Unet

Semantic Segmentation (Technical)

The process of assigning a class label to every pixel in an image to delineate supraglacial hydrological features.

Semantic Segmentation (Communication)

A method of categorising a satellite image pixel by pixel so that water bodies can be clearly separated from the surrounding ice.

Attention-Based U-Net (Technical)

A deep learning machine learning model implemented in Python using TensorFlow and Keras designed for precise pixel-level classification by focusing on the most relevant spatial features.

Attention-Based U-Net (Communication)

A highly targeted computer program trained to automatically identify and outline specific water features by 'paying attention' only to the most important parts of a satellite image.

Sentinel-1 SAR Imagery (Technical)

Synthetic Aperture Radar data used to penetrate cloud cover and distinguish between different surface types based on backscatter intensity for automated drainage detection.

Sentinel-1 SAR Imagery (Communication)

Satellite imagery that uses radar to see through clouds and darkness allowing for continuous monitoring of the Greenland Ice Sheet regardless of the weather.

Custom Data Pipeline (Technical)

An engineered automated workflow developed to process raw satellite data and feed it into Scikit-Learn and TensorFlow models for training and prediction.

Custom Data Pipeline (Communication)

A streamlined system built to automatically collect, organise, and prepare raw satellite images so the artificial intelligence can analyse them quickly and efficiently.

ESRI Story Map (Technical)

A non-technical output leveraging interactive web mapping to translate complex machine learning predictions and geospatial data into accessible narratives for stakeholders.

ESRI Story Map (Communication)

A visual and interactive presentation designed to explain the key findings of the project in simple, engaging terms for non-specialists.