Landslides are ubiquitous geomorphological phenomena with potentially catastrophic consequences. In several countries, landslide mortality can be higher than that of any other natural hazard. Predicting landslides is a difficult task that is of both scientific interest and societal relevance that may help save lives and protect individual properties and collective resources. The session focuses on innovative methods and techniques to predict landslide occurrence, including the location, time, size, destructiveness of individual and multiple slope failures. All landslide types are considered, from fast rockfalls to rapid debris flows, from slow slides to very rapid rock avalanches. All geographical scales are considered, from the local to the global scale. Of interest are contributions investigating theoretical aspects of natural hazard prediction, with emphasis on landslide forecasting, including conceptual, mathematical, physical, statistical, numerical and computational problems, and applied contributions demonstrating, with examples, the possibility or the lack of a possibility to predict individual or multiple landslides, or specific landslide characteristics. Of particular interest are contributions aimed at: the evaluation of the quality of landslide forecasts; the comparison of the performance of different forecasting models; the use of landslide forecasts in operational systems; and investigations of the potential for the exploitation of new or emerging technologies e.g., monitoring, computational, Earth observation technologies, in order to improve our ability to predict landslides. We anticipate that the most relevant contributions will be collected in the special issue of an international journal.
Tue, 24 May, 08:30–10:00
Chairpersons: Filippo Catani, Kushanav Bhuyan, Binod Tiwari
Landslides are among the top five natural disasters in terms of casualties and property damage; therefore, landslide susceptibility mapping is vital in landslide-prone areas, particularly hilly terrain. Globally, landslides alone take away the life of 17% of the total death caused by natural hazards. The death count in numbers is approximately 1000 per year, with property damage of about US$ 4 billion. This makes the study of landslides extremely significant. Important factors contributing to the reported global increase in landslides are the rapid growth of the world's population, urbanisation in the developing world, and global climate change.
Landslides are natural and anthropogenic hazards that have impacted Indian subcontinent, especially the Himalayas and other mountainous areas. Comparative evaluations of the landslide susceptibility mapping models are necessary for landslide susceptibility mapping to find the best fit model for the specific area. The present study has been conducted in the West Sikkim district of India, in the Indian Himalayan Region, using a data-driven statistical model of information value method (IVM) and frequency ratio method (FRM), as well as a knowledge-driven heuristic approach of analytic hierarchy process (AHP). The combination of the statistical and the knowledge-based approach is applied because the former gives the unbiased result based on the pixel value of the satellite data used, whereas the knowledge-based method gives the value based on the knowledge and experience of the expert, so a very good comparison can be made. In this study, eleven landslide conditioning factors were analysed in the remote sensing (RS) and geographic information system (GIS) environment, which are slope aspect, slope gradient, slope curvature, drainage density, elevation, lithology, land use and land cover (LULC), normalised difference vegetation index (NDVI), geomorphology, lineament density, and soil type. The Resourcesat 2A satellite images were used from Indian remote sensing agency having LISS 4 sensors of 5.8 m resolution data.
A total of 685 landslides were identified in a satellite image, and the polygons of the same in the shapefile format mapped in the GIS environment. Landslides mapped from the satellite data has also been validated in the google earth images and selected sites are also validated by ground truthing. 70% of the total landslide polygons were taken as the training data the remaining 30% landslide polygons were taken for the validation purpose The studies were validated using a receiver operating characteristic curve that fit the model with acceptable values of more than 60% for all three models, with the highest value of 74% being obtained for the information value method. The density distribution method has validated the result, confirmed by the landslide density increased from the low susceptibility zone to the high susceptibility zone. These types of studies are helpful for the decision-makers and the planners for the developmental projects that are ongoing in the state and future projects.
How to cite: Biswakarma, P. and Joshi, V.: A comparative study using bivariate statistical method and knowledge-driven heuristic approach for the comparison of landslide susceptibility mapping in West Sikkim district of Sikkim Himalaya, India, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-488, https://doi.org/10.5194/egusphere-egu22-488, 2022.
Landslides are associated with severe losses on the Loess Plateau of China. Providing hazard mitigation decision support for stakeholders and ensuring the safety of personnel play essential roles in risk management for landslides. Although early warning systems and escape guidelines have mitigated the risk to some extent, most methods are qualitative or semi-quantitative in the sitescale. Therefore, we propose a quantitative simulated-based spatial distribution model and scenario‐based human vulnerability probabilistic model for site-specific loess landslide risk assessment. For spatial distribution, coupled with multi-temporal remote sensing images and high-precision UAV cloud point data, a total of 98 loess landslides have occurred since 2004 on the Heifangtai terrace (North-West China) were collected to establish a landslide volume-date and retreating distance database. Eleven loess landslides are selected to construct a numerical model for parameter analysis. The centroid distance and overlapping area can quantitatively evaluate the accuracy of the simulation results. Different volumes and receding distance rates of landslides are fitted to determine the relationship between cracks and potential volume. Different volumes and parameters are combined to simulate the spatial distribution of potential loess landslides. Following the obtained hazard zone, a scenario-based model for evaluating the escape behavior and human vulnerability was proposed using a Python platform. Based on sampling surveys and field investigations, a database that includes detailed information for the hazard zone’s demographic structure and behavioral characteristics were established. The probability of scenario input parameters, such as the escape route and speed, were calculated and quantified by classic probability theory. In the selected slope slide case, farmland near the toe of the slope primarily includes exposed hazards with probabilities greater than 0.7. The registered population over 65 years old accounted for 13.46% of the total, and most residents had no more than a primary school education background. Older adults were inclined to escape a moving landslide by running parallel to the sliding direction, although the public considers this direction to be the most dangerous. The model simulation revealed that cumulative mortality could be significantly reduced by promoting disaster prevention awareness and improving the advance warning time. The developed quantitative hazard and human vulnerability framework provide a useful reference for local disaster reduction and disaster prevention rehearsal guidelines.
How to cite: Zhou, Q., Xu, Q., Peng, D., Zeng, P., Fan, X., Ouyang, C., Zhao, K., Yuan, S., Zhu, X., and Li, H.: Quantitative spatial distribution and human vulnerability assessment for site-specific loess landslide, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-2935, https://doi.org/10.5194/egusphere-egu22-2935, 2022.
Shallow landslides are a frequent type of mass movement in mountains regions. The recognition and mapping of shallow landslides are very important to better understand the characteristics of the hazard (e.g., triggering factors, conditional factors) and the magnitude of the event, as well as to facilitate susceptibility, vulnerability, and risk analysis. In Brazil, they are one of the most frequent and destructive natural hazards; each year numerous shallow landslides are triggered by rainfall, particularly in the south and southeastern regions of the country, resulting in social and economic impact. The construction of landslide inventories in Brazil is still incipient since not all mass movement events that occurred are documented and no mapping guidelines exist. Thus, this research aims to investigate how the inclusion/exclusion of the deposition area in a shallow landslide inventory mapping influences the results of a susceptibility assessment. The study area is the Gurutuba watershed, located in the municipality of Itaóca, São Paulo state, southeastern Brazil. Two shallow landslide inventories of the 2014 high magnitude mass movement event were created based on Google Earth Pro images dated 2014/10/08. The criteria applied for visual mapping were the absence of vegetation, shape, size, drainage network distance, planar rupture surface, altimetric variation, and slope position. The inventories were constructed based on the same visual guidelines, the difference between them is regarding the deposition area. Inventory 1 (INV1) includes rupture, transport, and deposition area, while inventory 2 (INV2) only includes rupture and transport area but excludes the deposition area. A bivariate statistical approach, i.e., the informative value method, was applied to create a susceptibility map and compare the performance of INV1 and INV2. Besides the inventories, four morphological thematic variables (aspect, slope, elevation, and curvature) derived from a digital elevation model (DEM) from the SRTM mission, re-sampled to 12.5 m, were used for this analysis. The thematic variables slope, aspect, and elevation did not generate a substantial difference with the inclusion/exclusion of the deposition area and showed similar statistical results for both inventories. The morphological classes with high susceptibility were slope between 40°and 50°, E and SE orientation, and elevation between 400 and 500 m. Curvature presented different results for each inventory, while in INV1 convex areas were the most susceptible, with INV2 both convex and concave areas were considered susceptible. The validation indicated slightly better performance of INV2 for the susceptibility mapping based on the success rate (AUC 0.775) and prediction rate (AUC 0.758) than INV1, which resulted in a lower success rate (AUC 0.758) and prediction rate (AUC 0.740). These results indicate that considering the deposition area for shallow landslide recognition and mapping affects the assessment of susceptibility mapping in a tropical environment. The criteria applied for shallow landslide mapping are not always mentioned in Brazilian studies despite the landslide inventory being the most important input variable for susceptibility assessment. Further analysis should be carried out in other regions of the country, as well as with more accurate resolution data if available.
How to cite: Dias, H. C., Hölbling, D., and Grohmann, C. H.: Investigating the effect of the landslide deposition area for susceptibility assessment in Brazil, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-1542, https://doi.org/10.5194/egusphere-egu22-1542, 2022.
Landslide susceptibility maps (LSMs) depict the probability of occurrence of a given type of landslide in a given area, based on the spatial distribution of a set of selected predisposing factors. Therefore, the susceptibility assessment is very sensitive to the parameters chosen and the identification of new parameters to be used as input data is a promising field of research in susceptibility studies as it may contribute to enhance the results.
In this work the machine learning algorithm called Random Forest (RF) has been applied, employing, in addition to the most common predisposing factors, a set of newly proposed parameters, with the aim of verifying their applicability in the landslide susceptibility analysis. The study area, 3100 km2 wide, contains the provinces of Lucca, Prato and Pistoia, in northern Tuscany (Italy).
The first innovative parameter introduced is the soil sealing map, derived from the national map updated yearly by ISPRA (Italian Institute for Environmental Protection and Research). Soil sealing represents the degree of anthropization of the soil, which can radically alter the geotechnical equilibrium or the hydrological system of hillslopes. This may be directly or indirectly linked to an increased landslides hazard.
In addition, multi-parametric geological information has been included. Usually, LSMs exploit only the lithological information provided by geological maps, neglecting potentially relevant geological information (e.g. degree of weathering or tectonic stress history). We created a set of geologically-based explanatory variables by reclassifying a high resolution geological map (where 194 lithostratigraphic units were mapped at the 1:10,000 scale) using five different approaches: lithological, genetic, paleo-environmental, structural and chronological.
The model was run twice, with and without these innovative parameters, and the two resulting LSMs were compared with three approaches: (1) the area under receiver-operator characteristic curve (AUC) highlighted that the advanced parameterization increases the effectiveness of the model; (2) the Out-of-Bag Error (OOBE). OOBE was used to assess the relative importance of each predisposing factors, and the new parameters showed high predictive power; (3) the resulting maps were compared, and the main differences could be explained by local complex geological settings, which are better accounted for using the multi-criteria geological parameterization.
How to cite: Nocentini, N., Luti, T., Rosi, A., and Segoni, S.: Landslide susceptibility assessment including a set of novel explanatory variables: soil sealing, and multi-criteria geological parameterization., EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-4051, https://doi.org/10.5194/egusphere-egu22-4051, 2022.
