The structure of LEWS can be schematized as an interrelation of the following main components: monitoring, modelling, forecasting, warning, response. However, tools, instruments, methods employed can vary considerably with the scale of analysis, as well as the characteristics and the aim of the warnings/alerts issued. For instance, at local scale instrumental devices are mostly used to monitor deformations and hydrogeological variables with the aim of setting thresholds for evacuation or interruption of services. At regional scale hydro-meteorological thresholds are widely used to prepare a timely response of civil protection and first responders. Concerning modelling techniques, analyses on local slopes generally allow for the use of numerical models, while statistical, probabilistic and physical-based models are widely used for large areas.
This session focuses on LEWS at all scales and stages of maturity, from prototype to active and dismissed ones. Test cases describing operational application of consolidated approaches are welcome, as well as works dealing with promising recent innovations, even if still at an experimental stage.
Contributions addressing the following topics will be considered positively:
- real-time monitoring systems (IoT)
- prediction tools for warning purposes
- in-situ monitoring instruments and/or remote sensing devices
- analysis of hydro-meteorological drivers to enhance forecasting
- warning models for issuing warning
- operational applications and performance analyses
- machine learning techniques applied for early warning purposes
Orals: Thu, 7 May, 14:00–17:45 | Room N2
Landslides in densely urbanized areas pose significant risks to infrastructure, services, and public safety, motivating the development of operational Landslide Early Warning Systems (LEWS). Within a collaboration between ISPRA and the Civil Protection Department of the Municipality of Rome, three sites affected by rainfall-induced shallow landslides are currently being investigated to support early warning activities. Among them, the Mt. Mario site is of particular interest due to the occurrence of two severe wildfires (July 2024 and June 2025) that damaged vegetation over approximately 12.5 ha, followed by intense rainfall events that triggered shallow landslide trenches and scarps. Such disturbances alter hydro-mechanical soil properties by modifying root reinforcement, hydraulic conductivity, ash deposition, and surface runoff dynamics, thereby affecting slope stability.
In collaboration with the Civil Protection Department, a multi-scale monitoring strategy has been deployed across the three sites, including IoT in-soil sensors (soil moisture, water potential, biaxial clinometers), meteorological stations (rainfall intensity, solar radiation, temperature, humidity, wind), piezometer, inclinometer, and geophysical surveys. Concurrently, an extensive laboratory campaign is characterizing the site through direct shear tests, permeability measurements, soil water retention curves, and physical property analyses on undisturbed samples – typically made of silty/clayey sands. During the first monitoring year, the aim is to assess the coupled hydro-mechanical response of the slope under varying meteorological conditions.
Digital twins of the monitored sites are being developed by combining finite-element and limit-equilibrium modelling to investigate the behaviour of the slopes in consequence to external meteorological inputs, vegetation presence and root decay, and set the basis for threshold definition for LEWS. The final goal of the study is to inform the design of warning thresholds, optimize sensor deployment, and improve risk mitigation strategies for urban slopes.
How to cite: Fraccica, A., Maggi, M., Bonasera, M., Chiessi, V., D'Angiò, D., Niceforo, D. M. A., Ruscito, V., Ferri, G., and Romeo, S.: Early Warning Systems for Landslides in the urban area of Rome (Italy): an integrated approach, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-17898, https://doi.org/10.5194/egusphere-egu26-17898, 2026.
We present a collaborative research initiative involving Politecnico di Torino (Italy), Ilia State University (Georgia), and Telavi State University (Georgia) aimed at developing and implementing customized Early Warning Systems (EWS) for landslide risk mitigation. The research is carried out at strategic pilot sites in Georgia, a region characterized by high geological complexity and significant susceptibility to slope instability. Two Georgian pilot sites are located in the Kakheti Region, within the Gombori Range of the Alazani River basin (eastern Georgia), and in the Vere River basin, a right-bank tributary of the Kura River, southwest of the capital city, Tbilisi.
A first phase focuses on site characterization based on integrated geological, geophysical, geotechnical, and geomatics surveys. This phase aims to define lithological sequences, material properties, and slope geomorphological features to identify the dominant failure mechanisms. The geomatics methodology involves the use of GPS devices, photogrammetric analysis and drone-based LiDAR surveys. The adopted geophysical methods mainly combine electrical resistivity tomography (ERT) and seismic refraction and surface wave analyses. Geological characterization and modeling further include lithological and structural analyses, identification and mapping of existing landslide and debris-flow bodies using photogrammetry and satellite image analyses, estimation of the approximate volume of mobilized sediments in different catchments within the study area, and the collection of geological information required to model the potential distribution of landslide-related debris flows.
The second phase addresses the EWS design for the development of a monitoring framework primarily based on geophysical and geomatics methodologies. Attention is also given to the monitoring of the landslide-induced microseismicity associated with fracture processes and slope movements. The EWS is customized with sensor configuration and threshold parameters specifically designed for the types of landslide phenomena identified during the first phase. The overall goal is to develop a methodology for protecting infrastructures and local communities from landslide triggering; by integrating multi-sensor data fusion with site-specific geological and hydogeological models, the project aims to establish a robust framework for real-time monitoring and early warning, providing a scalable approach to landslide risk management.
How to cite: Godio, A., Colombero, C., Chiampo, F., Di Toro, L., Fiorucci, A., Strallo, V., Vagnon, F., Merebashvili, G., Sukhishvili, L., Gogoladze, S., Akubardia, D., Javakhishvili, Z., Boichenko, G., Lazariashvili, E., Vardoshvili, R., Tsiklauri, D., Davitashvili, M., and Berdzenishvili, N.: Integrated multi-disciplinary approach for landslide Early Warning Systems: a collaborative framework in Georgia, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-10019, https://doi.org/10.5194/egusphere-egu26-10019, 2026.
This review is timely in offering a comprehensive assessment of rainfall-induced landslide early warning systems through the perspective of the United Nations Early Warnings for All framework. Existing rainfall-induced early warning systems are operational in a limited number of settings and are unevenly distributed geographically. Across different implementation levels, locally based systems are frequently fragmented and operationally burdensome. Most functioning systems prioritise debris flows and shallow landslides and rely predominantly on rainfall-based thresholds. Although susceptibility mapping is commonly included, explicit risk mapping remains largely neglected. Real-time monitoring using instruments such as piezometers and inclinometers is present in some systems but is constrained by substantial maintenance demands, which restrict wider deployment. Persistent challenges include limited data availability, the absence of harmonised forecasting methodologies, insufficient forecast validation, and the underutilisation of artificial intelligence, all of which undermine overall system robustness. Engagement with communities and relevant stakeholders is generally weak, and consideration of multi-hazard environments is rare. This review highlights a range of critical areas requiring further development and underscores the importance of collaborative, context-sensitive, and geographically adaptable approaches to advance reliable and inclusive landslide early warning systems at a global scale. While the Early Warnings for All initiative offers a potentially transformative framework, its application to landslide early warning remains constrained by funding limitations, inadequate localisation, and enduring regional inequalities. Without prioritising regionally tailored strategies and securing sufficient resources, the universal establishment of effective landslide early warning systems is unlikely to be achieved.
How to cite: Salam, R., Ahmed, B., and Sammonds, P.: A Review of Rainfall-induced Landslide Early Warning Systems in the Context of Early Warnings for All Framework, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-20522, https://doi.org/10.5194/egusphere-egu26-20522, 2026.
Regional Landslide Early Warning Systems (LEWS) typically rely on rainfall thresholds, that correlate precipitation data with past landslide occurrences to forecast future events. While these systems are simple and accessible, they often lack spatial resolution and fail to capture the complex relationships driving landslides, as they consider only rainfall as input, neglecting critical hydrogeological soil properties. On the other hand, Machine Learning (ML) techniques offer the advantage of incorporating multiple geoenvironmental factors, and have been widely applied to generate landslide susceptibility maps. However, these methods are constrained to spatial predictions, limiting their applicability to LEWSs.
This study presents a dynamic ML methodology using the Random Forest (RF) algorithm to generate daily Landslide Hazard Maps (LHMs), which allow to predict the probability of landslides occurrence in both space and time. The proposed approach integrates dynamic rainfall data (both daily and antecedent rainfall) with static geoenvironmental attributes.
The proposed dynamic methodology involves using a temporally-explicit landslide inventory and identifying non-landslide events over time and space. This allows the inclusion of dynamic variables, such as daily and antecedent rainfall, in the model. It also allows the inclusion of traditional static parameters such as lithology and geomorphologic attributes.
