A conceptual framework is provided by a hypothetical earthquake on the Alpine Fault in the western Southern Alps, New Zealand (Fig. 1). The topography of the west coast of the Southern Alps is obtained from the SRTM3 digital elevation model (DEM). This location is selected due to prior work documenting the role of landsliding in sediment and OC mobilization (Hilton et al., 2008a; Hovius et al., 1997; Korup et al., 2010) and the role of large earthquakes in long-term sediment and OC records in lakes (Frith et al., 2018; Howarth et al., 2012, 2014). The Southern Alps formed as a result of oblique plate convergence between the Pacific and Australian plates at a rate of 39.7 mm yr-1 (DeMets et al., 2010). The Alpine Fault accommodates up to 80 % of the plate convergence (Norris and Cooper, 2007) in earthquakes of magnitude >7.5, with the last major earthquake occurring in 1717 CE and an average recurrence time of 263±68 years as determined from paleoseismic studies (Howarth et al., 2016, 2018). The Southern Alps form a barrier to westerly winds that leads to high precipitation rates of up to 13 m yr-1 along the west coast (Tait and Zheng, 2007). Landscapes in the Southern Alps are characterized by steep hillslopes with modal gradients of approximately 35∘ (Korup et al., 2010), and hillslope erosion is dominated by landsliding in the current aseismic period (Hovius et al., 1997).

The high rainfall sustains temperate rainforest on steep slopes at elevations ≤800 m, while shrubs, herbs and grassland persist up to and above the regional snowline at ∼1250 m (Reif and Allen, 1988). The carbon stocks of aboveground biomass in the western Southern Alps have been estimated at 17500±5500 tC km-2 based on local forest plot inventories (Hilton et al., 2011). The OC stocks of soils in the Southern Alps have been recently estimated at ∼13000±4300 tC km-2 (Harvey, 2019). This estimate was derived from measured OC concentrations in soil profiles collected from 52 sites in the mountain range. The OC stocks are estimated to the point of refusal of the soil auger profiles (average thickness 0.42±0.09 m) and comprise soil OC in the upper organic-rich horizons (mean thickness 0.10±0.01 m) and in the mineral-dominated B and C horizons beneath. The average value is slightly lower than previous estimates, which were inferred from sparse data (Frith et al., 2018; Hilton et al., 2011; Tonkin and Basher, 2001).

Populations of landslides triggered by large earthquakes are generated in the Quakos model, with an empirical relationship between the spatial patterns of the simulated peak ground acceleration (PGA) and landslide density (see Croissant et al., 2019b, for further details). The resulting landslide density map allows for the quantification of the total area of landsliding (Als,tot) and determines the spatial extent of mass wasting. The co-seismic landslides are described by several components: their geometrical properties of their scar area (Als, m2) and deposit volume (Vls, m3), their spatial distribution within the landscape, and their connectivity to the drainage network. Based on previous empirical studies (Malamud et al., 2004; Tanyaş et al., 2017), the full distribution of landslide areas can be described by a three-parameter inverse gamma function of the form 1PDFAls=1aΓaaAls-sρ+1exp⁡-aAls-s, where a is the position of the probability density function (PDF) maximum, s controls the rollover for small landslides, ρ is a positive exponent controlling the slope of the tail and Γ is the gamma function. The parameterization of Eq. (1) has been done using values obtained for a west coast landslide inventory (Hovius et al., 1997). In Quakos, a landslide population is generated by sampling the PDF until the sum of the landslide areas reaches the value of Als,tot. The landslides are spatially distributed within catchments following a landslide density map that depends on the PGA patterns triggered by the earthquake (Meunier et al., 2008). Landslide area is converted to volume with an empirical law of the form 2Vls=αAlsγ, where the empirical constants α and γ are set to 0.05 and 1.5 (Hovius et al., 1997; Larsen et al., 2010) to reflect the deep-seated bedrock landslides that are likely to be the dominant mechanism of landslide generation during large earthquakes in New Zealand (Larsen et al., 2010). Vls is a measure of the total volume of sediment mobilized by landslides. However, in the subsequent modeling we consider only the fine-grained portion of this landslide volume that can be transported as suspended load (Vls,fine), here set to 50 % of Vls, as an average value considering empirical estimations varying between 10 % and 90 % (Dadson et al., 2003). The landslide volume is distributed in a single model cell.

