This session revolves around methodologies and state-of-the-art approaches in landslide prediction, encompassing aspects like location, timing, magnitude, and the impact of single and multiple slope failures. It spans a range of landslide variations, from abrupt rockfalls to rapid debris flows, and slow-moving slides to sudden rock avalanches. The focus extends from local to global scales.
Contributions are encouraged in the following areas:
Exploring the theoretical facets of predicting natural hazards, with a specific emphasis on landslide prognosis. These submissions may delve into conceptual, mathematical, physical, statistical, numerical, and computational intricacies.
Presenting applied research, supported by real-world instances, that assesses the feasibility of predicting individual or multiple landslides and their defining characteristics, with specific reference to early warning systems and methods based on monitoring data and time series of physical quantities related to slope stability at different scales.
Evaluating the precision of landslide forecasts, comparing the effectiveness of diverse predictive models, demonstrating the integration of landslide predictions into operational systems, and probing the potential of emerging technologies.
Should the session yield fruitful results, noteworthy submissions may be consolidated into a special issue of an international journal.
Geohazard mass flow runout prediction is critical for protecting lives, infrastructure, and ecosystems. Rapid mass flows such as landslides, and avalanches are among the most destructive geohazards, often travelling many kilometres from their source. Uncertain initial conditions and strong sensitivity to topography make these events difficult to anticipate, particularly for downstream communities that may be exposed to severe impacts with little warning. In this context, computational speed is essential for enabling timely forecasting and scenario-based risk assessment.
Accurately predicting runout requires models that are both physically realistic and computationally efficient. However, existing approaches face a fundamental trade-off between realism and speed, limiting their use for large-scale forecasting, ensemble analysis, and operational early warning.
Here we demonstrate that neural networks can emulate the final outcomes of mass flow runouts across diverse real-world terrains. Our model is trained on approximately 90,000 high-fidelity simulations spanning more than 5,000 globally representative topographies. The model predicts both flow extent and deposit thickness with high spatial accuracy while achieving computation speeds orders of magnitude faster than numerical solvers. Importantly, the emulator reproduces key emergent physical behaviours, including avulsion and heterogeneous deposition patterns, and generalizes across a wide range of rheologies, volumes, and terrain types. Probabilistic outputs further enable scalable uncertainty quantification.
These results show that data-driven emulation can shift geohazard runout forecasting from site-specific analysis towards rapid prediction frameworks, supporting impact-based early warning and regional-scale hazard assessment. We anticipate that this approach will form a foundation for next-generation forecasting models that integrate physical simulation and machine learning to address transient dynamics, multi-hazard interactions, and cascading effects relevant to landslide hazard forecasting in space and time.
How to cite: Nava, L., Chen, Y., and Van Wyk de Vries, M.: Mass flow runout prediction using neural network emulators, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-4676, https://doi.org/10.5194/egusphere-egu26-4676, 2026.
Landslides are among the most frequent natural hazards worldwide, significantly threatening human life, infrastructure, and ecosystems. Identifying areas prone to slope failures is therefore essential for effective land management, particularly under changing climatic conditions. This study develops a robust statistical model for assessing shallow landslide susceptibility at the slope-unit level across the Aosta Valley, while explicitly evaluating the role of spatial autocorrelation. A comprehensive shallow landslide inventory, compiled by integrating the Italian Landslide Inventory (IFFI) database with the Regional Inventory of Instabilities of Aosta Valley, was used as the binary response variable indicating shallow landslide occurrences. A broad set of geo-environmental predictors was assembled and optimised through a novel structured variable selection workflow, combining multicollinearity analysis, stepwise selection, Random Forest classification, and Generalised Additive Models (GAMs). GAMs were used for modelling susceptibility and exploring predictor–response relationships via smoothing functions. To assess spatial autocorrelation effects, the coordinates of slope-unit centroids were incorporated into the GAM framework using a tensor-product smooth. This resulted in two models: model_A, excluding the spatial term, and model_B, including it. Model performance was evaluated using spatial and non-spatial k-fold cross-validation, assessed through mean Decrease in Deviance explained (mDD%), Effective Degrees of Freedom (EDF), and Area Under the Receiving Operating Characteristic curve (AUROC). Both models are statistically significant and exhibit high discriminatory power (AUROC > 0.85) under both validation schemes. Including the spatial tensor modestly improved model fit and predictive capacity for model_B relative to model_A, with higher deviance explained (39.0 vs. 35.9), R² (0.42 vs. 0.39), and lower AIC (714.2 vs. 724.5). Distributions of mDD% and EDF indicate greater stability for model_B, whereas model_A shows higher variability. However, the improved training performance of model_B likely reflects sensitivity to local spatial structure rather than enhanced generalisation. Under spatial cross-validation, testing performance decreases relative to non-spatial validation and becomes variable for both models, while the performance gap between model_A and model_B narrows (testing AUROC: 0.877 vs. 0.890; training AUROC: 0.854 vs. 0.856), highlighting the influence of spatial partitioning and the limited generalisation gains once spatial dependence is accounted for. Model predictions were used to generate shallow landslide susceptibility maps for the Aosta Valley. Although both models assign similar proportions of slope units to each susceptibility class, notable differences emerge in their spatial distribution, with class-specific discrepancies reaching up to 30%. Standard error analysis shows that the model including spatial tensor does not uniformly improve prediction confidence: uncertainty is reduced only in spatially clustered areas with potentially homogeneous geomorphological conditions and worsens elsewhere. This confirms a spatially selective benefit due to the inclusion of the spatial tensor, along with its limited contribution to the overall spatial generalisation. Landslide density patterns across susceptibility classes are consistent between training and testing subsets, supporting the robustness of the classification framework. Model_B yields slightly higher densities in the highest susceptibility class, whereas calibration analysis indicates marginally better probabilistic accuracy and stability for model_A. Overall, both models provide comparable and reliable representations of landslide susceptibility, revealing a trade-off between spatial sensitivity and calibration performance.
How to cite: Pompili, L., Camera, C. A. S., Sorichetta, A., Economou, T., Bondarenko, M., and Yankey, O.: Testing a Coordinate Tensor-Product Descriptor for Spatial Autocorrelation in a Shallow Landslide Susceptibility Model, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-5465, https://doi.org/10.5194/egusphere-egu26-5465, 2026.
Earthquake-triggered landslides are among the most destructive secondary seismic hazards, yet their rapid prediction at global scale remains elusive due to the limitations of existing physical and statistical models. Current approaches typically depend on regional inventories, simplified assumptions, or retrospective calibration, preventing timely and reliable assessments immediately after large earthquakes. To address this gap, we compiled the largest global database to date of ~400,000 coseismic landslides from 38 major earthquakes spanning diverse tectonic and climatic settings. Using this unified dataset, we developed a multi-scale fully convolutional deep-learning framework capable of predicting coseismic landslide probability worldwide with no prior local labels.
The model integrates 14 primary control indicators, representing topography, geo-ecology, hydrology and seismology, and learns nonlinear relationships governing slope failure across global environments. Independent testing shows that the global model achieves an average AUC of ~0.83 and spatial accuracy of ~0.77, while regional models trained within specific environmental domains achieve slightly higher performance. The predictions successfully reproduce both the extent and spatial pattern of landslides for events such as the 2015 Gorkha, 2016 Kaikoura, 2021 Nippes, 2022 Luding and 2002 Denali earthquakes. Sensitivity analyses further demonstrate that model performance is robust to inventory uncertainty but strongly influenced by the quality of input seismic and fault data.
Our framework predicts landslide probability for a new earthquake in less than one minute, enabling actionable early hazard intelligence well before cloud-free satellite imagery becomes available. A hypothetical Mw 7.5 earthquake scenario in Sichuan, China illustrates that rapid prediction can identify high-impact areas and populations exposed to landslide cascades within seconds. This study establishes the first globally scalable and operational deep-learning model for earthquake-triggered landslide prediction, offering transformative potential for rapid hazard response, seismic risk management, and global multi-hazard preparedness.
How to cite: Wang, X., Fan, X., Fang, C., and Dai, L.: A Unified Deep Learning Framework for Rapid Global Prediction of Coseismic Landslides, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-1514, https://doi.org/10.5194/egusphere-egu26-1514, 2026.