Co-seismic landslides are triggered by strong ground shaking in mountainous areas, resulting in threats to human activity and infrastructure. Co-seismic landslide susceptibility assessment plays a vital role in disaster prevention and mitigation. However, existing physical models for susceptibility assessment do not involve the dynamic nature of seismicity and the progressive processes of landslide initiation. The challenge of linking the development of internal cracks caused by dynamic seismic loading with the process of localized failure from abrupt mass movement will be addressed by a new physically based model that bridges the limit-equilibrium stability analysis with the fibre bundle model (FBM), which the FBM is a mathematical framework to simulate the highly nonlinear behaviour of the progressive damage and breakdown of disordered media statistically. Each hillslope in a catchment is depicted as an assembly of virtual bundles of fibres that represented the soil columns. The vibrating seismic load exerted on the mechanical connections causes the fibres to break progressively until restraining forces are exceeded. Since cracks occur at the interface of different soil layers, load redistribution occurs from the broken column to its neighbours through intact mechanical linkages, resulting in a new mechanical state. When the ground columns lose their balance, a load-bearing column can liquefy and trigger a landslide which could spread downstream. primary purpose of this study is to develop a semi-physical model for simulating the earthquake-induced landslides by incorporating earthquake time histories into a spatially distributed slope stability method on the basis of the FBM to represent the localized failure occurring prior to landslide release or after the ground shaking. The study has four specific objectives: (1) Development of a model framework assembled with the limit-equilibrium analysis and FBM for seismic effect simulation on hillslopes; (2) Development of an efficient regional method for physically based simulation of co-seismic slope instability; (3) Derive a method for predicting the increase in susceptibility to rainfall-induced landslides after seismic shaking, taking into account the soil healing process; (4) Determine the effect of vertical variation in soil strength parameters and groundwater table depth on the fibre bundle model by implementing a multi-layer approach. The proposed model framework linking limit-equilibrium stability analysis and fibre bundle model should sufficiently consider the dynamic characteristics of seismicity and progressive slope failure processes of landslide triggering.
How to cite: Chen, Y., van den Bout, B., van Westen, C., and Lombardo, L.: Co-seismic Landslide Susceptibility Modelling Based on the Fibre Bundle Model, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-8382, https://doi.org/10.5194/egusphere-egu22-8382, 2022.
Landslides in urban areas are conceived as phenomena capable of tearing the physical structure as well as the networks of socio-economic, cultural, material and immaterial relations that make up the life of cities. Landslide hazard analysis is usually mandatory for proper land use planning and management. Nevertheless, in some cases (e.g., municipality of Rome in Italy) regulatory plans lack detailed thematic mapping of geohazard-related data. In Italy, the safety of urban areas has become a very important issue in the last decade, therefore projects of national interest have been funded for the mitigation of geological risks.
Shallow landslides are common mass movements in urban areas. They can be triggered by earthquakes, heavy rains or induced by proximity to specific urban assets, like road cuts or retaining walls. Reliable quantification of landslide hazardous areas is often associated with the existence of static specific predisposing factors, such as local terrain variables, land use, lithology, proximity to roads and streams as well as dynamic factors related to trigger (e.g., antecedent rainfalls). Predictive multivariate statistical analysis, among which Machine Learning (ML) models, takes as input several predisposing and conditioning factors that may reveal patterns with the spatial and temporal distribution of different types of landslides. Therefore, ancillary landslide databases are the key-data to investigate the distribution, types, pattern, recurrence, and statistics of slope failures and consequently to determine the overall landslide hazard. However, the amount and quality of available data may be inadequate to build accurate large-scale predictive models. Open-source landslide inventories are often incomplete in spatial and temporal terms, with heterogeneous geometries, thus generating a data sparse environment consisting of a variety of low-accuracy datasets that need to be integrated and cross-validated to gain reliability.
In this study, the adoption of a combined approach based on GIS tools and Machine Learning techniques allowed to estimate landslide susceptibility based on both real and synthetic Landslide Initiation Points (LIPs). Open-source landslide inventories have been collected, cross-validated, and integrated in a unique database, thus creating a richer data product that contains the strengths but overcomes the weakness of each contributing dataset. As the number of LIPs was too low to train reliable ML models, we developed a methodology based on the features of occurred landslides in order to derive synthetic LIPs to boost the original database by three times. This approach has been applied to the Metropolitan area of Rome (Lazio, Central Italy), where rainfall-induced shallow landslides have been widely overlooked.
The final database with LIPs and predisposing factors has been used to create and validate different ML models and the most accurate one was then deployed to estimate landslide susceptibility for the whole area of the municipality of Rome with a resolution of 5 meters. The obtained results were then compared with pre-existing, regional, national, and European scale susceptibility maps to assess their reliability in case more detailed studies are not available. Eventually, rainfall probability curves were estimated to evaluate the temporal dependence of rainfall-induced shallow landslides.
How to cite: Mastrantoni, G., Caprari, P., Esposito, C., Marmoni, G. M., Mazzanti, P., and Bozzano, F.: Data requirements and scientific efforts for reliable large-scale assessment of landslide hazard in urban areas, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-4669, https://doi.org/10.5194/egusphere-egu22-4669, 2022.
Landslide susceptibility can be evaluated by using different statistical approaches. Among these, the methods based on conditional analysis exploit the observed incidence of landslides into homogeneous statistical domains (corresponding to single classes of each geo-environmental variable or to multivariate Unique Condition Units) to estimate their landslide susceptibility. Thus, the results of these types of analysis can be heavily compromised by the completeness or representativeness of the adopted landslide archive. On the other hand, inference-based frequentist methods allow scoring landslide susceptibility by using limited samples of cases, provided the calibration samples are statistically representative of the whole population, assuming that the lacking cases are missing completely at random.
This research aims to evaluate the effect of incomplete inventories in assessing landslide susceptibility, by using conditional analysis (Weight of Evidence, WoE; Frequency Ratio, FR) and inference-based (Binary Logistic Regression, BLR; Multivariate Adaptive Regression Splines, MARS) methods. In particular, we analysed the effects in terms of prediction skill of each of the four methods by reducing and randomly hiding the training calibration cases (and increasing the related validation cases).
The study was conducted in the Imera Settentrionale river basin (Sicily, Italy), by exploiting two different landslide archives (5134 earth flow and 1608 rotational/translational slides) and a set of 10 physical-environmental predictors. Cutoff-dependent and -independent metrics (ROC-curve analysis and confusion matrixes) were used to estimate the performance of the models.
As general assumptions, MARS and BLR modeling resulted as markedly more performing with moderately and asymptotically AUC improving up to 30-40% of the whole dataset, corresponding to the reaching of the relative optimal performance. A similar asymptotic AUC-increasing trend is described for WoE and FR, but with a lower performance. In particular, the optimal AUC values for rotational/translational slides range between 0.77 and 0.90, for BLR, 0.82 and 0.90, for MARS, 0.78 and 0.80, for FR, 0.76 and 0.78, for WoE. At the same time, a general lower model performance resulted for earth flows, with AUC values ranges of 0.69 and 0.75, for BLR; 0.75 and 0.79 for MARS; 0.67 and 0.70 for FR; 0.56 and 0.6, for WoE. Furthermore, differences in the selected predictors produced by the cases reduction were also explored through the analysis of the variable importance and the response curves.
How to cite: Martinello, C., Mercurio, C., Cappadonia, C., Mineo, G., Bellomo, V., Azzara, G., and Rotigliano, E.: Exploring the effect of inventory un-completeness in landslide susceptibility assessment: a test for conditional analysis- and regression-based models., EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-5190, https://doi.org/10.5194/egusphere-egu22-5190, 2022.
An inventory of more than 2000 mass movement events from the last 20 years from the canton Ticino, in the south of Switzerland, was analysed. The pre-Alpine to Alpine setting, combined with a mild temperate climate makes for a large number of natural events per year. The inventory consists of entries for spatially located movement types corresponding to rockfalls (43 %), debris flows (28%), landslides (17%), and avalanches (12%), with some recorded variables (date, coordinates, etc.). Additional geometrical data, as well as data from four categories (topography, hydrography, land use, and geology) was collected and pre-processed. Both a simple analysis and a more complex ones were carried out. From the initial statistical analysis, we determined that the relevant controlling parameters in this context are slope, aspect, terrain roughness index, topographic wetness index, and general lithology; while geometrical aspects of importance are area, length, height difference, volume, and angle of reach. We also conclude that the most affected districts are those of Blenio, Mendrisio, Locarno and Bellinzona, where debris flows and avalanches, debris flows, rockfalls and rockfalls prevail, respectively. From the geometrical aspects, we conclude that that rockfalls and landslides tend to have smaller areas and perimeters than avalanches and debris flows, as expected, due to their mobility. However, the deposit lengths, height differences and volumes show similar patterns. The calculated angle of reach shows similar median and mode values at around 26º/30º, 33º, 34º/35º and 41º, for debris flows, avalanches, landslides, and rockfalls and respectively. Significant power law correlations were found between deposit length and the height difference (cf. Corominas, 1996), deposit volume and the movement area (cf. Guzzeti et al., 2009), and the distribution of rockfall volumes (cf. Dussauge et al., 2003). Possible further work with this inventory includes probabilistic approaches and the application of machine learning techniques for the establishment of the precise relationships between the different controlling parameters and each movement type.
Corominas, J. (1996). The angle of reach as a mobility index for small and large landslides. Canadian Geotechnical Journal, 33(2), 260-271.
Dussauge, C., Grasso, J. R., & Helmstetter, A. (2003). Statistical analysis of rockfall volume distributions: Implications for rockfall dynamics. Journal of Geophysical Research: Solid Earth, 108(B6).
Guzzetti, F., Ardizzone, F., Cardinali, M., Rossi, M., & Valigi, D. (2009). Landslide volumes and landslide mobilization rates in Umbria, central Italy. Earth and Planetary Science Letters, 279(3-4), 222-229.
How to cite: Gutierrez, A., Derron, M.-H., Jaboyedoff, M., and Pedrazzini, A.: Aspects derived from the geological, geometrical, and statistical analysis of the Ticino landslide inventory, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-5963, https://doi.org/10.5194/egusphere-egu22-5963, 2022.
The identification of assets susceptible to landslide-related damage is critical for planners, managers, and decision-makers in developing effective mitigation strategies. Recent applications of machine learning and data mining methods have demonstrated their use in geotechnical assessments including the spatial evaluation of landslide susceptibility.
At Climate X, we utilise tree-based machine learning techniques alongside geographic information system and remote sensing data to map landslide susceptibility across Great Britain. We compile several conditioning factors—including topographic, subsurface, land use, and climate-related data—and combine them with over 18,000 landslide instances, recorded in National Landslide Database. We evaluate the capabilities of several techniques including, decision tree, bagged tree, random forest, and balanced random forest (applies random undersampling of the majority, non-landslide class) for landslide susceptibility modelling. Several performance evaluation indices (area under receiver operator characteristic curve (AUC), precision, recall, F1 score) were used to assess and compare the performance of models. We show that the random forest is the most accurate of our models with an AUC of 94.7%. Our results demonstrate that tree-based algorithms form a robust method to analyse regional landslide susceptibility and provide new insights into locations susceptible to landslide-related damage across Great Britain.
How to cite: Mitchell, H., Brennan, J., Burke, C., Kluza, K., Ramsamy, L., and Zeneli, M.: Regional landslide susceptibility mapping using tree-based machine learning techniques, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-8286, https://doi.org/10.5194/egusphere-egu22-8286, 2022.