Key innovations achieved are: (1) integration of dynamic rainfall variables as model input, (2) interpretation of model decisions through Partial Dependence Plots (PDPs) to assess their geomorphological plausibility, (3) iterative training on imbalanced datasets to improve predictive accuracy, and (4) the identification of a warning criterion for integrating the generated LHMs into a prototype LEWS.
The methodology was applied using the ITALICA landslide inventory, which provides spatiotemporal information for each event, along with satellite-based precipitation data (GPM IMERG). The use of slope units instead of pixels enhances the representation of geomorphological processes. The model was trained and tested in the Ligu-C Alert Zone (Liguria, Italy), an area with complex geology and high annual rainfall (>3000 mm). Subsequently, the generated predictor model was applied to an independent dataset to obtain daily LHMs for the period February-March 2024, (a period affected by several landslide events), demonstrating the predictive capabilities of the model.
Results confirm the potential of dynamic RF models to overcome the limitations of static ML approaches, providing actionable and interpretable outputs for operational LEWS.
How to cite: Nocentini, N., Rosi, A., Segoni, S., Gariano, S. L., Brunetti, M. T., Peruccacci, S., Melillo, M., Palazzolo, N., Peres, D. J., and Cancelliere, A.: A dynamic machine learning–based early warning system for daily landslide hazard prediction, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-7877, https://doi.org/10.5194/egusphere-egu26-7877, 2026.
Rainfall-induced landslides are among the most damaging natural hazards in humid and mountainous regions, posing severe threats to human life, infrastructure, and environmental systems. Despite extensive research, accurately identifying the rainfall conditions that trigger slope failures remains a major scientific challenge, particularly during extreme hydrometeorological events. Between late April and early May 2024, the state of Rio Grande do Sul (southern Brazil) experienced an unprecedented rainfall episode, resulting in one of the largest documented clusters of rainfall-triggered mass movements in the world. Over 15,000 landslides were recorded across an area of approximately 63,000 km², causing severe social, environmental, and economic impacts. The spatial extent of the affected area enables an investigation into whether spatial variability can be used in place of temporal information to establish intensity–duration thresholds for landslide triggering.
This study investigates the rainfall characteristics associated with landslide triggering during the 2024 extreme event, with emphasis on the temporal, spatial, and pluviometric conditions preceding and coinciding with slope failures. The research integrates high-resolution rainfall records from selected meteorological stations with detailed landslide occurrence data obtained through field campaigns and interviews with residents directly affected by the event, allowing the reconstruction of failure timing with sub-hourly precision. This integration enables a direct comparison between landslide occurrence and rainfall dynamics, including intensity, duration, cumulative rainfall, and antecedent precipitation.
The primary objective is to identify rainfall patterns linked to landslide initiation and to estimate empirical rainfall thresholds, defined as critical values of rainfall intensity and/or accumulation beyond which landslides are likely to occur. Thresholds are derived using conventional intensity–duration and cumulative rainfall approaches, focusing on empirical methods supported by historical observations. Particular attention is given to the role of antecedent rainfall conditions and to the clustering of landslides triggered by a single rainfall episode.
In addition to the temporal analysis, this study investigates the relationship between rainfall thresholds and the spatial extent of affected areas, evaluating whether the magnitude of the impacted area can serve as a complementary or alternative indicator to classical time-based thresholds. Across the landslide-affected area previously identified, landslides were triggered at different times over a three-day period, reinforcing the central hypothesis that spatial variability can be used in place of temporal information under spatially extensive extreme rainfall conditions. Accordingly, a threshold curve was derived from the same extreme event. Despite being based on a single event, the resulting threshold is consistent with intensity–duration relationships commonly reported in the literature.
Overall, these results highlight the potential of spatially informed approaches to refine rainfall threshold analyses during widespread landslide events.
How to cite: Bresolin Junior, N. A. and Michel, G. P.: Spatio-temporal rainfall controls on landslide triggering: can space be used in place of time in threshold definitions?, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-15752, https://doi.org/10.5194/egusphere-egu26-15752, 2026.
The risk of rainfall-induced landslides is expected to rise as climate change intensifies and increases the frequency of extreme precipitation. In this context, territorial landslide early warning system (Te-LEWS) represent effective non-structural measures for landslide risk mitigation at regional scale. Currently, most operational territorial warning models worldwide are based on rainfall thresholds. Since a trigger-cause conceptual framework of hydro-meteorological thresholds was proposed, a growing number of studies report that such thresholds outperform conventional rainfall thresholds. Nevertheless, hydro-meteorological thresholds have rarely been implemented in operational Te-LEWSs, because real-time monitoring of hydrological variables require dense in-situ networks, whereas the use of satellite/reanalysis products is constrained by latency.
Recently, the availability of weather and hydrological forecast products allows incorporating soil moisture information into an operational Te-LEWS. In this work, we present an operational hydro-meteorological warning model developed employing multiple hydro-meteorological thresholds derived from a probabilistic analysis, using soil saturation and precipitation data retrieved from the ERA5-Land product for one of the warning zones defined by Civil Protection for landslide risk management in Campania region, Italy. The performance of the developed model was demonstrated using the real-time forecasts from the Integrated Forecasting System High-Resolution (IFS-HRES) product and compared with the rainfall-only warning model currently operational in Campania in the period 2021–2024. The performance demonstration highlights that the hydro-meteorological model outperforms the regional model, reducing false alarms by 4.1% and shortening the duration of first warning levels not associated to landslides. In addition, the hydro-meteorological model decreases missed alarms by 1.2% and detects a large landslide event missed by the regional model.
How to cite: Zhang, S., Pecoraro, G., and Calvello, M.: Performance demonstration of a hydro-meteorological warning model for landslides at regional scale, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-9410, https://doi.org/10.5194/egusphere-egu26-9410, 2026.
Atmospheric rivers (ARs) are long, narrow, and transient corridors of intense water vapor transport in the lower atmosphere. By driving precipitation in the mid-latitudes, ARs sustain freshwater supply but also cause precipitation-induced disasters such as floods and landslides. In western North America, where precipitation-induced landslides (PILs) are a major geological hazard, ARs have been identified as key drivers of the precipitation regime and frequent precursors of landslide activity. Yet their value as predictors of PILs remains unknown.
In this study, we assess whether AR conditions can inform landslide early-warning efforts in western North America. We employ PIKART—a state-of-the-art AR catalogue at 0.25° and 6-hourly resolution—and a compilation of landslide catalogues across the region—the USGS Landslide Inventories Across the United States, the NASA Cooperative Open Online Landslide Repository, and the Preliminary Canadian Landslide Database—to investigate the association between ARs and PILs from 1996 to 2018. Based on their intensity and persistence, we classify ARs on an AR-strength scale from AR1 to AR5 and analyze how PIL occurrence varies across these strength ranks.
We find that AR conditions preceded more than 80% of days with reported PILs along the West Coast, yet most landslides were associated with weak, primarily beneficial ARs. Both isolated ARs and multi-event AR families contributed comparably to PIL occurrence. Despite this high co-occurrence, ARs exhibit little predictive power because most ARs do not trigger landslides: forecast skill is below 4% across most landslide-prone locations and does not exceed 15% even in regions with dense reporting, such as Portland, Oregon. Although neither the most frequent nor the most hazardous, moderate ARs of rank AR3 show the highest predictive skill. These results reveal a fundamental disconnect between the prevalence of ARs before landslides and their ability to predict them, highlighting both the challenges and opportunities of AR-based landslide forecasting in western North America.
How to cite: Vallejo-Bernal, S. M., Luna, L., Wolf, F., and Kurths, J.: Atmospheric rivers are common precursors but poor predictors of precipitation-induced landslides in western North America, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-18577, https://doi.org/10.5194/egusphere-egu26-18577, 2026.
The landslide meteorological early warning model based on empirical rainfall thresholds(ERT) always has a low warning accurate, and the temporal probability model(TPM) is expected to make up forthis shortcoming. In order to verify this idea, this research conducted a comparative experiment. First, we used accumulated effective rainfall-duration(EE-D) and rainfall on the day-accumulated effective rainfall in the previous 4 days(R0-AE4) as variables to construct two sets of TPM models, the receiver operating characteristic(ROC) curve and correlation coefficient were then used to evaluate the discriminative and predictive abilities of ERT/TPM. Then，the conditional probability formula was used to couple the spatiotemporal probability of landslides, and a probabilistic landslide meteorological early warning model(P-LEWM) was proposed. Finally, through the way of simulated warning, P-LEWM was compared with the matrix-based landslide early warning model(M-LEWM), which was constructed with ERT, the results show that: (1) The ERT/TPM constructed by R0-AE4 is more accurate in judging the hazard level of rainfall to trigger landslides, the area under the ROC curve increased by 6.8% to 12.5% compared to EE-D, (2) The TPM proposed in this paper can predict the probability of rainfall triggering landslides accurately, the correlation coefficient between the predicted amount of triggering-rainfall and the recorded amount is above 0.83，moreover, the EE-D type TPM is more accurate for heavy rainfall prediction, while the R0-AE4 is more suitable for regular rainfall events, (3) The EE-D type ERT will underestimates the hazard level of long-lasting heavy rainfall triggering landslide, which caused M-LEWM missed lots of landslides which happened in two typical rainfall events in 2018, with an missed rate of more than 50%, while P-LEWM constructed with TPM has a correct alert rate of over 90%, (4) Because of the accurate TPM and reasonable spatiotemporal model coupling method, the correct alert rate of the P-LEWM proposed in this article has been significantly improved compared to M-LEWM, the correct alert rate increased by 20.7% to 26%, the reasonable correct alert rate increased by 15.6% to 28.6%, and the missed alert rate decreased by more than 20.5%.
