Leveraging Machine Learning algorithms for improved Arctic sea-ice prediction using the Met Office suite of models

This project leverages advanced satellite remote sensing and machine learning (ML) to improve Arctic sea-ice forecasting. Traditional sea-ice models often develop significant biases in summer, particularly near the ice edge, limiting their accuracy. Additionally, these models are computationally expensive, restricting the frequency of forecast updates.

The goal is to apply ML techniques, such as deep learning, to enhance sea-ice prediction accuracy and efficiency. ML emulators can learn error-correction patterns from historical model outputs and observational data, including satellite-derived sea ice thickness (SIT) maps from summer months. This approach helps address summer biases and supports the initialization of climate models, improving their ability to overcome the "Spring Sea Ice Predictability Barrier."

Furthermore, the project aims to reduce computational costs by using ML-based emulators to approximate complex sea-ice models. This could allow for faster and more frequent forecast updates, improving real-time decision-making for Arctic operations and climate modelling. By combining physical understanding with ML, the project builds on recent advancements to revolutionize sea-ice prediction capabilities.