In the aftermath of the catastrophic 1877 eruption of Cotopaxi volcano, Ecuador, lahars triggered in the summit cone, after making havoc of the city of Latacunga, flowed into the Pastaza River gorge, eventually reaching the Amazon lowlands. Presently, just downstream the town of Baños, the Hidroagoyán Dam impounds the water of the upper Pastaza, creating a reservoir of about two million m3, and annually producing 2520 GWh of energy, or about the 10% of the national demand of Ecuador. Should an 1877-scale Cotopaxi eruption occur nowadays, which is not unlikely after the 2015 reactivation of the volcano, similarly originated lahars might impact the dam, overwhelming the protective bypass system designed to contain anomalous flood waves of the Pastaza river. We present here an assessment of the hazard that such lahars may imply to the very functioning of Hidroagoyán. The investigation exploits the predictive power of LLUNPIY, a Cellular Automata model for primary and secondary lahars, already validated when simulating the 1877 Cotopaxi north and southward lahars as far as Tumbaco and Latacunga, respectively. Specifically, the present preliminary simulation succeeds for the first time to describe the flow of the lahars along the Pastaza gorge, thus reaching the dam in Baños and beyond. LLUNPIY simulates lahars in a discretized space-time, where the values of altitude, erodible soil depth, lahar thickness, kinetic head and lahar outflows are updated for each cell at each step according to the following processes: 1) Lahar flows determination and shift, 2) Detrital cover erosion, 3) Energy dissipation by turbulence, and 4) Melting of Cotopaxi ice cap by pyroclastic bombs, the latter process being limited to the cells corresponding to the glacier. Simulation inputs are morphology, erodible pyroclastic cover, extension of the Cotopaxi ice cap, pyroclastic bombs’ duration and frequency; by modifying their values we are able to predict several different hazard scenarios, which as a whole represent a reliable forecast of what might happen to the Hidroagoyán dam and the energy production of Ecuador in the case of a novel eruption of Cotopaxi volcano.
Lupiano V. et al. (2018). Revisiting the 1877 Cataclysmic Lahars of Cotopaxi Volcano by a Cellular Automata Model and Implications for Future Events. CSAE'18.
Lupiano V. et al. (2021). LLUNPIY Simulations of the 1877 Northward Catastrophic Lahars of Cotopaxi Volcano (Ecuador) for a Contribution to Forecasting the Hazards. Geosciences 2021, 11, 81.
Frimberger T. et al. (2021). Modelling future lahars controlled by different volcanic eruption scenarios at Cotopaxi (Ecuador) calibrated with the massively destructive 1877 lahar. Earth Surface Processes and Landforms.
How to cite: Chidichimo, F., Lupiano, V., Catelan, P., Straface, S., and Di Gregorio, S.: Natural hazard assessment for strategic infrastructures: a study of Cotopaxi lahars’ impact upon the Hidroagoyán Dam in Ecuador, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-5595, https://doi.org/10.5194/egusphere-egu22-5595, 2022.
Landslides are one of the most common and dangerous natural hazards that occur worldwide. Their occurrence may cause material losses and even death. Therefore, it is important to incorporate any mitigation action to ensure safety. One of the first steps can be generation of the landslide susceptibility maps which portrays the terrain probability to landsliding. There are numerous methods for creating landslide susceptibility maps, and machine learning methods are recently widely used. Therefore, in this study, the XGBoost machine learning algorithm was also implemented.
However, many scientists reported that the most critical step in any prediction model is the selection of the most appropriate features. In the case of landslide susceptibility modelling, they are called landslide conditioning factors (LCFs). LCFs are selected based on expert knowledge, literature review, or based on various statistical approaches for feature selection. Among statistical approaches, Symmetrical Uncertainty (SU), Analysis of variance (ANOVA) or Pearson correlation index (PI) can be applied.
Therefore, the objective of this experiment was to evaluate the effect of the feature selection method on the accuracy of the maps of susceptibility to landslides. For the experiment, two various areas of interest have been evaluated in the area of Polish Flysch Carpathians. Also, various accuracy measures were used to evaluate model performance among them Area Under the Curve (AUC), precision, Recall, and F1-score.
Accuracy measures indicated that the best method for feature selection is Pearson correlation (F1 score on the level of 77.2 % and 79.4 %) for both study cases, however, the difference between these feature selection methods are not significant.
How to cite: Lewandowski, T. and Pawluszek-Filipiak, K.: Landslide susceptibility mapping by using various selection strategies of landslide conditioning factors and XGBoost , EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-12676, https://doi.org/10.5194/egusphere-egu22-12676, 2022.
Tue, 24 May, 10:20–11:50
Chairpersons: Xuanmei Fan, Sansar Raj Meena, Lorenzo Nava
In the framework of the CARG (Geological and Geomorphological Mapping of Italy) project, landslides are also mapped as constituting one of the main surficial layers, masking the bedrock lithologies and related stratigraphic/tectonic contacts. As such, they are frequently mapped with a low resolution both in terms of spatial pattern and typology characterization. In particular, typical landslides affecting slopes in the Italian Apennines (slides and flows) are frequently grouped inside large polygons sometimes at a small catchment scale. However, the possibility to exploit such a reference landslide inventory for landslide susceptibility assessment is of great importance.
In this test, the existing CARG landslide dataset for the “Visso” map (Marche, Italy) was split according to the movement typology by exploiting topography maps and orthophotos, thus producing rotational slides, earth flows, and complex landslides archives (198, 91, and 51 cases, respectively). Multivariate Adaptive Regression Splines (MARS)-based susceptibility models were following prepared by regressing each systematic landslide archive to a specific set of physical-environmental predictors, considered as determining for landslides activation. Furthermore, multicollinearity and variables importance analyses were carried out to verify their relevance and influence in landslide susceptibility assessment. Besides, a new type (LCL_SLU) of slope units, obtained by crossing classic hydrological partitioning with landform classification, was used as mapping units.
The results show good AUC (Area under the ROC curve) for all models when prediction skill is evaluated, with values of 0.82, 0.77, 0.78 for rotational slides, earth flows, and complex landslides, respectively; the same AUC became outstanding when success skill is detected, with 0.91, 0.95, and 0.99 scores, respectively. Finally, for potential use in territorial planning, an integrated map was produced by adding up the single-landslide susceptibility scores and ranking the output on a classical 0-1 scale. The final map reaches an AUC value of 0.89, confirming the high performance of the models.
The results of the test in the “Visso” map suggest as potentially very worth processing the landslide inventories already available from the CARG project to assess landslide susceptibility on a regional to national scale.
How to cite: Bufalini, M., Martinello, C., Cappadonia, C., Pambianchi, G., Rotigliano, E., and Materazzi, M.: From landslide mapping to susceptibility modeling: a test in central Italy, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-12253, https://doi.org/10.5194/egusphere-egu22-12253, 2022.
For many years, statistical based landslide susceptibility maps have been used to spatially display the relative landslide probability of large areas. Consequently, such maps serve as guidance for strategic territorial planning. In Lower Austria (approx. 19200 km²) a complete set of landslide susceptibility maps for all municipalities has been implemented in 2014. These maps resulted from using 12889 slides as observations and fitting a generalized additive model (GAM) with a variety of geomorphically meaningful explanatory variables. Aiming at easy interpretable maps, the three susceptibility classes minor (78% of all pixels within Lower Austria), moderate (16%) and major (6%) were defined. In these classes, 5%, 25% and 70% of the landslides were in the categories 1, 2 and 3, respectively. Since the completion of these susceptibility maps, nearly eight years have passed, and many new landslides have been mapped. This study investigates, if and to which degree the existing landslide susceptibility maps can correctly predict these new events.
This research aims to quantify the accuracy of the spatial predictions. Recently mapped landslides were obtained from two different sources: damage reports related to the “Baugrundkataster", and landslides mapped from hillshades of a high-resolution LiDAR DTM. Additionally, information on the quality of the original landslide inventory and the new ones is used to analyze the effects of only using high quality inventories in this explorative comparison.
First results give a similar occurrence percentage of recently mapped landslides in the same classes, in comparison with the original classification design. Depending on the inventory the occurrence percentage varies especially in the 3rd class. Preliminary analysis indicates that, depending on the inventory, 34 to 63% of the new landslides are situated in the 3rd category (designed to contain 70%). However, it is also observed even for the lower quality inventories, that more than 90% of the landslides are not more than 30 meters away from merged 2nd and 3rd category susceptibility class. Depending on the new inventory, this percentage can reach 97%, while up to 94% of the points are at 0m distance of the 2nd and 3rd classes. This is of major importance for the application of these maps, e.g. within spatial planning. Additionally other preliminary analyses already indicate a better proportional correspondence of landslides coinciding with the most landslide-prone 3rd category, when excluding lower quality samples.
The landslide susceptibility map will be recalculated based on the newly recorded events. The potential change of the spatial prediction will be quantified, and the causes of these potential changes will be analyzed. The identical methodological design is applied to ensure comparability and quality control.
How to cite: Lima, P., Steger, S., Petschko, H., Goetz, J., Schweigl, J., Bertagnoli, M., and Glade, T.: Exploiting newly available landslide data to verify existing landslide susceptibility maps a decade after their implementation, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-7351, https://doi.org/10.5194/egusphere-egu22-7351, 2022.
Landslide is one of the most destructive natural hazard in Himalaya. It is mainly caused by numerous geological, geomorphological and hydrological characteristics of the terrain, and generally triggered either by rainfall or earthquake. It poses a serious threat to human lives, environment and the built infrastructures of the region. It has been reported that every year around 300 - 400 fatalities occur in the Himalayan region and monetary loss incurred is ~ 100 million USD. Therefore it is necessary to demarcate different landslide susceptible zones in the region. This will help in the sustainable development of region and minimize the destruction caused by landslides. For the present study, large scale landslide susceptibility mapping for the state of Sikkim encompassing northern and eastern districts using Artificial Neural Network has been carried out.
Landslide susceptibility, the relative probability of occurrence of landslides in an area, is one of the prerequisites for the development of the area in this mountain terrain. To assess the landslide susceptibility in a region, it is essential to understand the spatial distribution of the active landslides and landslide deposits, and their controlling factors. The relative weightage to each landslide controlling factor is determined using appropriate models and finally the landslide susceptibility map is prepared.
Geologically, the area encompasses the rocks of the Lesser Himalaya and Higher Himalaya, demarcated from one another by Main Central Thrust (MCT) and mainly constitutes phyllite, schist, quartzite, schist and gneiss. An inventory of 247 active landslides and landslide deposits ranging in area from ~ 200 m2 to ~ 450700 m2 and thematic layers of fifteen possible causative factors of landslides viz. lithology, slope angle & aspect, elevation, curvature-plan, curvature-profile, topographic wetness index, stream power index, distance to drainage, road & thrusts, land use and land cover, normalized difference vegetation index (NDVI), and peak ground acceleration (PGA) have been prepared. Of the 247 landslides, 70% were randomly selected for the assessment of landslide susceptibility, and the remaining 30% were used for validating the model. The dependency rate of landslides on each causative factor were estimated using information gain value analysis and subsequently landslide susceptibility map was computed using artificial neural network (ANN) algorithm.
It has been noted that high and very high susceptible zones are mainly concentrated along the strike of the MCT, on south facing slopes as these are slopes experience concentrated rainfall due to the orographic barrier. The success rate of our model is 92% and prediction rate is 89%.
How to cite: Gupta, V., Kumar, S., Hermanns, R., Penna, I., Dehls, J., Sengupta, A., and Bhasin, R. K.: Spatial Prediction of Landslide susceptibility zones using Artificial Neural Network in the Sikkim Himalaya, India, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-11496, https://doi.org/10.5194/egusphere-egu22-11496, 2022.
The purpose of this work was the definition of a new set of environmental indicators for a fast estimation of landslide risk over very wide areas. The proposed methodology was performed in GIS environment using Italy (301.340 km2) as test case since it is a country characterized by a very high exposure to hydrogeological disasters and where landslides are very common.
