A hybrid ensemble-based deep-learning framework for landslide susceptibility mapping

International Journal of Applied Earth Observation and Geoinformation, Journal Year: 2022, Volume and Issue: 108, P. 102713 - 102713

Published: Feb. 11, 2022

Landslides are highly hazardous geological disasters that can potentially threaten the safety of human life and property. As a result, landslide susceptibility mapping (LSM) plays an important role in prevention system. Recently, many deep learning (DL) models have been adopted for LSM, but they also face problems such as sensitivity to overfitting lower accuracy. In this paper, novel hybrid LSM framework is proposed based on four heterogeneous ensemble (HEL) methods with three single DL models: belief network (DBN), convolutional neural (CNN) residual (ResNet). The model tested at Three Gorges Reservoir area, China. 202 historical landslides ten conditioning factors were selected construct geospatial dataset LSM. high-correlation low importance removed from by using Spearman Correlation Index random forests. was then divided into two subsets: 70% training 30% testing. Then results carried out HEL-based models. quantitative evaluation showed improved accuracy, outperformed Stacking achieved highest AUC value (0.984), Kappa (86.95%), overall accuracy (94.17%), precision (88.87%), Matthews correlation coefficient (87.03%) F1-score (91.34%) among all testing dataset, while Boosting obtained Recall (96.02%). At same time, study show better stability avoid effectively. addition, Gini index elevation factor contributes most area. general, has promising applicability improving

Language: Английский

---

Machine learning-based landslide susceptibility assessment with optimized ratio of landslide to non-landslide samples

Gondwana Research, Journal Year: 2022, Volume and Issue: 123, P. 198 - 216

Published: May 25, 2022

Language: Английский

---

Flood susceptibility mapping using an improved analytic network process with statistical models

Geomatics Natural Hazards and Risk, Journal Year: 2020, Volume and Issue: 11(1), P. 2282 - 2314

Published: Jan. 1, 2020

Flooding is a natural disaster that causes considerable damage to different sectors and severely affects economic social activities. The city of Saqqez in Iran susceptible flooding due its specific environmental characteristics. Therefore, susceptibility vulnerability mapping are essential for comprehensive management reduce the harmful effects flooding. primary purpose this study combine Analytic Network Process (ANP) decision-making method statistical models Frequency Ratio (FR), Evidential Belief Function (EBF), Ordered Weight Average (OWA) flood City Kurdistan Province, Iran. frequency ratio was used instead expert opinions weight criteria ANP. ten factors influencing area slope, rainfall, slope length, topographic wetness index, aspect, altitude, curvature, distance from river, geology, land use/land cover. We identified 42 points area, 70% which modelling, remaining 30% validate models. Receiver Operating Characteristic (ROC) curve evaluate results. under obtained ROC indicates superior performance ANP EBF hybrid model (ANP-EBF) with 95.1% efficiency compared combination FR (ANP-FR) 91% OWA (ANP-OWA) 89.6% efficiency.

Language: Английский

---

DEM resolution effects on machine learning performance for flood probability mapping

Journal of Hydro-environment Research, Journal Year: 2021, Volume and Issue: 40, P. 1 - 16

Published: Nov. 9, 2021

Floods are among the devastating natural disasters that occurred very frequently in arid regions during last decades. Accurate assessment of flood susceptibility mapping is crucial sustainable development. It helps respective authorities to prevent as much possible their irreversible consequences. The Digital Elevation Model (DEM) spatial resolution one most base layer factors for modeling Flood Probability Maps (FPMs). Therefore, main objective this study was assess influence DEMs 12.5 m (ALOS PALSAR) and 30 (ASTER) on accuracy probability prediction using three machine learning models (MLMs), including Random Forest (RF), Artificial Neural Network (ANN), Generalized Linear (GLM). This selected 14 causative independent variables, 220 locations were dependent variables. Dependent variables divided into training (70%) validation (30%) modeling. Receiver Operating Characteristic Curve (ROC), Kappa index, accuracy, other statistical criteria used evaluate models' accuracy. results showed resolving DEM alone cannot significantly affect regardless applied MLM independently model performance In contrast, such altitude, precipitation, distance from river have a considerable impact floods region. Also, evaluation RF (AUC12.5,30m = 0.983, 0.975) more accurate preparing FPM than ANN 0.949, 0.93) GLM 0.965, 0.949) models. study's solution-oriented findings might help water managers decision-makers make effective adaptation mitigation measures against potential flooding.

