Evaluating landslide inventories is crucial and the first step in assessing the extent of landslide event damage. Despite the several studies in landslide inventory preparation and assessment, there is a lack of standardised criteria for measuring their quality and completeness. This study aims to introduce an integrated approach for analysing different event inventories prepared by different geomorphologists. We considered five landslide inventories prepared by various authors following the 2015 Gorkha earthquake in Nepal [1-5]. We prepared susceptibility maps using multiple realisations of logistic regression, with slope units of the areas as a spatial basic unit for the analysis [6]. The goal was to analyse their differences or similarities and comprehend the influence of using them to prepare landslide susceptibility maps [7].

The key questions we explored were: How can the quality and reliability of landslide inventories be evaluated? And what are the similarities or differences in the landslide susceptibility maps generated using inventories from different research teams for the same event? To this end, we utilised three evaluation criteria: (i) an error index to check the discrepancies between inventories, (ii) statistical analysis to examine the inconsistencies in predisposing factors and susceptibility map performance, (iii) geospatial analysis to evaluate differences among inventories and their corresponding.

The study highlighted differences in landslide inventories and attributed them to differences in data collection methods and subjective judgments. It emphasised the need to address subjectivity for more accurate and consistent landslide mapping. The results from statistical analysis showed substantial differences in the areal extent and overlapping degree between inventories. The geospatial analysis, such as hot spots and cluster/outlier analysis, highlighted the distinctive differences in spatial patterns of landslide susceptibility maps corresponding to different inventories. The suggested geospatial methods offer investigators a viewpoint for quantitatively analysing earthquake-triggered landslide inventories and related susceptibility maps.

References

[1] Zhang et al., Journal of Mountain Science (2016). https://doi.org/10.1007/s11629-016-4017-0

[2] Gnyawali et al., Springer International Publishing (2017). https://doi.org/10.1007/978-3-319-53485-5_10

[3] Roback et al., US Geological Survey Data Release (2017). https://doi.org/10.5066/F7DZ06F9,   

[4] Kargel et al., Science (2016). https://doi.org/10.1126/science.aac8353

[5] Pokharel and Thapa, Journal of Nepal Geological Society (2019). https://doi.org/10.3126/jngs.v59i0.24992

[6] Alvioli et al., Journal of Maps (2022). https://doi.org/10.1080/17445647.2022.2052768

[7] Pokharel et al., Scientific Reports (2021). https://doi.org/10.1038/s41598-021-00780-y

How to cite: Pokharel, B., Alvioli, M., and Lim, S.: Statistical and geospatial approaches to evaluate the quality of earthquake-triggered landslide inventories, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-17851, https://doi.org/10.5194/egusphere-egu24-17851, 2024.

Accurate and timely landslide inventory is crucial, particularly after large-scale disasters like earthquakes or heavy rainfalls. While remote sensing enhances mapping speed, accuracy is vital to avoid missing or falsely identifying landslides. Effective mapping depends on factors like immediate access to high-quality imagery and skilled surveyors for ground truth definition.

In May 2023, Italy's Emilia-Romagna region experienced a severe hydrogeological emergency, which triggered thousands of landslides. The landslide inventory, crucial for emergency management, faced challenges due to the high number of landslides. Initial efforts using Copernicus Emergency Management Service and national resources faced limitations in completeness and reliability. Ultimately, the official inventory was based on a detailed manual mapping from high-resolution aerial imagery.

This work presents the magnitude of the triggering event, the types of the landslides occurred with respect to the geological constraints and discusses the potential benefits and limitations of automated landslide mapping methods in such scenarios. Specifically, more than 50.000 landslides have so far been mapped over an area of around 1000 km², which range from debris slides/avalanches to debris flows and rock block slides. The impact on infrastructures was significant especially on the road network. With respect to automatic mapping, two distinct techniques have been tested: the conventional NDVI (Normalized Difference Vegetation Index) method and the more sophisticated U-Net algorithm using different remote sensing images ranging in resolution from 10 to 0.2 m.

Results show that time-consuming creation of an extensive ground truth datasets is essential in order to evaluate the accuracy of automatic landslide mapping based on images of different resolution and quality, so to determine whether these methods can offer efficient alternatives to manual mapping in large-scale emergency situations.

