Introduction

The geological carbon cycle is characterized by carbon fluxes between atmospheric, biospheric and lithospheric reservoirs that mediate atmospheric CO2 and global climate over millennial to million-year timescales1,2. In canonical views of this carbon cycle, weathering of silicate minerals freshly exposed at Earth’s surface by uplift and erosion of mountains draws down CO2 from the atmosphere, balancing the emissions from Solid-Earth degassing, and keeping global climate in a habitable state2,3. However, by quantifying the role of other weathering reactions, a growing body of work suggests weathering may be a net source of CO2 to the atmosphere because the net drawdown of CO2 from silicate weathering (47–72 Mt C yr-1 globally after accounting for carbonate mineral formation)4 is offset or even exceeded by CO2 emissions from oxidative weathering of petrogenic organic carbon (OCpetro; 62–86 Mt C yr−1)5, and sulfide minerals (30–40 Mt C yr−1)6. Therefore, to maintain a habitable climate over the Phanerozoic, a large additional sink of CO2 is required, which re-emphasises the role of organic carbon burial2,7.

Erosion of the continents contributes to organic carbon burial by mobilizing and transporting biospheric organic carbon (OCbio) from vegetation and soils to the ocean (107–231 Mt C yr-1 globally; henceforth referred to as ‘OCbio export’))7,8,9, and by increasing mass accumulation rates on continental margins enhancing OCbio burial efficiency10. Thus, quantifying OCbio yields from terrestrial ecosystems and identifying the processes that control them over millennial timescales is essential for understanding the global carbon cycle2. However, almost all previous estimates of OCbio export from terrestrial environments and inferences about the processes that control it have been derived from river gauging of suspended sediment (SS) and carbon yields that span years to a decade at most8,9. This is a major challenge because key carbon cycle, geomorphic and sedimentary processes play out over millennia and may have periods of adjustment to extreme events11 or climate change12 over decades to millennia, which cannot be assessed from river gauging records over timeframes capturing less than a decade.

Temporal variability in OCbio export from terrestrial ecosystems to the ocean may occur due to pulsed perturbations (e.g. rapid changes in magnitude of fluxes over timescales of years to decades) or more gradual change (e.g. secular change over timescales of 100–1000 s of years). Despite their rarity, large earthquakes and storms significantly perturb Earth surface dynamics by triggering landslides that mobilize sediment and carbon on hillslopes and increase loads in rivers13,14,15. River suspended sediment yields can increase by up to an order of magnitude and persist for years to decades after the triggering events11,13,14,15, but comparatively little is known about how earthquake- and storm-induced landsliding contributes to long-term OCbio export from terrestrial ecosystems due to the brevity of most records16,17,18,19. In addition, secular changes in climate (e.g. late Holocene climate change) could impact runoff and associated OCbio export9, yet we lack assessments of how yields have changed over these longer periods of time.

Relatively short records of riverine OCbio yields have also precluded direct assessment of the controls on OCbio export variability through time. The relationship between OCbio yields, SS yields (as a proxy for physical erosion) and runoff from both global8 and mountain9 rivers has been revealed using spatial correlations between these variables. These correlations suggest the rate of OCbio export scales positively with physical erosion8 and runoff9 implying there may be feedbacks between OC export by rivers, tectonic processes and climatic change that could regulate net CO2 transfers through space and time2. However, implicit in this interpretation is the assumption that apparent spatial correlations between variables inform controls on temporal variability, a “space for time” substitution20,21. This assumption may not be valid, especially in cases where scatter in the scaling between OCbio yields and runoff spans an order of magnitude, potentially obscuring multiple inter-related controls. These assumptions can only be tested with records of OCbio export that span decades to millennia. It is clear that longer term records of OCbio export that span decades to millennia are required to refine our understanding of how OCbio export mediates the global carbon cycle.

Here we use an alternative approach to global datasets of river yields: we keep the physiographic setting constant while reconstructing past sediment and carbon yields from catchments using the sediment record. A high spatial density coring approach reconstructs mass accumulation in two lakes basins22, allowing SS, OCbio and OCpetro export to be quantified from two (~60 km2) range front catchments of the Southern Alps, New Zealand over the last millennium (Fig. 1). During this time there have been four Mw > 7.9 earthquakes on the range-bounding Alpine Fault (recurrence interval: 251 ± 48 (1σ) years)23, storms that have resulted in widespread slope instability24, and late Holocene changes in hydroclimate25. We also consider these findings in the context of global rates of earthquake-triggered landsliding, and in the context of models of OCbio export, which suggest that hydroclimate moderates carbon export from mountain forests. By combing observations from instrumental and natural archives that span years to a millennium, we establish links between tectonic and climatic forcing and carbon export from forests by quantifying how: 1) earthquakes and storms drive pulsed increases in sediment and OCbio export over decades; and 2) precipitation variability due to climatic change sets longer-term patterns in OC export from terrestrial ecosystems in the western range front of the Southern Alps.

Fig. 1: Tectonic settings and catchment maps for lakes Paringa and Mapourika.
figure 1

a The tectonic setting and topography of the Southern Alps. b, c The rangefront catchments of lakes Paringa and Mapourika. Coloured areas show lake catchments, and colour range shows hillslope angle thresholded to >15° to depict slopes from which most sediment is sourced26. The red line is the surface trace of the range-bounding Alpine Fault, blue polygons in the lakes represent 10 m isobaths, and red circles show the location of core sites used to reconstruct sedimentary fill volumes.

