IIT Delhi Develops AI Tool for Mapping Landslides

New Delhi- A team of researchers from the Civil Engineering Department at IIT Delhi has developed and made publicly available a cloud computing and machine learning-based tool for mapping landslide extent using satellite data. Named ML-CADE, the tool requires only two inputs, including the approximate date and location of a landslide event, and it can accurately map a complex cluster of landslides within 5 minutes and a simple landslide within 2 minutes.

The researchers trained the underlying model on a large amount of satellite, terrain, vegetation, and soil data. They say that the traditional methods of mapping landslides by manually digitizing satellite imagery are labour intensive, time-consuming, and costly. This is not feasible in large and remote areas. They also found that existing simple models developed using vegetation indices are not effective in areas with minimal vegetation. This is where machine learning comes into the picture.

The proposed tool is based on 19 features, including pre- and post-landslide transitions, slope and aspect of terrain, Normalized Difference Vegetation Index (NDVI), and differential Bare Soil Index, which can identify fresh landslide development. The decision-tree-based model allows for both pixel and object-based segmentation. It is powered by Google Earth Engine and does not require any pre-trained models, making it highly accessible and adaptable to local terrains.

The researchers say that the tool has the potential to be used by disaster managers in the early warning of landslides. It can also be used for mapping flood inundations, deforestation, sand mining, and other environmental challenges. The team is currently working on developing a national historical landslide inventory with spatial extents that will be invaluable for developing landslide early warning systems.