Landslide Susceptibility Prediction Research Articles

Game-theoretic optimization of landslide susceptibility mapping: a comparative study between Bayesian-optimized basic neural network and new generation neural network models.

Landslide susceptibility mapping is essential for reducing the risk of landslides and ensuring the safety of people and infrastructure in landslide-prone areas. However, little research has been done on the development of well-optimized Elman neural networks (ENN), deep neural networks (DNN), and artificial neural networks (ANN) for robust landslide susceptibility mapping (LSM). Additionally, there is a research gap regarding the use of Bayesian optimization and the derivation of SHapley Additive exPlanations (SHAP) values from optimized models. Therefore, this study aims to optimize DNN, ENN, and ANN models using Bayesian optimization for landslide susceptibility mapping and derive SHAP values from these optimized models. The LSM models have been validated using the receiver operating characteristics curve, confusion matrix, and other twelve error matrices. The study used six machine learning-based feature selection techniques to identify the most important variables for predicting landslide susceptibility. The decision tree, random forest, and bagging feature selection models showed that slope, elevation, DFR, annual rainfall, LD, DD, RD, and LULC are influential variables, while geology and soil texture have less influence. The DNN model outperformed the other two models, covering 7839.54 km2 under the very low landslide susceptibility zone and 3613.44 km2 under the very high landslide susceptibility zone. The DNN model is better suited for generating landslide susceptibility maps, as it can classify areas with higher accuracy. The model identified several key factors that contribute to the initiation of landslides, including high elevation, built-up and agricultural land use, less vegetation, aspect (north and northwest), soil depth less than 140 cm, high rainfall, high lineament density, and a low distance from roads. The study's findings can help stakeholders make informed decisions to reduce the risk of landslides and ensure the safety of people and infrastructure in landslide-prone areas.

Modelling antecedent soil hydrological conditions to improve the prediction of landslide susceptibility in typhoon-prone regions

AbstractMost regional landslide susceptibility models do not consider the evolving soil hydrological conditions leading up to a multiple occurrence regional landslide event. This results in inaccurate predictions due to the non-linear behaviour of the terrain. To address this, we have developed a simple and efficient model that incorporates the mid-term evolution of soil hydrological conditions. The model combines a water balance model and a geotechnical model based on infinite slope theory. The analysis of 561 high-intensity rainfall events in a typhoon-prone region of the Philippines revealed that the percolation of water during the 5-month wet season is crucial in determining landslide susceptibility. Consequently, high-intensity rainfall events at the start of the wet season are less likely to trigger landslides, while later events are more hazardous. We analysed the change in landslide susceptibility during the 2018 rainy season by comparing the probability of failure (PoF) before and after three high-intensity rainfall events (July, August and September). Only the event in September caused a significant increase in the probability of failure (PoF). The model showed an accuracy of 0.63, with stable cells better represented than unstable cells. The antecedent hydrological conditions on the lower soil layers are responsible for changes in landslide susceptibility. Our findings support the hypothesis that new approaches to developing hydro-meteorological thresholds for landslide early warning systems should be evaluated, especially in regions with strong seasonality.

Uncertainties of landslide susceptibility prediction: influences of different study area scales and mapping unit scales

This study aims to investigate the effects of different mapping unit scales and study area scales on the uncertainty rules of landslide susceptibility prediction (LSP). To illustrate various study area scales, Ganzhou City in China, its eastern region (Ganzhou East), and Ruijin County in Ganzhou East were chosen. Different mapping unit scales are represented by grid units with spatial resolution of 30 and 60 m, as well as slope units that were extracted by multi-scale segmentation method. The 3855 landslide locations and 21 typical environmental factors in Ganzhou City are first determined to create spatial datasets with input-outputs. Then, landslide susceptibility maps (LSMs) of Ganzhou City, Ganzhou East and Ruijin County are produced using a support vector machine (SVM) and random forest (RF), respectively. The LSMs of the above three regions are then extracted by mask from the LSM of Ganzhou City, along with the LSMs of Ruijin County from Ganzhou East. Additionally, LSMs of Ruijin at various mapping unit scales are generated in accordance. Accuracy and landslide susceptibility indexes (LSIs) distribution are used to express LSP uncertainties. The LSP uncertainties under grid units significantly decrease as study area scales decrease from Ganzhou City, Ganzhou East to Ruijin County, whereas those under slope units are less affected by study area scales. Of course, attentions should also be paid to the broader representativeness of large study areas. The LSP accuracy of slope units increases by about 6%–10% compared with those under grid units with 30 m and 60 m resolution in the same study area's scale. The significance of environmental factors exhibits an averaging trend as study area scale increases from small to large. The importance of environmental factors varies greatly with the 60 m grid unit, but it tends to be consistent to some extent in the 30 m grid unit and the slope unit.Graphic abstract