How to cite: Berti, M., Corsini, A., Pizziolo, M., Tommaso, S., Ciccarese, G., Critelli, V., Dal Seno, N., Fabbiani, C., Generali, M., Ioriatti, E., Lelli, F., Mulas, M., Rani, R., Ronchetti, F., Scaroni, M., Tondo, M., and Zuccarini, A.: Landslide inventory following the May 2023 Romagna hydrometeorological event (Northern Italian Apennines): the unavoidable requirement for laborious manual mapping, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-11385, https://doi.org/10.5194/egusphere-egu24-11385, 2024.

Efforts to understand the controls on landslides rely heavily on manually mapped landslide inventories, but these are costly and time-consuming to collect, and their reproducibility is not typically well constrained. To test the performance of manual mapping we compare two or more manually mapped inventories of landslides triggered by five recent earthquakes: Kashmir in 2005, Aysén in 2007, Wenchuan in 2008, Haiti in 2010, and Gorkha in 2015. We find surprisingly poor agreement between these maps (at worst 8 % overlap and at best 30 %). This has implications both for how future models and/or classifiers are tested and for the interpretations that are based on these inventories. We then test the ability of a new automated landslide detection index (ALDI) to recover landslide locations. In more than 50% of cases, ALDI more skilfully reproduces landslide locations from one inventory (treated as the ground truth) than a second inventory for the same site based on ROC curve analysis. Finally we examine the spatial pattern of landslides identified in the different maps and the patterns of their agreement and disagreement to show that: 1) much of the disagreement appears to be due to georeferencing rather than landslide identification; and 2) the ALDI map, which is quick and easy to produce, can be used to identify georeferencing errors and thus post-process and improve manual maps.

How to cite: Milledge, D., Bellugi, D., and Densmore, A.: Manual Landslide Maps are Surprisingly Inaccurate but Automated Detection Could Help, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-17066, https://doi.org/10.5194/egusphere-egu24-17066, 2024.

For the development of accurate shallow landslide (translational debris and earth slides with a depth < 2 m) susceptibility assessments and further hazard or risk analyses, it is essential that complete and accurate landslide inventory data is available. Various methods are applied for the construction of shallow landslide inventories. However, it is known that the most used methods underreport landslides in forests, e.g. with visual interpretation of satellite/aerial imagery and manual mapping of landslides during field visits. To address this issue, several studies have instead used topographic Light Detection and Ranging (LiDAR) data to create their landslide inventories. These studies showed that landslides under forest cover can be mapped using topographic LiDAR, as LiDAR can penetrate the vegetation cover. The methods used in these studies can be divided into (1) methods using raster data derived from filtered LiDAR point-cloud data and (2) methods working directly on point-cloud datasets. The benefit of the raster-based methods is their computational speed and scalability, while point-cloud based methods are difficult to apply to larger areas, due to their high computational requirements, but have a greater measurement accuracy (e.g., landslide depth). This difference in accuracy is especially important for the mapping of shallow landslides, which often leave only limited traces in the landscape.

This study investigates how both methods can be combined to derive a semi-automatic workflow for mapping shallow landslides using LiDAR data that is accurate and scalable. The investigation focusses on mapping shallow landslides under forest, and on how the derived workflow for mapping landslides needs to be adapted to forested and non-forested areas. In a first step, potential landslide-prone areas are identified using the difference of pre- and post-event digital terrain models, an after-event digital terrain model and their related topographic derivatives such as the roughness coefficient and slope. In the next step, the identified areas are segmented and man-made topographic changes are removed, before they are further analyzed with a more accurate mapping technique using point-cloud data from the multiscale model-to-model cloud comparison (M3C2) algorithm. In addition to the M3C2 distances, the point-cloud based mapping will also make use of 3D shape features describing point location and orientation to increase the accuracy and robustness of the topographic change detection and estimation. The scalability of the workflow is tested by applying the workflow to several areas in the Tyrolean Alps (Austria).

First results, derived with a logistic regression model using the raster-based derivatives, show a distinct difference in the feature importance of the topographic derivatives when forested and non-forested areas are compared. In addition, the performance of the model also greatly benefits from a separate training in forested and non-forested areas, with an increase in the Area Under the Curve (AUC) value from 0.84 to 0.89 for, respectively, unseparated and separated training.

How to cite: de Vugt,  ., Jia, S., Mayr, A., Schneider-Muntau, B., Zieher, T., Perzl, F., Adams, M., and Rutzinger, M.: A scalable workflow for shallow landslide inventory construction based on multitemporal LiDAR data with the explicit inclusion of landslides in forests, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-9020, https://doi.org/10.5194/egusphere-egu24-9020, 2024.