Extreme rainfall in the granitic hilly region of southeastern China often triggers clustered shallow landslides characterized by strong spatial concentration, high density, and near-synchronous occurrence, while exhibiting pronounced differences in failure type. Yet, a unified explanation for why failures aggregate and how different types evolve during the same storm remains limited. Focusing on Xiaba Township as a case study, this work investigates the key predisposing controls and triggering processes of rainfall-induced clustered landslides. Field surveys and geomorphic interpretation indicate that, above shallow surficial residual and weathered layers, the coupled effects of lithology, landform morphology, flow accumulation/convergence, vegetation, and related factors form a shallow, continuous landslide-prone strata (LPS) that is readily mobilized under heavy rainfall, making accurate prediction of LPS burial depth practically important. We compile a point dataset of LPS burial depth from numerous observed landslides and propose a Random-Forest–based ensemble regression framework to address label scarcity and imbalance, spatial autocorrelation, observational noise, and the lack of interpretable uncertainty in conventional approaches. Spatially blocked cross-validation paired with grouped bootstrap resampling, together with robust standardization, mild resampling, and sample weighting, improves the model’s ability to characterize scarce yet critical depth intervals. At inference, a multi-submodel ensemble with Monte Carlo input perturbations yields the median LPS depth and an accompanying uncertainty metric; exceedance-probability curves are used to quantify how predictors alter the probability of surpassing specified depth thresholds. On the validation set, the model achieves Pearson’s r = 0.587, MAE = 0.281 m, RMSE = 0.411 m, and Lin’s CCC = 0.5065, capturing the spatial pattern of LPS burial depth reasonably well; Bland–Altman analysis indicates limits of agreement of about ±0.8 m, mainly at extremes. To link the predicted LPS depth field to geomorphic processes and clustered-failure behavior, we derive ridge-line cross-section metrics from a high-resolution DEM and find that landslides preferentially occur on ridges with larger deflection angles and steeper slopes, with many sites showing signatures of historical reactivation. Spatial topological descriptors of landslide boundaries capture systematic differences between planar- and convergent-type failures and enable robust classification. Building on these insights, we develop a rainfall infiltration–hillslope runoff model that explicitly incorporates geomorphic convergence and apply it to the 16 June storm. Simulations suggest that failures cluster where the LPS approaches saturation and local convergence is high; planar-type landslides activate in a quasi-linear cumulative manner, whereas convergent-type landslides require longer preconditioning before failing abruptly under sustained rainfall. Overall, this field–data–process framework balances accuracy and robustness under imbalance and noise, provides regional LPS-depth mapping with uncertainty, and offers a physically based foundation and parameter constraints for dynamic prediction of clustered landslide risk in granitic hilly terrains.
How to cite: Luo, S., Huang, Y., Mao, W., Meena, S. R., and Floris, M.: Preconditioning Mechanisms and Triggering Processes of Rainfall-Induced Clustered Landslides Controlled by the Coupling Between Landslide-Prone Strata and Micro-Geomorphology, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-2464, https://doi.org/10.5194/egusphere-egu26-2464, 2026.
Against the global backdrop of transitioning to clean energy, China has established the world's largest clean energy power transmission network. However, the stable operation of these clean energy transmission networks is increasingly threatened by landslides under extreme climatic conditions. Given the current lack of clarity regarding the extent of landslide impacts on power transmission lines, it is crucial to systematically assess the potential dynamic spatiotemporal distribution of landslide susceptibility. This study presents the first comprehensive dynamic spatiotemporal prediction of landslide susceptibility for transmission lines in China's loess region, highlighting the urgent need to enhance the resilience of transmission infrastructure in response to escalating extreme climatic events. To address this issue, a boosting ensemble framework was initially employed to construct a preliminary susceptibility model, incorporating comprehensive landslide inventory data and twelve influencing factors. Furthermore, MT-InSAR technology and the K-Means clustering algorithm were utilized to derive long-term surface deformation patterns from 2020 to 2024. Finally, the initial susceptibility assessment was refined by integrating deformation zoning based on slope units, generating the final landslide susceptibility map. The results demonstrate that the Categorical Boosting (CatBoost) model outperformed other methods within the boosting ensemble framework (AUC = 0.914). MT-InSAR analysis revealed a maximum deformation rate of 77 mm/year in the study area, with a cumulative displacement of 373 mm. Time-series deformation clustering further indicated that regions dominated by the second deformation pattern were most prevalent. The enhanced matrix incorporating time-series deformation clusters modified the initial assessment by reclassifying slope units from "very high" susceptibility, resulting in a net reduction from 1,496 units to 394 units—a decrease of 1,102 units. This study refines traditional landslide susceptibility models by incorporating diverse surface deformation trends, thereby addressing the risk overestimation inherent in static models and supporting more precise disaster mitigation along transmission lines.
How to cite: Jin, B., Gui, L., and Yin, K.: Spatiotemporal modeling framework for landslide susceptibility assessment along Clean Energy Transmission Corridors, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-6909, https://doi.org/10.5194/egusphere-egu26-6909, 2026.
Landslides affecting natural and engineered slopes pose a growing challenge for disaster risk reduction, particularly under the increasing frequency and intensity of rainfall and snowmelt events driven by climate change. Operational slope stability forecasting requires the integration of meteo-hydro-geological data sources, physical understanding of failure mechanisms, and frameworks capable of delivering timely predictions. This abstract summarizes our research activities of creating an integrated real-time cloud-based operational framework that combines slope-and regional-scales digital twins for landslide forecasting, leveraging real-time monitoring, numerical modelling, and data-driven methods.
At the regional scale, slope stability forecasting is addressed through a hybrid methodology that merges physically-based infinite slope models with data-driven landslide susceptibility and probability models (Abraham et al., 2025). The regional framework operates across first-order catchments within a selected study area in Norway. A physically-based model computes pixel-wise Factor of Safety (FoS) values using precipitation, topography, and subsurface parameters, calibrated through back-analysis and applied in forward forecasting model. In parallel, a machine learning data-driven model estimates the probability of landslide occurrence at the catchment scale. Both model types are deployed as automated cloud services that generate daily forecasts, overcoming key operational challenges related to model integration, parameter updating, and large-scale data handling. Forecast outputs are disseminated through NGI Live, the Norwegian Geotechnical Institute’s data platform, supporting Landslide Early Warning Systems (LEWS).
Complementing the regional framework, slope-scale forecasting is achieved through the development of a digital twin of an instrumented slope in Norway (Piciullo et al., 2022; Piciullo et al., 2025). The digital twin integrates real-time monitoring of hydrological variables, such as volumetric water content and pore water pressure, with numerical slope stability modelling and machine learning. The numerical model is continuously validated against monitored data and used to calculate the FoS. To enable efficient operational forecasting, data-driven models, including Polynomial Regression and Random Forest, are trained on simulated FoS values, monitored hydrological conditions, and meteorological inputs to forecast the rolling three days FoS. These data-driven models replace the computationally intensive numerical model within the cloud service, enabling rapid and reliable FoS forecasts. A performance evaluation demonstrates that the data-driven surrogates provide accurate and robust FoS predictions comparable to the numerical model, highlighting their suitability for operational early warning applications.
By integrating detailed slope-scale digital twins with scalable regional-scale forecasting, we illustrates a coherent multi-scale approach to landslide prediction. The proposed framework is readily transferable to other sites and regions, offering a practical pathway for enhancing real-time landslide early warning and risk management.
The authors gratefully acknowledge the support received from The HuT EU project (ID101073957, https://thehut-nexus.eu/), which played a crucial role in facilitating and advancing our research.
References
Abraham, M. T., Piciullo, L., Liu, Z., Drøsdal, et al. (2025). Operational regional scale landslide forecasts: Physics-based and data-driven models. Proceedings of the 9th International Symposium on Geotechnical Safety and Risk (ISGSR 2025). Research Publishing, Singapore. https://doi.org/10.3850/981-973-0000-00-0-isgsr2025-paper.
Piciullo, L., Abraham, M. T., Drøsdal, I. N., and Paulsen, E. S. (2025). An operational IoT-based slope stability forecast using a digital twin. Environ. Model. Softw. 183, 106228. https://doi.org/10.1016/j.envsoft.2024.106228.
Piciullo, L., Capobianco, V., and Heyerdahl, H. (2022). A first step towards a IoT-based local early warning system for an unsaturated slope in Norway. Nat. Hazards 114. https:// doi.org/10.1007/s11069-022-05524-3.
How to cite: Piciullo, L. and Abraham, M. T.: Real-Time Multi-Scale Slope Stability Forecasting, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-16966, https://doi.org/10.5194/egusphere-egu26-16966, 2026.
Forecasting the evolution of slow-moving landslides is a challenge because landslide motion is modulated by hydrometeorological forcing (rainfall, snowmelt, groundwater fluctuations) acting across multiple timescales, resulting in complex and strongly non-linear forcing–response relationships. By leveraging long-term multi-parameter monitoring, AI-based models can help characterise and simulate these dynamics. However, two limitations persist. First, many approaches rely on deep-learning architectures (RNNs, GRUs, LSTMs) that successfully reproduce non-linear dynamics, but do not constrain landslide physics and have limited interpretability and transferability. Second, few AI applications address landslides governed by the combined influence of multiple hydrometeorological drivers. Existing applications remain largely site-specific, relying on tailored predictor sets and local calibration. Addressing these limitations requires interpretable modelling frameworks capable of operating across multiple landslide sites including data-scarce settings.
