Introduction

Most regions in the world are dominated by either a winter or a summer rain regime. One of the few exceptions is the Indus River Basin and surrounding regions in northwest South Asia that are climatically and environmentally diverse, and feature abrupt transitions between a dominant summer versus winter rain regime (Supplementary Fig. S1 and Supplementary Table S1). Here, we compare the six months of the summer monsoon (May through October, i.e., summer rain) with the remaining winter and spring months (November through April, hereafter referred to as winter rain). For agricultural societies, the seasonal timing and intensity of precipitation are more relevant than total annual rainfall, because a reliable and somewhat predictable timing of water supply is vital for agricultural planning1,2. With a robust multi-seasonal water supply during the mid-Holocene3,4,5,6, the Indus River Basin emerged as a favorable area for the development of the complex society known as the Indus Civilization (5000–3600 years ago). However, the intensity of seasonal rainfall in this region appears to have changed considerably since the mid-late Holocene transition c. 4200 years ago. The impact of these changes in seasonal water supply on rainfed agriculture warrants detailed investigation, particularly within the context of archaeological research about societal resilience and decline7,8,9,10,11,12.

The seasonality component of precipitation variability has increasingly garnered attention in studies from South Asia13,14. Still, the contribution of the Indian Winter Monsoon (IWM) is often overlooked because the Indian Summer Monsoon (ISM) provides >80% of annual rainfall where most studies have been undertaken. Yet, IWM precipitation is particularly important in the Indus River Basin, where precipitation from the winter westerlies dominates the snowpack15 and late dry season intensity affects agricultural practices and yields. Water availability on the floodplain is also influenced by snow and ice melt during the summertime, which contributes a considerable portion of the annual runoff that feeds the headwaters of the massive Indus/Punjab and Ganges River systems16. In the western extent of the Indus River Basin, for example, the ratio of winter to summer rainfall exceeds 1:1 (Fig. 1), emphasizing the need to better understand past changes in hydroclimate seasonality of this region.

Fig. 1: Local proportion of winter:summer precipitation.
figure 1

Large Indus cities are labeled and shown with the site distribution during the Mature Harappan (c. 4.6–3.9 ka BP, orange) and Late Harappan (c. 3.9–3.6 ka BP, red) periods. Contour lines represent the ratio of IWM (Nov–Apr) to ISM (May–Oct) precipitation based on the 0.25° GPCC v2018 dataset from 1951–2000 rain gauge data86.

Full size image

A consensus is emerging that some form of increased drought affected the landscape of northwest South Asia during the mid-late Holocene transition (i.e., the 4.2 ka event)17,18,19,20, although the exact timing and magnitude of the drier period(s) remain uncertain. Paleoclimatic reconstructions suggest a drier than normal mid-late Holocene transition in the winter-rain dominated parts of the Middle East21,22,23, as well as in parts of the Indian and Asian Summer Monsoon domains of eastern India and China1,24,25,26,27. With previous studies exploiting paleoclimate records from either the westerlies-dominated parts of Western Asia or the ISM domain, little is known about climate dynamics in the region of interaction between these two climate regimes. In this study, we focus on the bimodal rainfall regime of northwest South Asia, including the timing and duration of multi-season drought that has been often hindered by large age uncertainties in available paleoclimate records from the region and an overemphasis on the ISM. Specifically, we aim to reconstruct indications of both the winter and summer rainy seasons from our multi-proxy time series and evaluate the impact these changes may have had on the Indus Civilization.

Although several speleothem records from the Himalaya and northeast India track Holocene ISM strength, and some even cover the mid-late Holocene transition at 4.2 ka BP (kilo annumbefore present, with respect to 1950 CE), e.g., refs. 25,27, very few are located in the western Indian domain28 that simultaneously receives substantial precipitation from the IWM. Furthermore, none of these records incorporate trace element data to independently check conclusions based on speleothem δ18O. Perhaps due to the difficulty of interpretation, δ13C data are also often omitted from discussion although this proxy can inform our understanding of local environmental conditions29,30. Consequently, there is a great need for seasonally-resolved, multi-proxy paleohydroclimatic reconstructions from the Indus River Basin. Here, we aim to characterize hydrological seasonality during the millennium after the mid-late Holocene transition using a speleothem record from the west-central Himalayas. We track the evolution of summer and winter moisture in a seasonally resolved, multi-proxy stalagmite record that begins at 4.2 ka BP from Dharamjali Cave (29.5°N, 80.2°E).

Dharamjali Cave is a shallow (<14 m deep), climatically-responsive cave system with sub-seasonal infiltration dynamics (see Supplementary Discussion for additional information). The 25-cm-long DHAR-1 stalagmite was located at the far end of the cave. The stalagmite is predominantly aragonite and variations in oxygen isotope values (δ18Ostal) are used here to reconstruct regional hydrological changes. The δ18Ostal record of DHAR-1 reflects seasonal variations in the δ18O of rainfall, which is dominated today by the ISM (Supplementary Fig. S2). In northwest South Asia, the moisture source region is the first-order control on ISM δ18Oprecip, which includes the highly evaporated surface waters of the Arabian Sea (0.4 to 0.9‰) and the Bay of Bengal (−2 to 0.5‰)31,32, as well as 18O-enriched rainfall recycled by evapotranspiration from the continent33 (Supplementary Fig. S3). Strong ISM seasons are generally characterized by low δ18Oprecip resulting from an increased transport path length (Rayleigh distillation) and reduced sub-cloud evaporation34,35. In contrast to the ISM, the δ18O of IWM rainfall is high because it draws more of its moisture from the proximal Arabian Sea36,37 and it loses 16O through partial re-evaporation of rainfall in dry air. Winter precipitation is predominantly carried by western disturbances (WDs) that originate over the Mediterranean Sea or mid-Atlantic and travel eastward38,39.