The proposed indicators aim to characterize landslide risk by quantifying how much urban expansion interferes with geomorphological processes; to this end a landslide susceptibility map and a soil sealing/land consumption map were combined to derive a spatially distributed indicator over the whole Italian country (namely, Landslide Risk Index - LRI). LRI emphasizes how much anthropic elements are exposed to landslide processes, and it is a basic element which can be aggregated over larger spatial units to characterize them respect to risk. To this aim, LRI was aggregated at the municipal scale in order to define two more indexes named Average Landslide Risk (ALR) and Total Landslide Risk (TLR).
ALR was defined by the mean value of LRI for each municipality: it represents how hazardous is the area of the territory where the exposed elements have been located. TLR was defined as the sum of susceptibility values of all cells with land consumption within each municipality: it expresses how much the urbanization of a municipality involves areas which can be affected by landslides.
The highest values of ALR are located in small municipalities renowned as international holiday destinations located by the sea in rocky coasts; on the contrary, highest values of TLR are in large and densely urbanized municipalities and where large portions of the territory urbanized are located in hazardous areas. The obtained results are supported by evidence collected from other national databases of landslide hazard and risk.
Both indexes showed to be useful to evaluate if local administrations have been prudent in planning urban development or if they ignored the geomorphological hazards threatening its territory. The proposed indexes are simple to understand and they can be adapted to various contexts and at various scales (e.g provinces, districts or basins) and updated with very low efforts. Obviously, they represent an oversimplification of the complexity of landslide risk and they cannot substitute a detail quantitative risk assessment, nevertheless a thorough national-scale risk assessment is not yet feasible in Italy and this is the first time that a set of landslide risk indicators have been defined in Italy at national scale combining landslide susceptibility and land consumption maps allowing to gain preliminary insights about the landslide risk produced by the interaction between hillslope dynamics and urban expansion.
How to cite: Caleca, F. and Segoni, S.: Estimation of landslide risk at national scale by means of environmental indicators, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-11907, https://doi.org/10.5194/egusphere-egu22-11907, 2022.
Landslide susceptibility maps are often not validated after significant landslide events. In this work, we analyse the impact of the Vaia windstorm on landslide activity in Belluno province (Veneto Region, NE, Italy). The storm hit the area on October 27-30, 2018, causing 8,679 ha of damaged forests and widespread landslides. As shown in the case of windstorm Vivian (1990) and Lothar (1999) (Switzerland), extreme meteorological events can influence slope stability after three to ten years (Bebi et al 2019). Through multi-temporal landslide inventory mapping post Vaia event, we want to access and validate the landslide susceptibility maps produced by using pre-event data from the Italian Landslide Inventory IFFI and assess if the susceptibility has increased in the areas affected by the storm. We used artificial intelligence techniques to prepare multi-temporal inventory and susceptibility maps pre and post-event. In the pre-event event inventory, 5934 landslides and 14 landslide conditioning factors were used to prepare the susceptibility models. We then validate the pre-event landslide susceptibility maps using post-event inventory from the 2018 Vaia windstorm and a following intense rainfall event that occurred in the same area in December 2020. A total of 542 landslides were mapped after the 2018 Vaia storm event, and an update to the landcover map as forest damage layer was used for post-event susceptibility analysis. This study is one of the first attempts to validate pre-event susceptibility maps by utilising multi-temporal artificial intelligence-based landslide inventories in Belluno province (Veneto Region, NE, Italy).
Bebi, P., Bast, A., Ginzler, C., Rickli, C., Schöngrundner, K., and Graf, F., 2019, Forest dynamics and shallow landslides: A large-scale GIS-analysis: Schweizerische Zeitschrift fur Forstwesen, v. 170, p. 318–325, doi:10.3188/szf.2019.0318.
How to cite: Meena, S. R., Puliero, S., Bhuyan, K., Nava, L., Faes, L., Floris, M., Catani, F., and Lingua, E.: Multi-temporal landslide inventory for validation of landslide susceptibility maps after 2018 Vaia windstorm event in Belluno province (Veneto Region, NE, Italy). , EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-9847, https://doi.org/10.5194/egusphere-egu22-9847, 2022.
Predictive simulations of rapid complex geohazards remains a challenge as it requires multiple computationally demanding tasks – such as model selection, parameter inversion or uncertainty quantification. Complexity of the geohazard herein refers to the dynamics of the event, i.e. 1962 and 1970 Huascarán events in Peru, both of which started as rock-ice falls – latter with a much larger release volume – and resulted in debris – ice avalanches. Recent efforts demonstrated the promising high estimation capability of inexpensive-to-built Gaussian process emulators to replace expensive-to-run landslide run-out simulations for predictive modelling. Furthermore, parameter inversion based on active Bayesian learning has recently been shown to greatly benefit from the developed surrogate models. Such demonstrations were conducted on rather simplistic cases with flow models that require low number of parameters. Inclusion of entrainment, complex topography, and higher number of model parameters inevitably increases the dimension of input parameter space. This study investigates the estimation ability of Gaussian process emulators to estimate the run-out characteristics of 1962 and 1970 Huascarán events by considering the spatial variation of model parameters and entrainment. A GIS-based open source landslide run-out model, r.avaflow v2.3, was used to simulate both events. Effects of high dimensionality of input parameter space on the performance metrics of emulation has been addressed by increased number of simulations and parameter reduction techniques. Parameter inversion has been performed to calibrate the model by using a synthetic simulation as ground truth. Inverting synthetic field observations for a known ground truth simulation result now allows us to assess the information content of different candidate data.
How to cite: Yildiz, A., Zhao, H., and Kowalski, J.: Physics-informed machine learning to model rapid complex geohazards, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-2682, https://doi.org/10.5194/egusphere-egu22-2682, 2022.
Improving landslide prediction in time is key to reducing damage and fatalities in areas susceptible to landsliding. While most landslide early warning research has focused on establishing hydro-meteorological landslide thresholds on hourly to daily timescales, few studies globally have attempted to model or predict landslide seasonality. We use probabilistic models based on two intuitive metrics — counts of landslides and presence or absence of landslides — to predict landslide activity at monthly resolution. Our focus area is the Pacific Northwest region of the United States, which has one of the highest densities of landsliding in the country, and where seasonal landslide activity has been recognized but hardly quantified. We use Bayesian inference to combine data from five landslide inventories from the region with varying spatial and temporal coverage, data density, and reporting protocols to learn the regional pattern of seasonal landslide activity. Results of logistic and negative binomial regression show that the landslide season in the Pacific Northwest begins in November and is marked by credible increases in the probability of landsliding, average landslide intensity, and inter-annual variability. Landslide activity is highest between November and February, decreases from March through May, and stays low between June and October. Inter-annual variability in landslide activity is higher in winter than in summer months. These flexible models could be easily adapted to learn diverse seasonal patterns from other regions of the world, such as the East Asian Summer Monsoon peak observed in Japan or the Atlantic hurricane season fall peak seen in the Caribbean. Our results also show that Bayesian multi-level models are a promising way to combine data from multiple, seemingly incompatible landslide inventories from a single region with potentially wide-ranging future applications.
How to cite: Luna, L. and Korup, O.: Predicting seasonal landslide activity with Bayesian inference, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-4260, https://doi.org/10.5194/egusphere-egu22-4260, 2022.
The overarching goal of the study was to test the influence of climate-related spatially distributed predictors on rockfall susceptibility in an Alpine environment. The study focused over the central part of Aosta Valley (Western Italian Alps), where a large historical rockfall inventory and an extensive, multi-variable meteorological dataset are available for the period 1990-2020.
The first part of the study regarded the definition of process-based climate predictors for the susceptibility model. Based on previous studies (Bajni et al., 2021), three climate indices were known to influence rockfall occurrence in the area: effective water inputs (EWI, including rainfall and snow melting), wet-dry episodes (WD), and freeze-thaw cycles (FT). For each index, the spatially-distributed predictor for the susceptibility analysis was calculated as the mean annual exceedance frequency of previously defined thresholds. Such predictors were produced both starting from a station-based hourly dataset, and consequent regionalization, and a grid-based hourly dataset.
The second part of the study comprised the set-up of a rockfall susceptibility model by means of Generalized Additive Models (GAM), including topographic, climatic and two additional snow-related predictors (derived from a Snow Water Equivalent weekly gridded dataset, Camera et al., 2021). The validation of the produced models was carried out through a k-fold cross-validation (CV), while the evaluation of its performance was expressed in terms of area under the receiver operating characteristic curve (AUROC). Variable importance was assessed through the decrease in explained deviance (mDD%).
To improve and optimize the model, stepwise modifications of its setup were carried out:
- a visibility mask related to roads and main infrastructures was introduced to reduce the rockfall inventory bias.
- Models including alternatively the station-based and grid-based climatic predictors were compared. The evaluation was based both on the physical plausibility of the smoothing functions describing predictors behaviour, and in terms of quantitative performance. For the grid-based model, performance and predictors transferability were evaluated comparing a random CV, a spatial CV and a holdout CV.
- Concurvity among predictors was reduced through the implementation of a Principal Component Analysis.
The key results were: (i) the use of climate predictors (both station-derived and gridded-derived) resulted in an improvement of the model performance (AUROC up to 3%) in comparison to a topographic-only model; (ii) the climate predictors with the strongest physical significance were EWI and WD, with a mDD%= 5-10% each, followed by the maximum cumulated snow melting over a 32-day period (mDD%= 3-5%); (iii) the effect of FT was masked by elevation; (iv) the station-based models were more strongly affected by concurvity issues; (v) the PCA derived predictors maintained explainable physical meanings while consistently decreasing concurvity.
The presented procedure is reproducible in other environmental and climatic conditions and allows to implement process-related non-stationary susceptibility models, making them adaptable for future climate change scenarios.
Bajni et al., 2021. Landslides 18, 3279–3298. https://doi.org/10.1007/s10346-021-01697-3
Camera et al., 2021. Science of The Total Environment 147360. https://doi.org/10.1016/j.scitotenv.2021.147360
This study was supported by the RISK-Gest project operating under the INTERREG ALCOTRA 14/20 Programme.
How to cite: Bajni, G., Camera, C. A. S., and Apuani, T.: Precipitation, temperature and snow related predictors for a potentially dynamic rockfall susceptibility model in Aosta Valley, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-3840, https://doi.org/10.5194/egusphere-egu22-3840, 2022.
Landslides are among the most dangerous natural hazards, especially in developing countries. In these areas, where rain gauge networks are scarce, satellite rainfall products can be a viable alternative for landslide prediction. To date, only a few studies have investigated the capability and effectiveness of these products in regional-scale landslide prediction. We performed a comparative study on the reliability of ground-based rainfall products and satellite rainfall products for landslide prediction in India. We used a catalog of 197 rainfall-induced landslides over the 13-year period between April 2007 and October 2019. We calculated frequentist rainfall thresholds using GPM, SM2RAIN-ASCAT satellite products, and their merging, at daily and hourly temporal resolution, and ground-based data from the rainfall network of the Indian Meteorological Department (IMD) at daily resolution. The results indicate that satellite rainfall products outperform ground-based observations in the prediction of landslides due to their improved spatial (0.1° vs. 0.25°/pixel) and temporal (hourly vs. daily) resolutions. The best performance is achieved through the merging of GPM and SM2RAIN-ASCAT. These results open up the possibility for using satellite rainfall products in landslide early warning systems, particularly in poorly gauged areas.
How to cite: Brunetti, M. T., Melillo, M., Gariano, S. L., Ciabatta, L., Brocca, L., Amarnath, G., and Peruccacci, S.: Performance of satellite rainfall products for landslide prediction in India, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-2757, https://doi.org/10.5194/egusphere-egu22-2757, 2022.