Each landslide is assigned a probability that it is initially connected to the fluvial network (χ), which is a function of the drainage area of landslide pixels following an empirical relationship of the form (e.g., 2008, Wenchuan – Li et al., 2016; 2015, Gorkha – Roback et al., 2018): 3χ=cAlsμ, where c and μ are empirical constants. Equation (3) is used to assign the initial connectivity status of all landslides with an area smaller than 106 m2, while larger landslides are assumed to be initially connected (Croissant et al., 2019b). Using c=0.86 and μ=0.345, only 12 % of the landslides are initially connected to the drainage network (Fig. 1). However, because the largest landslides are more likely to be initially connected, between 40 % and 60 % of the landslide-derived sediment volume is initially available. We assume that sediment derived from initially connected landslides is immediately available for fluvial export toward the catchment outlet.

To account for processes which mobilize sediment from landslide deposits that are not initially connected to the river network (including soil creep, overland flow, shallow landsliding and debris flows), which are known to act (Dietrich et al., 2003; Roering et al., 2001) but are otherwise difficult to parametrize, our approach allows landslide deposits on hillslopes to progressively reach the river network at a rate defined as a constant “connection velocity” ucon (m yr-1) (Croissant et al., 2019). The path taken by the landslides to reach the closest river is computed using a steepest descent algorithm, and their time of connection (tcon) is determined by dividing the distance to the river network by ucon. The river network is extracted from the DEM using a single flow algorithm and by setting a critical drainage area of 0.5 km2.

To quantify the erosion of soil OC, the area of each individual landslide scars is combined with the local soil organic carbon stock (Cstock, tC m-2) determined in this study to quantify the total mass of eroded soil OC (tC): 4Moc,tot=∑0NAls,nCstock,n, where N is the total number of earthquake-triggered landslides. Simulations account for the uncertainties related to the OC content of soils. As outlined at the opening of Sect. 2, the transport of the aboveground biomass is not considered here as we assume it is mobilized as large woody debris, whose transport dynamics remain poorly constrained (Wohl, 2011).

In the model framework, once a landslide deposit reaches the fluvial network, fine sediment and OC are transported at a rate set by the local transport capacity of the river (Wang et al., 2015). Here, we only consider the evacuation of the fine sediment and OC that are transported as suspended load, with a nominal grain size <1 mm (see Sect. 2). First, we consider the event-based sediment discharge of the suspended load (Qs, m3 s-1) that is commonly empirically described as a power-law function of water discharge (Q, m3 s-1): 5Qs=ksQε, where ks is a constant (s m-3) and ε is a positive exponent. Here, the water discharge is a linear function of the local drainage area and surface runoff. The constants ks and ε can be determined from measurements across a range of discharges. For our study location, we use gauging data from the Hokitika, Haast, Whataroa and Poerua rivers, and we find values of ks=300 and ε=2. To encompass the contribution of all discharge events to sediment transport, the annual long-term sediment flux is computed as the integral of the convolution between sediment flux (Eq. 5) and the probability density function (PDF) of daily discharge events (PDFQ) (DiBiase and Whipple, 2011; Lague, 2014; Lague et al., 2005): 6Qs‾=∫0QmaxQsPDFQ, where Qmax is the maximum discharge in the range. The PDF of daily discharges is described by an inverse gamma function of the form 7PDFQ=kk+1Γk+1exp⁡-kQ/Q‾Q/Q‾-2+k, where k is a positive constant expressing the variability of the discharge events. A low value of k corresponds to higher discharge variability.

For each landslide, the evolution of the mass of fine sediment remaining in the landslide deposit while fluvial transport removes mass is then described by 8dMlsrivdt=-Qs‾. Equation (8) is solved to get the temporal evolution of the fine sediment mass remaining in the deposit as 9Mls,triv=Mls,0-Qs‾t=Qs‾t0-t, where Mls,0 is the initial mass of fine sediment mobilized by landsliding and t0=Mls,0Qs‾ the time necessary to export the entirety of the fine sediment.

