part of the full range of their variability and therefore result in uncertainty and sometimes controversy in the understanding of the trend and variation of TCs. For example, a study showed that since the 1980s, the global trends in TC frequency exhibit a distinct spatial pattern, with decreases in the southern Indian Ocean and western North Pacific and increases in the North Atlantic and central Pacific⁷. But another study suggested a declining trend in TC frequency at global and regional scales³. Understanding the mechanisms by which TCs have varied in response to past forcing will help us understand how TC risk might change with enhanced global warming. However, current reconstructions of TCs rely on paleoclimate proxies with low resolution, primarily suited for discussing TC activities and their link to climate change over centennial or millennial scales. TC events that actually occurred at daily to weekly timescales are opaque at coarse time resolution studies^8,9,10. Furthermore, model simulations of paleo-TCs also face numerous challenges. For example, the simulation results of different models vary greatly, most publicly available models have low spatial resolution, omitting many nearshore areas such as the South China Sea (SCS)^11,12. If natural proxies can be developed to provide TC data with daily resolution similar to modern instrumental data, it would extend the length of observational records and be of great significance for understanding the full range of mechanisms of TC occurrence and development, validating climate models, and predicting future trends in TC development under global warming.

The marine bivalve, Tridacna spp., is the largest bivalve species, which is widely distributed in the coral reefs of the Indo-Pacific Ocean from the Eocene to the present. The symbiotic algae (zooxanthellae) in their mantles provide nutrients for the growth of the shell through photosynthesis. It usually stays in one location of a coral reef during its lifetime and thus accumulates a record for a point in space^13,14,15,16. Tridacna spp. have hard and dense aragonite shells with visible annual growth lines and even daily growth lines in their inner shell layer, implying that they have the potential to record the daily changes in the surrounding environment, thereby reconstructing past extreme weather events, such as TCs. Komagoe et al.¹⁷ suggested that TCs interfered with the daily growth rate (DGR) of the Tridacna shell and the positive peaks of shell Ba/Ca ratio and δ¹⁸O_shell reflected the upwelling and lower sea surface temperature (SST) caused by TCs. Yan et al.¹⁸ found that the hourly-daily resolution biogeochemical proxy records matched well with TC events in the northern SCS. When a TC approached the sampling site, the DGR of Tridacna shell decreased abruptly due to weather changes. High Fe/Ca ratios and strong fluorescence intensity raised in the inner layer of the shell due to seawater mixing. These studies improve the resolution of TC reconstructions and indicate that Tridacna shells have significant potential for prehistorical TC reconstructions, particularly for individual TC events.

An ideal site for the investigation of TC events is the northern SCS, which is a critical region for TC genesis in China and one of the primary pathways for Northwest Pacific TCs to reach landfall (Fig. 1). Our recent study showed that extreme variation of DGR of modern Tridacna shells is a reliable indicator for TC season reconstruction¹⁹. However, current research in this area is still in its early stages and little is known about long-term changes of the TC season in the northern SCS. Fossil giant clam samples are widely deposited in the coral reefs of the SCS, which have the potential to provide high-resolution records (at daily resolution or even higher) of past TC changes. This study introduced a Tridacna proxy with the daily resolution that captured TC events in the northern SCS. We reconstructed the monthly estimates of TC occurrence at circa 3 ka BP, characterized its seasonal pattern, compared it with the modern condition and explored the possible physical mechanisms regulating seasonal changes of TCs in the study region.

Fig. 1: Site location and local TC frequency.

a Location of North Reef (Red dot, 17°05′, 111°30′E) and Langhua Reef (black dot, 16°3′N, 112°33′E). Blue lines are TC pathways during 1991–2020 AD. b Monthly mean TC frequency based on TC best track data from 1991 to 2020 AD.