The fact that Slovenia is highly exposed to landslides underlines the need for preventive measures to reduce the hazard associated with landslides. For this reason, in 2011 the Geological Survey of Slovenia (GeoZS) started developing the MASPREM system to predict landslides hazard due to increased rainfall at the national level.
In 2021, the MASPREM system was upgraded to a local landslide early warning system which was specifically developed for landslide-prone area in the hinterland of the settlement of Koroška Bela (Karavanke mountain, NW Slovenia). This area is known by numerous landslides, that represent the source area of a potential debris flows that could pose a threat to the settlement bellow. The triggering mechanisms behind this kind of landslides are related to various environmental conditions (e.g. geological conditions, tectonic settings, topography, etc.) and triggering factors such as prolonged and/or intense precipitation, changes in groundwater levels, erosion and earthquakes.
Since we cannot avoid the risk of landslides and have to adapt, it is important to understand and predict landslide behaviour. With the help of landslide monitoring early landslide activity can be detected and landslide impacts can be reduced.
To meet this need, we have implemented real-time geotechnical (extensometers), hydrometeorological (piezometers, rain gauges) and geodetic (GNSS antennas) sensors that enable temporal prediction of landslide dynamics. Based on analyses of monitoring data and reconstruction of previous event, threshold values (precipitation, displacements) were determined.
Additionally, we set up customised dashboards that allow access to all real-time monitoring sensors. In this way, GeoZS emergency service and stakeholders can access daily updated data presented on webpage at any time. In the future, we plan to upgrade the local warning system with emails alerts sent to registered users when determined threshold values will be exceeded.
Acknowledgement: The research was funded by the Slovenian Research Agency (Research Program P1-0419, project Z1-2638, Infrastructure programme I0-0007), the Administration of the Republic of Slovenia for Civil Protection and Disaster Relief, the Ministry of Environment and Spatial Planning, and the Municipality of Jesenice.
How to cite: Peternel, T., Zupan, M., Šegina, E., Jemec Auflič, M., and Šinigoj, J.: Development of a first local landslide early warning system in Slovenia , EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-1599, https://doi.org/10.5194/egusphere-egu22-1599, 2022.
Storm-induced landsliding is a global and recurrent hazard, likely to increase with the strengthening of extreme precipitation events associated with current climate change. Risks associated with landslide hazard could be mitigated, for example with early warning systems or forecasting procedures. However, these approaches require to have constrained a tight relation between rainfall characteristics and the occurrence of landsliding. A traditional approach has been to derive such relationships from the failure of individual landslides, but the development of landslide mapping from satellite imagery allows now to constrain large landslide inventories triggered by single storm. Thus, at regional scale, forecasting the region of occurrence of a widespread landsliding event may be easier than forecasting the failure of individual slopes.
In turn, this regional approach requires spatially and temporally resolved rainfall information about the storms which caused landsliding. In-situ measurements are often too sparse for this and rainfall estimates derived from satellite observations have been proposed as a potential solution to this problem. However, only few studies have assessed the ability of satellite multi-sensor precipitation products (SMPPs) to characterize adequately the rainfall events which caused landsliding. Here, we address this issue by testing the rainfall pattern retrieved by 2 SMPPs (IMERG and GSMAP) and a hybrid product (MSWEP) against a large, global database of 18 comprehensive landslide inventories associated with well identified storm events. We use the nearly 20 years of data of the products to compute local rainfall anomaly over each area during the events and in every year of available data, and assess if the spatial pattern of intense anomaly corresponds to the landslide pattern, and if years without reported landslides have low level of anomalies. We found that after converting event rainfall to anomaly, the three products do retrieve the largest anomaly (of the 20 years) during the major landslide event for a number of cases. Still, the spatial pattern is often at least partially offset from the landslide areas, and that in many cases large anomalies are retrieved in years without substantial landsliding. Typically short, intense and localized storms are often missed by the three products, while large scale storms (e.g., hurricanes) are mostly retrieved, although the quality of the retrieval varies with each product. Using radar measurements or lightning records, we also show that in a number of cases where the SMPPs rainfall anomaly is poorly collocated with the landsliding, this is likely due to a biased retrieval of the rainfall rather than some variations in the landscape propensity to rainfall-induced landslides. We conclude on some potential avenue to improve SMPPs, typically including space-borne lightning measurement and better accounting for orographic precipitations.
In conclusion, rainfall estimates derived from satellite may be helpful in analyzing and understanding the pattern of landsliding, provided they are normalized by local extreme rainfall to obtain rainfall anomaly. Still, to advance toward regional scale landsliding, such methods of rainfall anomaly should also be applied to nowcast products from SMPPs and possibly to forecast issued from modern weather model.
How to cite: Marc, O., Juca-Oliveira, R., Gosset, M., Emberson, R., and Malet, J.-P.: Global assessment of the skills of satellite precipitation products to retrieve extreme rainfall events causing landsliding, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-5022, https://doi.org/10.5194/egusphere-egu22-5022, 2022.
Tue, 24 May, 13:20–14:50
Chairpersons: Maria Cuevas, Fausto Guzzetti, Filippo Catani
Forecasting rainfall-induced landslides, whilst challenging, is increasingly important due to the impact these hazards can have on society. The difficulty in forecasting arises from the inherent variability of geo-environmental factors and the scale at which underlying processes operate. The availability of data required to develop and validate thresholds for operational purposes is often limited. In regions where data (e.g. meteorological, or geotechnical) is sparse or incomprehensive, it is important to have a framework to systematically fuse the incomplete datasets to aid the development of a threshold model or to supplement an existing preliminary trigger threshold model.
For this study, a bespoke conceptual hydrological model called the ‘BGS water balance model’ is used in Nilgiris (Tamil Nadu state, India) to integrate the ground and meteorological information for informed decision making on the landscape saturation condition. This simple conceptual model with applicability over a large area provides an approximation of the degree of saturation value that can be used to map the potential antecedent wetness pathway leading to the initiation of landslides.
In this session, the BGS water balance model features along with the study area geological characteristics, landslide controls, input datasets and sensitivity analysis will be discussed. Further, we will show the results of the back-analysed landslides and explore the value of this approach in the context of landslide forecasting.
How to cite: Nedumpallile Vasu, N., Banks, V., Mathiyalaghan, R., Kumar, S., Karmarkar, R., Singh, G., Kumar Mishra, A., Rossi, M., Arnhardt, C., Dashwood, C., Ghosh, S., and Bee, E.: Value of ground information in regions with limited landslide inventory for trigger threshold development — Application in Nilgiris, Tamil Nadu State, India, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-12728, https://doi.org/10.5194/egusphere-egu22-12728, 2022.
Shallow landslides of the slide-type movement represent potentially damaging events in mountain areas all over the world. These geomorphic processes are caused by a combination of predisposing factors (e.g., hillslope material), preparatory conditions (e.g., prolonged snow-melt), and triggers (e.g., heavy rainfall). Data-driven methods have been used to model shallow landslides at regional scales. Traditional approaches are mainly focused on the spatial dimension, whereas the space-time component remains a challenge.
This contribution is built upon data on past landslide occurrence from 2000 to 2020 events in the province of South Tyrol, Italy (7400 km²). The inventoried information systematically relates to damage-causing and infrastructure-threatening events. The methodical procedure included an initial delineation of slope units that were subsequently replicated in time (2000 to 2020) and randomly subsampled to generate balanced distributions of landslide presence/absence observations across space and time. Different spatial static factors and cumulative daily precipitation time windows were aggregated into the mapping units. A Generalized Additive Mixed Model (GAMM) was implemented to derive statistical relationships between the different static and dynamic factors and the occurrence in space-time of shallow landslides. The resulting predictions were validated from multiple perspectives and transferred into space for different combinations of dynamic factors (e.g., triggering and preparatory precipitation, seasonal effects).
The first results are promising. The exploratory analysis has revealed that from a temporal viewpoint, the best-performing model consists of a combination of preparatory and triggering factors while additionally accounting for seasonal effects. The further inclusion of the spatial static factors improved the modeling results. The developed approach shows the potential to integrate static and dynamic landslide factors for large areas by also accounting for the underlying data structure (e.g., repeated observations nested in space) and data limitations (e.g., accounting for spatial data incompleteness). The proposed method is expected to enhance the predictability of shallow landslides at multiple spatial and temporal scales and provide a better understanding of the role of the environmental processes. This study is framed within the PROSLIDE project, that has received funding from the research program Research Südtirol/Alto Adige 2019 of the Autonomous Province of Bozen/Bolzano – Südtirol/Alto Adige.
How to cite: Moreno, M., Steger, S., Lombardo, L., Crespi, A., Zellner, P. J., Pittore, M., Mair, V., and van Westen, C.: Space-time modeling of rainfall-induced shallow landslides in South Tyrol, Italy, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-9175, https://doi.org/10.5194/egusphere-egu22-9175, 2022.
Rainfall-induced landslides are a damaging natural hazard occurring worldwide. Generally, slope failure mechanisms are quite established, as they are related to pore water pressure increase or gradients (either in saturated or unsaturated soil conditions), while the hydrological processes that control the conditions that predispose the slopes to landslide triggering are rarely, or only indirectly, considered. In fact, understanding and modelling these processes, usually developing over spatial and temporal scales much larger than the landslide itself, have been neglected for decades by the scientific community involved in landslide hazard assessment.
More recently, increasing attention has been given to the driving hydrological processes in landslide field research, but several challenging aspects are still open: the inclusion of large scale (in time and space) processes in the assessment of the hydrological balance of the potentially unstable soil mass; the effects of drainage processes through the soil-bedrock interface at slope scale; the mismatch of soil mechanics and hydrological models, in terms of scale and process conceptualization; the inclusion of catchment hydrological information in landslide hazard assessment.
Identification of predisposing hydrological processes in hillslopes and their inﬂuence on landslide triggering can significantly improve the predictive performance of landslide models, whatever their application scale (i.e., from hillslope to regional) and level of complexity (i.e., from physically-based distributed to lumped empirical). Recently, studies that consider the role of predisposing hydrological processes in landslide triggering have been rising, and landslide hydrology is progressively establishing itself throughout the scientific community. A brief overview of some significant recent results of landslide hydrology is presented, with specific reference to: assessment of slope water balance for the identification of major hydrological processes predisposing slopes to failure; definition of empirical hydrometeorological thresholds for landslide prediction, by coupling triggering precipitation depth with either antecedent water content at slope scale, or catchment water storage.
How to cite: Marino, P., Greco, R., and Bogaard, T. A.: Landslide hydrology: new challenges in landslide prediction, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-6393, https://doi.org/10.5194/egusphere-egu22-6393, 2022.
Forecasting the time of imminent slope failures is a powerful component in local early warning systems. Different prediction methods have been developed and applied successfully since the 1960s, but the most used and commonly accepted is the inverse velocity method after Fukuzono (1985). Technical developments in real-time and remote monitoring in the last decade offer new possibilities to monitor the displacement of unstable slopes with high accuracy and high frequency. However, state-of-the-art failure time forecasting methods are not yet ready to simply use such data for prospective predictions. The inverse velocity method has not been developed with high-frequency and therefore usually noisy measurement data which require automatism and filtering which in turn influences the outcome of the forecasts. Also, it does not indicate the uncertainty of its forecasts by default. Furthermore, defining the starting point for the calculation of reasonable forecasts (onset of acceleration) in real time remains challenging while many studies in literature used the method retrospectively in post-event analyses.
