Landslide susceptibility prediction modelling based on semi‐supervised XGBoost model DOI

Qiangqiang Shua,

Hongbin Peng,

Jingkai Li

et al.

Geological Journal, Journal Year: 2024, Volume and Issue: 59(9), P. 2655 - 2667

Published: March 8, 2024

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

Language: Английский

Refined and dynamic susceptibility assessment of landslides using InSAR and machine learning models DOI Creative Commons

Yingdong Wei,

Haijun Qiu, Zijing Liu

et al.

Geoscience Frontiers, Journal Year: 2024, Volume and Issue: 15(6), P. 101890 - 101890

Published: July 9, 2024

Landslide susceptibility assessment is crucial in predicting landslide occurrence and potential risks. However, traditional methods usually emphasize on larger regions of landsliding rely relatively static environmental conditions, which exposes the hysteresis refined-scale temporal dynamic changes. This study presents an improved approach by integrating machine learning models based random forest (RF), logical regression (LR), gradient boosting decision tree (GBDT) with interferometric synthetic aperture radar (InSAR) technology comparing them to their respective original models. The results demonstrated that combined improves prediction accuracy reduces false negative positive errors. LR-InSAR model showed best performance at both regional smaller scale, particularly when identifying areas high very susceptibility. Modeling were verified using data from field investigations including unmanned aerial vehicle (UAV) flights. great significance accurately assess help reduce prevent risk.

Language: Английский

Citations

44
Improving pixel-based regional landslide susceptibility mapping DOI Creative Commons
Xin Wei, Paolo Gardoni, Lulu Zhang

et al.

Geoscience Frontiers, Journal Year: 2024, Volume and Issue: 15(4), P. 101782 - 101782

Published: Jan. 12, 2024

Regional landslide susceptibility mapping (LSM) is essential for risk mitigation. While deep learning algorithms are increasingly used in LSM, their extensive parameters and scarce labels (limited records) pose training challenges. In contrast, classical statistical algorithms, with typically fewer parameters, less likely to overfit, easier train, offer greater interpretability. Additionally, integrating physics-based data-driven approaches can potentially improve LSM. This paper makes several contributions enhance the practicality, interpretability, cross-regional generalization ability of regional LSM models: (1) Two new hybrid models, composed modules, proposed compared. Hybrid Model I combines infinite slope stability analysis (ISSA) logistic regression, a algorithm. II integrates ISSA convolutional neural network, representative techniques. The module constructs explanatory factor higher nonlinearity reduces prediction uncertainty caused by incomplete inventory pre-selecting non-landslide samples. captures relation between factors inventory. (2) A step-wise deletion process assess importance identify minimum necessary required maintain satisfactory model performance. (3) Single-pixel local-area samples compared understand effect pixel spatial neighborhood. (4) impact on performance explored. Typical landslide-prone regions Three Gorges Reservoir, China, as study area. results show that, testing region, using account neighborhoods, achieves roughly 4.2% increase AUC. Furthermore, models 30 m resolution land-cover data surpass those 1000 data, showing 5.5% improvement optimal set includes elevation, type, safety factor. These findings reveal key elements offering valuable insights practices.

Language: Английский

Citations

21
Uncertainties in landslide susceptibility prediction modeling: A review on the incompleteness of landslide inventory and its influence rules DOI Creative Commons
Faming Huang,

Daxiong Mao,

Shui‐Hua Jiang

et al.

Geoscience Frontiers, Journal Year: 2024, Volume and Issue: 15(6), P. 101886 - 101886

Published: July 1, 2024

Landslide inventory is an indispensable output variable of landslide susceptibility prediction (LSP) modelling. However, the influence incompleteness on LSP and transfer rules resulting error in model have not been explored. Adopting Xunwu County, China, as example, existing first obtained assumed to contain all samples under ideal conditions, after which different sample missing conditions are simulated by random sampling. It includes condition that whole study area randomly at proportions 10%, 20%, 30%, 40% 50%, well south County aggregation. Then, five machine learning models, namely, Random Forest (RF), Support Vector Machine (SVM), used perform LSP. Finally, results evaluated analyze uncertainties various conditions. In addition, this introduces interpretability methods explore changes decision basis RF Results show (1) certain (10%–50%) may affect for local areas. (2) Aggregation cause significant biases LSP, particularly areas where missing. (3) When 50% (either or aggregated), mainly manifested two aspects: first, importance ranking environmental factors slightly differs; second, regard modelling same test grid unit, weights individual drastically vary.