Full size image

Results and discussion

Reconstructed sediment and carbon yields

The sedimentary records of lakes Paringa and Mapourika are characterized by a repeating sequence of sedimentary phases that can be related to the Alpine Fault earthquake cycle11,24,26(Fig. 2a, d): (i) co-seismic megaturbidites formed by shaking-induced subaqueous mass wasting (henceforth termed co-seismic phase); (ii) post-seismic hyperpycnite stacks formed during periods of elevated river sediment flux in response to earthquake-induced landsliding (henceforth termed post-seismic phase); and (iii) inter-seismic layered silts formed between earthquakes when the catchments are responding to secular changes in environmental forcing (henceforth termed inter-seismic phase). Within some inter-seismic phases there were isolated hyperpyncite stacks inferred to be associated with storm-driven landsliding24. We used a dense network of ~6 m sediment cores from the lakes to establish detailed stratigraphic correlations that formed the basis for volumetric and density models of the sedimentary fill (Methods; Fig. 2; Supplementary Figs. 1–3; Supplementary Data 1).

Fig. 2: Core sedimentology, inter-core correlations and volume models for lakes Paringa and Mapourika.
figure 2

a, d Images, visual logs and dry density for the master core from the depocenter of each lake. b, e Examples of the high-resolution correlations that can be traced across the entire lake basins. Lower panels show volumetric models for lakes Paringa (c) and Mapourika (f). Accumulation over four Alpine Fault earthquake cycles is reconstructed in the Windbag Basin of Lake Paringa, while only two cycles are reconstructed in Lake Mapourika. Insets in b and c show the spatial distribution and thickness of co- (red), post- (yellow), inter-seismic (brown) and storm-induced landslide (green) accumulation over the earthquake cycle between the 1717 CE earthquake and 1950.

Full size image

Combined the volumetric and density models for Lake Paringa show that 6.7 × 106 ± 0.2 × 106 t (2 Standard Errors (SE)) of sediment accumulated over the last 986 years (965-1007 years; 95% Highest Probability Density Function (HPDF) range). 6 ± 0.4% (2SE) of the total mass was partitioned into co-seismic phases, 39 ± 3.0% (2SE) into post-seismic phases and 55 ± 3.0% (2SE) into inter-seismic phases. Equivalent models for Lake Mapourika have 6.6 × 106 ± 0.3 × 106 t (2SE) of sediment over the last 553 years (547–549 years; 95% HPDF range), 17 ± 1.0% (2SE) in co-seismic phases, 34 ± 3.0% (2SE) in post-seismic phases and 38 ± 4.0% (2SE) in inter-seismic phases. Two hyperpycnite stacks related to storm-induced landsliding within the inter-seismic phases make up the remaining volume.

Dry sediment masses combined with measurements of the proportion of organic carbon in these phases and high resolution age-depth models (Methods; Supplementary Figs. 6–8) allowed SS, OCbio and OCpetro yields to be quantified (Fig. 3a). Coseismic deposits were formed by reworking of lacustrine sediment on deltas and subaqueous slopes and do not relate directly to catchment yields through time so are omitted11. In Lake Paringa the average SS and OCbio yields during the post-seismic phases were significantly higher than the average inter-seismic yields (SS yields: χ2 = 40.1, df = 1, p < 0.01; OCbio yields: χ2 = 27.5, df = 1, p < 0.01). Inter-seismic SS and OCbio yields also varied significantly between earthquake cycles (SS yields: χ2 = 24.7, df = 3, p < 0.01; OCbio yields: χ2 = 12.4, df = 3, p < 0.01). The lowest value for inter-seismic SS yields occurred during the 1400 CE earthquake cycle, while OCbio yields were lowest during the 1150 CE cycle. Maximum inter-seismic SS and OCbio yields occurred during the earthquake cycle between the 1717 CE earthquake and 1950.

Fig. 3: Relationship between suspended sediment (SS), biospheric organic carbon, (OCbio) and petrogenic organic (OCpetro) yields between post-seismic, inter-seismic and storm-driven landsliding phases.
figure 3

a Scaling between SS, OCbio and OCpetro yields for post-seismic (squares), inter-seismic (circles) and storm-driven landsliding (triangles) phases in lakes Paringa and Mapourika. Lower panels show the temporal evolution of SS, OCbio and OCpetro yields following the 1717 CE earthquake for lakes Paringa (b) and Mapourika (c). Error bars show 95% Highest Probability Density Function ranges.

Full size image

The average post-seismic SS yields into Lake Mapourika were three times higher than those from Lake Paringa (Fig. 3a). Again, post-seismic sediment yields were significantly higher than the average inter-seismic yields (χ2 = 27.2, df = 1, p < 0.01). There were also two non-earthquake related increases in SS yields, one within each seismic cycle, most likely caused by storm-induced landsliding24. These occurred in the late 1800s and mid-1500s and had SS yields that were close to an order of magnitude higher than inter-seismic values. OCbio yield was only quantified for the earthquake cycle between the 1717 CE earthquake and 1950. For that cycle the post-seismic yield was significantly higher than the inter-seismic yield (χ2 = 8.4, df = 1, p < 0.01), but not substantially different from those recorded in the equivalent phase from Lake Paringa. OCpetro yields were substantially lower than OCbio yields in both lakes Paringa and Mapourika but recorded similar contrasts between post-seismic and inter-seismic periods.

