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
- warning models for issuing warning
- operational applications and performance analyses
- machine learning techniques applied for early warning purposes
Orals: Thu, 1 May | Room 1.15/16
Taiwan's precipitous landscape, susceptible geological composition, and high-intensity rainfall contribute significantly to the prevalence of slope instabilities. The amplification of extreme precipitation events, driven by climatic changes in recent years, has further escalated the risks associated with slope failures. In the context of large-scale landslide monitoring systems, the strategic positioning of monitoring instruments and the calibration of alert thresholds present critical challenges. Effective placement of these instruments is paramount, targeting zones of notable displacement and active slope dynamics to ensure the acquisition of timely and precise data necessary for managing emergent landslide risks. Additionally, the establishment of scientifically grounded warning thresholds can markedly improve the efficiency of disaster prevention mechanisms.
This study integrates a Material Point Method (MPM) numerical model with recorded monitoring data to construct a comprehensive model that accurately reflects the physical behaviors and existing conditions of the Guanghua landslide area. The MPM is adept at addressing large deformation scenarios and provides a detailed depiction of slope displacement behaviors, which are verified through comparisons with field monitoring data. Incorporating engineering reliability analysis into the model allows for the consideration of uncertainties, enhancing discussions on optimal monitoring strategies and the determination of effective warning thresholds. The outcomes of this research are instrumental in refining slope disaster monitoring systems, advancing early warning capabilities, and developing sophisticated risk management strategies.
How to cite: Lin, P.-Y. and Liao, K.-W.: Optimizing Landslide Monitoring and Alert Systems through Material Point Method Modeling: A Case Study of the Guanghua Landslide in Taiwan, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12637, https://doi.org/10.5194/egusphere-egu25-12637, 2025.
Possible landslide zone is chosen from historical events, triggering factors, terrain and many others. Only rainfall induced landslide is considered as the candidate site, and it could be as many as thousand cases for any incoming typhoon or storm in Taiwan. Movements gathered by satellite In-SAR; rock strength estimation of slope is managed by GIS to select the possible sites of the most dangerous slope. Install a pair of 50 Hz GNSS on the slope to obtain the baseline change to reveal the stability and meanwhile the derived wet delay of GNSS signals at various azimuth and directions offer better prediction of rainfall intensity than QBSUM. Once the estimate rainfall will over the threshold of slope unit then the Ku band GB-SAR is deployed to capture the real-time slope movement. The average slip rate of slope surface movement is recorded via different amount of rainfall, whenever continuous 3-time detected slip rate or 3 monitored sub-zone are exceeded than averaged slip value then a landslide warning is issued. It took 5 minutes to calculate the GNSS wet delay for the next half hour rainfall estimation; 3 minutes by GB-SAR to detect and validate the slip rate of slope. This configuration could detect mm level slope movement within hundreds meter distance or cm level in 2 Km away from the slope.
How to cite: Yu, T.-T. and Peng, W.-F.: Landslide Early Warming Configuration with 50Hz GNSS and GB-SAR in Mountain Region of Taiwan, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-2026, https://doi.org/10.5194/egusphere-egu25-2026, 2025.
In January 2024, a significant earthquake struck the Noto Peninsula, causing numerous landslides across the affected region. Subsequently, in September 2024, the area experienced a 100-year extreme rainfall event, which triggered further large-scale landslides. These phenomena are considered to be compound disasters resulting from multiple interrelated factors.
The strong seismic shaking during the major earthquake weakened the ground strength across a wide area, leaving unstable sediment from collapsed slopes accumulated on mid-mountain regions. Even in areas where immediate damage appeared minimal, the seismic event significantly heightened the potential risk of landslides in mountainous regions. Under such conditions, subsequent extreme rainfall poses a high likelihood of triggering large-scale landslides.
To mitigate damage, it is essential to monitor potentially hazardous areas in mountainous regions following an earthquake and assess the degree of downstream risk to communities. Effective risk communication is vital to ensure residents take appropriate evacuation measures during heavy rainfall.
This study proposes a localized Early Warning System (EWS) that considers watershed dynamics and evaluates its significance. Drawing from the 2016 Kumamoto Earthquake, where IoT-based sensors were employed in a localized EWS, the effectiveness and challenges of such systems are discussed.
Based on the above, this paper explores monitoring methodologies aimed at preparing for compound slope disasters caused by post-earthquake heavy rainfall, with a focus on fostering safe and resilient communities.
How to cite: Sakai, N.: Localized Early Warning Systems for Compound Slope Disasters: Insights from Rainfall-induced landslides after major earthquake, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-2669, https://doi.org/10.5194/egusphere-egu25-2669, 2025.