Local environmental conditions can also be traced using speleothem carbon isotopes (δ13Cstal). Shifts in δ13Cstal are driven by a combination of factors, including: overlying vegetation, soil respiration, bedrock dissolution regime, cave ventilation, drip rate, and prior calcite or aragonite precipitation (PCP/PAP) during the driest season or intervals30. The dense C3-type forest above Dharamjali Cave today40 may well have once featured more drought-resistant C4-type vegetation, as reconstructed for parts of the northwestern Himalaya c. 4 ka BP41,42. In particular, the growth and resilience of the Chir pine species (Pinus roxburghii) dominantly present above Dharamjali cave today40 is known to be particularly sensitive to winter and spring rainfall43,44. Thus, we might expect increasing δ13Cstal values in response to long-term drought-induced vegetation changes, particularly driven by winter and spring moisture limitation, e.g., refs. 43,44,45,46. Other factors influencing δ13Cstal are relevant at the seasonal to interannual scale, and fluctuate more dynamically during the winter season. For example, in winter the cave air temperature exceeds that of the outside air (promoting enhanced ventilation), drip rate slows to 3–8 drops/min compared to 20–30 drops/min in summer40, and PAP is most likely to occur in air-filled voids in the epikarst as potential evapotranspiration exceeds precipitation during the drier winter months. Taken together, increased cave ventilation, lower drip rates, and PAP in wintertime all lead to higher δ13Cstal and more bedrock-like values, though the colder winter season also features minimal soil respiration that acts to lower δ13Cstal. Cave CO2 monitoring data are not available at Dharamjali Cave, but the sum of mechanisms suggests that cooler, longer, and/or drier winters lead to higher δ13Cstal values (see Supplementary Discussion for a more extensive analysis of δ13C). Seasonal cave monitoring would be required to disentangle the primary effects responsible for δ13Cstal values and provide more certainty about their paleoclimatic interpretation.

Shifts in speleothem calcium isotopes, specifically δ44/40Ca, are similarly useful aridity proxies via PCP/PAP processes in caves, and have been used to interpret past rainfall amount more quantitatively than is possible with δ18O47,48,49. During carbonate precipitation, 40Ca is preferentially incorporated into the solid phase relative to the heavier isotope 44Ca. Thus, water that infiltrates more slowly and is given more time to degas and precipitate before reaching the stalagmite under drier conditions will yield a smaller difference between the two isotopes and a resulting higher δ44/40Castal. This concept underlies the interpretation of any PCP or PAP proxy, and directly connects this proxy to rainfall amount50. Importantly, an aridity threshold is required before any PCP/PAP effects are noticeable—therefore, such proxies tend to mainly reflect changes in the length or severity of the driest season because only an extreme change in wet season rainfall amount would reduce cave saturation to an appreciable degree to cross the threshold for this proxy14,48.

Additionally, we use trace element changes of DHAR-1 as independent tracers of past hydrologic changes50,51,52. The distribution of U2+, Sr2+, and Ba2+ in aragonite speleothems reacts to PAP dynamics in the epikarst53,54. With intensified and/or prolonged dry-season (winter) aridity and associated PAP, concentrations of uranium (DU(Ar) » 1) and strontium (DSr(Ar) > 1) are expected to decrease in the resulting speleothem, while barium (DBa(Ar) ~ 1) appears to be a less reliable indicator, although it may also decrease54. Research in calcite stalagmites further suggests that transition metals such as Zn2+, Ni2+, Cu2+, and Co2+ increase with lower drip rates due to the dominance of ligand-bonded metal dissociation over prior calcite precipitation (PCP)55,56. We investigate preliminary transition metal relationships in the DHAR-1 data to understand whether such drip rate effects may also relate to ISM rainfall intensity or wintertime PAP in this aragonite speleothem.

Results and discussion

Stalagmite age model and mineralogy

The age model for the multi-proxy record from stalagmite DHAR-1 is based on twelve U-series dates (Supplementary Table S2) that span 4.2–2.55 ka BP with an average 2σ error of ±18 years. Depth-age modeling was completed for each proxy using the COPRA routine57. Low growth rates of 90 µm/year in the older section (4.2–3.1 ka BP) were followed by higher deposition rates of 200 µm/year for 3.1–2.55 ka BP (Supplementary Fig. S4). The higher growth rate after 3.1 ka BP mainly reflects an increase in porosity associated with changes in trace element concentration and crystal growth fabric. The simultaneous shift in growth axis at 3.1 ka BP suggests a change in drip flow path, related to a climatic or tectonic trigger (Supplementary Discussion), and we therefore focus our climatic interpretation on the earlier undisturbed interval between 4.2 and 3.1 ka BP (all data are available in the online repository).

