Published May 9, 2025 | Version v1
Dataset Open

RiverScope: High-Resolution River Masking Dataset

Authors/Creators

  • 1. EDMO icon University of Massachusetts, Amherst
  • 2. ROR icon University of North Carolina at Chapel Hill
  • 3. EDMO icon Brown University
  • 4. ROR icon University of Colorado Boulder
  • 5. ROR icon University of Massachusetts Amherst
  • 6. University of Massachusetts

Contributors

Annotator:

  • 1. ROR icon University of Colorado Boulder

Description

You can find a detailed documentation of the data in the following repository: https://github.com/cvl-umass/riverscope

1. Summary

RiverScope is a global-scale, high-resolution satellite image dataset focused on rivers using PlanetScope and co-registered with other sensors SWOT, Landsat (via SWORD), and Sentinel-2. Our dataset contains expert-labeled water segmentation on PlanetScope images to capture fine details that high-resolution imagery better reveals. There are 1,145 images with pre-defined train/valid/test splits that can be used for research on fine-scale water segmentation, width estimation, and future multi-sensor hydrological modeling. The data splits were determined such that each split contains different locations on Earth to avoid data leakage.

For sample models and evaluations that use this dataset, you can refer to the following repository: https://github.com/cvl-umass/riverscope-models

2. Datset Format

Inside RiverScope, there are 4 folders corresponding to each data source: (1) PlanetScope, (2) Sentinel-2, (3) SWORD, (4) SWOT. Each has its own folder. All satellite image files (including the labels) are GeoTIFFs--georeferenced raster imagery.

In the upper level, there are 3 csv files defining the data for each split: train/valid/test. Each csv file contains the following columns:

  1. reach_id (int) - The ID of the reach contained in the image
  2. mid_lat (float) - Latitude of the middle of the tile crop
  3. mid_lon (float) - Longitude of the middle of the tile crop
  4. planetscope_id (str) - Identifier of the PlanetScope image used for the tile. Can be used to download the raw data.
  5. normalized_planetscope_path (str) - Relative path of the normalized PlanetScope data, cropped to 500x500 pixels that matches the same area as the label.(.tif)
  6. label_path (str) - Relative path of the water segmentation label for the given PlanetScope tile. [0=background, 1=river, 2=other water] (.tif)
  7. sword_path (str) - Relative path of the SWORD vector file that contains the nodes and reaches. (.shp)
  8. s2_path_raw (str) - Relative path of the raw Sentinel tile that contains the area captured by the PlanetScope tile (.tif)
  9. s2_path_reprojected (str) - Relative path of the Sentinel tile that was reprojected and cropped to the same area as the PlanetScope tile (.tif)
  10. swot_node_path (str) - Relative path of the node data contained in the given image/area (.geojson)
  11. swot_node_id (str) - Identifier of the SWOT node data, can be used to download the full,raw data
  12. swot_pixc_path (str) - Relative path of the rendered pixel cloud from SWOT (.tif)
  13. swot_pixc_id (str) - Identifier of the SWOT pixel cloud, can be used to download the full,raw data

For the full documentation for each data source, including how to load and view each data type, please refer to the repository https://github.com/cvl-umass/riverscope

Data versions:

  • SWORDv16
  • SWOT PIC0, V1 (for PIXC, RiverSP)

3. References:

Altenau, E. H., Pavelsky, T. M., Durand, M. T., Yang X., Frasson, R. P. d. M., & Bendezu, L. (2021). The Surface Water and Ocean Topography (SWOT) Mission River Database (SWORD): A global river network for satellite data products". Water Resources Research.

Biancamaria, S., Lettenmaier, D. P., & Pavelsky, T. M. (2016). The SWOT mission and its capabilities for land hydrology. In Remote Sensing and Water Resources (pp. 117-147). Springer, Cham.

ESA. Sentinel-1-missions-sentinel online-sentinel online. Eur. Sp. Agency, 2022

Planet Labs PBC. Planet application program interface: In space for life on earth, 2024. URL https://api.planet.com

 

Files

RiverScope.zip

Files (8.1 GB)

Name Size Download all
md5:910f3ffc046cbf60df30c27937555fa5
8.1 GB Preview Download

Additional details

Related works

Is source of
data (Other)

Dates

Created
2025-05-09

Software

Repository URL
https://github.com/cvl-umass/riverscope
Programming language
Python
Development Status
Active