An automated, regional-scale rapid mapping tool for multiple landslide events: testing a proof-of-concept case study from New Zealand 2023.

Multiple, concurrent, landslide events occur on a regional scale due to severe weather and earthquakes. During disaster scenarios, data latency (the time between the event occurring and spatial data being made available to users) can be several hours to several days depending on satellite return paths, the route they take, image quality, processing and interpretation time and communication by specialists. Freely available spaceborne Earth Observation data products are increasing exponentially, providing the opportunity to exploit them for the rapid generation of a global or regional scale landslide inventory. Social sensor data allow access to a rich source of human information such as text, videos, photographs, timestamps and coordinates and can report disaster information quicker than media outlets and conventional observatories. These data, while inherently imperfect, provide near-real-time information in large quantities and at spatial densities that can exceed conventional observations across vast areas. Our work tests a proof of concept to combine complementing data from social media and satellite Earth observations in order to focus automated landslide data collection and mapping in real time. We take the case study in New Zealand where an estimated 5,000 landslide events occurred in February 2023 during and after Cyclone Gabrielle. We describe: (1) the challenges with and use of machine learning to continuously mine, in real-time and on the global scale, social media for landslide-related content and images and (2) the use of a CNN (Convolutional Neural Network) for mapping the failed slope using satellite data. This work highlights that data latency in disaster scenarios could be lessened considerably through the combination of automated tools, providing information in real-time on the worldwide scale.