At the base of DHAR-1, X-ray diffraction analysis confirms a mineralogical transition from a 3-mm basal calcite layer to primary aragonite for the rest of the sample over 4.2–2.55 ka BP. Trace element data corroborates this shift with distinct changes of several element/calcium ratios, where Mg/Ca is higher in calcite compared to aragonite and Sr/Ca, U/Ca, and Ba/Ca are all lower in calcite than aragonite (Fig. 2a). This calcite-to-aragonite transition in itself supports a shift towards drier conditions because aragonite is preferentially deposited with more extensive evaporation and PCP or PAP58,59,60. The darker, dirty layer between the calcite-aragonite transition marks a hiatus between the two deposition phases (Supplementary Fig. S5).

Fig. 2: DHAR-1 proxies from 4.2 to 3.1 ka BP.
figure 2

a 4.27–3.1 ka BP DHAR-1 record with high-resolution δ18O, δ44Ca, and δ13C stable isotope time series (note reverse axes) and Ba/Ca, Sr/Ca, U/Ca, Ni/Ca, Zn/Ca, and Cu/Ca trace element data shown as 20-year LOESS smoothing curves. Dominant processes for the proxies are indicated with arrows on the right (Indian Summer Monsoon, prior aragonite precipitation, and drip rate). δ13C is not specified because it is affected by a (variable) combination of prior aragonite precipitation, cave ventilation, and long-term soil carbon changes driven by vegetation above the cave. b 4.27–3.85 ka BP in more detail covering the 4.2 ka event, with δ18O and δ44Ca stable isotope time series, and U/Ca and Ni/Ca with 10-year LOESS smoothing curves. The asterisk represents the transition from calcite to aragonite after a hiatus. For each series, shaded envelopes represent the 2.5 and 97.5% proxy confidence intervals, and horizontal dashed lines show the mean value over 4.2–3.1 ka BP. Vertical shaded bars denote relatively dry (yellow) and wet (blue) intervals between 4.27 and 3.85 ka BP. U-series dates with ±2σ error bars are shown at the bottom of the graphs above the lower age axes.

Full size image

Oxygen, carbon, and calcium stable isotopes

A profile of 750 oxygen and carbon stable isotope samples was resolved at 100–300 µm (annually) over the entire sample 4.2 to 2.55 ka BP (GFZ Potsdam), and extended with 876 samples milled at 50 µm (sub-annual) resolution between 4.2 and 3.6 ka BP at the University of Cambridge (Fig. 2). Laser ablation element data was obtained at c. 25 µm resolution at the University of Waikato (Supplementary Fig. S6). Both the low and high resolution DHAR-1 isotope series show excellent agreement in the region of overlap. Furthermore, DHAR-1 replicates the lower resolved and less-well dated DH-1 record from the same cave61 (Supplementary Fig. S7).

The majority of the δ44Castal measurements (49 out of 61 samples) focused on the period of greatest interest from 4.2 to 3.6 ka BP, yielding a nearly decadal resolution (1 sample per 12.6 years). Values during this period ranged from −1.06 to −0.27‰ (mean = −0.84‰). The prominent positive excursion in the δ44Castal results closely mirror both δ18Ostal and δ13Cstal between 4.2 and 3.97 ka BP (Fig. 2b), a strong indication for a lengthy, multi-season period of lower rainfall. The data from the following period between 3.97 and c. 3.4 ka BP shows more negative values in all proxies, though with divergent trends: δ18Ostal trends negative, δ44Castal has no discernible trend, and δ13Cstal trends positive. This pattern suggests a less dominant PAP forcing for δ44Ca and δ13C after crossing a moisture threshold (e.g., an extended ISM season could ameliorate PAP processes)14,48. The positive trend in δ13Cstal may relate to strengthening winter cave ventilation driven by cooling temperatures outside the cave and related CO2 dynamics, decreasing cave drip rates and thus enhanced CO2 degassing in winter (summer cannot be a candidate because δ18Ostal shows the ISM rainfall increased), and/or a long-term shift to more winter-drought-resistant C4-type vegetation above the cave, e.g., refs. 43,44,45,46 (see Supplementary Discussion for additional δ13C analysis).

Trace elements tracking prior aragonite precipitation, drip rate, and redox conditions

The trace element profiles (seasonal resolution at 25 µm) add further evidence to our interpretation of the DHAR-1 proxies. A notable grouping of U2+, Sr2+, and Ba2+ in PC2 of a Principal Component Analysis (PCA) on the trace element profile over 4.2–3.1 ka BP (Fig. 3), and their positive correlation (Supplementary Fig. S8), suggest that PAP influences the concentration of these elements. In particular, U2+ and Sr2+ consistently follow the other PAP proxy, δ44Ca (Fig. 2a). The cluster of transition metals Zn2+, Ni2+, and Cu2+ in PC3 (Fig. 3) highlights elements that may respond to drip-rate changes in aragonite stalagmites55 (see Supplementary Discussion for additional transition metal discussion). Indeed, the transition metals all indicate lower drip rates during the period of less rainfall from 4.2–3.97 ka BP. By extension, low drip rates also occur around 3.6 ka BP and again after 3.3 ka BP.