Landslide susceptibility prediction and mapping in Loess Plateau based on different machine learning algorithms by hybrid factors screening: Case study of Xunyi County, Shaanxi Province, China

Machine learning models are widely used for landslide susceptibility prediction (LSP). To better develop a landslide susceptibility prediction model for the Loess Plateau, this study compared different machine-learning models (Logistic regression (LR), support vector machines (SVM), random forests (RF), Naïve Bayes (NB), and multilayer perceptron (MLP)). Using Xunyi County, China as the study area, 1096 landslides were collected and divided into a training set (70%) and a test set (30%). Significant influencing factors were selected from 20 landslide factors using GeoDetector. The relationships between these factors and landslides were determined by normalized frequency ratios. Five trained machine learning models calculated landslide susceptibility indices and generated landslide susceptibility maps. Additionally, the accuracy (ACC), precision (PRE), recall (REC), F1 score, AUC values, LSI distribution characteristics, and overlay analysis of historical landslides with LSM were used to evaluate the performance of the models. The results show that the LSP results of the five models are generally reasonable. The results further indicated that the use of GeoDetector was able to filter out the main factors of eolian landslides, improving the predictive performance of the model and obtaining better LSMs. The RF model had the best overall performance, followed by the SVM, and the MLP model created LSM high susceptibility zones that could contain the highest number of historical landslides. The q-values of factor contributions indicate that elevation, NDVI, rainfall, POI kernel density, and RDLS are the main factors controlling landslides, which need to be paid attention to for landslide control in the Loess Plateau region. The results of this study can provide a reference for the follow-up study of landslide prevention and control, especially landslide susceptibility modeling in the Loess Plateau.

Landslide Susceptibility Zoning: Integrating Multiple Intelligent Models with SHAP Analysis

In this study, we aim to delineate landslide susceptibility zones within Dien Bien province, Vietnam, leveraging the capabilities of various machine learning models including Light Gradient Boosting Machine (LGBM), K-Nearest Neighbors (KNN), and Gradient Boosting (GB). Harnessing a dataset comprising 665 data points and encompassing 14 influential factors such as slope, aspect, curvature, elevation, geological composition, Normalized Difference Vegetation Index (NDVI), and proximity to geological features like faults, rivers, and roads, a comprehensive database for landslide modeling was constructed. The analysis entailed rigorous evaluation and comparison of model accuracy employing established statistical metrics, notably Receiver Operating Characteristic (ROC) curves and Area Under the Curve (AUC). The findings underscore the efficacy of the Light Gradient Boosting Machine model, exhibiting superior performance with an AUC score of 0.85, surpassing both the Gradient Boosting model (AUC = 0.81) and the K-Nearest Neighbors model (AUC = 0.79). Notably, the Light Gradient Boosting Machine model emerges as a promising tool for precise landslide prediction within the study area, offering significant potential for the creation of a robust landslide susceptibility map. The resulting spatial forecast map for Dien Bien province holds considerable utility for informing land use planning initiatives aimed at mitigating the impact of landslide disasters in the region. Moreover, the application of SHAP (Shapley Additive explanation) values to quantify the contribution of each factor to landslide susceptibility prediction, offering novel insights into model interpretation and feature importance. The resulting spatial forecast map holds significant implications for land use planning and disaster mitigation efforts in Dien Bien province, showcasing the potential of advanced machine learning techniques in enhancing landslide risk management strategies.

