Department Systems Analysis, Integrated Assessment and Modelling

Feature Extraction From High Dimensional Noisy Datasets

This project explores the use of machine learning techniques to uncover low-dimensional features within high-dimensional datasets, both simulated and observed. Our aim is to identify relevant features that provide insight into the underlying parameters for simulated datasets, improving Bayesian inference methods such as ABC and HMC. In the context of catchment hydrology, the focus is on removing the impact of meteorological drivers such as precipitation and temperature in order to uncover the unique fingerprints of a catchment in runoff data.