How to cite: Song, Y.: Refined Meteorological Early Warning for Rainfall-Induced Landslide Based on Probabilistic Rainfall threshold, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-11334, https://doi.org/10.5194/egusphere-egu26-11334, 2026.
Rainfall-induced landslides are one of the most widespread and destructive types of geohazards worldwide, and improving the spatiotemporal accuracy and timeliness of early warning remains a persistent challenge in disaster risk management. This study develops a three-dimensional landslide rainfall threshold model framework based on hourly rainfall time series using 216 rainfall-triggered landslide events recorded in Anhua County, China, during 2022–2024. We further provide a systematic assessment of how rainfall observations with different spatiotemporal resolutions, including regional automatic weather stations (RWS), national meteorological stations (NMS), and GPM satellite precipitation, affect threshold-model performance. The proposed three-dimensional framework is then compared against conventional two-dimensional threshold models, including the intensity–duration (I–D), cumulative event rainfall–duration (E–D), and cumulative event rainfall–intensity (E–I). The results indicate that the spatiotemporal resolution of rainfall data is the key determinant of warning performance. The three-dimensional model driven by RWS performs best, achieving a false negative rate (FNR) of 7.46% and a minimum description length (MDL) close to zero (−0.03), and significantly outperforming the counterparts based on NMS and GPM. Moreover, the proposed three-dimensional models also remain stable under both short-duration, high-intensity rainfall and prolonged, cumulative rainfall conditions, with overall performance consistently superior to that of the two-dimensional models. These findings demonstrate that an hourly three-dimensional threshold model supported by high spatiotemporal rainfall observations can substantially improve the accuracy and timeliness of landslide early warning, providing an effective methodological basis for more precise regional warning of rainfall-induced landslides.
How to cite: Sun, Q., Xiao, T., and Liu, X.: Three-Dimensional Landslide Rainfall Threshold Model Driven by Multi-Source Rainfall Data: Development and Performance Evaluation, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-15739, https://doi.org/10.5194/egusphere-egu26-15739, 2026.
Taiwan is located at the junction of the Eurasian Plate and the Philippine Sea Plate and has an extremely complex geological structure, resulting in frequent earthquake activity. Taiwan is situated in the central region of the Northwest Pacific, where typhoons often form and develop during the summer and autumn seasons. The combination of heavy rainfall and earthquakes exposes Taiwan’s mountainous regions to landslides and debris flow disasters, which cannot be completely prevented through engineering measures and significantly impact people's lives and property. Therefore, an effective debris flow warning system is urgently needed. In this paper, the maximum hourly rainfall depth (Im), the maximum 24-h rainfall amount (Rd) and RI (RI=Im ✕ Rd) were analyzed, and the relationship between RI and debris flow triggering is presented. The rainfall-based warning model RI was compared with the RTI model, which is currently adopted by the Taiwanese government. The RTI model is defined as the product of hourly rainfall intensity and the sum of the 24h-rainfall depth and prior-rainfall depth. The two models were applied to the Chen-Yu-Lan River Watershed in Nantou County, central Taiwan, to evaluate the debris-flow occurrence probability during several extreme rainfall events. The results show that the RI model can effectively evaluate temporal variations in debris-flow occurrence probability in response to hourly rainfall intensity during a rainfall event.
How to cite: Huang, W.-S., Chen, J.-C., Fan, J.-Q., Lai, X.-Z., Li, F.-B., Zhang, X.-D., and Li, G.-L.: A Warning Model of Rainfall Characteristics for Debris Flow Occurrence in Taiwan, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-16712, https://doi.org/10.5194/egusphere-egu26-16712, 2026.
Rapid movements of sediment mass (e.g. shallow landslides, debris flows, rockfall) pose an imminent risk to settlements, infrastructure and human life in mountain regions. Forecasting such intermittent hazards on a large-scale is still challenging, yet essential to ensure effective risk management. Landslide early warning systems can benefit from the predictive power of dynamic hydroclimatic controls to better anticipate the initiation of these events.
This study characterizes distinct hydroclimatic triggering conditions for rapid alpine mass movements, their exceptionality, and their predictability in time.
We base our analysis on an inventory of ca. 1900 observations of shallow landslides, debris flows, and rockfalls in the Swiss Alpine Rhine basin (approx. 4300 km²) over the past 25 years. Utilizing hydrometeorological time series derived from gridded soil and climate products at a 1×1 km spatial and daily temporal resolution, we retrieve distinct families of predisposing and triggering conditions allowing us to objectively identify different process types. Our results show that a significant proportion of events are not exclusively rainfall-driven: approximately 20% of both shallow landslides and debris flows occurred under the influence of snow cover and snowmelt, suggesting that the hillslope response to precipitation and soil wetness varies seasonally. This underscores the necessity of a multivariate and sequential modeling approach.
In a second step, we expand the methodology into a data-driven modelling framework by employing a recurrent neural network (long short-term memory LSTM). It simulates the probability of mass movements occurring over time by decoding the temporal dynamics of the catchment’s hydroclimatic conditions. We demonstrate the algorithm’s potential to internally reproduce hydrogeomorphic catchment states based solely on input time series of precipitation, temperature, and soil wetness. We report an area under the curve-receiver operating characteristic (AUC-ROC) metric of 0.94 (landslides) and 0.84 (debris flows) for testing.
The findings of this study offer novel insights into hydroclimatic and hydrogeomorphic controls on the predisposing and triggering conditions of rapid alpine mass movements. Modern computational techniques allow to simulate seasonally varying contributions of multivariate hydrometeorological variables to the initiation of such events. This will enable predictions of changes in sediment mass movement distributions under a future climate and will offer an opportunity for plugging into early warning systems for landslides.
How to cite: Demmel, S., Mair, D., and Molnar, P.: Predictive power of hydroclimatic controls on the initiation of rapid alpine sediment mass movements, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-20514, https://doi.org/10.5194/egusphere-egu26-20514, 2026.
Landslides are a widespread hazard in Italy, and Early Warning Systems (LEWS) help mitigate this risk through non-structural measures. Recent advances in monitoring and data analysis have improved LEWS but identifying spatially and temporally variable triggering factors remains challenging. Integrating low-cost GNSS with precipitation networks can enhance system reliability. In this study, a continuous early warning system focusing on rainfall as a triggering factor was applied to a complex deep-seated landslide in the Carnic Alps, north-eastern Italy. The Cazzaso landslide monitoring system, installed in 2016 by the CRS (OGS) in collaboration with the Regional Civil Protection, continuously collects displacement data from 12 GPS and 2 GNSS stations. Time series of displacement and precipitation data from two rain gauges were analyzed to identify landslide reactivation events using a velocity threshold—a novel approach that provides valuable insights for updating LEWS protocols. The Cazzaso landslide was found to be primarily rainfall-triggered, leading to the application of empirical Intensity–Duration (I–D) rainfall thresholds for early warning. Validation showed limited reliability, likely due to the landslide’s complex geometry and depth, which are not fully captured by simple statistical methods. To address this, a Random Forest (RF) model combined with Explainable AI (XAI) techniques was employed. Out-of-Bag Error (OOBE) assessed variable importance, and Partial Dependence Plots (PDPs) illustrated their influence. The analysis identified 8-day cumulative rainfall as the most effective predictor of landslide reactivation, enabling the definition of more reliable thresholds for the GNSS-based warning system. This integrated approach improves the operational effectiveness of LEWS and can be adapted to evaluate short- and long-term rainfall impacts in diverse geological and climatic contexts. While site-specific, the methodology provides a transferable framework for other landslide-prone areas.
