Document Type : Original Article
Authors
Department of Soil Science, College of Agriculture, Shiraz University, Shiraz, Iran
10.22034/iwm.2026.2096421.1281
Abstract
Extended Abstract
Introduction: Landslides are among the most important natural hazards, with widespread human, economic, and environmental consequences. Landslide susceptibility mapping plays an important role in disaster risk reduction, land-use planning, and sustainable environmental management. With the development of geographic information systems, remote sensing, and artificial intelligence, machine learning algorithms have become an important approach in this field because of their ability to model complex and nonlinear relationships. Despite the rapid growth of research in this area, a comprehensive bibliometric analysis of its scientific evolution and knowledge structure has not yet been conducted. Therefore, this study aimed to analyze the evolutionary trends, scientific performance, and knowledge structure of global research on machine-learning-based landslide susceptibility mapping.
Materials and Methods: Bibliometric data were retrieved from the Scopus database for the period 2002–2026. A total of 785 scientific documents published in 260 sources, involving 2,378 researchers and comprising 1,588 author keywords, were analyzed. The analyses were conducted using the R statistical environment and the Bibliometrix package, employing conventional knowledge-mapping methods, including descriptive analysis of research performance, temporal publication trends, scientific life-cycle analysis, Bradford’s Law, and international scientific collaboration analysis. Publication characteristics, citation counts, collaboration patterns, and journal distribution were also examined to identify the evolutionary trends and knowledge structure of this research field.
Results and Discussion: The average number of authors per article was 5.03, while 18.36% of the articles involved international collaboration. The mean document age was 2.84 years, and the average citation count was 74.38 citations per article, indicating the dynamism and favorable scientific impact of this field. Publication trends revealed three distinct phases: slow growth during 2002–2016, rapid growth during 2017–2020, and substantial growth during 2021–2026. Publications increased from approximately 40 articles in 2020 to nearly 150 articles in 2025. The scientific life-cycle model showed a good fit (R² = 0.888), with 2024 identified as the peak period of scientific growth. Bradford’s Law analysis showed that, among the 260 sources, 12 journals constituted the core zone, 39 journals formed the second zone, and 209 journals comprised the third zone. The Bradford coefficient was K = 3.4, with a high coefficient of determination (R² = 0.992), while the Kolmogorov–Smirnov test (p = 0.491) confirmed the consistency of the observed distribution with the theoretical model. China emerged as the leading hub of scientific collaboration, maintaining extensive collaborative networks with several countries, including the United States, India, Australia, and Iran.
Conclusion: Machine-learning-based landslide susceptibility mapping has emerged as a rapidly growing interdisciplinary field, and the findings indicate that it has entered a stage of scientific maturity. China occupies a central position in knowledge production and exchange, while international collaboration plays an important role in advancing the field. The findings can provide a basis for future research planning, particularly by focusing on interpretable artificial intelligence, deep learning, foundation models, and multimodal spatial data analysis.
Keywords
Subjects
