Skip to content

Landslide4sense

Back to geoai datasets

Field Value
Description Dataset Description
This dataset, originally introduced in the Landslide4Sense-2022 GitHub repository, contains multispectral and elevation data for landslide detection. It consists of three splits: training (3799 patches), validation (245 patches), and test (800 patches). Each patch is a 128x128 image composed of 14 bands combining Sentinel-2 multispectral channels (B1–B12), ALOS PALSAR slope (B13), and ALOS PALSAR digital elevation model (B14), all resampled to ~10m spatial resolution.

Dataset Structure
Annotations and images are provided in HDF5 format across train, validation, and test folders, each with corresponding image_X.h5 and mask_X.h5 files. Masks provide pixel-wise binary labels (0: Non-landslide, 1: Landslide).

Annotation Description
Mapping classes: 0 = Non-landslide, 1 = Landslide.

Purpose
The dataset supports benchmarking of geospatial machine learning models for pixel-wise landslide segmentation and environmental hazard analysis.
Folder /datasets/geoai/ibm-nasa-geospatial/Landslide4sense
Discipline GeoAI / Remote Sensing / Environmental Science
DOI
Link Access Data
Public True
Publication Date 2025-02-11
Downloaded 2025-09-10
Data Type HDF5
Dataset Size 23G
Number of Files 17500
Usage 
$ module avail
$ module load datasets
$ module load geoai/ibm-nasa-geospatial/Landslide4sense/2025-02-11
Usage Policy Link https://choosealicense.com/licenses/mit/
Usage Policy This dataset is released under the MIT License
.
The MIT License allows use, modification, and distribution of the dataset for both research and commercial purposes, provided that the original copyright notice and license text from the Institute of Advanced Research in Artificial Intelligence (IARAI) are retained. Users should cite the Landslide4Sense dataset and acknowledge the IBM–NASA Geospatial team when using it in derived works, analyses, or models.
Citation IBM NASA Geospatial. (2024). Landslide4Sense Dataset (v1.0) [Dataset]. Hugging Face. https://huggingface.co/datasets/ibm-nasa-geospatial/Landslide4sense
BibTeX
📜 View BibTeX citation
@dataset{ibm_nasa_landslide4sense_2024,
  title        = {Landslide4Sense Dataset (v1.0)},
  author       = {IBM NASA Geospatial},
  year         = {2024},
  howpublished = {\url{https://huggingface.co/datasets/ibm-nasa-geospatial/Landslide4sense}},
  note         = {Available on Hugging Face Datasets}
}