Mountainous regions such as the Austrian Alps face a constant threat of natural hazards. Over time, this persistent danger has prompted a transition from heuristic hazard management strategies towards a more quantified risk culture. Since quantitative risk assessment heavily relies on understanding the occurrence frequency of the hazard processes under consideration, knowledge about past events and their characteristics becomes pivotal, thereby shaping the effectiveness and broader applicability of methodological workflows employed in this context.

We present challenges, and insights gleaned from the research project “gAia”, focusing on a data-driven susceptibility assessment for shallow landslides in Austria. The identified challenges mainly revolve around the quality of landslide inventories, which is influenced by factors like underreporting, inconsistent documentation, and lack of standardized data management practices. We thus recommend adopting FAIR (Findability, Accessibility, Interoperability, Reusability) principles and developing Data Management Plans to address these issues, and propose a general data management workflow:

• Identify data sources and contents: Collect information about data sources and characteristics in a (machine-readable) DMP to obtain an overview of all data sources and most important characteristics (e.g. format, size, license, context, bias limitations). This should support the contextualization and ability to reuse this data.
• Define processing activities: Explicitly define processing workflows to enhance reproducibility and transparency, using established standards such as Business Process Management (BPMN) or semantic web technologies to represent complex processes formally and make them more comparable and accessible to users.
• Define (meta)-data and process activities trace templates: Provide metadata templates for datasets and trace processing activities to improve interoperability and reusability. Define domain-specific vocabularies and use concepts such as datasheets, model cards, ML experiment tracking and model registry tools as well as task orchestration platforms for data engineering pipelines to make results more traceable and reviewable. 
• Monitoring processes for natural hazard event data: Implement processes to ensure adherence to quality metrics, with results published in machine-readable formats.

We detail the implementation of these steps using established concepts of traceability and provenance, and encourage to implement workflow tasks using common open source programming languages. In addition, we endorse the use of Git for version control and GitLab/GitHub as tools for facilitating collaboration and structuring technical tasks.

The benefits of the proposed data management strategies for enhancing quality and reliability of data as well as increasing overall transparency of processes are showcased in the gAia project. The project workflow, represented as a P-Plan, demonstrates the application of these strategies in different phases. Specifically, the importance of proper data management and adherence to FAIR principles for data-driven research and practical usability is highlighted using landslide inventories as a core example.

In summary, we provide insights into the complexities of geospatial data management in mountain hazard research and offer practical solutions to enhance the integrity and reliability of data for supporting effective risk assessment and disaster risk reduction.

gAia is funded through the KIRAS Security Research Program for Cooperative Research and Innovation Projects by the Austrian Research Promotion Agency (FFG) and the Federal Ministry of Finance, under grant agreement FO99988636910.

How to cite: Waltersdorfer, L., Siposova, A., Schlögl, M., and Mayer, R.: Adopting FAIR data management practices in mountain hazard research: Strategies for ensuring data quality for landslide susceptibility modeling, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-18951, https://doi.org/10.5194/egusphere-egu24-18951, 2024.

Due to the impact of climate change, the increasing frequency of extreme rainfall events, with concentrated rainfalls, commonly cause landslide hazard in the mountain areas of Taiwan. However, there are uncertainties for the predicted rainfall as well as the landslide susceptibility analysis.

This study employs machine learning approached, including the logistic regression method LR and deep learning method CNN, to analyze the landslide susceptibilities. Together with the predicted temporal rainfall, the predictive analysis of landslide susceptibility was performed in the adopted study area in Central Taiwan. The uncertainties within the rainfall prediction was firstly investigated before applied to the landslide susceptibility analysis. To assess the susceptibility of the landslides, logistic regression method LR and deep learning method CNN were applied. The results of predictive analysis, with the discussions on the accuracy and uncertainties, can be applied for a better landslide hazard management in the study area.

How to cite: Shou, K.-J.: On the Predictive Analysis of Landslide Susceptibility by ML Approached– for the Case in Taiwan, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-4938, https://doi.org/10.5194/egusphere-egu24-4938, 2024.