Here, we introduce a scalable and eXplainable Artificial Intelligence (XAI) modelling framework using eXtreme Gradient Boosting (XGBoost) and based on a set of 248 and physically grounded, non site-specific hydrometeorological predictors computed from net rainfall, effective rainfall, and groundwater level time series. Predictors are designed to represent three complementary aspects of landslide water-related forcing: (i) the hydrological state of the system, (ii) hydrological memory effects, and (iii) short-term hydrological transient processes. To capture multi-timescale hydromechanical dependencies, predictors are computed over multiple time windows ranging from 1 to 90 days. The approach simulates daily landslide velocities, evaluates predictive skill using RMSE and MAE metrics, and provides interpretable and explainable constraints on the predictor influence using features importance ranking and SHAP-based attribution tools.
We evaluate the framework on three slow-moving landslides in France: Séchilienne (fractured miscaschist), Viella (morainic and colluvial deposits), and Villerville (chalk, sand and colluvial deposits ovelying marl substrate) spanning contrasting lithologies, deformation mechanisms and kinematics to demonstrate the scalability of the approach.
The XAI framework accurately reproduces landslide velocity time series across sites and testing periods with small residual errors relative to the amplitude of observed velocity variations (Séchilienne, 0.005-0.015 cm.d-¹ ; Viella, 0.01-0.035 cm.d-¹ ; Villerville, 0.02-0.06 cm.d-¹). The identified predictors per landslide align with contrasting physical processes, including delayed pore-water pressure build-up driven by slow matrix infiltration in impermeable slope material (Villerville) and rapid responses to rainfall in more permeable (Viella) or fractured (Séchilienne) slope materials. Together, these results show that XAI frameworks can recover site-specific landslide behaviour while preserving physical interpretability across diverse settings, and demonstrate one of the first applications of a common model structure and non-site-specific predictor set across multiple distinct landslide case studies.
How to cite: Béjean-Maillard, O., Bertrand, C., Malet, J.-P., Dubois, L., Batailles, C., Lespine, L., Maquaire, O., Fressard, M., and Ducasse, J.: Scalable XAI-based forecasting of landslide surface velocities from environmental forcings, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-18592, https://doi.org/10.5194/egusphere-egu26-18592, 2026.
Vietnam faces substantial landslide risk, with the highest number of reported landslide-related fatalities in Southeast Asia. Approximately 70% of the country’s territory is mountainous or hilly, and landslides recur annually during the rainy season from June to November, particularly in northern and central provinces. The severe impacts of Typhoon Yagi in September 2024, which caused 323 fatalities and an estimated USD 3.47 billion in damages, further highlighted systemic gaps in landslide risk information and early warning.
In response, the Vietnam Disaster and Dyke Management Authority, the Swiss State Secretariat for Economic Affairs, and GIZ initiated a scoping study to explore the development of a national landslide risk information system. The study’s primary objective is to assess the feasibility of establishing a landslide risk information system in Vietnam through a systematic review of existing data, tools, systems, and methodologies. It seeks to define a practical framework covering technical design, institutional arrangements, and capacity-building needs, and to develop a phased roadmap with indicative cost estimates and implementation timelines to guide future investment and system development
The aim of this contribution is to document the first stage of the scoping study, including initial stakeholder consultations and preliminary findings, and to define how these will inform the subsequent assessment and development of recommendations. The study applies a data maturity assessment framework based on a structured questionnaire covering seven dimensions of a landslide risk information system: data access and sharing, digital applications and services, information and communication technology infrastructure, staff competencies, institutionalisation and partnerships, governance, and disaster risk reduction collaboration.
The inception phase confirms that effective landslide early warning in Vietnam requires a multi-level system that links national technical capacity with provincial coordination and commune-level action. At the national level, the Department of Geology and Mines has been identified as a potential nodal agency for maintaining a national landslide database, working in coordination with the National Hydro-Meteorological Forecasting Centre for forecasting, the Disaster Management Policy and Technology Centre for capacity development, and the National Remote Sensing Department for satellite-based monitoring. At the provincial level, significant capacity strengthening is needed to digitise commune-level data, integrate scientific and community-based risk maps, and translate national warnings into village-specific advisories. At the communal level, priorities include the use of simple smartphone-based reporting tools, the development of community-based disaster risk management maps, and the dissemination of warnings through established platforms such as Zalo.
Several structural and technical challenges constrain the development of such a system. These include restrictions on data sharing and protection, limited and short-term funding arrangements, high staff turnover, the absence of unified technical standards, and regulatory constraints that limit innovation. Critically, the lack of systematic and georeferenced landslide reporting impedes the development of reliable thresholds and evidence-based risk assessments. In addition, the absence of digitised village-level risk maps and real-time monitoring capacity limits local decision-making and increases the likelihood of overly generalised or inaccurate warnings at the commune level.
How to cite: van Westen, C., Meszarosova,, S., Nguyen Thanh, L., Vuong Thu, H., Pham Tran, M., Mai Ky, V., Bui Van, H., Angelino, C., Lombardo, L., Tanyas, H., and Dahal, A.: Designing a National Landslide Risk Information System for Vietnam, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-14716, https://doi.org/10.5194/egusphere-egu26-14716, 2026.
Rapid flow-like geohazards pose acute threats to communities and infrastructure, yet physics-based runout simulators remain computationally prohibitive for operational impact-based risk analysis. Even if high-resolution datasets with extensive coverage are available, high computational costs direct the decision makers into using scenario-based assessments, which can significantly miscalculate the expected risk given the highly uncertain nature of such events. This study investigates Gaussian-process (GP) emulators for extremely high-dimensional outputs (exceeding 103 to 104 spatio-temporal grid points), systematically quantifying the trade-offs introduced by dimensionality reduction (DR). We compare three GP variants—Parallel Partial Gaussian Process (PPGaSP), Batch-independent GP (BiGP), and Multitask GP (MTGP)—and apply an established DR–GP workflow to assess the impact of different DR approaches on emulation accuracy and efficiency. This workflow first compresses spatio-temporal fields into low-dimensional latent representations, then performs GP emulation in latent space, and finally reconstructs predictions with uncertainty quantification in the original grid space. Three benchmark cases, synthetic and real-world problems, are used to validate the framework. Our findings provide actionable guidance for selecting appropriate emulation models in high-dimensional geohazard problems. We also investigate the balance between computational efficiency and prediction fidelity for risk-informed early-warning integration.
How to cite: Chang, C.-H., Yildiz, A., and Kowalski, J.: High-dimensional predictions for impact-based risk analysis of geohazards, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-1288, https://doi.org/10.5194/egusphere-egu26-1288, 2026.
Traditional landslide classifications, such as those by Varnes (1978) and Cruden & Varnes (1996) are primarily focused on material type and movement style. The new scheme presented here, inspired by Leroueil et al. (1996), organizes mass movements into sequential stages: Pre-failure: Damage and deformation processes that weaken the slope; Failure: The point at which mechanical properties are altered enough to cause instability; Activation: The initial movement triggered by failure; Post-failure: Changes in propagation style or further movement; Quiescence: A period of inactivity, but with potential for remobilization; Remobilization/Reactivation: Renewed movement after quiescence by new types or following the previous movement styles; Stabilization: The final, stable state.
This approach allows for a more nuanced understanding of landslide evolution, supporting both forensic analysis and predictive modeling. The expanded classification explicitly incorporates ice, snow (Locat et al., 2024), and rock debris as distinct material types, recognizing their growing importance in mass movement processes: Ice: Behaves similarly to rock, with unique rheological properties (e.g., ice creep, fracture). Snow: Treated analogously to soil, with subtypes (dry, wet, slush) based on water content and mechanical behavior. Rock debris: Recognized for its distinct propagation and initiation mechanisms, differing from both classical rockslides and debris slides. It also considers the significance of ambient fluid (subaerial vs subaqueous landslides), which has important implications during the pre-failure, failure and post-failure stages as well as cascading events such as tsunamis.
Several new types and refinements are introduced: Damaging: Cohesive masses breaking away in an indefinable manner, not previously formalized; Detachment: A cohesive solid body that separates either through an indeterminate process or by means of tearing. Glide: Solid or cohesive masses slipping over gentle slopes, including phenomena like rock blocks sliding on grassland. Secondary effects: Air blasts, entrainment, and erosion are now explicitly included, acknowledging their significant impact during and after mass movement events.
The classification also clarifies and expands definitions for slides, flows including Flow ± Slide, water-supported and density currents, the latter being specific for subaqueous landslides, snow avalanches and pyroclastic flow, ensuring that a broader range of real-world scenarios are covered.
By structuring landslide classification around stages and integrating new materials and types, the proposed scheme: Facilitates scenario-based hazard and risk assessment; Supports both retrospective (forensic) and predictive analyses; Addresses the increasing complexity of mass movements in a changing climate, including cascading and sequential events.
References
Cruden, D.M., & Varnes, D.J. 1996. Landslide Types and Processes. In: Turner, A.K.S., R.L. (ed.) Landslides: Investigation and Mitigation, 36-75.
