Gruobenwald (46.89025° N; 9.85585° E; 1050–1300 m a.s.l.) is a roughly 35° steep beech spruce forest near Klosters, Switzerland (see Fig. ). Due to the steep slope, the cliffs above the forest, which reach heights of up to 130 m, and the exposed infrastructure (road of national importance, railway, residential area), rockfall hazards have been an ongoing issue for many years. As the outdated protective structures needed to be replaced, the construction of new measures started in 2022, consisting of 650 m of reinforced earth dams with a capacity of 5000 kJ and 400 m of rockfall barriers with a capacity of 2000 kJ. There are many dead trees, so called snags, without bark and branches, but which are still standing. In June 2022, a storm with heavy rainfall and wind caused a dead tree to fall and slide approximately 120 m down the slope until it came to rest on the national road (Fig. c). The mobilised tree had an approximate diameter at breast height DBH = 75 cm and a stem height hS = 30 m. Two days later, another deadwood log of similar dimensions slid down the slope and came to rest in a wooden palisade next to a building (see Fig. b). Due to these two incidents and the associated danger to the construction site, the local forestry operators decided to cut down potentially dangerous snags without bark and branches and to transport them down to the valley by helicopter. During this operation, two more logs were mobilised after being cut down, slid down the slope and came to rest on the road or hiking trail, respectively. After completion of the intervention, there were no more recorded events as of December 2025.

Plattawald (46.77072° N; 9.41829° E; 600–1000 m a.s.l.) is a mountain forest near Rothenbrunnen, Switzerland. The average slope gradient is 35–40°, and a road and railway line run along the foot of the slope. There are numerous cases of rockfall and sliding deadwood on this slope, which cause only minor damage as the rocks and deadwood are caught by a series of rockfall barriers above the road. In February 2022, a tree without bark and branches was mobilised, slid down approximately 200 m in altitude and came to rest in the valley-side guardrail of the road, blocking both the road and the railway. It is unclear whether the mobilised tree was initially standing or lying. Due to a thin layer of snow on the day of the incident, the tracks of the sliding log were clearly visible and were documented by the local forestry operations. As the section of road where the log was deposited is protected by a 3.5 m high rockfall barrier and no traces were found in the immediate vicinity of the barrier, the log must have jumped over the barrier. This claim is reinforced by the fact that the last traces were found on a small terrace about 20 m above the barrier, where the log was presumably launched into the air. The next tracks were then found on the retaining wall next to the road.

Following , the motion of the log is described using two coordinate frames. The inertial frame I is fixed in space and anchored at the origin O, which is defined at the origin of the digital elevation model (DEM). The axes exI and eyI span the horizontal plane, while the axis ezI extends along the vertical, counter-gravitational direction. The body-fixed frame K is attached to the centre of mass S of the log. The three axes exK, eyK and ezK are along the log's principal axes of inertia. The minor principal axis of inertia for a truncated cone lies in its longitudinal direction, which we define as ezK. The two coordinate systems, their origins and axes are shown in Fig. a.

At any time t, the position and orientation of the log, modelled as a rigid body, is given by the generalised coordinates q: 1q:=IrOSpIK∈R7, where IrOS is the position vector from the origin O to the centre of mass S, expressed in the inertial frame I, and pIK=(e0eT)T∈R4 is a unit quaternion describing the orientation of the log. A linear mapping of the dimension-reduced, generalised velocities u leads to q˙: 2q˙=F(pIK)u,u:=IvSKΩ∈R6,F(pIK)=I3×300-12eT012e0I+12ẽ, where IvS is the translational velocity vector expressed in the frame I, and KΩ is the rotational velocity vector represented in the frame K.

