Landslide Hunter: a fully automated EO platform for rapid mapping of landslides in semi-cloudy conditions

Landslides are a common natural hazard mostly triggered by seismic, climatic, or anthropogenic factors. The impacts of landslides on the nature, the built environment, and the society call for effective hazard management to improve our preparedness and resilience. Accurate landslide risk analysis methods are necessary to identify the elements at risk, and effective early warning systems are needed to prevent loss of life and economic damage. Landslide catalogs provide valuable information on past events that can be exploited for better hazard assessment and early warning. However, creating a landslide catalog is a time-consuming process, especially after major disasters. Several semi-automated landslide mapping methods using cloud-free optical satellite images have been developed recently that benefit from the advancements in image processing and AI technologies. However, such methods are mostly tested only in specific study areas, and it is uncertain if they can respond to the analysis needs globally. Compiling cloud-free images by combining many semi-cloudy images also requires significant time. Additionally, most landslides occur in mountainous regions, which are typically characterized by heavy rainfall patterns. As a result, finding cloud-free images that cover these areas in their entirety is quite difficult.

The Landslide Hunter is a prototype online platform designed to rapidly detect landslides using an innovative method, which analyzes consecutive partially cloudy optical Earth observation (EO) images to identify visible landslide extents and then automatically integrate these partial extents to determine the complete extent of the landslides. The platform continuously monitors online resources for events capable of triggering landslides (e.g., major earthquakes), pinpoints regions where landslides are likely to have occurred following such events, and initiates the collection of EO data for these identified areas from public EO data portals. Whenever a new image becomes available, it is downloaded and processed automatically to detect landslide areas. Proximity to cloudy regions is used to determine if a landslide is partially visible or not, and partial extents are marked for further tracking. By combining information from successive analyses, the full extents of landslides are determined. This allows timely first detection of landslides and their effective monitoring under cloudy conditions. The platform allows the integration of various models for landslide detection, ranging from simple index-based approaches (e.g., NDVI) to advanced machine learning and deep learning techniques utilizing image segmentation. The results are published in an open-access landslide catalog, available through a user-friendly web portal for individuals and a REST API for machine access. This catalog is continuously updated and offers faster updates compared to any existing conventional catalog. The platform enables stakeholders, such as researchers, public authorities, and international organizations, to receive notifications when new landslides are detected in their areas of interest. In addition to supporting and expediting rapid damage assessment efforts, the data provided can contribute to landslide prediction initiatives, ultimately enhancing the safety of communities and the built environment.

This presentation offers an in-depth exploration of the design principles and operational framework of the Landslide Hunter platform. It covers the platform's core features, functional capabilities, and user interface, along with a comprehensive overview of the data access methods designed to enhance interoperability and seamless integration with other systems. Furthermore, a live demonstration of the operational platform highlights its practical applications and effectiveness. The demonstration showcases how the platform enables the automatic identification and tracking of landslides without relying on cloud-free optical satellite imagery and how it facilitates near real-time monitoring of landslide evolution, contributing to the global mapping and cataloging of such events.