Landslide planforms are commonly used as a chunk in various applications, from volume estimates to hazard/susceptibility modelling. These oversimplifications may decrease the accuracy of predictive models. We developed two complementary models that leverage a landslide’s topology and morphology to improve information in existing landslide databases by distinguishing movement types such as slides, flows, and falls and delineating the kinematic zones, source versus runout.
The first model identifies underlying movements by examining the 3D shapes of landslides (Bhuyan et al., 2024). Tested on inventories across Italy, the United States Pacific Northwest, and Türkiye, the method achieves >80% accuracy in distinguishing various and even complex coupled movement types. Further application to undocumented landslides in the 2008 Wenchuan earthquake-affected region illustrates the method’s potential to inform hazard evaluations.
The second model classifies source and runout zones of landslides with margins of error below 15–20% (Bhuyan et al., 2025). The initial model is developed and validated in geomorphologically diverse regions such as Dominica, Türkiye, Italy, Nepal, and Japan. Subsequent deployments in Chile, Japan (Hokkaido), Colombia, Papua New Guinea, and China reveal source areas commonly occupy less than 30% of a landslide’s total footprint.
These complementary steps hence provide robust and scalable solutions for missing landslide data, which are essential for improving predictive models. They lead to better hazard assessments and a deeper understanding of landslide initiation and propagation. To ease reusability, we will soon integrate these modelling steps into the existing classifier library (Rana et al., 2022).
References
Bhuyan, K., Rana, K., Ferrer, J. V., Cotton, F., Ozturk, U., Catani, F., and Malik, N.: Landslide topology uncovers failure movements, Nat Commun, 15, 2633, https://doi.org/10.1038/s41467-024-46741-7, 2024.
Bhuyan, K., Rana, K., Ozturk, U., Nava, L., Rosi, A., Meena, S. R., Fan, X., Floris, M., Van Westen, C., and Catani, F.: Towards automatic delineation of landslide source and runout, Engineering Geology, 345, 107866, https://doi.org/10.1016/j.enggeo.2024.107866, 2025.
Rana, K., Malik, N., and Ozturk, U.: Landsifier v1.0: a Python library to estimate likely triggers of mapped landslides, Nat. Hazards Earth Syst. Sci., 22, 3751–3764, https://doi.org/10.5194/nhess-22-3751-2022, 2022.
How to cite: Ozturk, U., Bhuyan, K., and Rana, K.: Decoding Landslide Movements and Kinematic Zones from Landslide Planforms, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-3649, https://doi.org/10.5194/egusphere-egu25-3649, 2025.
Landslide Early Warning Systems (LEWSs) are cost-effective solutions designed to prevent loss of life and economic damage caused by landslides by issuing timely warnings to communities. Traditionally, LEWSs rely on rainfall thresholds, which, while simple and accessible, consider only rainfall data and overlook critical hydrogeological soil properties. To improve accuracy, Machine Learning (ML) algorithms have been adopted to generate landslide susceptibility maps by integrating multiple geoenvironmental factors. However, susceptibility maps lack a temporal dimension, limiting their applicability to LEWSs.
Recent advancements in ML have enabled the creation of Landslide Hazard Maps (LHMs) that incorporate spatial and temporal predictions, significantly enhancing their relevance for LEWSs. Despite these improvements, their practical implementation into LEWSs faces two challenges: i) the absence of standardised validation methods to ensure reliability, and ii) a mismatch between pixel-based LHMs and the larger spatial units used for regional warnings. This discrepancy limits the use of LHMs by civil protection authorities, who require simplified and reliable data to effectively coordinate warnings and responses over wide areas.
This study introduces a standardised and automatic validation approach using the Double-Threshold Validation Tool (DTVT). This tool aggregates pixel-based LHMs into broader spatial units called Pixel Aggregation Units (PAUs). Each PAU is classified as unstable based on two thresholds: the Failure Probability Threshold (FPT), indicating the probability above which a pixel is considered unstable, and the Instability Diffusion Threshold (IDT), defining the minimum number of unstable pixels required to classify an entire PAU as unstable.
The DTVT automatically iterates through FPT-IDT combinations, calculating performance metrics to identify the optimal pair that ensures zero missed alarms and minimizes false positives. This process transforms detailed, pixel-based maps into practical hazard assessments suitable for regional LEWSs. Furthermore, the DTVT allows for the calibration of three criticality levels (low, moderate, and high) by adjusting FPT and IDT values.
To demonstrate its effectiveness, the study applies the DTVT in Florence, Italy, using LHMs developed with advanced ML techniques incorporating temporal dimensions. The case study illustrates how the DTVT simplifies complex landslide probability data into actionable warnings, enabling real-time decisions by civil protection agencies.