Leroueil, S., Locat, J., Vaunat, J., Picarelli, L., Lee, H. & Faure, R. 1996. Geotechnical characterization of slope movements. Proc., 7th International Symposium on Landslides, Trondheim, 53-74.
Locat J., Urgeles R., Isler D., Jaboyedoff M., Lee H., Leroueil S., Mulder T., 2024. The Varnes’ classification of mass movement types to include the subaqueous environment and snow/ice materials. In: Merrien V. and Nicot F. (Eds.): 14TH INTERNATIONAL SYMPOSIUM ON LANDSLIDES, 8th - 12th July 2024, Chambéry, France. 189-192.
Varnes, D.J. 1978. Slope movement types and processes. Special report, 176, 11-33.
How to cite: Jaboyedoff, M., Locat, J., Issler, D., Mulder, T., and Urgeles, R.: Extending Varnes' mass movement classification from pre-failure through post-failure, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-21106, https://doi.org/10.5194/egusphere-egu26-21106, 2026.
Landslides are among the most destructive natural hazards in mountainous regions. Their occurrence is jointly governed by predisposing factors such as topography, geology, and soil properties, as well as external triggers such as rainfall. The temporal evolution of rainfall plays a crucial role in controlling pore-water pressure build-up and slope instability. However, most existing data-driven studies rely on metrics that condense complex information into scalar quantities – such as accumulated precipitation or maximum intensity – that fail to capture the “memory effect” of antecedent rainfall and wet–dry cycles on slope stability. This leaves an important question unresolved: how do the accumulation and temporal patterns of historical rainfall across different time scales influence the likelihood that a subsequent rainfall event will trigger landslides?
To address this problem, we propose a binary time-encoding approach for long- and short-term rainfall sequences. The method transforms continuous rainfall records into binary indicators that describe the occurrence, persistence, and temporal arrangement of rainfall. By summarizing rainfall history across multiple time windows, the approach preserves key antecedent information while reducing noise in long rainfall series and substantially lowering computational cost, making it suitable for large-scale, multi-event landslide susceptibility and spatio-temporal forecasting models.
We designed case studies using open-access landslide inventories, such as Northeastern Turkey, Italy, Switzerland, and precipitation datasets to compare (i) models built with conventional cumulative or intensity-based rainfall metrics and (ii) models incorporating the proposed binary time-encoded rainfall features. The analysis is implemented within the SHIRE framework (Edrich et al., 2024), while introducing a novel binary time-encoding strategy for long- and short-term rainfall sequences. Here, we present results demonstrating how antecedent rainfall at different temporal scales influences landslide occurrence and show that binary time encoding provides a compact and transferable representation of rainfall “memory” for regional landslide hazard assessment and early-warning frameworks.
References
Edrich, AK., Yildiz, A., Roscher, R., Bast, A., Graf, F. & Kowalski, J., A modular framework for FAIR shallow landslide susceptibility mapping based on machine learning. Natural Hazards 120, 8953–8982 (2024). https://doi.org/10.1007/s11069-024-06563-8
How to cite: Zhu, F., Kowalski, J., and Yildiz, A.: Assessing the impact of rainfall memory on landslide susceptibility using binary time encoding, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-1875, https://doi.org/10.5194/egusphere-egu26-1875, 2026.
Landslide susceptibility mapping (LSM) at the global scale is a prerequisite for hazard risk management but has long been hindered by inventory bias and unquantified model uncertainties. Existing global products are often constrained by substantial spatial sampling biases, leading to inconsistent prediction performance across data-scarce and vegetated regions. Addressing these challenges, this study presents a robust 1-km global susceptibility model derived from a dataset of over 2 million landslide events aggregated from 24 diverse sources.
To resolve data heterogeneity, we applied an LLM-driven framework (utilizing Qwen2.5-7B) to extract and standardize attributes from unstructured descriptions across 14 languages, significantly elevating metadata completeness . Leveraging this enriched inventory and 34 environmental predictors (comprising 17 static and 17 dynamic variables), we implemented a rigorous spatial block cross-validation strategy to strictly evaluate model transferability. We evaluated nine machine learning algorithms (e.g., CatBoost, ExtraTrees) coupled with Optuna tuning. Furthermore, Monte Carlo simulations (N=50) were integrated to propagate input uncertainties, generating explicit pixel-level confidence intervals.
Our results demonstrate high predictive accuracy (spatial CV AUC > 0.99), suggesting that the density of the training data effectively bridges generalization gaps found in previous studies. Feature optimization confirms the model’s robustness even with reduced dimensions. Spatially, the model identifies high-susceptibility zones in complex tropical highlands (e.g., the Andes and Southeast Asia), aligning with independent records of fatal landslide clusters. By providing a bias-corrected and uncertainty-aware spatial baseline, this study offers a critical foundation for global hazard monitoring.
How to cite: Jin, R. and Zhang, S.: Global landslide susceptibility mapping: a 1 km resolution model derived from a 2-million-event inventory with uncertainty quantification, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-4351, https://doi.org/10.5194/egusphere-egu26-4351, 2026.
Landslide inventories are often incomplete and biased due to limited personnel and financial resources, which constrains the development of high-quality, long-term spatio-temporal landslide datasets. In comparison, event-based landslide inventories, which are typically compiled shortly after triggering storms, can be mapped more comprehensively and tend to be internally consistent. Leveraging such inventories is thus valuable for exploring the interconnections between extreme precipitation events and environmental characteristics on slope instability.
Here, we evaluate the temporal transferability of event-based landslide susceptibility models to another landslide event, and the sensitivity of transferability to the choice of landslide absence sampling time windows. Accounting for spatial landslide collection bias and temporal biases in landslide absence sampling, we trained three Generalized Additive Models (GAMs) on landslides triggered by the September 2024 extreme precipitation event "Boris" to the Pielachtal region, Lower Austria. The models differ only in their temporal windows for landslide absence sampling: (M1) from the onset of the precipitation event until the observation date of the last inventoried landslide (September 12-17, 2024), (M2) from from the start of the month until the observation date of the last inventoried landslide (September 1-17, 2024), and (M3) only on the dates of landslide occurrence (September 16 -17, 2024). The models were then validated against an independent event, the May 2014 precipitation-triggered landslide inventory, to assess temporal generalization.
This research provides insights into how absence sampling design influences event-based, spatio-temporally dynamic landslide susceptibility modelling and its transferability across events. Our findings support cost-effective protocols for inventory compilation and model development, and enhancing readiness for future extreme precipitation events.
How to cite: Sternath, S., Steger, S., Schlögl, M., and Glade, T.: The effect of different landslide absence sampling time windows in event-based landslide susceptibility models, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-17136, https://doi.org/10.5194/egusphere-egu26-17136, 2026.
Abstract
Rainfall-triggered landslides constitute a major natural hazard worldwide and are especially prevalent in mountainous regions experiencing intense rainfall. Despite substantial progress in the development of empirical hydrometeorological thresholds for landslide initiation, a central challenge remains the definition of spatially distributed thresholds that adequately represent both hydrological preconditioning, rainfall triggering and spatial variability in hillslope response. Existing regional approaches often rely on antecedent rainfall as a proxy for subsurface conditions or treat slope susceptibility as spatially homogeneous, thereby limiting their physical interpretability and operational robustness.
This study develops a susceptibility-informed hydro-meteorological threshold framework for rainfall-triggered landslides in Rwanda, a mountainous country of tropical Africa in an under-researched type of climate. The framework explicitly integrates rainfall triggering, hydrological preconditioning, and spatial variability in slope response within the cause–trigger concept. Rainfall forcing is derived from IMERG and downscaled from its native 0.1° (~10 km) spatial resolution to 1 km to better capture local-scale rainfall variability in complex terrain. Hydrological preconditioning is represented using a simple leaky-bucket water-balance model, providing spatially distributed proxy indicators of soil moisture and subsurface water storage that explicitly characterize antecedent wetness conditions relevant for slope stability.
Hydro-meteorological thresholds are formulated by combining rainfall intensity–duration and cumulative rainfall metrics with hydrological state indicators derived from the water-balance model. The threshold behavior is explicitly conditioned on an existing regional landslide susceptibility map, allowing identical hydro-meteorological forcing to produce different threshold responses depending on terrain predisposition. A landslide inventory comprising 82 documented events of exact known date of occurrence from 2000 to 2024 is used to analyze trigger–response relationships and to evaluate threshold behavior across susceptibility classes. Thresholds are explored using empirical and statistical techniques, including cumulative rainfall analysis, multi-dimensional trigger plots, and receiver operating characteristics (ROC)-based performance assessment.
Preliminary results show that observed landslides are strongly concentrated in moderate to high susceptibility classes, with frequency ratio (FR) values increasing from 0.24 in very low susceptibility areas to 4.1 in very high susceptibility areas. This supports conditioning hydro-meteorological thresholds on spatial predisposition, enabling more spatially differentiated and physically interpretable early warning thresholds.