Language: Английский

---

Multi-Hazard Exposure Mapping Using Machine Learning for the State of Salzburg, Austria

Remote Sensing, Journal Year: 2020, Volume and Issue: 12(17), P. 2757 - 2757

Published: Aug. 25, 2020

We live in a sphere that has unpredictable and multifaceted landscapes make the risk arising from several incidences are omnipresent. Floods landslides widespread recurring hazards occurring at an alarming rate recent years. The importance of this study is to produce multi-hazard exposure maps for flooding federal State Salzburg, Austria, using selected machine learning (ML) approach support vector (SVM) random forest (RF). Multi-hazard were established on thirteen influencing factors flood such as elevation, slope, aspect, topographic wetness index (TWI), stream power (SPI), normalized difference vegetation (NDVI), geology, lithology, rainfall, land cover, distance roads, faults, drainage. classified inventory data landslide into training validation with widely used splitting ratio, where 70% locations training, 30% validation. accuracy assessment was derived through ROC (receiver operating curve) R-Index (relative density). RF yielded better results both 0.87 0.90 compared 0.89 SVM. However, map Salzburg SVM provides planners managers plan regions affected by floods landslides.

Language: Английский

---

Snow avalanche susceptibility mapping using novel tree-based machine learning algorithms (XGBoost, NGBoost, and LightGBM) with eXplainable Artificial Intelligence (XAI) approach

Stochastic Environmental Research and Risk Assessment, Journal Year: 2023, Volume and Issue: 37(6), P. 2243 - 2270

Published: March 13, 2023

Language: Английский

---

Spatial modeling of flood probability using geo-environmental variables and machine learning models, case study: Tajan watershed, Iran

Advances in Space Research, Journal Year: 2021, Volume and Issue: 67(10), P. 3169 - 3186

Published: Feb. 21, 2021

Language: Английский

---

Determination of flood probability and prioritization of sub-watersheds: A comparison of game theory to machine learning

Journal of Environmental Management, Journal Year: 2021, Volume and Issue: 295, P. 113040 - 113040

Published: June 18, 2021

Language: Английский

---

Optimization of statistical and machine learning hybrid models for groundwater potential mapping

Geocarto International, Journal Year: 2021, Volume and Issue: 37(13), P. 3877 - 3911

Published: Jan. 4, 2021

Determining areas of high groundwater potential is important for exploitation, management, and protection water resources. This study assesses the spatial distribution in Zarrinehroud watershed Kurdistan Province, Iran using combinations five statistical machine learning algorithms – frequency ratio (FR), radial basis function (RBF), index entropy (IOE), evidential belief (EBF) fuzzy art map (FAM). To accomplish this, 1448 well locations area were randomly divided into two data sets training (70%= 1013 locations) validation (30%= 435 based on holdout method. Fourteen factors that can affect presence or absence identified, measured, mapped ArcGIS SAGA-GIS software. The models used to predict suitable conditioning produce maps. probability at any location was classified as low, moderate, high, very natural breaks spectrum. model predictions tested validity their success determined receiver operating characteristic (ROC) curves, standard errors (SE), positive predictive value (PPV), negative (NPV), sensitivity (SST), specificity (SPF) accuracy (ACC), Friedman test. performance assessments under curve (AUC) (ACC) showed FR-RBF had good (AUC= 0.889, ACC= 87.51). FR-FAM 0.869, 84.67), EBF-FAM 0.864, 84.42), EBF-RBF 0.854, 83.94), FR-IOE 0.836, 83.62), EBF-IOE 0.833, 80.42) also acceptable performance. results test show there are significant differences between highest mean rank generated by (3.642). Therefore, hybrid be increase groundwater-prediction region perhaps similar settings. HighlightsThe studied watershedA combination methods including FR, RBF, IOE, EBF FAMVery located northern halfThe development

Language: Английский

---

Decision support tools, systems and indices for sustainable coastal planning and management: A review

Ocean & Coastal Management, Journal Year: 2021, Volume and Issue: 212, P. 105813 - 105813

Published: July 21, 2021

Language: Английский

---