The precision of the chronology was highest immediately after the 1717 CE earthquake in both lakes, which allowed the most precise yield estimates. Median SS yield peaked in the first year after the earthquake at 21,400 t km−2 yr−1 (516–31,000 t km−2 yr−1, henceforth all uncertainties are 95% Highest Probability Density Function (HPDF) ranges unless otherwise stated) in Lake Paringa and 11,600 t km−2 yr−1 (893–36,400 t km−2 yr−1) in Lake Mapourika (Fig. 3b, c). The SS yields then decrease to values that were within the 95% HPDF range of the inter-seismic yields in 4 yrs (0.4–10 yrs) in Lake Paringa and 8 yrs (5–16 yrs) in Lake Mapourika. Both OCbio and OCpetro yields followed similar trends to SS yields in both lakes.

Our reconstructed yields could be biased by incomplete retrieval of the sequence in some cores, erosion of previously deposited material by mass transport processes or heterogeneity in carbon and sediment mass not captured by our sampling approach. Despite these potential sources of error our reconstructed yields agree well with independent denudation rate estimates from cosmogenic nuclides and global scaling between OCbio, OCpetro and SS yields (Methods, Supplementary Fig. 9). This demonstrates that lake stratigraphic records can preserve quantitative records of yields from mountain catchments that bridge the gap between instrumental and millennial timescales.

Implications for extrapolating short-term yields over millennial timescales

Our reconstructions of sediment and carbon yields over decades to a millennium can be used to assess how these yields deviate from the long-term average, defined here as the average yield over the record duration. This informs on the suitability of extrapolating short-term yield estimates from river gauging data to timescales relevant to the geological carbon cycle (>103 year timeframes). For Lake Paringa the average SS yield over the record duration was 151 t km−2 yr−1 (138–163 t km−2 yr−1), while the OCbio yield over the same timeframe was 3.6 t km−2 yr−1 (3.1–4.2 t km−2 yr−1). For comparison, the lowest inter-seismic SS and OCbio yields were 0.6 times these long-term averages. Conversely, over the decades where the lake was responding to earthquake-induced landsliding the SS and OCbio yields were up to 2.2 and 1.7 times the long-term yield, respectively. The maximum post-seismic SS and OCbio yields were 142 and 203 times the long-term yield, respectively.

In Lake Mapourika the average SS yield was 434 t km−2 yr−1 (386–482 t km−2 yr−1) over the record duration, and the OCbio yield 4.5 t km−2 yr−1 (3.4–5.6 t km−2 yr−1) over the last 236 years (233–238 years). Here too, the lowest inter-seismic SS and OCbio yields were a factor of 0.5 and 0.7 times the long-term yields, respectively; while they were up to 4.7 and 2.1 times the long-term yields when responding to earthquake or storm-induced landsliding. The maximum post-seismic SS and OCbio yields were a factor of 27 and 16 times the average long-term yields, respectively. Our reconstructed yields clearly show that yields estimated over short timescales can substantially under- or overestimate the long-term sediment and carbon yields. We therefore recommend caution when extrapolating sediment and carbon yields calculated over years to a decade from river gauging data to longer timescales.

Our approach to reconstructing sediment and carbon yields from lake sediments could be deployed to determine how extreme events influence these yields over timeframes relevant to the carbon cycle in a broader suite of physiographic settings. Lakes have previously been used to quantify changes in sediment yields over post-glacial timescales27,28 or to anthropogenic landuse change over centuries22. However, our study shows that quantitative yield estimates can be generated for the landscape response to discrete extreme events over decadal to millennial timeframes. Lakes preserve diverse extreme events, including earthquakes29,30, volcanic eruptions31,32, fires33,34, storms32,35,36, and glacial outburst floods, as well as more secular changes in climate and land use change, providing rich avenues for better understanding how these phenomena modulate yields over annual to millennial timescales. Our approach will be most applicable where sedimentary evidence of these extreme events occurs in lakes with simple basin morphology, facilitating more accurate volumetric reconstructions, and the potential to develop high-resolution chronology (e.g. varved lake sediments37).

Extreme event forcing of SS, OCbio and OCpetro yields

Previous work has assessed the role of earthquake-triggered landsliding in the carbon cycle for single events in the modern day18,19 or by relying on changes in accumulation rate in a single core from one lake38. Our sediment and OC yields allow us to now quantify how these yields have varied over a millennium, across multiple events. In both lakes Mapourika and Paringa, earthquake and storm-driven landsliding increased yields by up two orders of magnitude over years to decades (Fig. 3). In total, these extreme events caused 42% of SS, 39% of OCbio and 41% of OCpetro export to the lake in just 20% of the record duration in Lake Paringa. The same proportions in Lake Mapourika were 54% of SS, 38% of OCbio, and 57% of OCpetro. The similarity between the two catchments demonstrates the importance of pulsed disturbance in driving temporal variation in sediment and carbon yields from the range front of the Southern Alps over century to thousand-year timescales.

