Historical Landslide Inventories Research Articles

Landslide Susceptibility Mapping Using Machine Learning: A Case Study of Oregon

Landslide susceptibility assessment, a crucial component of disaster prevention and mitigation, has traditionally relied on geological datasets and commonly used evaluation metrics. However, these metrics, often applied to 2D susceptibility maps, may not capture the full complexity of the issue. In this study, we incorporated spatial distribution data, including a historical landslide inventory, and integrated morphological, geological, hydrological, and land-use features. This approach enabled the construction of predictive models using eight distinct algorithms. The evaluation of classifier performance scores unveiled the superior performance of gradient tree boosting. In contrast, logistic regression struggled to handle the nonlinear aspects of landslide susceptibility analysis. A finding in our research was the consistent disparity in predictive susceptibility maps, even when similar performance scores were achieved across different algorithms. This underscores the necessity of combining evaluation metrics with spatial mapping in landslide susceptibility model assessment. We also highlighted the critical role of digital elevation models in constructing effective models when feature data was limited. Conversely, when abundant information was available, the integration of multi-source data significantly enhanced the precision of landslide susceptibility mapping. Overall, this study provides valuable insights for spatial prioritization in landslide field investigations and enhances our understanding of landslide risk assessment.

Seismic Landslide Susceptibility Assessment Using Newmark Displacement Based on a Dual-Channel Convolutional Neural Network

Landslide susceptibility assessment (LSA) is an essential tool for landslide hazard warning. The selection of earthquake-related factors is pivotal for seismic LSA. In this study, Newmark displacement (Dn) is employed as the earthquake-related factor, providing a detailed representation of seismic characteristics. On the algorithmic side, a dual-channel convolutional neural network (CNN) model is built, and the last classification layer is replaced with two machine learning (ML) models to facilitate the extraction of deeper features related to landslide development. This research focuses on Beichuan County in Sichuan Province, China. Fifteen landslide predisposing factors, including hydrological, geomorphic, geological, vegetation cover, anthropogenic, and earthquake-related features, were extensively collected. The results demonstrate some specific issues. Dn outperforms conventional earthquake-related factors such as peak ground acceleration (PGA) and Arias intensity (Ia) in capturing seismic influence on landslide development. Under the same conditions, the OA improved by 5.55% and AUC improved by 0.055 compared to the PGA; the OA improved by 3.2% and AUC improved by 0.0327 compared to the Ia. The improved CNN outperforms ML models. Under the same conditions, the OA improved by 4.69% and AUC improved by 0.0467 compared to RF; the OA improved by 4.47% and AUC improved by 0.0447 compared to SVM. Additionally, historical landslides validate the reasonableness of the landslide susceptibility maps. The proposed method exhibits a high rate of overlap with the historical landslide inventory. The proportion of historical landslides in the very high and high susceptibility zones exceeds 87%. The method not only enhances accuracy but also produces a more fine-grained susceptibility map, providing a reliable basis for early warning of seismic landslides.

An Ensemble Approach of Feature Selection and Machine Learning Models for Regional Landslide Susceptibility Mapping in the Arid Mountainous Terrain of Southern Peru