We developed a prospective failure time forecasting model (PFTF model) based on the linear inverse velocity method which can handle high frequency data in real time or simulated real time. The model uses multiple smoothing windows for the input data and the inverse velocity calculation. This minimizes the influence of subjective decisions on the sensitive smoothing process and enables a statistical quantification of uncertainties. The onset of acceleration is detected automatically and in real time by using different quantiles of inverse velocities. The model runs a new calculation with every new available datapoint. The completely open-source code is written in R and will be available online after publication. To perform sensitivity analyses and calibrate the model, we used GNSS and inclinometer observations from before the acceleration phase until failure of a rock block at the Grabengufer (Randa, CH). We also tested the model with data from other historical events characterized by different geological settings, measurement techniques, and sampling rates ranging from 2 minutes to multiple hours.
Here, we show the potential of the developed PFTF model as a tool for prospective slope failure time forecasting. Our multiple smoothing approach minimizes subjective decisions, improves forecasting after automatic detection of the onset of acceleration, and enables a statistical evaluation of the forecasts´ uncertainty. The most essential pattern here is the transition from diverging, unreliable and unstable forecasts to converging, reliable and certain forecasts. After further validation with multiple datasets, the model will be applicable to many slope failure processes (slide, topple, fall, flow), different materials (rock, earth, ice, other) and different scales (m³-km³).
Reference: Fukuzono, T. (1985): A Method to Predict the Time of Slope Failure Caused by Rainfall Using the Inverse Number of Velocity of Surface Displacement. – Journal of Japan Landslide Society, 22, 2: 8–14.
How to cite: Leinauer, J., Weber, S., Cicoira, A., Beutel, J., and Krautblatter, M.: Towards prospective failure time forecasting of slope failures, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-7673, https://doi.org/10.5194/egusphere-egu22-7673, 2022.
Large areas of Campania (southern Italy) are characterized by steep slopes covered with shallow deposits of loose pyroclastic materials, laying upon bedrocks with different characteristics (i.e., limestones, dolomites, volcanic tuff). The pyroclastic covers, usually in unsaturated conditions, are frequently affected by rainfall-induced shallow landslides, which cause heavy damage to property and infrastructures and sometimes casualties. Owing to the brittle behavior of the involved soils, hardly exhibiting any deformation before failure, the occurrence of such landslides is not easily predictable, so that operational early warning systems for rainfall-induced landslides (LEWS) usually rely only on empirical thresholds based on precipitation information (i.e., intensity and duration of triggering rainfall event). Anyway, the reliability of landslide prediction would benefit from the inclusion of hydrological information about the condition of the slope cover before the onset of the triggering rainfall (e.g., Marino et al., 2020a).
Three years of continuous field monitoring carried out at the slope of Cervinara, located around 40 km north-east of the city of Naples, where a destructive flowslide occurred in December 1999, have provided insight of the hydrological processes controlling the water balance of the pyroclastic deposits, laying upon a densely fractured limestone bedrock, where a temporary perched aquifer develops during the rainy season (Marino et al., 2020b). This knowledge allowed setting up a physically based model capable of identifying the seasonality of the predisposing conditions leading to slope failure (Greco et al., 2018; Marino et al., 2021). Aiming at identifying the hydrological processes mostly affecting landslide triggering, the model is coupled with a stochastic rainfall generator (i.e., the Neyman-Scott rectangular pulse model), previously calibrated based on 20 years hourly rainfall data, obtaining a 1000 years long synthetic series of the slope cover response to precipitations (in terms of soil suction, water content, perched aquifer water level, and leakage through the soil-bedrock interface). The obtained synthetic dataset of rainfall and hydrological variables have been analyzed with machine-learning techniques, so to identify the most effective combination of variables for landslide predictions.
The analysis of the synthetic data allows identifying the most suitable variables to be monitored, for assessing the hydrologic conditions predisposing the slopes to failure. In fact, the obtained results are confirmed by the analysis of the available field monitoring data, indicating that coupling rainfall measurements with field and remote hydrological monitoring significantly improves landslide prediction.
Greco R, Marino P, Santonastaso GF, Damiano E (2018). Interaction between perched epikarst aquifer and unsaturated soil cover in the initiation of shallow landslides in pyroclastic soils. Water 10:948.
Marino P, Peres DJ, Cancelliere A, Greco R, Bogaard TA (2020a). Soil moisture information can improve shallow landslide forecasting using the hydrometeorological threshold approach. Landslides 17(9): 2041-2054.
Marino P, Comegna L, Damiano E, Olivares L, Greco R (2020b). Monitoring the Hydrological Balance of a Landslide-Prone Slope Covered by Pyroclastic Deposits over Limestone Fractured Bedrock. Water 12(12): 3309.
Marino P, Santonastaso GF, Fan X, Greco R (2021). Prediction of shallow landslides in pyroclastic-covered slopes by coupled modeling of unsaturated and saturated groundwater flow. Landslides 18(1): 31-41.
How to cite: Roman Quintero, D. C., Giudicianni, C., Marino, P., Santonastaso, G., and Greco, R.: Identification of hydrological monitoring variables for improvement of shallow landslides prediction in pyroclastic slopes of Campania, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-8547, https://doi.org/10.5194/egusphere-egu22-8547, 2022.
Landslide processes often cause great economic losses, infrastructure damage and numerous casualties in many mountains and hilly landscapes worldwide. Landslide processes are very diverse, and may be shallow or deep, slow, or fast, with translational or rotational movements, and can sometimes even have a compound nature, with a single event behaving in different ways along time or space. For example, under certain conditions, slow-moving landslides can increase their speed, becoming flows with a large mobility range and destructive energy.
Although the methods for creating landslide susceptibility and hazard maps are now well advanced, they often do not represent the diversity of the landslide processes. Moreover, they do not represent hazard to the different stages of land sliding sub-processes, like failure, movement, and deposition area. Even though these sub-processes are connected, the final outcome of a disastrous event can differ greatly according to the movement mechanisms and pre-event conditions. This way, reliable hazard maps for single landslides, that account for their changing behavior during motion, still faces significant challenges.
The core purpose of this research is to evaluate the mobility and hazard scenarios of three slow-moving landslides with varying extensions, depths, and topography. All the study areas are located in Lower Austria. The run-out of the landslides was estimated using r.avaflow, a physically based mass flow model. The depth and soil structure of the landslides has been previously investigated by geotechnical and geophysical analysis. Different scenarios were considered for the modelling, including different factors like landslide extent, soil depth, and assumed water saturation, that determines the flow velocity, extent, and viscosity and thus the spatial extent of the run-out. The temporal probability of failures was analyzed using a physically based slope stability model. Using rainfall, snow, and temperature records from nearby gauging stations in Lower Austria, each landslide event was linked to different triggering rainfall or snowmelt events, and the slope stability was evaluated in terms of their Safety Factor.
The output of the analysis is a set of different landslide run-out maps for each of the three study areas. These maps also include the temporal probabilities for each landslide, considering several extent and mobility scenarios. The results support the decision-making policies, including risk reduction measures, and the implementation of landslide early warning systems.
How to cite: Arango, M. I., Lima, P., Mergili, M., and Glade, T.: Mobility and hazard analysis of selected landslides in Lower Austria , EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-8646, https://doi.org/10.5194/egusphere-egu22-8646, 2022.
Hydrogeological instability is among the effects of climate change with major impact on people and built environments security. Among instabilities, landslides are responsible for significant human and economic losses worldwide.
Landslide dynamic is characterized by a broad range of velocity-scales, from the steady creeping slip to a catastrophic avalanche passing through the intermittent rapid slip. During these phases, the landslide undergoes different mechanical behaviours. In particular, during the triggering phase, the landslide behaves roughly like a rigid body and the driving process is the pore-pressure diffusion that causes the intermittent slipping of the involved material. Once the landslide is initiated, it follows various behaviours, e.g. we may have a flow-like motion typical of debris and mud flows, where the landslide follows a visco-plastic behaviour and the overall process becomes advection dominated.
We propose an efficient multi-core numerical framework solving a two-dimensional depth-integrated fluid dynamic model for the simulation of flow-type landslides such as debris and mud flows. The governing equations are solved on adaptive quadtree meshes via the classical two-step second order Taylor-Galerkin scheme with a classical flux correction finite element strategy to avoid spurious oscillations near discontinuities and wetting-drying interface. Possible extensions considered by the author, such as an implicit-explicit operator splitting strategy, to deal with stiff diffusion and source terms will be discussed. Extensions that however do not affect the data locality of the scheme so do not affect the efficiency of the parallel implementation. To avoid excessive refinement in non-interfacial regions, we implement an interface tracking strategy that ensures detail preservation at the wetting-drying interface. We test the numerical framework on a real case study located in the Northern Italy to show its ability to deal with real problems.
How to cite: Gatti, F., Perotto, S., De Falco, C., and Formaggia, L.: An efficient parallel depth-integrated adaptive numerical framework with application to flow-type landslides, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-6411, https://doi.org/10.5194/egusphere-egu22-6411, 2022.
This research has developed a system that monitors social media continuously for landslide-related content, using a landslide classification model to identify and retain the most relevant information. The system harvests photographs in real-time and interprets each image as landslide or not-landslide. To achieve this, a training model was developed and tested through independent and collaborative working to establish a large image dataset that has then been applied to the live Twitter data stream. This paper presents results from interdisciplinary research carried out by computer scientists at the Qatar Computing Research Institute (QCRI), earthquakes and social media specialists at the European-Mediterranean Seismological Centre (EMSC) and landslide hazard expertise from the British Geological Survey (BGS).
How to cite: Pennington, C., Bossu, R., Ofli, F., Imran, M., Qazi, U., Roch, J., and Banks, V.: A global landslide incident reporting demonstrator using AI to interpret social media imagery in near-real-time , EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-9752, https://doi.org/10.5194/egusphere-egu22-9752, 2022.
Deformation monitoring has been proven to be an effective way to forecast and mitigate landslide geohazards. With the development of monitoring technology and equipment, the GPS technology have been widely adopted in landslide surface displacement monitoring, and borehole inclinometer methods are often used to measure deep displacements. However, for landslides with large and abrupt deformations, a large amount of landslide deep displacement data can hardly be processed by traditional methods because of the shearing failures of inclinometers, which cause serious data redundancy. Considering the time-frequency characteristics of deep displacement data obtained from typical rainfall-reservoir induced landslides in China Three Gorges Reservoir (CTGR) area, a quadratic wavelet reconstruction and bispectrum analysis (QWRBA) method is designed for feature extraction and landslide state classification. During this process, two wavelet decompositions are first used to decompose the input deep displacement data into components with different physical meanings. Then, some reconstructed components and non-reconstructed components are analysed with a bispectrum. The deep displacement bispectrum features generated by the bispectrum analysis of each component are fused to obtain the eigenvalues of these bispectrum features, and the eigenvalues of the fused bispectrum features are used as the characteristic landslide deep displacement data. By utilizing the fused bispectrum features as the inputs of an adaptive moment estimation-based convolutional neural network (CNN), different deep displacement conditions are recognized as corresponding deformation states.
How to cite: Long, J., Li, C., Liu, Y., and Feng, P.: Evolution state identification of deep landslide displacement based on a quadratic wavelet reconstruction and bispectrum analysis method, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-4781, https://doi.org/10.5194/egusphere-egu22-4781, 2022.