These transient increases may have important implications for extrapolating sediment and OC yields derived from short-term instrumental records over longer timeframes if they are representative of mountain catchments more generally8,39,40. The transient increase in SS fluxes from extreme event-driven landsliding depends on the ratio of catchment area influenced by the disturbance to the total catchment area41. Large earthquakes and high magnitude storms can generate high landslide densities over thousands of square kilometers42,43. Hence, we speculate that extreme event-driven landsliding may play an important role in driving transient increases in SS, OCbio and OCpetro yields from catchments with areas up to ~103 km2 that have similar physiography to those studied here. Catchments with areas in the order of 102-103 km2 dominate the western flank of the Southern Alps and, more generally, the high-standing islands of Oceania44, where large earthquakes and storms are important drivers of erosion11,13,14, and short transport pathways transfer highly transient discharges directly to depocenters with little intermediate storage45. These small, high-standing, and forested catchments have OCbio yields that are, on average, four times higher than yields from other physiographic settings2, and have been estimated to produce one-third of organic carbon exports from land to the global ocean44. Therefore, our findings, though based on only two catchments, may have broader implications. They suggest SS and OCbio yields determined using river gauging data over years to a decade may underestimate long-term yields from similar catchments in Oceania, which combined are a significant source of OCbio export to the global ocean.

Our findings may be less relevant for larger continental-scale catchments. In these settings, where catchment areas exceed ~104 km2, the spatial extent of event-driven landsliding is often less than the total catchment area (e.g. ref. 18). If the ratio of disturbed catchment area to the total catchment area controls the magnitude of transient increases in yield41, then transient increases in OCbio export caused by extreme-event driven landsliding may not be as pronounced. These systems also have longer sediment and carbon residence times that may result in higher rates of OCbio oxidation as it passes through long floodplains before eventual burial in the ocean10,46. For example, the 32,000 km2 Narayani R. catchment draining the Nepalese Himalaya showed a negligible increase in SS or OCbio yields following the 2016 Mw7.8 Gorka earthquake that triggered extensive landsliding18. There, co-seismic landslides impacted the eastern part of the catchment and represented only a modest increase over annual monsoon-driven rates of landsliding47. Determining the suite of physiographic settings where extreme events drive substantial temporal variation in yields requires future research that applies our reconstruction approach in more diverse settings.

To provide a first-order estimate of the global importance of earthquake-induced landslides (EQIL) as a control on OCbio erosion, we combine an EQIL database48 with the global mean biomass and soil stocks (Methods). A global database of rainfall-induced landsliding does not exist, precluding an equivalent analysis of that trigger49. The 38 mapped events of EQIL have disturbed a total area of 5748 km2 globally between 1976 and 2019, at a rate of 133 km2 yr−1(ref. 48). A biome-based classification of vegetation and soil carbon stocks is used to estimate the OCbio mobilized by these landslides50. We estimate a total of 103 MtC (43 MtC and 60 MtC mobilized from vegetation and soil, respectively) by these events. This represents an average OCbio yield by EQIL of 2.4 MtC.yr−1. The uncertainties on this estimate are difficult to quantify because events may be missing from the EQIL database48, while temporal variations in EQIL rate are not likely to be captured by the ~50-year timeframe. Nevertheless, this flux is similar in magnitude to the annual OCbio flux from the Ganges – Brahmaputra, one of the largest sources of OCbio to the global Ocean8. EQIL may thus be a globally significant driver of OCbio erosion relevant to the sedimentary carbon cycle if this OCbio is effectively exported from hillslopes to ocean depocentres. Our analysis does not constrain how much earthquake-mobilised OCbio is oxidised as it traverses the transport pathway from hillslopes to the ocean17. Modelling approaches that capture sediment transport and transient storage17 have suggested that even with rapid organic matter degradation, the majority of earthquake mobilized OCbio can be exported from mountain catchments, yet the detailed fate of widespread-landslide mobilized sediment and organic carbon through downstream fluvial systems remains a major uncertainty10.

Climate drivers of inter-seismic SS and OCbio yields

Hydroclimatic variability has been invoked as an important driver of sediment and carbon yields in mountain rivers9 based on space for time substitution experiments from global river yield datasets. Here, the thousand-year timeframe over which the Lake Paringa record covers provides an opportunity to test this hypothesis. Indeed, we find there is a significant difference in SS and OCbio yields between at least one of the individual inter-seismic phases of the Lake Paringa record i.e. the time periods when the Lake Paringa catchment is not responding to enhanced SS and OCbio yields from earthquake-induced landsliding (Fig. 4). In Lake Mapourika, the shorter length of the OCbio yield record (a single inter-seismic period) precludes any assessment of temporal variability.

Fig. 4: Inter-seismic biospheric organic carbon (OCbio) yields compared to reconstructed hydroclimate, Southern Hemisphere Westerly Wind (SHWW) intensity, and related climate indices.
figure 4

a Inter-seismic OCbio yields, and (b) sediment yield (SS) from Lake Paringa (solid lines indicate medians; shaded regions represent 95% Highest Probability Density Function ranges; phases sharing the same letter are not significantly different based on Bonferroni-corrected pairwise chi-squared tests). c Reconstructed precipitation indices derived from pollen records25, and d fiord carbon stable isotope ratios from the Western South Island Climate Zone (WSI)51. e Reconstructed SHWW intensity over southern New Zealand52, a primary driver of positive precipitation anomalies in the WSI during the instrumental period92,93. f Variations in SHWW intensity driven by interactions between the Southern Annular Mode (SAM; black line)53 and the Southern Oscillation Index (SOI; red line)54, which together enhance SHWW intensity over southwest New Zealand when both SAM and SOI are in phase and negative55.