This study evaluates the utility of the ensemble framework of feature selection and machine learning (ML) models for regional landslide susceptibility mapping (LSM) in the arid climatic condition of southern Peru. A historical landslide inventory and 24 different landslide influencing factors (LIFs) were prepared using remotely sensed and auxiliary datasets. The LIFs were evaluated using multi-collinearity statistics and their relative importance was measured to select the most discriminative LIFs using the ensemble feature selection method, which was developed using Chi-square, gain ratio, and relief-F methods. We evaluated the performance of ten different ML algorithms (linear discriminant analysis, mixture discriminant analysis, bagged cart, boosted logistic regression, k-nearest neighbors, artificial neural network, support vector machine, random forest, rotation forest, and C5.0) using different accuracy statistics (sensitivity, specificity, area under curve (AUC), and overall accuracy (OA)). We used suitable combinations of individual ML models to develop different ensemble ML models and evaluated their performance in LSM. We assessed the impact of LIFs on ML performance. Among all individual ML models, the k-nearest neighbors (sensitivity = 0.72, specificity = 0.82, AUC = 0.86, OA = 78%) and artificial neural network (sensitivity = 0.71, specificity = 0.85, AUC = 0.87, OA = 79%) algorithms showed the best performance using the top five LIFs, while random forest, rotation forest, and C5.0 (sensitivity = 0.76–0.81, specificity = 0.87, AUC = 0.90–0.93, OA = 82–84%) outperformed other models when developed using all twenty-four LIFs. Among ensemble models, the ensemble of k-nearest neighbors and rotation forest, k-nearest neighbors and artificial neural network, and artificial neural network and rotation forest outperformed other models (sensitivity = 0.72–0.73, specificity = 0.83–0.84, AUC = 0.86, OA = 79%) using the top five LIFs. The landslide susceptibility maps derived using these models indicate that ~2–3% and ~10–12% of the total study area fall within the “very high” and “high” susceptibility. The obtained susceptibility maps can be efficiently used to prioritize landslide mitigation activities.

Inventory and Spatial Distribution of Ancient Landslides in Hualong County, China

The establishment of a regional historical landslide inventory plays an indispensable role in landslide assessment and prevention. In this study, based on the Google Earth platform, an inventory of ancient landslides in Hualong County, Qinghai Province was established. The inventory includes 3517 ancient landslides with individual areas ranging from 2354.6 m2 to 12.44 km2. The dominant characteristics include an elevation of 2600–2800 m, slope of 10–20°, aspects SW, W, and NW, mudstone and sandstone of Paleoproterozoic, Neoproterozoic and Quaternary loess, 8–10 km from faults, 0–1 km from rivers, cultivated and grassland types, NDVI of 0.25–0.3, and an average precipitation in the range of 480–500 mm. In addition, the geometric analysis of landslides shows that the average height and length of ancient landslides in the study area are 151.92 m and 429.52 m, respectively. The power law relationship between the two is L = 0.41 × H1.37. The ancient landslide inventory of this study exhibits an integrated pattern of the development characteristics and spatial distribution of landslides in the Tibetan Plateau and the upper Yellow River basin, as well as providing a significant reference for subsequent landslide susceptibility mapping in the area.

CNN-LSTM-ATTENTION DEEP LEARNING MODEL FOR MAPPING LANDSLIDE SUSCEPTIBILITY IN KERALA, INDIA

Abstract. As a typical type of natural disaster, landslides may result in injuries to humans, threats to property security, and economic loss. As such, it is important to understand or predict the probability of landslide occurrence at potentially risky sites. Using typically machine learning (ML) to estimate landslide susceptibility based on a landslide inventory and a set of factors that impact the occurrence of landslides is a common practice. However, in landslide susceptibility assessment, existing DL-based neural network methods use a fully connected layer to optimize the selection of factors, which limits their efficiency in extracting features of those contributing factors. In response to those problems, this study proposed a CNN-LSTM model with an attention mechanism (AM) to avoid the complex optimization of input factors while the same or even better prediction accuracy can be achieved. To compare our method with the existing ones, the historical landslide inventory and the remote sensing data of Kerala, India were used to produce the input variables needed in the methods considered. The results show that our method produced more accurate results, compared to those existing neural network methods (e.g. CNN, LSTM and CNN-LSTM).

Towards the Use of Land Use Legacies in Landslide Modeling: Current Challenges and Future Perspectives in an Austrian Case Study