How to cite: Nocentini, N., Segoni, S., Rosi, A., and Fanti, R.: Double-Threshold Validation Tool (DTVT): a tool for automatically validating and converting pixel-based landslide hazard maps into actionable warning criteria, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6630, https://doi.org/10.5194/egusphere-egu25-6630, 2025.
Rainfall threshold is an effective way for landslide early warning (LEW). Many threshold calculation models based on statistical principles have been proposed, or have been applied in national or regional early warning of geological hazard. However, due to global warming, frequent extreme rainfall and other factors, the triggering conditions of geological hazards have changed. And a fixed single rainfall threshold may no longer be applicable. In addition, the traditional threshold model requires a large number of high-quality landslide records in the region, and it is easy to ignore the heterogeneity of geological environment. Therefore, taking the slope unit as the object, we explore the dynamic updating model of rainfall threshold based on machine learning.
Based on 170 high-risk slope units in Lin 'an District of Zhejiang Province and 65625 warning data from 2021 to 2023, we collected landslide records, rainfall station information and hourly rainfall data simultaneously. According to E-D threshold model and effective antecedent rainfall model, the general law of rainfall-induced landslide in Lin 'an District is derived, which is used as prior knowledge for model training. Then, through the warning data recorded in each slope unit, Decision Tree (DT), Bayesian Ridge (BR), Support Vector Machine (SVM), and Multi-Layer Perceptron (MLP) are selected as base learners, and ensemble strategies such as Bagging, Boosting, and Stacking are considered to dynamic updating of rainfall thresholds.
Taking R2, Mean Square Error (MSE), Root Mean Square Error (RMSE), and Mean Absolute Error (MAE) as evaluation metrics, the Stacking model shows the best prediction performance. In addition, two new landslides occurred in 170 slope units in 2024 were used to verify, and it was found that the updated threshold reduced the redundant workload and gave an accurate early warning of landslides. However, the volume of warning data and its distribution on different rainfall indicators are important factors affecting the threshold update. The accuracy of updating threshold needs to be tested with more experience and practice.
The warning data reflects the response of slope to rainfall under different rainfall conditions, which is of great significance to the threshold update of slope unit. The dynamic updating of rainfall thresholds using machine learning meets the application requirements of current climate change and provides new ideas for disaster prevention and mitigation in the new era.
How to cite: Zhang, X. and Chen, L.: A threshold updating model for rainfall-induced landslide, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-8036, https://doi.org/10.5194/egusphere-egu25-8036, 2025.
Empirical thresholds are crucial tools for predicting the occurrence of rainfall-induced shallow landslides, debris flows and flash floods at territorial scale. These thresholds are typically based only on rainfall data, but this approach overlooks the influence of predisposing conditions and complex hydrogeological processes in the area of interest. The soil response under intense meteorological events can be better investigated by using local monitoring data; indeed, a deeper knowledge of the possible effects in the ground of different rainfall events could provide fundamental support to decision makers towards warning for potential critical events over a relatively wide area (e.g., a catchment or a municipality).
To this aim, IoT monitoring networks have been installed within two small catchments in the municipalities of Amalfi and Sorrento (Campania region, southern Italy). The two monitoring networks―active since autumn 2023 and spring 2024 respectively―can be defined as multifactor networks; in fact, they include sensors installed to monitor the following variables: rainfall, soil water content, soil suction and water level in streams. The sensors have been installed at several locations, covering both the upstream and the downstream sections of the two catchments. This allows the combined use of widespread meteorological data and local real-time measurements coming from monitoring devices installed at specific spots of geomorphological interest. To fully characterize the weather conditions and their potential to cause shallow landslides, debris flows and flash floods, data from satellite observations and reanalysis products are also considered in the analysis. The multifactor time-series analysis is aimed at establishing correlations between the collected variables and at defining a relationship between the local meteorological conditions and the hydrogeological response in the shallower soil layers.
The final aim is the identification of proxies of “critical conditions” over time, that can be used to improve the performance of territorial warning models for rainfall-induced shallow landslides, debris flows and flash floods.
How to cite: Menichini, R., Pecoraro, G., Rianna, G., and Calvello, M.: Multifactor analysis of IoT, satellite and reanalysis time-series for early warning of rainfall-induced shallow landslides, debris flows and flash floods at municipal scale, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-8157, https://doi.org/10.5194/egusphere-egu25-8157, 2025.