Acknowledgements
This study was carried out within the RETURN Extended Partnership and received funding from the European Union Next-GenerationEU (National Recovery and Resilience Plan – NRRP, Mission 4, Component 2, Investment 1.3 – D.D. 1243 2/8/2022, PE0000005)
How to cite: Rosi, A., Franceschini, R., Nocentini, N., Tunini, L., Zuliani, D., Peressi, G., and Rossi, G.: Integrating GNSS and Explainable AI for rainfall thresholds in large landslide monitoring, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-10010, https://doi.org/10.5194/egusphere-egu26-10010, 2026.
Given the increase in the frequency of disasters caused by landslides due to extreme precipitation events and unplanned urbanization, landslide early warning systems (LEWS) have been shown to be increasingly necessary as effective and cost-beneficial risk-reduction and damage mitigation tools. Recently, the use of machine learning techniques in the prediction of landslide triggering for application in LEWS has shown promise, with several examples in the literature demonstrating good results. However, the need for large volumes of data for training models for this purpose is a considerable obstacle to their broader application, especially in regions that lack good landslide inventories. This study tests the use of civil defense service records related to landslides as a proxy for the actual triggering of landslides, since peaks in the number of service calls are observed during such events. Supervised machine learning models (Support Vector Machine, Multilayer Perceptron, and Random Forest) were used and their performances were compared with that of a LEWS based on empirical thresholds already in operation in the study area. To this end, records from the Civil Defense of Petrópolis, Rio de Janeiro, Brazil, regarding occurrences registered during the period from 2015 to 2019 were obtained, as well as a historical precipitation series from a rain gauge situated in the same municipality with the same temporal coverage. After data processing, which removed spurious readings in both data sources, an input was created whose features consisted of precipitation accumulations and maximum intensities recorded in different temporal windows and whose label was the presence or absence of civil defense records on the same date. The results confirm the potential of using machine learning algorithms in LEWS, since the models based on Random Forest and Multilayer Perceptron presented Recall, F1-Score, and Balanced Accuracy considerably superior to those of the LEWS operating in the municipality. They indicated, as well, the need for improvements to the empirical thresholds used in Petrópolis, particularly the ones for the activation of warning sirens, whose activations were concentrated in only 2 of the 4 thresholds in use.
How to cite: Nonato Vieira Cereto, A., Michel, G. P., Zanandrea, F., and Lache Salcedo, I.: Towards the Development of Machine-Learning-Based LEWS for Data-Scarce Environments, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-1245, https://doi.org/10.5194/egusphere-egu26-1245, 2026.
Landslides are common and highly destructive geological hazards, and accurately identifying landslide-prone areas is of great significance for disaster prevention and mitigation. To address the limitations of traditional landslide susceptibility models—such as insufficient generalization capability, strong spatial heterogeneity, and high predictive uncertainty—this study proposes an integrated landslide susceptibility modeling approach that incorporates spatial matrices and uncertainty analysis. The proposed method ensembles four base models, including Logistic Regression, Random Forest, Maximum Entropy, and a Graph Neural Network. Node-level uncertainty is quantified using prediction variance. Three types of adjacency matrices—geographical, environmental, and prediction-based—are constructed and adaptively fused via an attention mechanism. Within a two-layer graph convolutional network framework, multi-source information is jointly propagated and probability estimates are calibrated. A case study in Linxiang City, Hunan Province, China demonstrates that the proposed model achieves an AUC of 0.937 and a landslide identification rate of 94.96%, significantly improving the accuracy and reliability of landslide recognition.
How to cite: Huang, W. and Xiao, T.: Ensemble Landslide Susceptibility Modeling Based on Spatial-Matrix Coupling and Uncertainty Analysis, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-16507, https://doi.org/10.5194/egusphere-egu26-16507, 2026.
Displacement-based early warning of deep-seated landslides requires displacement forecasts that are not only accurate but also physically consistent under rapidly changing hydrological conditions. Variations in rainfall infiltration, pore-water pressure, and subsurface moisture dynamics can modify the effective stress state and creep rates, leading to complex and often nonlinear displacement responses. Despite extensive modeling efforts, reliable and physically interpretable displacement forecasting remains challenging. Data-driven models often lack process consistency, while physics-based approaches are limited by uncertain parameters and simplified assumptions when applied to real-world conditions. In this study, we develop a physics-informed machine learning model that integrates physical process constraints with landslide monitoring data. A hydrologically driven deformation relationship dominated by seepage-related effects is incorporated as a model constraint to guide the prediction of displacement. The model is trained using cumulative displacement observations and hydrological forcing from an IoT-enabled in situ monitoring system deployed on a landslide, and is subsequently applied to forecast displacement over unseen periods. Results show that embedding physical constraints improves the temporal generalization and physical plausibility of predicted displacement trajectories, particularly during hydrologically triggered acceleration phases. The inferred model parameters exhibit physically interpretable and internally consistent behavior, indicating that dominant hydrological controls on deformation are captured. This framework improves both robust displacement forecasting and physical interpretability, thereby supporting the development of operational landslide early-warning systems.
How to cite: Tang, X., Zhang, Z., Kibirige, D., Wei, Z., Hu, Y., Raj Meena, S., and Catani, F.: A Physics-Informed Machine Learning Model for Displacement Forecasting of Deep-Seated Landslides, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-21578, https://doi.org/10.5194/egusphere-egu26-21578, 2026.
The growing availability of IoT-enabled sensor networks has transformed how slope stability is monitored within Landslide Early Warning Systems (LEWS), producing vast datasets on pore pressures, groundwater levels, displacements, and external drivers such as rainfall. In the literature, there is an increased trend to apply advanced data analysis approaches and surrogate models to support slope stability assessment and early warning. However, the sheer volume and complexity of these data often limit users’ ability to interact with them in a flexible and intuitive way. Emerging advances in multi-modal generative AI models and agentic frameworks suggest a new paradigm: chat-with-your-data.
In this approach, users interact directly with slope monitoring data through natural language, requesting tailored visualizations, summaries, analyses, or forecasts without the need for bespoke coding or rigid workflows. In the context of slope stability assessment and early warning, a practitioner could ask for recent pore pressure trends, rainfall-displacement correlations, threshold exceedances, or anticipated changes in stability conditions based on forecasted meteorological inputs for a specific site. The system identifies the relevant data sources, retrieves data, performs the required operations, and returns insights in user-friendly formats such as maps, diagrams, or downloadable datasets.
The potential benefits include more direct access to relevant data and analyses, uncovering correlations, and enabling real-time decision support. However, challenges remain. These include ensuring that project-level access controls are respected, handling heterogeneous geospatial references, providing tailored data representations across spatial scales, and maintaining transparency and reliability in automatically generated outputs. Addressing these issues requires combining domain-specific knowledge in slope stability and landslide processes with expertise in generative AI and data governance.
This work outlines a vision for how conversational interfaces could enhance slope-scale Landslide Early Warning Systems by supporting monitoring, modelling, and forecasting activities through intuitive human–data interaction. By allowing experts to query their data directly, we move toward systems that are more adaptable, interpretable, and insight-driven, promoting more effective use of monitoring data for targeted warning and risk mitigation.
How to cite: Mathisen, A., Granitzer, A.-N., and Piciullo, L.: Conversational AI for Slope Stability Monitoring: Enabling “Chat-with-Your-Data” as a Decision Support Tool, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-20914, https://doi.org/10.5194/egusphere-egu26-20914, 2026.
Slow-moving landslides are a widespread hazard in coastal and mountainous settings, causing damage to property and infrastructure, and sometimes loss of life. Mechanisms driving rare catastrophic failure events are poorly understood, highlighting the need for effective monitoring systems. Traditional landslide monitoring techniques include remote sensing (e.g. InSAR), geotechnical instrumentation (e.g. piezometers), and geophysical monitoring (e.g. electrical resistivity). Although numerous and varied, traditional methods cannot always provide the high spatiotemporal resolutions required for real-time monitoring. Remote sensing techniques can be spatially and temporally coarse, and ground-based instrumentation is costly and susceptible to damage during ground failure.
We have established a novel IoT-based wireless sensor network (WSN) for slow-moving landslide monitoring which has been operational for 4 years. It consists of motion-triggered, low-power, low-cost inertial measurement unit (IMU) sensors that are embedded in artificial boulders (SlideCubes) and distributed across the landslide body. The sensors communicate via LoRaWAN (Long Range Wide Area Network) with a gateway, and data are uploaded to a server in near real-time. This research focuses on the western portion of the Black Ven-Spittles landslide complex at Lyme Regis, Dorset where a small earthflow propagates from a disused landfill site. The site is a suitable ‘field laboratory’ in which to test the WSN and SlideCubes; the earthflow is self-contained and somewhat isolated from the surrounding complex, reaching comparatively high velocities (c. 52.62 m y-1) and retrogressing westward towards the town allotments, car park and other infrastructure. Our SlideCubes are deployed on the landslide surface and ‘go with the flow’ during gradual failure. Two brands of IMU sensor are deployed across the earthflow, allowing comparison between similar sensors and evaluation of their suitability for monitoring landslides.