Land use/land cover (LULC) changes may alter the risk of landslide occurrence. While LULC has often been considered as a static factor representing present-day LULC, historical LULC dynamics have recently begun to attract more attention. The study objective was to assess the effect of LULC legacies of nearly 200 years on landslide susceptibility models in two Austrian municipalities (Waidhofen an der Ybbs and Paldau). We mapped three cuts of LULC patterns from historical cartographic documents in addition to remote-sensing products. Agricultural archival sources were explored to provide also a predictor on cumulative biomass extraction as an indicator of historical land use intensity. We use historical landslide inventories derived from high-resolution digital terrain models (HRDTM) generated using airborne light detection and ranging (LiDAR), which are reported to have a biased landslide distribution on present-day forested areas and agricultural land. We asked (i) if long-term LULC legacies are important and reliable predictors and (ii) if possible inventory biases may be mitigated by LULC legacies. For the assessment of the LULC legacy effect on landslide occurrences, we used generalized additive models (GAM) within a suitable modeling framework considering various settings of LULC as predictor, and evaluated the effect with well-established diagnostic tools. For both municipalities, we identified a high density of landslides on present-day forested areas, confirming the reported drawbacks. With the use of LULC legacy as an additional predictor, it was not only possible to account for this bias, but also to improve model performances.

Atmospheric triggering conditions and climatic disposition of landslides in Kyrgyzstan and Tajikistan at the beginning of the 21st century

Abstract. Landslide is a major natural hazard in Kyrgyzstan and Tajikistan. Knowledge about atmospheric triggering conditions and climatic disposition of landslides in Kyrgyzstan and Tajikistan is limited even though this topic has already been investigated thoroughly in other parts of the world. In this study, the newly developed, high-resolution High Asia Refined analysis version 2 (HAR v2) data set generated by dynamical downscaling was combined with historical landslide inventories to analyze the atmospheric conditions that initialized landslides in Kyrgyzstan and Tajikistan. The results indicate the crucial role of snowmelt in landslide-triggering processes since it contributes to the initialization of 40 % of landslide events. Objective thresholds for rainfall, snowmelt, and the sum of rainfall and snowmelt (rainfall + snowmelt) were defined. Thresholds defined by rainfall + snowmelt have the best predictive performance. Mean intensity, peak intensity, and the accumulated amount of rainfall + snowmelt events show similar predictive performance. Using the entire period of rainfall + snowmelt events results in better predictive performance than just considering the period up to landslide occurrence. Mean annual exceedance maps were derived from defined regional thresholds for rainfall + snowmelt. Mean annual exceedance maps depict climatic disposition and have added value in landslide susceptibility mapping. The results reported in this study highlight the potential of dynamical downscaling products generated by regional climate models in landslide prediction.

Landslide types in the Slani Potok gully, Croatia

The Slani Potok gully (0.48 km2) is situated in the central part of the Vinodol Valley (64.57 km2) in Croatia, and it was formed in the Eocene flysch deposits. So far, the area of Slani Potok has been mainly referred in the scientific literature as being subjected to unusually intense soil erosion processes (i.e., ‘’excessive erosion’’), associated with landslides. However, the landslides were never investigated in detail, given the available research methods only involved field investigations. Therefore, the landslide types according to the most commonly used landslide classifications have remained undetermined. In this paper, landslide types in the Slani Potok gully are presented, identified and mapped based on the visual interpretation of seven different LiDAR topographic derivatives computed from the 1 x 1 m DTM available from March 2012. The geomorphological historical landslide inventory of the Slani Potok gully was created, consisting of 181 landslide phenomena. Landslides cover 69 % of the area (0.33 km2) of the Slani Potok gully. The size of the smallest landslide is 65 m2, and of the largest is 10,563 m2. Landslides are very small to moderate-small, shallow to moderate-shallow, and mainly successive in distribution. Most of the landslides initiate along the margins of the gully channel walls, and extend to the gully channel bottom. Such a large number of identified landslides, as well as their specific spatial arrangement within the gully, indicates that sliding processes predominantly affect the morphologic development of the Slani Potok gully, and that the soil erosion is the secondary process in the study area.

Mine landslide susceptibility assessment using IVM, ANN and SVM models considering the contribution of affecting factors.