The mass of fine sediment exported by the river at any point in time is given by 10mls,triv=Q‾sdt, with dt being the time step over which the computation is completed.

We use a reduced-complexity approach to describe the physical processes in play, which allows us to explore the large spatial and temporal scales over which earthquake-triggered landslides impact landscapes. However, this means that the transport of fine sediment and particulate OC is subject to several assumptions. First, by considering only the transport of fine, soil-derived particles as suspended load, each landslide volume is partitioned between fine grains (that can be transported in river suspension) and coarse grains (which are transported by saltation and other bed-load processes). Second, sediment transport is treated as a detachment-limited regime, and entrained particles are never redeposited within the catchment. This assumption is likely to be reasonable for the Southern Alps, which are characterized by steep and short rivers that promote efficient suspended sediment transport (Korup et al., 2010). We also assume that the fine sediment is exported at fluvial transport capacity (Eq. 8). We do not consider in situ stabilization of landslide materials, regrowth of vegetation on those materials or subsequent remobilization by landsliding. Therefore, all landslide-derived fine grains that were initially mobilized will be eventually exported by rivers at the timescale controlled by the local transport capacity. In addition, the climatic context (mean annual runoff and runoff variability) that controls the fluvial transport capacity is assumed to be constant over the duration of the simulations. Finally, any other event that might be triggering landslides during the post-seismic phase (rainstorms, earthquakes of lower magnitude, etc.) are not considered.

Our model postulates that the fate of landslide-mobilized OC is controlled by its physical export by rivers versus its biogeochemical oxidation that can release CO2 back to the atmosphere. The long-term sink of CO2 by erosion comes about because a fraction of the mobilized OC is stored in sediments, while the vegetation and soil OC are renewed at the site of erosion (Berhe et al., 2007; Stallard, 1998). If eroded materials remain on hillslopes, the OC contained in the landslide deposits can be subjected to heterotrophic respiration and OC oxidation, either by soil fauna present in the original forest soil and translocated to the landslide deposit or by colonization by new fauna communities. When landslides are connected to the river network, the model framework considers the OC remaining in the deposit to experience oxidation and physical transport simultaneously. OC oxidation can be generally described by either discrete organic matter pools with a specific rate of degradation or by continuum models that seek to explore a range of oxidation rates (e.g., Arndt et al., 2013; Manzoni et al., 2009). In this section, we describe the different models of organic matter degradation used in our approach.

The spatial variation of the local transport rate (Qs‾), combined with a landslide volume (Vls) population, leads to the emergence of a distribution of landslide export times t0 and of distances between deposits and the fluvial network. This in turn controls the connection time tcon. Here, we explore the fate of OC as a function of the connection velocity of landslides to the river network (ucon), which also controls the distribution of tcon.

For each individual landslide generated by Quakos, we calculate the proportion of OC exported by rivers as opposed to that oxidized to CO2 as a function of ttot=tcon+t0. The general pattern that emerges is that the proportion of OC exported by rivers decreases with ttot, ranging from 100 % of OC exported by rivers for values of ttot<1 year to 0 % for ttot>300 to 10 000 years depending on the value of tcon (Fig. 5). This approach allows us to assess an envelope of possible values of the proportion of OC exported by rivers as a function of ttot, defined by two cases: (i) when tcon is large compared to the export time (red line in Fig. 5) and (ii) when the landslides are directly connected to the rivers when tcon=0 (blue line in Fig. 5). The scatter in the data emerges from the competition between the two timescales controlling the fate of the landslide-mobilized OC. A dominance of tcon promotes the oxidation (and/or storage on hillslopes) of the organic matter, while a lower value of t0 leads to a higher proportion of OC that is exported by the fluvial network. Thus, for a fixed climate (mean annual runoff of 1 m yr-1 and runoff variability k=1 in Eq. 7) the quantity of OC that is exported by the fluvial network is sensitive to the chosen connection velocity of the landslides to the rivers (Fig. 5).

