Sophie de Roda Husman

PhD candidate at Geoscience and Remote Sensing, Delft University of Technology

• 📧: [email protected]
• 🐦: https://twitter.com/SdeRodaHusman
• 🐱: https://www.tiktok.com/@zuidpool_sophie

This repository provides material used for the paper: A high-resolution record of surface melt on Antarctic ice shelves using multi-source remote sensing data and deep learning. The paper offers an elaborate exposition of the model development.

Within this GitHub repository, you will find the following components:

1. Data 📊

The repository includes the UMelt record, which can be accessed through Google Earth Engine as assets or as GeoTIFFs via 4TU.ResearchData. Additionally, a set of simple scripts are provided to aid in the retrieval of the data. The UMelt record is available in two formats: either as a twice-daily resolution or as one product per season, representing the summer melt occurrence.

2. Scripts 📃

These scripts were instrumental in the creation of the UMelt model. The folder contains four scripts that were used to create the UMelt model. Script 1 preprocesses the training, validation, and testing data and exports it as TensorFlow record format in a Google Cloud Bucket. Script 2 develops and trains the U-Net model. Script 3 enables the deployment of the trained TensorFlow model in Google Earth Engine. Script 4 generates UMelt predictions using the trained U-Net model. These scripts collectively provide the necessary functionality for data preprocessing, model development, deployment, and prediction generation in the UMelt model pipeline.

3. Figures 🌌

This repository includes all the figures featured in the paper, including both the main figures and the supplementary figures.