Informative landslide hazard estimates are needed to support landslide mitigation strategies to reduce landslide risk across the United States (U.S.). While existing national-scale landslide susceptibility products assess where landslides are likely to occur, they do not address how often, which is a critical element of landslide hazard and risk assessments. In particular, the U.S. Federal Emergency Management Agency’s National Risk Index (NRI) requires landslide frequency estimates by county, which are U.S. administrative regions ranging from 120 km² to 377,055 km² in size, to inform expected annual loss estimates. In this study, we present county-level landslide frequency (landslides area^-1 y^-1) estimates for the 50 U.S. states. We applied Bayesian negative binomial regression to estimate both the expected (average) landslide frequency and full distribution of annual landslide counts for each county as a function of landslide susceptible area, frequency of potentially triggering precipitation, and propensity for triggering earthquakes. We trained our model with 62,720 reported landslides from 316 counties with the most comprehensive records available nationwide and used zero-inflated negative binomial distributions as an incompleteness model to correct for temporal reporting gaps. We found that average annual landslide frequencies vary by nearly three orders of magnitude across U.S. counties, ranging from 0.05 (0.04–0.07) landslides 1000 km^-2y^-1 in Midland County, Texas to 31 (21–43) landslides 1000 km^-2y^-1 in Lake County, California and reflecting the country’s strong variations in landslide susceptibility, earthquake probability, and precipitation frequency. Counties with estimated frequencies in the top 20% of all counties are predominately along the West Coast of the continental United States, in mountainous regions of the Pacific Northwest and Intermountain West, in locally steep or earthquake prone regions of the Midwest and South, along the Appalachians, in southern Alaska, and on the big island of Hawaii. By examining the number of landslides predicted in 99^th percentile years for each county, we identified that 31% of U.S. counties have potential for widespread landsliding, even when such large events have not been reported in the training data for that county. Overall, our results better represent the range of possible landslide frequencies and spatial variations across the entire United States than previous national-scale estimates reported in the NRI and can inform other risk reduction and loss mitigation efforts across the United States.

How to cite: Luna, L., Woodard, J., Bytheway, J., Belair, G., and Mirus, B.: Constraining landslide frequency across the United States to inform county-level risk reduction, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-1724, https://doi.org/10.5194/egusphere-egu25-1724, 2025.

In May 2024, a Mega Disaster hit 96% of the municipalities in Rio Grande do Sul (RS) state, southern Brazil, causing 182 casualties and impacting approximately 2.4 million people. In addition to the floods that hit the capital Porto Alegre, more than 15,000 landslides were recorded due to the extreme rainfall event (> 600 mm in some regions), severely impacting an area of nearly 18,000 km². Although other landslide events have been recorded in RS in the past, none of them have matched the magnitude of this one. In this sense, we aimed to generate a landslide susceptibility model, based on historical data and evaluate its capacity to forecast the areas affected by landslides in 2024. This retrospective assessment was performed using an inventory of four past events between 1995 and 2017, totalling 1,211 landslides, represented by 15,580 points. We randomly selected the same number of points (15,580) over the RS to represent non-landslide areas and split the entire sample set into training (70%) and validation (30%). A Random Forest model, leveraging seven morphometric parameters, was employed to generate the map, which was evaluated with the validation sample set. A second validation was carried out considering the landslides in 2024, represented by 324,500 points. This validation was based on the relationship closeness between 2024 landslides and each susceptibility class using frequency ratio. The last evaluation consisted of analysing landslide areas (rupture, propagation, and deposition) and their distribution in each susceptibility class. To achieve this, we automatically divided the 2024 landslide points into three sets, according to the altitude difference found in each polygon. Our landslide susceptibility map presented a high performance, with an overall accuracy of 0.9, being capable of correctly classifying 64% of 2024 landslides into susceptible areas (very high, high, and moderate susceptibility classes). The very high susceptibility class accounted for 31% of the 2024 landslides and had a frequency ratio of 13.04, showing a high correlation between landslide locations and the analysed class. Further analysis revealed that the model successfully predicted 79% of rupture zones, highlighting its robustness in identifying key prone areas. While the model performed well in identifying rupture and propagation areas as susceptible, its predictions for deposition zones were less accurate, likely due to limitations in the historical inventory, which was carried out after 2017, when most landslide deposition areas were no longer visible in remote sensing imagery. Furthermore, even though the 2024 Mega Disaster was responsible for 12.5 times more landslides than all the previous inventory, our model based on 1,211 landslides correctly classified around 9,600 landslides (64%) in susceptible areas. Therefore, although the 2024 extreme rainfall event was much larger than any previously recorded in the region, many areas could have already been identified as susceptible. Finally, the existence of a more complete landslide inventory (including rupture, propagation, and deposition areas) provides more accurate susceptibility maps, which can support territorial planning, contributing to disaster risk management, mitigation strategies, and land use policies.