Keywords: Rainfall-triggered landslides, Hydro-meteorological thresholds, Antecedent wetness, Landslide susceptibility
How to cite: Dusabimana, J. D., Dewitte, O., Uwihirwe, J., Bogaard, T., Bugenimana, E. D., Musemakweri, J., Vanmaercke, M., Van Weverberg, K., and Reinoso Rondinel, R.: Susceptibility-informed hydro-meteorological thresholds for rainfall-triggered landslides in Rwanda, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-6632, https://doi.org/10.5194/egusphere-egu26-6632, 2026.
Landslides represent a significant geohazard worldwide, whose frequency and impacts are being amplified by climate change materialized through more intense and extreme rainfall. Projecting climate-driven landslide risk in tropical mountains such as the Colombian Andes requires methodologies that integrate climate projections with geomorphological triggers, going beyond traditional static susceptibility maps toward dynamic process-based frameworks. This study presents a novel methodology to assess future landslide propensity, integrating statistically downscaled climate projections with climate-informed probabilistic landslide models. A performance-weighted multi-model ensemble was constructed from 20 global models from the CMIP6 project (GCMs), selected according to their ability to reproduce observed rainfall patterns and trends during a historical baseline period (1981–2014). This ensemble provided future monthly climate data (2024–2100) for three shared socioeconomic pathways (SSP1-2.6, SSP2-4.5, and SSP5-8.5). These data enabled the calibration of monthly generalized additive models (GAMs) for landslide probability, trained with more than 10,000 events and using 15 extreme rainfall indices as explanatory variables, along with slope gradient as topographic control. To improve interpretability and robustness, the model results, originally at the climate model grid scale, were aggregated into slope units, generating maps of relative landslide propensity in probabilistic terms, a more appropriate spatial representation for future risk assessment than point estimates.
Our analysis revealed strong seasonal control: landslide triggers shift from high-intensity rainfall during the main wet seasons (April-May, October-November) toward antecedent dryness metrics in transition months. Future projections indicate a marked intensification in landslide propensity, especially in the Central and Western mountain ranges. Projected increases in mean rainfall, from approximately 20% in the short term (2024–2040) to more than 50% toward the end of the century (2081–2100) under SSP5-8.5, were correlated with a notable expansion of areas classified with high landslide propensity. Critically, the methodological framework identified not only where, but also when, propensity is highest within the annual rainfall cycle. This work improves landslide risk assessment by providing continuous probabilistic forecasts over time (monthly), which are highly sensitive to climate variability. Our results provide practical, scenario-based information to identify critical time windows and geographical priorities that support adaptive land use planning and early warning systems in a region highly vulnerable to geological hazards. Future work in progress will aim to refine and expand this framework, considering the inclusion of additional predictors, such as soil moisture, temperature, and changes in land cover, in order to address the occurrence of the phenomenon under study in a more holistic manner.
How to cite: Vega, J.: Modeling Future Landslide Propensity in the Colombian Andes: A GAM-based Projection from GCM Multi-Model Extreme Rainfall Indices, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-8167, https://doi.org/10.5194/egusphere-egu26-8167, 2026.
Mass movements are a primary process in the evolution of landforms in the Paute River basin in southern Ecuador, where lithological, structural, and topographical factors drive terrain instability. Rainfall is a significant triggering factor due to its direct influence on increasing interstitial pressure and reducing material resistance, particularly on slopes previously modified by anthropogenic activities, such as the mega-landslide La Josefina, which occurred in 1993 (Bonnard, 2011). This research aims to establish precipitation thresholds to enhance understanding of the activation and reactivation of mass movements in the central Paute River basin, with a focus on events that impact infrastructure and human settlements (Torres Ramírez, 2021). The methodology involves collecting and refining rainfall records from manual stations operated by the National Institute of Meteorology and Hydrology (INAMHI) and comparing these with the Integrated Multi-satellite Retrievals for GPM (IMERG) satellite products to address data discontinuities and improve spatial rainfall coverage. Statistical analyses were conducted to identify critical precipitation thresholds associated with the initiation of mass movement processes, based on correlations between event occurrence and antecedent and accumulated precipitation conditions (Iza-Wong et al., 2025; WMO, 2017). Preliminary findings indicate that precipitation thresholds vary across the study area by season (Marco Molina et al., 2000; Zaragozí et al., 2025). During the rainy months of March, April, and May, rainfall is more concentrated in the southwestern region, with precipitation ranging from the 95th percentile value of 6 mm/day up to 14 mm/day. In contrast, during September, October, and November, higher rainfall is observed in the northeastern region. This spatial heterogeneity underscores the influence of geology, soil texture, and land-use changes on mass-movement processes; in addition, the evaluation of precipitation patterns further distinguishes the types of events that trigger landslides in the area. These conclusions offer a technical foundation for enhancing preparedness, monitoring, and early warning systems for climate risk management in the Paute River basin.
Keywords: Hydrogeomorphology, mass movements, precipitation, rainfall thresholds, climate risk, Paute River basin
References
Bonnard, C. (2011). Technical and Human Aspects of Historic Rockslide-Dammed Lakes and Landslide Dam Breaches (pp. 101–122). https://doi.org/10.1007/978-3-642-04764-0_3
Iza-Wong, A., Moldovan, G., Ben-Bouallègue, Z., Hemingway, R., Chantry, M., & Lavers, D. (2025). Evaluation of precipitation observations across Ecuador. Atmospheric Science Letters.
Marco Molina, J.A.; Matarredona Coll, E.; Padilla Blanco, A. (2000) La dimensión espacial de los riesgos geomorfológicos. Boletín de la Asociación de Geógrafos Españoles. Available online: https://dialnet.unirioja.es/servlet/articulo?codigo=1122897
Torres Ramírez, R. (2021). Análisis espacio-temporal de los eventos ocurridos (movimientos en masa), en el período 2012-2020, en la zona centro de la cuenca del río Paute-Ecuador. http://rua.ua.es/dspace/handle/10045/114795
WMO. (2017). Guide to the Global Observing System. WMO-No. 488. https://community.wmo.int/en/wmo-no-488-guide-global-observing-system
Zaragozí, B., Font, P., Cano-Aladid, J., & Marco Molina, J. (2025). A Small Landslide as a Big Lesson: Drones and GIS for Monitoring and Teaching Slope Instability. Geosciences, 15, 375. https://doi.org/10.3390/geosciences15100375
How to cite: Iza Wong, A. M., Torres-Ramírez, R., Marco-Molina, J. A., Mayacela-Salazar, B., and Vásquez-Morante, S.: Rainfall thresholds triggering mass movements in the central Paute River basin, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-15268, https://doi.org/10.5194/egusphere-egu26-15268, 2026.
In India, landslides are one of the severe disasters with the highest fatality rate. Over the past few years, due to the heavy and prolonged rainfall events, there has been a surge in landslides in the Northwestern Himalayan region. Himachal Pradesh has faced an economic loss of $60 million alone in the 2021 monsoon season, with more than 200 casualties, followed by severe damage caused in the 2023 and 2025 monsoons. To mitigate the risk, landslide susceptibility mapping (LSM) has emerged as a fundamental step that can help in formulating policies for high-risk areas. Statistical methods, deterministic approaches and remote sensing techniques have been extensively employed by various researchers to forecast landslides. This paper introduces a novel LSM framework which utilises both natural and anthropogenic conditioning factors to develop pixel-based site-specific susceptibility. The natural parameters include topography (elevation, slope, aspect), geomorphology, distance to streams, water table depth. Anthropogenic factors include Normalized Difference Vegetation Index (NDVI) change, distance from roads. This study integrates the quantitative methods along with the qualitative expert knowledge to develop enhanced susceptibility maps for the 3 landslide events that occurred in the months of July and August 2023 in Mandi district. To overcome the simplicity and uncertainty of parameters probability of failure is utilized to reframe the susceptibility. The buffer zone for each landslide is categorized into 3 zones based on risk associated: low risk (green), medium risk (yellow), and high-risk (red) zone. Cross-validation is employed to evaluate the generalization capability of models across the landslide sites, to understand their inter-site transferability.
Keywords: Rainfall induced landslides, Probability of failures, susceptibility mapping, uncertainty analysis
How to cite: Chauhan, T., Thakur, V., and Uday, K. V.: Probabilistic framework for enhanced Landslide Susceptibility Mapping for Rainfall-Induced Landslides, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-21883, https://doi.org/10.5194/egusphere-egu26-21883, 2026.