Results

AMS ¹⁴C date and Sr/Ca profile of fossil T. squamosa

AMS ¹⁴C dating can determine the basic age range of fossil giant clam samples and provide a climate background for TC activity research. The AMS ¹⁴C dating of T. squamosa A114 was 3157 ± 22 yr BP and the radiocarbon age was corrected to 2980 ± 40 yr BP using Marine20 of Calib 8.20^20,21.

The Sr/Ca ratios of specimen A114 showed clear annual cycles and about 26 annual cycles were detected in Sr/Ca profiles of A114 (Fig. 2a). The average Sr/Ca ratios of A114 was 1.63 ± 0.08 mmol/mol (n = 668), ranging from 1.41–1.87 mmol/mol. Previous study showed that Sr/Ca ratios of T. squamosa are positively correlated with SST in the northern SCS, where the monthly SST minima occur near January, and have stable high values from May to September^22,23,24. Therefore, the time for the data point of minimum Sr/Ca was assigned in January.

Fig. 2: Cross-check of low-resolution Sr/Ca data and daily growth rate of the Tridacna sample A114.

a The raw data of Sr/Ca. The blue line is the part with clear daily growth layers from the juvenile stage, about 12 years. b The raw data of daily growth rate (DGR). The thick line is fitted using the smoothing spline model. The number of daily growth layers in each annual cycle is marked below the profile. c The annual layer width obtained from the distance of Sr/Ca minimum value in each annual cycle (blue, solid) and the annual layer width (green, hollow) obtained by summing up the DGR in each year.

Daily growth rate and daily resolution shell age

LSCM provides the micro-images of daily growth bands of Tridacna shell. The counting results of sample LHJ2 showed there were 2047 daily growth increments over about 5.6 years (1 year = 365 days) with the DGR ranged from 4.0 to 23.2 μm, and the average thickness was 11.4 μm¹⁹. As age increased, the daily growth rate tended to decline, resulting in a significantly narrow growth bands, which are often difficult to observe under LSCM. Therefore, only the clearest part of the daily growth layers from the juvenile stage of fossil giant clam A114 was selected for analysis. The DGR counting results from 3-times replication, which was 4231, 4248 and 4274, respectively, about 12 years (1 year = 365 days). The DGR of specimen A114 also showed clear annual cycles as that of Sr/Ca ratios, ranging from 2.0 to 27.2 μm. The data points of DGR per year were obtained from the annual cycle of DGR, ranging from 320 to 366 days, which means the maximum difference between the number of DGR per year and the calendar year is 45 days (Fig. 2b).

The daily resolution chronology of modern T. squamosa LHJ2 has been established, from 12 May 2013 to 18 December 2018¹⁹. The average difference between the number of calendar days and daily growth increments was 9.5 days per year, and the age-related error negligibly affected the recognition of the seasonal frequencies of tropical cyclones¹⁹. The daily resolution chronology of fossil T. squamosa A114 was established based on the number of DGR, annual cycles of Sr/Ca and AMS ¹⁴C dating results. The basic age of the A114 sample was determined by AMS ¹⁴C dating and calibrated to calendar ages using Marine20, yielding 2980 ± 40 cal yr BP. The slice daily resolution chronology was cross-checked by DGR and Sr/Ca records. On the one hand, the annual growth width, the distance of Sr/Ca within each annual cycle, was calculated based on the sampling density (samples per year) and sampling interval (0.1 mm). On the other hand, the annual growth width can also be derived by summing DGR. The results obtained by the two methods are similar (Fig. 2c), which indicates that the results of Sr/Ca and DGR are comparable and that the identification of daily and annual growth increments is correct.

In order to assign the daily resolution shell age to the calendar date for time series analysis, linear interpolation was performed on the DGR profile. Modern Tridacna shell showed that the month when the minimum DGR occurred corresponded to the coldest month of SST in the SCS²⁵. In the Xisha islands, the coldest month for average SST is January, and sometimes February. We assigned the annual DGR minimum as January 31, and the time for the data between two adjacent DGR minimums was assigned using linear interpolation with an equal time span to ensure that there are 365 DGR data points per year. The uncertainty in the coldest SST month (January or February) introduces approximately one month of dating error in the daily resolution Tridacna chronologies.