How to cite: Quevedo, R. P., Maciel, D. A., Andrades Filho, C. O., Mexias, L. F. S., Oliveira, G. G., Herrmann, P. B., Alves, F. C., and Glade, T.: Could such a large landslide event be expected in Rio Grande do Sul, southern Brazil? Using past events to predict the area impacted by the 2024 Mega Disaster, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-15304, https://doi.org/10.5194/egusphere-egu25-15304, 2025.

Estimating the potential runout distance of landslides and their associated impacted areas is a critical component of landslide hazard and risk analysis. Traditionally, back-analysis of past landslides has been employed to predict the runout behaviour of potential future events. To refine landslide runout models and characterize co-seismic landslide dynamics, we conducted an in-depth analysis of a subset of landslides triggered by the Mw 7.8 Kaikōura earthquake in New Zealand (14 November 2016), focusing on the Kowhai Valley in Kaikōura.

First, we mapped polylines connecting landslide sources to their corresponding deposits. Given that all landslides were triggered during the same seismic event within steep upland catchments, source areas did not consistently correspond directly to mapped debris trails. Second, we attributed these polylines with information on confinement, substrate type, connectivity, geometry, and physiographic attributes, analysing their relationships with travel length and fall height to identify controls on runout distance. Third, we applied three regional-scale runout modelling approaches—1) a Fahrböschung angle method, 2) the Gravitational Path Process Model, and 3) Flow-R—to evaluate their effectiveness in predicting travel distances and patterns of co-seismic landslide runout.

Our mapping identified 3,535 landslide polylines linking 3,105 source areas to 2,652 debris trails. Approximately two-thirds of the landslides exhibited a one-to-one relationship between source and deposit, while the remainder displayed more complex linkages, including multiple deposits from a single source, single deposits from multiple sources, or interactions involving multiple sources and deposits. Statistical analysis revealed significant relationships between runout distance and factors such as substrate type, confinement, coupling, and geometry, although no significant relationship was observed with landslide volume.

Model accuracy assessments, using goodness of fit metrics, showed that most approaches either displayed weak accuracy or overestimated landslide runout areas. The best fit models indicated that the landslides triggered in the Kaikōura earthquake travelled a shorter distance than expected from the international literature. Further analysis revealed considerable variability in model accuracy for individual landslides, with larger landslides showing better goodness-of-fit metrics than smaller ones. Landslides located in the lower reaches of the Kowhai Valley also demonstrated higher model accuracy, potentially as a function of landscape relief. These findings underscore the complex controls influencing co-seismic landslide runout and highlight the importance of accounting for uncertainties in regional-scale landslide runout models.

How to cite: de Vilder, S., Wolter, A., Lukovic, B., Leith, K., Hill, S., and Cox, S.: Runout characteristics of landslides triggered by the 2016 Kaikoura Earthquake, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-14509, https://doi.org/10.5194/egusphere-egu25-14509, 2025.

    Rapid, giant landslides, or sturzstroms, are among the most powerful natural hazards on Earth. They are produced by catastrophic, deep-seated slope collapses with minimum volumes on the order of 10⁶–10⁷ m³. Such collapses are often the final stage of accelerating slope movement that may have continued for years. Measurements made over 60 years ago before the failure of Mt. Toc into the Vajont reservoir in the Italian Alps remain one of the best records of pre-collapse slope movement. Numerous studies have recognized that the rate of movement increased hyperbolically during at least two months of heavy rainfall before the mountainside collapsed on 9 October 1963. Two hundred million m³ of rock sent a wave of water over the Vajont dam, killing approximately 2,500 people in the downstream communities of Longarone, Pirago, Villanova, Rivalta, and Fae. Analysis of the extended record shows that the hyperbolic trend was preceded by an exponential acceleration during 1962. The earlier trend was interrupted in December 1962 when the reservoir was temporarily drained to install engineering safety measures. The acceleration resumed in July–August 1963 after the reservoir was refilled to its pre-drainage level. This combined exponential-hyperbolic acceleration trend is consistent with the activation and eventual linkage of cracks along the future failure plane.This suggests that the surface movements were a consequence of fracturing as deep as 200 m underground, rather than cracking being a result of slope movement. This interpretation points to the weakening of deep rock as the primary driver of failure, caused by factors such as increases in pore water pressure and water-induced corrosion, rather than the destabilization from the weight of a water-saturated slope.Since rock cracking occurs within a restricted range of physical conditions, this case study demonstrates that medium-term forecasts of catastrophic slope failure are a feasible goal. By identifying and quantifying these conditions, we can advance predictive capabilities and mitigate the devastating impacts of rapid landslides, such as tsunamis, seismic shocks, and downstream flooding.

