SKLGP大讲堂第121期 | Evaluating the Role of AI-Generated Landslide Inventories in Hazard and Risk Management: Progress, Challenges, and Perspectives

报告题目：Evaluating the Role of AI-Generated Landslide Inventories in Hazard and Risk Management: Progress, Challenges, and Perspectives

报告人：Sansar Raj Meena

单位：帕多瓦大学（ Padova）

时间：2025年6月27日10:00—11:30（周五）

地点：地灾国重211教室（珙桐对面）

个人简介：

Sansar Raj Meena博士现任意大利帕多瓦大学地球科学系研究员，并担任荷兰特温特大学ITC研究所访问科学家及波士顿大学遥感中心客座科学家。长期致力于地质灾害智能监测研究，在滑坡自动识别、多时相动态制图与AI风险评估领域取得多项创新成果：主导开发了全球首个高分辨率滑坡数据集（HR-GLDD），建立了基于深度学习的滑坡智能解译体系，研究成果被纳入联合国灾害评估指南。作为核心成员参与欧盟Horizon 2020等10余项国际科研项目，担任多个国际期刊编委及学术会议召集人，展现了突出的学术影响力与工程应用价值；在《Nature Communications》《Remote Sensing of Environment》等顶级期刊发表论文80余篇，总引用2800+，h指数为28。

报告简介：

Landslide inventories serve as fundamental datasets for susceptibility assessment, hazard modeling, and risk-informed decision-making. Conventional approaches relying on manual visual interpretation of remote sensing imagery are not only time-consuming and subjective, but also exhibit significant inconsistencies across different regions and temporal phases. This lecture focuses on exploring the application potential of artificial intelligence (AI) technologies, particularly deep learning, in developing semi-automatic/fully-automatic landslide identification frameworks capable of achieving rapid and accurate large-scale landslide detection.

First, we systematically evaluates the current state of AI-powered landslide identification technologies. While existing models can achieve over 80% recognition accuracy in regions similar to training data, their F1-scores typically drop to the 50%-80% range when confronted with diverse geomorphological types and disaster scenarios. In-depth analysis reveals that inconsistent annotation standards, lack of high-quality benchmark datasets, and incomplete validation systems significantly constrain the operational application effectiveness of these models.

To address these challenges, we proposes systematic solutions including establishing unified annotation standards, developing multi-region shared datasets, and improving model validation frameworks to substantially enhance the generalization capability of AI models. Particular emphasis is placed on the necessity to integrate AI technologies with traditional geoscience analysis methods, ensuring result reliability through rigorous validation processes before practical application in early warning systems and risk management.

Finally, this research provides important theoretical foundations for promoting the practical application of AI technologies in landslide risk reduction, effectively bridging the gap between algorithmic innovation and geoscientific application requirements.

欢迎广大师生踊跃参加！

地质灾害防治与地质环境保护全国重点实验室

2025年6月26日