A noticeable feature of the equatorial Pacific climate is its east-west asymmetry, with warmer ocean surface conditions to the west that supports atmospheric deep convection, and a colder east. Easterlies converge near the equator and drive surface waters to the west, deepening the thermocline to the west and shoaling it to the east; this allows equatorial upwelling to bring cooler waters on the eastern side. This asymmetry periodically breaks down in the modern-day climate, leading to transient El Niño conditions1,2. Persistent changes to this asymmetry that resemble the pattern of ENSO occurred in past climates, leading to the so-called ‘permanent El Niño-like’ conditions. While the surface climate pattern of these changes may resemble El Nino, they occur on longer timescales and through mechanisms that differ from the interannual case3. Marked “permanent El Niño-like” conditions were thought to occur in the Pliocene4,5, and they have also been proposed to occur at various points in the Quaternary6,7,8,9.

Understanding past mean changes to the east-west asymmetry is directly relevant to understanding tropical Pacific mean state changes under future warming10. Virtually all climate models project a pronounced warming in the eastern equatorial Pacific relative to the western equatorial Pacific. However, these projections fly against recently observed trends that show the opposite to have occurred during the last decades, suggesting that the model projections may be in error10. This potential error has strong implications for future regional projections of climate. Like interannual El Niño, multidecadal to multicentennial changes in sea surface temperatures (SST) in the tropical Pacific that mirror the ENSO pattern can lead to changes to regional climates throughout the globe mediated through atmospheric teleconnections. Paleoproxy information that documents changes to the tropical Pacific mean climate, in particular its east-west asymmetry, can provide a valuable perspective on this problem.

There is currently little consensus on the mean state of the Tropical Pacific during the Little Ice Age (LIA). SST reconstructions indicate relative warming in the eastern11,12,13 and cooling in the western equatorial Pacific14,15, suggesting a weak zonal gradient. In contrast, the majority of hydrological records suggest wetter conditions in the western tropical Pacific16,17,18,19,20, and drier conditions in the central21 and the eastern equatorial Pacific22,23 during the LIA, consistent with a strong zonal SST gradient phase during this time. However, a number of records are not consistent with this hydrological trend, either indicating drier (wetter) conditions in the western (eastern) equatorial Pacific during the LIA24,25, or showing no apparent trend26.

This apparent discrepancy could be explained by other potential drivers of precipitation changes in the tropical Pacific, namely the Intertropical Convergence Zone (ITCZ) and East Asian Summer Monsoon (EASM) and Australian Summer Monsoon (ASM)22,27. Presently, it is difficult to differentiate the contribution of the different drivers to shaping the observed paleohydrological patterns during the LIA, because a sufficiently distributed set of paleo proxies that are able to differentiate between these drivers is currently lacking. However, one feature of all of these other drivers is that they are associated with warm-season convective rainfall. A proxy that primarily measures autumn-time orographic rainfall—as we will show for our proxy—would not be as influenced by these various drivers, and thus the climatic interpretation would be less ambiguous.

Here we report a new rainfall record, based on leaf-wax hydrogen isotope data (expressed as δDwax values) from a 1,400-year-old sedimentary sequence from Bulusan Lake (N 12°45′, E 124°6′) in northeastern Philippines (Figs. 1 and 2). The region is a member of a recently identified global network of “Autumn Monsoon precipitation” regions28. Autumn monsoon regions occur on the eastern coast of land masses when trade winds impinge on orography and the temperature is still sufficiently warm to support moist convection. Peak rainfall occurs when these regions are poleward of the ITCZ, and thus embedded in easterly low-level trades, usually between October and February at our site (Supplementary Fig. 1a). While the western North Pacific monsoon also operates over this region, its contribution to the annual rainfall amount is smaller. Additionally, typhoons are a source of episodic precipitation, and can potentially account for up to 30% of northeastern Philippines rainfall in exceptional years29. ENSO is the main driver of interannual rainfall variability at Bulusan, resulting in reduced (increased) precipitation during El Niño (La Niña) years (Supplementary Fig. 1b). Consequently, a rainfall record from Bulusan is highly suitable to provide insights into the mean changes to the east-west asymmetry of the Tropical Pacific climate system during the last millennium.