The fragile ecological environment near mines provide advantageous conditions for the development of landslides. Mine landslide susceptibility mapping is of great importance for mine geo-environment control and restoration planning. In this paper, a total of 493 landslides in Shangli County, China were collected through historical landslide inventory. 16 spectral, geomorphic and hydrological predictive factors, mainly derived from Landsat 8 imagery and Global Digital Elevation Model (ASTER GDEM), were prepared initially for landslide susceptibility assessment. Predictive capability of these factors was evaluated by using the value of variance inflation factor and information gain ratio. Three models, namely artificial neural network (ANN), support vector machine (SVM) and information value model (IVM), were applied to assess the mine landslide sensitivity. The receiver operating characteristic curve (ROC) and rank probability score were used to validate and compare the comprehensive predictive capabilities of three models involving uncertainty. Results showed that ANN model achieved higher prediction capability, proving its advantage of solve nonlinear and complex problems. Comparing the estimated landslide susceptibility map with the ground-truth one, the high-prone area tends to be located in the middle area with multiple fault distributions and the steeply sloped hill.

Implications of Landslide Typology and Predisposing Factor Combinations for Probabilistic Landslide Susceptibility Models: A Case Study in Lajedo Parish (Flores Island, Azores—Portugal)

The main objective of this study is to better understand and quantify the consequences for landslide susceptibility assessment caused by (i) the discrimination (or not) of landslide typology and (ii) the use of different predisposing factor combinations. The study area for this research was Lajedo parish (Flores Island, Azores—Portugal). For the landslide susceptibility modeling, 12 predisposing factors and a historical landslide inventory with a total of 474 individual landslide rupture areas were used as inputs, and the Information Value method was then applied. It was concluded that susceptibility models developed specifically for each landslide typology achieve better results when compared to the model developed for the total inventory, which suffers from a bias caused by the strong spatial abundance of one landslide typology. A total of 4095 susceptibility models were tested for each typology, and the best models were selected according to their goodness of fit. The best model for both falls and slides has seven predisposing factors, some of which do not correspond to the factors that have the best individual discriminatory capabilities. The number of expected and observed unique terrain conditions for each model allowed us to conclude that with the successive addition of predisposing factors, there is an inability of the territory to generate new observed unique terrain conditions. This consequence was directly related to the inability to increase the goodness of fit of the computed models. For each landslide typology, the predictive capacity of the best susceptibility model was assessed by computing the Prediction Rate Curves and the Area Under the Curve.

Landslide characteristics and spatial distribution in the Rwenzori Mountains, Uganda

In many landslide-prone regions, data on landslide characteristics remain poor or inexistent. This is also the case for the Rwenzori Mountains, located on the border of Uganda and the DR Congo. There, landslides frequently occur and cause fatalities and substantial damage to private property and infrastructure. In this paper, we present the results of a field inventory performed in three representative study areas covering 114 km2. A total of 371 landslides were mapped and analyzed for their geomorphological characteristics and their spatial distribution. The average landslide areas varied from less than 0.3 ha in the gneiss-dominated highlands to >1 ha in the rift alluvium of the lowlands. Large landslides (>1.5 ha) are well represented while smaller landslides (<1.5 ha) are underrepresented. The degrees of completeness of the field inventories are comparable to those of similar historical landslide inventories. The diversity of potential mass movements in the Rwenzori is large and depends on the dominant lithological and topographic conditions. A dominance of shallow translational soil slides in gneiss and of deep rotational soil slides in the rift alluvium is observed. Slope angle is the main controlling topographic factor for landslides with the highest landslide concentrations for slope angles above 25–30° in the highlands and 10–15° in the lowlands. The undercutting of slopes by rivers and excavations for construction are important preparatory factors. Rainfall-triggered landslides are the most common in the area, however in the zones of influence of the last two major earthquakes (1966: Mw = 6.6 and 1994: Mw = 6.2), 12 co-seismic landslides were also observed.