How to cite: Pan, F. and Kilburn, C. R. J.: Controls on rates of slope movement before catatsrophic collapse, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-5374, https://doi.org/10.5194/egusphere-egu25-5374, 2025.

Landslide failures pose a severe threat to society, especially when valley bottoms become blocked, ponding rivers and burying critical infrastructure. The erratic and spatially distributed occurrence of those rapid mass wasting processes makes it eminent to understand major drivers and find reliable predictors that can help early warning.

Here, we present results of a systematic study on a progressively developing landslide near the town of Müsch, in one of the narrowest sections of the Ahr Valley, Germany. The slope instability had been reactivated by the 2021 summer flood and shows accelerated toppling and rotational movement at the 100 m wide front, as well as surface evidence of distributed movement in the 200 m long hinterland. Partial failure of the frontal sector had been modelled, indicating the damming of the 30 m wide valley bottom, causing rapid inundation of upstream settlements.

We analyse 2.5 years of continuous seismic data from a small geophone network. Seismic coda wave interferometry and resonance frequency analysis yields insights to cyclic and progressive rock stress evolution as well as the effect of water content at and below the surface. More than 3000 discrete crack emissions due to brittle rock mass failure were detected, located and quantified. The precise timing of the crack signals reveals a strong control of working time hours, suggesting an external anthropogenic forcing of the slope instability. We discuss the generic applicability of the multi-proxy seismic approach in light of further, post-flood reactivations of slope instabilities in the Ahr Valley and elsewhere.

How to cite: Dietze, M., Fracica Gonzalez, L., Bell, R., Schrott, L., and Hovius, N.: Strong daily landslide cracking activity – Does traffic drive slope failure?, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-8438, https://doi.org/10.5194/egusphere-egu25-8438, 2025.

Landslide Susceptibility Modeling (LSM) is an important method for mitigating regional landslide risks. However, the scarcity of landslide inventories and the prevalence of low-quality non-landslide samples significantly limit the further development of traditional LSM frameworks. To address this issue, this paper develops a next generation of LSM framework that redefines landslide and non-landslide samples from the perspective of deformation. By integrating deformation time-series data from Global Navigation Satellite System (GNSS) and InSAR, the framework introduces deformation samples defined by deformation rates and obtains a greater number of landslide samples and high-quality non-landslide samples through the establishment of appropriate deformation thresholds. A series of ablation experiments were conducted in Wanzhou District, Chongqing, China. The results indicate that when the deformation threshold is set to 0.6, the proposed LSM framework achieves an AUC value of 0.94, a TPR of 0.92, and a TNR of 0.94, representing a significant improvement compared to the traditional LSM framework (AUC = 0.85, TPR = 0.74, TNR = 0.58). Additionally, the experimental results demonstrate that when using InSAR data to obtain deformation samples, either a large number of low-quality InSAR deformation samples or a small number of high-quality but spatially uneven InSAR deformation samples can result in the proposed LSM framework performing worse than the traditional LSM framework. Therefore, special attention must be paid to balancing the quality and quantity of InSAR data.

How to cite: Feng, X., Du, J., Chai, B., and Bogaard, T.: Combining landslide inventories with deformation time series: A methodology, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-8480, https://doi.org/10.5194/egusphere-egu25-8480, 2025.

Landslide inventories are fundamental for susceptibility mapping, hazard modeling, and risk management. For decades, the geoscientific community has relied on manual visual interpretation of satellite and aerial imagery for landslide inventory generation. However, manual methods pose significant challenges, including subjectivity in landslide boundary delineation, limited data sharing within the scientific community, and the substantial time and expertise required for accurate mapping. Recent advancements in artificial intelligence (AI) have spurred a surge in research on semi-automated and fully automated landslide inventory mapping. Despite this progress, AI-generated inventories remain in their developmental phase, with no existing models capable of consistently producing ground-truth representations of landslide events following a triggering event. Current studies utilizing AI-based models report F1-scores ranging between 50% and 80%, with only a few achieving over 80%, often limited to the same study areas used for model training. This highlights a significant research gap in the reliability and generalizability of AI-generated inventories for hazard and risk assessments. The geoscientific community must critically assess the accuracy and transferability of AI-generated landslide data to ensure their applicability in subsequent phases of landslide response and mitigation. Further collaborative efforts and benchmark datasets are needed to establish standardized protocols for validating AI-generated landslide inventories across diverse geomorphological settings.