The time evolution of IvS is given by the balance of linear momentum: 3mRIv˙S=IFg+IFc, where mR is the mass of the deadwood log. IFg is the gravitational force and IFc is the sum of all contact forces between the log and the terrain. The air resistance is neglected. Accordingly, the time evolution of KΩ obeys Euler's equation: 4KΘSKΩ˙+KΩ×KΘSKΩ=KMS, where KΘS defines the inertia tensor for the log expressed in the frame K, and KMS is the net torque generated by contact forces acting on the boundary of the log. The geometry of the log is approximated by a point cloud defining a polytope. Each vertex may come in contact with the terrain, inducing a contact force at that vertex. The contact forces of all vertices are assembled in a tuple λ. When the sliding log is not in contact with the terrain or surrounding trees, both IFc and KMS are zero. The equation of motion of the sliding deadwood can be summarised as: 5Mu˙-h(q,u)=W(q)λM=mRI3×303×303×3KΘSh(q,u)=IFg,-KΩ×KΘSKΩT Here, M is the diagonal mass matrix, h(q,u) contains the forces acting on the centre of mass (i.e. gravity and gyroscopic terms) and W(q) is the matrix of generalised force directions that transfers all the contact forces at the vertices on the log boundary to a corresponding force and torque at its centre of mass. Equation () together with contact laws is solved iteratively using Moreau's time-stepping scheme with a Gauss–Seidel iteration method .

A contact between the log and the terrain is detected whenever at least one vertex of the log lies on or below the terrain surface . At every contact point, a local contact frame is defined with one normal and two tangential directions relative to the terrain surface. In the contact normal direction, the Signorini condition correlates the normal contact force λN and the normal gap gN as follows: 60≤λN⟂gN≥0.

The two tangential directions are mutually orthogonal and span the local tangent plane, while their orientation within this plane is arbitrary. The tangential contact behaviour within this plane is governed by the spatial Coulomb friction law: 7-γT∈{0}if∥λT∥<∥μλN∥,stickingR0+λTif∥λT∥=∥μλN∥,slipping where μ is the sliding friction coefficient. The negative sign indicates that the resulting tangential contact force always acts opposite to the relative tangential contact velocity within the terrain tangent plane. The contact force vectors λi=(λT,iTλN,i)T for 4 vertices are visualised in Fig. c.

Additionally, the hard contact approach requires an impact law, which, in combination with an impact equation, defines a mapping from pre- to post impact velocities. The impact law used for this model is based on the Newtonian impact law, which includes two cases : 8γN++εNγN-=0∧ΛN>09γN++εNγN-≥0∧ΛN=0 where γN- and γN+ are the normal pre- and post-impact velocity components, and εN is the normal restitution coefficient. The first case corresponds to an active impulsive contact force ΛN>0 whereas the second case describes a separation induced by impacts at other contact points. Consistent with rockfall modelling, we use a normal restitution coefficient of εN=0, following the underlying rockfall model.

The contact detection between a sliding log and standing trees follows the algorithm by , with its interaction also considered as hard contact. The standing trees are modelled as rigid, truncated cones with a DBH and hS (see Fig. a). The energy dissipation caused by the uprooting and swaying of standing trees during an impact is not considered. Unlike the original tree model for rockfall applications, the standing trees do not have an energy threshold indicating a breaking failure during impact because of the assumption that the sliding deadwood will always break before the standing tree.

In addition, we include existing rockfall barriers in the model as thin, platy obstacles with the same impact behaviour as the standing trees. The barriers are rigid, have an infinite energy threshold and follow hard contact mechanics.

Given the initial conditions derived in Sect. , the only unknown model parameter left is the friction coefficient μ. The observed deposition pattern can be reproduced with a friction coefficient in the range of μ = 0.2–0.3. With higher friction, the road cannot be reached by the sliding deadwood. With lower friction, the logs do not stop on the road and continue sliding. The corresponding simulation results for a friction coefficient of μ = 0.25 are shown in Fig. a. We assume that all four events occurred with the same deadwood geometry and the same friction coefficient. The simulated deadwood logs show average velocities of 14 m s⁻¹ with maximum velocities reaching 30 m s⁻¹. Out of the 20 000 simulated logs, 1829 reach the national road, as determined by counting the number of trajectories intersecting the pink reference line shown in Fig. . This corresponds to a relative frequency of 9.1 % according to Eq. (). The lateral spread, calculated using Eq. (), amounts to 40, 60, 60, and 55° for events 1–4, respectively. Figure b presents the simulation results under the same conditions as Fig. a, but with all newly implemented protection measures in place. In the depicted section, these measures include two embankments and six rockfall barriers completed in 2024, which successfully reduce the relative frequency of reaching the road to 0 %. Figure c shows the simulation results for the scenario with a randomly generated forest based on the forest density as well as the DBH and tree height distributions obtained from the ALS data. In this case, the relative frequency of reaching the road is reduced to 7 % without a change in lateral spread compared to the case with the real forest structure. An exemplary magnified section below the first release polygon highlights the structural differences between the real forest (Fig. e) and the randomly generated forest (Fig. f). Lastly, Fig. d presents the scenario without any standing trees. In this case, the relative frequency of reaching the road increases to 74.5 %, while the lateral spread for events 1–4 amounts to 23, 31, 25, and 18°, respectively.