The identification of landslide triggering conditions is a fundamental step for the development of effective landslide early warning systems (EWSs), essential for reducing the risks and impacts of these natural disasters. Enhancing the predictive accuracy of these systems requires advanced methodologies such as artificial neural networks (ANNs) that can dynamically assess landslide triggering conditions. Recent advancements have demonstrated significant improvements in landslide prediction when using ANNs fed with observed precipitation and multilayered soil moisture data from ERA5-Land at the onset of rainfall events. However, ERA5-Land data are typically available with a delay of approximately five days, making their direct application to real-time prediction systems challenging. This study investigates the feasibility of utilizing lagged ERA5-Land soil moisture data for real-time landslide prediction and evaluates impacts on predictive performance. Neural networks were developed using soil moisture data lagged by 0 to 15 days prior to rainfall events. The test-application focused on the case study of Sicily, Italy, and revealed that lagged soil moisture data affect prediction accuracy, which still significantly higher than using just precipitation data. For the lags of interest, the reduction of performance is modest. Specifically, with a 5-day lag, the True Skill Statistic index decreased only marginally, from 0.78 to 0.72. These findings highlight the potential for incorporating ERA5-Land multilayered soil moisture data into operational LEWSs, even when using lagged datasets, with potential real-time applications.
How to cite: Palazzolo, N., Peres, D. J., Buonacera, G., Zofei, R. D., and Cancelliere, A.: Exploring the potential of lagged ERA5-Land Soil Moisture Data for Real-Time Landslide Prediction Using Neural Networks , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-10481, https://doi.org/10.5194/egusphere-egu25-10481, 2025.
This work addresses the complex challenge of forecasting the Time of Failure (ToF) for tailings dam and a heap leach facility using Satellite InSAR data. These structures can be susceptible to sudden instabilities due to the contractive and brittle behavior of tailings, and constant remodeling in mining areas can introduce bias in data retrieval from satellites. Moreover, these structures are seldom monitored and the cost-effective coverage of satellite InSAR data can be a useful tool to deploy.
We developed a probabilistic and multi-model approach to forecast ToF and assess the confidence of these predictions, both qualitatively and quantitatively. Our method, the Reiterative Algorithm (RA), provides ToF forecasts using different forecasting method employing iteratively each new data available and yielding a broad range of predictions. The main purpose of this methodology is to fully exploit the predictive potential from slight acceleration signals before failure, monitored with low-temporal resolution instruments such as Satellite InSAR. Some recent papers claim the possibility of predicting brittle failures of such structures.
Emphasis is placed on both the predictability of failure occurrence and the confidence of these forecasts, which is crucial for issuing warnings.
Results are presented for four tailings dam – Dam B1 at the Córrego do Feijão mine in the town of Brumadinho (Brazil), Jagersfontein tailings dam (South Africa), NTSF Embankment at Cadia (Australia), and Zelazny Most (Poland) - and a heap leach facility at Copler Mine (Turkey). All except Zelazny Most experienced failure.
The findings advance the understandings of ToF predictability and highlight the need for further research to improve the accuracy of such forecasts.
How to cite: Szakolczai, I., Intrieri, E., Carlà, T., Piciullo, L., Frauenfelder, R., and Vöge, M.: Considerations on Forecasting the time of failure of Tailings Dams and a heap leach facility through Satellite InSAR Data, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-16264, https://doi.org/10.5194/egusphere-egu25-16264, 2025.
Territorial Landslide Early Warning Systems (Te-LEWSs) globally rely on meteorological and hydro-meteorological thresholds for effective warning dissemination. In India, the widely used Intensity-Duration (ID) curve serves as a primary meteorological threshold for landslide forecasts, while hydro-meteorological thresholds, such as the Soil-Water Index (SWI), remain under evaluation. Challenges persist in threshold attribution due to uncertainties in meteorological and landslide datasets. To address this gap, this study employs data-science-based approaches to differentiate triggering and non-triggering rainfall events across multiple Indian regions: Kerala, Maharashtra, Uttarakhand, and Himachal Pradesh. The analysis identifies ID thresholds for landslide forecasting with over 90% confidence and an accuracy exceeding 85%. Additionally, SWI-based hydro-meteorological thresholds are derived, though further refinement is needed for enhanced accuracy. Using hourly meteorological data from multiple sources, the study demonstrates the robustness of data-driven methodologies in resolving uncertainties and improving the reliability of Te-LEWS thresholds in India.
How to cite: Sagar, N. and Siva Subramanian, S.: Data Science-based Separation of Triggering and Non-Triggering Rainfall of Landslides for Threshold Attribution, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18175, https://doi.org/10.5194/egusphere-egu25-18175, 2025.
