Landslide susceptibility mapping using Partial Decision Tree-Based hybrid artificial intelligence models

This research is focused on developing landslide susceptibility, risk and management zonation map in the Ilam dam basin, in the west of Iran. For this purpose, all existent landslide locations in the basin (50 landslides) were registered using GPS device and 70% of these points (35 landslides) were used for landslide susceptibility modeling and the rest (15 events) were used to evaluate the model. In order to prepare landslide susceptibility map, eight key factors were used for landslide occurrence such as distance to fault, distance to stream, distance to road, lithology, land use/cover, slope percent, aspect and precipitation derived from the spatial database in Arc GIS 9.3. A hybrid method of logistic regression and Analytic Hierarchy Process (AHP) were respectively used to determine the weight and rate of different factors and their classes. After applying rate to classes of parameters using the AHP method, landslide susceptibility map was prepared by means of logistic regression analysis tool in IDRISI software. The model accuracy was assessed by receiver operating characteristic (ROC) indicator and the pseudoR 2 and used as a basis for risk mapping. The landslide risk map was prepared using Varnes equation through combining three maps of susceptibility, vulnerability and the elements at risk. In order to provide the landslide management map, multi-criteria evaluation (MCE) method was used incorporating susceptibility and risk variables. The results suggest the logistic regression and AHP model has high accuracy (ROC= 81.2%, pseudo-R 2 = 0.32). We found that 39.84, 72.45, and 76.33 km 2 of Ilam dam basinare located in the high and very high classes of lands lide susceptibility, risk and management maps, respectively.

Review on landslide susceptibility mapping using support vector machines

Landslide susceptibility mapping is a helpful tool for assessment and management of landslides of an area. In this study, we have applied first time Forest by Penalizing Attributes (FPA) algorithm-based Machine Learning (ML) approach for mapping of landslide susceptibility at Muong Lay district (Vietnam). For this aim, 217 historical landslides locations were identified and analyzed for the development of FPA model and generation of susceptibility map. Nine landslide topographical and geo-environmental conditioning factors (curvature, geology/lithology, aspect, distance from faults, rivers and roads, weathering crust, slope, and deep division) were utilized to construct the training and validating datasets for landslide modeling. Different quantitative statistical indices including Area Under the Receiver Operating Characteristic (ROC) curve (AUC) were used to evaluate the performance of the model. The results indicate that the predictive capability of the FPA is very good for landslide susceptibility mapping on both training (AUC = 0.935) and validating (AUC = 0.882) datasets. Thus, the novel FPA based ML model can be utilized for the development of accurate landslide susceptibility map of the study area and this approach can also be applied in other landslide prone areas.

Landslide susceptibility maps are very important tools in the planning and management of landslide prone areas. Qualitative and quantitative methods each have their own advantages and dis-advantages in landslide susceptibility mapping. The aim of this study is to compare three models, i.e., frequency ratio (FR), Shannon’s entropy and analytic hierarchy process (AHP) by implementing them for the preparation of landslide susceptibility maps. Shimla, a district in Himachal Pradesh (H.P.), India was chosen for the study. A landslide inventory containing more than 1500 landslide events was prepared using previous literature, available historical data and a field survey. Out of the total number of landslide events, 30% data was used for training and 70% data was used for testing purpose. The frequency ratio, Shannon’s entropy and AHP models were implemented and three landslide susceptibility maps were prepared for the study area. The final landslide susceptibility maps were validated using a receiver operating characteristic (ROC) curve. The frequency ratio (FR) model yielded the highest accuracy, with 0.925 fitted ROC area, while the accuracy achieved by Shannon’s entropy model was 0.883. Analytic hierarchy process (AHP) yielded the lowest accuracy, with 0.732 fitted ROC area. The results of this study can be used by engineers and planners for better management and mitigation of landslides in the study area.