Despite inherent uncertainties in annual daily growth layer counts, the interannual error accumulation can be constrained through linear interpolation between successive minima in the DGR sequence. The specimen A114 was a fossil sample and the exact date of death was unknown. Therefore, we assumed an arbitrary time series as years 1, 2, 3…, 12 for the daily resolution shell age (Fig. 2).

Discussion

Giant clams can contain a record of a high-quality and high-resolution reconstruction of marine climate and environment due to their unique daily growth layers and sensitivity to changes in the surrounding environment. Modern giant clams have promising potential for accurately recording individual TC events and TC seasonal characteristics in the Xisha Islands of the SCS^17,19. TCs can swiftly alter SST, surface wind patterns, cloud cover, and wave characteristics. These environmental changes can significantly impact the growth of a Tridacna shell, prompting rapid and notable responses. Previous studies have demonstrated that TCs generally lead to a decrease in SST within their vicinity, accompanied by increased rainfall and wind conditions, which are unfavorable for the growth of Tridacna shells^17,18. While the cyclonic circulation associated with TCs can facilitate transport of nutrient-rich subsurface water to the ocean surface, resulting in elevated concentrations of nutrients and phytoplankton blooms in the surface seawater^17,18,26. These conditions can stimulate the growth of shell carbonate. Therefore, theoretically, the passage of a TC can cause abnormal fluctuations in the growth of Tridacna shells, including increases and decreases. This study referred to our previous study¹⁹ and optimized the definition of TC events, in which each TC event should contain abnormally positive GIV and negative GIV for two consecutive days or more, with an interval greater than 3 or more days.

Summer and autumn are seasons with frequent TC activities in the Northern Hemisphere, and the June–November period is considered the major TC season in the SCS^27,28 (Fig. 1b). Considering the seasonal variation of TCs and to avoid the impact of other extreme weather events on the growth of giant clams, such as cold waves in winter (December–February), this study only focuses on the TC season, from June to November. We compared the TC frequency records respectively obtained from the modern T. squamosa shell LHJ2 and TC best track data near the sampling site from 2013 to 2018 AD. The multi-year monthly average TC frequency derived from modern T. squamosa shell during the TC season was 0.78 per month, consistent with the statistical results from contemporaneous TC best track data (2013–2018 AD). The multi-year monthly average TC frequency obtained from TC best track data during 1991–2020 AD was also 0.78. And the occurrence of peak TC frequency at modern times was in September, followed by October (Fig. 3a, Fig. 1b). The results are clear as intense TCs generally occur more frequently in autumn than in summer due to their reliance on sufficient heat from the surface ocean for formation²⁹. Modern calibration is inherently constrained by the relatively short lifespan of contemporary Tridacna specimens, potentially limiting their capacity to fully capture the complete TC climatology. Nevertheless, a statistically robust positive correlation between monthly mean TC frequencies derived respectively from TC best track data and proxies in modern shell LHJ2, suggesting a causal relationship, that is, the abnormal fluctuations of shell were caused by TC events. The monthly number of abnormal fluctuations of the shell can be mainly attributed to variations in TC frequency. These results are further supported by the Superposed Epoch Analysis (SEA) of the TC response of modern shell LHJ2 (Fig. 3b). In the immediate days following a TC event, the median bootstrap shell response is significant (p < 0.01). But the peak appearing 5 days after the event, which may be affected by the age error (average 9.5 days per year) of T. squamosa LHJ2.

Fig. 3: Seasonal distribution of the TC frequency in the northern SCS at the present day.

a Reconstructed TC frequencies respectively from the modern Tridacna shell LHJ2 (Orange bars) from 2013 to 2018 AD¹⁹ and contemporaneous TC best track data (Bule bars). b SEA shows the response of TC_LHJ2 event within the study area. Dashed lines indicate threshold for statistically significance (p < 0.01, one-tailed).