How to cite: Meena, S. R., Singh, S., Bhookya, R., and Floris, M.: Evaluating the Potential of AI-Generated Landslide Inventories for Hazard and Risk Management: Advancements and Limitations, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17609, https://doi.org/10.5194/egusphere-egu25-17609, 2025.

Forecasting catastrophic slope failures is one of the most challenging tasks in landslide hazard analysis. Reliable landslide forecast is essential for civil authorities to effectively inform the public about potential mountain collapses and their timing, facilitating timely evacuations and the implementation of other safety measures. Over the past decades, great efforts have been devoted to develop and deploy high-precision monitoring technologies to observe unstable slope movements. Various empirical or physical approaches have also been proposed to forecast imminent slope collapses, the predictability of which, however, still remains elusive. One major uncertainty arises from the intermittency of geomaterial rupture behaviour, which is typically characterised by a series of progressively shorter quiescent phases interrupted by sudden accelerations, rather than a smooth continuous progression of deformation and damage. This seemingly erratic pattern complicates landslide prediction. Here, we propose a generalised failure law based on the log-periodic power law singularity model for more reliable time-to-failure forecast of catastrophic landslides. Incorporating a discrete hierarchy of time scales and rooted in the fundamental principles of statistical physics, this novel failure law accurately captures the intermittent rupture dynamics of heterogeneous geomaterials at the site scale. It ensures robustness while maintaining a strong connection to the underlying physical processes. By "locking" into the oscillatory structure of rupture dynamics, this parsimonious model transforms intermittency from traditionally perceived noise into essential information to constrain its prediction. We extensively validate this new failure law on a large dataset of 49 historical landslide events, across a wide range of contexts including rockfalls, rockslides, clayslides, and embankment slopes. The results indicate that our method is general and robust, with significant potential to mitigate landslide hazards and enhance existing early warning systems.

How to cite: Lei, Q. and Sornette, D.: A physics-based generalised failure law for forecasting catastrophic landslides, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18378, https://doi.org/10.5194/egusphere-egu25-18378, 2025.

Forecasting landslides induced by rainfall is a challenging task that involves the interaction of multiple factors, such as soil conditions, topography, and rainfall intensity. The complex nature of these events, combined with the lack of complete data on landslide occurrences, makes it difficult to produce accurate predictions. Traditional deterministic models struggle to account for the variability and uncertainty inherent in the processes leading to landslides. On the contrary, probabilistic approaches can incorporate uncertainty and provide more reliable description of this phenomenon. In this work we develop a probabilistic framework for forecasting rainfall-induced landslide occurrence addressing the challenges of data sampling, uncertainty, and variability. For the study, we collected a dataset of shallow rainfall-induced landslides in an area in southern Italy, spanning 22 years of rainfall records. The dataset includes the locations and date of occurrences of the landslides, and daily rainfall measurements. Using a Bayesian approach, we calculate the posterior probability of landslide occurrence given specific daily cumulated rainfall thresholds. To account for the uncertainty in the landslide and rainfall data, we employed probabilistic distributions i.e., uniform and beta distributions, to model the uncertainty in the prior and likelihood functions. The uncertainty was further addressed through random sampling techniques, allowing for the integration of data variability and the dependencies between landslides and rainfall, obtaining posterior probability distributions of landslide occurrence for each rainfall threshold. The results offer a probabilistic approach to landslide forecasting that can be used for better-informed decision-making in risk management and early warning systems. By accounting for the uncertainties in the data and model parameters, our approach provides a more robust method for landslide prediction under varying rainfall conditions. 

How to cite: Ferriero, F., Guzzetti, F., Urciuoli, G., and Marzocchi, W.: Bayesian probabilistic forecasting of rainfall-induced landslides, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-10443, https://doi.org/10.5194/egusphere-egu25-10443, 2025.

Landslides triggered by earthquakes evolve over time, leading to repeated damage in the affected areas. These slope movements are influenced by a range of factors, including climatic, seismic, and terrain conditions, which vary both temporally and spatially [1]. To predict the likelihood of landslides occurring across different times and locations, statistical models must account for these spatial and temporal dependencies. In this study, we employ the Markov Switching Spatiotemporal Generalized Additive Model (MSST-GAM), as introduced by Sridharan et al. [2]. Their research highlighted how this model effectively captures the spatial and temporal influences of various landslide-related factors, offering accurate susceptibility estimates for the Wenchuan area in China.

