Showing papers in "Landslides in 2022"

Landslide detection using deep learning and object-based image analysis

Abstract: Abstract Recent landslide detection studies have focused on pixel-based deep learning (DL) approaches. In contrast, intuitive annotation of landslides from satellite imagery is based on distinct features rather than individual pixels. This study examines the feasibility of the integration framework of a DL model with rule-based object-based image analysis (OBIA) to detect landslides. First, we designed a ResU-Net model and then trained and tested it in the Sentinel-2 imagery. Then we developed a simple rule-based OBIA with only four rulesets, applying it first to the original image dataset and then to the same dataset plus the resulting ResU-Net heatmap. The value of each pixel in the heatmap refers to the probability that the pixel belongs to either landslide or non-landslide classes. Thus, we evaluate three scenarios: ResU-Net, OBIA, and ResU-Net-OBIA. The landslide detection maps from three different classification scenarios were compared against a manual landslide inventory map using thematic accuracy assessment metrics: precision, recall, and f1-score. Our experiments in the testing area showed that the proposed integration framework yields f1-score values 8 and 22 percentage points higher than those of the ResU-Net and OBIA approaches, respectively.

Landslide detection in the Himalayas using machine learning algorithms and U-Net

Abstract: Abstract Event-based landslide inventories are essential sources to broaden our understanding of the causal relationship between triggering events and the occurring landslides. Moreover, detailed inventories are crucial for the succeeding phases of landslide risk studies like susceptibility and hazard assessment. The openly available inventories differ in the quality and completeness levels. Event-based landslide inventories are created based on manual interpretation, and there can be significant differences in the mapping preferences among interpreters. To address this issue, we used two different datasets to analyze the potential of U-Net and machine learning approaches for automated landslide detection in the Himalayas. Dataset-1 is composed of five optical bands from the RapidEye satellite imagery. Dataset-2 is composed of the RapidEye optical data, and ALOS-PALSAR derived topographical data. We used a small dataset consisting of 239 samples acquired from several training zones and one testing zone to evaluate our models’ performance using the fully convolutional U-Net model, Support Vector Machines (SVM), K-Nearest Neighbor, and the Random Forest (RF). We created thirty-two different maps to evaluate and understand the implications of different sample patch sizes and their effect on the accuracy of landslide detection in the study area. The results were then compared against the manually interpreted inventory compiled using fieldwork and visual interpretation of the RapidEye satellite image. We used accuracy assessment metrics such as F1-score, Precision, Recall, and Mathews Correlation Coefficient (MCC). In the context of the Nepali Himalayas, employing RapidEye images and machine learning models, a viable patch size was investigated. The U-Net model trained with 128 × 128 pixel patch size yields the best MCC results (76.59%) with the dataset-1. The added information from the digital elevation model benefited the overall detection of landslides. However, it does not improve the model’s overall accuracy but helps differentiate human settlement areas and river sand bars. In this study, the U-Net achieved slightly better results than other machine learning approaches. Although it can depend on architecture of the U-Net model and the complexity of the geographical features in the imagery, the U-Net model is still preliminary in the domain of landslide detection. There is very little literature available related to the use of U-Net for landslide detection. This study is one of the first efforts of using U-Net for landslide detection in the Himalayas. Nevertheless, U-Net has the potential to improve further automated landslide detection in the future for varied topographical and geomorphological scenes.

Exploring a landslide inventory created by automated web data mining: the case of Italy

Abstract: Abstract Nowadays, several systems to set up landslide inventories exist although they rarely rely on automated or real-time updates. Mass media can provide reliable info about natural hazard events with a relatively high temporal and spatial resolution. The news publication about a natural disaster inside newspaper or crowdsourcing platforms allows a faster observation, survey, and classification of these phenomena. Several techniques have been developed for data mining inside social media for many natural events, but they have been rarely applied to the automatic extraction of “landslide events”. This source of information allows continuous feedback from real world, and news concerning landslide events can be rapidly collected. In this work, the newspaper articles about landslides in Italy are automatically collected by an existing data mining algorithm, based on a semantic engine. The news has been analysed to assess their distribution over the territory and to verify the possibility of using them for hazard mapping purpose. In 10 years, from 2010 to 2019, the algorithm identified and geolocated 184322 articles referring to 32525 generical events (“news”). At first, the collected data underwent to a manual verification, followed by a classification based on news relevance, localization accuracy and time of publication. Then, these data have been used to identify the areas and the periods most affected by landslide phenomena. The analyses show that almost 42% of Italian municipalities have been affected by landslide. According to the results, the use of data mining is helpful for the creation of landslide databases where the day and the approximative location (municipality) of the possible landslide triggers are known. This database, in turn, can be used for scientific purposes, as the definition of the meteorological condition associated with landslide initiation, the validation of risk maps. It can also be used for a proper land use or risk mitigation planning, since the most landslide-prone municipalities can be defined.