Previous studies have reported monthly changes in TCs affecting Chinese coastal areas over the Holocene. A historical record for the past 2000 years showed a relatively early occurrence of landfall typhoons in China during the Medieval Warm Period than compared to the Little Ice Age³⁰. Coral records from the northern SCS suggested that the timing of the rainy season in this region changed during the middle to late Holocene³¹, where the rainy season is synchronous with the TC season in modern times. These results indicate that the timing of the TC season over the northern SCS was variable during the Holocene. However, historical studies of TC season are few and are limited by the resolution of paleoclimate proxies.

The significantly positive correlation (R = 0.74, p < 0.001) between TC frequencies reconstructed by the modern sample LHJ2 and calculated from the instrumental data underpinned the reliability of the number of Tridacna shell abnormal fluctuations as a proxy for the seasonal distribution of TC frequency (Fig. 3a). Using this approach, the occurrence of peak TC frequency at circa 3 ka BP can be inferred from the monthly abnormal fluctuations of the T. squamosa A114. However, the calibrated period was limited by the survival age of modern living T. squamosa LHJ2, which was only 6 years. To better characterize the average climatological state of TC season at circa 3 ka BP, the changes of monthly TC frequency were calculated by the 6-year moving average of the reconstructed TC frequency from A114 shell. The results revealed an average of 0.90 TCs per month during June-November, with peak frequency concentrated in June-August (Fig. 4a). Then, we employed a block bootstrap approach (1000 iterations) to extend the record to a 30-year climatological period, generating robust confidence intervals for the reconstructed TC frequency. The bootstrap-derived seasonal pattern consistently showed June-August as the predominant TC months, confirming the reliability of our reconstruction (Fig. 4b). In contrast to the modern-day September–October TC season (Fig. 3, Fig. 1b), it appeared that the predominant timing of TCs circa 3000 years before was somewhat earlier. In a word, the Tridacna-based reconstruction indicates a notable increase in TC frequency at around 3 ka BP, with an estimated 15% enhancement relative to modern climatology. This raises the question of what caused the predominant timing of TCs in the northern SCS to shift from June–August circa 3000 years ago to September–October in recent decades. From a climatic mean-state perspective, one possible explanation is that the Intertropical Convergence Zone (ITCZ) was located at a higher latitude during 3 ka BP, which may have contributed to earlier TC formation^30,32.

Fig. 4: Seasonal distribution of the TC frequency in the northern SCS at circa 3 ka BP.

a TC frequency reconstructed from the T. squamosa A114 at circa 3 ka BP. Dashed lines represent the 6-year moving average with a 1-year interval. The thick line indicates the overall 12-year average. b Reconstructed TC frequency based on that in (a), extended to 30 years by applying block bootstrap method to the origin data (number of bootstrap iterations: 1000). Blue shading represents 1st–99th percentile and blue line indicates the median value.

Based on TC observations and model simulations, most ocean basins have seen a poleward migration of TC activity regions in recent decades, and this migration is linked to variations in TC seasonality^33,34. A recent study identified a significant seasonal advance of intense TCs since the 1980s in most tropical oceans²⁹. Using climate model simulations, that study suggested that the earlier onset of favorable oceanic conditions is driven by greenhouse gas forcing, whereas the contributions of natural and anthropogenic aerosol forcings are negligible or even negative²⁹. However, Cao et al.³⁵ emphasized the role of anthropogenic aerosols in TC changes. Open questions remain on how anthropogenic and natural factors may affect TC development with continuous global warming.