Landslide Risks to Bridges in Valleys in North Carolina

This research delves into the intricate dynamics of landslides, emphasizing their consequences on transportation infrastructure, specifically highways and roadway bridges in North Carolina. Based on a prior investigation of bridges in Puerto Rico after Hurricane Maria, we found that bridges above water and situated in valleys can be exposed to both landslide and flooding risks. These bridges faced heightened vulnerability to combined landslides and flooding events due to their low depth on the water surface and the potential for raised flood heights due to upstream landslides. Leveraging a dataset spanning more than a century and inclusive of landslide and bridge information, we employed logistic regression (LR) and random forest (RF) models to predict landslide susceptibility in North Carolina. The study considered conditioning factors such as elevation, aspect, slope, rainfall, distance to faults, and distance to rivers, yielding LR and RF models with accuracy rates of 76.3% and 82.7%, respectively. To establish that a bridge’s location is at the bottom of a valley, data including landform, slope, and elevation difference near the bridge location were combined to delineate a bridge in a valley. The difference between bridge height and the lowest river elevation is established as an assumed flooding potential (AFP), which is then used to quantify the flooding risk. Compared to traditional flood risk values, the AFP, reported in elevation differences, is more straightforward and helps bridge engineers visualize the flood risk to a bridge. Specifically, a bridge (NCDOT ID: 740002) is found susceptible to both landslide (92%) and flooding (AFT of 6.61 m) risks and has been validated by field investigation, which is currently being retrofitted by North Carolina DOT with slope reinforcements (soil nailing and grouting). This paper is the first report evaluating the multi-hazard issue of bridges in valleys. The resulting high-fidelity risk map for North Carolina can help bridge engineers in proactive maintenance planning. Future endeavors will extend the analysis to incorporate actual flooding risk susceptibility analysis, thus enhancing our understanding of multi-hazard impacts and guiding resilient mitigation strategies for transportation infrastructure.

Application of machine learning in the assessment of landslide susceptibility: A case study of mountainous eastern Mediterranean region, Syria

Landslide is a considerable geomorphological risk in terrain systems worldwide. Advanced techniques present a unique tool for predicting landslide susceptibility with unbiased and precise outputs. However, the application of this tool to analyze landslide susceptibility in the eastern Mediterranean landscape is still not sufficiently understood. This study aimed to assess the implementation of three machine learning (ML) algorithms, i.e., support vector machine (SVM), random forest (RF) and extreme gradient boost (XGBoost), in landslide susceptibility mapping in the mountainous area of western Syria. In this regard, 200 landslide events were inventoried from historical data, aerial images and conducted fieldworks. Sixteen triggering factors were selected according to the literature and geographical features (Monsoon period). The receiver operating characteristic (ROC) outcomes revealed that the RF achieved better performance with an area under curve (AUC) of 0.96, pursued by XGBoost and SVM with AUC of 0.94 and 0.90, respectively. This assessment presents an essential understanding of the effective implementation of ML in landslide assessment in the mountainous region of the eastern Mediterranean. We emphasized, hence, that the RF algorithm has the most robust performance of landslide susceptibility prediction in the western Syria. Moreover, the outputs of this study will provide local decision-makers with insights to produce regional management strategies for the landslide, especially after the Syrian war phase.

Landslide susceptibility prediction modelling based on semi‐supervised XGBoost model

In the process of landslide susceptibility prediction (LSP) modelling, there are some problems in the model dataset relating to landslide and non‐landslide samples, such as landslide sample errors, subjective randomness and low accuracy of non‐landslide sample selection. In order to solve the above problems, a semi‐supervised machine learning model for LSP is innovatively proposed. Firstly, Yanchang County of Shanxi Province, China, is taken as the study area. Secondly, the frequency ratio values of 12 environmental factors (elevation, slope, aspect, etc.) and the randomly selected twice non‐landslides are used to form the initial model datasets. Thirdly, an extreme gradient boosting (XGBoost) model is adopted for training and testing the initial datasets, so as to produce initial landslide susceptibility maps (LSMs) which are divided into very low, low, moderate, high and very high susceptibility levels. Next, the landslide samples in initial LSMs with very low and low susceptibility levels are excluded to improve the accuracy of landslide samples, and the unlabelled twice non‐landslide samples in initial LSMs with low and very low susceptibility levels are randomly selected to ensure the accuracy of non‐landslide samples. These new obtained landslide and non‐landslide samples are reimported into XGBoost model to construct the semi‐supervised XGBoost (SSXGBoost) model. Finally, accuracy, kappa coefficient and statistical indexes of susceptibility indexes are adopted to assess the LSP performance of XGBoost and SSXGBoost models. Results show that SSXGBoost model has remarkably better LSP performance than that of XGBoost model. Conclusively, the proposed SSXGBoost model effectively overcomes the problems that the accuracy of landslide samples needs to be further improved and that non‐landslide samples are difficult to select accurately.