Fig. 3: Principal component analysis from 4.2-3.1 ka BP.
figure 3

a Principal components 2 and 3 of the LA-ICP-MS dataset for the aragonite-only bottom half of DHAR-1B (4.2–3.1 ka BP) with a 95% confidence ellipse (black circle), showing distinct groups of the drip-rate-sensitive transition metals Ni, Zn, and Cu, the PAP-sensitive trace elements Ba, Sr, and U, and the redox-related trace elements of Fe, S, Cr, and Mn. b Loadings for PC2 (12.4% of variance) and PC3 (9% of variance). Statistics were calculated and the figure was generated using PAST 4.10 software87. Note that PC1 (28.8% of variance) is not shown—the loadings indicate a dominance of U/Ca over other elemental ratios.

Full size image

Furthermore, the redox-sensitive ions of S2+, Mn2+, Fe2+, and Cr2+, together with Mg2+ and Si2+, can be separated from the other two groups in PC3 (Fig. 3). While higher Mg and Si likely reflect increased input of detrital material or enhanced weathering of fresh bedrock material, the grouping of redox-sensitive elements suggests that the soil above the epikarst may be sensitive to saturated conditions that enable a series of redox processes62. Notably, the S/Ca record deviates after 3.72 ka BP (marking a period of higher and more variable sulfur content), and occasionally aligns with the PAP (U/Ca, Sr/Ca) and drip rate (Zn/Ca, Ni/Ca) proxies (Supplementary Figs. S6 and S9). This dynamic suggests that another environmental factor emerges after 3.72 ka BP—possibly related to redox conditions (e.g., waterlogging) in the overlying soils. Concurrent peaks in the redox-sensitive elements (Supplementary Fig. S6) are thus interpreted as periods when a more saturated drip flow path promoted water retention—suggestive of a hydrological (and perhaps ecological) shift associated with increased soil moisture and change in the seasonal timing of precipitation.

Multi-proxy interpretation

Correlations between individual proxies (δ44Ca, δ13C, δ18O, Sr/Ca, S/Ca, Ba/Ca, Zn/Ca, and Ni/Ca) were examined for three discrete time periods over 4.14–3.61 ka BP (Supplementary Fig. S9). This analysis highlights incongruous relationships between several pairs of proxies after 3.94 ka BP or 3.72 ka BP (e.g., δ13C with δ18O, Sr/Ca, and U/Ca; δ18O with Zn/Ca and Ni/Ca; Sr/Ca with Zn/Ca and Ni/Ca), suggesting that certain proxies display threshold behavior (e.g., PAP occurs only during extreme drought). A nuanced interpretation may be appropriate in some periods; for example, c. 3.6 and 3.3 ka BP when trace elements indicate a decreased drip rate but also less PAP, which could result from a lengthened ISM season that distributes the same amount of moisture over a longer wet season. After 3.4 ka BP, the δ13C and drip rate proxies (Cu/Ca, Zn/Ca, and Ni/Ca) consistently indicate wetter and/or warming conditions (with δ13C suggesting either vegetation that thrives in wetter conditions, less CO2 degassing due to wet conditions, or less ventilation from warming), and this appears related to winter precipitation because the ISM proxies show a decrease in summer precipitation. Drip rate could reflect both ISM and IWM precipitation, and thereby requires reliable ISM or IWM proxies to distinguish a seasonal interpretation. In contrast, the correlation between δ13C and the drip rate proxies breaks down from 3.7 to 3.4 ka BP. Given that δ13C is influenced by a number of factors30, it is plausible that ventilation or CO2 degassing could be the primary influence on δ13C during some periods, while gradual shifts in soil carbon above the cave could influence the overall (centennial-scale) trends. For example, if winter precipitation decreased over decades to centuries, it is possible that winter- and spring-drought sensitive trees43,44,45,46 would be increasingly replaced by C4-type plants, leading to higher δ13C (as seen between 3.9 and 3.4 ka BP).

Of the entire DHAR-1 record, the 230-year period after 4.2 ka BP stands out for its above-average δ18O, δ44Ca, δ13C, Ni/Ca, Zn/Ca, and Cu/Ca, while Ba/Ca, Sr/Ca, and U/Ca values are all below-average (Fig. 2a). This agreement across all proxies reveals three distinct dry periods lasting 25–90 years each, which correspond to the general timing of the ‘4.2 ka event’ that is associated with the mid-late Holocene transition (Fig. 2b). After 3.97 ka BP, some divergent trends emerge in ISM (δ18O), PAP (δ44Ca, δ13C, U/Ca, Sr/Ca) and drip rate (Zn/Ca and Ni/Ca) proxies, suggesting that the drought threshold was not consistently passed in both seasons and other environmental factors assumed a more prominent role. While ISM rainfall increases after 3.97 ka BP, PAP proxies indicate a more muted recovery (or simply stabilization) of annual precipitation, with drip rate recovering until a decrease around 3.6 ka BP, and δ13C trends point to cooling (enhanced ventilation) and/or drying conditions in wintertime (either via more degassing from lower drip rate, or a shift in soil carbon). Such a shift in seasonality towards more pronounced rainfall contrast between summer and winter may have contributed to long term changes in vegetation composition (particularly away from winter-drought-sensitive C3-type trees to more drought-tolerant C4-type grasses) that ultimately altered the water retention and drainage properties of the soil above the cave (redox conditions indicated by S/Ca, Fe/Ca, Mn/Ca, Cr/Ca). We note that the δ13C proxy is susceptible to multiple influences, while the Ni/Ca, Zn/Ca, and Cu/Ca seem to be reliable proxies for drip rate, even though they are novel and untested in aragonite speleothems. Keeping this caveat in mind, we cautiously interpret the multi-proxy record as an increase in seasonality with stronger ISM seasons and longer or drier winters between 3.97 and 3.4 ka BP, followed by decreased seasonality as the ISM weakens and winter moisture picks up again from 3.4 to 3.1 ka BP.