Regional coseismic landslide hazard assessment without historical landslide inventories: A new approach

Currently, regional coseismic landslide hazard analyses require comprehensive historical landslide inventories as well as detailed geotechnical data. Consequently, such analyses have not been possible where these data are not available. A new approach is proposed herein to assess coseismic landslide hazard at regional scale for specific earthquake scenarios in areas without historical landslide inventories. The proposed model employs fuzzy logic and geographic information systems to establish relationships between causative factors and coseismic slope failures in regions with well-documented and substantially complete coseismic landslide inventories. These relationships are then utilized to estimate the relative probability of landslide occurrence in regions with neither historical landslide inventories nor detailed geotechnical data. Statistical analyses of inventories from the 1994 Northridge and 2008 Wenchuan earthquakes reveal that shaking intensity, topography, and distance from active faults and streams are the main controls on the spatial distribution of coseismic landslides. Average fuzzy memberships for each factor are developed and aggregated to model the relative coseismic landslide hazard for both earthquakes. The predictive capabilities of the models are assessed and show good-to-excellent model performance for both events. These memberships are then applied to the 1999 Chi-Chi earthquake, using only a digital elevation model, active fault map, and isoseismal data, replicating prediction of a future event in a region lacking historic inventories and/or geotechnical data. This similarly results in excellent model performance, demonstrating the model's predictive potential and confirming it can be meaningfully applied in regions where previous methods could not. For such regions, this method may enable a greater ability to analyze coseismic landslide hazard from specific earthquake scenarios, allowing for mitigation measures and emergency response plans to be better informed of earthquake-related hazards.

Landslide hazard and risk assessment using semi-automatically created landslide inventories

Landslide inventories prepared manually from remote sensing data or through field surveys have shown to be useful for preparation of landslide susceptibility and hazard maps. Recent literatures show several studies have been carried out to prepare landslide inventories from satellite data by automatic methods. However, almost no attempt has been made to validate the effect of such inventories on landslide hazard and risk assessment. In this paper we have shown how landslide inventories prepared by semi-automatic methods from post-event satellite images can be used in the assessment of landslide susceptibility, hazard and risk in the High Himalayan terrain in India. A susceptibility map was made using the weights-of-evidence method, wherein weights were derived using the semi-automatically prepared historical landslide inventories combined with a series of pre-disposing factor maps. Seven evidence layers were used for the calculation of weights, selected in such a way that the majority could be derived from satellite data. Validation was done using the test data created through a temporal subsetting of the inventories. Temporal probability was calculated through Gumbel frequency distribution analysis using daily rainfall data of a 13year period for which landslide inventories were prepared from satellite data. Spatial probability was determined by calculating landslide density for the inventories per susceptibility class that represent a given return period. Elements at risk, such as buildings and roads, were interpreted from a high resolution Cartosat-1 (2.5m) image. Absolute vulnerability of the buildings and roads were multiplied with landslide spatial probability to derive the total loss for different return-period scenarios and shown in a risk curve. This study has shown that the inventories prepared semi-automatically can be used for landslide hazard and risk assessment in a data-poor environment.

How can statistical models help to determine driving factors of landslides?

Landslides are a hazard for humans and artificial structures. From an ecological point of view, they represent an important ecosystem disturbance, especially in tropical montane forests. Here, shallow translational landslides are a frequent natural phenomenon and one local determinant of high levels of biodiversity. In this paper, we apply weighted ensembles of advanced phenomenological models from statistics and machine learning to analyze the driving factors of natural landslides in a tropical montane forest in South Ecuador. We exclusively interpret terrain attributes, derived from a digital elevation model, as proxies to several driving factors of landslides and use them as predictors in our models which are trained on a set of five historical landslide inventories. We check the model generality by transferring them in time and use three common performance criteria (i.e. AUC, explained deviance and slope of model calibration curve) to, on the one hand, compare several state-of-the-art model approaches and on the other hand, to create weighted model ensembles. Our results suggest that it is important to consider more than one single performance criterion.Approaching our main question, we compare responses of weighted model ensembles that were trained on distinct functional units of landslides (i.e. initiation, transport and deposition zones). This way, we are able to show that it is quite possible to deduce driving factors of landslides, if the consistency between the training data and the processes is maintained. Opening the ‘black box’ of statistical models by interpreting univariate model response curves and relative importance of single predictors regarding their plausibility, we provide a means to verify this consistency.With the exception of classification tree analysis, all techniques performed comparably well in our case study while being outperformed by weighted model ensembles. Univariate response curves of models trained on distinct functional units of landslides exposed different shapes following our expectations. Our results indicate the occurrence of landslides to be mainly controlled by factors related to the general position along a slope (i.e. ridge, open slope or valley) while landslide initiation seems to be favored by small scale convexities on otherwise plain open slopes.