We can verify the hypothesis of the log jumping over the rockfall barrier by fitting a flight parabola between the presumed take-off and landing points. To obtain the 3D coordinates of the mapped points, they are projected onto a 0.5 m DEM. Since the log was sliding, the lift-off angle must have been equal to the local slope angle. We assume that only gravity acts on the log and that the longitudinal axis of the log is always parallel to the resulting velocity vector. The two projected points, together with the take-off angle, define a unique flight parabola. This parabola passes over the rockfall barrier (see Fig. c), which proves that such a jump is indeed possible. In addition, the uniquely defined take-off velocity of 24 m s⁻¹ is within realistic bounds. In order to reach the required take-off velocity to jump over the barrier with the proposed 3D model, a friction coefficient of μ<0.3 is required, which corresponds to the same maximum allowable friction to reconstruct the events in Klosters. Contrary to the Klosters case, however, it is not possible to derive a minimum required friction, as the log was not stopped gradually by friction between log and terrain but abruptly by a guardrail. The 3D simulations confirm that it is possible that a mobilised tree jumps over the rockfall barrier. According to the simulation shown in Fig. a, 1442 of the initially 20 000 released logs reached the southernmost barrier of which 1264 were stopped and 178 jumped over. This results in a relative frequency of the observed event of 0.9 % according to Eq. (). This frequency is small but reasonable given the fact that such an event has never been recorded despite the high number of mobilisations observed. A plausible 3D trajectory of the observed event can be found in Appendix , Fig. . Additionally, we can demonstrate the efficiency of the existing rockfall barriers by removing all protection measures in a simulation (see Fig. b). In this case, 4658 logs (23.3 %) reach the road.

The documented cases of sliding deadwood highlight three key factors contributing to the risk of sliding deadwood impacting infrastructure below: (1) low friction; (2) steep slopes; and (3) early-stage wood decomposition.

Low friction either occurred due to terrain conditions (e.g. wet soil, snow, foliage cover), the condition of the deadwood (wet log without bark and branches) or a combination of both. Physically, a sliding movement is possible for a friction coefficient of μ<tan⁡ψ, where ψ is the slope inclination. The proposed model can reconstruct both case studies with a friction coefficient in the range of μ = 0.2–0.3, which therefore describes the typical contact friction between wet terrain and a wet log without bark and branches. A friction coefficient of μ = 0.3 means that under these conditions, a mobilisation is theoretically possible at slope angles ψ>18°. For a slope angle of 30°, a mobilisation is possible for μ<0.58.

All events occurred in very steep forests with slope angles ψ≥35°. While this is directly linked to the first criterion, slope angle warrants a separate discussion. In a typical multibody simulation of relatively smooth contacting bodies, the Coulomb friction model is used to account for sliding resistance due to various abrasive processes which take place on a microscopic scale. These frictional processes are highly complex, depend heavily on the microscopic roughness, and are quite bluntly described by the Coulomb friction model through a single parameter, being the friction coefficient μ. In the modelling of sliding deadwood (and also rockfall), the macroscopic roughness is modelled through the chosen resolution of the DEM. The roughness effects which fall below the resolution of the DEM are to be accounted for by the Coulomb friction model. This macroscopic roughness of the DEM implies that very steep forests with ψ≥35° are a requirement for deadwood mobilisation even though the calibrated μ = 0.2–0.3 suggests a minimum required slope angle of 12–18°. This also implies that the calibrated μ is only valid for the DEM resolution of 0.5 m and should be increased accordingly when working with lower resolution DEMs. A sensitivity test for the Gruobenwald case study showed that the relative reach frequency of the national road increased from the initial 9.1 % for the 0.5 m DEM to 9.5 % and 18 % for the 2 and 5 m DEMs, respectively, when using the same friction coefficient of 0.25. This illustrates the effect, although its magnitude most likely depends on the terrain roughness present at a given site.

