IndLands : A Spatiotemporal dataset for region-aware landslide analysis from Multi-Source Remote Sensing Imagery

This repository contains the complete workflow and supporting files for generating a landslide-prone area dataset using remote sensing and machine learning techniques. The dataset has been prepared for the following Indian states:

Uttarakhand Sikkim Himachal Pradesh Mizoram Maharashtra Karnataka Arunachal Pradesh

The final output is a multi-modal dataset containing terrain, spectral, and texture-based features extracted from satellite data, along with manually annotated landslide zones for training machine learning models.The complete workflow and steps can be accessible from here

IndLands Repository

Structure of the ML Ready Dataset

Each ZIP archive corresponds to a single state and contains multi-modal data derived from remote sensing sources, including:

• Raw Sentinel-2 image tiles
• Digital Elevation Model (DEM) data
• GLCM-based texture features
• Spectral index feature maps
• Spatial subsets
• Manually annotated landslide regions
• dataset.csv, providing a consolidated tabular view of all extracted features, annotations, and associated latitude–longitude coordinates

Note: The full dataset, including intermediate preprocessing outputs generated at each stage of the pipeline, is available upon request.

---