Records of monthly TC frequency reconstructed from fossil Tridacna shells provide a new perspective to probe the characteristics of TC seasonality. This proxy record represents mean climatic conditions over the shell’s lifespan. To investigate the reason for the earlier and increasing shift in TC occurrence at circa 3 ka BP, we examined the oceanic and atmospheric conditions favorable for the seasonal changes of TCs. For the oceanic conditions, ocean heat content should be higher and mainly moderated by SST. For the atmospheric part, increased water vapor content in the atmosphere provides essential moisture for the formation of TCs, aiding their formation and development.

Paleoclimate studies suggested that the seasonal variation of TCs may be related to the position of the ITCZ^30,32. ITCZ is a tropical belt of deep convective clouds near the equator where the winds from both hemispheres converge and rise. The annual mean position of ITCZ is north of the equator and seasonally, it migrates towards a hemisphere that warms relative to the other. In the SCS, ITCZ generally migrates between 20°N in boreal summer and 5°N in boreal winter^36,37. On longer timescales, paleoclimate records indicate that the ITCZ migrates similarly towards a differentially warming hemisphere. Therefore, ITCZ position has been closely connected to interhemispheric thermal gradients^37,38. Reconstructed global temperature anomalies showed that after the mid-Holocene (around 6 ka), the Northern Hemisphere reconstructed temperature gradually decreased, and the temperature gradient between the Northern and Southern Hemisphere was reduced³⁹ (Fig. 5b). And reconstructed and modeled SST contrast between the northern and southern tropics was stronger than it is today (Fig. 5d)^40,41. Therefore, at circa 3 ka BP, the temperature gradient derived from reconstructions was larger than it was at the preindustrial era. These conditions may have provided more favorable heat conditions for TCs in the and could have contributed to a northward shift of the thermal equator, thereby reflecting the northward of ITCZ at circa 3 ka BP³⁷. Clear evidence for ITCZ migrations was also supported by the dark terrestrial detritus rich in elements such as titanium from the Cariaco basin and reconstructed East Asian Summer monsoon (EASM) records (Fig. 5b, c). Higher titanium concentrations indicate more terrestrial runoff and rainfall, interpreted as a farther northward ITCZ excursion in boreal summer. Declining titanium concentrations indicate a southward of ITCZ position since the mid-Holocene (Fig. 5b)⁴². Paleoclimate records (including Chinese stalagmite records, the loess magnetic susceptibility records from the Chinese Loess Plateau, and lake sediment records) exhibit consistent trends, showing a weakening trend in the EASM or monsoon precipitation since 3 ka BP^43,44,45, which probably reflects a southward migration of the ITCZ. Furthermore, Fig. 4a showed that the tropical insolation during the boreal June to August was much more intense around 3000 years ago compared to the present day⁴⁶. This increased insolation may have accelerated the attainment of oceanic heat conditions favorable for TC formation in the Northern Hemisphere, potentially contributing to the earlier onset of the TC season.

Fig. 5: Late Holocene climate forcing and paleoclimate records.

a Monthly mean insolation for 0°–30°N⁴⁶. b Northern Hemisphere (30°–90°N) temperature anomaly (blue line) and the temperature contrast (red line) between the extratropics of the Northern and Southern Hemisphere. The temperature and temperature contrast are derived from there constructions in Marcott et al.³⁹, for extratropical areas of 30°–90°. Titanium content (orange) from the Cariaco Basin off the Venezuelancoast⁴². c East Asian monsoon records indicating the ITCZ position since the Late Holocene, including the speleothem δ¹⁸O record (green line) from Sanbao cave⁴³, reconstructed precipitation (blue line) from Lake Gonghai in northern China⁴⁴ and stacked normalized magnetic susceptibility (black line) from Chinese Loess Plateau⁴⁵. d SST contrast (red line) between the tropics (0–30°) of the Northern and Southern Hemisphere in the Pacific (100°–180°E)⁴⁰. Difference in modelled SST (gray line) of Western Pacific between Northern Hemisphere (100–180°E, 0–85°N) and Southern Hemisphere (100–180°E, -70°S–0). Insolation contrast between 30°N and 30°S (purple line) during the TC season of the Northern Hemisphere (June to November) and Southern Hemisphere (December to May)⁴⁶.

