ice floe validation dataset

The Ice Floe Validation Dataset is a collection of randomly sampled 100 km by 100 km MODIS scenes from the Aqua and Terra satellites. This repository contains the dataset itself along with the notebooks and scripts used to produce the dataset.

contents

data

The data directory has subdirectories

1. /metadata containing specification/config files
2. /modis containing truecolor, falsecolor, landmask, and cloudfraction images
3. /masie containing images from the MASIE 4 km data product, cropped and aligned with the MODIS projection via nearest-neighbors interpolation
4. /validation_dataset containing the binary PNG, labeled TIFF files, and Photoshop files. The subdirectories are
   • binary_floes
   • binary_landfast
   • binary_landmask
   • binary_water_samples
   • labeled_floes
   • labeling_psd_files
   • property_tables

The files all follow a strict, informative naming convention.

1. MODIS and MASIE images: NNN-region_name-100km-YYYYMMDD.satellite_name.image_type.250m.tiff NNN is the zero-padded, three digit case number. Each case specifies a date and location, and contains two satellite images (Aqua and Terra). MODIS image types are cloudfraction (RGB), truecolor (Bands 1-4-3), falsecolor (Bands 7-2-1), and landmask. MASIE images have masie instead of satellite, since there's only one MASIE image each day, and have image types seaice and landmask.
2. Validation images: NNN-region_name-YYYYMMDD-satellite-image_type.file_type. Image types with file type png are binary_floes, binary_landfast, binary_landmask. Labeled floe masks are tiff with image type labeled_floes and have integer labels corresponding to the labels in the property tables.

Finally, the /property_tables contains the main metadata table with all case information and manual image assessments ice_floe_validation_dataset.csv as well as automatically generated property tables for each image with ice floes in it, and a table of hybrid automatic/manual floe matching for paired analyses. Passtimes for each image identified using the Satellite Overpass Identification Tool and NASA Earthdata Worldview are included.

notebooks

1. bookkeeping.ipynb Scripts and diagnostics to check the completeness of the datasets.
2. sample_selection.ipynb Code for iterative random sampling

scripts

1. setup_locations.py
2. calculate_sea_ice_fraction.py
3. binarize_landmasks.py
4. extract_features_and_pair_floes.py
5. plot_locations.py
6. plot_sif_climatology.py

figures

The figures directory has diagnostic and explanatory figures, including the sea ice fraction climatology, the map of analyzed cases, and a summary of the metadata.

citation

The dataset will be described in an upcoming paper in IEEE Transactions on Geoscience and Remote Sensing "Automated segmentation and tracking of sea ice floes from optical remote sensing imagery". If you use the dataset in your work, please cite the dataset as

Watkins, D.M. (2025) "Ice Floe Validation Dataset", Zenodo, doi: 10.5281/zenodo.17716713

The Zenodo DOI will always link to the newest version of the dataset.

acknowledgments

Creation of the dataset was supported by grants from NASA and the US Office of Naval Research, under the direction of Dr. Monica Wilhelmus wilhelmuslab.me. Contributors to the data validation and labeling include Ellen Buckley, Minki Kim, Emma Slaght, and Ethan Register.

We acknowledge the use of the NASA Worldview Snapshots tool for MODIS data access.