Language: Английский

Citations

19
Predicting Max Scour Depths near Two-Pier Groups Using Ensemble Machine-Learning Models and Visualizing Feature Importance with Partial Dependence Plots and SHAP DOI
Buddhadev Nandi, Subhasish Das

Journal of Computing in Civil Engineering, Journal Year: 2025, Volume and Issue: 39(2)

Published: Jan. 11, 2025

Language: Английский

Citations

2
Uncertainties of landslide susceptibility prediction: Influences of random errors in landslide conditioning factors and errors reduction by low pass filter method DOI Creative Commons
Faming Huang,

Zuokui Teng,

Chi Yao

et al.

Journal of Rock Mechanics and Geotechnical Engineering, Journal Year: 2023, Volume and Issue: 16(1), P. 213 - 230

Published: Nov. 20, 2023

In the existing landslide susceptibility prediction (LSP) models, influences of random errors in conditioning factors on LSP are not considered, instead original directly taken as model inputs, which brings uncertainties to results. This study aims reveal influence rules different proportional uncertainties, and further explore a method can effectively reduce factors. The firstly used construct factors-based then 5%, 10%, 15% 20% added these for constructing relevant errors-based models. Secondly, low-pass filter-based models constructed by eliminating using filter method. Thirdly, Ruijin County China with 370 landslides 16 case. Three typical machine learning i.e. multilayer perceptron (MLP), support vector (SVM) forest (RF), selected Finally, discussed results show that: (1) decrease uncertainties. (2) With proportions increasing from 5% 20%, uncertainty increases continuously. (3) feasible absence more accurate (4) degrees two issues, errors, modeling large basically same. (5) Shapley values explain internal mechanism predicting susceptibility. conclusion, greater proportion higher uncertainty, errors.

Language: Английский

Citations

31
Autonomous prediction of rock deformation in fault zones of coal roadways using supervised machine learning DOI
Feng Guo, Nong Zhang, Xiaowei Feng

et al.

Tunnelling and Underground Space Technology, Journal Year: 2024, Volume and Issue: 147, P. 105724 - 105724

Published: March 22, 2024

Language: Английский

Citations

12
Comprehensive review of remote sensing integration with deep learning in landslide forecasting and future directions DOI

Nilesh Suresh Pawar,

Kul Vaibhav Sharma

Natural Hazards, Journal Year: 2025, Volume and Issue: unknown

Published: March 10, 2025

Language: Английский

Citations

1
Landslide spatial prediction using cluster analysis DOI
Zheng Zhao, Hengxing Lan, Langping Li

et al.

Gondwana Research, Journal Year: 2024, Volume and Issue: 130, P. 291 - 307

Published: March 1, 2024

Language: Английский

Citations

8
Optimization method of conditioning factors selection and combination for landslide susceptibility prediction DOI Creative Commons
Faming Huang,

K Y Liu,

Shui‐Hua Jiang

et al.

Journal of Rock Mechanics and Geotechnical Engineering, Journal Year: 2024, Volume and Issue: unknown

Published: Aug. 1, 2024

Landslide susceptibility prediction (LSP) is significantly affected by the uncertainty issue of landslide related conditioning factor selection. However, most literature only performs comparative studies on a certain selection method rather than systematically study this issue. Targeted, aims to explore influence rules various commonly used methods LSP, and basis innovatively propose principle with universal application for optimal factors. An'yuan County in southern China taken as example considering 431 landslides 29 types Five methods, namely, correlation analysis (CA), linear regression (LR), principal component (PCA), rough set (RS) artificial neural network (ANN), are applied select combinations from original The results then inputs four common machine learning models construct 20 combined models, such CA-multilayer perceptron, CA-random forest. Additionally, multifactor-based multilayer perceptron random forest that selecting factors based proposed "accurate data, rich types, clear significance, feasible operation avoiding duplication" constructed comparisons. Finally, LSP uncertainties evaluated accuracy, index distribution, etc. Results show that: (1) have generally higher performance lower those selection-based models; (2) Influence degree different accuracy greater methods. Conclusively, above not ideal improving may complicate processes. In contrast, satisfied combination can be according principle.

Language: Английский

Citations

8
A landslide susceptibility assessment method considering the similarity of geographic environments based on graph neural network DOI
Qing Zhang, Yi He, Lifeng Zhang

et al.

Gondwana Research, Journal Year: 2024, Volume and Issue: 132, P. 323 - 342

Published: May 16, 2024

Language: Английский

Citations

7