Since Dharamjali Cave sits in a transition zone influenced by winter westerlies and the ISM, it is likely that multidecadal shifts in Northern Hemisphere insolation, the Atlantic Meridional Oscillation (AMO) and North Atlantic Oscillation (NAO), as well as the Pacific Decadal Oscillation (PDO) and El Niño-Southern Oscillation (ENSO) would play a role in modulating the strength of the dry and wet season rainfall systems and total annual rainfall over Dharamjali Cave, as it does in northeast India63. Along with negative NAO conditions, atmospheric blocking patterns over Europe can translate to drier westerlies in the Mediterranean and western Asia, which is one mechanism proposed for the 4.2 ka event1,64,65. Furthermore, warm PDO or ENSO phases have been linked to a weaker ISM with less rainfall in northeast India14,66,67.

4.2 ka event

DHAR-1 provides vital insight into the seasonality of the 4.2 ka event and the millennium thereafter. The unambiguous agreement between all moisture proxies (ISM, PAP, drip rate) from 4.2 to 3.97 ka BP provides a convincing set of evidence for a multi-season drought that overprinted all other environmental forcings in the DHAR-1 cave system. Although we lack early-Holocene stalagmite records older than 4.2 ka BP from Dharamjali Cave, we are still able to put the DHAR-1 reconstruction into perspective with post-4.2 ka BP climate conditions. Our multi-proxy time series allows us to characterize seasonality changes during the 4.2 ka event (dry in both seasons), the duration of the dry periods (25–90 years), the timing of ISM recovery (after 3.97 ka BP), and environmental and seasonality changes after 3.97 ka BP.

We compare DHAR-1 with nearby records to provide a more (pan)regional view of mid-Holocene climate change (see locations of studies from Fig. 4 in Supplementary Fig. S1). A highly resolved and well dated stalagmite ML.1 from Mawmluh Cave shows a decrease in rainfall closer to 4.0 ka BP that does not recover over the next centuries27 (Fig. 4a), whereas another stalagmite record (KM-A) from Mawmluh suggests lower rainfall at 4.1 ka BP and subsequent recovery by 3.9 ka BP25. The discrepancy between both Mawmluh records may be due to dating and dissolution issues in the KM-A record27, thus we refer to the more recent and well-dated ML.1 reconstruction from this cave. The discrepancy in timing of drought between the similarly well-dated and highly-resolved ML.1 and DHAR-1 records is most likely due to regional variation in ISM behavior. The distance between Mawmluh and Dharamjali caves is approximately 1200 km (east-west), where Mawmluh cave sits much closer to the Bay of Bengal moisture source for the ISM and is completely removed from the westerlies/IWM influence in winter. Therefore, it would be reasonable to expect a reduction in ISM rainfall to affect more distant locations like Dharamjali more severely. We also cannot exclude the possibility that changes to the Arabian Sea branch of the ISM might have decreased ISM moisture in the western regions of India (and thereby primarily impact the Dharamjali record), while the Bay of Bengal branch may have been impacted differently32,68.

Fig. 4: Comparison of northwest South Asian records over the 4.2 ka event.
figure 4

a Indian Mawmluh stalagmite ML.1 δ18O27, b Arabian Sea marine core 63KA δ18O of G. ruber 400–500 μm69, c Red Sea marine core GeoB 5836-2 δ18O of G. ruber70, d Indian stalagmite DHAR-1 δ18O (this study), e DHAR-1 δ13C (this study), f Iranian Gol-e-Zard stalagmite GZ-14-1 δ18O23, g marine core 63KA Δδ18O of N. dutertrei—G. sacculifer69, h Indian Sahiya stalagmite SAH-2 δ18O28, i Gulf of Oman marine core M5-422 CaCO371, j Kotla Dahar lake δ18O19, and k Early Harappan (EH, c. 5.0–4.6 ka BP), Mature Harappan (MH, c. 4.6–3.9 ka BP), and Late Harappan (LH, c. 3.9–3.6 ka BP) periods, shown with interpreted favorability for summer and winter crop cultivation based on the climate records in the figure. Dates are shown above each record with ±2σ error bars.

Full size image

Similar to Dharamjali, the ISM record of marine core 63KA from the northeast Arabian Sea shows a double-peaked increase in δ18O values centered around 4.1 and 4.0 ka BP that coincides with reduced Indus River outflow, which is followed by a recovery after 3.9 ka BP18,69 (Fig. 4b). In the Red Sea, high-salinity conditions are observed over the same period70 (Fig. 4c). The 4.2 ka event is typically characterized as an undifferentiated multi-century-scale drought, but the DHAR-1 record provides considerably finer-grained detail over the 4.2 ka event and reveals three major phases of lower rainfall peaking at 4.19, 4.11, and 4.02 ka BP, each lasting 25–90 years and separated by 20–30-year-long recovery phases (Figs. 2b and 4d, e).

