Leveraging Near Real-Time Remote Sensing and Explainable AI for Rapid Landslide Detection: A Case Study in Greece

Landslides, triggered by severe rainfall events, pose significant risks to both life and infrastructure. Timely and accurate detection of such landslides is crucial for effective disaster management and mitigation. This study presents an innovative approach combining near real-time remote sensing data with advanced machine learning techniques to rapidly identify landslide occurrences following severe rainfall events, specifically focusing on a recent case in Greece.
Our methodology harnesses the capabilities of pre and post-event satellite imagery to capture the landscape's transformation due to landslides. We compute remote sensing indices, including the Normalized Difference Vegetation Index (NDVI) and the Normalized Difference Water Index (NDWI), among others, to detect changes indicative of potential landslide areas. This approach leverages the temporal resolution and wide-area coverage of satellite data, enabling a swift and comprehensive assessment immediately after a triggering rainfall event.
To enhance the accuracy of our detection model and reduce false positives, we incorporate a landslide susceptibility map generated via a Weight of Evidence (WoE) model. This map is based on historical landslide occurrences and helps to exclude areas with very low to low susceptibility, thereby refining our detection process.
Central to our study is the implementation of an eXplainable AI (XAI) framework. This aspect is particularly crucial, as it provides insights into the influence of various landslide-related factors on the model's predictions. The factors considered include elevation, slope angle, aspect, plan and profile curvature, distance to faults and river networks, lithology, and hydrolithology cover. By employing XAI techniques, we unravel the complex interactions between these variables and their relative importance in predicting landslide occurrences. This not only enhances the trustworthiness and transparency of our model but also aids in understanding the underlying geophysical processes leading to landslides.
The model's architecture is built upon advanced machine learning algorithms capable of processing large datasets efficiently. This setup is particularly suited to handle the high-dimensional and multi-temporal nature of remote sensing data. Furthermore, the model's ability to rapidly process and analyze data aligns well with the urgency required in disaster response scenarios.
Our case study in Greece demonstrates the model's efficacy in rapidly identifying landslide-prone areas post-severe rainfall events. The results show a significant improvement over traditional methods in terms of speed and accuracy. Moreover, the inclusion of XAI provides valuable insights for local authorities and disaster management teams, enabling them to make informed decisions for emergency response and long-term land-use planning.
This research contributes to the evolving field of rapid landslide detection by integrating cutting-edge remote sensing technologies with the latest advancements in machine learning and AI interpretability. It offers a novel, efficient, and transparent approach to landslide detection, which is vital for enhancing disaster preparedness and resilience in landslide-prone regions.