Smart Operational Systems for Disaster Response: A Landslide Monitoring Perspective Dr. Serkan Girgin1,2
[email protected] https://linkedin.com/in/serkan-girgin/ Multi-Hazard Risk Mitigation Symposium 14 October 2025, Ankara, Türkiye Faculty of Geo-information Science and Earth Observation (ITC) 1 Centre of Expertise in Big Geodata Science (CRIB) 2
Effective disaster response depends on the rapid collection, verification, and dissemination of accurate and up-to-date information from the affected areas. Illustration by Storyset.com
The data collected during and after events also serve as a foundation for comprehensive hazard assessment, helping to identify risks, vulnerabilities, and potential impacts. Illustration by Storyset.com
Moreover, these datasets are essential for the development, training, and validation of predictive modeling approaches to improve, e.g., early warning systems. Illustration by Storyset.com
Despite advances in technology, manual data collection and mapping involving numerous organizations, agencies, and individuals remain the predominant method. Illustration by Storyset.com
https://hotosm.org
https://mapping.emergency.copernicus.eu
This approach, while valuable for capturing on-the-ground insights, is typically time-consuming, labor-intensive, and prone to inconsistencies due to varying standards, methodologies, and capacities among participating actors. As a result, coordination and data integration can be challenging, limiting the speed and effectiveness of situational analysis and decision-making. Illustration by Storyset.com
New techniques leveraging Earth Observation (EO) data and Geospatial Artificial Intelligence (GeoAI) are beginning to automate and refine the process. Illustration by Storyset.com
There are several critical challenges involved in operating and maintaining large-scale landslide monitoring platforms. Illustration by Storyset.com Robust Data Pipeline: Managing and processing massive, heterogeneous datasets from satellites, sensors, and crowdsourced inputs. Model Integration and Deployment: Ensuring algorithms maintain accuracy and efficiency when applied to large-scale, high-variety data. Validation and Benchmarking: Obtaining consistent, high-quality reference data for verification across diverse geographic and environmental contexts. User Interface and Accessibility: Delivering rapid, up-to-date analyses while minimizing latency and system downtime. Maintenance: Securing long-term funding, infrastructure, and partnerships among research, operational, and policy communities.
Research Continued innovation in models and data fusion, along with open sharing of methods and datasets. Policy Supportive frameworks for data sharing, sustained funding, interoperability, and long-term system maintenance. Operations Reliable infrastructure, standardized workflows, and integration of research outputs into scalable, real-time systems. Cross-sector collaboration and partnerships are needed to align technological capabilities with real-world needs Illustration by Storyset.com Center of Expertise in Big Geodata Science
The Landslide Hunter platform aims to enable faster detection of landslides minimizing delays from obstructions and to support the testing and comparison of EO-based methods. Illustration by Storyset.com The project "Landslide Hunter: the first fully automated AI-based platform to map and monitor landslides remotely" with file number OCENW.XS23.3.145 of the research programme NWO Open Competition Domain Science XS is financed by the Dutch Research Council (NWO)
The platform monitors online sources for events that could potentially trigger landslides and identifies areas where landslides are likely to have occurred
Preand post-event EO images are collected and analyzed to detect visible landslide extents using various landslide detection methods
The Landslide Hunter uses a three-step approach to address the challenging task of accurately determining landslide extents by integrating incomplete and often unreliable information Illustration by Storyset.com
In the first step, model outputs are categorized into cell-based semantic classes and associated certainty indicators Model Inference Landslide Not Landslide Model Probability High Low Low High Occlusion No Yes No Yes No Yes No Yes Occlusion Probability High Low Low High High Low Low High High Low Low High High Low Low High Classification L! L L A L L? L? A B BBU B+ B B U Classes Unknown Anomaly Not Landslide Landslide U A B! BB? L? L L!
In the second step, a rule-based decision-making process assigns a single representative class to the time series of each data cell R1. Immediate Repeat R2. Majority Vote R3. U Dominance
Finally, partial extents are marked for further tracking until complete landslide coverage is obtained through successive analyses
This enables the timely first detection and effective monitoring of landslides, even under obstructed conditions* (e.g., cloudy weather). Illustration by Storyset.com
