The Classification for Landslide Detection challenge that ran from April to August of 2025 answered this question. Nearly 1,000 participants from 90 countries, who submitted more than 8,600 models using multi-source satellite data to tackle a major challenge in Earth observation: accurately detecting landslides from space, even through cloud cover.
This challenge was not only about building an accurate model. The Global Initiative to develop and promote international standards for AI in disaster management. Typically, AI challenges appeal to the competitive nature of developers. As a result, many participants focus on topping the scoreboard rather than following best practices. This raises a critical question: how do we encourage AI developers to embed best practices when creating solutions?
We designed this challenge to bridge innovation with standardization. In addition to leaderboard rankings, participants were required to document how their best-performing solutions aligned with AI best practices, focusing on:
These best practices were drawn from the Technical Reports produced by the ITU/WMO/UNEP Focus Group on AI for Natural Disaster Management, the predecessor project to the Global Initiative. Conducting this step after leaderboard scoring preserved experimentation while reinforcing internationally agreed guidelines. It was inspiring to see how the winners went beyond performance metrics and thoughtfully considered AI standards, demonstrating an integrated approach that aligns with real-world requirements. The winners also went ahead to develop AI solutions that consider adoptability and reusability with limited redesign for other disasters, not just landslides.
Following international standards makes AI solutions more scalable and interoperable with existing systems. Yet standards are often introduced top-down, through policies that many developers only encounter if they move into operational use. We believe that everyone developing AI-based systems–whether for research or operational use–should be aware of international standards and try to integrate them into their workflow. By embedding standards in an online competition, Zindi and ITU introduced them directly to a global community of AI developers. This bottom-up approach encouraged nearly 1,000 participants to integrate best practices into their workflows - helping to build AI solutions that are not just accurate, but also transparent, ethical, and impactful, with tangible benefits for society and disaster resilience efforts.
This challenge was run in partnership with the International Telecommunication Union (ITU) AI for Good platform, the European Space Agency (ESA), MISSLab (University of Padova), ComHaz (University of Cambridge), and the World Meteorological Organization (WMO), and coordinated by the Global Initiative on Resilience to Natural Hazards.
Thank you to our authors and challenge hosts, Lorenzo Nava, Jennifer Selby, Chester Karwatowski, Rakiya Babamaaji, Katharina Weitz, Monique Kuglitsc and Silvia Raquel Garcia Benitez