On 27 January 2021, at 21:00 (UTC+8), a shallow loess landslide occurred in Heifangtai, Yongjing County, Gansu Province, northwest of the Chinese Loess Plateau. Fortunately, the independently developed GNSS system predicted the landslide 7 hours in advance. Although farmland and channels were buried and destroyed, no damage has been done to the lives and houses of residents. In order to explore the triggering factors and movement process of the landslide, based on the field investigation, we collected the precipitation and temperature data more than one year before the landslide and comprehensively used UAV photogrammetry, numerical simulation, and laboratory test for comprehensive research. It was found that as the temperature rose and freeze-thaw cycles, changes in mechanical properties of loess and unique stratum structure were the main factors triggering the landslide. The rise of temperature led to an increase in groundwater levels, and the strength of soil decreased gradually until shear liquefaction occurred. This landslide caused a substantial topographic change, which provided conditions for slope instability in the future. The process of landslide movement can be divided into three stages: start-up stage, severe sliding stage, and deceleration stage. Simulation results show that the maximum velocity was 22 m/s, and the maximum sliding distance was 393 m. The main movement period was 40 s, and the apparent friction angle was 5°. Finally, this study provides a reliable basis for studying dynamic process and failure mechanism of loess landslide.
How to cite: jiaxu, K., jianqi, Z., jianbing, P., jia, Z., jiaqi, M., shibao, W., and yuting, F.: A landslide in Heifangtai, Northwest of Chinese Loess Plateau: Triggered Factors, Movement Characteristics and Failure Mechanism, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-5517, https://doi.org/10.5194/egusphere-egu22-5517, 2022.
Deep learning is a data-driven approach that requires high-quality labeled data to construct training and evaluation datasets. However, there are few open landslide data sets at present, and the degree of standardization of data sets is low. Now, the most advanced instance segmentation algorithms require strongly supervised learning. The cost of acquiring new categories of images is prohibitive. A question is raised: Is it possible to train high-quality instance segmentation models for early landslide disaster identification on the premise that not all categories are marked with complete instance segmentation annotations?
This article mainly deals with the intelligent identification of the small and medium-scale loess-bedrock historical landslides in the east Gansu Province. We proposed a modified instance segmentation algorithm based on transfer learning. Specifically, (1) A self-made landslide dataset was constructed. Google Earth images were used as the data source, and Arc GIS was selected as the landslide interpretation software. Based on DEM and 1:50,000 detailed regional geological hazard survey data, landslide boundaries were manually circled using the dataset annotation software（label me）according to the landslides' features of color, spectrum, vein, and surface roughness in optical images. The method of regional separation of datasets was used, with Anding district of Dingxi city as the validation set (15%), and Tianshui city, Longnan city, and Qingyang city as the sampling areas of the training set (70%) and testing set (15%) in the dataset. (2) A novel segmentation algorithm for landslide instances was proposed. The algorithm combined partially supervised training with weight transfer function to achieve high precision landslide classification and boundary recognition on data set constructed by mixed label annotation method. (3) A new method of Mask scoring was adopted to solve the problem that the accuracy of instance segmentation was affected by the lack of Mask scoring in Mask-RCNN.
The results show that the proposed method is superior to other algorithms in precision, accuracy, and recall rate. In addition, the Mask-IOU threshold value of 0.5 was used to estimate the average accuracy higher than the Mask-IOU threshold value of 0.75. The improved algorithm is in the segmentation of small and medium-sized landslides better than for large landslides, which will help solve the problem that it is difficult to comprehensively monitor the small and medium-sized landslides in the geological field survey. And our algorithm is not sensitive to the diffident backbone network and can achieve stable improvement on different Backbones. The average accuracy is about 3.1. The result of the experiment verified with the landslide field survey data in the validation area demonstrates this algorithm is stable and adaptable.
How to cite: Wang, J., Chen, G., Derron, M.-H., and Jaboyedoff, M.: A Modified Mask-RCNN Algorithm for Intelligent Identification of Landslide Based on High-resolution Remote Sensing data, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-5493, https://doi.org/10.5194/egusphere-egu22-5493, 2022.
Tue, 24 May, 15:10–16:40
Chairpersons: Sansar Raj Meena, Maria Cuevas, Lorenzo Nava
Literature has been widely enriched lately by results of research on the effects of vegetation on slope stability. The abundant research carried out in the field arose from different purposes: many works were performed with the finality of developing slope stability models and improving their capability of represent the soil behavior, while for many others, the priority was deepening the knowledge on the vegetation effects for bioengineering purposes. All those studies have in common the consequences of having confirmed, deepened, and expanded our knowledge on the subject, in some cases exploring some aspects not considered in the past. Some authors focused on certain plant species, other on the influence of the forest management, still others on the effect of the moisture gradient and wildfires, exploiting the numerical modelling and/or the field work.
The present work aims to summarize the most recent studies about the vegetation effects in slope stability dynamics, focusing on the root reinforcement effect and its parameterization into slope stability models: the evaluation of root reinforcement in wide areas is analyzed with reference to the most recent studies; studies dealing with development of slope stability models that consider root reinforcement are reviewed, followed by works on the influence on slope stability of some plant species, forest management techniques, wildfires and moisture gradients.
The vast spatial and temporal variability characterizing the root reinforcement still represents an open challenge for research in distributed slope stability modelling of wide areas and every new research in the field is much needed. The results of the studies conducted to assess the root reinforcement impact of different plant species highlighted the high species-specific character of the parameter. That points out the importance to pursue the study of new plant species root reinforcement impacts as well as already studied plant species, but in different environmental conditions. The impact of forest structure disturbances due to sylviculture or wildfires on root reinforcement emerged as significative and further studies are therefore needed in this direction. Lastly, some recent works pointed out that soil moisture has a significant control on root tensile strength.
How to cite: Masi, E. B., Segoni, S., and Tofani, V.: Regional slope stability simulations: recent advances in root reinforcement modelling, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-10550, https://doi.org/10.5194/egusphere-egu22-10550, 2022.
Geomorphic hazards such as landslides and flash floods (hereafter called GH) often co-occur
and interact imposing significant impacts in the landscape. Particularly in the tropics, where
GH are under-researched while impact is disproportionally high, establishing regional-scale
inventories of GH events is essential to better understand the behaviour and the patterns in
GH event occurrence. Robust AI-based detection tools such as the IMCLASS classifier
provide an excellent solution to accurately determine the location of GH events. However,
they rely on accurate training samples and require some knowledge on the timing of the event.
This information is regularly unavailable when exploring for new GH events in inaccessible
areas such as the tropics. Here we present our first endeavours into an automated workflow
for detecting unknown events in the tropics using the IMCLASS detection tool associated to
an unsupervised building of training samples using time series of Copernicus Sentinel 2
imagery. Per pixel, we investigate the cumulative difference from the mean over time for a
multitude of spectral index time series (e.g. NDVI, BI, SAVI) and their related z-score time
series. The method allows us to distinguish GH-affected and non-affected pixels based on the
prominence of the peak, and determine an approximate timing based on the location of the
peak within the timeseries. Both information are then used as input for the IMCLASS
classifier. The method is highly optimized in terms of computation time allowing to process
large regions of interest. Preliminary results over Uvira, DRC and the Mahale Mountains,
Tanzania, have shown to be encouraging and provide insight into a more automated workflow
applicable on the regional scale where event occurrence and timing is yet unknown. Further
steps will consist of adapting the workflow to different landscape, topography and climatic
How to cite: Michea, D., Deijns, A., Deprez, A., Dewitte, O., Kervyn, F., and Malet, J.-P.: Endeavours into a more automated workflow for regional scale landslide and flash flood event detection in the tropics using IMCLASS, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-4637, https://doi.org/10.5194/egusphere-egu22-4637, 2022.
Cost-effective spatial landslide models play a critical role in landslide mapping after an event and landslide susceptibility modelling for spatial planning and hazard mitigation. Challenges faced by many researchers in compiling the necessary landslide inventories are the time-consuming instance labelling and imbalanced data when training machine-learning models. Active learning is a practical way of reducing labelling costs by selecting more informative instances for labelling by an expert. Although this method has increasingly been adopted in remote-sensing classification, it is relatively new in the context of landslide mapping. To test the performance and potential benefits of active learning in this context, we combined two common active learning strategies, uncertainty sampling and query by committee with a state-of-the-art machine-learning technique, the support vector machine (SVM). Their utility is illustrated in a case study in the Ecuadorian Andes by comparing their performances to SVMs with simple random sampling of training locations. Based on the mean AUROC (area under the receiver operating characteristic curve) as a performance measure, SVMs with uncertainty sampling tended to perform better than random sampling and query-by-committee strategies. Meanwhile, uncertainty sampling achieved more stable performances according to a lower AUROC standard deviation across repetitions. Taken together, under limited data conditions, active learning with uncertainty sampling is more efficient by selecting more informative instances for SVM training. Therefore, we suggest that this strategy can be incorporated into the workflow of interactive landslide modeling not only in emergency response settings but also to more efficiently generate landslide inventories for event-based landslide susceptibility modeling.
How to cite: Wang, Z. and Brenning, A.: Combining active-learning approaches with support vector machines for landslide mapping, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-1478, https://doi.org/10.5194/egusphere-egu22-1478, 2022.
Multi-temporal landslide inventories are crucial for understanding the changing dynamics and states of activity of landslide masses. However, mapping landslides over space and time is challenging as it requires lots of time and resources to delineate landslide bodies for affected areas. With the current advances in artificial intelligence models and acquisition of very high-resolution satellite imageries, the need to map landslides not just spatially, but also temporally, has become evident. Generating multi-spatiotemporal landslide inventories can allow to improve our understanding of evolving landslides and landslide re-activations, addressing the changing susceptibilities, and the associated risks to elements-at-risk. Furthermore, as a result of having multi-temporal inventories, the temporal probability of occurrence of landslides can also be investigated with the help of envelop curves based on variables like rainfall duration, intensity, cumulative rainfall, antecedent rainfall. Therefore, in this endeavour, we have developed a model that generates multi-temporal landslide inventories for some of the most affected landslide regions by using several inventories around the world, for example, in Nepal (Gorkha earthquake of 2015). This study is the first attempt to map landslides over space and time using the state-of-the-art artificial intelligence models and gives a new perspective at mapping landslides through a temporal lens. Subsequently, the modelled outputs are utilised to assess and understand the changing dynamic behaviour of past landslides.
How to cite: Bhuyan, K., Tanyas, H., Nava, L., Puliero, S., Meena, S. R., Floris, M., Catani, F., Van Westen, C., and Pareek, T.: Multi-spatiotemporal landslide mapping for landslide evolutionary investigation, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-1356, https://doi.org/10.5194/egusphere-egu22-1356, 2022.
Where landslide hazard mitigation is impossible, Early Warning Systems are a valuable alternative to reduce landslide risk. Nowcasting and Early Warning Systems for landslide hazard mitigation have been implemented mostly at local scale, as such systems are often difficult to implement at regional scale or in remote areas due to dependency on fieldwork as well as local sensors. In recent years, various studies have demonstrated the effective application of machine learning for deformation forecasting of slow-moving, non-catastrophic, deep-seated landslides. Machine learning, combined with satellite remote sensing products offers new opportunities for both local and regional monitoring of deep-seated landslides and associated processes.
We tested the opportunities for machine learning on a multi-sensor monitored Austrian landslide. Our goal was to link conditions on the slope to the deformation pattern, to nowcast the deformation accelerations four days ahead of time. The in-situ sensors enabled us to test various model configurations based on combinations of local, remote sensing and retrospective analysis data sources. Our early results with shallow neural networks provide important context for future attempts. The complexities encountered were twofold: the machine learning model is poorly constrained due to the limited time span of five years of observations, and standard error metrics, like mean squared error, are unsuitable for model optimizations for landslide nowcasting.