In this work, we further extend the model for hazard prediction. The model is used on a multitemporal dataset of landslides that occurred in New Zealand during and following the 2016 Kaikoura earthquake [3]. The years in which the landslides were mapped were used to separate the temporal units. Twelve covariates were used, including terrain (slope, aspect, curvature, distance from features like faults, etc.), climatic (rainfall and soil moisture), and seismic (when the year coincided with a major seismic event). We employ zero-inflated Poisson and Gaussian emission probabilities [4] for the dependent variables, which are the areas and counts of landslides in slope units. A Markov-switching GAM is used to predict the dependent variables from the covariables based on two hidden risk states (high risk and low risk). We introduce soil moisture as an additional dynamic variable to parametrize the transition probabilities between the hidden states. 

We tested the model using a five-fold spatiotemporal cross-validation. The results compare favourably to a number of cross-sectional models [5]. The model predictions indicate that MSST-GAM can capture the spatial and temporal dependence of the landslide occurrences in slope units when compared with other cross-sectional and spatiotemporal models in literature.

References

[1] Keefer, D., “Investigating landslides caused by earthquakes - A historical review,” Surv. Geophys., vol. 23, no. 6, pp. 473–510, 2002 

[2] Sridharan, A., Gutjahr, G., and Gopalan, S., “Markov–Switching Spatio–Temporal Generalized Additive Model for Landslide Susceptibility,” Environ. Model. Softw., vol. 173, no. August, p. 105892, Feb. 2024 

[3]  Bhuyan, K., Tanyaş, H., Nava, L. et al. “Generating multi-temporal landslide inventories through a general deep transfer learning strategy using HR EO data”. Sci Rep 13, 162, 2023

[4] Wagh, Y.S. and Kamalja, K.K., 2018. “Zero-inflated models and estimation in zero-inflated Poisson distribution”. Communications in Statistics-Simulation and Computation, 47(8), pp.2248-2265.

[5] Reichenbach, P., Rossi, M., Malamud, B., Mihir, M., Guzzetti, F. “A review of statistically-based landslide susceptibility models”. Earth-science reviews. 2018 May 1;180:60-91.

How to cite: Gutjahr, G., Sridharan, A., and Gopalan, S.: A Markov Switching Spatiotemporal GAM for Landslide Hazards in New Zealand, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12309, https://doi.org/10.5194/egusphere-egu25-12309, 2025.

Rock avalanches are large volume landslides composed of flowing fragments of rock that can reach velocities in excess of 50 m/s, impact large areas, and can seriously threaten the safety of people and infrastructure. Numerical models play a crucial role in forecasting the hazard and risk associated with rock avalanches. The Orin3D model, based on the equivalent fluid concept, can be used to simulate rock avalanche motion, however it is unknown what the best model parameterization is for forecasting.  However, Orin3D is implemented to run on a graphical processing unit (GPU), which improves simulation times by two orders of magnitude, making large-scale calibration feasible, as is investigated herein. 
In the present work, we use a posterior analysis based on Bayesian statistics to calibrate Orin3D for three different parameterizations: 1) Frictional rheology, 2) Voellmy rheology, and 3) the combination of Frictional and Voellmy rheology, using a data set containing 22 historical rock avalanche cases, and requiring over 450,000 model runs. Based on the calibration results, a probabilistic prediction framework is then tested that generates pseudo-predictions for the cases in the database, incorporating key features of rock avalanches, such as path materials and topographic constraints. We find that, among these three rheological settings, the best prediction results for most cases are obtained with the combination of Frictional and Voellmy rheology. We further use these results to suggest a prediction procedure that considers the volume, path material and topographic confinements of rock avalanches, which provide guidance for the rheological setting in the model and important basis for the prediction and mitigation of rock avalanche hazards in practice.

How to cite: Deng, G. and Aaron, J.: Calibration and prediction procedure of rock avalanche through advancing numerical simulation, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-8845, https://doi.org/10.5194/egusphere-egu25-8845, 2025.

The field of landslide susceptibility modelling has seen the adoption of many different data-driven approaches, spanning from linear models to the most recent deep-learning solutions. In short, simpler models offer greater interpretability, while predictions derived from complex architectures are more difficult to explain. For this reason, complex algorithms are often referred to as black-box models. However, in the context of landslide susceptibility mapping, the ability to provide highly accurate results along with interpretable predictions is highly valuable. In light of these considerations, this study presents a landslide susceptibility mapping by exploring the capabilities of a new generation of interpretable models, namely Explainable Boosting Machines (EBMs). Unlike the majority of explainable approaches that unveil the decisions of a complex model in a post-processing phase, EBMs offer direct interpretability and full transparency. As a consequence, EBMs fall into the category of glass-box models. Notably, the incorporation of these models within studies focusing on the relationship between landslide occurrence and extreme rainfall events raises considerable interest and represents the aim of this work. Therefore, this contribution focuses on landslides triggered by a heavy rainfall event on September 15, 2022, in Central Italy. To analyze the interaction between landslide occurrence and the event, a novel rainfall variable is introduced among the set of predictors, capturing the event’s intensity relative to historical rainfall patterns. Specifically, this rainfall variable is computed as the percentage of precipitation attributed to the event compared to the mean annual rainfall. The rainfall variable also introduces a dynamic component to the proposed modelling, since it may vary at every future rainfall event. As a consequence, by combining the dynamic nature of the rainfall variable with the exact intelligibility of EBMs, the study also presents a landslide susceptibility mapping under potentially different rainfall scenarios with respect to the September 15, 2022 event.