The sensors precisely capture motion onset which is transmitted in near real-time. From this, we examine spatial patterns of SlideCube motion and extract relative trigger magnitudes, producing a holistic picture of earthflow failure events as well as a preliminary assessment of potential catastrophic collapse. The IoT network also comprises an onsite rain gauge, with potential for integration of additional sensors, that supplies information about possible drivers of this motion. We draw on third-party meteorological and wave data to further support our process understanding. We categorise types of motion recorded by the IMU sensors and validate this with trail camera imagery, providing insight into the geomorphological processes occurring on the landslide surface and subsurface. Our WSN is a successful test case of low-cost landslide monitoring which has potential for development into a continuously operational early warning system.
How to cite: Newby, K., Bennett, G., Roskilly, K., Luo, C., Manzella, I., and Sgarabotto, A.: Real-time monitoring of slow-moving landslides using novel IoT-based wireless sensor networks, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-424, https://doi.org/10.5194/egusphere-egu26-424, 2026.
Landslides and earthquakes are among the most significant natural hazards, often generating seismic signals with partially overlapping characteristics that complicate reliable discrimination, particularly in mountainous and tectonically active regions. Accurate identification of landslide-induced seismic signals is essential for developing reliable landslide catalogs, improving hazard assessment, and enabling real-time monitoring systems. In this study, we present an advanced machine learning and deep learning–based framework for the classification of seismic signals associated with landslides and earthquakes, using real observational data and dimensionality-reduction techniques. Seismic waveform data were collected from permanent seismic stations operated by the Seismological Observatory and the Earthquake Engineering Department at the Indian Institute of Technology Roorkee. Earthquake events were identified using established regional and global earthquake catalogues, while the landslide catalogue was independently developed by our research group through systematic analysis of seismic records, field evidence, and event validation. This self-developed landslide catalogue provides a high-confidence dataset for supervised learning and represents a significant contribution to regional mass-movement monitoring efforts. The seismic signals were initially characterized using a comprehensive set of signal descriptors derived from previous studies on landslide and earthquake seismology. Approximately 97 time-domain, frequency-domain, and statistical parameters were extracted for each event, capturing waveform amplitude, energy distribution, spectral content, and temporal evolution. While these features effectively describe seismic signal behavior, their high dimensionality introduces redundancy and may degrade classification performance. To address this challenge, multiple Principal Component Analysis (PCA) approaches, including conventional and kernel-based PCA, were employed to reduce dimensionality while preserving the most informative components relevant for class discrimination.
Following dimensionality reduction, advanced machine learning classifiers were applied to distinguish between landslide- and earthquake-generated seismic signals. The classification framework was trained using combinations of real data and synthetically augmented samples generated through CTGAN (Conditional Tabular Generative Adversarial Network) to improve class balance and model robustness. Model performance was evaluated using independent test datasets derived from raw, unseen seismic signals, ensuring a realistic assessment of generalization capability. Across different PCA–classifier combinations, the proposed framework achieved high classification accuracy, consistently exceeding 95% and reaching values close to 97% for optimal model configurations. Precision, recall, F1-score, and ROC–AUC metrics further demonstrate the reliability and stability of the classification results. Importantly, the trained models were validated directly on raw seismic data, highlighting their ability to generalize beyond feature-engineered training sets. This result indicates strong potential for operational deployment. The proposed methodology provides a scalable and automated approach for discriminating landslide-induced seismicity from earthquakes and can be integrated into continuous seismic monitoring systems.
Overall, this study demonstrates the effectiveness of combining seismic signal processing, dimensionality reduction, and advanced machine learning for landslide detection. The developed framework has significant implications for the real-time development of accurate landslide catalogs and offers a promising pathway toward improving early warning capabilities using continuous data streams from regional seismometer networks.
How to cite: Sharma, P. M. L. and Rawat, Dr. D.: Automated Classification of Seismic Signals for Real-Time Hazard Monitoring, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-5459, https://doi.org/10.5194/egusphere-egu26-5459, 2026.
Recent decades have witnessed a marked increase in the frequency, intensity, and cascading disasters, resulting in severe social and economic losses, particularly when coupled with unpreparedness and social vulnerability. This contribution presents theoretical and applied advances in landslide disaster risk reduction, with emphasis on trigger analysis and the transformation of heterogeneous real-time data streams into actionable early warning intelligence.
Amrita’s AI-Enabled Real-Time Landslide Early Warning System (A-LEWS) is designed for the real-time monitoring, detection, and early warning of landslides (Ramesh, 2014). U.S. Patent No. 8,692,668). This system features Intelligent Wireless Probes (IWPs) equipped with various hydro-geophysical sensors, which are deployed deep beneath the earth’s surface to capture critical landslide-triggering parameters in vulnerable areas. The landslide detection system is founded on the integration of hydro-geophysical sensors that directly capture the physical processes governing rainfall-induced slope failure. Pore pressure transducers and dielectric soil-moisture sensors quantify rainfall infiltration, transient pore pressure buildup, and loss of effective stress, which are primary controls on slope instability (Figure 1). Tiltmeters and strain gauges measure slow ground deformation and changes in slope geometry associated with progressive failure, while geophones detect vibration signatures linked to material movement and subsurface fracturing. These heterogeneous sensors are interfaced through enhanced subsurface sensor columns and connected to wireless sensor nodes, enabling in situ, high-resolution monitoring across crown, middle, and toe regions of the slope. Given the constraints of remote deployments, limited power availability, difficult terrain, and long-term operation, the system adopts an energy-aware wireless sensor network design. Low-power operation is further supported by state-based node transitions, time synchronization, and selective high-rate sensing only during elevated-hazard conditions. Together, this sensor science and energy-efficient network architecture enable reliable, scalable, and long-duration landslide monitoring while preserving power resources without compromising early warning capability (Ramesh 2009).
Figure 1: Context-Aware IoT Edge Node Integrated with Adaptive Energy Management & Dynamic Sensor Prioritization
A multilevel warning dissemination architecture ensures timely alerts to the relevant vulnerable community and stakeholders. This system in Munnar, Kerala, has been successfully providing warnings to the community since 2005, 2009, 2011, 2013, 2018, 2020, 2021, 2022, 2023, 2024, and 2025. A scalable version of LEWS has been implemented in Chandmari, Gangtok, Sikkim, where landslides are induced by both rainfall and earthquakes. Fully deployed in 2018, this system includes 11 IWPs with over 200 geophysical sensors. The system has been operational, with continuous monitoring, analysis, and reporting to the Sikkim State Disaster Management Authority (SSDMA). The effectiveness of the system in issuing successful warnings and supporting informed decision-making is illustrated in Figure 2.
Figure 2: Landslide Early Warnings Issued in 2020 for the Munnar area, Idukki, Kerala
How to cite: Ramesh, M. V., vasudevan, N., kumar, S., hariharan, B., kumar, N., tirugnanam, H., pullarkat, D., singh, B., Guntha, R., Ug, G., Premaja kalesan, I., Jijilal, A., Madhu, D., and Rangan, V.: A Multilevel AI-IoT Operational System for Landslide Early Warning: Transitioning from Heterogeneous Data Streams to Actionable Risk Intelligence, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-20814, https://doi.org/10.5194/egusphere-egu26-20814, 2026.
Posters on site: Fri, 8 May, 08:30–10:15 | Hall X3
Geodynamics of Himalaya is always associated with intense rainfall, aggressive slopes, fragile lithology, and active tectonism, in which linear infrastructure expansion has combined and amplify the landslide risk at threshold levels. 15 % land area of India (including snow cover) is prone to landslide hazards, in which Uttarakhand state is the most susceptible part of the country. According to the Geological Survey of India, Uttarakhand, has witnessed 4,654 landslides, 92 avalanches, 67 cloudbursts and 12,758 flood events which resulted as over 1,200 fatalities and 1.3 billion US dollars damage between 2015 to 2025. This underlines a major wake call for the geoscientists and policy makers regarding settlements in and around himalayan regions.