Full size image

Paleoclimate reconstructions from within the same regional climate zone as Lake Paringa show rainfall25,51 and its main driver Southern Hemisphere Westerly Wind (SHWW) intensity52 co-vary with OCbio and SS yields (Fig. 4). While the low number of inter-seismic yield estimates and variable temporal resolution of the paleoclimate data preclude formal time series analysis, first-order qualitative trends are clear. OCbio yields were lowest (between 2.1 ± 0.5 and 2.3 ± 0.5 tC km−2 yr−1) during the drier period of the record when SHWW intensity over the southern South Island was low. Conversely, the highest inter-seismic OCbio yields (between 4.1 ± 0.8 and 2.8 ± 0.8 tC km−2 yr−1) were found during wetter periods when SHWW intensity was higher. Precipitation and SHWW location and intensity are influenced by interaction between the Southern Annular Mode (SAM)53 and Southern Oscillation Index (SOI)54 that cause higher precipitation and SHWW intensity over southwest New Zealand when both SAM and SOI are in phase and negative55 (Fig. 4). Time intervals where these climate indices were in phase and negative correspond to periods when reconstructed sediment and OCbio yields are higher, providing further support for the link with climatic change. This suggests that precipitation moderates OCbio export over century to millennial timescales during time periods when the landscape is not responding to large magnitude earthquakes and storms. As precipitation is strongly correlated with runoff in the high precipitation regime of the Southern Alp’s western flank56, our findings also support hypotheses from space for time substitution experiments that use OCbio yield from river gauging and spatial gradients in climate to show runoff controls OCbio export9.

To further explore the inter-seismic data from Lake Paringa, we use a runoff-driven erosion model of OCbio, which has previously been used to describe the climate sensitivity of OCbio export in mountain catchments9. Informed by a shear-stress erosion model and fit to empirical data from mountain river catchments globally, river particulate OCbio concentrations ([POCbio], mg L−1) are described by a function of instantaneous runoff, R (mm day−1):

$$[{{{{\boldsymbol{POC}}}}_{{{{\bf{bio}}}}}}]=\alpha {{R}^{\gamma }}$$

where \({{{\boldsymbol{\alpha }}}}\) (units in of mg L−1 mm−1γ dayγ) is a pre-factor that is proposed to be related to the slope angle of the catchment9, and the exponent γ = 1.37 ± 0.17 (95% CI)9. To provide a first order estimate of fluxes and the controls, we assume \({{{\boldsymbol{\alpha }}}}\) = 0.052, which reflects the Lake Paringa catchment with some areas of lower slope compared to trunk valley catchments9. We use a mean annual runoff of 3848 ± 440 mm yr−1 (95% CI) calculated from the Paringa catchment rain gauge57, and corrected for the trend of increasing rainfall between 1960 and 2022 of 154 ± 114 mm (95% CI) per decade in the Western South Island Climate Zone58. The model returns an OCbio concentration of 1.3 ± 0.6 (95% CI) mgC l−1, which equates to an annual load of 201 ± 87 (95% CI) tC yr−1 when multiplied by the annual flow (l yr−1), and an OCbio yield of 5.0 ± 2.2 (95% CI) tC km−2 yr−1 when normalised by catchment area. The modelled yield agrees within uncertainty with the lake-derived estimate of 4.1 ± 0.8 tC km−2 yr−1 (Fig. 4a) for the most recent inter-seismic period (1770–1950). The slightly lower central tendency of the reconstructed yield can be explained by material not trapped by the lake, which has a trapping efficiency of 92% (Methods).

Building on the observed relationship between reconstructed OCbio yields and hydroclimate, combined with the agreement between reconstructed and modelled yields, we use the runoff-driven model to quantify the change in OCbio export predicted for future climate change. For a mid-range scenario (i.e. RCP6.0), streamflow in the western Southern Alps is projected to increase by 10–20% from the 1986–2005 average to 2080–2099 (ref. 59). Based on the parameterized runoff-erosion model used here, OCbio export could increase by 39 ± 30% (95% CI) over this timeframe (from 5.0 ± 2.2 tC km−2 yr−1 to 7.0 ± 4.5 tC km−2 yr−1). Our analysis is likely conservative because it doesn’t account for the predicted increase in severity of the largest flow events59, which will amplify OCbio erosion because it scales non-linearly with runoff9,16. These events are also more likely to cause widespread landsliding, which we have shown results in dramatic increases in OCbio yields.

Implications for the carbon cycle

The erosion of OCbio from the terrestrial biosphere and transfer into sedimentary storage acts as a net CO2 sink and O2 source over geological time8,9. Nevertheless, it still remains unclear how orogenesis, erosion and weathering contribute to the carbon cycle, and the feedbacks which link surface processes to climate change2, which is usually viewed through the lens of the silicate weathering feedback (e.g. ref. 60). In our reconstructed timeseries spanning the last millennium we see an important control on OCbio export by active tectonics. Co-seismic landsliding and subsequent sediment and OCbio export is responsible for ~40% of the OCbio and SS transfer in these steep forested catchments. However, during the inter-seismic periods that dominate total mass export from these catchments (~60% of the long-term flux in Lake Paringa), the fluxes are mediated by hydroclimatic variability. Wetter periods have higher OCbio yields, and the most recent climate change from the Little Ice Age (~1450 CE–1850 CE) coincided with a doubling of OCbio fluxes into the lake.