How to cite: Caleca, F., Confuorto, P., Raspini, F., Segoni, S., Tofani, V., Casagli, N., and Moretti, S.: Modelling landslide susceptibility through a glass-box machine learning , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-5730, https://doi.org/10.5194/egusphere-egu25-5730, 2025.

Shallow landslides are frequently observed at natural slopes and often lead to more destruction through flow-like disasters. Traditionally, physically-based landslide susceptibility models utilized infinite slope stability analysis to determine slope stability in terms of factor of safety (FS) over regional scales. Although the infinite slope model is computationally less demanding, it cannot account for the spatial variability of soil properties and the three-dimensional (3D) effects arising from complex topography. However, using 3D slope stability models is computationally demanding and suffers from discontinuity introduced by abrupt changes in soil thickness. Therefore, this research proposes a new Three-Dimensional Translational Shallow (3DTS) slope stability model to overcome these drawbacks of the existing models with complex 3D sliding surfaces.

The developed 3DTS model utilizes the Green-Ampt (GA) infiltration model and the 3D extension of the Janbu simplified method of slope stability. The 3DTS utilizes a generalized GA model to account for non-uniform infiltration history and compute the surface runoff. In 3D limit equilibrium slope models, the failing soil mass must be subdivided into rigid soil columns; however, the developed 3DTS uses the cells from a digital elevation model (DEM) as the rigid soil columns. The shear strength, modeled with the Mohr-Coulomb criterion, is provided by the soil frictional resistance on the base and the side regions of the outermost soil columns. Additional strength from the vegetation roots at the shallow surfaces is modeled.

The method used in the developed 3DTS model for generating slip surfaces from DEM cells was verified by comparing computed FS with the 3-Dimensional Probabilistic Landslide Susceptibility (3DPLS) model, which uses ellipsoidal slip surfaces. A parametric study analyzed the sensitivity of the slip surface's shape, the side soil resistance, and the vegetation resistance to shallow translational failures. The applicability and computational efficiency of the developed 3DTS for large-scale landslide susceptibility assessment were demonstrated by analyzing landslide case studies in Norway and South Korea.

This work is the result of collaboration between Norwegian Geotechnical Institute (NGI) and Korea Advanced Institute of Science and Technology (KAIST) through the project GEOMME (2021-2026; Pnr. 322469), “Climate-induced geohazards mitigation, management, and education in Japan, South Korea, and Norway”, supported by the Research Council of Norway.

How to cite: Cheon, E., Ahmet Oguz, E., DiBiagio, A., Piciullo, L., Kwon, T. H., and Lee, S. R.: A Physically-based 3D Landslide Susceptibility Model for Shallow Translational Landslides using DEM, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17059, https://doi.org/10.5194/egusphere-egu25-17059, 2025.

Landslides pose significant threats to human lives and economies, with their frequency and intensity increasingly exacerbated by climate change. This was demonstrated in May 2023 in the Emilia-Romagna region (northern Italy), where 80,000 landslides were triggered by two rare, high-intensity rainfall events with a return period of 300 years, occurring just 14 days apart. The landslides exhibited diverse types and materials, necessitating tailored risk management approaches due to differences in volume, velocity, and post-event behaviour. To address these complexities, susceptibility maps must integrate both static predictors and dynamic triggering factors to better understand the relationships between rainfall and landslide types.

Using a detailed landslide inventory developed through collaboration between the Emilia-Romagna Geological Service and the universities of Modena and Bologna, we analysed the relationship between rainfall and five distinct mapped landslide types: debris slide, debris flow, earth slide, earth flow, and rock slide. This study introduces a Transformer Neural Network (TNN) to integrate static predictors (e.g., slope, aspect, geology, land cover) with dynamic rainfall data from the 30 days preceding the second rainfall event (16^th May), capturing the influence of antecedent wet/dry conditions. The TNN processes rainfall time series data similarly to speech recognition algorithms, allowing it to model temporal dependencies effectively.