UAV applications to assess short-term dynamics of slow-moving landslides under dense forest cover

Abstract: Abstract The paper presents a methodology to rapidly assess and map the landslide kinematics in areas with dense vegetation cover. The method uses aerial imagery collected with UAVs (Unmanned Aerial Vehicles) and their derived products obtained from the structure from motion technique. The landslide analysed in the current paper occurred in the spring of 2021 and is located in Livadea village from Curvature Subcarpathians, Romania. This landslide affected the houses in the vicinity, and people were relocated because of the risk of landslide reactivation. To mitigate the landslide consequences, a preliminary investigation based on UAV imagery and geological-geomorphological field surveys was carried out to map the active parts of the landslide and establish evacuation measures. Three UAV flights were performed between 6 May and 10 June using DJI Phantom 4 and Phantom 4 RTK UAVs (Real-Time Kinematic Unmanned Aerial Vehicles). Because it is a densely forested area, semi-automated analyses of the landslide kinematics and change detection analysis were not possible. Instead, the landslide displacement rates and the changes in terrain morphology were assessed by manually interpolating the landmarks, mostly tilted trees, collected from all three UAV flights. The results showed an average displacement of approximately 20 m across the landslides, with maximum values reaching 45 m in the transport area and minimum values below 1 m in the toe area. This approach proved quick and efficient for rapid landslide investigations in a densely forested area when fast response and measures are necessary to reduce the landslide consequences.

Impact of extreme rainfall events on landslide activity in Portugal under climate change scenarios

Abstract: Abstract Rainfall is considered the most important physical process for landslide triggering in Portugal. It is expected that changes in the precipitation regimes in the region, as a direct consequence of climate change, will have influence in the occurrence of extreme rainfall events that will be more frequently, throughout the century. The aim of this study relied on the assessment of the projected future changes in the extreme precipitation over Portugal mainland and quantifying the correlation between extreme rainfall events and landslide events through Rainfall Triggering Thresholds (RTTs). This methodology was applied for two specific locations within two Portuguese areas of great geomorphological interest. To analyze the past frequency of landslide events, we resorted to the DISASTER database. To evaluate the possible projected changes in the extreme precipitation, we used the Iberia02 dataset and the EURO-CORDEX models’ runs at a 0.11° spatial resolution. It was analyzed the models’ performance to simulate extreme values in the precipitation series. The simulated precipitation relied on RCM-GCM models’ runs, from EURO-CORDEX, and a multimodel ensemble mean. The extreme precipitation assessment relied on the values associated to the highest percentiles, and to the values associated to the RTTs’ percentiles. To evaluate the possible future changes of the precipitation series, both at the most representative percentiles and RTTs’ percentiles, a comparison was made between the simulated values from EURO-CORDEX historical runs (1971–2000) and the simulated values from EURO-CORDEX future runs (2071–2100), considering two concentration scenarios: RCP 4.5 and RCP 8.5. In the models’ performance, the multimodel ensemble mean appeared to be within the best representing models. As for the projected changes in the extreme precipitation for the end of the century, when following the RCP 4.5 scenario, most models projected an increase in the extreme values, whereas, when following the RCP 8.5 scenario, most models projected a decrease in the extreme values.

Uncertainty quantification of landslide runout motion considering soil interdependent anisotropy and fabric orientation

Abstract: Abstract Natural soils often exhibit an anisotropic fabric pattern as a result of soil deposition, weathering, or filling. This paper aims to investigate the effect of soil interdependent anisotropy and fabric orientation on runout motions of landslides and evaluate the most critical fabric orientation for the post-failure behavior. The shear strength properties of soil deposit (i.e., cohesion $$c$$ c and friction angle $$\varphi$$ φ ) are modeled as negatively cross correlated bivariate random fields. The results reveal that the spatial variability and the negative cross-correlation of $$c$$ c and $$\varphi$$ φ notably influence the post-failure behavior. In addition, the rotation of soil layer orientation significantly affects the runout motion. Based on the analyses, the deposition orientation of $${30}^{^\circ }$$ 30 ∘ is identified to produce the highest mean value and standard deviation of the runout distance. The findings from this study highlight the importance of considering the orientation of soil stratification, rather than only the magnitude of shear strength, in assessing the post-failure behavior of a landslide.