Fig. 1: Map of mean annual precipitation in the tropical Pacific showing the location of our study site, Bulusan Lake, and the locations of other hydrological and sea surface temperature records.
figure 1

In the western Pacific, Bulusan Lake (BL), Clear Lake (W1)25, and Spooky Lake (W2)21 in Palau Island, Dongdao Island (W3)19, Makassar Strait (W416, W514,15), and Liang Luar (W6)51. In the central Pacific, Washington Island (C1)21, and in the eastern Pacific, El Junco Lake (E1)11, Poza de la Diablas (E2)24, and ocean multicore near Española Island (E3)13 in the Galapagos Islands. Overall wetter and drier conditions during the Little Ice Age, derived from the mapped hydrological records (circles), are depicted in blue and red, respectively. Relevant sea surface temperature records (triangles), with overall warmer and cooler conditions during the Little Ice Age, are depicted in red and light blue, respectively. The band of heavy precipitation indicates the ITCZ. Modified figure from ref. 95.

Fig. 2: Geographical and sedimentological features of the Bulusan site.
figure 2

a Topographic map of the Philippines showing the location of the Bulusan site. Data from PhilGIS96, the map generated in QGIS version 2.14.397. b Map showing Bulusan Lake (blue) within its larger landscape, with Mt Bulusan volcano in the west (green)98. c Age-depth model for the Bulusan sediment sequence, constructed with Bacon version 2.287,88 in R version 3.2.499. Probability distributions for the radiocarbon dates are shown in blue. The top of the sediment sequence is assigned the year the cores were collected. The shaded areas represent the 1σ probability age ranges by interpolating between AMS14C age control points. The red dashed line represents the age-depth calibration used in further analysis. SUERC-64011 is not shown here as it lies below the analysed section. d Optical and radiograph profiles of the Bulusan sediments, generated in ItraxPlot program100. The radiograph images depict the level of organic content, with higher values shown as lighter bands. The green line represents the validation of the measurement quality along the cores. For more details on the physical and chemical characterization of Bulusan cores, see Supplementary Note 8.

We show that a dramatic and abrupt rise in δDwax values occurred around 1630 AD and persisted until roughly 1900 AD. This shift is interpreted as a movement towards substantially drier conditions in the western tropical Pacific during the latter part of the Little Ice Age, due to changes in the mean state of the tropical Pacific associated with zonal sea surface temperature gradients. Our results reveal a major influence of multidecadal to multicentennial changes of zonal SST gradients on the hydrology of the tropical Pacific over the past millennium. We also show that pronounced changes in the mean state of the tropical Pacific can transpire within a span of a few decades, a behavior that existing fully coupled climate models fail to replicate.


Changes of δDwax in the Bulusan core

The top 300 cm from two overlapping sediment cores (BUL1 and BUL2) collected from Bulusan Lake provide a continuous, undisturbed sedimentary sequence spanning the last 1400 years (Fig. 2d and Supplementary Fig. 2) (see Methods). An independent and robust age-depth model (Fig. 2c) has been established using 10 AMS14C dates (Supplementary Table 1 and Supplementary Note 1). Bulusan sediments contain abundant terrestrial-lipid biomarkers, specifically the n-alkanes with 27, 29, and 31 carbon atoms (nC27–31) over this time period. We report D/H isotope ratios of those n-alkanes (expressed as δ2H or δD values), as well as an independent record of fossil pollen abundance from the same sedimentary sequence to ensure adequate identification of biomarker sources30 (Fig. 3) (see Methods).

Fig. 3: Paleorecords from the Bulusan Lake.
figure 3

A record of H/D isotope ratio measurements of leaf-wax alkanes (nC27-31), expressed as δD values with error bars of 1 standard deviation, and a record of fossil pollen abundance expressed as the percentage of grasses (Poaceae) and sedges (Cyperaceae), representing C4 plants, vs. other terrestrial plant taxa (trees, shrubs, and herbs), representing predominately C3 plants. Data plotted with ggplot2 package101 in R version 3.2.499.

The nC27–31 alkanes are a major component of the leaf waxes of higher terrestrial plants31. δD values of these alkanes (δDwax) in the surface sediments of smaller lakes fed by precipitation and groundwater, such as Bulusan Lake, are significantly correlated with local rainfall δD values, offset by a biosynthetic fractionation32. δDwax values can be additionally affected by evaporative deuterium enrichment of leaf water due to increased evapotranspiration under low atmospheric humidity32,33. At Bulusan, where the relative humidity is high year-round, leaf water deuterium enrichment is expected to be minimal33, and therefore δDwax at the site should faithfully record changes in δD values of precipitation (δDprecip).