Accordingly, at around 3 ka BP, paleoclimate records suggested that the SST contrast between the northern and southern tropics was larger and the ITCZ was located at a higher latitude compared to recent decades. This may have resulted in the convergence zone reaching the study area (17°N) earlier and extending further northward. Within the convergence zone, the convergence of airflow generated strong upward motion, which promoted the release of unstable energy and favored the occurrence and development of TCs. This could contribute to the seasonal advance of TCs from autumn to summer and an increase in TC frequency (Fig. 3). At circa 3 ka BP, due to the northward shift of the ITCZ, the timing of its northernmost reach should be delayed compared to modern conditions. As the ITCZ continued its seasonal migration, during the autumn months, even though the SST remained warm, the Xisha Islands might be positioned south of ITCZ, potentially resulting in decreased deep convection and insufficient circulation conditions to support the formation and development of TCs. In the winter and spring seasons, the ITCZ should begin to move southward, and the cooling SST likely reduced the potential for TC genesis. As a result of the above process, at circa 3 ka BP, the TC season in the northern SCS was concentrated in June-July-August (Fig. 4).

Using observations and model simulations, previous studies have attempted to understand seasonal changes of TCs in a warming climate. The study observed a significant seasonal advance of intense TCs since the 1980s in most tropical oceans. A link was found between increasing greenhouse gas and seasonal advance of intense TCs under a high-emission scenario²⁹. Under future warmer climate conditions characterized by an expanded land area in the Northern Hemisphere, it is plausible that the Northern Hemisphere may experience relatively greater warming compared to the Southern Hemisphere. Such a temperature gradient could potentially amplify the meridional temperature difference, which might, in turn, drive a northward migration of the ITCZ. This hypothetical mechanism could contribute to earlier seasonal occurrences and increased frequencies of TCs in the northern SCS. Although further research is needed to confirm this hypothesis, our reconstruction results still provide a new perspective for understanding the mechanisms and dynamics of seasonal changes in TC patterns.

Methods

Tridacna squamosa samples

Two Tridacna squamosa (abbreviated form: T. squamosa) shells were used in this study, both collected from the Xisha Islands, northern SCS. The SCS is the largest semi-enclosed marginal seas near China where TCs occur frequently (Fig. 1a). The modern Tridacna squamosa shell LHJ2 used for calibrating the relationship between Tridacna and TCs was collected in the Langhua Reef (16°3′N, 112°33′E), with a lifespan spanning from 12 May 2013 to 18 December 2018. Relative data were obtained from our previous study¹⁷. The T. squamosa shell A114 used for reconstruction of TC activity in the northern SCS was collected from the North Reef (17°05′, 111°30′E). The sample was sectioned in the field before transportation to the laboratory in Xi’an.

The T. squamosa shell A114 is 9.5 cm high and 14.8 cm long. Consistent with the pretreatment method of shell LHJ2¹⁷, one side of the shell A114 was cut from the umbo to ventral margin along the axis of maximal growth. Then a radial section was cut using a diamond-wire cutting machine to obtain a slab of 1 cm thick for imaging and Sr/Ca analysis (Fig. S1). The slab was polished by waterproof abrasive paper of 400 mesh, 800 mesh, 1200 mesh, 1500 mesh, and 2000 mesh, in sequence. Final polishing was done with a 0.5 μm diamond compound until the slab surface was flat and sleek.

Laboratory analysis

One powder sample was taken in the middle of the A114 sample for accelerator mass spectrometry ¹⁴C (AMS ¹⁴C) dating to determine the age of the T. squamosa A114. ¹⁴C dates were calibrated to calendar ages by the Marine20 calibration data set using CALIB rev. 8 (http://calib.org/calib/) and were corrected for a marine reservoir effect using ΔR = -142, σ = 25, determined for the average of nearly 5 sites ΔR from North Reef^18,19.

