This project focuses on developing methods to improve extreme weather forecasting for flood prediction. The ultimate goal is to mitigate the human impact of flooding by advancing deep-learning approaches that enable more accurate and reliable water resources management in vulnerable regions.
Traditional process-based models are limited in their ability to capture the complex, non-linear dynamics that govern flood systems. Furthermore, real-world environmental datasets introduce challenges, such as noisy sensor measurements and data uncertainty that degrade predictive performance. Conversely, while modern deep learning approaches offer powerful predictions, they often lack interpretability for high-stakes decision-making.
To address these interconnected challenges, this project will develop advanced machine learning frameworks by exploring several methodological directions: i) integrating heterogeneous data sources to leverage complementary information across diverse data modalities, ii) modelling uncertainty through probabilistic and uncertainty-aware learning techniques, iii) bridging physics-based and data-driven frameworks to build models that are both physically consistent and data-adaptive, iv) investigating explainable artificial intelligence to improve transparency and the understanding of model behaviour, v) supporting broader spatio-temporal forecasting tasks.
Machine learning, Deep learning, Time-series forecasting, Signal processing, Hydrological-Hydraulic modelling, Deep Multimodal fusion, Uncertainty quantification, Extreme predictions, Explainable AI, Extreme distributions, Explainable Anomaly Detection
Worked 3 years as an Analytics and Software Developer, Arcadis, UK
MSc in Data Science, Newcastle University, UK
BSc in Physics, Aristotle University of Thessaloniki, Greece
Dr Georgios Exarchakis
Prof Mike Tipping
Dr Thomas Kjeldsen