Our reconstructions support the hypothesis based on global scale space for time substitution experiments on modern observations that erosion of OCbio could act as a negative feedback in the Earth System9. Wetter climates associated with warmth can enhance erosion of OCbio from terrestrial ecosystems, promote its transfer into long-term sedimentary storage, and drawdown CO2 from the atmosphere (Fig. 4). The lake timeseries is consistent with an estimated ratio between changing runoff and CO2 drawdown fluxes from OCbio export and burial of ~1:1.6 (ref. 9). By comparison, a ratio of <1:0.6 (ref. 61) is thought to be typical of silicate weathering which is the other key negative feedback on geological timescales60,62. If temperature and hydroclimate are coupled at the global scale over geological timescales63, then OCbio export and burial represents a sensitive negative feedback that is not currently accounted for in Earth System models that seek to quantify the carbon cycle64. Understanding the operation of the geological carbon cycle once terrestrial ecosystems established in the Phanerozoic and erosion provided an additional negative feedback will require concerted research effort.

Methods

Study site

The Southern Alps are formed by oblique convergence between the Australian and Pacific plates at a rate of 39.7 mm yr−1 on a bearing of 245° (ref. 65). Up to 80% of the plate motion is accommodated on the range-bounding Alpine Fault66, which produces major and great earthquakes (Mw >7) with a return period of 251 ± 48 (1σ) years23. The western flank of the Southern Alps is dominated by steep slopes developed in tectonically fractured metasedimentary bedrock that support high rates of landsliding67,68,69. Hillslopes are covered in largely undisturbed temperate rainforest below a treeline at ~1000 m. High rates of net ecosystem productivity of 94 ± 11 (2SE) tC km−2 yr−1 support carbon stocks of 17,500 ± 5500 (2SE) tC km−2 in above-ground biomass and 18,000 ± 9000 (2SE) tC km−2 in soils70. Moisture derives predominantly from the Tasman Sea71 and is transported by north westerly flows that result in orographic precipitation of between 5 m yr−1 and 12 m yr−1 (ref. 72) in the present day. Paleoclimate reconstructions show late Holocene variation in hydroclimate with a pronounced precipitation minimum between 1750 CE and 1350 CE25,51. The climate and tectonic setting produce erosion rates of up to 10 mm yr-1 (refs. 73,74). Lakes Paringa and Mapourika are located ~3–5 km west of the Alpine Fault, have range front catchments with areas of ~60 km2 (Fig. 1). Erosion in these catchments occurs predominantly (~90%) on slopes above 15˚26, so contributions from alluvial fans and floodplains is minimal.

Sediment coring and analysis

The sediment coring approach aimed to provide unprecedented 4D resolution on the sediment fill in these lakes. A Mackereth corer was used to retrieve sediment cores up to 6 m long from 28 sites in Lake Paringa and 22 in Lake Mapourika75, providing a spatial coverage of approximately one core per ~500 m (Fig. 1 and Supplementary Fig. 1). Sediment cores were digitally imaged using a GEOTEK linescan camera, logged visually and X-ray computed tomography (CT) scanned using a GE BrightSpeed medical CT scanner set to 120 kV, 250 mA, pitch of 0.625 mm and a 100 cm2 window. Stratigraphic correlations between cores were achieved by mapping lithofacies stacking patterns that are easily identified in digital imagery and CT tomography and can be traced laterally throughout the lake basins due to the close core spacing24 (Fig. 2). These correlations were mapped in Corelyzer 2.1.276 and validated using radiocarbon dates to produce basin-wide fence diagrams of contacts between co-seismic, post-seismic, and inter-seismic sedimentary units (that we subsequently term phases) to inform volumetric and density modelling of the sedimentary fill (Fig. 2a, c; Supplementary Figs. 2 and 3). Higher resolution intra-phase correlations were also achieved within the 1717 CE post-seismic phase and during inter-seismic phases that contained periods of increased sediment and OC yields from storm-induced landsliding.

In Lake Paringa we focused the coring and sedimentary fill volume modelling on the Windbag Basin. The major fluvial tributaries of the lake flow into this basin and it is where the majority of terrestrial sediment accumulates in the lake11,23 (Supplementary Fig. 1). Cores from the sill that separates the Windbag Basin from the Hall Basin show minimal accumulation, which suggests that the amount of material that enters the Hall Basin from the Windbag Basin is negligible. This observation supports our decision to model volumes and yields for the Windbag basin alone, though we acknowledge that our yield estimates from Lake Paringa are conservative.