Notably, relatively wet winter conditions are apparent for several centuries preceding the 4.2 ka event, as demonstrated by the Gol-e-Zard speleothem record from Iran23 (Fig. 4f) as well as enhanced upper ocean mixing and more evaporative, windy winter conditions in the NE Arabian Sea inferred from the 63KA marine record69 (Fig. 4g). During the mid-late Holocene transition, upper ocean mixing was subdued69 and regionally arid and dusty conditions affected Western Asia for 290 years after 4.26 ka BP23.

In contrast, the Sahiya Cave speleothem record (SAH-2), located c. 250 km further west than Dharamjali Cave, shows relatively low δ18O values between 4.2 and 3.5 ka BP that would indicate a stronger ISM or IWM28. However, a prominent growth rate minimum (2–3 µm/year) in the SAH-2 record during this interval suggests that a reduced water supply to the stalagmite may have affected its growth and signal a depositional hiatus, which could mask a drought and explain the discrepancy with the DHAR-1 record (Fig. 4h). Consequently, additional records with multiple proxies from this cave or nearby sites would be useful additions to clarify its interpretation.

Overall, the dry winter and summer conditions between c. 4.2 ka BP and 3.9 ka BP resulted in aeolian dust spikes in the Arabian Sea71 (Fig. 4i) and drying of lakes in continental India, e.g., refs. 19,20 (Fig. 4j). While ISM strength was already slowly deteriorating prior to the 4.2 ka event due to decreasing Northern Hemisphere summer insolation, e.g., refs. 4,69, the shift from exceptionally wet to markedly dry winter conditions would be most perceptible in regions such as the Indus River Basin that receive a high proportion of winter rain.

Cultural implications

Importantly, DHAR-1’s precise age model allows us to sub-divide the 4.2 ka event into separate severe arid phases within a 230-year drier-than-normal period. Records of the 4.2 ka event often portray it as a single mega-drought that lasted around 100–200 years. The high resolution of the DHAR-1 record advances our understanding by revealing at least three major dry periods within this period lasting 25–90 years each. Since the 4.2 ka event is significant in part because of its impact on large, complex Bronze Age civilizations, DHAR-1’s level of temporal resolution is applicable to the human decision-making timescale. While farmers and traders may be able to temporarily adjust practices in the face of a multi-year drought, a severe multi-decadal dry period affecting several generations of people would prompt more far-reaching and permanent adaptations or even population movement—particularly after the peak of the final and longest 90-year drought by 4.02 ka BP.

Paleoclimate data suggest that the Early Harappan phase (c. 5.0–4.6 ka BP) and the first half of the Mature Harappan phase (c. 4.6–3.9 ka BP) of the Indus Civilization were accompanied by relatively strong winter westerlies and associated IWM precipitation23,69, but concurrently declining ISM precipitation69,72,73,74. After 4.2 ka BP, Harappa (an urban center within the winter-rain dominated region) began to show signs of decline such as disease outbreaks and deteriorating urban systems8,9. Based on the DHAR-1 record, the three major phases of lower rainfall after 4.2 ka BP each lasted >25 years over a c. 230-year period, and would have had long-term environmental impacts on daily and year-round access to water, predictability of rainfall, and the extent, timing, and recurrence of river flooding. Moreover, the adverse effects of droughts on rainfed and floodplain agriculture would have been amplified if both rainfall seasons weakened or failed entirely. These periods of lower rainfall are particularly long in human timescales, and would have impacted multiple generations of individual populations and influenced their subsistence practices. The diversity of crops and farming practices of some populations of the Indus Civilization made them more resilient to such changes10,11,75.

Populations of the Indus Civilization were already adapted to cope with unpredictable climate conditions, and had the capacity to utilize a range of agricultural strategies involving both summer and winter crops10,75. However, through an extended, multi-generational drought, such adaptation strategies would have become necessary measures of last-resort, leading to reduced surpluses, decreased margins of error, and elevated vulnerability to environmental hazards and rapid hydrological changes7,11,76. Such changes are reflected in a reduction in craft activities and innovation in Indus urban centers, the decline of long-distance exchange and trade, and a trend towards deurbanization and the proliferation of smaller and more flexible rural settlements8,11. The DHAR-1 record suggests that both summer and winter crops would have become increasingly challenging to grow after 4.2 ka BP, though an ISM recovery by 3.7 ka BP would have favored summer crops in ISM-dominated regions (Fig. 4k). Aridity-adapted crops like winter barley and summer millet would have been the most successful under these changed circumstances, while crop diversity (including rice, which continued being used in the Late Harappan and post-Indus period) would have helped mitigate risk. The likelihood that cropping strategies were designed to mitigate risk is supported by the available archaeobotanical evidence7,8,11,75,77,78. The observed decreasing reliance on winter wheat after 4.2 ka BP and increasing presence of more drought-tolerant summer crops benefited smaller communities that were self-reliant11,79,80, and perhaps even encouraged pastoralism8,81. This socio-economic transformation was combined with a spatial displacement of population towards settlements in the ISM-dominated northeastern and southeastern Indus regions, which also offer a higher total annual rainfall10. In this respect, the stressors of a climatic shift with associated environmental changes over a multi-generational period would have led to sustainability through tactical and strategic subsistence choices, as well as adaptation through movement away from cities into new parts of the rural hinterland. The side effect of such strategic choices, however, was a transition away from an urban way of life that had seen the floruit of new technologies.