This work focuses on preparing a Landslide Hazard Zonation (LHZ) map/model for the NH-109K corridor of Lesser Himalaya, India using Geographic Information System (GIS) and Machine Learning (ML) techniques. The approach involves integrating multiple geo-environmental and terrain parameters that influence slope-instability. The primary thematic layers considered in this study include slope, aspect, rainfall, normalized difference vegetation index (NDVI) and land use/land cover (LuLc). Additional factors such as lithology, drainage density, proximity to roads, and rainfall are also incorporated. SRTM DEM and Sentinel 2 satellite imagery are used to derive topographic and derivative parameters, while rainfall and landslide inventory are obtained from Indian Meteorological Department and Bhukosh portal (Open-source data archive of Geological Survey of India). The thematic layers are standardized, weighted, and integrated within the GIS environment and simulated into data driven Machine Learning environment to establish their spatial association with observed landslide occurrences. Through this integration, the study aims to delineate zones exhibiting varying degrees of landslide prone across the NH-109K.
The resulting LHZ map categorizes the area into five susceptibility zones (very high, high, moderate, low and very low) reflecting the degree of terrain instability. The work emphasizes the significance of ML techniques in assessing complex natural hazards like landslides. Such an approach contributes to informed decision-making for infrastructure development and hazard mitigation in mountainous regions. Adopted methodologies also holds potential for replication in other mountain-corridors facing similar geomorphic and climatic conditions. Thus, this study supporting sustainable and resilient road network planning in landslide-prone areas with special reference to the Lesser Himalayan belt of India.
Keywords: GIS, Landslide, Lesser Himalaya and Machine learning (ML).
How to cite: Mishra, P., Singh, R., Sharama, P., and Dwivedi, S.: ML and GIS approach for Landslide Hazard Assessments in Lesser Himalaya, India, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-654, https://doi.org/10.5194/egusphere-egu26-654, 2026.
Rainfall-induced landslides are a major hazard in mountainous regions such as the Pyrenees, where intense or prolonged precipitation frequently triggers slope failures. The SPIRAL project (EFA039/01,POCTEFA 2021–2027) aims to improve preparedness and response capacity by developing an operational Landslide Early Warning System (LEWS) that integrates meteorological and geological data streams into a unified workflow for civil protection agencies in Spain, France, and Andorra.
The system combines dynamic rainfall information—observed and forecast—with static susceptibility maps to estimate hazard levels at two scales: territorial (1 km²) and regional (30 m). Data sources include rain gauge networks (AEMET, SMC, Meteo-France, CHE), radar observations, and numerical weather prediction models (ECMWF-IFS, Harmonie). Observed precipitation is processed hourly, generating accumulations over 1 h, 6 h, 12 h, and 24 h. For real-time analysis, rainfall fields are derived using inverse distance weighting (territorial domain) and Conditional Merging of radar and gauge data (regional domain), ensuring spatial continuity and quantitative accuracy. Forecast horizons up to 72 h are incorporated using ECMWF outputs blended with radar-based nowcasting to maintain temporal consistency.
Hazard estimation relies on decision matrices that cross rainfall thresholds with susceptibility values for landslides and rockfalls. Products are generated in raster and slope-unit formats at both scales. Each hour, the system updates hazard maps and computes maximum risk levels across all accumulation intervals. Alerts are classified into four qualitative levels (Very Low, Low, Medium, High) and visualized through the Argos platform—a cloud-based multi-hazard early warning system enabling real-time monitoring, intuitive map visualization, and automated notifications to civil protection agencies.
SPIRAL demonstrates the feasibility of integrating heterogeneous data streams into a unified operational workflow. Key innovations include: (i) dynamic blending of observed and forecast precipitation for seamless short-term prediction; (ii) multi-scale hazard modeling combining susceptibility and triggering factors; and (iii) full interoperability with existing risk management platforms. Preliminary tests using historical rainfall episodes confirm the system’s ability to capture spatial and temporal variability of hazard conditions, supporting timely decision-making for emergency response.
Future developments will focus on refining rainfall thresholds, incorporating real-time in-situ monitoring (e.g., piezometers, crackmeters), and validating performance under operational conditions. This work contributes to advancing LEWS design by coupling meteorological forecasting with geospatial susceptibility analysis in a transboundary mountain environment.
How to cite: Marturià, J., Becerra, J., Buxo, P., Yannick, T., Colas, B., Echevarria, A., Guerrero, J., and Gasc, M.: Integrating Meteorological and Geological Data for Landslide Early Warning in the Pyrenees: The SPIRAL Project, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-21702, https://doi.org/10.5194/egusphere-egu26-21702, 2026.
The eruption of the Tajogaite volcano on the island of La Palma lasted for 85 days, between September and December 2021. Large quantities of lava and pyroclasts were emitted, exceeding 200 Mm3. It has been estimated that 45 Mm3 corresponded to pyroclasts.
The stability conditions of the new Tajogaite volcanic edifice were analysed to determine its hazard potential. The volcanic cone is composed of pyroclastic materials, mainly lapilli (2-64 mm) and scoria (> 64 mm), with intercalated layers of ash (< 2 mm) and encrusted sulphate and carbonate precipitates. The estimated height of the cone reaches 200 m, with slopes of 30-35º, which have fractures that favour the emission of gases. The stability analysis under unsaturated conditions yielded a safety factor of 1.2, which in geotechnical terms is equivalent to stable conditions; however, under saturated conditions, the safety factor is less than 1.00, indicating instability or failure under very heavy rainfall.
According to rainfall records and historical data, this region could experience heavy rainfall of more than 100 mm in several hours, with a possible frequency of once every 10 years, and exceptionally, accumulated rainfall of more than 400 mm could occur over several days in a 50-year interval. If these conditions occur, the pyroclastic materials of the cone may become saturated and unstable, and lahars may occur.
Given the risk of lahars, whose probability is low or very low, and the instability of the volcanic cone slopes, this volcano has been included as a study area within the PRISMAC project, which plans to establish an early warning system for landslides using geospatial technologies.
The PRISMAC project (1/MAC/2/2.4/0112), co-financed by the INTERREG VI D Madeira-Azores-Canary Islands MAC 2021-2027 Territorial Cooperation Program, aims primarily to analyze, mitigate, and manage natural hazards, with a particular focus on landslide movements, which are increased by the effects of climate change. To achieve this, harmonized methodologies for susceptibility and risk analysis are being developed, enabling the identification of high-risk areas within the participating Macaronesian regions. This will facilitate the creation of monitoring systems, early warning, and alarm mechanisms, which are essential for reducing the impact of these phenomena on populations and infrastructure.
The early warning and alarm system proposed by PRISMAC is based on the development of algorithms that take into account critical rainfall thresholds, combined with aerial geospatial techniques (drones with LiDAR systems), terrestrial techniques (high-precision 3D laser scanners) and satellites (Sentinel-1 radar using the InSAR technique to detect millimetric ground movements).
How to cite: Hernández-Gutiérrez, L. E., Pérez-Martín, Ó., González de Vallejo, L. I., Medina-Dávila, J., Padilla, G. D., Ortega, V., Álvarez, A., García-Hernández, R., Hernández, P. A., Hernández-Martín, H., and Pérez, N. M.: Early warning system for monitoring landslides of pyroclast and lahars from the 2021 eruption of the Tajogaite volcano on the island of La Palma, Canary Islands, Spain, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-8069, https://doi.org/10.5194/egusphere-egu26-8069, 2026.
Rainfall-induced natural hazards are widespread worldwide and pose significant threats to society and infrastructure. Reliable assessment of these hazards strongly depends on the availability and quality of rainfall information. In regions characterized by complex topography, rainfall patterns are highly heterogeneous, which complicates hazard evaluation. This is particularly evident in the mountainous areas of Campania (southern Italy), where pyroclastic soil deposits are widespread and rainfall-triggered shallow landslides and debris flows frequently occur. In such settings, the spatial variability of rainfall plays a crucial role in controlling the spatio-temporal distribution of landslides, affecting the performance of hazard assessment tools.
This study investigates the spatial variability of rainfall at the event scale in the Partenio Massif and the Sarno Mountains. The study area is characterized by coarse-grained pyroclastic soils, consisting of variable layers of volcanic ash and pumice deposited over densely fractured limestone bedrock. Rainfall records from 23 rain gauges operating between 2002 and 2024 were used to define rainfall event series. Rain events were separated using a minimum inter-event time of 24 hours with rainfall amount lower than 2 mm. The study area was subdivided into zones by grouping rain gauges that share the same probability distribution of rainfall event depth and duration, as identified through Kolmogorov-Smirnov tests.
Within each defined zone, the Pearson correlation coefficient and the spatial variability of rainfall were evaluated for all pairs of rain gauges, considering both rainfall depth and duration of events overlapping for at least one hour. Strong correlations were observed for both depth and duration among closely located rain gauges. However, both the correlation strength and the number of overlapping events progressively decreased with increasing inter-station distance. For each pair of stations, the differences in rainfall depth and duration of overlapping events at two stations were found to be normally distributed around their mean values, with a clear dependence of the standard deviation on the square root of the mean. Moreover, the standard deviation was observed to increase following a power-law relationship with inter-station distance across all zones.
