Search

The robustness of landslide prediction models has become a major focus of researchers worldwide. We developed two novel hybrid predictive models that combine the self-organizing, deep-learning group method of data handling (GMDH) with two swarm intelligence optimization algorithms, i.e., cuckoo search algorithm (CSA) and whale optimization algorithm (WOA) for spatially explicit prediction of landslide susceptibility. Eleven landslide-causing factors and 334 historic landslides in a 31,340 km2 landslide-prone area in Iran were used to produce geospatial training and validation datasets. The GMDH model was employed to develop a basic predictive model that was then restructured and its parameters were optimized using the CSA and WOA algorithms, yielding the novel hybrid GMDH-CSA and GMDH-WOA models. The hybrid models were validated and compared to the standalone GMDH model by calculating the area under the receiver operating characteristic (AUC) curve and root mean square error (RMSE). The results demonstrated that the hybrid models overcame the computational shortcomings of the basic GMDH model and significantly improved landslide susceptibility prediction (GMDH-CSA, AUC = 0.909 and RMSE = 0.089; GMDH-WOA, AUC = 0.902 and RMSE = 0.129; standalone GMDH, AUC = 0.791 and RMSE = 0.226). Further, the hybrid models were more robust than the standalone GMDH model, showing consistently excellent performance when the training and validation datasets were changed. Overall, the swarm intelligence-optimized models, but not the standalone model, identified the best trade-offs among objectives, accuracy, and robustness.

This study investigates a pixel-based image analysis methodology built on unsupervised Deep Learning (DL) for rapid landslide detection. The utilized data includes the Minimum Noise Fraction (MNF) and Normalized Difference Vegetation Index (NDVI) derived from Sentinel-2 images and the topographic slope factor derived from the ALOS PALSAR sensor. We used a Convolutional auto-encoder (CAE) for extracting deep features from our input data. The Mini Batch K-means is then used for clustering the resulting deep features. The resulting landslide detection maps were then compared with a landslide inventory dataset for accuracy assessment. The proposed approach achieved the highest values of 76%, 91%, 83%, and 70% in terms of precision, recall, f1-score, and mIOU, respectively. This is the first study investigating unsupervised DL for landslide detection using Sentinel-2 images to the best of our knowledge.

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.

This paper proposes a new approach based on an unsupervised deep learning (DL) model for landslide detection. Recently, supervised DL models using convolutional neural networks (CNN) have been widely studied for landslide detection. Even though these models provide robust performance and reliable results, they depend highly on a large labeled dataset for their training step. As an alternative, in this paper, we developed an unsupervised learning model by employing a convolutional auto-encoder (CAE) to deal with the problem of limited labeled data for training. The CAE was used to learn and extract the abstract and high-level features without using training data. To assess the performance of the proposed approach, we used Sentinel-2 imagery and a digital elevation model (DEM) to map landslides in three different case studies in India, China, and Taiwan. Using minimum noise fraction (MNF) transformation, we reduced the multispectral dimension to three features containing more than 80% of scene information. Next, these features were stacked with slope data and NDVI as inputs to the CAE model. The Huber reconstruction loss was used to evaluate the inputs. We achieved reconstruction losses ranging from 0.10 to 0.147 for the MNF features, slope, and NDVI stack for all three study areas. The mini-batch K-means clustering method was used to cluster the features into two to five classes. To evaluate the impact of deep features on landslide detection, we first clustered a stack of MNF features, slope, and NDVI, then the same ones plus with the deep features. For all cases, clustering based on deep features provided the highest precision, recall, F1-score, and mean intersection over the union in landslide detection.

Earthquakes and heavy rainfalls are the two leading causes of landslides around the world. Since they often occur across large areas, landslide detection requires rapid and reliable automatic detection approaches. Currently, deep learning (DL) approaches, especially different convolutional neural network and fully convolutional network (FCN) algorithms, are reliably achieving cutting-edge accuracies in automatic landslide detection. However, these successful applications of various DL approaches have thus far been based on very high-resolution satellite images (e.g., GeoEye and WorldView), making it easier to achieve such high detection performances. In this study, we use freely available Sentinel-2 data and ALOS digital elevation model to investigate the application of two wellknown FCN algorithms, namely the U-Net and residual U-Net (or so-called ResU-Net), for landslide detection. To our knowledge, this is the first application of FCN for landslide detection only from freely available data. We adapt the algorithms to the specific aim of landslide detection, then train and test with data from three different case study areas located in Western Taitung County (Taiwan), Shuzheng Valley (China), and Eastern Iburi (Japan). We characterize three different window size sample patches to train the algorithms. Our results also contain a comprehensive transferability assessment achieved through different training and testing scenarios in the three case studies. The highest f1-score value of 73.32% was obtained by ResU-Net, trained with a dataset from Japan, and tested on China’s holdout testing area using the sample patch size of 64 × 64 pixels.