Generating a landslide susceptibility map taking into account casual factors like slope geometry, soil/rock types, river location, groundwater conditions, rainfall data and human activities, also including the infrastructures at risk, in order to accurately evaluate the proneness to landslides at fine spatial resolution is a highly-demanding task. Using high-quality data from 3298 different generations of landslides and non-landslides, a framework using Google Earth Engine has been efficiently developed for evaluating landslide susceptibility in the Daunia area of the Italian Southern Apennines, a sector extensively affected by gravitational phenomena of different typologies in Apulia region (Southern Italy). Casual factors including internal (predisposing) and external (preparatory and triggering) factors have been considered to be used within Spatial Data Modellers (SDM). Further, a cloud computing platform via algorithmic models, easily to update, has been created to derive a susceptibility map at the regional scale, especially useful in areas with highly complicated topography. To this purpose, different methods have been compared, including Fuzzy logic methods (Gamma, Product, Sum, And, and Or), as well as machine learning algorithms, such as RF (random forest) and GTB (GradientTreeBoost). The results have been represented via classification and regression modes. A performance analysis has been also carried out and the best modeling performance is observed to belong to the RF algorithms, as provided by Root Mean Squared Error (RMSE): 0.05, R-squared: 0.9825 (regression mode), and Accuracy: 0.8409 (84.09%), 95% CI : (0.8102, 0.8683), P-Value : < 2.2e-16, Kappa : 0.7626 (classification mode), Kappa statistic measures the agreement between the observed accuracy and the accuracy that would be expected by chance. A Kappa of 0.7626 indicates substantial agreement. Considering the Pearson correlation matrix heatmap, visually representing the Pearson correlation coefficients between pairs of variables, it is observed that rainfall, lithology and slope geometry can have the strongest impact on the occurrence of landslides in Daunia. The framework developed in this study is supposed to be applied not only in the region under study, but also in other landslide-prone areas around the world.
How to cite: Sabaghi, M., Lollino, P., and Parise, M.: Developing a framework through Analytical Models on Google Earth Engine for Landslide Susceptibility Assessment in the Daunia area, southern Italy, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6005, https://doi.org/10.5194/egusphere-egu25-6005, 2025.
Situated in the South-Eastern Alps, Slovenia belongs to the (high-)mountainous countries of Europe, with Mount Triglav marking the highest elevation at 2,864 masl. The country is crossed by several mountain ranges in the north and the Dinaric Alps stretching from the center towards the south-east; further south-west, a karst plateau elevates the topography of the country. Throughout all the lithologic and topographic diversity, mass movements are common and frequently associated with a variety of spatial factors favoring their formation.
Slovenia recognizes the need for comprehensive mapping of mass movements and analyzing the associated contribution factors to provide safe frameworks for land use planning, and the Slovenian Geological Survey (GeoZS) has been working regularly in accordance with these guidelines for several years. Initial works by Komac (2000–2009) set prominent contributing factors (e.g., lithology, elevation, slope aspect, slope inclination, terrain roughness, terrain curvature, distance to streams, distance to faults, and land cover) in relation to landslide formation by uni- and multivariate statistics. Here, the term “landslide” covers different types of mass movements without further differentiation.
Based on the relations of the univariate statistical analysis, further refined individual algorithms were developed for landslides, rock falls and debris flows, selecting and gradually adjusting contributing factors for each of the three phenomena.
Applying fuzzy logic and linear membership functions, weights are attributed to relevant factors for landslides, rock falls, and debris flows, respectively. This concept entails that the GeoZS’ approach to the national risk assessment of mass movements is based on susceptibility (i.e., not on probabilities) and, hence, the notion of return periods for events of specific characteristics does not apply.
We present the Slovenian Landslide Susceptibility Map at the scale of 1:250,000 as well as some of the 95 already processed municipality Landslide Susceptibility Maps at the scale of 1:25,000 reflecting the geomorphologic variability of the country resulting in different mass movement patterns with respect to magnitude, frequency, type, contributing factors, and associated risk.
The map at the scale of 1:250,000 is one of the components of the Slovenian Landslide Forecasting and Warning System MASPREM.
How to cite: Domej, G., Jež, J., Kumelj, Š., Turk, D., Novak, A., and Kure, K.: Landslide Susceptibility Maps in Slovenia: landslides, rock falls & debris flows, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-8639, https://doi.org/10.5194/egusphere-egu25-8639, 2025.
Shallow landslides, characterized by sudden, superficial movements and the absence of precursors, pose an increasing threat to the population and local authorities, particularly in the context of climate change and rapid urbanization. A good understanding of the triggering mechanisms of these phenomena is essential to predict their occurrence and to design reliable early warning systems. For that purpose, the retrospective analysis of event inventories can provide critical information on predisposing and triggering factors, making it possible to establish, e.g., susceptibility maps and rainfall triggering thresholds.
This study presents the results of a statistical analysis of an inventory of 4786 shallow landslides that occurred in the Alpes Maritimes department (southeastern France). This area, characterized by Mediterranean and mountainous climates, is prone to intense and localized rainfall events. It was severely impacted by the Alex storm in October 2020, an exceptional event that triggered numerous landslides among other consequences. A total of 1656 landslides in our inventory are related to this storm. Our objective is to assess how landslides triggered by this exceptional event can integrated into hazard analyses.