First, even in Europe, with a six-day repeat cycle for Sentinel-1, there will be less than 500 InSAR deformation estimates from the start of the mission early 2015 to the end of 2022. As as consequence, there are only a few uniquely identifiable accelerations at the slope, and their timing is poorly defined within the six days between acquisitions. Therefore, the amount of training data is limited compared to the potentially large number of variables in more powerful machine learning models. On the Austrian slope we could rely on local, daily deformation measurements, to reveal sub-weekly minor accelerations, and to simulate potential, future, data availability.
Second, training of machine learning models is typically aimed at minimizing the average error. However, the average is a poor descriptor of the landslide accelerations that are deviations from the average, long-term behaviour. An alternative error metric was developed, that is more resiliant to slight timing errors.
Therefore, landslide deformation nowcasting is not a straightforward application of machine learning and there is a long road ahead for the large scale implementation of machine learning in landslide nowcasting and Early Warning Systems. Next step will be to evaluate our model on a landslide with a stronger deformation signal and more rapid onset of acceleration. We expect that these additional experiments will strengthen our preliminary conclusion that a successful nowcasting system requires simple, robust models and frequent, high quality and event rich data to train the system.
How to cite: van Natijne, A., Bogaard, T., and Lindenbergh, R.: Lessons learned from deformation nowcasting at a deep-seated landslide, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-3913, https://doi.org/10.5194/egusphere-egu22-3913, 2022.
The study of landslides spans from pre-failure mechanisms to post-failure propagation. The risk posed by landslides often relies more on the latter, and quantitative analysis for it can also describe the hazard caused by landslides more intuitively. Traditional numerical methods, such as the finite element method (FEM), suffer from severe mesh distortions when dealing with the highly nonlinear problems of landslides, especially in the post-failure propagation, resulting in inefficient or even failed computations. Meshfree methods such as the material point method (MPM) can efficiently describe the large deformation process of a structure using material points by reducing the dependence on the mesh. However, its computational efficiency is much lower compared to FEM. Currently, MPM programs are written in languages like C/C++/Fortran, which are performant but difficult to implement and read, and in languages like MATLAB/Python, which are flexible and easy to read but at the cost of much lower performance. This is known as the “two-language problem”. A new programming language, Julia, recently rose to prominence in scientific computing. It is designed for high-performance computing, has many of the features of advanced programming languages, and solves the "two-language problem". Benefiting from the native support for GPU computing in Julia, we can easily introduce GPU computing in the program to efficiently simulate the dynamic process in the post-failure of landslide. Consequently, for such a computationally intensive task, programming a high-performance MPM in Julia would be an attractive alternative. We use the Generalized Interpolation Material Point (GIMP) method, a variant of MPM, to perform the simulations and demonstrate the capabilities of the Julia language for high-performance scientific computing.
How to cite: Huo, Z., Jaboyedoff, M., Derron, M.-H., Wyser, E., and Mei, G.: High-performance Material Point Method for Landslide Simulation in Julia, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-10247, https://doi.org/10.5194/egusphere-egu22-10247, 2022.
Mapping landslides after major triggering events (earthquake, large rainfall) is crucial for disaster response, hazard assessment, as well as for having benchmark inventories on which landslide models can be tested. Numerous studies have already demonstrated the utility of very-high resolution satellite and aerial images for the elaboration of inventories based on semi-automatic methods or visual image interpretation. However, while manual methods are very time consuming, faster semi-automatic methods are rarely used in an operational contexts, partly caused by data access restrictions on the required input (i.e. VHR satellite images) and by the absence of dedicated services (i.e. processing chain) available for the landslide community.
From a data perspective, the free access to the Sentinel-2 and Landsat-8 missions offers opportunities for the design of an operational service that can be deployed for landslide inventory mapping at any time and everywhere on the Earth. From a processing perspective, the Geohazards Exploitation Platform –GEP– of the European Space Agency –ESA– allows the access to processing algorithms in a high computing performance environment. And, from a community perspective, the Committee on Earth Observation Satellites (CEOS) has targeted the take-off of such service as a main objective for the landslide and risk community.
Within this context, we present a largely automatic, supervised image processing chain for landslide inventory mapping. The workflow includes:
- A segmentation step, which performances is optimized in terms of precision and computing time and with respect to the input data resolution.
- A feature extraction step, consisting in the computation of a large set of features (spectral, textural, topographic, morphometric) for the candidate segments to be classified;
- A per object classification , based on the training of a random-forest classifier from a sample of manually mapped landslide polygons .
The service is able to process both HR (Sentinel-2 or Landsat-8) and VHR (Pléiades, SPOT, Planet, Geo-eyes or every multi-spectral image with 4 bands, blue, green, red, NIR) sensors. The service can be operated in two modes (bi-dates, single-date; the bi-dates mode is based on change detection methods with images before and after a given event, whereas the mono-date mode allows a mapping of landcover at any given time).
The service is presented on use cases with both medium resolution (Sentinel-2, Landsat-8) and high-resolution (Spot-6,7, Pléiades) images corresponding landscapes recently impacted by landslide disasters (e.g. Haiti, Mozambique, Kenya). The landslide inventory maps are provided with uncertainty maps that allows identifying areas which might require further considerations.
Although the initial focus and the main usage of ALADIM is associated with the landslide analyses, there is a large panel of possible applications. The processing chain was already tested in different others contexts (urbanization, deforestation, agricultural land change, …) with very promising results.
How to cite: Deprez, A., Marc, O., Malet, J.-P., Stumpf, A., and Michéa, D.: ALADIM – A change detection on-line service for landslide detection from EO imagery., EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-3536, https://doi.org/10.5194/egusphere-egu22-3536, 2022.
Landslide inventories are used for multiple purposes including landscape characterisation and monitoring, or landslide susceptibility, hazard, and risk evaluation. Their quality can depend on the data and the methods with which they were produced. The poor visibility of the territory to investigate offered by the point of observation from which landslides are interpreted is not frequently considered as a source of error in manually produced inventories. In this work, we present an approach to relate visibility and spatial distribution of the information collected in field work based inventories and inventories obtained through interpretation of satellite images.
We first used the r.survey tool and a digital elevation model to model and classify the visibility of the territory explored by field work based inventories. Furthermore, we assumed uniform visibility for inventories obtained through interpretation of satellite images.
Then, we measured the landslide density in the different visibility classes of the field based inventories. Last, we simulated visibility classes for the image based inventories using the road net of the area as virtual observation points, and we measured the relative landslide density.
We applied this approach to four inventories: one was produced by photo-interpretation, another one concerns to a regional multi-temporal database and the other 2 were done by direct field-mapping.
Our results show that 1) the density of the information is strongly related to the visibility in inventories obtained through field work, where landslides are abundant in high visibility classes but rarely reported in low visibility classes; and 2) the density of information is almost constant in inventories obtained by photo-interpretation of images, but they suffer from a marked under representation of small landslides in areas with potentially high visibility, e.g. close to roads. We maintain that the proposed procedure can be useful to evaluate the quality of landslide inventories and drive their correct use.
How to cite: Bornaetxea, T., Marchesini, I., Mondini, A., Kumar, S., and Karmakar, R.: Terrain visibility can affect landslide data collection, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-5250, https://doi.org/10.5194/egusphere-egu22-5250, 2022.
According to the Centre for Research on Epidemiology of Disasters, every year landslides are to be blamed worldwide for at least 17% of all fatalities from natural disasters. Rainfall-induced shallow landslides are responsible for a significant number of those: they mobilize the first few meters (usually <2m) of soil, have high velocities and occur after abundant and prolonged rainfall events.
The runout of a landslide, defined as the difference between the total area of a landslide and its source area, from which the sediment is first mobilized, is what determines how far a landslide travels and how big the affected area is, and yet the runout is often neglected when it comes to analysing the overall hazard caused by potential landslides.
The land use practices have been proven as one of the factors which impact the susceptibility of an area to the formation of shallow landslides, it is however less clear if the land use also plays a role in influencing the size of the area of runout.
The aim of the present work is to investigate the correlation between the runout area and the land use in which the shallow landslide develops.
To do so, two inventories of landslides, which occurred in neighbouring regions in Northern Italy (Lombardy and Piedmont), comparable for lithology, land use, geomorphology and climate, were analysed.
The result of the analysis was that there were statistical differences in the distribution of the runout among different land use classes, meaning that an influence of the land use on the runout was highly probable. Such results could improve the comprehension on shallow landslides mobility and runout and could lead to the development of possible models of assessment of the runout at different scales.
How to cite: Giarola, A., Bordoni, M., Tarolli, P., Zucca, F., Galve, J. P., and Meisina, C.: Analysis of the influence of land use on the runout area of shallow landslides, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-10338, https://doi.org/10.5194/egusphere-egu22-10338, 2022.
The Norwegian Mass Movements inventory is crucial for producing landslide susceptibility maps and early warning thresholds. However, it has significant sampling and spatial bias, with approximately 90% of registered landslides found within 100 m of a road. Applying AI, and the computing power of Google Earth Engine, to extract information from earth observation data, has great potential to improve our understanding of the true spatial distribution of landslides in Norway. Recently, globally-trained generalised ML algorithms have been developed, aiming to detect landslides from satellite images in regions where they have not been previously trained. Here we investigate how these tools can be applied in Norwegian conditions.
This study consists of two parts; 1) to evaluate how well existing generalised ML landslide detection algorithms perform in Norwegian conditions, and 2) to investigate methods for automatically back-dating and extracting trigger information for newly detected landslides using the Google Earth Engine platform. Two generalised ML methods using Sentinel-2 images, proposed by Prakash et. al (2021) and Tehrani et. al (2021), were tested on the Jølster case study (30.07.2019) from western Norway. This case study is a very well documented example of a multiple landslide event, triggered by extreme rainfall, and represents some of the ‘unique’ fjord- and mountainous-environments in Norway. In part two; backdating and extracting trigger information with Google Earth Engine - the investigated methods were tested on specific debris flow at Vassenden, using Sentinel-2 satellite images and global precipitation datasets (GSMaP and GPM).
Preliminary detection results were relatively poor. The Prakash algorithm vastly overestimated landslide activity, and the Tehrani algorithm did not detect any landslides. We found that snow cover, seasonal vegetation and lighting changes in the input images - factors that greatly affect detectability of landslides in Norway - were not sufficiently accounted for in the two methods tested. In the second part; extracting the date and trigger information - a mean-NDVI time-series of Sentinel-2 images within a one-year window was produced for the landslide area, and the date range of vegetation loss determined. The precipitation datasets were filtered to identify the magnitude and time of maximum precipitation at the landslide point, within the previously determined date range.
To conclude, these early, generalised ML landslide detection models show good potential to be applied in Norway, however they do require retraining and further development to perform well in the local conditions. Due to the strong seasonal effects, a more suitable approach for improving landslide inventories could be to conduct annual regional surveys, then backdate the newly detected landslides and assign most-likely-trigger information. Modifications to the preparation of input images are recommended to account for the seasonal conditions, including a) widening the time window for image selection to one year, b) creating a cloud-free composite based on a modified greenest-pixel approach, and c) filtering for snow. We plan to expand this study to include case studies from a diverse range of locations and seasonal conditions in Norway, and to retrain and modify the machine learning pipelines to further improve detection results.
How to cite: Lindsay, E., Jarna, A., Fredin, O., and Nordal, S.: Towards automated registration of climate-related landslides in Norway by combining Google Earth Engine, global precipitation datasets and AI, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-12540, https://doi.org/10.5194/egusphere-egu22-12540, 2022.
Handshake – who is here?
Nickname already in use
Nickname to be defined
You are not registered for EGU22
According to our records, your EGU22 registration is not valid for today.
How to access NH3.6
The session NH3.6 is running on a video chat platform. Please find the button Enter Zoom Meeting to access.