Gully erosion is a severe form of soil erosion that results in a wide range of environmental problems such as, dams’ sedimentation, destruction of transportation and energy transmission lines, decreasing and losing farmland productivity, and land degradation. The main objective of this study is to accurately map the areas prone to gully erosion, by developing two machine learning (ML) models, namely artificial neural network (ANN) and random forest (RF) models within 4-fold cross-validation (CV). Moreover, we used the multi-collinearity analysis to select 11 variables among 15 conditioning factors to train the ML models for gully erosion susceptibility mapping (GESM). Lamerd county, Iran, is chosen for a study area because Lamerd county is one of the most affected areas by gully erosion in this country. From 232 gully samples, 75% was used to train the two ML models and the rest of the samples (25%) were used to validate the generated GEMSs using 4-fold CV. The RF model produced a higher accuracy with an accuracy value of 93%. The GEMS generated by the RF model shows that the areas classified as highly vulnerable and very highly vulnerable are 1,869 ha and 5,148 ha, respectively. Results from the two models indicated that the most vulnerable land use/landcover class is bare land because of the low vegetation cover. The outcome of this study can help managers in Lamerd county to mitigate the soil erosion problem and prevent future gully erosion by taking preventive measures.

© 2019 Elsevier B.V. Floods are one of the most devastating types of disasters that cause loss of lives and property worldwide each year. This study aimed to evaluate and compare the prediction capability of the naïve Bayes tree (NBTree), alternating decision tree (ADTree), and random forest (RF) methods for the spatial prediction of flood occurrence in the Quannan area, China. A flood inventory map with 363 flood locations was produced and partitioned into training and validation datasets through random selection with a ratio of 70/30. The spatial flood database was constructed using thirteen flood explanatory factors. The probability certainty factor (PCF) method was used to analyze the correlation between the factors and flood occurrences. Consequently, three flood susceptibility maps were produced using the NBTree, ADTree, and RF methods. Finally, the area under the curve (AUC) and statistical measures were used to validate the flood susceptibility models. The results indicated that the RF method is an efficient and reliable model in flood susceptibility assessment, with the highest AUC values, positive predictive rate, negative predictive rate, sensitivity, specificity, and accuracy for the training (0.951, 0.892, 0.941, 0.945, 0.886, and 0.915, respectively) and validation (0.925, 0.851, 0.938, 0.945, 0.835, and 0.890, respectively) datasets.

© 2019 by the authors. This study proposes a hybrid computational intelligence model that is a combination of alternating decision tree (ADTree) classifier and AdaBoost (AB) ensemble, namely "AB-ADTree", for groundwater spring potential mapping (GSPM) at the Chilgazi watershed in the Kurdistan province, Iran. Although ADTree and its ensembles have been widely used for environmental and ecological modeling, they have rarely been applied to GSPM. To that end, a groundwater spring inventory map and thirteen conditioning factors tested by the chi-square attribute evaluation (CSAE) technique were used to generate training and testing datasets for constructing and validating the proposed model. The performance of the proposed model was evaluated using statistical-index-based measures, such as positive predictive value (PPV), negative predictive value (NPV), sensitivity, specificity accuracy, root mean square error (RMSE), and the area under the receiver operating characteristic (ROC) curve (AUROC). The proposed hybrid model was also compared with five state-of-the-art benchmark soft computing models, including singleADTree, support vector machine (SVM), stochastic gradient descent (SGD), logistic model tree (LMT), logistic regression (LR), and random forest (RF). Results indicate that the proposed hybrid model significantly improved the predictive capability of the ADTree-based classifier (AUROC = 0.789). In addition, it was found that the hybrid model, AB-ADTree, (AUROC = 0.815), had the highest goodness-of-fit and prediction accuracy, followed by the LMT (AUROC = 0.803), RF (AUC = 0.803), SGD, and SVM (AUROC = 0.790) models. Indeed, this model is a powerful and robust technique for mapping of groundwater spring potential in the study area. Therefore, the proposed model is a promising tool to help planners, decision makers, managers, and governments in the management and planning of groundwater resources.

