“Landslides, triggered by natural events like heavy rainfall and earthquakes, pose significant risks to lives, infrastructure, and the environment. Effective monitoring and mapping of landslides are crucial for mitigating these risks, guiding emergency responses, and supporting resilient infrastructure planning

Using multi-source satellite data, you will work to create an accurate landslide detection model. This model should differentiate landslide-affected areas from unaffected regions, leveraging both optical imagery and Synthetic Aperture Radar (SAR) data. The provided datasets, sourced from Sentinel-1 and Sentinel-2, include RGB and near-infrared bands as well as SAR bands (VV and VH) captured pre- and post-event. These data can reveal landscape changes and offer a unique view of the terrain, combining visual and radar-based insights to detect surface alterations and other indicators of landslides.

While optical data is precise and interpretable, SAR data is invaluable in cloud-covered regions. Combining these datasets can improve detection accuracy, particularly in challenging conditions.

The objective of this challenge is to create a model that effectively leverages SAR for cloud-covered regions while prioritising optical data where available, enabling accurate and reliable landslide detection.

You also need to adhere to trustworthy AI guidelines to create solutions that extend beyond performance metrics, ensuring models are transparent, ethical, and impactful, with tangible benefits for society and disaster resilience efforts.”

Link to challenge and news item this week from the University of Cambridge in comments.

https://www.esc.cam.ac.uk/news/using-ai-see-landslides-and-target-disaster-response

https://zindi.africa/competitions/classification-for-landslide-detection