Lake core chronology

Chronology was generated for master cores located in the depocentre of each lake (PA1 and MA1) using 137Cs, radiocarbon dating and earthquake age estimates from Howarth et al.23. 137Cs was measured using gamma spectrometry at the Institute of Environmental Science and Research, Christchurch, New Zealand with a high-purity germanium well detector. Activities are reported in Bq.kg−1 as mean values with 95% confidence limits (Supplementary Data 4). Radiocarbon dates on terrestrial macrofossils (23 for Paringa and 15 for Mapourika) were treated and measured by Accelerator Mass Spectrometry (AMS) according to the methods of Baisden et al.77 at Rafter Radiocarbon Laboratory and Fink et al.78 at ANSTO AMS facility and are reported in Howarth et al.23. The radiocarbon dates were calibrated with the ShCal20 calibration curve79. Independent age estimates for each earthquake were derived from Wells et al.80 for the 1717 CE event and Howarth et al.23 for the ~1400 CE, ~1150 CE and ~925 CE earthquakes, and were used as additional temporal constraints in the age-depth models. All chronological information was integrated with core depth using the P_Sequence prior model with an event thickness constant k derived empirically (cf. Ramsey81; k = 2 for Paringa and k = 3 for Mapourika) in OxCal 4.4 to produce an age depth model for each lake81 (Supplementary Fig. 6). Depth intervals associated with rapidly deposited layers (RDL) and voids generated by gas were removed from the depth sequence to produce RDL-corrected depth that was used for age modelling purposes because the formation of RDL is near instantaneous and they do not represent the passage of time. The age models were used to derive the duration and related uncertainty of post-seismic, inter-seismic and storm-induced landslide phases and sub-phases that were then used in the production of SS, OCbio and OCpetro yields.

Lake sediment volume modelling and sediment yield estimates

The volumes of co-, post- and inter-seismic phases as well as 1717 CE post-seismic and storm-induced landslide sub-phases were modelled using lake bathymetry82,83 and down core logs in the 3D geological mapping software LeapfrogTM using radial basis function interpolation in the Offset Surface tool84 (Supplementary Figs. 4 and 5; Supplementary Movies 1 and 2). The boundaries of the model were set using bathymetric slopes of 20°, above which geophysical data revealed that minimal sediment accumulates. 3D models of sediment bulk density variation across the sedimentary fill were also generated in Leapfrog using the downcore sediment bulk density derived from the CT tomography85.

The wet mass (ϒw, kg) of all sedimentary phases and sub-phases was calculated using:

$${{{{\Upsilon }}}}_{w}=\sum\limits_{i=1}^{n}{V}_{i}{\rho }_{i}$$
(1)

where Vi (m3) is the volume of a given voxel (5 m3) in the Leapfrog volume model for a phase and ρi is the bulk (kg m−3) density of that voxel derived from the Leapfrog bulk density model.

The dry mass (ϒd, kg) of all sedimentary phases and sub-phases was then calculated using:

$${{{{\Upsilon }}}}_{d}={{{{\Upsilon }}}}_{{{{\rm{w}}}}}\left(1-\frac{W}{100}\right)$$
(2)

and W is the average weight percent water content of the phase derived using mass loss on drying at 105 °C from samples from the master cores from each lake (Supplementary Data 1). Spatial variability in water content was quantified for each phase over the 1717 CE seismic cycle in each lake and showed that the standard errors of the mean (SE) W for each phase were always less than 2.5%. Consequently, a conservative uncertainty for the W of each phase of ±5% (2SE) was adopted and propagated in quadrature through ϒd calculations. We make no correction for trapping efficiency because it is estimated to be 92% based on the method of Gill86 that uses the ratio of lake capacity to mean annual inflow.

Suspended sediment mass per unit catchment area and per unit time (SS, expressed as t km−2 yr−1) were generated using:

$${SS}={{{{\Upsilon }}}}_{{{{\rm{d}}}}}/A/t$$
(3)

where ϒd is the dry mass of the phase or sub-phase, A is the catchment area with hillslope angle above 15° (~40 km2 for Lake Paringa and ~20 km2 for Lake Mapourika), and t is the phase duration derived from the chronology (Supplementary Data 2). Chronological and ϒd uncertainties were propagated through the calculations using a Monte Carlo approach in OxCal 4.4 and reported as median values with 95% highest probability density function (HPDF) ranges unless otherwise stated.

Lake sediment geochemical analysis and OC yield estimates

The average OC content (weight %) and carbon stable isotope ratios of each depositional phase or sub-phase were quantified using samples from core PA1 in Lake Paringa and core MA6m1 from Lake Mapourika that were processed and measured by EA-IRMS according to the methods of Frith et al.38. For Lake Mapourika, we focus on the post 1717 CE section in MA6m1 (Supplementary Data 3). In addition, we have measured the radiocarbon activity of bulk organic carbon in MA6m1 (n = 10) and catchment soils (n = 6) using methods described by Wang et al.26, reported as the fraction modern, F14C (Supplementary Data 3).

Total OC (OCtotal, tC) of each phase or subphase was estimated using Eq. (3)

$${{OC}}_{{{{\rm{total}}}}}={{{{\Upsilon }}}}_{{{{\rm{d}}}}}\left(\frac{{OC} \% }{100}\right)$$
(4)

where \({{{\Upsilon }}}\)d (t) is the dry sediment mass and OC% is the weighted average organic carbon weight percent of the phase. Uncertainties in ϒd and OC% (2SE) were propagated in quadrature through the calculation of OCtotal. OCtotal from each phase was partitioned into OCbio and OCpetro fractions. For Lake Paringa, the fraction of rock-derived organic carbon (Fpetro) was quantified using measured total organic carbon (TOC) of the lake sediment samples and a binary mixing model with a bedrock end-member (0.13 ± 0.1% (2SE)) as discussed in previous work38,87. This was appropriate because of the high OC% in the lake sediments (mean = 2.64 ± 0.55% (2SE)), which preclude significant inputs of OCpetro.

