Iceberg Segmentation Data

This repository contains (1) Image patches and masks used to train, validate, test UNet++ models for delineating icebergs in Greenland ice mélange. The image patches and masks were extracted from Sentinel-1, and Sentinel-2 data, to train two separate UNet++ models for each dataset. The train_validate_test folder contains pickle files that can be read as numpy arrays (images, masks that were split for train/validate/test). The relevant Python code is located here: https://github.com/hfisser/IcebergSegmentation. (2) Delineations of sub-regions used in the paper. (3) Terminus positions (points at the center of the glacier) extracted from the Automated Greenland Glacier Termini Position Time Series, Version 1, dataset. (4) Iceberg outlines analyzed in the Worldview-Sentinel-2 validation.