δD values of the three abundant terrestrial-lipid biomarkers, nC27–31, are highly correlated in the Bulusan core, with a coefficient of determination (r) ranging between 0.69 and 0.81 (p < 10−6; Supplementary Fig. 3 and Supplementary Note 2), confirming a common origin of these alkanes. The average chain length of n-alkanes between nC25 and nC33 is stable throughout the record up until the 19th century when there is a slight decrease, possibly caused by the changes in the composition of surrounding vegetation due to increasing human activities in the area34 (Supplementary Fig. 4 and Supplementary Note 3). Fossil pollen from the same sedimentary sequence reveals that the plant assemblages surrounding the lake during the last 1,400 years contained only a small fraction of grasses and sedges (C4 plants) (Fig. 3) and were dominated by trees and shrubs (Supplementary Note 4), the likely source of the nC27-31 alkanes35. The percentage of grass pollen remained stable between 0–5% throughout the record, indicating that there was no major change in the proportion of C4 to C3 plants, and rules out the possibility that vegetation changes have significantly affected δDwax values32. However, the nC27 alkane (average relative abundance of 20%) in the record shows higher variability in comparison to nC29 (average relative abundance of 35%) and nC31 (average relative abundance of 45%) indicating more variable sources (Fig. 3, Supplementary Fig. 4, and Supplementary Note 5). Furthermore, previous studies have suggested that nC31 is a major constituent of grasses and herbs36 in addition to trees35, which may at least partly explain its reduced variance compared to nC29. While δDwax values of all three alkanes show the same pattern, we focus our interpretation on the nC29 alkane in order to minimize the potential distortion of the precipitation signal due to variability in alkane sources and deuterium enrichment differences between C3 and C4 plants.

δD values of nC29 varied by up to 45‰ during the study period indicating significant hydrological changes at Bulusan Lake over the last 1400 years (Fig. 3). Relatively low and stable δDwax values occurred during AD 600–1600, followed by a distinct and abrupt increase in δD values at AD 1630 ± 35 (i.e., 20‰ within a period of a few decades, see Supplementary Table 2b, Supplementary Fig. 5, and Supplementary Note 6). These high δDwax values persisted until ~AD 1900 ± 40 only to drop just as abruptly to their lowest values in the record over the last 100 years. Based on these elevated δDwax values, we infer significantly drier conditions during the second half of the LIA (~AD 1630–1900), compared to the rest of the record.

Drivers of precipitation δD values at Bulusan

We use a range of instrumental datasets and global circulation model outputs to identify drivers of δDwax (and therefore δDprecip) variations at Bulusan, using the interannual variability to isolate enriched and depleted years. First, we use an isotope-enabled general circulation model (isoGSM2; see Methods) to understand the large-scale drivers of δDprecip values in the northeastern Philippines. Since the northeastern Philippines is spatially too limited for the rainfall to be adequately resolved in isoGSM2, we analyze instead the δD of precipitable water (hereafter δDpw) for the isoGSM2 gridpoint closest to that region (N 13°, E 124°); variations in δDpw are found to closely match variations in the δD of station rainfall measured over that region (Supplementary Fig. 6 and Supplementary Table 3). We generate an interannual index of amount-weighted δDpw (using July and the following June as the start and end months), and compare years with enriched δDpw with depleted years (Fig. 4a). There is a clear El Niño influence for enriched years—the 1982–83, 1997–98, and 2015–16 El Niños stand out (Fig. 4a), and moreover the amount-weighted δDpw index is correlated with an interannual index of ENSO (Nino3.4 averaged over the peak months of October-February, Fig. 5a) at r = 0.75 (p < 0.001; Fig. 4b). Years with more positive δDpw values are drier over the northern Philippines during the autumn and winter rainfall months (October-February), because of reduced moisture flux from the northeasterly trades impinging on the northeastern Philippines (Fig. 4c, d). SST and rainfall changes associated with enriched relative to depleted δDpw years (Fig. 4e) shows warming SST in the eastern equatorial Pacific and slight cooling over the western equatorial Pacific indicative of a weak zonal SST gradient, and commensurate changes to rainfall.

Fig. 4: Examination of δD values of precipitable water (δDpw) of enriched vs. depleted years using isoGSM2 output.
figure 4