For that purpose, we analyse the relations between two groups of landslides (i.e Alex or non-Alex landslides) and both meteorological variables and predisposing factors. Cumulative rainfall/duration thresholds are computed using the CTRL-T algorithm (Melillo et al., 2018). Results show that rainfall thresholds for landslides triggered by Storm Alex are significantly higher than those for other landslides. In addition, predisposing factors, in particular the geology derived from the harmonized geological map of France, show different distribution The formations corresponding to surface deposits, such as debris material, were more heavily mobilized during storm Alex. The implications of these outcomes for improving the reliability of susceptibility maps and early warning systems are discussed.
How to cite: Armand, L., Chambon, G., Bernardie, S., and Cerdan, O.: Statistical analysis of shallow landslides in the Alpes-Maritimes (France) : predisposing factors, meteorological variables, and extreme events, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-10800, https://doi.org/10.5194/egusphere-egu25-10800, 2025.
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 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 successfully simulating the September 2015 event, a period of intense rainfall, demonstrating its high reliability in distinguishing stable from unstable conditions.
Results confirm the potential of dynamic RF models to overcome the limitations of static ML approaches, providing actionable and interpretable outputs for operational LEWS. Future research will focus on extending this methodology across Italy and validating it against independent datasets to ensure robust predictions in different geoclimatic contexts.
Work supported by PRIN-ITALERT project, funded by European Union – Next Generation EU M4.C2.1.1 - CUP: B53D23006720006
How to cite: Rosi, A., Nocentini, N., Segoni, S., Gariano, S. L., Brunetti, M. T., Peruccacci, S., Melillo, M., Palazzolo, N., Peres, D. J., and Cancelliere, A.: Landslide early warning system based on a dynamic machine learning approach: a case study in Italy, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-16489, https://doi.org/10.5194/egusphere-egu25-16489, 2025.
Landslides pose a critical threat to Colombia’s Andean region, where steep topography and intense rainfall events frequently disrupt road infrastructure. Although data-driven models are widely used for landslide susceptibility, they often focus on static conditioning factors without fully capturing the temporal dimension essential for early warning. Integrating space and time into a single model remains challenging due to data heterogeneity, incomplete inventories, and the complexity of rainfall triggers. In this study, we address these gaps by developing a space-time data-driven landslide model tailored for an Early Warning System (EWS) that targets roadblocks.
We address this challenge by combining multiple landslide inventories, satellite rainfall estimates (CHIRPS), and 15-day ensemble rainfall forecasts (CHIRPS-GEFS), the project aims to provide forecasted landslide probabilities. The workflow is structured into three phases. First, a landslide inventory is compiled by harmonizing multiple datasets—each varying in quality, completeness, and spatial-temporal granularity. We address inconsistencies across institutional, academic, and regional inventories to derive a consolidated database of over 17,000 rainfall-induced landslides. Second, with this inventory, we extract data on static and dynamic predictors such as slope steepness, geology, land cover, and rainfall. Using generalized additive models (GAMs), we estimate daily landslide probabilities at a spatial resolution suitable for critical road segments. We compare short-term (1–3 days) to medium-term (up to 15 days) forecasting accuracy to assess model performance. Third, results are translated into spatial dynamic probability thresholds. These thresholds are designed to alert authorities about imminent or escalating risks of landslide-induced roadblocks.
Preliminary tests indicate that this type of space-time model is particularly suitable for integrating forecast-based rainfall data and testing multi-day lead times. The final outcome is a prototype EWS component where probabilistic landslide alerts are updated daily, contributing to risk-informed decision-making for road infrastructure management in Colombia. This contribution discusses the methods, preliminary results, and future steps.
How to cite: Urueña Ramirez, D. A., Moreno, M., Lombardo, L., Gómez, D., Vega, J., and van Westen, C.: A Dynamic Landslide Model for Early Warnings in Colombia's Roads, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-20285, https://doi.org/10.5194/egusphere-egu25-20285, 2025.
Rainfall-induced landslides and debris flow as one of the most common geohazards, causing significant societal and economic impacts. To enhance the accuracy of early warning systems and reduce the risks associated with these events, it is essential to establish precise and regionally adaptive rainfall thresholds. This study addresses the challenges in defining rainfall thresholds by integrating rainfall data with landslide datasets that include large-scale landslides caused by Typhoon Morakot, which are defined as those with an area larger than 10 hectares, a volume exceeding 100,000 cubic meters, or a depth greater than 10 meters, small-to-moderate sized (defined as those with a sliding area of less than 10 hectares, a soil volume of less than 100,000 cubic meters, and a sliding depth of less than 10 meters) landslides triggered by Typhoons Sinlaku and Kongrey, and recent debris flow events (2019–2023) in the Putunpunas River area of Kaohsiung, Taiwan. By incorporating diverse landslide magnitudes and climatic conditions, this study seeks to improve the reliability and adaptability of rainfall thresholds.