As the Late Harappan period transitioned into the Painted Grey Ware period after 3.5 ka BP, the DHAR-1 record suggests a state of increased seasonality between 3.97 and 3.4 ka BP, where summer crops may have been favored over winter crops (Fig. 4k), but there is little archaeobotanical data available for this period as yet. From 3.4 to 3.1 ka BP, the trend may have reversed to a state of decreased seasonality with a weaker ISM and warmer/wetter winters. It was during this period that larger settlements began to appear at Charsadda, Taxila, and in the Bannu region, which all lie along the western edge of the Indus River Basin82. Larger settlements would develop in a range of locations in the Ganges River Basin after 3.0 ka BP. The nature of adaptations in these later periods remain to be explored in detail, but together with the example of the Indus Civilization, they provide informative analogs for the modern situation where we again face a changing climate, albeit one that is largely being impacted by human actions.

Conclusions

The DHAR-1 paleoclimate record indicates that the 4.2 ka event consisted of three distinct dry phases that involved a decrease in both the Indian Summer Monsoon (ISM) and Indian Winter Monsoon (IWM). Prior studies have focused exclusively on either ISM or winter rainfall depending on their location, so additional multi-proxy studies from the overlapping region (particularly the western region, covering 5.0–3.0 ka BP) are critically needed to better understand the interaction of these two rainfall systems. The impact of the 4.2 ka event is especially detectable in northwest South Asia because the ISM and IWM domains overlap in this region, and both seasons were affected by decreased rainfall during the mid-late Holocene transition. The role of winter aridity is particularly noteworthy as it enhances and prolongs the driest growing season. The DHAR-1 record shows a 230-year period from 4.2 to 3.97 ka BP where all moisture proxies (ISM, PAP, and drip rate) unambiguously indicate aridity, punctuated by three 25–90 year-long phases of lower rainfall that lasted long enough to affect multiple generations of individual populations and their subsistence strategies. After 3.97 ka BP, diverging trends in these proxies suggest yearlong aridity was replaced by more complex shifts in rainfall seasonality and environmental conditions—including a recovery of the ISM by 3.7 ka BP alongside drier and/or cooler winters that may have also promoted a long-term ecological shift towards more drought-resistant grassy vegetation. Eventually, this may have increased waterlogging of soil above the cave to the point of altering local redox conditions. Shifting seasonality of precipitation appears to be a key factor influencing the populations in the Indus River Basin over the mid-late Holocene, encouraging cropping adaptations and shifting population centers based on the availability of food and water throughout the year.

Methods

Dating and mineralogy

U-series dating was performed at Caltech on 22 samples (Supplementary Table S2). No date has yet been analyzed from the calcite segment older than 4.14 ka BP due to a lack of specialized equipment needed for such small and very likely Th-rich samples. Twelve U-series ages (between 2.55 and 4.14 ka BP) were used to construct the final age model (Supplementary Fig. S4), which was built using ensembles of 2000 Monte Carlo simulations for each proxy using the MATLAB-based COPRA routine that explicitly considers individual proxy uncertainties57. The Piecewise Cubic Hermite Interpolating Polynomial (PCHIP) interpolation method was used for all proxies. COPRA output has the distinct advantage of showing both the age uncertainties as well as confidence envelopes for all the proxy time series. X-ray diffraction (XRD) was used to check for mineralogical changes of aragonite v. calcite. Four 20 mg samples were milled from DHAR-1A, mixed with ethanol and evenly smeared onto a glass plate. Samples were loaded onto a Bruker D8 XRD instrument equipped with a MoKα source Lynxeye XE-T PSD detector. Measurements spanned 0–40° angle at 0.037° steps, 206.5 s per step. Phases were identified using the PDF2 (Powder Diffraction File) database in Eva V10.0 software.

Stable isotopes

Using the DHAR-1A half of the speleothem, 750 samples were milled for stable isotope analysis (δ18O and δ13C) at 100–300 µm resolution and analyzed at GFZ Potsdam. Further high-resolution stable isotope analysis at the University of Cambridge included 876 samples from the bottom 4 cm of the mirroring slab DHAR-1B, covering c. 4.2–3.6 ka BP. In the high-resolution series, 74 samples were taken from the basal section of DHAR-1A because this portion of the speleothem is better represented and preserved on the DHAR-1A slab. As the mirror image of DHAR-1A, slab DHAR-1B is minimally offset (Supplementary Fig. S5). The curvature of the speleothem and the high-resolution nature of the stable isotope dataset from DHAR-1B (50 µm) renders some inevitable offsets, and requires adjustment prior to comparison of the high-resolution and lower-resolution profiles. A master depth scale was created based on the LA-ICP-MS transects from slab DHAR-1B.