Landslides pose a considerable threat to urban environments, and understanding where and how they may impact critical infrastructure is essential for risk management and early warning. This study presents an integrated methodological framework for data-driven landslide analysis in Tbilisi, Georgia, combining initiation susceptibility mapping, empirical runout path assessment, and exposure analysis. The approach focuses on the Tbilisi area (~505 km²) and first models landslide initiation susceptibility separately for slides, flows, and falls using a range of topographic and geological predictors. Generalized Additive Models (GAMs) were applied to produce continuous probability maps of initiation, which were subsequently classified into low, medium, and high susceptibility classes to define potential source locations for process path simulations. Based on these release locations, potential downslope propagation was estimated using a simplified, empirical energy-line approach based on the angle-of-reach principle. Multiple stochastic simulations per release cell captured variability in runout paths. The resulting potential process path maps then formed the basis for exposure assessment by intersecting them with spatial data on buildings, roads, and railway lines. The analysis identifies areas most likely to be impacted, providing an evaluation of multi-landslide exposure across the area. Beyond serving as a baseline for spatial planning, the results are being evaluated for integration with real-time meteorological nowcasting products to support impact-based early warning. Overall, the study demonstrates the potential of a straightforward landslide modelling chain to support risk management and early warning, contributing to enhanced resilience in Tbilisi. The analysis was conducted within the MedEWSA project funded by Horizon Europe (Grant Agreement No. 101121192).
How to cite: Gaprindashvili, G., Steger, S., Kienberger, S., Kinkladze, I., Gaprindashvili, M., Kurtsikidze, O., Rikadze, Z., and Bairamovi, T.: A Straightforward Integrated Assessment of Landslide Initiation, Potential Process Paths, and Exposure in Tbilisi, Georgia, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-6948, https://doi.org/10.5194/egusphere-egu26-6948, 2026.
To improve and automate the shallow landslide component of the already operating Austrian early warning system AMAS (Austrian Multi-Hazard Impact-based Advice Service), regional precipitation thresholds are needed. Both the existing warning system and the precipitation thresholds developed in this study do not target individual landslides, but focus on severe regional events involving multiple landslides. Here, we present preliminary results of precipitation threshold modelling at national scale.
Historic regional events were extracted from Austria-wide landslide inventories, including GEORIOS (GeoSphere Austria), the WLK database of the Austrian Torrent and Avalanche Control, as well as landslide inventories from different Austrian federal states. Landslide absence observations were identified by selecting landslide-free precipitation events with more than 20 mm in 24 h for which no indications of landslides were found after screening additional sources such as fire brigade reports, police records, local authorities, and VIOLA – the severe weather database of GeoSphere Austria.
Taking into account the diverse environmental conditions under which landslides occur, Austria was divided into 21 geo-climatic regions using hierarchical cluster analysis, which considered geological, geomorphological, pedological and climatic factors, complemented by expert knowledge. While our aim is to model the precipitation thresholds for each of the 21 geo-climatic regions in Austria, we present preliminary results for two study areas of the Fischbacher Alps and the Vorarlberger Molasse. Lessons learned in these areas will be applied to nationwide modeling.
Precipitation threshold modeling was performed using two different techniques: (i) a data-driven approach based on generalized additive models (GAMs), which combines triggering and antecedent precipitation, and (ii) a quantile regression approach, which defines the onset of relevant precipitation following a dry period. For both approaches, precipitation data from INCA (Integrated Nowcasting through Comprehensive Analysis, combined radar and station data) were used with hourly resolution.
To optimize the results, the durations of triggering and antecedent precipitation in the GAM model, as well as the dry-period duration and the maximum precipitation threshold during the dry period in the quantile regression model, are systematically varied. Additional model variants consider the inclusion of the antecedent precipitation index (API) and the use of different landslide samples (e.g., representatively sampled points across different rainfall events) for both models. The best modeling results are selected via ROC-based cross-validation complemented with expert plausibility checks (e.g., longer antecedent precipitation for fine-grained soils).
First GAM results showed very high predictive performance, with mean cross-validation AUROCs exceeding 0.9. Including a third variable in the model, namely peak 1-hour rainfall within the triggering window, alongside cumulative triggering and antecedent precipitation further improved the model, and the modeled relationships appeared plausible. Early quantile-regression estimates of intensity-duration (ID) thresholds are consistent with prior work (e.g., Guzzetti et al., 2008; Marra et al., 2014) but exhibit a steeper power-law decay. These results are sensitive to event-sample representativeness as well as the delineation of triggering rainfall, and they reveal spatial heterogeneity consistent with differing geological and meteorological predisposition.
How to cite: Schwarz, L., Steger, S., Spiekermann, R., Enigl, K., Schlögl, M., and Tilch, N.: National-scale shallow landslide precipitation thresholds in Austria for early warning: A comparison of two modelling approaches , EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-7890, https://doi.org/10.5194/egusphere-egu26-7890, 2026.
Slow-moving landslides have recently gained attention as geological hazards requiring long-term monitoring, as they can trigger large-scale slope failures or debris flows. Consequently, various studies have identified slow-moving landslides as precursors to large-scale landslides. However, conventional field instrumentation or GPS-based monitoring has limitations for long-term monitoring of large-scale areas. Consequently, SAR-based time-series displacement analysis is being utilized as an alternative. SAR time-series analysis offers the advantage of enabling long-term monitoring of ground displacement across extensive regions. Nevertheless, research on the interaction between the long-term displacement patterns of slow-moving landslides and their triggering factors remains insufficient. In particular, systematic research is needed on how the displacement observed over time interacts with static factors (topography, geology, etc.) or dynamic factors (precipitation, temperature). Existing statistical-based time series models are useful for clearly analyzing trends and seasonality in displacement data and understanding the underlying structure. However, they have limitations in fully reflecting nonlinear displacement patterns or complex interactions with various triggering factors.
This study aims to perform time-series prediction using long-term SAR-based displacement data and analyze the relationship between displacement patterns and triggering factors from a data mining perspective. Specifically, it applies deep learning-based LSTM, capable of learning long-term dependencies, alongside existing statistical models for comparison and analysis. LSTM is evaluated as a model suitable for complex prediction of slow-moving landslides, as it considers the long-term cumulative effects of time-series data and can comprehensively learn nonlinear displacement patterns and multivariate data.
Applying the method proposed in this study, the Gangwon Province area of South Korea was designated as the study region, and displacement data was constructed using Sentinel-1 SAR imagery acquired from 2014 to 2024. We examined the interactions between static and dynamic data expected to influence the constructed displacement data. We then performed long-term predictions using SAR-based displacement time series via deep learning-based LSTM to evaluate the potential for landslide monitoring from a long-term perspective.
Acknowledgement
This work was supported by the National Research Foundation of Korea(NRF) grant funded by the Korea government(MSIT) (No. RS-2024-00358026 and RS-2025-00515970).
How to cite: Lee, J.-H., You, H.-Y., Park, H.-J., Kim, S.-W., Jeong, C. H., and Chae, S. H.: Long-Term Displacement Prediction of Slow-Moving Landslides Using SAR-Based Time-Series Displacement Data, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-9455, https://doi.org/10.5194/egusphere-egu26-9455, 2026.
The frequency and magnitude of landslide damage have increased due to the impact of heavy rainfall, which has been exacerbated by climate change. Consequently, the importance of landslide susceptibility analysis for identifying high-risk areas is being further emphasized. Previous susceptibility studies have utilized various data-driven analyses, including machine learning and deep learning, to understand the complex nonlinear relationships among landslide-influencing factors. In particular, ensemble techniques have been shown to enhance overall performance and stability by combining the prediction results of individual models. However, previous bagging and boosting-based ensemble techniques have primarily focused on improving average classification performance. Further examination is necessary to assess the stability and interpretability of decision boundaries under varying threshold values and the distribution characteristics of prediction probabilities. This is especially challenging in landslide datasets with significant class imbalance, where pixels in the boundary region can exhibit highly sensitive prediction changes depending on threshold settings.
To address these limitations, this study employed the gcForest (multi-grained cascade forest) model, also known as Deep Forest. gcForest is a deep learning alternative that utilizes a cascade structure, comprising multiple layers of random forests. Each layer receives both original features and class probability outputs from the preceding layer. This structure facilitates the incremental updating of probability information for samples near decision boundaries, enabling iterative reclassification. This structure is distinct from existing ensemble techniques in that it enables stepwise improvement of decision boundaries for samples with high prediction uncertainty. This is in contrast to the existing ensemble techniques that determine predictions at a single stage. In order to make a comparison with existing ensemble techniques, this study has set bagging-based random forest and boosting-based XGBoost as the base model of deep forest.
The proposed analysis approaches were applied to Pohang City, Gyeongsangbuk-do, South Korea, where a large-scale landslide occurred in 1998. The analysis results demonstrated that the gcForest-based model exhibited enhanced prediction performance (gcForest_RF AUC = 91.62%, gcForest_XGBoost AUC = 91.40%) in comparison to the prevailing ensemble methods, random forest and XGBoost. Specifically, the XGBoost-based gcForest model demonstrated enhanced accuracy, improving from 0.797 to 0.814, and an elevated f1-score from 0.789 to 0.814 when compared to the prevailing XGBoost model. These results indicate that gcForest's stepwise improvement structure contributes to enhanced performance in classifying uncertain samples near decision boundaries, thereby enabling more stable landslide susceptibility prediction.
Acknowledgement
This work was supported by a National Research Foundation of Korea (NRF) grant funded by the Korean government (MSIT) (RS-2025-00515970).