For rainfall data preprocessing, the first step was to determine the climatic season (cold season: October to April; warm season: May to September) for each rainfall record. Based on the season, rainfall events were initially separated using intervals of 3 hours (warm season) or 6 hours (cold season). Hourly rainfall measurements below 0.2 mm were excluded (set to 0). Subsequently, rainfall events were reconstructed using adjusted interval criteria of 6 hours for the warm season and 12 hours for the cool season. Valid rainfall events were required to have cumulative rainfall greater than 1 mm. Only events meeting this condition were further processed. Finally, rainfall events were redefined based on adjusted intervals of 5 hours for the warm season and 10 hours for the cool season to better capture event continuity.
This study employed the bootstrap technique to estimate rainfall thresholds under various exceedance probabilities (0.005% to 50%). The threshold curve is expressed as E=(α±∆α)∙D^(γ±∆γ) , where α represents the baseline proportional constant between cumulative rainfall (E,unit:mm) and event duration(D,unit:hr), reflecting the vertical shift of the threshold under different probability conditions. ∆α represents the standard deviation of α , quantifying its uncertainty. Additionally,γ=-β+1, where β is the average slope of the best-fit line (T50), and ∆γ is the standard deviation of γ . These parameters effectively describe the uncertainty range of the thresholds across different probabilities.
The results show that α and γ under different exceedance probabilities provide a reliable description of rainfall thresholds, which can be adjusted regionally based on local topographical and different scales and corresponding event types conditions, Typhoon Gaemi in 2023 can serve as a validation case. Ultimately, this study provides a robust scientific foundation for rainfall threshold estimation, supporting the implementation of early warning systems for rainfall-induced landslides and contributing to regional risk management and disaster mitigation strategies.
Key words: Landslide、Rainfall、Debris flow、Empirical rainfall thresholds
How to cite: Chu, C. H. and Chao, W. A.: Reconstruction of Rainfall Events and Empirical Rainfall Threshold Modeling for Landslide and Debris Flow Forecasting, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-10590, https://doi.org/10.5194/egusphere-egu25-10590, 2025.
We propose a national scale landslide nowcasting system for Italy (300,000 km2) by combining rainfall thresholds with a set of spatially explicit risk indicators. The combination of these two very different elements is obtained through a dynamic matrix, which was purposely calibrated to provide an output in the form of five possible levels of expected risk (from R0 to R4). These levels are connected to the growing intensity of expected impacts and a pre-defined confidence in issuing warnings without omitting alarms.
A specific set of rainfall thresholds is defined for each of the 150 alert zones (AZ) in which Italy is divided. The risk indicator is defined at a municipality level. The calibration of the dynamic risk matrix is carried out independently for each AZ, following predefined operational criteria.
The verification of the matrix outputs was satisfactory as no AZs experienced landslides at the R0 level; only two of them had more than 10% of landslides at the R1 level, and most of the AZs had more than 90% of the landslides in the R2 to R4 risk classes. A comparison with a nation-wide dataset of very severe hydrogeological disasters further proved the consistency of the model outputs with the scenarios that occurred during past events, as most part of the impacts occurred in places and times when the matrix outputs were at the highest levels.
The proposed methodology represents a reliable improvement for state-of-the-art territorial warning systems, as it brings two main advances: (1) the spatial resolution is greatly improved, as the basic spatial unit for warning is downscaled from AZs to municipalities (whose average extension, in Italy, is about 1770 and 38 km2, respectively); (2) the outputs can better address the needs of landslide emergency management, as the warning are specifically addressed to small areas based on the expected impacts (since risk indicators are used in the dynamic matrices), rather than on the probability of landslide occurrence.
How to cite: Segoni, S., Nocentini, N., Barbadori, F., Medici, C., Gatto, A., Rosi, A., and Casagli, N.: A nation-wide nowcasting system for Italy combining rainfall thresholds and risk indicators, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-7441, https://doi.org/10.5194/egusphere-egu25-7441, 2025.
The movement of rain-induced landslides is a complicated process with largely unknown mechanisms that can have devastating consequences globally every year. Aiming to unravel the hydrological-mechanical interactions that govern the dynamics of such landslides, we studied a rain-induced clayey landslide in the Greater Bay Area of China, where the sliding surface was located beneath the groundwater level, representative of many landslides in clay rich lowland slopes near human activities. Aligning with previous studies, there is a strong correlation between observed groundwater levels and landslide displacement. However, our field measurements accented an intriguing pattern: the rising rate of groundwater levels, rather than their absolute values, exhibited a remarkably synchronized relationship with landslide motion. Slope stability modeling suggests that the observed landslide behavior could be predicted by including rainfall infiltration based on initial groundwater levels, whereas modeling considering solely transient groundwater levels may fail to capture landslide movement. The changes of groundwater levels in all landslides involve rainfall infiltration processes, and the speed of groundwater level rise may actually reflect the saturation state of slopes. Our findings suggest that the saturation state of slopes likely modulates landslides movement and should be considered to improve predictions of clayey landslides initiation and mobility.