A landslide susceptibility modelling has been carried out by applying two machine learning regression algorithms (SVR and CatBoost), and later two population-based optimization algorithms (metaheuristics) such as PSO and GWO were integrated to assess whether the integration improved the performance of the two regression algorithms. A total of 18 predisposing factors were selected for the study. After the multicollinearity assessment and feature selection applying the information gain (IG) method, four predisposing factors (three factors with collinearity issues and one irrelevant factor) were excluded. Hence, 14 predisposing factors were selected for the modelling. The landslide susceptibility maps were thus created by applying the CatBoost, CatBoost-PSO, CatBoost-GWO, SVR, SVR-PSO, and SVR-GWO models. The validation employing different techniques (MAE, MSE, RMSE, and R²) confirmed that the CatBoost model (MAE = 0.065 and 0.071, MSE = 0.027 and 0.032, RMSE = 0.165 and 0.180, and R² = 0.890 and 0.869) is better than the SVR model (MAE = 0.179 and 0.181, MSE = 0.063 and 0.063, RMSE = 0.251 and 0.252, and R² = 0.746 and 0.745). The integration of optimization algorithms improved the performance of these two regression models, and the GWO has the best performance when compared to the PSO algorithm. Also, CatBoost-GWO (AUC = 0.910) has the best performance, followed by CatBoost-PSO (AUC = 0.909), CatBoost (AUC = 0.899), SVR-GWO (AUC = 0.868), SVR-PSO (AUC = 0.858), and SVR (AUC = 0.840). The Friedman and Wilcoxon-signed rank tests confirmed that the models are significant. The feature importance assessment using the CatBoost confirmed elevation, slope, geomorphology, road, and soil bulk density as the top five important predisposing factors.

How to cite: Ajin, R. S., Segoni, S., and Fanti, R.: Optimization of ML-based regression models applying metaheuristic algorithms to determine the landslide susceptibility, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-19268, https://doi.org/10.5194/egusphere-egu24-19268, 2024.

The integration of Machine Learning (ML) into susceptibility mapping and hazard modelling has unveiled complex relationships between landslide predisposing factors and occurrence. However, understanding the practical implications of landslide susceptibility results still crucial. One concern is the conventional treatment of landslide inventory data, where various landslide types are uniformly addressed during model training, proving unrealistic given diverse geological and geomorphological conditions for distinct landslide types. Moreover, each landslide type requires a unique mitigation strategy, and valuable information is lost when treating all landslides uniformly. Another challenge is how susceptibility models typically present probabilities, while practical applications demand clear categories to avoid underestimation or overestimation of risks.

This study explores the practical application of landslide susceptibility to linear infrastructure, specifically railway design. The investigation centres on two established models, comparing each other: firstly, the classical statistical-based method, Weight of Evidence (WoE); and secondly, a ML method, the Generalized Additive Model (GAM). WoE was chosen for its clarity, while GAM accommodates continuous variables, offering a nuanced understanding of non-linear relationships. The study area encompasses a new 22.11 km railway stretch in the central Marche Region, Italy. We assessed the susceptibility to five landslide types, as classified in the Italian Landslide Inventory (IFFI). The models (WoE and GAM) are applied using different landslide types as distinct training datasets, resulting in unique susceptibility maps for each type. Additionally, the study evaluates the models using the Area Under the Receiver Operating Characteristic (AUROC) curve, providing insights into their performance. The rockfall susceptibility map demonstrates high reliability (AUROC of 0.942 with WoE and 0.978 with GAM), while slide-type landslides show more modest but fairly good results (AUROC of 0.696 with WoE and 0.784 with GAM).

To provide a comprehensive understanding of the study area, overall landslide susceptibility was calculated, corresponding to the probability of failure for any type of landslide. The overall probability was obtained by implementing a complementary probability approach for landslide probability analysis. A method is then proposed to classify the sensitivity of landslide types by considering the difference in their influences, facilitating a clearer understanding of their contributions to the overall susceptibility assessment.

The second practical concern involves the accurate definition of hazard classes, pivotal due to its direct impact on risk management, decision-making, and overall infrastructure resilience. The study introduces a novel approach, using mode calculation to consolidate results from various reclassification methods. This strategic use of mode calculation ensures a reliable representation of hazard classes, addressing the limitations of individual methodologies.

The results underscore the importance of considering multiple models and methodologies to obtain a comprehensive perspective in decision-making processes. Importantly, the study highlights the use of both classical statistical methods, exemplified by WoE, and ML methods like GAM, showcasing the benefits of a diverse analytical approach in landslide risk assessment for linear infrastructure.