All of the recorded events occurred with fresh deadwood with early-stage wood decomposition. More specifically, the decay was advanced to the point where the trees have lost their branches and bark to reduce sliding friction. At the same time, the trees still showed sufficient structural strength so that they did not break on impacts while sliding. According to the decay stage classification by , this deadwood state corresponds to a late stage II or early stage III. In this critical decay stage, a pocket knife can penetrate the wood only in the direction of the fibres but not in perpendicular direction . For a case study on a spruce forest with bark beetle infestation, this critical decay stage was reached 2–5 years after death . The critical time period, however, heavily depends on the tree species. showed that beech has significantly higher decay rates than spruce or pine. Additional influencing factors are the stem diameter , temperature, humidity and contact to the ground . Distance to the ground may affect local moisture conditions and accessibility for microbial colonisation, particularly at low heights and where dense vegetation, such as shrubs, tall herbs or grasses, surrounds the dead stems. , for example, included sample distance to ground as a potential predictor of deadwood decay. While they did not find a significant effect, they noted that this variable was measured only at the time of sampling, although it is temporally variable, and that multi-year averages may better represent the conditions experienced by a log throughout the decay process. This is particularly relevant in the present context, as branches may break and the stem may settle closer to the ground over time, potentially increasing moisture availability and contact with decomposer communities and thereby accelerating decay. Slope exposure represents another relevant but complex site factor. Exposure can influence both temperature and moisture, which may have opposing effects on decay rates. South-facing slopes may favour decay through higher temperatures and longer biologically active periods. However, they may also lead to faster drying of stems, potentially constraining decomposition, whereas north-facing slopes may favour decay through higher humidity and more favourable conditions for fungal activity .

This critical decay stage also explains why the majority of deadwood mobilised in the presented events were snags triggered by an active mobilisation mechanism (e.g. windthrow, rockfall, snow weight or logging). Only in the Rothenbrunnen case study, it is unclear from the documentation whether the deadwood was initially standing or lying. Active mobilisation is important especially for snags because if a snag falls purely due to decay without any additional loads, it is very likely that it will immediately break apart on impact and therefore cannot slide over long distances. Lying deadwood may also be mobilised at sufficient structural strength (e.g. by rockfall or erosion). However, since it has already withstood a major mobilisation event when the respective tree or snag initially fell, it must be held back by significant resistance, such as terrain roughness, blocking trees, or other deadwood, making remobilisation inherently more difficult. In fact, lying deadwood may even help reduce the hazard potential of sliding deadwood, as it acts as a natural barrier, blocking or slowing down mobilised logs, provided it is itself firmly locked in place.

While all three factors are prerequisites for risks from sliding logs, local natural conditions can vary significantly. The first case study shows a typical critical combination of conditions which fulfil all the criteria for sliding deadwood, where a steep forest with previous bark beetle infestation led to several snags without bark and branches (reduced friction) and is later hit by a storm (reduced friction due to rain, active mobilisation mechanism due to wind). The second case study shows that wet snow can also be critical, as it causes mechanical stress on trees and at the same time reduces friction.

Standing trees influence sliding deadwood in two distinct ways, each affecting the hazard differently. On the one hand, trees act as effective physical barriers that can block or arrest sliding logs, particularly when the longitudinal axis of a log is misaligned with its velocity vector. This protective effect is clearly reflected in the simulation results. In the Klosters case study, the relative frequency of sliding logs reaching the road increases from 9.1 % for the simulation with standing trees to 74.5 % when the forest is removed. In the Rothenbrunnen case study, this frequency increases from 23.3 % with standing trees to 100 % without forest (see Appendix , Fig. ). These results demonstrate that the forest itself represents a highly effective protective measure against hazards caused by its own mobilised deadwood.