a Time Series of the amount-averaged δDpw over Bulusan, and correlation. δD at the gridpoint closest to Bulusan was amount-averaged from the preceding July through the following June. The mean (−151.23 per mil) was subtracted and the time series detrended prior to plotting. Years with high δDpw are enriched. b Scatterplot between the amount-weighted δDpw with an interannual index of ENSO (Nino3.4 averaged over the peak months of October-February). The two time series are correlated at r = 0.75 (p < 0.001). c Climatological vertically-integrated moisture flux during the northern Philippines wet months of October-February (vectors; reference vector is 300 (g kg−1).(m s−1)), and the October-February precipitable water (shaded; units are kg m−2). d Difference in the vertically-integrated moisture flux (vectors; reference vectors is 100 (g kg−1).(m s−1)) and precipitable water (shaded; units are kg m−1) averaged over October-February, enriched minus depleted years; the eight most enriched years in (a) are used for the enriched composite; similarly for depleted years. The precipitable water field is shaded only if the difference is significant at the 5% level, using a two-sided t-test; vectors are shown only if the difference for either the zonal or meridional flux is significant at the 5% level. For enriched years, there is reduced moisture flux by the northeasterly trades into the northern Philippines. e The difference in sea surface temperature (shaded, contour interval 0.5 K) and rainfall (contour interval 1.5 mm d−1) averaged over October-February, for enriched minus depleted years. The sea surface temperature field is shaded only if the difference is significant at the 5% level, using a two-sided t-test.

Fig. 5: Relationship between ENSO, precipitation, and precipitable water δD at Bulusan.
figure 5

a Nino3.4 averaged over October-February which we use as an interannual index of ENSO. We use +0.7 K as the threshold to identify El Niño years, and −0.7 K for La Niña years (horizontal dashed lines); the remainder of the years are neutral years. b δD of precipitable water for the El Niño (red), neutral (black), and La Niña (blue) composites at Bulusan. The thicker lines indicate the months when the difference between the El Niño and La Niña composites are significant at the 5% level using a two-sided t-test. Here, we use the isoGSM2 gridpoint nearest Bulusan Lake (N 13°, E 124°) for the δD field. c Precipitation for the El Niño (red), neutral (black), and La Niña (blue) composites at Bulusan. The thicker lines indicate the months when the difference between the El Niño and La Niña composites are significant at the 5% level using a two-sided t-test. Here, we use the CRU TSv4 rainfall gridpoint closest to Bulusan Lake, at N 12°45’, E 124°15’ (see Methods). Together, (b) and (c) shows that rainfall is reduced, and also isotopically heavier, for El Niño years compared to La Niña years.

To further assess the influence of ENSO on δDpw at Bulusan, we composite δDpw of El Niño and La Niña years over the Bulusan gridpoint (N 13°, E 124°) and find that precipitable water is heavier (lighter) during El Niño (La Niña) years compared to neutral ENSO years, in particular over the autumn and winter months (Fig. 5b). Despite being highly negative and a mismatch in absolute values, as they are averaged for the entire atmospheric column of water vapor, the relative change between El Niño and La Niña years in modeled δDpw values of about 11‰ is consistent with surface water observations37. Regionally, precipitable water is isotopically enriched over the western Tropical Pacific and Maritime continent for El Niño years relative to La Niña years (Supplementary Fig. 7), accompanying the El Niño-driven reduction to rainfall over those regions. The two are causally related: El Niño reduces deep convection over the western tropical Pacific, and since convection depletes the isotopic composition of water vapor, water vapor in this region becomes isotopically enriched (Supplementary Fig. 7).

Similarly, differences in precipitation amount at Bulusan occur during the autumn and winter months between El Niño and La Niña years. Rainfall data from the Climate Research Unit TS version 4.0338 dataset for the gridpoint nearest Bulusan (N 12°45′, E 124°15′) (see Methods) show that La Niña years have more precipitation than neutral ENSO years, and El Niño years have less (Fig. 5c). A similar relationship for precipitation and δDprecip can be observed in the Global Network of Isotopes in Precipitation (GNIP) data (Supplementary Fig. 8 and Supplementary Note 7), where the major difference between 1997 and 1998 is during the autumn and winter.

Our results demonstrate that δDpw (and therefore also δDwax) values are isotopically enriched, and rainfall is reduced, during El Niño years (weak zonal gradient) relative to La Niña years. The bulk of the rainfall changes occur during the autumn and winter months (October-February). This relationship to ENSO allows us to use Bulusan δDwax to monitor past changes in the east-west asymmetry of the equatorial Pacific climate.


Based on our analysis above, the distinct and abrupt increase in nC27-31 alkane δD values in the Bulusan paleorecord between ~AD 1630–1900 suggests an abrupt shift in the hydroclimate of the western tropical Pacific during the late LIA towards significantly drier conditions. Since the lowland tropics are characterized by low-temperature variability during the Holocene26,39, the ‘temperature effect’ on the isotopic composition of precipitation in these regions should be negligible27 and is therefore not normally included in the interpretation of the δD changes in the paleorecords from the tropical Pacific from this time period.