We evaluated the TNN with rainfall data at different temporal resolutions (daily and hourly intervals) and compared its performance against models using only static predictors or cumulative rainfall. The TNN was trained on 70% of the dataset, targeting specific landslide types to generate susceptibility maps tailored for each type. Model performance was assessed using a comprehensive set of metrics, including Area Under the Curve (AUC), Accuracy, Recall, F1 and F2 scores, Matthew’s Correlation Coefficient, and Kappa Coefficient. Additionally, we applied the SHAP (Gradient Explained) method to analyse the influence of rainfall on susceptibility values, revealing the model's internal decision-making processes.

The results demonstrate that integrating rainfall time series significantly enhances susceptibility mapping accuracy. The TNN using daily rainfall data produced the most reliable maps for all landslide types, except debris flows, where hourly intervals yielded slightly better results. SHAP analysis further illuminated the role of rainfall in susceptibility variations, providing valuable insights into the TNN's functionality. Overall, the TNN outperformed models using only static predictors or cumulative rainfall, offering a robust framework for understanding and predicting landslide susceptibility in diverse scenarios.

How to cite: Rani, R., Dahal, A., Lombardo, L., Tanyas, H., and Berti, M.: Unveiling the Complex Relationship Between Rainfall and Landslide Types Using a Transformer Neural Network, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6943, https://doi.org/10.5194/egusphere-egu25-6943, 2025.

Landslide susceptibility is traditionally determined by analyzing various topographic, geological, and hydrological factors, which influence the probability of landslide occurrence. Recent research in Italy and Nepal has shown that landslide susceptibility is also controlled by landslide path dependency (LPD), where previous landslides locally and temporally influence the future landslide susceptibility. Our study focusses on Collazzone (Italy), a region predominantly affected by shallow landslides, supported by multi-temporal landslide inventory from 1939 to 2014. Here, we are comparing the impact of earlier landslides on landslide susceptibility in the downslope and lateral directions. We hypothesize that the LPD has more impact in downslope direction than in lateral direction due to the crucial role played by formation of positive feedback loop of the soil-landslide system. In the downslope direction, landslides can create weakly permeable soil layers that increase the water saturation, thus increasing the probability of subsequent landslides. In contrast, the lateral direction lacks this feedback mechanism, making subsequent landslides less likely.

For testing our hypothesis, we used simulated annealing to make artificial landslide inventories which approximate the real landslide inventory in terms of topographic positioning, but that lack any LPD. After that we calculate Ripley’s K by using a space-time cuboid for these control inventories and for the real inventory. Generalized additive models (GAM) were used to analyze the ratio between real and control Ripley’s K values. GAM results indicate that there is a nonlinear relationship between ratio of real to control Ripley’s K and time difference (dT), lateral (dL) and downslope distance (dD) between consecutive landslides. This ratio has a negative relationship with dL, and dD, while the relationship with dT is weak. Moreover, we found that earlier landslides have a stronger impact on future occurrence of landslides in downslope direction than in lateral direction. Our results provide clear evidence that downslope direction plays a significant role in landslide path dependency.

How to cite: Sodhi, H. S., Temme, A., Samia, J., Rossi, M., and Ardizzone, F.: Comparing the strength of Landslide Path Dependency in the downslope and lateral directions using simulated annealing and space-time clustering , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-4137, https://doi.org/10.5194/egusphere-egu25-4137, 2025.

Landslide susceptibility assessment at scales wider than the single slope has been so far carried out mainly through heuristical/geomorphological and/or statistical methods, except for applications limited to shallow landslide predictions by means of infinite-slope limit equilibrium models (Godt et al. 2008). Owing to the complexity of developing quantitative deterministic susceptibility models at wide scales, taking into account also deep and complex landslide mechanisms, the application of limit equilibrium methods as well as numerical stress-strain methods have been historically limited to the scale of the single slope. However, the increased availability of powerful computational tools as well as the existence of detailed geological and geotechnical databases at scales that are intermediate between the single slope and the regional scales, as for example the scale of a single urban centre, allow for extending the application of three-dimensional limit equilibrium analysis to the assessment of landslide susceptibility at such scale, also taking into account the failure susceptibility of deep and complex landslide mechanisms. This contribution presents a physically-based methodology aimed at assessing landslide susceptibility at the urban area scale, for both shallow and deep instability processes involving urbanized areas that are diffusely affected by landsliding processes (Ugenti et al. 2025). The proposed methodology has been applied to the municipality of Carlantino (Daunia Apennines, Southern Italy) as a test case study, using the available geological and geomorphological datasets as well as the soil geotechnical property data. Based on a three-dimensional geotechnical model, 2.5 km² wide, a three-dimensional limit equilibrium analysis has been develop to obtain a mechanically-based map of the safety factors at the urban area scale, assuming different scenarios related to the groundwater table depth, which has been validated against geomorphological evidence and remote sensing data. The proposed approach, which is supposed to be exportable to other geological environments, provides a valuable tool for quantitative assessment of the slope stability conditions of an overall urban area to be used for a more rational approach of urban planning policies and risk management activities.