On the other hand, standing trees can deflect sliding logs, increasing lateral spread. Due to their elongated shape, logs can traverse slopes over long distances and often deviate substantially from the steepest descent direction. Once a log enters a traverse motion, realignment with the fall line becomes unlikely, as its length and repeated interactions with standing trees hinder rotational adjustment. In the Klosters case study, the simulated lateral spread amounts to 40, 60, 60, and 55° for events 1–4. When the simulations are repeated without forest, the lateral spread decreases to 23, 31, 25, and 18°, respectively. For the Rothenbrunnen case study, the same comparison shows a reduction in lateral spread from 49 to 18° (see Appendix , Fig. ).

A key observation from the events in Klosters is that the deadwood tends to slide further when the simulation is run with tree locations and DBH from the ALS instead of a randomly generated forest with the same tree density and DBH distribution. For the simulation shown in Fig. a, the relative frequency of reaching the road decreases from 9.1 % with the ALS tree data to 7.0 % with the randomly generated forest. The reason for this decline is that the preferred path of the sliding deadwood is determined by the micro-topography (e.g. sliding in ditches, valleys, gullies). In a randomly generated forest, these preferred paths can be blocked by standing trees. In reality, however, these gullies are often treeless or less densely forested due to rockfall, avalanche dynamics, and other channel processes, as well as unfavourable site conditions related to persistent snow cover and competition from herbaceous vegetation . The comparison between Fig. e and f highlights the importance of using ALS–derived tree locations: while the preferential sliding path is largely vegetation-free in the real forest, it is obstructed in the randomly generated forest, leading to an underestimation of the hazard potential.

In contrast, the Rothenbrunnen case study shows a smaller difference between simulations using ALS tree data and a randomly generated forest. For the simulation shown in Fig. b, the relative frequency of reaching the road even increases slightly from 23.3 % with the ALS tree data to 23.8 % with the randomly generated forest (Appendix , Fig. ). This negligible difference can be explained in this case by the fact that there are no clear preferential sliding paths in gullies that could be blocked by randomly generated trees. This suggests that in non-channelised terrain, precise tree locations are less important, as there are no distinct preferential sliding paths that could be blocked by randomly placed trees. Nonetheless, individual tree data remains valuable for accurately determining forest stem density.

These results furthermore suggest that the spatial correlation between forest structure and terrain-controlled preferential process paths may also be relevant for rockfall modelling, as rocks, similarly to sliding deadwood, tend to follow preferential paths controlled by the terrain. With increasingly detailed forest data and improved processing methods , explicitly accounting for exact tree locations and forest structure could therefore represent a promising direction for future hazard assessments.

Steep mountain forests, where sliding deadwood is possible, often serve as protection against natural hazards. These areas, prone to sliding trees, are thus typically also exposed to rockfalls, avalanches or landslides. Compared to rocks in rockfall hazard scenarios, deadwood logs have a less favourable shape, lower strength and density than rocks, generally resulting in reduced velocities, kinetic energies and run-out distances. This may suggest that, in most cases, the danger from sliding deadwood plays a relatively minor role in the planning of protective measures such as barriers or embankments. However, there are some conditions where a hazard analysis of sliding deadwood should be considered:

According to our work, the risks posed by sliding deadwood in protection forests and natural hazard management are relevant only in specific situations. In most forested areas with relatively gentle slopes (< ∼ 35°) and high roughness, deadwood can mitigate rockfall and avalanche risks and support sustainable forest regeneration . However, along steep gullies with low surface roughness, it may be necessary to assess whether removing dying or dead trees or actively cutting and securing them across the slope is appropriate to reduce overall gravitational hazard risks.