Object-oriented analysis of multi-temporal panchromatic images for creation of historical landslide inventories

Object-oriented analysis (OOA) has been demonstrated to produce more accurate results than pixel-based image processing. Studies carried out by previous researchers have shown how landslide inventories can be prepared from multispectral satellite images using OOA. However, panchromatic images are frequently the only data available after a landslide event. Furthermore, preparation of historical inventories relies on the analysis of satellite images and aerial photographs acquired over past few decades that are also mostly only available in black and white. In such cases the methodology developed using multispectral data cannot be used directly due to limited spectral information, in particular in near-infrared bands. In this paper we present a new methodology that addresses some of these issues. Using high resolution panchromatic images from Cartosat-1 (2.5 m) and IRS-1D (5.8 m), and a 10 m gridded DTM extracted from Cartosat-1, we developed a new approach which uses change detection techniques and a global contextual criteria in an object-based environment to detect and classify landslides into five different types. Continuous time series images from 1998 to 2006 were used to prepare annual landslide inventories in a highly rugged Himalayan terrain. The maximum and minimum detection percentages achieved for all landslides are 96.7% and 71.5%, respectively, with corresponding quality percentages of 88.1% and 55.3%, respectively. However, the lack of spectral information proved to be a hurdle resulting in a high branching factor that indicates that further work is required to eliminate false positives. Nevertheless, the method was able to create much needed historical landslide inventories, which are critical for landslide hazard and risk assessment studies.

Application and verification of a fractal approach to landslide susceptibility mapping

The existing methods for landslide susceptibility mapping, whether statistic method or physics-based method, require many data such as lithology, topography, soil properties, land use and so on. However, in many regions of the world, the abundance of data is not available, but the need for landslide susceptibility maps is great. For these regions, how should reliable susceptibility maps be produced from the limited data? We addressed the issue of the problem and developed a new landslide susceptibility analysis technique using historical landslide inventories and fractal statistics on a GIS platform. The aim of this article is to apply and verify the use of this new technique to landslide susceptibility mapping in the Zhejiang Province (101,800 km2 in area), China.

Characteristics of the size distribution of recent and historical landslides in a populated hilly region

Despite the availability of studies on the frequency density of landslide areas in mountainous regions, frequency–area distributions of historical landslide inventories in populated hilly regions are absent. This study revealed that the frequency–area distribution derived from a detailed landslide inventory of the Flemish Ardennes (Belgium) is significantly different from distributions usually obtained in mountainous areas where landslides are triggered by large-scale natural causal factors such as rainfall, earthquakes or rapid snowmelt. Instead, the landslide inventory consists of the superposition of two populations, i.e. (i) small (< 1–2 · 10 − 2 km 2), shallow complex earth slides that are at most 30 yr old, and (ii) large (> 1–2 · 10 − 2 km 2), deep-seated landslides that are older than 100 yr. Both subpopulations are best represented by a negative power–law relation with exponents of − 0.58 and − 2.31 respectively. This study focused on the negative power–law relation obtained for recent, small landslides, and contributes to the understanding of frequency distributions of landslide areas by presenting a conceptual model explaining this negative power–law relation for small landslides in populated hilly regions. According to the model hilly regions can be relatively stable under the present-day environmental conditions, and landslides are mainly triggered by human activities that have only a local impact on slope stability. Therefore, landslides caused by anthropogenic triggers are limited in size, and the number of landslides decreases with landslide area. The frequency density of landslide areas for old landslides is similar to those obtained for historical inventories compiled in mountainous areas, as apart from the negative power–law relation with exponent − 2.31 for large landslides, a positive power–law relation followed by a rollover is observed for smaller landslides. However, when analysing the old landslides together with the more recent ones, the present-day higher temporal frequency of small landslides compared to large landslides, obscures the positive power–law relation and rollover.