SPATIALLY AWARE LANDSLIDE SUSCEPTIBILITY PREDICTION USING A GEOGRAPHICAL RANDOM FOREST APPROACH

Abstract. Landslide susceptibility prediction practices have been increasingly reliant on non-geographically-oriented (i.e., aspatial) machine learning algorithms. While these approaches have exhibited increasing success, they have often faced criticism for their limited consideration of spatial autocorrelations and local variations across geographical space, thereby neglecting the concept of spatial non-stationarity. To fulfill the research gap, this work applies a geographical random forest (GRF) approach, contrasting it with the conventional random forest (RF) algorithm. To this end, the study area, encompassing the Lake Sapanca Basin and its surroundings, was subdivided into 4,452 slope-based mapping units. The effectiveness of both predictive models was then measured by using overall accuracy (OA) and area under the curve (AUC). The results revealed that the GRF (OA = 80.82% and AUC = 85.22%) outperformed the RF algorithm (OA = 75.34% and AUC = 82.50%) by approximately 5% in OA, and demonstrated a 3% improvement in AUC score. The Wilcoxon signed-rank test confirmed significant differences (95% level) between the predictions of both models. The slope parameter emerged as the globally most influential factor, but local interpretations disclosed notable variations in the importance of causative factors contingent upon location. For instance, the curvature parameter was the most important geospatial covariate in around one-third (34.23%) of the slope units, mostly concentrated in the northernmost zones of the study area. On the other hand, elevation was the most important factor for 14.67% of the slope units primarily located in the southern region.

Uncertainties in landslide susceptibility prediction: Influence rule of different levels of errors in landslide spatial position

The accuracy of landslide susceptibility prediction (LSP) mainly depends on the precision of the landslide spatial position. However, the spatial position error of landslide survey is inevitable, resulting in considerable uncertainties in LSP modeling. To overcome this drawback, this study explores the influence of positional errors of landslide spatial position on LSP uncertainties, and then innovatively proposes a semi-supervised machine learning model to reduce the landslide spatial position error. This paper collected 16 environmental factors and 337 landslides with accurate spatial positions taking Shangyou County of China as an example. The 30–110 m error-based multilayer perceptron (MLP) and random forest (RF) models for LSP are established by randomly offsetting the original landslide by 30, 50, 70, 90 and 110 m. The LSP uncertainties are analyzed by the LSP accuracy and distribution characteristics. Finally, a semi-supervised model is proposed to relieve the LSP uncertainties. Results show that: (1) The LSP accuracies of error-based RF/MLP models decrease with the increase of landslide position errors, and are lower than those of original data-based models; (2) 70 m error-based models can still reflect the overall distribution characteristics of landslide susceptibility indices, thus original landslides with certain position errors are acceptable for LSP; (3) Semi-supervised machine learning model can efficiently reduce the landslide position errors and thus improve the LSP accuracies.

Offshore application of landslide susceptibility mapping using gradient-boosted decision trees: a Gulf of Mexico case study

AbstractAmong natural hazards occurring offshore, submarine landslides pose a significant risk to offshore infrastructure installations attached to the seafloor. With the offshore being important for current and future energy production, there is a need to anticipate where future landslide events are likely to occur to support planning and development projects. Using the northern Gulf of Mexico (GoM) as a case study, this paper performs Landslide Susceptibility Mapping (LSM) using a gradient-boosted decision tree (GBDT) model to characterize the spatial patterns of submarine landslide probability over the United States Exclusive Economic Zone (EEZ) where water depths are greater than 120 m. With known spatial extents of historic submarine landslides and a Geographic Information System (GIS) database of known topographical, geomorphological, geological, and geochemical factors, the resulting model was capable of accurately forecasting potential locations of sediment instability. Results of a permutation modelling approach indicated that LSM accuracy is sensitive to the number of unique training locations with model accuracy becoming more stable as the number of training regions was increased. The influence that each input feature had on predicting landslide susceptibility was evaluated using the SHapely Additive exPlanations (SHAP) feature attribution method. Areas of high and very high susceptibility were associated with steep terrain including salt basins and escarpments. This case study serves as an initial assessment of the machine learning (ML) capabilities for producing accurate submarine landslide susceptibility maps given the current state of available natural hazard-related datasets and conveys both successes and limitations.