In both case studies, sliding deadwood reached infrastructure despite protective measures. In Klosters, this was due to outdated protective structures (wooden palisades) or a lack of protective measures in some places. In Rothenbrunnen, deadwood was able to jump over a barrier positioned below a terrace. When considering implications for natural hazard management, it is important to note that: (1) a rockfall in both cases could have caused similar or greater damage, and (2) perfect protection is hardly ever guaranteed due to construction constraints like barrier height or length. Nevertheless, the comparisons between scenarios with and without protection, as shown in Figs.  and , highlight the overall efficiency of the current measures. Although events involving sliding deadwood highlight significant hazards and, depending on topography, can expand the lateral extent of risk zones, the need for additional structural measures must be carefully evaluated.

The decision to leave deadwood in the forest depends on the local natural conditions, the relative importance assigned to biodiversity and various forest functions and potential effects on subsequent disturbances . Regarding the deadwood's role in natural hazard protection, its changing effects over time are particularly significant . showed from large-scale laboratory tests that the fracture impact energy of a log is reduced by up to 89 % in 8–10 year old deadwood with rot fungus compared to fresh deadwood. As the structural strength is also important for the protective effect of deadwood, the age of the deadwood is linked to competing positive and negative effects, emphasising the need for an integral view on deadwood from a forest management perspective. Deadwood in its earliest stages shows high structural strength and thus a high protective effect against rockfall and avalanches, but temporarily offers often suitable habitat for bark beetles in spruce-dominated forests . In a later decay stage (2–5 years after death for spruce), structural strength and protective effect still remain intact, while at the same time, the hazard of sliding logs increases. In the advanced decay stage (5+ years after death in spruce), the risk of sliding deadwood decreases, but so does its protective effect. Conversely, the ecological value of deadwood continues to increase, particularly as a habitat and, typically after more than 20 years , as a substrate for seedling establishment, thereby promoting forest regeneration.

This trade-off also applies spatially. Although deadwood logs in gullies have a higher sliding hazard potential, they also offer increased protection against other natural hazards. For rockfall, these gullies are preferred rolling paths and therefore an optimal location for a natural barrier. In the case of avalanches, deadwood logs stabilise the snow cover in these less vegetated areas . Our study sheds light on an often overlooked but, in specific situations, important effect of deadwood, helping to decide where and when retaining deadwood is overall beneficial, and where alternative management approaches, such as removal or controlled relocation and stabilisation, may be more appropriate. The targeted felling and stabilisation of trees as protective structures was extensively investigated by . The configuration of such structures can be optimised, for example by supporting and securing felled trees on stumps or standing trees using wire cables, to enhance their protective effect against rockfall while preventing subsequent downslope mobilisation.

In addition, the strong protective effect of standing trees against sliding deadwood implies a spatial differentiation of hazard within forested slopes. Logs mobilised high on a forested slope still traverse a long forested runout path and therefore have multiple opportunities to be intercepted by standing trees. In contrast, logs mobilised in the lower parts of the forest encounter little or no forest remaining downslope, which substantially increases the likelihood of reaching exposed infrastructure. This suggests that snags and deadwood located in lower forest sections that are still sufficiently steep may require particular attention in hazard assessments, and that interventions such as felling should be carried out with caution in these areas.

Despite the high level of process detail incorporated through a physically based modelling framework coupled with high-resolution ALS tree data, several limitations remain. Firstly, sliding deadwood is represented as rigid bodies that do not fracture upon impact. While this neglects potential breakage at advanced decay stages, it is fully consistent with the objective of this study. The purpose of the model is not to simulate all possible deadwood conditions, but to quantify runout, lateral spread, and hazard potential under those specific conditions that make sliding deadwood hazardous. As shown in Sect. , these conditions correspond to a narrow decay-stage window in which sufficient residual structural strength is still present to allow long-distance transport without immediate fragmentation. The model is therefore explicitly conditioned on a prior assessment of decay stage: only logs that fall within this critical structural state are considered potential hazards and simulated. Under this premise, assuming non-breaking rigid logs is appropriate. At the same time, this assumption yields conservative, worst-case estimates of runout and impact frequency, since even structurally strong logs may still break depending on impact velocity and configuration. Logs in more advanced decay stages are expected to fragment early and thus pose a substantially lower hazard, but are outside the intended scope of the present model.