While several paleohydrological records from the region have had changes in their isotopic signatures linked to shifts in the geographic position and extent of ITCZ22,27, our isoGSM2 analysis confirms that at Bulusan δDprecip is primarily an autumn-time rainfall proxy28, with ITCZ as a secondary contributor to variability in the record through its influence on summer precipitation. Specifically, our results indicate that ENSO is the key driver of present-day precipitation δD variations at Bulusan, as the decrease in convective activity over the western tropical Pacific during El Niño years leads to an enrichment of water vapor in the region (Supplementary Fig. 7). Furthermore, while previous interpretations of changes in precipitation stable isotopes in the region have focused on precipitation amount, our findings are supported by several recent studies which have suggested that ENSO can directly influence precipitation stable isotopic values on an interannual scale via alterations in the strength of convective activity across the tropics (e.g., refs. 40,41,42). Additionally, a number of studies have demonstrated a mismatch between local precipitation and precipitation stable isotopes in the western tropical Pacific43,44,45,46, instead implicating factors beyond the amount effect such as vapor transport history and the degree of vapor parcel distillation. Therefore, the variability in precipitation stable isotopes in the western tropical Pacific seem to be reflective of regional-scale atmospheric circulation rather than local rainfall amount.

Based on our modern climate system analysis showing ENSO having a dominant role in shaping interannual precipitation δD dynamics at Bulusan, we propose a shift in the tropical Pacific mean state tied to the establishment of a weak zonal SST gradient in the tropical Pacific was responsible for the hydrological changes recorded during the second half of LIA in the Bulusan paleorecord. The reduction in the east-west asymmetry in the tropical Pacific mean state results in a reduced Walker circulation, leading to drier conditions over the western Pacific similar to what occurs during El Niño conditions today. Similarly, there would also be changes to the structure of convection in the tropical Pacific whereby the ITCZ of the northern tropical Pacific and the South Pacific Convergence Zone (SPCZ) collapses to a single ITCZ near the equator in the western and central equatorial Pacific, similar to what is seen during strong El Niño events today47. In the eastern Pacific, the northern ITCZ shifts equatorward as well. Since ENSO and the ITCZ are known to be strongly linked on interannual scales48,49,50, we cannot exclude the potential for secondary influence by summer precipitation changes on the Bulusan record driven by the ITCZ dynamics, though our results suggest that changes in this climatic phenomenon is not something that Bulusan δD proxy predominately monitors.

Our results are supported by SST-based records from the tropical Pacific. The period of drier conditions at the Bulusan site corresponds to the period of relative warming in the eastern11,12,13 and further cooling in the western equatorial Pacific14,15 during the LIA, interpreted as a weak zonal SST gradient (Fig. 6). Importantly, the onset of cooling temperatures recorded in the Makassar Strait SST record14,15 and of warming temperatures detected in the Galápagos multicore KNR195-5MC42C record13 during the LIA period closely match the timing of the distinct increase in alkane δD values in the Bulusan paleorecord.

Fig. 6: Comparison of the Bulusan record (nC29 alkane δD values) with regional SST records.
figure 6

In the western Pacific, Makassar Strait (W5, the error bars represent the standard error)14,15,102, and in the eastern Pacific, El Junco Lake (E1)11,12,103 and ocean multicore near Española Island (E3, the error bars represent 1 standard deviation)13,104 in the Galápagos. Data plotted with ggplot2 package101 in R version 3.2.499. The gray strip denotes the period of significant pronounced change in the Bulusan record (~AD 1630–1900).

While several paleohydrological records from the region indicate the opposite situation, i.e., overall wetter conditions in the western Pacific16,17,18,19,20, and overall drier conditions in the central21 and the eastern Pacific22,23 during the LIA, a number of these and other records indicate a change to reduced precipitation in the western Pacific (i.e., Makassar Strait16) and increased precipitation in the eastern Pacific (i.e., Poza de la Diablas24) between AD 1600 and 1900, which supports our interpretation (Fig. 7). Arguably, the strongest support from the tropical Pacific for our results comes from the Liang Luar cave stalagmite δ18O record from Flores51, which also suggests a marked shift from wetter to drier conditions at ~1600 AD. Data from ref. 52 from Eastern Africa offer additional support with strong climatic connections to ENSO today. The authors argued that the East African hydrological changes formed part of a larger tropics-wide climate change linked to warmer SSTs in the western Indian Ocean and cool SSTs in the eastern Indian Ocean and western Pacific warm pool, analogous to what is seen from teleconnections during El Niño events in present-day climate. The timing and nature of the East African hydrological changes match those seen in the Bulusan record, consistent with the onset of a weak zonal SST gradient.