We next investigate the impacts of the connection velocity. The fate of carbon is considered to be either returned to the atmosphere as CO2, retained in the landscape (in landslide deposits) or exported in river sediments (potential longer-term OC burial). Not surprisingly, the results for the single pool OC degradation model show that a lower connection velocity and a higher oxidation constant produce a larger proportion of C that has been oxidized and released in the atmosphere (Fig. 6). Interestingly, scenarios that show a significant storage of OC at the end of a seismic cycle only emerge for a low oxidation constant coupled to a slow connection velocity (Fig. 6). The multi-pool degradation model outputs do not differ significantly from the single pool results. However, predicted OC storage in the mountain landscape is larger when the low-reactivity pool is the largest (Fig. 7). For both oxidation models, high connection velocities (>100 m yr-1) systematically promote the river transport of OC, with between 80 % and 98 % of OC being exported out of the catchments. The physical meaning of the connection velocity for the case of fine sediment is discussed in Sect. 7.1.

By considering a full landslide population, we can also consider the effect of landslide size on the fate of OC. Our results suggest that intermediate landslides (i.e., with an area ranging from 103 to 105 m2) are responsible for most of the export of river carbon (Fig. 8). While smaller landslides (Als<103 m2) are removed efficiently by rivers, they mobilize a lower initial quantity of OC and are less important in the total carbon budget. In contrast, large landslides (Als>105 m2) individually mobilize large quantities of OC but have longer export times as t0 is a function of Vls,fine, potentially reducing their impact on the carbon export by fluvial transport.

In our model, the mean annual runoff and runoff variability determine the sediment transport rate (Eq. 6 that depends on Eq. 7) and the river export of OC (Hilton, 2017; Wang et al., 2019). The fate of landslide-mobilized OC is sensitive to the export time, which is a ratio between the volume of fine sediment available for transport and the sediment transport rate. Here, we assess how the fate of OC is moderated by these climatic parameters. For the mean annual runoff, we chose a range of values that span the observed values of different mountain ranges (e.g., Bookhagen and Burbank, 2010; Hicks et al., 2011). For the runoff variability, k varies between 0.5, a value implying a high recurrence of large daily runoff events typical of Taiwan (Lague et al., 2005), and 4, a value representative of temperate regions that rarely experience flooding events (Croissant et al., 2019a).

The role of runoff and runoff variability is explored for a slow landslide connection velocity (ucon= 1 m yr-1), ranging to the “full connectivity” case. Generally, the proportion of OC exported by rivers increases for higher mean annual runoff and higher runoff variability (Fig. 9). For mean annual runoff values below 1 m yr-1, the proportion of OC export by rivers is not significantly affected by runoff variability. In general, the results show more sensitivity to the mean annual runoff than to runoff variability. This emerges from the parameterization of the long-term fluvial transport capacity (Eqs. 6 and 7). The suspended sediment transport law (Eq. 5) does not have a threshold, which limits the role of large runoff events in the long-term transport capacity. Therefore, the long-term transport capacity is controlled by intermediate runoff events, which are set by the mean annual runoff value (see further explanation in Croissant et al., 2019a).

The connection velocity has a major impact on the modeled fate of OC. However, climate strongly moderates this response. For instance, the full connectivity scenario shows that the proportion of the landslide-mobilized OC that is exported by rivers ranges from 30 % for low runoff and low discharge variability to ∼95 % for high runoff and variability (Fig. 9c). These model outputs predict that most feasible combinations of the climate parameters lead to a dominance of the fluvial export of OC. The low connection velocity produces a specific response (Fig. 9a). This is because the amount of exported material is limited to only the initially connected landslides (e.g., ∼30 % in these simulations). We discuss the broader implications of these results in Sect. 7.