The Himalayan region and hilly areas face severe challenges due to landslide occurrences during the rainy seasons in India, and the study area, i.e., the Rudraprayag district, is no exception. However, the landslide related database and research are still inadequate in these landslide-prone areas. The main purpose of this study is: (1) to prepare the multi-temporal landslide inventory map using geospatial platforms in the data-scarce environment; (2) to evaluate the landslide susceptibility map using weights of evidence (WoE) method in the Geographical Information System (GIS) environment at the district level; and (3) to provide a comprehensive understanding of recent developments, gaps, and future directions related to landslide inventory, susceptibility mapping, and risk assessment in the Indian context. Firstly, 293 landslides polygon were manually digitized using the BHUVAN (Indian earth observation visualization) and Google Earth® from 2011 to 2013. Secondly, a total of 14 landslide causative factors viz. geology, geomorphology, soil type, soil depth, slope angle, slope aspect, relative relief, distance to faults, distance to thrusts, distance to lineaments, distance to streams, distance to roads, land use/cover, and altitude zones were selected based on the previous study. Then, the WoE method was applied to assign the weights for each class of causative factors to obtain a landslide susceptibility map. Afterward, the final landslide susceptibility map was divided into five susceptibility classes (very high, high, medium, low, and very low classes). Later, the validation of the landslide susceptibility map was checked against randomly selected landslides using IDRISI SELVA 17.0 software. Our study results show that medium to very high landslide susceptibilities had occurred in the non-forest areas, mainly scrubland, pastureland, and barren land. The results show that medium to very high landslide susceptibilities areas are in the upper catchment areas of the Mandakini river and adjacent to the National Highways (107 and 07). The results also show that landslide susceptibility is high in high relative relief areas and shallow soil, near thrusts and faults, and on southeast, south, and west-facing steep slopes. The WoE method achieved a prediction accuracy of 85.7%, indicating good accuracy of the model. Thus, this landslide susceptibility map could help the local governments in landslide hazard mitigation, land use planning, and landscape protection.

Ensemble techniques were developed, applied and validated for the analysis of landslide susceptibility in Jinbu area, Korea using the geographic information system (GIS). Landslide-occurrence areas were detected in the study by interpreting aerial photographs and field survey data. Landslide locations were randomly selected in a 70/30 ratio for training and validation of the models, respectively. Topography, geology, soil and forest databases were also constructed. Maps relevant to landslide occurrence were assembled in a spatial database. Using the constructed spatial database, 17 landslide-related factors were extracted. The relationships between the detected landslide locations and the factors were identified and quantified by frequency ratio, weight of evidence, logistic regression and artificial neural network models and their ensemble models. The relationships were used as factor ratings in the overlay analysis to create landslide susceptibility indexes and maps. Then, the four landslide susceptibility maps were used as new input factors and integrated using the frequency ratio, weight of evidence, logistic regression and artificial neural network models as ensemble methods to make better susceptibility maps. All of the susceptibility maps were validated by comparison with known landslide locations that were not used directly in the analysis. As the result, the ensemble-based landslide susceptibility map that used the new landslide-related input factor maps showed better accuracy (87.11% in frequency ratio, 83.14% in weight of evidence, 87.79% in logistic regression and 84.54% in artificial neural network) than the individual landslide susceptibility maps (84.94% in frequency ratio, 82.82% in weight of evidence, 87.72% in logistic regression and 81.44% in artificial neural network). All accuracy assessments showed overall satisfactory agreement of more than 80%. The ensemble model was found to be more effective in terms of prediction accuracy than the individual model.