The rigid-body representation furthermore neglects flexural deformation of the log during impacts. In reality, long stems may bend and temporarily store strain energy, which may subsequently be dissipated or released and thereby influence the post-impact translational and rotational motion.

The initial conditions of the simulations introduce an additional limitation. Logs are released in a simplified, upright configuration and randomly positioned within the specified release area, without explicitly modelling the true mechanical triggering process or the transition from toppling to sliding. This idealisation may influence the very early phase of motion, particularly the initial orientation of the log. However, this is fully consistent with the intended application of the model: the mobilisation potential of deadwood is first assessed based on decay stage, terrain inclination, and local site conditions. The simulation tool is then applied under the explicit condition that mobilisation is already deemed possible, in order to quantify runout and lateral spread. This separation between mobilisation assessment and runout modelling follows standard hazard-modelling practice.

Additional simplifications include the rigid and indestructible representation of standing trees, which neglects energy dissipation due to tree swaying, uprooting, breaking or soil–root interactions during impacts. This again leads to slightly conservative estimates of runout and lateral spread. Furthermore, friction is represented by a single effective Coulomb coefficient that implicitly incorporates microscopic surface roughness effects. As discussed in Sect. , this parameter is resolution-dependent and valid only for the DEM resolution of 0.5 used in this study. Despite these limitations, the model reliably reproduces the observed deposition patterns of documented events and fulfils its intended purpose: providing conservative, physically consistent estimates of runout and lateral spread conditional on a prior mobilisation assessment.

The number of well-documented sliding deadwood events remains small, and the recurring patterns identified in the presented case studies should therefore not be interpreted as an exhaustive description of all possible controls. The observed events consistently indicate the importance of low friction, steep terrain, and a suitable decay stage, but additional mechanisms may become relevant under different site or tree conditions.

One such aspect is the influence of branches, which may introduce competing effects on sliding mobility. They may partially suspend the stem above the ground and help bridge terrain roughness or obstacles, but may conversely increase resistance through interlocking with terrain or standing trees or even penetrating into soft ground. The balance between these mechanisms is likely to depend on terrain roughness, local obstacles, forest structure and crown architecture of the sliding tree. In the presented case studies, all sliding events involved branchless logs, while terrain roughness along the main sliding paths appeared to be of secondary importance. This supports the interpretation that the very low friction of smooth, barkless and wet stems was particularly important under these conditions. At other sites with more pronounced roughness or different crown architectures, the influence of branches may differ.

Additional mobilisation mechanisms may also occur beyond those observed in the presented case studies. Rockfall impacts, for example, could plausibly act as an active mobilisation mechanism if the other necessary predispositions are met. Unlike mobilisation by wind or snow load, a rockfall impact may additionally transfer momentum to the stem and thereby alter its initial motion. These triggering mechanics are not resolved by the present model, which idealises the initial conditions and focuses on runout after mobilisation.

Further systematic documentation of future events would therefore be valuable for testing the generality of the patterns identified here and for better constraining these additional mechanisms.

The sliding deadwood model can be applied directly by any user with access to RAMMS::Rockfall, without requiring code-level modifications or a special software version. A simulation is set up following the standard rockfall workflow, including the definition of release areas and forest. To suppress scarring effects, the ground mechanical properties are set to a high ME (e.g. ME = 1000 MPa) and a zero drag coefficient (CD = 0). The friction coefficient of the slope is then defined by manually editing the input XML file and setting μmin⁡=μmax⁡ equal to the desired effective Coulomb friction coefficient.

The geometry of the sliding log is generated externally, for example using a Python script, based on a chosen DBH and log length, which can then be imported as a custom rock shape in RAMMS.

The presented sliding deadwood model is physically tailored to a rigid-body solver that explicitly resolves frictional contact and interactions with individual trees. In principle, it can only be transferred to numerical frameworks that fulfil three essential requirements:

Representation of the moving object as an extended rigid body rather than a lumped mass or spherical particle