Individual risk evaluation for landslides: key details

Abstract: Risk-taking is an essential part of life. As individuals, we evaluate risks intuitively and often subconsciously by comparing the perceived risks with expected benefits. We do this so commonly that it passes unnoticed, like when we decide to speed home from work or go for a swim. The comparison changes, however, when one entity (such as a government) imposes a risk evaluation on another person. For example, in a quantitative risk management framework, the estimated risk is compared with a tolerable risk threshold to decide if the person is ‘safe enough’. Landslide risk management methods are well established and there is consensus on tolerable life-loss risk thresholds. However, beneath this consensus lie several key details that are explored by this article, along with suggestions for refinement. Specifically, we suggest using the risk unit, micromort (one micromort equals a life loss risk of 1 in 1 million), in describing risk estimates and thresholds, to improve risk communication. For risk estimation, we provide guidance for defining and combining landslide scenarios and for recognizing where unquantified risk from low-probability/high-consequence scenarios ought to inform risk management decisions. For risk tolerance thresholds, we highlight the pitfalls of selecting unachievably low thresholds and suggest that there is no single universal threshold. Additionally, we argue that gross disproportion between costs and benefits of further risk reduction, which is integral to the As Low As Reasonably Practicable (ALARP) principle, is a commonly unachievable and counter-productive condition for risk tolerance, and other conditions centered on proportionality often apply. Finally, we provide several figures that can be used as risk communication tools, to provide context for risk estimates and risk tolerance thresholds when these values are reported to decision makers and the public.

Rockfalls: analysis of the block fragmentation through field experiments

Abstract: Abstract Fragmentation is a common feature of rockfall that exerts a strong control on the trajectories of the generated blocks, the impact energies, and the runout. In this paper, we present a set of four real-scale rockfall tests aimed at studying the fragmentation of the rocky blocks, from the global design of the field procedure to the data analysis and the main results. A total of 124 limestone, dacite, or granite blocks ranging between 0.2 and 5 m 3 were dropped from different heights (8.5 to 23.6 m) onto four slopes with different shapes (single or double bench) and slope angles (42º to 71º). The characteristics of the blocks, in particular the size, surface texture and joint condition, were measured before the drops. The trajectories of the blocks and both the initial and the impact velocities were tracked and recorded by means of three high-speed video cameras. A total of 200 block-to-ground impacts have been studied. On average, 40% of the blocks broke upon impact on the slope or on the ground, making it necessary to measure the fragments. The initial and final sizes of the blocks/fragments were measured by hand with tape, though photogrammetric techniques (UAV and terrestrial) were also used for comparison purposes. The information gathered during the field tests provides a deep insight into the fragmentation processes. On the one hand, the high-resolution slow-motion videos help to describe when and how the block breakage takes place and the spatial distribution of the pieces. On the other hand, it is possible to compute the block trajectories, the velocities, and the energy losses using videogrammetry. The results include, for instance, a block average fragmentation of 54% and 14% for the limestone and granitoids, respectively; the systematic inventory of the size fragments, which may be used for fitting the power law distributions; and after each breakage, the total angle of aperture occupied by the fragments has been measured, with values in the range 25º–145º. To figure out the different behavior of the blocks in terms of breakage/no breakage, each block-to-ground impact has been characterized with a set of parameters describing the energy level, the robustness of the substrate, and the configuration of the block contact at the impact point, among others. All these terms are combined in a function F , which is used to adjust the field data. The adjustment has been carried out, first, for the whole 200 events and later for a subset of them. The procedure and the results are described in the paper. Although the discrimination capability of F is moderately satisfactory, it is very sensitive to the test site and setup. It must be highlighted that these field tests are a unique source of data to adjust the parameters of the numerical simulation models in use for rockfall studies and risk mitigation, especially when fragmentation during the propagation is considered.

Geological and soil engineering properties of shallow landslides occurring in the Kutupalong Rohingya Camp in Cox’s Bazar, Bangladesh

Abstract: Abstract The Forcibly Displaced Myanmar Nationals (FDMN), historically known as ‘Rohingya’ who fled the 2017 ethnic atrocities and genocide in the Northern Rakhine State of Myanmar, took shelter in Cox’s Bazar District of Bangladesh. The camp network, known as Kutupalong Rohingya Camp (KRC), is situated in the tectonically active tertiary hilly terrain. The KRC has been experiencing hydrometeorological hazards, where landslides are frequent. This study investigated the slopes’ geological condition, engineering properties and human interventions, which influence the landslides. The exposed slopes were relatively high (> 10 m) and steep ranging from 40° to 60° that have numerous polygonal tension cracks and fissures. From the geological and geotechnical aspects, there are three successive units of slope materials: (1) residual soils of sandy silt with clay, (2) highly weathered silty sandstones and (3) shale/clay with silt and fine sand intercalations at the bottom of the slopes. Field observations revealed that most slope failures occurred in the residual soil and weathered silty sandstone units. The residual soils have a bulk density of 1.49–1.97 g/cm 3 , a liquid limit of 25–48%, a plasticity index of 5–16% and an undrained shear strength of 23–46 kPa. The silty sandstones have a bulk density of 1.44–1.94 g/cm 3 , an internal friction angle of 34°–40° and a cohesion of 0.5–13 kPa. The mineralogical composition determined by the X-ray diffraction shows low clay mineral content, which does not affect landslides. However, the slope geometry, low shear strength with strain softening properties and torrential rainfall accompanied by anthropogenic factors cause numerous landslides every year. This study will help take proper mitigation and preparedness measures for slope protection in the KRC area and surroundings.