Improving Past Climate Reconstruction Using Machine Learning

This PhD project at the University of Reading focuses on advancing past climate reconstruction using state-of-the-art machine learning techniques. Understanding historical climate states is essential for deciphering the mechanisms and rates of climate change. Traditional climate reanalyses, which are based on weather forecasting methods, often suffer from abrupt shifts due to the limited incorporation of observational data. Kalman Smoothing methods, which can utilize all available data (including future observations relative to the analysis time), offer a promising alternative but have been underutilized in climate science due to computational constraints. With the rapid development of machine learning algorithms, this project aims to make Kalman Smoothing and related data assimilation methods more accurate and computationally efficient. The research will leverage numerical weather prediction data and integrate observations from the atmosphere, oceans, cryosphere, and land surface. A key aspect will be representing and estimating uncertainties in both the assimilated observations and the resulting climate reconstructions. Students will have the opportunity to develop and test new methods on idealized computational models before applying them to large datasets from the Met Office and ECMWF. The project is embedded within the data assimilation research centre at Reading, offering collaboration with experienced scientists and potential engagement with the Met Office (as part of their CASE studentship scheme) and ECMWF. This position is ideal for candidates with a strong background in mathematical physics and computational skills, particularly those interested in optimisation methods or with prior experience in machine learning. The project is supported by a full UKRI stipend and home-level PhD tuition fees, and is associated with the EPSRC Centre for Doctoral Training in the Mathematics for our Future Climate. References for further reading include recent publications on ocean reanalyses and Kalman smoother algorithms, highlighting the scientific foundation and ongoing advancements in this field. Applications are accepted year-round, and interested candidates are encouraged to contact the supervisors for more information and guidance on the application process.