How to cite: Lee, S.-H., Lee, J.-H., and Park, H.-J.: Analysis of Rainfall induced Landslide Susceptibility Using Deep Forest Model for Decision Boundary Interpretation, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-9508, https://doi.org/10.5194/egusphere-egu26-9508, 2026.
Landslides rank among the most destructive geological hazards globally, with their frequency and intensity increasingly exacerbated by the dual pressures of climate change and rapid anthropogenic land modification. Traditional static landslide hazard mapping often relies on global feature importance rankings, which obscure the spatial heterogeneity of driving mechanisms. This black box nature limits the physical interpretability of hazard evolution. This study aims to establish a long-term Dynamic Landslide Hazard assessment framework to decouple the causal mechanisms of rainfall and land use in landslide evolution.
Focusing on Hubei Province (1980–2024), we integrated XGBoost for dynamic prediction with Double Machine Learning (DML) for causal attribution. To address high dimensionality, Principal Component Analysis (PCA) was employed to reconstruct comprehensive indices (cumulative variance > 90%). Central to our methodology is the proposal of a novel "Consistency-Interaction Diagnostic Framework." By coupling the global trends derived from Partial Dependence Plots (PDP) with the local heterogeneity of SHAP values, this framework constructs a 2D metric system to diagnose the physical stability and spatial interaction strength of drivers.
Application of this diagnostic framework successfully decoupled the dual physical attributes of landslide drivers, a distinction missed by traditional methods:
- The framework accurately identified land use intensity and static terrain factors as "Stable Background Stress." These factors exhibit high PDP-SHAP correlations (Consistency > 0.95) with low spatial variance, confirming their roles as domain-wide controls regardless of local micro-environments.
- In contrast, rainfall factors were diagnosed as "High-Sensitivity Pulses." For instance, antecedent summer precipitation exhibited an extremely high SHAP interaction Coefficient of Variation (CV) of 3.78. This quantitative diagnosis proves that rainfall is not a uniform stressor but a spatially selective trigger whose hazard efficiency is intensely modulated by local topography.
- Diagnostic results further reveal that the majority of environmental factors fall into the "Heterogeneous Effect" quadrant. This indicates that landslide incubation is not a linear superposition of single factors but a complex non-linear process strongly modified by local environments.
This study demonstrates that the proposed framework offers a new physical perspective for opening the black box of machine learning. By distinguishing between globally consistent factors and locally sensitive perturbations, the findings provide a scientific basis for shifting landslide risk management from homogenized meteorological warnings to fine-grained control based on spatial heterogeneity and ecological resilience.
How to cite: Zhu, S., Chen, L., and Segoni, S.: A Causal Analysis based on Dynamic Landslide Hazard Assessment from 1980 to 2024 in Hubei, China, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-12891, https://doi.org/10.5194/egusphere-egu26-12891, 2026.
Hydrometeorological forcing (rainfall, snowmelt, groundwater fluctuations) acts across multiple timescales and is a primary driver of surface velocity dynamics in slow-moving landslides. Many studies use trained AI-based models to simulate daily-to-monthly velocities over validation periods defined by specific historical hydrometeoroligical contexts. Although these models achieve accurate predictive skill, they are typically deterministic and therefore provide limited insight into the range of plausibly velocity responses under alternative, yet realistic, forcing conditions.
To address this gap, we introduce a probabilistic framework built around two axes. Forcing variability is represented by generating 500 plausible meteorological time series using a modified Richardson-type weather generator (rainfall and air temperature). These series are then propagated through a transfer-function hydrological model to simulate groundwater-level variability driven by generated effective rainfall. Second, daily velocities are simulated using a trained XGBoost model based on a set of hydrometeorological predictors. The resulting ensemble is summarised as monthly velocity distributions over a one-year horizon, thereby capturing distinct dynamics across a full hydrological cycle. Distributional performance is evaluated using the Prediction Interval Coverage Probability (PICP) and the Mean Interval Score (MIS).
We evaluate the framework on three slow-moving landslides spanning contrasting lithologies, deformation mechanisms and kinematics: Viella (morainic and colluvial deposits ; France), Villerville (chalk, sand and colluvial deposits ovelying marl substrate ; France), and Vallcebre (clayey siltstone and colluvial debris overlying limestone substrate), to demonstrate the scalability of the approach.
The modified Richardson-type generator reproduces key statistical properties of historical meteorological records. Calibrated groundwater models capture the main dynamics of groundwater fluctuations, with R2 values of 0.84 (Viella), 0.76 (Villerville) and 0.53 (Vallcebre). The simulated monthly velocity distributions exhibit clear seasonality, with more contrasted annual cycles at Viella and Villerville, consistent with site-specific hydrogeological behaviour. On average, prediction intervals encompass a substantial fraction of observed monthly velocities (mean PICP: 53% for Viella, 40% for Villerville and 76% for Vallcebre), with strong variability across months. Remaining discrepancies mainly reflect data availability, limitations in groundwater simulations, and constraints in the learned forcing–velocity relationships within the XGBoost model, highlighting priorities for further methodological improvements. Overall, the proposed framework provides a first practical tool to quantify the range of probable landslide-velocity responses under multiple plausible hydro-meteorological scenarios.
How to cite: Bertrand, C., Maillard-Bejean, O., Malet, J.-P., Moya, J., and Maquaire, O.: Probabilistic Simulation of Monthly Landslide Velocity Under Hydro-meteorological Variability, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-18603, https://doi.org/10.5194/egusphere-egu26-18603, 2026.
Landslides pose significant threats to infrastructure and communities in the Republic of Moldova, yet until now no comprehensive national-scale inventory or susceptibility assessment has been available. This study presents the first complete landslide inventory and AI-based susceptibility model for the entire country, integrating multi-source remote sensing data with presence-only machine learning techniques.
We developed a new landslide inventory comprising 246 polygons through visual interpretation of aerial imagery, orthophotos, and LiDAR data (5m resolution in central regions), complemented by field verification. This inventory was integrated with existing databases to create a comprehensive dataset of 1,523 landslide polygons for susceptibility modeling. Landslides were classified following international schemes, focusing on slide- and flow-type movements in medium- to deep-seated failures, while excluding shallow landslides, rockfalls, and debris flows.
Susceptibility analysis employed the MaxEnt presence-only machine learning algorithm with environmental variables including slope, elevation, valley depth, topographic wetness index, normalized height, Gaussian and Casorati curvature, lithology, and land cover derived from 30m resolution JAXA DEM and 1:200,000 geological maps. The model demonstrates strong predictive performance, with 68% of mapped landslides exhibiting mean susceptibility values exceeding 0.7.
Results reveal distinct spatial patterns: high-susceptibility zones (susceptibility values 0.7-0.997) form continuous corridors along valley networks in the central and northern hilly regions (Codrii Hills, Ciuluc Plateau, Dniester Hills), while southern and northern plains exhibit consistently low susceptibility (~8.27×10⁻¹¹ to 0.3). Geomorphometric analysis shows landslides preferentially occur at mid-slope positions (normalized height 0.3-0.6), in areas with moderate valley depths (15-28m median), and intermediate topographic wetness index values (7-10), reflecting strong structural control by cuesta landforms and Miocene clay-rich lithologies.
The bimodal distribution of susceptibility values within the inventory, with peaks at both low (<0.3) and high (>0.8) values, suggests the presence of both active landslides under current environmental conditions and relict features formed during wetter Pleistocene climates. This interpretation aligns with regional studies from adjacent Romanian territories.
This research provides the first national-scale susceptibility map for Moldova and establishes a scalable framework for landslide risk assessment in regions with heterogeneous geomorphology and incomplete historical data. The results support strategic planning for hazard mitigation, infrastructure development, and land-use management, particularly in densely populated agricultural regions where landslide impacts are already documented. Future work should focus on incorporating temporal triggering factors, anthropogenic influences, and climate change scenarios to enhance predictive capabilities.
Acknowledgements: This work was supported by a grant of the Ministry of Research, Innovation and Digitization, CNCS – UEFISCDI, project number 40PCBROMD within PNCDI IV.
How to cite: Ilinca, V., Nicoara, I., Daia-Creinicean, T., Tambur, A., Spian, C., Jeleapov, V., and Sandric, I.: National-Scale Landslide Susceptibility Mapping in the Republic of Moldova: A Presence-Only Machine Learning Framework, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-20026, https://doi.org/10.5194/egusphere-egu26-20026, 2026.
Landslide susceptibility is the likelihood of a landslide occurring in a given area based on local terrain conditions (Brabb, 1984). It is fundamental for land-use planning and risk mitigation strategies and can be assessed through various approaches, including statistical and physically-based methods (Guzzetti et al., 1999; Reichenbach et al., 2018). Statistical approaches are preferred for small scale zoning as they rely on landslide inventories and thematic maps that are easier to gather, while physically-based methods remain challenging as they demand detailed geomechanical and hydrological inputs that are time-consuming and costly to acquire.