How to cite: Huang, T. and Shen, P.: Key Features of Rain-induced Clayey Landslide: Groundwater Level Rising Rate Highly Synchronized with Landslide Speed, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-16110, https://doi.org/10.5194/egusphere-egu25-16110, 2025.
Landslide early warning depends on the availability of reliable real-time monitoring data. In this context, high reliability means optimal data continuity (minimal data loss), sufficient redundancy of sensors observing a variety of parameters, and a high accuracy of monitoring techniques. Nowadays, most manufacturers can provide nearly perfect reliability for modern IoT monitoring devices under laboratory or calibration conditions. However, under challenging high alpine conditions, the actual effective reliability of a monitoring system remains unknown as long as the system is not fully operational or is even kept confidential from the public.
Here, we analyse the effective reliability of the real-time monitoring system at the Hochvogel summit (2,592 m a.s.l.) where conditions combine limited access throughout the year especially in winter, inaccessible steep areas, no permanent power supply, high snow loads, high probability of lightning strikes, and highly jointed and weathered rock mass. The monitoring system is operational since October 2019 transmitting data every 10 min via wireless LoRa technology from 10-12 geotechnical sensors (crack meters, laser distance meters, inclinometers, rain gauge). Many sensors operate at the edge of radio range (2,800 m horizontal and 1,500 m vertical distance to the gateway, mostly without direct line of sight). We analyse the probability of data loss in three categories: (i) daily average, i.e. days on which at least one measured value was transmitted are valid; (ii) hourly average; and (iii) all 10-minute data. Generally, the probability of missing data increases with higher temporal resolution, as suboptimal conditions and transmission problems are often short-lived. Due to its magnitude and failure process, we expect the Hochvogel instability to accelerate several hours to few days before failure. Therefore, hourly and daily datasets are most important. The daily transmission reliability for most sensors is 97.3–100 %. From the laser distance gauges, less data can be used for early warning, as they are covered by snow for several months per year. On an hourly basis, the transmission reliability is 96.7–99.4 % for crack meters, and 56-65 % for the laser distance sensors.
This analysis of more than 5 years of data allows us to quantify the effective reliability of the Hochvogel monitoring system and to identify the most important reasons for data loss and particularly critical periods in which several sensors fail simultaneously. This will help decision-makers and responsible parties to plan or adapt their systems and give guidance on how much financial means they must spend to reach the desired level of resilience, reliability, or redundancy.
How to cite: Leinauer, J. and Krautblatter, M.: The actual effective reliability of a high-alpine real-time monitoring system, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11121, https://doi.org/10.5194/egusphere-egu25-11121, 2025.
Posters virtual: Mon, 28 Apr, 14:00–15:45 | vPoster spot 3
EGU25-851 | ECS | Posters virtual | VPS12
Using volumetric water content measurements with the implementation of machine learning for monitoring shallow landslides induced by rainfallMon, 28 Apr, 14:00–15:45 (CEST) | vP3.25
Improved and affordable prediction techniques are required because the growing frequency of shallow landslides caused by shifting weather patterns poses severe dangers to ecosystems, infrastructure, and communities. Although comprehensive monitoring systems are available, their high costs and complexity often make them impractical in resource-constrained regions. This study aims to evaluate the predictive potential of volumetric water content (VWC) measurements for shallow landslides and leverage machine learning techniques to develop cost-effective prediction models. The study employed one-dimensional modified column tests to simulate various scenarios (e.g., soil densities, drainage conditions) using a one-meter-high acrylic column to measure VWC, pore water, and air pressure. Key findings include the identification of VWC-related parameters (e.g., steady-state VWC and its gradient) as effective predictors of slope failure. When integrated with ML models, these parameters demonstrate the potential for enhancing prediction accuracy. This study provides a pathway to developing cost-effective early warning systems for slope instability, offering a practical solution for improving safety, using volumetric water content measurements to protect infrastructure, and enhancing resilience in landslide-prone regions, mainly where comprehensive monitoring systems are infeasible.
How to cite: Avzalshoev, Z., Ahmad, W., and Ahmad, T.: Using volumetric water content measurements with the implementation of machine learning for monitoring shallow landslides induced by rainfall, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-851, https://doi.org/10.5194/egusphere-egu25-851, 2025.