© 2019 by the authors. Licensee MDPI, Basel, Switzerland. In this study, we introduced a novel hybrid artificial intelligence approach of rotation forest (RF) as a Meta/ensemble classifier based on alternating decision tree (ADTree) as a base classifier called RF-ADTree in order to spatially predict gully erosion at Klocheh watershed of Kurdistan province, Iran. A total of 915 gully erosion locations along with 22 gully conditioning factors were used to construct a database. Some soft computing benchmark models (SCBM) including the ADTree, the Support Vector Machine by two kernel functions such as Polynomial and Radial Base Function (SVM-Polynomial and SVM-RBF), the Logistic Regression (LR), and the Naïve Bayes Multinomial Updatable (NBMU) models were used for comparison of the designed model. Results indicated that 19 conditioning factors were effective among which distance to river, geomorphology, land use, hydrological group, lithology and slope angle were the most remarkable factors for gully modeling process. Additionally, results of modeling concluded the RF-ADTree ensemble model could significantly improve (area under the curve (AUC) = 0.906) the prediction accuracy of the ADTree model (AUC = 0.882). The new proposed model had also the highest performance (AUC = 0.913) in comparison to the SVM-Polynomial model (AUC = 0.879), the SVM-RBF model (AUC = 0.867), the LR model (AUC = 0.75), the ADTree model (AUC = 0.861) and the NBMU model (AUC = 0.811).

© 2019 by the authors. Floods are some of the most dangerous and most frequent natural disasters occurring in the northern region of Iran. Flooding in this area frequently leads to major urban, financial, anthropogenic, and environmental impacts. Therefore, the development of flood susceptibility maps used to identify flood zones in the catchment is necessary for improved flood management and decision making. The main objective of this study was to evaluate the performance of an Evidential Belief Function (EBF) model, both as an individual model and in combination with Logistic Regression (LR) methods, in preparing flood susceptibility maps for the Haraz Catchment in the Mazandaran Province, Iran. The spatial database created consisted of a flood inventory, altitude, slope angle, plan curvature, TopographicWetness Index (TWI), Stream Power Index (SPI), distance from river, rainfall, geology, land use, and Normalized Difference Vegetation Index (NDVI) for the region. After obtaining the required information from various sources, 151 of 211 recorded flooding points were used for model training and preparation of the flood susceptibility maps. For validation, the results of the models were compared to the 60 remaining flooding points. The Receiver Operating Characteristic (ROC) curve was drawn, and the Area Under the Curve (AUC) was calculated to obtain the accuracy of the flood susceptibility maps prepared through success rates (using training data) and prediction rates (using validation data). The AUC results indicated that the EBF, EBF from LR, EBF-LR (enter), and EBF-LR (stepwise) success rates were 94.61%, 67.94%, 86.45%, and 56.31%, respectively, and the prediction rates were 94.55%, 66.41%, 83.19%, and 52.98%, respectively. The results showed that the EBF model had the highest accuracy in predicting flood susceptibility within the catchment, in which 15% of the total areas were located in high and very high susceptibility classes, and 62% were located in low and

© 2019 by the authors. We used a novel hybrid functional machine learning algorithm to predict the spatial distribution of landslides in the Sarkhoon watershed, Iran. We developed a new ensemble model which is a combination of a functional algorithm, stochastic gradient descent (SGD) and an AdaBoost (AB) Meta classifier namely ABSGD model to predict the landslides. The model incorporates 20 landslide conditioning factors, which we ranked using the least-square support vector machine (LSSVM) technique. For the modeling, we considered 98 landslide locations, of which 70% (79) were used for training and 30% (19) for validation processes. Model validation was performed using sensitivity, specificity, accuracy, the root mean square error (RMSE) and the area under the receiver operatic characteristic (AUC) curve. We also used soft computing benchmark models, including SGD, logistic regression (LR), logistic model tree (LMT) and functional tree (FT) algorithms for model validation and comparison. The selected conditioning factors were significant in landslide occurrence but distance to road was found to be the most important factor. The ABSGD model (AUC= 0.860) outperformed the LR (0.797), SGD (0.776), LMT (0.740) and FT (0.734) models. Our results confirm that the combined use of a functional algorithm and a Meta classifier prevents over-fitting, reduces noise and enhances the power prediction of the individual SGD algorithm for the spatial prediction of landslides.