A Sherline micromill with a Ø 1 mm drill bit was used for stable isotope sub-sampling. Sampling lines were selected near the central drip point of the speleothem, where the laminations showed minimal curvature. To eliminate as much cross-sample integration error as possible during the high-resolution milling process, we first milled a 1 mm deep trench along the sampling lines, as well as a parallel trench 5 mm away from the sample line following established procedures29 (Supplementary Fig. S5). All sampling equipment was cleaned with ethanol before each sample. An air duster was used to remove residual dust. This trenching process ultimately resulted in a 4-mm-wide section between the trenches that was sampled at 50 µm resolution, at 0.75 mm depth.

For each sample, c. 200 µg of material was milled, of which c. 100–200 µg were sealed in a Borosilicate glass exetainer vial with a silicone rubber septum, and loaded onto the Thermo Gasbench autosampler in batches of 40 samples. Each batch of samples included 10 reference carbonates of the in-house standard Carrara Z (calibrated to VPDB using the international standard NBS 19) and 2 control samples of Fletton Clay. Samples and standards were first flushed with helium and then acidified with 104% orthophosphoric acid for 1 h at 70 °C, and finally analyzed with a Thermo Delta V mass spectrometer in continuous flow mode. Precision of Carrara Z was ±0.06‰ (1σ) for δ18O and δ13C.

To merge the isotope data from the aragonitic and calcitic parts of the stalagmite, a +1.16‰ carbon isotope correction and a +0.81‰ oxygen isotope correction83 was applied to the measurements from the basal section of speleothem (247–250.3 mm) that consists primarily of calcite instead of aragonite. The original, uncorrected calcite data are reported in data files, but the corrected-to-aragonite data are plotted in all figures.

The δ44/40Ca measurements were made on 60 samples of aragonite and 1 sample of calcite milled along the stalagmite growth axis between 4.2–2.8 ka BP. Of these samples, 23 were initially milled at point locations on the DHAR-1A slab, followed up by 38 high-resolution measurements on the DHAR-1B slab focused on the period 4.2–3.6 ka BP (achieving a resolution of one sample per 12.6 years in this period). In addition, cave host rock (−0.36‰), drip water (−0.43‰), and modern carbonate samples (mean of −0.78‰) were measured. The δ44/40Ca measurements were made on a ThermoFisher Scientific Triton Plus Multicollector Thermal Ionization Mass Spectrometer (MC-TIMS) at the University of Cambridge following established methods84. Carbonate samples were dissolved in 2% nitric acid, and a double spike of 42Ca and 48Ca (1:1) was added at a 10:1 sample to spike ratio. A dose of c. 4 µg of Ca was loaded onto double rhenium filaments and activated with phosphoric acid. The NIST 915 A or 915B standard was measured about every 10 samples, and yielded a 2σ error of 0.1‰ for 11 total measurements.

Trace element analysis

The elemental composition of DHAR-1B was determined using laser ablation inductively-coupled plasma mass spectrometry (LA-ICP-MS). Elemental data was measured at the University of Waikato under supervision of Dr. Amanda French. The speleothem was ablated using a RESOlution SE series 193 nm excimer laser ablation system. Helium was used as the carrier gas to move the aerosol to an Agilent 8900 Triple-Quad ICP-MS. DHAR-1B was ablated using a 50 μm diameter laser spot size, which traversed the speleothem parallel to the growth axis at 24.3 μm/second to achieve a final spatial resolution of c. 25 μm. Before the measurement, each line was traversed by a rapid 100 μm spot size laser ablation cleaning sweep to remove potential contaminants. For the base of slab DHAR-1B, sampling was done parallel to the sampling tracks generated for the stable isotope milling, and beyond this the laser ablation analysis was continued along the growth axis in 8 segments (Supplementary Fig. S5). The glass standards NIST610 and NIST612 bracketed measurement transects at least every 15 min to correct for instrumental drift. Raw data was processed using the IOLITE data-processing software85, and trace element/Ca mass ratios were calculated using Ca as an internal standard assuming stoichiometry (40% Ca in CaCO3).

The LA-ICP-MS data was further post-processed in MATLAB to remove any obvious outliers by identifying points ±4σ away from a 5-point running mean of the dataset. The software program PAST 4.10 (Hammer et al., 2001) was used for the principal component analysis (Fig. 3). We used 16 of the element ratios over the time interval 4.2–3.1 ka BP in the aragonite portion of the speleothem (Ba/Ca, Co/Ca, Cr/Ca, Cu/Ca, Fe/Ca, K/Ca, Mg/Ca, Mn/Ca, Ni/Ca, Pb/Ca, S/Ca, Si/Ca, Sr/Ca, Ti/Ca, U/Ca, and Zn/Ca). The correlation matrix was used to compare standardized variables, and a Kaiser–Meyer–Olkin value of 0.84 indicates that the dataset and sampling resolution is well-suited for principal component analysis. PC1 explains 28.8% of the variance, highlighting the unique properties of U/Ca compared to the rest of the elements (a strong effect of PAP). PC2 explains 12.4% of the total variance (highlighting the PAP v. transition metal element clusters), and PC3 explains 9% of the variance (distinguishing the detrital/weathering/redox indicators).