Spatial prediction of landslide hazard at the Yihuang area (China) using two-class kernel logistic regression, alternating decision tree and support vector machines

Landslide susceptibility mapping is considered a useful tool for planning, disaster management, and natural hazard mitigation of a region. Although there are different methods for predicting landslide susceptibility, the bivariate statistical analysis method is considered to be simple and popular. In this study, the main aim is to evaluate the performance of Shannon entropy (SE) and weights of evidence (WOE) statistical models in landslide susceptibility mapping of Pithoragarh district of Uttarakhand state, India. For this purpose, ten landslide affecting factors, namely, slope degree, aspect, curvature, elevation, land cover, slope forming materials, geomorphology (landforms), distance to rivers, distance to roads, and overburden depth were used for the development of landslide susceptibility maps using the SE and WOE methods. Data extracted from the Google Earth images, Aster Digital Elevation Model, and Geological Survey of India report were used for the construction and evaluation of landslide susceptibility models and maps. The landslide data of 91 locations were randomly divided into two parts in the ratio of 70 : 30 using GIS software that is 70% data was used for training the models and 30% data was used for testing and validating the models. Performance of the applied models was evaluated using area under the AUC (area under the curve) ROC (receiver operating characteristics) curve. Results indicated that the WOE model is having better accuracy (AUCWOE = 68.75%) than the SE model (AUCSE = 52.17%) in the development of landslide susceptibility maps. Hence, WOE model can be used for the development of accurate landslide susceptibility maps which can provide useful information to decision maker and policy planner in better development of landslide prone areas.

Applying deep learning and benchmark machine learning algorithms for landslide susceptibility modelling in Rorachu river basin of Sikkim Himalaya, India

Landslide susceptibility mapping is among the first works for disaster management and land use planning activities in a mountain area like Ganzhou City. The aims of the current study are to assess GIS-based landslide spatial modeling using four models, namely data-driven evidential belief function (EBF), frequency ratio (FR), maximum entropy (Maxent), and logistic regression (LR), and to compare their performances. At first, a landslide inventory map was prepared according to aerial photographs, satellite images, and extensive field surveys. In total, 3971 landslide events were recognized in the study area that used 2979 landslides (75 %) for modeling and 992 landslide events (25 %) for validation. In the next step, the landslide-conditioning factors, namely slope angle, slope aspect, altitude, plan curvature, profile curvature, topographic wetness index (TWI), slope-length (LS), lithology, normalized difference vegetation index (NDVI), distance from rivers, distance from faults, distance from roads, and rainfall, were derived from the spatial database. Finally, landslide susceptibility maps of Ganzhou City were mapped in ArcGIS based on EBF, FR, Maxent, and LR approaches and were validated using the receiver operating characteristic (ROC) curve. The ROC plot assessment results showed that in the landslide susceptibility maps using the EBF, FR, Maxent, and LR models, the area under the curve (AUC) values were 0.7367, 0.7789, 0.7903, and 0.8237, respectively. Therefore, it can be concluded that all four models have AUC values of more than 0.70 and can be used in landslide susceptibility mapping in the study area. Also, the LR model had the best performance in the current study. Meanwhile, the mentioned models (EBF, FR, Maxent, and LR) showed almost similar results. The resultant susceptibility maps produced in the current study can be useful for land use planning and hazard mitigation purposes in the study area.