It is important to note that in these simulations we do not attempt to capture any role of climate in the build-up of the soil OC stock in the landscape. Mean annual temperature and rainfall act to moderate the production of OC and its preservation in soil (e.g., Carvalhais et al., 2014). In addition, these variables control the development of mineral soil (e.g., Mudd and Yoo, 2010), and mineral–OC interactions are critical to organic matter stabilization (Hemingway et al., 2019). As such, a wetter climate might act to enhance erosion and export of landslide-mobilized OC (Fig. 9) but also increase its export from hillslopes by controlling soil OC stocks.

Previous work has sought to quantify the provenance and flux of particulate OC following large storm events based on river gauging station data (Clark et al., 2017; Hilton et al., 2008b), remote sensing approaches (Clark et al., 2016; Ramos Scharrón et al., 2012; West et al., 2011) or lake stratigraphic records (Frith et al., 2018; Wang et al., 2020). These studies provide insights on the rates and processes of OC transport to the oceans (Galy et al., 2015; Hilton, 2017). However, they are generally unable to constrain the roles of OC oxidation in the landscape (Fig. 2) and aboveground biomass regrowth (Fig. 4) in the OC budget of a landscape recovering from these damaging events. In our study, we propose a theoretical description of these processes using a simplified numerical approach.

The export of OC during the post-seismic period depends strongly on the fine sediment export rate (Hovius et al., 2011; Tolorza et al., 2019; Wang et al., 2015, 2016). A few examples have shown that, in wet climates, suspended load fluxes after a large earthquake are characterized by a rapid increase directly after the seismic event, which is sustained for less than a decade, before returning to background levels (Hovius et al., 2011). This behavior is reproduced by our simulations (Fig. 10). However, this does not imply that all the fine sediment has been evacuated out of the epicentral area. For instance, abundant landslide deposits located in headwater areas (Ramos Scharrón et al., 2012), or fine particles trapped under coarser grains in the deposits, would slow down the evacuation of sediment once the easily accessible particles have been entrained. In that regard, the scenarios that we propose in Sect. 6 show possible trajectories of the partitioning between oxidation and fluvial export, with oxidation being the dominant process after an initial pulse of fluvial evacuation of OC. However, we note that the initial pulse is present in all simulations, independent of the connection velocity, and implies that in all cases fluvial export is the dominant fate of OC mobilized by landslides in a setting like the Southern Alps. These findings are consistent with the available measurements from the 2008 Wenchuan earthquake (Wang et al., 2016) and fit the pattern of organic matter accumulation seen in lake deposits that record four past Alpine Fault earthquakes (Frith et al., 2018).

Other research has sought to assess the role of landslide erosion as a CO2 source or sink. For instance, Ramos Scharrón et al. (2012) proposed a post-landsliding OC evolution model that accounts for the competition between carbon gains and losses. Their modeling approach differs from ours because the processes that result in a CO2 sink only include soil formation, aboveground biomass growth on landslide scars and OC storage in landslide deposits. In their approach, fluvial transport contributes to oxidation as part of the OC source term. Despite these important differences, their results also suggest that widespread mass wasting is a CO2 sink over long timescales in settings where landslide deposits have long residence times (centuries), which would correspond to our low connection velocity scenario. Our results differ from theirs in that we model connection of landslides to the mountain river network, leading to scenarios in which up to 100 % of the OC mobilized by landslides is removed by rivers rather than oxidized in the landscape.

The mobilization of OC by events that trigger widespread landsliding can produce large pulses of carbon export from the biosphere (Clark et al., 2016; Frith et al., 2018; Wang et al., 2016; West et al., 2011). These carbon fluxes are large enough to raise questions about the role of large erosive events associated with earthquakes and storms in the long-term carbon cycle (Hilton and West, 2020). In this study, we propose a framework to examine the fate of OC eroded during a large earthquake and isolate how the climatic context plays a role in promoting fluvial transport. We do not attempt to quantify any subsequent degradation of OC in the marine realm or within sedimentary deposits. In the context of “short” mountain ranges along active continental margins with sediment transport distances of <100 km and rivers that are coupled to deep-ocean sedimentary sinks (e.g., Mountjoy et al., 2018), it has been shown that 70 % or more of the OC exported in river sediments is buried in offshore sedimentary deposits and is therefore considered a long-term sink (Blair and Aller, 2012; Galy et al., 2015; Kao et al., 2014).