In contrast, the average OC% in Lake Mapourika was lower (mean %OC = 1.38 ± 0.18% (2SE)), meaning OCpetro may contribute more to the carbon mass in the core. As such, the Fpetro for Lake Mapourika was constrained using a binary mixing model based on 14C fraction modern measurements (F14C, n = 10; Supplementary Fig. 7), where the OCpetro end member is 0 and the OCbio end member is derived from measurements of soil O horizons26. In this subset of the samples, we find that Fpetro correlates with δ13C in the lake core, reflecting the fact that the OCpetro end member has a high δ13C value, distinct from the vegetation88. As such, we use a linear regression between Fpetro and δ13C (r2 = 0.84) to apply to the wider sample set for which we have δ13C values but not F14C (Supplementary Fig. 8). Uncertainty of two times the standard error of the predicted value was propagated in quadrature through the calculation of Fperto for Lake Mapourika. For both lakes uncertainty derived from Fpetro estimates (2SE) was propagated in quadrature through the calculation of OCbio and OCpetro masses. We assume negligible lake-derived (autochthonous) OCbio because biomarker and bulk sediment stable isotope analysis indicate only minimal OCbio from aquatic biomass38.

OCbio and OCpetro yields per unit catchment area and per unit time were calculated using equations analogous to Eq. (2) (Supplementary Data 2). Uncertainties for OCbio and OCpetro masses and the chronology were propagated through calculations using Monte Carlo sampling and reported as median values with 95% highest probability density function (HPDF) ranges unless otherwise stated. Statistical differences between distributions of SS, OCbio, and OCpetro yields were assessed using chi-squared (χ²) tests of depositional phase means and standard deviations under the assumption of approximate normality76. A global chi-squared test evaluated agreement among all phases (α = 0.05), while Bonferroni-corrected pairwise tests identified specific differences (α = 0.0083).

The accuracy of reconstructed sediment and organic carbon yields

In lieu of overlapping instrumental and reconstructed timeseries the accuracy of our reconstructed yields was assessed using comparisons to independent denudation rate estimates from cosmogenic nuclides and global scaling between SS, OCbio, and OCpetro yields. 10Be-derived denudation yields from two range front catchments near Lake Mapourika with similar physiographic characteristics (morphometry, rock type, vegetation and climate) as Potters Creek provide a mean yield of 4870 ± 1340 t km−2 yr−1 (2SE) averaged over 485 ± 182 years (2SE)89. This estimate is an order of magnitude greater than the reconstructed SS yields over the same timeframe from Lake Mapourika, which is expected because the coarse sediment and dissolved loads are not captured in our lake-based reconstruction. If we take the reconstructed SS yield from Lake Mapourika, we can estimate the missing coarse sediment load component assuming that: 1) landsliding drives erosion and sediment production in the Southern Alps68; 2) 12 ± 6% (2SE) of landslide grainsize distributions are <2 mm based on landslide grainsize measurements from a similar physiographic setting (New Zealand and Taiwan)90,91, and 3) chemical denudation is an order of magnitude less than physical denudation and hence a small component of total denudation in these systems89. The total sediment yield for the Lake Mapourika catchment in this case would be 3630 ± 1880 t km−2 yr−1 (2SE), which is similar to the 10Be denudation rates. Unfortunately, there is no comparable data for validating the Lake Paringa SS yields, though the lower yields compared to Lake Mapourika are comparable to the reduction in landslide-derived erosion rates in the Southern Alps between these loctions68, and consistent with a decrease in mean annual precipitation72 moving south west.

The lake-derived estimates of SS, OCbio and OCpetro yield are also consistent with relationships derived from modern-day fluxes in global rivers8,9 (Supplementary Fig. 9). Our estimates of OCbio yields for a given SS yield are within the range of global values, though they sit above the regression line8. This is expected given the high organic carbon stocks of soils in the western Southern Alps70, which results in high OCbio yields when mobilized by landsliding88. The reconstructed OCpetro yields for a given SS yield fall closer to the global regression line than the OCbio yields, probably because rock OC% of Southern Alps bedrock is close to the average for most major rock types globally5.

Global earthquake-induced landslide mobilized OCbio estimate

We use the EQIL (“earthquake induced landslide”) database of Seal et al.48 and filter it for events where landslide area has been mapped. Since the satellite era, 38 earthquakes have been studied and landslide maps produced between 1976 and 2019. Many more EQIL events have metrics such as the total impacted area, and number of landslides, but we have focused only on those with reported landslide areas. This approach is likely to have missed events over this time interval. We recognize that the time period is short compared to the typical return interval of earthquakes and the full magnitude frequency distribution of events will not have been sampled.

To calculate the eroded mass of organic carbon from soil and vegetation (OCbio) we follow the approach of Hilton et al.70, who use average carbon stocks for soils and vegetation. Because the landslide event locations are linked to the earthquake epicenter in the EQIL database33 and that individual landslide polygons are not available, we assess the carbon stocks for an event using biome estimates of organic carbon stocks in vegetation and soils from Watson et al.50. As such, we report a total mobilized mass and do not focus on individual events which may represent over- or under- estimates based on the local soil and vegetation carbon stocks.