<p>The landslide development laws vary in different landslide-prone areas, hence the susceptibility models often perform in varied ways in different regions. Due to the periodic regulation of reservoir water level, a large number of landslides occur in the Three Gorges Reservoir area (TGRA). These landslides seriously threaten the safety of local residents and their property. It is crucial to find the model that can generate a landslide susceptibility map with higher accuracy in the TGRA. The main objective of this study was to explore the preference of machine learning models for landslide susceptibility mapping in the TGRA.</p><p>The Wushan segment of TGRA was selected as a case study, which is located in the middle reaches of the TGRA, the southwest of China. In this study, 165 landslides were identified and 14 landslide causal factors were constructed from different data sources at first, including altitude, slope, aspect, curvature, plan curvature, profile curvature, stream power index, topographic wetness index (TWI), terrain roughness index, lithology, bedding structure, distance to faults, distance to rivers, and distance to gully. Subsequently, multicollinearity analysis and information gain ratio model were applied to select landslide causal factors. After removing five factors (altitude, TWI, profile curvature, plan curvature, curvature), the landslide susceptibility mapping using the calculated results of four models, which were support vector machines (SVM), artificial neural networks, classification and regression tree, and logistic regression. Finally, the accuracy of the four models was evaluated and compared using the accuracy statistic methods and the receiver operating characteristic (ROC). The results of accuracy analysis showed that the SVM model performed the best. At the same time, the SVM performance behavior for susceptibility modelling in other areas were collected. In these regions, the accuracy of SVM was always larger than 0.8. We could see that SVM performed acceptably in different regions, and thus it can be used as a recommended model in TGRA and other landslide-prone regions.</p><p>In this study area, a total of 62% of the landslides were within 300 m from the Yangtze River, and the distance to rivers was the most important factor. The impoundment of the TGRA impacted the landslide development in three aspects: (1) the long-term immersion of reservoir water gradually reducing the strength of rock (soil) at the saturated zone (mostly near the Yangtze river), reducing the resistance force of landslide; (2) the strong dynamic action of water enhancing the lateral erosion on the bank slope, changing the slope shape, and thus reducing the slope stability; (3) the periodic fluctuation of the reservoir water making the self-weight, static, and dynamic water pressure of the landslide change, which could increase the resistance force or reduce the sliding force of the landslide and even cause overall instability and damage. Hence, in order to reduce the losses caused by landslides in TGRA, we should pay more attention to the early warning of reservoir bank landslides.</p>

Landslide susceptibility maps provide crucial information that helps local authorities, public institutions, and land-use planners make the correct decisions when they are managing landslide-prone areas. In recent years, machine-learning techniques have become very popular for producing landslide susceptibility maps. This study aims to compare the performance of these machine learning models with the traditional statistical methods used to produce landslide susceptibility maps. The landslide susceptibility for Ardanuc, Turkey was evaluated using three models: logistic regression (LR), support vector machine (SVM), and random forest (RF). Ten parameters that are effective in landslide occurrence are used in this study. The accuracy and prediction capabilities of the models were assessed using both the receiver operating characteristic (ROC) curve and area under the curve (AUC) methods. According to the AUC method, the success rate of the LR, SVM, and RF models was 83.1%, 93.2%, and 98.3%, respectively. Further, the prediction rates were calculated as 82.9% (LR), 92.8% (SVM), and 97.7% (RF). According to the verification results, RF and SVM models outperformed the traditional LR model in terms of success and prediction rate. The RF model, however, performed better than the SVM model in terms of success and prediction rates. The landslide susceptibility maps produced as a result of this study can guide city planners, local administrators, and public institutions related to disaster management to prevent and reduce landslide hazards.

A comprehensive landslide inventory and susceptibility maps are prerequisite for developing and implementing landslide mitigation strategies. Landslide susceptibility maps for the landslides prone regions in northern Pakistan are rarely available. The Hunza-Nagar valley in northern Pakistan is known for its frequent and devastating landslides. In this paper, we have developed a landslide inventory map for Hunza-Nagar valley by using the visual interpretation of the SPOT-5 satellite imagery and mapped a total of 172 landslides. The landslide inventory was subsequently divided into modelling and validation data sets. For the development of landslide susceptibility map seven discrete landslide causative factors were correlated with the landslide inventory map using weight of evidence and frequency ratio statistical models. Four different models of conditional independence were used for the selection of landslide causative factors. The produced landslides susceptibility maps were validated by the success rate and area under curves criteria. The prediction power of the models was also validated with the prediction rate curve. The validation results shows that the success rate curves of the weight of evidence and the frequency models are 82% and 79%, respectively. The prediction accuracy results obtained from this study are 84% for weight of evidence model and 80% for the frequency ratio model. Finally, the landslide susceptibility index maps were classified into five different varying susceptibility zones. The validation and prediction result indicates that the weight of evidence and frequency ratio model are reliable to produce an accurate landslide susceptibility map, which may be helpful for landslides management strategies.