References:

Godt J.W., Baum R.L., Savage W.Z., Salciarini D., Schulz W.Z., Harp  E.L. (2008). Transient deterministic shallow landslide modeling: requirements for susceptibility and hazard assessments in a GIS framework. Engineering Geology, 102 (3-4), 214-226.

Ugenti A., Mevoli F.A., de Lucia D., Lollino P., Fazio N.L. (2025). Moving beyond single slope quantitative analysis: a 3D slope stability assessment at urban scale. Engineering Geology, 344, 107841, doi: 10.1016/j.enggeo.2024.107841.

How to cite: Lollino, P., Ugenti, A., Mevoli, F. A., de Lucia, D., and Fazio, N. L.: 3D limit equilibrium analysis: an opportunity for quantitative landslide susceptibility assessment at the scale of the urban area, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-16171, https://doi.org/10.5194/egusphere-egu25-16171, 2025.

The increase in localized heavy rainfall and intense storms due to climate change has led to a continuous rise in landslide damages in South Korea, including slope failures and debris flows. While post-landslide recovery and damage site assessments are crucial, it is equally important to develop proactive and systematic landslide adaptation strategies to predict and prepare for landslides in advance. This study aims to develop an interpretable machine learning-based landslide susceptibility model and analyze landslide-prone areas under future climate change scenarios. Through this approach, it seeks to clearly identify the impact of forest management factors on landslides and establish effective adaptation strategies tailored to climate change scenarios. A dataset comprising 6,517 recorded landslide events from 2011 to 2024 was utilized. Various external and internal conditioning factors were obtained and constructed with a resolution of 100 meters. Climate scenario analysis employed SSP 1-2.6, SSP 2-4.5, and SSP 5-8.5, with extreme climate factors including rainfall intensity, the number of heavy rain days, daily rainfall, and 5day cumulative rainfall. Notably, changes in stand age class, DBH class, and growing stock under future forest management scenarios were calculated and integrated into the landslide model, enabling an evaluation of how management strategies affect landslide susceptibility. Results were validated by comparing past actual occurrence data. The SSP 5-8.5 scenario indicates a significant increase in landslide occurrences. These findings provide valuable insights into the effects of climate change on landslide susceptibility in South Korea and examine the potential of future landslide management strategies to mitigate associated susceptibility. 

How to cite: Kim, U., Lee, S., Roh, M., Kim, S., and Lee, W.: Spatial Prediction of the Future Landslide Susceptibility under the SSP Scenario Using Machine learning Algorithms, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-14735, https://doi.org/10.5194/egusphere-egu25-14735, 2025.

An active-layer detachment slide (ALDS) occurred on September 21, 2018, in the Fenghuoshan mountains of the Qinghai-Tibet Plateau (QTP) (34◦39.1′N, 92◦53.5′E). With the Sentinel-1A image from Copernicus Open Access Hub, we use small baseline subset to achieve the time series deformation map to analyze the thermo-spatial creep feature, motion pattern, trigger mechanism, and correlation of environmental changes in the ALDS. The SBAS (the Small Baselines Subset) results show that the trailing part of ALDS has the largest downward deformation rate; however, the leading area was small, and the creep feature shows a clear seasonal change corresponding to the freeze-thaw cycle. We also divide the motion pattern into three stages: moderate creep, steady creep, and rapid collapse, based on the deformation rate. Meteorological observation and reanalysis data, as well as borehole data, show that heavy precipitation in the summer of 2017 and 2018 promote the formation of underground ice, while high air temperatures allow the thaw plane to reach the ice-rich zone, and confined water generated by the two-way freezing process result in ALDS. Moreover, there exists a temporal delay of approximately one month in the association between deformation rate and both precipitation and temperature. Furthermore, there is a clear correlation between variations in thawing depth and deformation, which serves as the primary catalyst for ALDS in permafrost regions. Finally, we also identify that ALDS is a mixed-type landslide and that cumulative deformation and creep damage play the main roles in triggering ALDS. 

How to cite: wen, Z. and wang, F.: Creep features and mechanism of active-layer detachment slide on the Qinghai-Tibet Plateau by InSAR, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-2110, https://doi.org/10.5194/egusphere-egu25-2110, 2025.