Fig. 7: Comparison of the Bulusan record (nC29 alkane δD values) with other paleohydrological records in the region.
figure 7

In the western Pacific, Clear Lake (W1, the error bars represent 1 standard deviation)25, Makassar Strait (W4)16, and Liang Luar (W6)51,105, in the eastern Pacific, Poza de las Diablas (E2, the error bars represent 1 standard deviation)24, and outside the tropical Pacific, the East Africa Lakes MCEOF record (EAF, 68 and 95% uncertainty bounds are shown in dark and light blue, respectively)52. Data plotted with ggplot2 package101 in R version 3.2.499. The gray strip denotes the period of significant change in the Bulusan record (~AD 1630–1900).

A reduction in δDwax values in the Bulusan record around 1900 AD suggests a return to wetter conditions in the western tropical Pacific, indicative of a strong zonal SST gradient. Our results are in congruence with instrumental data which record a continued strengthening of the zonal SST gradient in recent decades, along with the sharp increase in the greenhouse gas (GHG) concentrations53. In contrast, state-of-the-art climate models predict that rising GHGs decrease the SST gradient across the tropical Pacific41. This striking mismatch between models and observational data has been recently explored by ref. 10, who attribute the erroneous model response to a well-known cold bias in the Pacific cold tongue as simulated by the coupled models. The Bulusan record provides a longer temporal perspective on this question, indicating that a trend of weak zonal SST gradient in the tropical Pacific may have started as early as early 1900s, though the low resolution of the age-depth model in the top section of the core hinders a more exact timing of the onset of this trend.

A remarkable feature of our record is the abrupt transition from low to high δDwax values between 1600 and 1650. While the resolution of the age-depth model precludes the estimation of the exact timing or rate of this transition in the Bulusan record (Supplementary Fig. 9), the results nevertheless emphasize that marked changes to the tropical Pacific mean state asymmetry can occur on decadal timescales. The shift from wetter to drier conditions at the Bulusan site happened over the period of only ~50 years, as did the subsequent return to wetter conditions ~270 years later. This finding is in accordance with SST-based zonal gradient reconstructions from the region. For instance, ref. 12 found that a change to a state of the weak zonal gradient of tropical Pacific SST at ~AD 1650 happened over the course of a few decades, very similar to the hydrological shifts recorded in the Bulusan record. Despite the increasing paleo-evidence, current fully coupled climate models do not reproduce such abrupt shifts in tropical Pacific mean state, underlining potential limits in our mechanistic understanding.

Studies to date tend to describe the LIA in the tropical Pacific as having either an overall weak or strong zonal SST pattern, implying the entire period should be characterized by one mode or the other. However, the Bulusan record suggests, along with a number of other records in the tropical Pacific region16,24,25,51, that there may have been a shift in the Pacific mean state dynamics approximately halfway through the LIA period. It is possible that the 1630–1900 AD event was caused by internal variability of the tropical Pacific climate system. However, while the relative roles of external forcing versus internal climate dynamics in generating (multi)decadal scale variability in the tropical Pacific have been widely discussed54,55,56,57, an internally generated process driving (multi)decadal oscillations detected in palaeoclimate records has yet to be conclusively determined. In contrast, various means by which ENSO may influence the tropical Pacific mean state, and vice versa, have been previously described in refs. 58,59,60,61,62,63,64,65, highlighting how the relationship between the tropical Pacific mean state and ENSO variability is complex and interactive13,66,67. For instance, the influence of volcanic activity on ENSO has been widely demonstrated, with most observations (e.g., refs. 68,69) and paleoclimate records (e.g., refs. 70,71,72,73), as well as several modeling studies, suggesting that major tropical volcanic eruptions give rise to El Niño events73,74,75,76,77,78,79,80. Thus, one can imagine how the increased volcanic activity during LIA81,82 may have amplified ENSO activity, and subsequently, caused or contributed to a shift to a weak zonal SST gradient in the tropical Pacific. While it is not possible to determine with existing data whether this shift was a result of internal variability or external forcing, we anticipate future studies focusing on this important question. For instance, high-resolution records of the subsurface thermal structure, a critical component of El Niño events83, especially on longer timescales84,85, would help to understand the patterns driving the SST gradient as well as the precipitation changes during this time period.