Powder samples for Sr/Ca analysis were micro-milled perpendicular to the annual growth bands of the T. squamosa sample at 0.1 mm intervals, which met the requirement of monthly resolution (~12 samples per year). For each test, about 20 μg powder sample was dissolved in 2 ml 5% HNO₃ and extracting 1 ml solution for the Sr/Ca measurement. The Sr/Ca ratios were determined by Inductively Coupled Plasma Optical Emission Spectrometer (ICP-OES) with radial plasma observation at the IEECAS. The spectral line of Ca is at 317.433 nm and Sr is at 407.771 nm. Lab standards N15 was inserted after every 5 Sr/Ca sample measurements to monitor the status of the instrument. The external precision for N15 is 0.29% (2.399 ± 0.007 mmol/mol, n = 246).

Micro-images at 20x magnification of the polished shell slab were produced on a Nikon A1 Laser Scanning Confocal Microscope (LSCM) at the IEECAS. The laser excitation wavelength is 488 nm. Scanning was performed along the Sr/Ca sampling path, and the images were concatenated using the built-in stitching function of LSCM. The micro-images derived from the LSCM showed that the shell had clear alternating layers of dark and light during its growth stage. A couplet of dark and bright layers indicates one daily growth layer (Fig. S1b, d).

Daily resolution shell age

As age increased, the daily growth layers became less distinct and difficult to distinguish. The reason may be that the juvenile shell is close to the boundary between the inner and outer layers of the giant clam, and is protected by the outer shell, making it less susceptible to seawater erosion than the shell of old age. In addition, the growth rate during the older stages of A114 was low, even 1/10 of the growth rate during the juvenile stages, so that the magnification of LSCM may not be able to clearly observe the daily growth layers. To reduce the error in counting the daily growth layers of A114 and minimize the year-to-year propagation of errors, we selected that part with clear daily growth layers, from the juvenile stage for counting.

The daily layers and the daily growth increments (referred to as daily growth rate, DGR) were measured from the LSCM images using CooRecorder 9.3 software (http://www.cybis.se/forfun/dendro/). Based on the number and width of DGR obtained by LSCM, combined with the AMS ¹⁴C dating result, a relative shell age with daily resolution was established, whose chronological length is the lifespan of T. squamosa A114. Previous studies have shown that the Sr/Ca ratios of Tridacna shells in the northern SCS exhibit obvious seasonal cycles. Among them, the Sr/Ca ratios of T. squamosa shells are positively correlated with SST, where the monthly SST minima occur near January, and have stable high values from May to September^20,21,22. We used the annual cycles of the monthly resolution Sr/Ca series of A114 to calibrate the daily resolution shell age. In order to conduct subsequent time series analysis, linear interpolation was performed on the DGR profile to ensure there were 365 data points per year.

TC events definition

Large fluctuations in the DGR over a short period are influenced by extreme weather events, such as cold surges during the boreal winter and TCs during the boreal summer–autumn^15,16,17. Previous studies have employed variation in the DGR of Tridacna (the deviation of DGR from its local mean) to represent these fluctuations^16,17. However, due to the relatively short lifespan of Tridacna, the influence of growth trends was not considered in their research. In fact, the DGR of Tridacna shells are influenced by environmental factors and their vital effects. Analysis of environmental signals in the DGR time series requires removal of any age-related growth trend. The growth trend was fitted by using a 6th polynomial, and indexing removes the age-related growth trend from the measured growth data by dividing measured growth by the estimated growth trend (Figs. S2a, S3a). Then, a standardized dimensionless growth index (SGI) was obtained by following equation^23,45:

$${SGI}=\frac{\frac{{measure}}{{trend}}-\bar{x}}{\sigma }$$

(1)