Landslides and other disastrous natural catastrophes jeopardise natural resources, assets, and people's lives. As a result, future resource management will necessitate landslide susceptibility mapping (LSM) using the best conditioning factors. In Aqabat Al-Sulbat, Asir province, Saudi Arabia, the goal of this study was to find optimal conditioning parameters dependent hybrid LSM. LSM was created using machine learning methods such as random forest (RF), logistic regression (LR), and artificial neural network (ANN). To build ensemble models, the LR was combined with RF and ANN models. The receiver operating characteristic (ROC) curve was used to validate the LSMs and determine which models were the best. Then, utilising random forest (RF), classification and regression tree (CART), and correlation feature selection, sensitivity analysis was carried out. Through sensitivity analysis, the most relevant conditioning factors were determined, and the best model was applied to the important parameters to build a highly robust LSM with fewer variables. The ROC curve was also used to evaluate the final model. The results show that two hybrid models (LR-ANN and LR-RF) were predicted the very high as 29.67-32.73 km2 and high LS regions as 21.84-33.38 km2, with LR predicting 22.34km2 as very high and 45.15km2 as high LS zones. The LR-RF appeared as best model (AUC: 0.941), followed by LR-ANN (AUC: 0.915) and LR (AUC: 0.872). Sensitivity analysis, on the other hand, allows for the exclusion of aspects, hillshade, drainage density, curvature, and TWI from LSM. The LSM was then predicted using the LR-RF model based on the remaining nine conditioning factors. With fewer variables, this model has achieved greater accuracy (AUC: 0.927). This comes very close to being the best hybrid model. As a result, it is strongly advised to choose conditioning parameters with caution, as redundant parameters would result in less resilient LSM. As a consequence, both time and resources would be saved, and precise LSM would indeed be possible.

Landslides, which usually occur in mountainous and hilly areas, occur as a result of the soil or rock material forming a slope moving down under the influence of gravity. Forested areas, mostly in mountainous regions, are susceptible to landslides. Forest roads are important infrastructure facilities to protect forest resources and to achieve sustainable management objectives. Forest roads provide many benefits such as facilitating the transportation of wood raw materials, preventing fires and providing access to areas where recreational activities are carried out. However, inappropriately opened forest roads in forest areas cause problems such as landslides, which cause both serious destruction of road networks and serious deformations in forest areas. Landslide-prone forest roads also cause serious economic losses due to disruption of product transport and road maintenance costs. Within the scope of this study, landslide susceptibility maps (LSMs) were produced to determine the relationship between landslides and landslide-causing factors in Handüzü Forest Management Unit of Kastamonu Regional Directorate of Forestry (KRDF) located in the Central Black Sea Region of Türkiye. Land use, altitude, slope, aspect, plan and profile curvature, topographic wetness index (TWI), distance to forest road, drainage networks and fault, crown closure and lithology were used as conditioning factors in the study. Logistic Regression (LR) and Support Vector Machine (SVM) based machine learning models were used to generate LSMs. The receiver operating characteristics (ROC) curve and area under the ROC curve (AUC) method were used to compare the performance of landslide susceptibility models. In the accuracy assessment using the prediction rate curve, the AUC value was 0.968 for the SVM model and 0.668 for the LR model. The AUC values confirmed that SVM performed much better than LR. In addition, the susceptibility of newly planned forest roads (not currently available in the field) in LSMs were determined in the study. As a result of the study, it was determined that the most effective factors affecting landslides in Handüzü Forest Management Directorate are distance to forest roads and drainage networks. In the analyses, it was found that 28.28% of the existing forest roads in the LSM produced with SVM and 56.57% in the LSM produced with LR were found to be in »high« and »very high« landslide susceptible areas. Similarly, 38.43% of the newly planned roads in the LSM produced with SVM and 52.23% in the LSM produced with LR were found to be in »high« and »very high« landslide susceptible areas. These findings showed that forest roads are the main factor in the occurrence of landslides in the study area. Therefore, taking LSMs into account in the planning of forest roads will contribute to reducing the damages that may occur in forest areas due to landslides.