Our modeling exercise has shown that earthquakes can contribute to long-term biospheric carbon sinks in the Southern Alps case, with minimal within-catchment oxidation of eroded OC (Fig. 10). The model is broadly relevant to other mountain catchments in the western Pacific: Taiwan and Papua New Guinea (Dadson et al., 2005; Ferguson et al., 2011; Hicks et al., 2011; Hilton et al., 2008b). However, these settings also share geomorphic, tectonic and climatic attributes with islands of the Caribbean (e.g., Allemand et al., 2014) and of Central America (e.g., Ramos Scharrón et al., 2012), as well as some catchments draining the western American continent (Leithold et al., 2006). In these conditions, the combination of short sediment travel distance and limited potential for fluvial sediment storage may promote efficient carbon transport from land through river systems (Scheingross et al., 2019). However, a complete understanding of the role of earthquakes in the carbon cycle must consider large floodplains, such as the river systems that drain the Himalayas and the Andes. For instance, Galy et al. (2008) showed evidence for the replacement of Himalayan biospheric OC in the Ganges floodplain, and particulate OC eroded from the Andes is thought to suffer a similar fate during transport through the Amazon floodplain (Bouchez et al., 2014; Ponton et al., 2014). Drier climates may also contrast with the findings we describe because the OC stocks in soils may be lower, and landslide debris may reside in catchments for much longer periods of time.

Our model has been applied to a single case study with a specific tectonic and climatic regime (Fig. 10). It thus represents a first step in understanding how widespread landslide-triggering events impact regional to global carbon fluxes. Ideally, future work can explore the diversity of mountain ranges that present different vegetation, soil cover, topography and climate, with a specific consideration of the recurrence time of earthquakes (Fan et al., 2018). Additionally, a complete carbon budget (Hilton and West, 2020) could be assessed if other long-term CO2 sinks via silicate mineral weathering that occurs in landslide deposits are considered (Emberson et al., 2017), alongside CO2 release from sulfide mineral oxidation (Emberson et al., 2016a, 2016b) and oxidation of rock OC (Hilton et al., 2014).

In this paper, we couple the reduced-complexity model Quakos to organic matter degradation models to assess the fate of OC following a large landslide-triggering event. As with any modeling approach, our model contains some limitations that are discussed here.

Prediction of the landslide pattern for an event that has not taken place during the period of modern instrumental records is a difficult endeavor. We were able to constrain the shape of the landslide distribution using observations of rainfall-triggered landslides in the Southern Alps (Hovius et al., 1997). Nevertheless, uncertainties remain as to the landslide spatial density that would emerge from a Mw∼8 earthquake on the Alpine Fault (Robinson and Davies, 2013). While this parameter does not significantly affect the amount of OC redistributed in the different reservoirs (Fig. 6), it would greatly impact the OC fluxes and therefore their comparison to background fluxes. Our computations show that for the “full connectivity” scenario, maximum post-seismic OC fluxes derived from the fluvial export would correspond to ∼2 times (for a peak density of 4 %) to ∼6.5 times (for a peak density of 8 %) the background fluxes computed from Hilton et al. (2008a).

One of the main limitations of our modeling approach is in the assumption of a connection velocity ucon. As discussed in Croissant et al. (2019b), the lack of a theoretical framework describing the evolution of non-cohesive material sitting on hillslopes after a landslide-triggering event makes the description of this process difficult. For fine particles, there are several key processes that need to be considered across a range of scales. These include those operating at the grain scale (e.g., creep, rain splash) to those operating over length scales of meters (e.g., dry raveling, runoff-driven erosion) to hundreds of meters (landsliding, debris flow). The availability of fine sediment is also likely to change over time, for instance through the initial removal of easily transportable debris (Wang et al., 2015; West et al., 2014; Zhang et al., 2019). To move more finer material, subsequent mass movements may be needed to expose it within deposits and/or within mountain river channels (Fan et al., 2019). It is tempting to compare the values of connection velocity chosen here (1 to 100 m yr-1, or 3×10-8 to 3×10-6 m s-1) to the velocity of water flow sediment export via a range of processes (e.g., DiBiase et al., 2017). In this case, the chosen connection velocities are very slow. However, they seek to describe the net motion of a large sediment pile that integrates a range of processes that occur discretely in space and time.