The outcomes of this study provide a quantitative basis for incorporating rainfall spatial uncertainty into hydrometeorological models for rainfall-induced hazard assessment over large areas. Additionally, the results offer valuable insights for optimizing rain gauge network design, contributing to the development of more effective early warning systems.
How to cite: Abdullah, A., Roman Quintero, D. C., Marino, P., and Greco, R.: Investigating the event-based rainfall spatial variability for reliable geohazard assessment, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-7415, https://doi.org/10.5194/egusphere-egu26-7415, 2026.
The Carpathian region of Ukraine is significantly at risk of landslides attributed to its complex geology, steep and rugged topography, high levels of precipitation, and human-induced alterations in land use. This modelling employed the CatBoost algorithm to evaluate landslide susceptibility in the Transcarpathian region (Zakarpattia Oblast) of Ukraine, and comprised two phases, along with a performance comparison. A landslide inventory featuring 697 recorded landslides was utilized, with a data split of 70:30. In the initial phase, ten predisposing factors were utilized, and multicollinearity was assessed based on Variance Inflation Factor (VIF) values to confirm that correlated factors were absent. Subsequently, the modelling was implemented, and the performance was evaluated.
In the second phase, the Boruta feature selection algorithm was applied to eliminate irrelevant factors. The CatBoost-based modelling was executed again, and the predictive performance was assessed. Finally, the performance of the models was compared to analyze how it varies before and after the implementation of the Boruta algorithm. The performance of the models was analyzed using the Receiver Operating Characteristic (ROC) curve and other metrics, including Accuracy, F1-score, Precision, and Recall.
All ten factors yielded VIF values under the threshold of 10, and consequently, they were retained for modelling. Before the implementation of the Boruta algorithm, the model exhibited poor performance, with an area under the ROC curve (AUC) value of 0.644 (64.4%), an Accuracy of 0.600, an F1-score of 0.643, a Precision of 0.614, and a Recall of 0.674. The Boruta-based selection led to the rejection of four irrelevant predisposing factors; consequently, six factors qualified for subsequent analysis. The performance after applying the Boruta algorithm is as follows: a fair AUC value of 0.731 (73.1%), an Accuracy of 0.683, an F1-score of 0.725, a Precision of 0.676, and a Recall of 0.781. The model performance improved by 0.087 (8.7%) in AUC, 0.083 in Accuracy, 0.082 in F1-score, 0.062 in Precision, and 0.107 in Recall.
Despite the improvement in performance, the model did not yield superior evaluation scores. A possible reason is the constraint related to the quality of input data, which ongoing research is attempting to resolve by refining datasets and updating landslide inventories. However, the enhancement emphasizes the need for accurately selecting relevant factors in generating robust outputs. Moreover, the application of machine learning techniques in the Transcarpathian region, where there are limited methodological advancements, signifies a crucial advancement for landslide risk management in Ukraine. The insights from this modelling are instrumental as a preliminary step towards the future design of regional-scale early warning systems.
How to cite: Ajin, R. S., Gatto, A., Nocentini, N., Hadiatska, K., Ivanik, O., Kravchenko, D., Petrushenko, E., and Fanti, R.: Enhancing landslide susceptibility modelling through feature selection: A machine learning approach in the Ukrainian Carpathians, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-16568, https://doi.org/10.5194/egusphere-egu26-16568, 2026.
This is apreliminary study to better refine Landslide Early Warning Systems effectiveness about very severe events. When issuing an alert at the maximum level possible, missed alarms may result in casualties, increased damages and delayed response, but also false alarms may have consequences that cannot be ovelooked (including economic costs of countermeasures activated in vain, suspended services, and a generalized loss of trust in the system, which undermines the effectiveness of the future warnings).
We therefore focus on the spatial patterns of major geo-hydrological disasters across Italy (for which national-level emergencies were issued), using an innovative target variable (Months in Emergency State - MES), which captures both the recurrence of disasters and the persistence of their impacts.
As explanatory variables, we initially consider 62 potential predisposing factors from different fields: environmental, territorial planning, soil sealing, and
socio-economic. A three-step feature selection process based on Pearson correlation, multicollinearity analysis, and ReliefF algorithm, was applied to reduce redundancy and identify the most relevant predictors (18), which were used in a CatBoost regression model.
Results highlight that combining parameters from different fields significantly improves model performance. Surprisingly, anthropogenic factors, such as territorial planning and socio-economic indicators, had a greater influence than physical characteristics in driving the recurrence of disasters and the persistence of their impacts.
A further analysis on the results (by means of Partial Dependence Plots) highlighted very complex and somehow counterintuitive relationships.
The most important driver is the amount of soil sealing in areas classified as “medium hazard” for landslides or floods. This factor is directly and sharply related to MES (more than high-hazard areas), suggesting a need to revise hazard classifications or existing planning regulations. Gross Domestic Product (GDP - a proxy for wealth and productivity) ranks second, showing a mixed effect: while wealthier areas face higher exposure, they also show
stronger resilience. TWI, a hydrological indicator, shows that disasters are more linked to minor watercourses than to large rivers, advising to reconsider the mitigation priorities.
This study provides new insights on hydro-geological disasters and the complex non-linear relationships between physical features, land planning and socioeconomic characteristics. The consequences of urbanization in fragile areas is clearly overlooked and we conclude that it should be better addresses in modern territorial landslide early warning systems. This study has tested and identified some prominent variables that are being intergated into prototypal warning systems under development in the framework of ongoing research programs.
How to cite: Segoni, S., Nocentini, N., Ajin, R. S., Gatto, A., and Fanti, R.: Investigating the patterns of major geo-hydrological disasters inItaly, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-17217, https://doi.org/10.5194/egusphere-egu26-17217, 2026.
Regional-scale prediction of shallow landslides is essential for operational early-warning systems. Rainfall-duration thresholds and susceptibility mapping are the most commonly used approaches for defining triggering conditions and predisposing factors, respectively. In this study, we investigate a joint approach that combines triggering and predisposing factors.
This study is conducted in the Southeast of France, which was severely affected by the Storm Alex, a millennial return period rainfall event, in 2020. It relies on a retrospective analysis of 1600 shallow landslides recorded in the study area. A random forest approach is applied to quantify the relative importance of landslide geomorphological factors, i.e. geology, parameters derived from Digital Elevation Model (slope angle, aspect, profile curvature…), and several landslide hydrometeorological factors, including the cumulative 1-day, 5-day, 10-day, 30-day and 90-day antecedent rainfall. The significance of the factors is analysed, as well as the performance of the prediction, for normal and extreme rainfall events. This study constitutes a step towards a real-time landslide prediction model, to be then integrated within an early warning system.
How to cite: Armand, L., Bernardie, S., Cerdan, O., and Chambon, G.: Landslide prediction based on jointly analysis of triggering and predisposing factors , EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-20752, https://doi.org/10.5194/egusphere-egu26-20752, 2026.
Monitoring systems are essential tools for landslide risk mitigation. Monitoring reduces vulnerability by providing the data required for slope instability characterization and hazard assessment. Moreover, alert and alarm systems rely on the ability of monitoring networks to timely deliver reliable and precise measurements. Remote monitoring systems are not mature enough to ensure reliable early-warning capabilities; therefore, information redundancy is achieved by deploying a set of sensors directly on the landslide body. Commonly used devices include total stations, GPS receivers, automatic inclinometers, and ground-based radar systems. However, the harsh environmental conditions typical of unstable slopes frequently affect the instruments performance and the data availability. The lack of a stable power supply is one of the main limitations of these systems, often preventing their operation precisely during the most critical situations, such as thunderstorms. In addition, adverse atmospheric conditions, including fog or low cloud cover, can compromise the visibility of topographic benchmarks, reducing data availability when it is most needed. To address these limitations a nonlinear parametric model for forecasting landslide displacements based on rainfall input has been developed. The model is trained and continuously updated using displacement data acquired by the monitoring system. In the event of system failure, the model is able to simulate landslide kinematics by means of “virtual sensors”, forecasting displacements and detecting sudden accelerations of the landslide, thereby preserving the early-warning functionality. The approach was successfully tested using data from the monitoring system installed at the Rotolon landslide, located in the municipality of Recoaro Terme (Vicenza, Italy).
