Climate is what you expect, while weather is what you get. Numerical Weather Prediction (NWP) has been the main computational tool for forecasting the weather, and is based on mathematical modelling and a physical understanding of the atmosphere to predict its evolution. These predictions are accurate for small time scales, but become more unreliable for longer periods due to the chaotic nature of the moisture processes within the atmosphere. As Artificial Intelligence (AI) has exponentially exploded, it has also made its way into the field of weather prediction. AI-based Weather Prediction Models (AI-WPM) consider data-driven approaches, by analysing a corpus of 40 years of hourly atmospheric analysis, and not relying on any physical laws to arrive at their predictions. These models are magnitudes faster than traditional NWP; however, their accuracy is still below satisfaction at a global scale. Additionally, the black-box nature of these methods is undesirable as these modelling techniques have no explainability or physical limitations in their predictions. This project aims to take the best from the physics-informed NWP processes and AI-WPM in order to produce models capable of enhancing the predictions of NWP models, while matching the speeds of AI-WPM. These models will pertain a higher level of transparency in comparison to purely data-driven AI-WPM by embedding governing equations and physical laws within the models explicitly and/or implicitly. This increase in transparency is projected to increase trust in stakeholders, potentially allowing AI-WPM to be used publicly for weather prediction. This project is of great interest to specialists at the Met Office, with the aims and objectives of this PhD being developed with the Met Office. The project greatly benefits from the input from their researchers (particularly Dr C. Morcrette) through their expertise on numerical weather prediction, including access to existing software for solving primitive equations and relevant weather data. The models produced during studies and their respective findings may be used by the Met Office to develop new systems, or improve on existing ones.
Machine Learning, Deep Learning, Transfer Learning, Uncertainty Quantification, Numerical Methods, Partial Differential Equations, Continuum Mechanics, Climate Modelling, Feature Engineering, Physical Constraint Integration, Data Assimilation, Hybrid Models, Explainable AI (XAI) For Weather Models.
Graduated from University of Birmingham with a Class I MSci in Mathematics.
Undergraduate Masters dissertation focused on using Non-Linear Feed-Forward Neural Networks to solve Partial Differential Equations.
Dr Lisa Maria Kreusser
Dr Xi Chen
Prof Corwin Wright
Dr Cyril Morcrette (University of Exeter)