Hybrid Integration of Bagging and Decision Tree Algorithms for Landslide Susceptibility Mapping

Landslides represent a significant global natural hazard, threatening human settlements and the natural environment. The primary objective of the study was to develop a landslide susceptibility modeling approach that enhances prediction accuracy and informs land-use planning decisions. The study utilized a hybrid ensemble-based methodology to improve prediction accuracy and effectively capture the complexity of landslide susceptibility patterns. This approach harnessed the power of ensemble models, employing a bagging algorithm with base learners, including the reduced error pruning decision tree (REPTree) and functional tree (FT) models. Ensemble models are particularly valuable because they combine the strengths of multiple models, enhancing the overall performance and robustness of the landslide susceptibility prediction. The study focused on Yanchuan County, situated within the hilly and gully region of China’s Loess Plateau, known for its susceptibility to landslides, using sixteen critical landslide conditioning factors, encompassing topographic, environmental, and geospatial variables, namely elevation, slope, aspect, proximity to rivers and roads, rainfall, the normalized difference vegetation index, soil composition, land use, and more. Model performances were evaluated and verified using a range of metrics, including receiver operating characteristic (ROC) curves, trade-off statistical metrics, and chi-square analysis. The results demonstrated the superiority of the integrated models, particularly the bagging FT (BFT) model, in accurately predicting landslide susceptibility, as evidenced by its high area under the curve area (AUC) value (0.895), compared to the other models. The model excelled in both positive predictive rate (0.847) and negative predictive rate (0.886), indicating its efficacy in identifying landslide and non-landslide areas and also in the F-score metric with a value of 0.869. The study contributes to the field of landslide risk assessment, offering a significant investigation tool for managing and mitigating landslide hazards in Yanchuan County and similar regions worldwide.

Modeling of landslides susceptibility prediction using deep belief networks with optimized learning rate control

To overcome critical issues in landslide susceptibility modeling, a multifactor landslide susceptibility prediction model based on deep belief networks (DBN) with optimized learning rate control (LRC-DBN) was introduced in this study. The LRC-DBN model was applied to predict landslide susceptibility in Zhidan County, Shaanxi and its performance was compared against random forest, support vector machine and logistic regression models. The results show that the LRC-DBN model achieves a maximum AUC value of 0.941, which demonstrating higher predictive performance. The interpretability of LRC-DBN reveals that the development of loess landslides is more likely in areas with elevations ranging from 894.7 m to 998.5 m, slopes ranging from 49.6° to 64°, terrain undulations spanning 31.7 m to 91.0 m, rock type T3y and terrain humidity ranging from 2.4 to 3.9. The result provides invaluable insights for local landslide prevention endeavors and can assist decision-making in urban planning.

Comparison of different open‐source Digital Elevation Models for landslide susceptibility mapping

AbstractIn this study, the application of open‐source digital elevation model (DEM) is explored for regional landslide susceptibility mapping (LSM), and the potential impact of different DEM choices on the mapping accuracy is also examined. With the advancements in remote sensing technology, an increasing number of global open‐source DEMs have been available, with improvement in the accuracy. However, the latest released data are rarely evaluated in LSM research. In this paper, DEM‐based factors, including elevation, aspect, slope, plan curvature and profile curvature, were generated from seven open‐source DEMs, including Advanced Spaceborne Thermal Emission and Reflection (ASTER) V2, ASTERV3, ALOS World 3D‐30 m (AW3D30), Copernicus DEM 30 m (COP) Forest and Buildings removed Copernicus DEM (FABDEM), NASADEM, and Shuttle Radar Topography Mission (SRTM). DEM‐based factors were coupled with the distance to road, distance to river, land use, lithology, rain and normalized difference vegetation index (NDVI). The significant difference between DEMs is determined by comparing the area proportion. Slope, plane curvature and profile curvature are found to have a maximum difference of 15%–20%. Then, K‐Nearest Neighbours (KNN) and Random Forest (RF) were used to predict landslide susceptibility with two sampling methods, namely, 70% for training and 30% for testing (S1); 67% for training and 33% for testing (S2). For KNN with S1, the prediction rate is range from 0.8299 to 0.8701, with a difference of 0.0402. The difference of prediction rate is decreased to 0.0207 for S2 and 0.0258 for RF. COP has the highest prediction rate of 0.8701, 0.9254 and 0.9461 for KNN with S1 and RF with S1 and S2, respectively. ASTERV2 is the worst with prediction rate of 0.8897 and 0.8996 for KNN with S2 and RF with S1, respectively. The research result provides valuable insights for the selection of open‐source DEMs in future LSM.