Numerous physically based and empirically based models have been proposed to explore and predict hillslope transport processes (e.g., Aksoy and Kavvas, 2005; DiBiase et al., 2017; Tucker and Bradley, 2010). Future work could seek to incorporate more sophisticated transport laws into the framework we describe here. This would also allow the links between landslide remobilization and climate to be explored, as there is likely to be a strong link between movement of existing landslide debris and events with high rainfall intensity and a high river stage (Marc et al., 2015). Useful insights could be gained using morphodynamic models with a realistic grain size distribution (Fan et al., 2019). An additional source of uncertainty relates to the dynamics of particulate OC transport versus clastic sediment. While the general link between the two phases has been demonstrated (Galy et al., 2015; Hilton et al., 2012), there is also evidence that discrete clasts of woody debris can be sorted by flowing water (Hilton et al., 2015; Lee et al., 2019), and its transport behavior is different from the clastic load due to a lower density and different particle shape (Turowski et al., 2016).

The model approach assumes that the majority of fine sediment is available for transport, i.e., located at the surface or near the subsurface of landslide deposits. In reality, fine sediments will be distributed within landslide deposits, and some are therefore shielded from entrainment. To export this buried fine sediment, removal of some coarser fraction would have to occur, most likely during high-intensity rainfall events that have the capacity to export the coarse fraction. This feature would have the effect of prolonging the storage time of fine sediment and OC on hillslopes. Accounting for this process would require prior knowledge of the distribution of fine sediment within the deposit, which is still largely unknown (Fan et al., 2019), and a more sophisticated sediment transport model to account for the export of the coarser fraction of the sediment (Croissant et al., 2017).

In our simulations, we simplify how the degradation of OC proceeds following its mobilization of soil from hillslopes. First, it is possible that the landsliding process sorts and mixes the organic matter in such a way to promote oxidation of some of the materials but to protect some from degradation. For instance, observations in the field suggest that large woody debris may concentrate near the surface and toe of the deposit after a landslide. Instead, finer material may mix deeper in the deposit (Hilton et al., 2008b). If the landslide deposit has low porosity and/or receives runoff focused from the landslide scar (Emberson et al., 2016a), waterlogging may promote preservation of this material under anoxic conditions. This would act to enhance the proportion of OC exported by erosion processes. Second, we do not account for OC contained within sedimentary bedrock. In the western Southern Alps, the OC content of rocks is low (∼0.15 %), but oxidative weathering is thought to be enhanced by high erosion rates (Hilton et al., 2014; Horan et al., 2017). In the future, improved constraints on the reactivity of rock-derived OC (Hemingway et al., 2018) and models that link the production of fine clastic sediment to weathering reactions (Carretier et al., 2018) could be used to explore the role of large landslide populations in this mechanism for CO2 release.

Finally, we assume there is no OC oxidation during river transport. Our New Zealand case study is focused on short rivers with maximum distances from headwaters to oceans of less than 100 km. This assumption is supported by recent work by Scheingross et al. (2019), who showed that within-river oxidation would be minimal, with between 0 and 10 % of mobilized OC being oxidized during transport over 1000 km. We also note that in the simulations, intermediate discharge events dominate the long-term transport rates (Croissant et al., 2019a), meaning that most of the OC would be exported by flows that do not escape the river channel and thus not go overbank onto the floodplains. This dynamic could be altered if a sediment bed-load wave significantly modifies the riverbed elevation (e.g., Hancox et al., 2005; Korup, 2004). Finally, we recognize that in the Amazon River floodplain, high-altitude-derived OC from mountain catchments is thought to be lost and/or overprinted by lowland OC (e.g., Bouchez et al., 2014; Ponton et al., 2014). This means the geomorphic configuration of mountain catchments impacted by earthquake events could play an additional role in the fate of the mobilized OC.