High-precision landslide susceptibility assessment based on the coupling of IHAOAVOA algorithm and BP neural network

• Published in: Earth science informatics Vol. 18; no. 2; p. 247
• Main Authors: Liang, Siyu, Li, Li, Qiang, Yue, Xu, Xinlong, Yang, Wenjun, Chen, Tao, Tan, Xinyi, Wang, Xi
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.02.2025; Springer Nature B.V
• Subjects: Accuracy; Algorithms; Annual rainfall; Artificial intelligence; Back propagation networks; Correlation coefficient; Correlation coefficients; Coupling; Decision making; Drainage density; Earth and Environmental Science; Earth Sciences; Earth System Sciences; Elevation; Emergency preparedness; Evaluation; Geography; Geology; Hazard assessment; Human engineering; Hydrology; Information Systems Applications (incl.Internet); Landslides; Landslides & mudslides; Lithology; Machine learning; Methods; Neural networks; Ontology; Simulation and Modeling; Space Exploration and Astronautics; Space Sciences (including Extraterrestrial Physics; Topography; China; Three Gorges Reservoir; Yangtze River; Negative samples; Landslide susceptibility; Information value; IHAOAVOA; Machine learning
• ISSN: 1865-0473; 1865-0481
• DOI: 10.1007/s12145-025-01753-9

The terrain and geological conditions in western China are complex, and landslide disasters occur frequently. Accurate assessment of landslide susceptibility is crucial for disaster prevention and control. Single machine learning models have been widely used in landslide susceptibility assessment; however, further optimization of their results remains a subject worth considering. Traditional BP neural networks encounter issues during the training process, such as getting trapped in local optima, slow training speeds, and a tendency to overfit. To address these challenges, this paper presents a case study from Yunyang County, Chongqing City, China. This study first investigates the geological environment and current conditions of the study area, selecting 11 factors such as elevation, slope, and lithology as influential factors for landslide susceptibility. Pearson’s correlation coefficient and multicollinearity analysis were applied to evaluate the relationships between the factors, leading to the elimination of highly correlated factors. The information gain ratio was applied to assess the importance of factors, and the results showed that elevation, NDVI, and annual rainfall were the most influential factors on the landslide susceptibility model, while drainage density and road density were the least influential factors. Next, the samples were divided into training and testing datasets in a 7:3 ratio, with slope units chosen as the evaluation units. BP, AO-BP, AVOA-BP, and IHAOAVOA-BP models were employed to construct the landslide susceptibility assessment model. The models’ accuracy was evaluated using AUC, Accuracy, Precision, Recall, and F-measure. The results indicate that integrated models, which combine algorithm coupling, outperform individual models. Among them, the IHAOAVOA-BP model performed the best, with an AUC value of 0.87, which is 0.08, 0.04, and 0.02 higher than the BP, AO-BP, and AVOA-BP models, respectively. The findings of this study suggest that the IHAOAVOA algorithm combines the strengths of both algorithms. By incorporating this algorithm into the BP neural network, it is possible to overcome the limitations of the BP model. The IHAOAVOA-BP coupled model demonstrates ideal predictive accuracy and strong applicability for landslide susceptibility evaluation.