How to cite: Rani, R., Sciarra, M., Rodani, S., and Berti, M.: Toward Pragmatic Landslide Susceptibility Mapping for Railway Planning: A Comparative Analysis of Statistical and Machine Learning Methods - The Case Study of the Marche Region, Italy, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-10842, https://doi.org/10.5194/egusphere-egu24-10842, 2024.

To mitigate the risk of landslides, building a model that can provide information on the spatial and temporal probabilities of landslides is essential yet challenging. Landslides are influenced by environmental factors, such as topography, geology, and mechanical properties of the soil, as well as triggering events like rainfall and earthquakes. This research leverages Random Forest algorithm for classification by creating multiple decision trees. Each tree is trained on a distinct, randomly selected subset of the dataset. The dataset includes specific static variables for each location, such as lithology, slope angle, aspect, curvature, and land use. Additionally, the study considers two dynamic variables for each location: high-resolution soil moisture data obtained from satellites to examine the impact of soil water content, and rainfall data.
By utilizing a unique rainfall-induced landslide database, which includes the location and time of landslide occurrences in the study area. The algorithm extracts the corresponding rainfall and soil moisture values preceding each landslide event and trains the model by adjusting both static and dynamic variables. The rainfall data is analyzed on two different time scales: short-term cumulative rainfall (1-72 hours before a landslide event) and medium-term cumulative rainfall (5-15 days before a landslide event). The outcomes are individual trees that determine the final class (landslide or non-landslide location) for each pixel based on the majority vote. The model's outputs, out-of-bag errors, and partial dependence plots provide insights into how each parameter influences the model's landslides predictions, and help to evaluate the impact of rainfall and soil saturation conditions on landslides occurrence both in space and in time.

How to cite: Peiro, Y., Ciabatta, L., Volpe, E., and Cattoni, E.: Spatiotemporal Modelling of Landslide Susceptibility Using Satellite Rainfall and Soil Moisture Products through Machine Learning Techniques, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-17519, https://doi.org/10.5194/egusphere-egu24-17519, 2024.

Landslide risk assessment is ineludible for communities in mountainous regions of the world. Life and habitat loss plague these communities due to rapid mass movements. Cloud burst conditions and strong earthquake tremors trigger thousands of landslides that can run amuck in the terrain. One way to assess the risk is by field investigations. Vast areas cannot be covered on foot and other limitations such as a lack of accessible roads. The next viable alternative is aerial photography and unmanned aerial vehicles (UAV). The range and accuracy of the deployed drones and aircraft also limit this method of earth observation. Overcoming these limitations, optical satellite images give us information for a vast area and can observe the surface daily, which is only limited by cloud cover. These satellites have advanced to give highly accurate images with a resolution of 3m/pixel.

Remote sensing image processing techniques are used to generate automated inventory of landslides. Most of the current algorithms use segmentation algorithms to map the landslide polygons but do not include the urban settlements that are vulnerable to these mass movements. Planet lab images (3m/pixel) and Google earth images (0.2m/pixel) can be acquired at a daily temporal scale. These images can be further processed to identify the scars caused by the landslides and the outlines of urban settlements. Sridharan et al 2020 and 2022 have looked at discrete wavelet transforms (DWT) and deep learning algorithms for assessing the proximity of these landslide scars to urban settlements.

In this work, we extend these algorithms with high-resolution satellite images and identify the risk caused by landslides to communities. We collected images for landslides from various parts of the world that are observed to be active and added them as one class of training dataset. From the urban settlements that are in steep terrain, we collect a second set of images as another training class. More than 3000 images are used for training and validation with a 70:30 train-test split. The models are then tested by classifying a set of images that contain both classes to assess their potential in identifying landslide risk to communities. Classic Support vector machine is used as a classifier after extracting features by DWT. Traditional deep learning algorithms such as AlexNET and ResNET are trained and tested on the satellite images. We observe both DWT and Deep learning algorithms  have good overall accuracy in extracting features and validating them while there are some accuracy differences in identifying the risk in mixed images that have both the classes. The models are rated based on the confusion matrix and AUC-ROC. While various DWT give an accuracy in range of 90 – 96%, the deep learning models have a range of 95 – 98%.

How to cite: Sridharan, A., Ajai A.S, R., and Gopalan, S.: The Use of Wavelet Transforms and Deep learning algorithms to identify risk caused by landslides from multitemporal satellite images, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-16687, https://doi.org/10.5194/egusphere-egu24-16687, 2024.