© 2019 China University of Geosciences (Beijing) and Peking University The sub-watershed prioritization is the ranking of different areas of a river basin according to their need to proper planning and management of soil and water resources. Decision makers should optimally allocate the investments to critical sub-watersheds in an economically effective and technically efficient manner. Hence, this study aimed at developing a user-friendly geographic information system (GIS) tool, Sub-Watershed Prioritization Tool (SWPT), using the Python programming language to decrease any possible uncertainty. It used geospatial–statistical techniques for analyzing morphometric and topo-hydrological factors and automatically identifying critical and priority sub-watersheds. In order to assess the capability and reliability of the SWPT tool, it was successfully applied in a watershed in the Golestan Province, Northern Iran. Historical records of flood and landslide events indicated that the SWPT correctly recognized critical sub-watersheds. It provided a cost-effective approach for prioritization of sub-watersheds. Therefore, the SWPT is practically applicable and replicable to other regions where gauge data is not available for each sub-watershed.

© 2018 by the authors. Licensee MDPI, Basel, Switzerland. In this study, land subsidence susceptibility was assessed for a study area in South Korea by using four machine learning models including Bayesian Logistic Regression (BLR), Support Vector Machine (SVM), Logistic Model Tree (LMT) and Alternate Decision Tree (ADTree). Eight conditioning factors were distinguished as the most important affecting factors on land subsidence of Jeong-am area, including slope angle, distance to drift, drift density, geology, distance to lineament, lineament density, land use and rock-mass rating (RMR) were applied to modelling. About 24 previously occurred land subsidence were surveyed and used as training dataset (70% of data) and validation dataset (30% of data) in the modelling process. Each studied model generated a land subsidence susceptibility map (LSSM). The maps were verified using several appropriate tools including statistical indices, the area under the receiver operating characteristic (AUROC) and success rate (SR) and prediction rate (PR) curves. The results of this study indicated that the BLR model produced LSSM with higher acceptable accuracy and reliability compared to the other applied models, even though the other models also had reasonable results.

© 2018 by the authors. Earthquakes are among the most catastrophic natural geo-hazards worldwide and endanger numerous lives annually. Therefore, it is vital to evaluate seismic vulnerability beforehand to decrease future fatalities. The aim of this research is to assess the seismic vulnerability of residential houses in an urban region on the basis of the Multi-Criteria Decision Making (MCDM) model, including the analytic hierarchy process (AHP) and geographical information system (GIS). Tabriz city located adjacent to the North Tabriz Fault (NTF) in North-West Iran was selected as a case study. The NTF is one of the major seismogenic faults in the north-western part of Iran. First, several parameters such as distance to fault, percent of slope, and geology layers were used to develop a geotechnical map. In addition, the structural construction materials, building materials, size of building blocks, quality of buildings and buildings-floors were used as key factors impacting on the building’s structural vulnerability in residential areas. Subsequently, the AHP technique was adopted to measure the priority ranking, criteria weight (layers), and alternatives (classes) of every criterion through pair-wise comparison at all levels. Lastly, the layers of geotechnical and spatial structures were superimposed to design the seismic vulnerability map of buildings in the residential area of Tabriz city. The results showed that South and Southeast areas of Tabriz city exhibit low to moderate vulnerability, while some regions of the north-eastern area are under severe vulnerability conditions. In conclusion, the suggested approach offers a practical and effective evaluation of Seismic Vulnerability Assessment (SVA) and provides valuable information that could assist urban planners during mitigation and preparatory phases of less examined areas in many other regions around the world.

© 2018 by the authors. Licensee MDPI, Basel, Switzerland. The main objective of this research was to introduce a novel machine learning algorithm of alternating decision tree (ADTree) based on the multiboost (MB), bagging (BA), rotation forest (RF) and random subspace (RS) ensemble algorithms under two scenarios of different sample sizes and raster resolutions for spatial prediction of shallow landslides around Bijar City, Kurdistan Province, Iran. The evaluation of modeling process was checked by some statistical measures and area under the receiver operating characteristic curve (AUROC). Results show that, for combination of sample sizes of 60%/40% and 70%/30% with a raster resolution of 10 m, the RS model, while, for 80%/20% and 90%/10% with a raster resolution of 20 m, the MB model obtained a high goodness-of-fit and prediction accuracy. The RS-ADTree and MB-ADTree ensemble models outperformed the ADTree model in two scenarios. Overall, MB-ADTree in sample size of 80%/20% with a resolution of 20 m (area under the curve (AUC) = 0.942) and sample size of 60%/40% with a resolution of 10 m (AUC = 0.845) had the highest and lowest prediction accuracy, respectively. The findings confirm that the newly proposed models are very promising alternative tools to assist planners and decision makers in the task of managing landslide prone areas.