Fast hybrid methods for modeling landslide susceptibility in Ardal County

Recently, machine learning models have received huge attention for environmental risk modeling. One of these applications is landslide susceptibility mapping which is a necessary primary step for dealing with the landslide risk in prone areas. In this study, a conventional machine learning model called multi-layer perceptron (MLP) neural network is built upon advanced optimization algorithms to achieve a firm prediction of landslide susceptibility in Ardal County, West of Iran. The used geospatial dataset consists of fourteen conditioning factors and 170 landslide events. The used optimizers are electromagnetic field optimization (EFO), symbiotic organisms search (SOS), shuffled complex evolution (SCE), and electrostatic discharge algorithm (ESDA) that contribute to tuning MLP’s internal parameters. The competency of the models is evaluated using several statistical methods to provide a comparison among them. It was discovered that the EFO-MLP and SCE-MLP enjoy much quicker training than SOS-MLP and ESDA-MLP. Further, relying on both accuracy and time criteria, the EFO-MLP was found to be the most efficient model (time = 1161 s, AUC = 0.879, MSE = 0.153, and R = 0.657). Hence, the landslide susceptibility map of this model is recommended to be used by authorities to provide real-world protective measures within Ardal County. For helping this, a random forest-based model showed that Elevation, Lithology, and Land Use are the most important factors within the studied area. Lastly, the solution discovered in this study is converted into an equation for convenient landslide susceptibility prediction.

A novel landslide susceptibility prediction framework based on contrastive loss

ABSTRACT Recently, the positive unlabeled (PU) learning algorithms have proven highly effective in generating accurate landslide susceptibility maps. The algorithms categorize samples exclusively into positive samples (landslides) and unlabeled samples for training, eliminating random or subjective selection of non-landslide samples. However, existing PU learning algorithms face limitations in capturing correct negative samples in multi-genesis landslide areas, leading to lower prediction accuracy. To address this issue, the PU-pullbaggingDT algorithm was proposed in this study. This integrated method combines the strengths of two techniques: the superior ranking performance of the PU-baggingDT algorithm and the low bias of the contrastive learning approach. The contrastive learning introduces a new contrastive loss of PU learning (PUpull loss), which pulls the distance of similar landslide samples in the projection space closer based on the validation accuracy threshold, without the need for data augmentation and class probability. The PUpull loss relaxes the tightness toward unlabeled samples, reducing the impact of incorrectly defined non-landslide samples on prediction results in multi-genesis landslide areas. The proposed algorithm outperforms existing PU-learning and machine learning methods (support vector machine, decision tree, logistic regression, AdaBoost, and XGBoost) with random selection of negative samples for predicting landslide susceptibility in China’s Zigui County, as demonstrated by comprehensive evaluation metrics. The landslide susceptibility mapping utilizes equal interval division and ranking, achieving approximately a 90% landslide percentage in areas with very high and high susceptibility in Zigui County. This demonstrates the capability of the proposed algorithm to effectively predict landslide susceptibility in complex geological settings.