The prominent role of meridional shifts in the ITCZ in shaping tropical Pacific climate during the last millennium, and especially during the LIA, has been well established in previous studies12,16,17,24,27. Less certain is the nature and extent of the influence of (multi)decadal zonal ENSO-like shifts in the tropics during this time, and its interaction with meridional forcing from outside the tropics51. Our study directly addresses this knowledge gap. Overall, although the evidence presented here originates from a single site, our data, combined with other relevant paleohydrological and SST records from the region and modern observational and modeling data, make a strong case for a rapid change in tropical Pacific mean state linked to zonal SST gradients during the second half of the LIA. Our findings emphasize the prominent role of the tropical Pacific ocean-atmosphere mean climate, particularly its east-west asymmetry, in shaping precipitation patterns in the tropical Pacific over the past millennium through rapid, decadal changes in its mean state. Through atmospheric teleconnections, these changes could then be communicated globally to shift regional climates elsewhere, akin to what happens during an ENSO event today.


Study site and material collection

Bulusan Lake is a 0.28 km2 oval-shaped lake located at 360 m.a.s.l., at the foot-slopes of Mount Bulusan volcano at the southeast end of the Bicol Volcanic Arc86. It consists of two basins: a small round basin, and a larger narrow basin, which are 12 and 24 m deep respectively. The two basins comprise a closed system fed primarily by precipitation and groundwater. The vegetation surrounding Bulusan Lake, and up the slope of Mt Bulusan, is classified as a Dipterocarp rainforest, i.e., aseasonal lowland tropical rainforest dominated by the Dipterocarpaceae tree family34. Two overlapping sedimentary sequences, 5.3 m BUL1 and 5.0 m BUL2, were collected as 1 m segments from the smaller basin with a Livingstone Piston corer using an anchored platform in March 2013. The sediments were sealed in protective plastic containers and shipped to the University of Oxford where they have been stored at 5 °C. Prior to subsampling, the cores were physically and chemically characterized (Supplementary Note 8).

Age-depth modeling

A robust age-model for the Bulusan sediments has been established with AMS14C dates (Supplementary Table 1) and the surface sediment set to 2013 (the year the core was collected). The radiocarbon analysis was done on identified terrestrial plant macrofossils (e.g., leaves, twigs) to avoid reservoir effects and minimize age uncertainty. AMS14C samples were measured at the NERC Radiocarbon Facility, East Kilbride. The age-depth modeling of the Bulusan core was performed using a Bayesian statistics approach to reconstruct accumulation histories for deposits with the Bacon 2.2 package in R87,88 at 1 cm resolution and the IntCal13 radiocarbon calibration curve for the Northern Hemisphere89, and model output estimates the age span per cm being between 4.5–6 years. For further details, see Supplementary Note 1.

Biomarker extraction, identification, and quantification

Sediment samples (n = 75) were sampled at the University of Oxford as a continuous series of 4 cm long slices (i.e., 2–6 g dry weight per sample) from a 300 cm long sequence, with each slice corresponding to a ~20-year interval. The samples were freeze-dried, and lipid biomarkers were extracted at the University of Potsdam with a dichloromethane (DCM): methanol mixture (9:1) at 100 °C and 103 bar using a Dionex accelerated solvent extraction system (ASE) 350. The extracts were separated into aliphatic and aromatic/alcohol/fatty acid fractions by solid phase extraction (SPE) in 8 ml glass columns filled with 2 g silica gel (0.040 to 0.063 mm mesh size) using hexane and DCM: methanol mixture (9:1) for elution, respectively, for a detailed description see the manual SPE extraction procedure in Rach et al.90.

The n-alkanes (nC23-33 homologs) from the aliphatic fraction were identified and quantified at the University of Potsdam using a gas chromatograph (Agilent GC 7890-A) equipped with a mass selective detector-mass and a flame ionization detector (FID) (Agilent MS 5975-C) coupled via an electronic split interface. The MSD trace was employed for compound identification. The peak area from the FID chromatogram was used for compound quantification by calibrating against the peak area of the internal standard 5α-androstane. Sulfur and unsaturated hydrocarbons were removed from the aliphatic fraction by elution through a pipette column of activated copper powder and a pipette column of an activated mix of silver nitrate and silica gel, respectively, by applying a combination of hexane and dichloromethane (DCM). The resulting aliphatic fraction produced clean chromatograms and no unresolved complex mixture (UCM) was detected (except in Samples 1–6).

Compound-specific hydrogen isotope measurement of n-alkanes

Compound-specific hydrogen isotope ratios of the aliphatic fraction were measured at the University of Potsdam using a Thermo Fisher Delta-V-Plus Isotope Ratio Mass Spectrometer coupled to a Thermo Fisher TraceGC 1310 gas chromatograph. The samples were measured on a 50 m HP ultra 1 column with an inner diameter of 0.2 mm and a film thickness of 0.33 μm. The samples were measured with the following GC-temperature program: (i) temperature was held for 2 min at 40 °C, (ii) temperature was increased to 256 °C at a rate of 4 °C per min, (iii) oven temperature was increased to 300 °C at 2 °C per min, and (iv) the final temperature was held for 45 min. The alkane standard mix was measured with a shorter program where the final temperature of 300 °C was held for 3 min only, as all standard peaks had eluted. The samples were measured in triplicates, with the exception of the top six samples which were measured in duplicates due to low compound concentrations.