Where $x$ is the measured value divided by the growth trend, σ is the standard deviation of $x.$ We calculated a daily growth index to indicate the fluctuations affected by extreme weather events over a few days. To avoid the influence of negative and decimal values in SGI on the results, we first adjusted the SGI average from 0 to 9.8 (the average of the raw DGR records), and the adjusted record was called the growth index (GI, Fig. 3b). The variation of daily growth index (GIV) is defined by following equation:

$${{GIV}}_{i}=\frac{{{GI}}_{i}-\overline{G{I}_{i-1}^{i+1}}}{\overline{G{l}_{i-1}^{i+1}}}$$

(2)

GIV indicates the changes of the daily growth in one day relative to the average of the most recent 3 days, the positive or negative values indicate the relative increase or decrease^16,17. Then, abnormal fluctuations were identified as the extreme GIV records, which included extremely positive GIV (exceeded the 95^th percentile) and negative GIV (below the 5^th percentile) (Figs. S2c, S3c). TC events in Tridacna shells were identified by the presence of abnormal fluctuation persisting for two or more consecutive days, with a minimum interval of three days between events.

Modern T. squamosa LHJ2 have clear daily resolution chronologies, enabling direct comparison of Tridacna-based proxies with individual TC events. In contrast, for the fossil sample analysed here, the daily resolution shell age was obtained through interpolation rather than being based on absolute calendar dates. During the interpolation process, the minimum point of annual DGR variation was assigned to January 31st as the control point. This assignment is based on the consistent seasonal variation between DGR and SST, and the observation that January represents the coldest month on average, although coldest SST occasionally occur in February^22,25. This methodological limitation would introduce a potential chronological uncertainty of approximately one month. In this case, dividing the reconstructed TC events by month is more reliable for discussing the seasonal variation of past TCs. Monthly TC frequency estimates were calculated by the number of reconstructed TC events by shell in a month. The relative change in TC frequency (Δr) is computed as:

$$\varDelta r=\frac{{F}_{{past}}-{F}_{{modern}}}{{F}_{{modern}}}\times 100 \%$$

(3)

where$\,{F}_{{past}}$ and ${F}_{{modern}}$ are the Monthly TC frequency estimates at circa 3 ka BP and modern, respectively.

Considering the seasonal characteristics of TCs in the SCS and to avoid the impact of other extreme weather events on the growth of giant clams, such as cold waves in winter (December–February), this study only focuses on the TC events that occur during TC season, from June to November^25,26.

Calibration of modern TCs

To characterize the seasonal pattern in the vicinity of our reconstruction site under modern climate conditions, we analysed 6-h best track data from the Joint Typhoon Warning Center (http://www.metoc.navy.mil/jtwc/jtwc.html). The statistics focused on the Xisha Islands (109°–114°E, 14.5°–19.5°N) for the period January 1991 to December 2020. The result is referred to as TC_obs.

Given the scarcity of modern Tridacna specimens with long lifespans, the LHJ2 sample near the North Reef, as documented in published research, was selected. Employing the definition of TC events based on the GIV record described above, we reconstructed the TC events from 2013 to 2018 AD, designated as TC_LHJ2. The relationship between instrumental TC data and modern Tridacna proxy was established by correlation analysis between TC_LHJ2 and contemporaneous TC_obs, which took into account their autocorrelation. Significance tests are presented for coefficient of correlation combining Monte-Carlo iterations with frequency. The analysis was performed using a MTALAB package (https://oxlel.zoo.ox.ac.uk/resources/reconstats).

We determine the statistical significance of the TC_LHJ2 response using superposed epoch analysis (SEA). The SEA is a statistical method to identify consistent responses to events, by testing for the possibility of random occurrence. It is commonly used to detect the significance of volcanically forced climatic anomalies, where the key year list (volcanic event years) needs to be accurate⁴⁷. In this study, the TC event day list was used for this purpose. We studied the 15-day anomalies before and after each event. Epochal anomalies are considered significant when the TC_LHJ2 response below the 1st percentile or above the 99th percentile (i.e., one-sided epochal anomaly at p < 0.01).