Analysis of landslide susceptibility prediction accuracy with an event-based inventory: The 6 February 2023 Turkiye earthquakes

Landslide susceptibility assessment is a complex challenge explored by various scientists, but not fully resolved. In this study, we produced the landslide susceptibility map of a large region covering 38,500 km2 area in South-East Turkiye severely affected by the 6 February 2023 Kahramanmaras Earthquakes (Mw 7.7 and Mw 7.6) using an inventory produced in previous years. We employed random forest regression with a total of nine geomorphological and environmental features and evaluated the results using the co-seismic inventory with 2611 landslides compiled here. Although high accuracy was obtained from pixel-based assessments of test data split from the learning set, the independent validation set of co-seismic landslides showed that attention needs to be paid unseen features such as rare lithological units. Given the significant damage caused by latent hazards with the Kahramanmaras earthquakes, producing reliable inventories and precise landslide susceptibility maps is crucial for risk reduction and minimizing damages.

The Prediction of Cross-Regional Landslide Susceptibility Based on Pixel Transfer Learning

Considering the great time and labor consumption involved in conventional hazard assessment methods in compiling landslide inventory, the construction of a transferable landslide susceptibility prediction model is crucial. This study employs UAV images as data sources to interpret the typical alpine valley area of Beichuan County. Eight environmental factors including a digital elevation model (DEM) are extracted to establish a pixel-wise dataset, along with interpreted landslide data. Two landslide susceptibility models were built, each with a deep neural network (DNN) and a support vector machine (SVM) as the learner, and the DNN model was determined to have the best pre-training performance (accuracy = 88.6%, precision = 91.3%, recall = 94.8%, specificity = 87.8%, F1-score = 93.0%, and area under curve = 0.943), with higher parameters in comparison to the SVM model (accuracy = 77.1%, precision = 80.9%, recall = 87.8%, specificity = 73.9%, F1-score = 84.2%, and area under curve = 0.878). The susceptibility model of Beichuan County is then transferred to Mao County (which has no available dataset) to realize cross-regional landslide susceptibility prediction. The results suggest that the model predictions accomplish susceptibility zoning principles and that the DNN model can more precisely distinguish between high and very-high susceptibility areas in relation to the SVM model.

Assessment of resampling methods on performance of landslide susceptibility predictions using machine learning in Kendari City, Indonesia

Abstract Landslide susceptibility projections that rely on independent models produce biased results. This situation will worsen class balance if working with a small population. This study proposes a landslide susceptibility prediction model based on resampling, cross-validation, bootstrap, and random subsampling approaches, which is integrated with the machine learning model, generalized linear model, support vector machine, random forest, boosted regression trees, classification and regression tree, multivariate adaptive regression splines, mixture discriminate analysis, flexible discriminant analysis, maximum entropy, and maximum likelihood. This methodology was applied in Kendari City, an urban area which faced destructive erosion. Area under the ROC curve (AUC), true skill statistics (TSS), correlation coefficient (COR), normalized mutual information (NMI), and correct classification rate (CCR) were used to evaluate the predictive accuracy of the proposed model. The results show that the resampling algorithm improves the performance of the standalone model. Results also revealed that standalone models had better performance with the CV algorithm compared to the Bt and RS algorithms. The Bt-RF model excels in statistical measures (AUC = 0.97, TSS = 0.97, COR = 0.99, NMI = 0.50, and CCR = 0.93). Given the admirable performance of the proposed models in a moderate scale area, promising results can be expected from these models for other regions.

Ensemble stacking: a powerful tool for landslide susceptibility assessment – a case study in Anhua County, Hunan Province, China

Traditional landslide susceptibility assessment methods often rely on single models, which can be biased and less accurate. In this article, we introduce a two-tiered strategy to enhance landslide susceptibility predictions. Initially, we employ an ensemble stacking technique that combines the strengths of three machine learning classifiers. This combination leverages the support vector classifier (SVC) as the key meta-classifier to optimize and refine predictions. Subsequently, we integrate the extreme gradient boosting (XGB), random forest (RF) and gradient boosting decision tree (GBDT) models with SVC to create hybrid approaches. In this study, we evaluate and compare the effectiveness of six machine learning algorithms for predicting landslide susceptibility in Anhua County, Hunan Province, China. The results demonstrated that the stacking ensemble model outperforms traditional models. The XBG+SVC model achieves the highest AUC value (0.9468), which is followed by the GBDT+SVC (0.9316), RF+SVC (0.9162), XGB (0.9393), GBDT (0.9009), and RF (0.8693). These findings indicate that stacking machine learning approaches hold promise for landslide susceptibility mapping.