A standard containing nC16 to nC30 alkanes with known δD values (mix A4, Arndt Schimmelmann, University of Indiana) was measured in duplicate at the beginning and the end of each sequence or after six sample injections, and then used for normalization of δD values to the Vienna Standard Mean Ocean Water (VSMOW) scale. The mean standard deviation of all measured samples for the nC27, nC29, and nC31 alkanes (n = 75) was about 1.6‰, 1.3‰, and 0.9‰, respectively. The H3+ factor was determined once a day and remained constant over the measurement period, indicating stable ion source conditions.

Isotope evaluation

Compounds with intensities below 800 mV (on mass 2) (i.e., nC23 and nC25) were not used for evaluation, since this was outside of the linear range of the GC-IRMS system. In addition, an exclusively higher terrestrial plant origin of nC25 is difficult to assume because the alkane can be produced by aquatic plants such as Potamogeton. Therefore, for our interpretations we only used δD values of alkanes with 27, 29, and 31 carbon atoms (nC27-31), which showed clean, baseline-separated peaks in all samples (except nC31 in Sample 2 which coeluted with an unidentified substance). The nC27-31 alkanes are the most widely used leaf-wax biomarkers and their hydrogen isotopic composition has been shown as a robust recorder of plant source water δD along extraordinarily steep climatic gradients31.

Extraction and processing of fossil pollen

Sediment samples were extracted as a continuous series of 4 cm long 1 cm3 slices from the top 3.0 m of the composite Bulusan core, with each slice corresponding to a ~20-year interval. Standardized protocols were employed for the extraction and processing of fossil pollen from the sediment samples91. 500 terrestrial pollen grains were counted in each sample (and identified to family or genus level where possible), and the relative abundance of grasses (Poaceae) and sedges (Cyperaceae) at each level was calculated as the proportion of Poaceae and Cyperaceae pollen in the total terrestrial pollen counts (expressed in percentages), at the same extent and resolution as the δDwax records.

Modern rainfall data

We use the Climate Research Unit gridded Time Series version 4.03 dataset (CRU TSv4)38 of monthly mean climatic quantities. Individual station rainfall data are combined and gridded onto a 0.5-degree horizontal grid using angular distance weighting. We use the gridpoint closest to Bulusan lake position (N 12°45′, E 125°15′), to represent the rainfall over that region. We also only use the time period overlapping with the isoGSM2 data, from 1979 to 2017.

Isotope-incorporated Global Spectral Model version 2 (IsoGSM2) simulation of precipitation isotopes

Output from an isotope-enabled general circulation model, the isoGSM version 292 between 1979 to 2017, is used in this study30. IsoGSM2 is based on a version of the Scripps Experimental Climate Prediction Center’s global spectral model, but with the large-scale fields nudged towards the National Center for Environmental Prediction version 2 (NCEP2) reanalyzes92,93. Nudging is done for temperature, zonal and meridional winds only, and only for large (>1000 km) spatial scales. In this respect, the isoGSM2 simulation is similar to NCEP2, but with the addition of precipitation isotopes. The simulated precipitation isotopes have been shown to be realistic when compared to observations94; however, because of the relatively coarse horizontal resolution of isoGSM2 (T62, corresponding to 1.89 degrees at the equator), northeastern Philippines rainfall is not adequately resolved. In lieu of using precipitation isotopes from isoGSM2, we use the simulated δD of precipitable water (δDpw) at the gridpoint closest to Bulusan Lake to assess changes to the δD of rainfall. The underlying assumption here is that changes to δD of rainfall reflect changes to the δD of vapor supplying the rainfall. We support this assumption by showing that changes in the simulated δDpw compared favorably with observed δD measured by GNIP (see Supplementary Fig. 6 and Supplementary Note 9).

ENSO analysis

We derive an interannual index of ENSO by taking the average of sea surface temperature monthly anomalies over the Nino3.4 region (5S-5N, 170W-120W), and then averaging over the ENSO peak months of October through February; the index is subsequently referred to as Nino3.4ONDJF. To obtain the El Niño and La Niña composites for Fig. 5, we define years where Nino3.4ONDJF > 0.7 as El Niño years, and Nino3.4ONDJF <−0.7 as La Niña years.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.