This study presents a novel physically-based methodology for large-scale landslide susceptibility assessment that integrates the limit equilibrium method (Borselli, 2023) with spatialisation criteria and statistical classification approaches (Mevoli et al., 2026). The procedure enables the generation of spatially distributed safety factor and failure surface depth maps, and susceptibility zoning. The methodology was applied to a ~40 km² area in Southern Italy, testing multiple scenarios to evaluate the influence of different geomechanical and hydraulic configurations. Model performance was assessed through a classification algorithm, revealing scenarios with optimal discrimination capability. The physically-based results were compared with those obtained from statistical approach, demonstrating the promising applicability of the proposed physically-based methodology for assessing landslide susceptibility at large scales.
This reproducible and adaptable framework offers a physically-based alternative for assessing ladslide susceptibility at large scales, proividing direct applications for landslide susceptibility zoning in research and operational contexts.
References
Borselli L. (2023). "SSAP 5.2 - slope stability analysis program". Manuale di riferimento. Del codice ssap versione 5.2. Researchgate. https://dx.doi.org/10.13140/RG.2.2.19931.03361
Brabb, E.E., 1984. Innovative approaches to landslide hazard and risk mapping. In: Proceedings 4th International Symposium on Landslides, vol. 1, Toronto, pp. 307–324.
Guzzetti, F., Carrara, A., Cardinali, M., & Reichenbach, P. (1999). Landslide hazard evaluation: a review of current techniques and their application in a multi-scale study, Central Italy. Geomorphology, 31(1-4), 181-216. https://doi.org/10.1016/S0169-555X(99)00078-1
Mevoli, F. A., Borselli, L., Santangelo, M., Monte, N., de Lucia, D., Ugenti, A., & Rossi, M. (2026). Landslide susceptibility zoning through physically-based limit equilibrium method modelling. CATENA, 263, 109726. https://doi.org/10.1016/j.catena.2025.109726
Reichenbach, P., Galli, M., Cardinali, M., Guzzetti, F., & Ardizzone, F. (2005). Geomorphological mapping to assess landslide risk: Concepts, methods and applications in the Umbria region of central Italy. Landslide hazard and risk, 429-468.
How to cite: Mevoli, F. A., Borselli, L., Santangelo, M., Monte, N., de Lucia, D., Ugenti, A., and Rossi, M.: A novel physically-based methodology for assessing landslide susceptibility at large scales, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-20796, https://doi.org/10.5194/egusphere-egu26-20796, 2026.
The Cares route (Picos de Europa National Park - Northern Spain) is a hiking trail subject to intense tourist pressure, where visitors are frequently exposed to landslides, with rockfalls being the most common events. Despite the high frequency of these processes, no systematic inventory had been compiled to date. Since 2024, the SAFETRACK Project has been developing a comprehensive inventory of landslides affecting the route, including both recent and historical events. The inventory is based on the review of multiple data sources: (1) regional and local newspapers, (2) social media, (3) technical notes, and (4) reports from national park rangers. Each data source introduces specific biases into the dataset. For instance, technical reports usually provide highly accurate spatial information but often lack precise data on the timing of the events. In contrast, press archives and social media typically offer reliable temporal information, although spatial details are often imprecise. To address these limitations, the methodology incorporates several procedures aimed at extracting objective information from the original sources, assessing data reliability and minimizing inventory bias. These procedures include: (1) use of multiple and complementary data sources; (2) geo-location of landslides based on spatial descriptions and photographic evidence, supported by free online cartographic platforms (Google Maps-Google Street View and Iberpix) and fieldwork; (3) temporal location of the landslides through cross-validation among sources and interviews with park rangers and local residents; and (4) classification of the spatio-temporal data according to a reliability scale. The proposed methodology has proven effective in obtaining objective and sufficiently reliable data, making the resulting inventory suitable for subsequent quantitative analyses and future research.
Funding: Research Project “Sensibilización ante los procesos de ladera y mejora de la seguridad en sendas de montaña de los Parques Nacionales: propuesta de innovación para la autoprotección y educación ambiental – SAFETRACK” financed by the University of León.
How to cite: Valenzuela, P., Colmenero-Hidalgo, E., Rodríguez, I., Cruz, J. A., Almendros, P., García-Meléndez, E., Domínguez-Cuesta, M. J., Ferrer-Julià, M., and Pereira, I.: Reducing spatio-temporal bias in the Cares landslide inventory (Picos de Europa National Park, Northern Spain), EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-18426, https://doi.org/10.5194/egusphere-egu26-18426, 2026.
Landslide Inventory Maps (LIMs) are the essential starting point for any hazard assessment, yet their statistical quality is often assumed rather than verified. A persistent issue in susceptibility modeling, particularly with the widely used Frequency Ratio (FR) method, is the assumption of conditional independence among factors. This simplification not only overlooks complex inter-dependencies between geology and terrain but also tends to hide the inherent limitations and biases of the underlying inventory.
In this study, we propose a shift toward a Multivariate Conditional Likelihood Ratio (MCLR) framework to explicitly evaluate and manage inventory representativeness. By estimating likelihoods over joint combinations of geomorphic, hydrologic, and land-cover factors, MCLR preserves the multivariate signals that drive landslide occurrence. Crucially, we treat the resulting "empirical sparsity" (data-poor environmental units) not as a mathematical hurdle, but as a diagnostic strength. By imposing minimum support criteria, we can pinpoint specific environmental domains where the inventory lacks representative power, effectively "exposing" the quality constraints of the input data.
To test how these patterns perform under real-world forcing, we introduce an event-based Rainfall Amplification Factor (RAF) as a diagnostic stress test. Using a terrain-trend-plus-residual interpolation, we capture the spatial heterogeneity and orographic enhancement of precipitation to dynamically modulate the MCLR-based susceptibility. This allows us to track how inventory limitations propagate from static maps into event-scale hazard interpretations.
Our findings demonstrate that MCLR produces more physically interpretable patterns than marginal FR, especially in complex landscapes where terrain and geology are tightly coupled. The RAF analysis further reveals where susceptibility models remain robust and where representativeness gaps become critical during extreme events. Ultimately, this framework provides a transparent bridge between static susceptibility mapping and event-oriented hazard assessment, offering a quantitative basis for evaluating the reliability of landslide inventory products under extreme forcing conditions.
How to cite: Chiou, R. B. and Liao, K. W.: Refining inventory-based frequency-ratio landslide susceptibility using multivariate conditional likelihood ratios and event-based rainfall amplification, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-11306, https://doi.org/10.5194/egusphere-egu26-11306, 2026.
Posters virtual: Mon, 4 May, 14:00–18:00 | vPoster spot 3
EGU26-16487 | ECS | Posters virtual | VPS12
Machine Learning–Driven Landslide Nowcasting for Operational Early Warning in the Himalayan RegionMon, 04 May, 14:36–14:39 (CEST) vPoster spot 3
The intensification of extreme rainfall has resulted in widespread landslide hazards in mountainous regions of the world. The Indian Himalayan Region, one of the most densely urbanized, has been facing an alarming increase in landslides, the prediction of which is difficult using existing empirical rainfall thresholds. This study develops a novel machine learning-driven landslide nowcasting system by integrating the landslide susceptibility (LSM) and probability of rainfall-induced landslides (P-RIL). The LSM provides the spatial location of future landslides by analyzing the terrain characteristics, anthropogenic factors, hydrological presence, and geological formations using the random forest (RF) method based on landslides occurring between 2017-2024. The results indicated that 7% of the area was under high susceptibility, followed by 12% under high susceptibility. To calculate the effect of rainfall in triggering landslides, the P-RIL was calculated considering R1 (rainfall on 1st day of occurrence), R3 (rainfall on 3rd day), R7 (7th day rainfall), R15 (15th day rainfall), Wetdays, Max_72 Hours, and antecedent rainfall index (ARI) as variables to train in the RF model. Finally, each day nowcasting results were obtained by integrating the LSM and P-RIL within a probabilistic framework. The landslide occurring in 2025 was used to validate the nowcasting results. The results indicated that the landslides were ranked within the forecasted hazard distribution, with percentile values of 87%, 90%, 93%, and 99%, respectively, denoting the occurrence of landslides within the top 13%–1% of the most hazardous slope units at the time of prediction. One event lay in the extreme hazard class (>99th percentile), highlighting the model’s strong discriminatory capability. Finally, the forecast results for each day were updated in a Google Earth Engine application to aid policymakers and planners in developing better mitigation and preparedness strategies. This study represents the first of its kind landslide nowcasting system in Mandi district using the information obtained from landslide susceptibility and rainfall-derived triggering parameters, thus offering meaningful insight into a practical decision-support tool for policymakers and disaster management authorities.
How to cite: Singh, A., Dhiman, N., Pathak, B., and Praise Shukla, D.: Machine Learning–Driven Landslide Nowcasting for Operational Early Warning in the Himalayan Region, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-16487, https://doi.org/10.5194/egusphere-egu26-16487, 2026.