How to cite: Bertorelle, E., Jeddi, M., Falcone, P., Giarrè, L., Ghirotti, M., Ballaera, A., Ceccotto, F., and Mantovani, M.: A Rainfall-Driven Virtual Sensors Model to preserve Landslide Early-Warning Capabilities under Monitoring Systems Failures., EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-10345, https://doi.org/10.5194/egusphere-egu26-10345, 2026.
This study proposes and statistically optimizes a definition of “continuous rainfall” that links rainfall records to soil saturation dynamics. The analysis uses hourly rainfall observations and volumetric water content (VWC) measured by eight sensors installed at a 1-m depth at a natural slope monitoring site in Songnisan National Park, Korea, from 2017 through October 2024. After calibrating the initial condition using the mean dry-season VWC (≈ 0.1) and normalizing the observations to represent relative changes in soil saturation, continuous rainfall was formulated using three parameters: (1) the maximum allowable rain-free period (RPmax), (2) the minimum hourly rainfall threshold included in the accumulation (HRmin), and (3) a moving-window duration (MWdur) that accounts for saturation decay due to drainage. Cumulative continuous rainfall amounts were generated for 27 parameter combinations (RPmax = 12/24/36 h; HRmin = 0/1/2 mm; MWdur = 48/60/72 h), and the correlations between these amounts and normalized VWC were evaluated. The results show pronounced differences in statistical performance across parameter sets: depending on the chosen combination, the same VWC trajectory was either fragmented into multiple rainfall events or consistently captured as a single continuous rainfall event. These findings indicate that an optimized continuous rainfall metric that represents soil hydrodynamics can improve the interpretation of rainfall inputs for shallow landslide prediction. Future work will extend the approach to diverse slope settings and link it to real-time early-warning systems.
How to cite: Park, J.-Y., Song, Y.-S., Kim, M., and Yun, D.: Optimizing a Statistics-Based Continuous Rainfall Definition to Represent Soil Saturation Dynamics for Shallow Landslide Prediction, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-18802, https://doi.org/10.5194/egusphere-egu26-18802, 2026.
Based on the climate change adaptation report, South Korea's annual mean temperature in the late 21st century is likely to rise by 2.3 to 6.3 °C compared to current levels, while annual average precipitation is expected to rise by 4 to 16 % relative to the current mean. Increased precipitation reduces frictional resistance and elevates moisture content in slopes, increasing the strain on top slope sections and greatly increasing the potential of landslides.
Under Article 32-6 of the Republic of Korea's Forest Protection Act Enforcement Decree, landslide warning system are classified into three stages: advisory, preliminary warning, and warning, based on soil moisture limits of 80, 90, and 100%, respectively. Nonetheless, soil moisture is heavily influenced by site-specific soil features like permeability and groundwater level, limiting its capacity to anticipate localized conditions.
In order to secure a golden time for readiness through early prediction of landslides, which are highly sensitive to climate change, and to reduce casualties by evacuating residents quickly and safely, the Korea Forest Service launched a field-response technology development project in 2025. The goal of this project is to build technology for AIoT-based real-time risk monitoring, an audio-visual warning system, and evacuation-route guidance to optimize safe evacuation from landslide. For this research, a test site was chosen in the Triassic granite area of Cheongsong, Gyeongsangbuk-do, South Korea. The project involves setting up in-situ soil moisture measurement devices, developing a next-generation alram technique that combines audio-visual warning devices with soil moisture measurement, and building an early warning system with a LoRa-based IoT wireless sensing network.
Acknowledgement
This study was conducted with the support of the R&D program for Forest Science & Technology (No.RS-2025-02233085)
How to cite: Jeong, C. and Chae, S.: Current Status of Landslide Early Warning and Safe Evacuation Research in South Korea, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-9356, https://doi.org/10.5194/egusphere-egu26-9356, 2026.
Agricultural reservoirs are increasingly exposed to disaster risks due to climate change-induced extreme precipitation and progressive facility aging. In South Korea, more than 20 reservoir collapse incidents have occurred since 2000, including 25 embankment failures during July–August 2020 alone. Of the 17,106 agricultural reservoirs nationwide, 50.8% were constructed before 1945 and 85.4% are over 30 years old, underscoring the urgent need for advanced safety monitoring systems. Analysis of 101 reservoir failures over the past two decades indicates that reservoirs aged 55–60 years exhibit the highest failure rates, with concentrated rainfall identified as the dominant triggering factor.
Current reservoir safety management systems rely primarily on deterministic approaches with simple threshold-based sensor decision rules, which are inadequate for addressing uncertainties in hydrological processes, geotechnical conditions, and structural behavior. Existing early warning concepts often assume automated spillway controls or movable weirs that are impractical for small- and medium-sized agricultural reservoirs, while fragmented implementation of rehabilitation projects, disaster monitoring, and warning systems hinders integrated risk management and effective disaster response.
This study presents the development of a modular integrated reservoir monitoring system designed to overcome these limitations through three core components: (1) heterogeneous modular sensor technology; (2) an on-device integrated operation platform; and (3) a big data-based disaster analysis framework.
The modular sensor system integrates three or more hybrid sensor types to enable simultaneous surface and subsurface monitoring. A modular architecture with interchangeable sensor blocks allows flexible deployment, independent replacement, and future system upgrades. Laboratory performance evaluations confirmed measurement accuracy and stability under diverse environmental conditions.
The on-device integrated operation platform resolves data heterogeneity through standardized data transformation and mapping protocols. A unified gateway supports real-time data streaming and messaging via broker-based communication, enabling bidirectional data processing for monitoring, control, and fault detection. Dual data backup mechanisms ensure system continuity during network disruptions, while edge computing capabilities reduce latency for critical decision-making when on-site access is restricted during extreme weather events.
The big data analytics framework focuses on minimizing measurement errors and processing anomalies inherent in heterogeneous sensor networks. By analyzing disaster-related anomaly patterns and applying multi-sensor data fusion techniques, the system enhances early warning detection capability beyond that of single-sensor approaches.
The proposed integrated system addresses key operational challenges, including multi-manufacturer sensor compatibility, remote accessibility under adverse conditions, cost-effective scalability, and automated decision support that reduces reliance on subjective operator judgment. Field implementation targets aging reservoirs with high-risk profiles identified through historical failure analysis, providing testbeds for system validation and refinement.
This research establishes a technical foundation for risk-based reservoir safety assessment that explicitly incorporates hydrological, geotechnical, and structural uncertainties, representing a transition from deterministic to probabilistic monitoring paradigms consistent with international best practices in dam safety management(Project No. RS-2025-02263904, second year).
How to cite: Lee, S.-Y., Hwang, S., Kim, W., and Park, B.: Modular Integrated Monitoring System for Agricultural Reservoir Embankment Safety Management, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-2870, https://doi.org/10.5194/egusphere-egu26-2870, 2026.
Rock shed is commonly used as rockfall protection structure for mountainous roads. In recent years, in order to reduce the damage caused by falling rocks to rock shed, the design has incorporated cushion materials such as discarded tires and steel plates. However, under clustered rockfall impact conditions, the dynamic response behavior of rock shed remains worthy of further investigation.
In this study, a physical model of a rock shed located beneath a slope with an inclination of 70 degrees is adopted as the research object. Clustered rockfall impact tests are conducted by varying the inclination angle of the slab, and signal analysis techniques are employed to examine the characteristics of the structural responses in both the time domain and frequency domain. Based on the observed signal response features, the dynamic amplification characteristics and frequency-domain response behavior associated with the dominant structural frequency are evaluated.
How to cite: Huang, Y.-L. and Lo, C.-M.: Signal Analysis for Rock Shed Induced by Landslide, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-6169, https://doi.org/10.5194/egusphere-egu26-6169, 2026.
This study is based on the linear elastic behavior of particulate materials in PFC and aims to establish a physical model for micro-energy signals. A series of physical experiments, including compression tests, friction tests, and rebound tests, were conducted using embedded miniature earth pressure cells. Based on experimental results and parameter conversion, a comprehensive analysis and calculation of micro-energy signals were performed, including strain energy, damping energy, frictional energy, and kinetic energy. The calculated micro-energy signal components were then compared with the corresponding results obtained from PFC numerical simulations to calibrate the proposed physical model of micro-energy signals. The comparative analysis demonstrates that the developed micro-energy signal–based approach can effectively estimate the characteristic micro-energy signal features of sliding and non-sliding surfaces, and that the results satisfy the requirements for field-scale applications. Finally, the potential applicability of micro-energy signals for slope monitoring was evaluated, and a corresponding layout methodology for monitoring instrumentation was proposed.
How to cite: Xie, C.-H., Lo, C.-M., and Wu, Y.-C.: Characteristics of Micro-Energy Signal for landslide, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-6167, https://doi.org/10.5194/egusphere-egu26-6167, 2026.
