Background
Extreme rainfall events are becoming more frequent and intense, posing major risks to communities, infrastructure, and economies. Traditional weather models often struggle to capture their complex spatial and temporal patterns, especially under a changing climate. This project focuses on advanced statistical modelling, specifically spatio-temporal generalised linear/additive mixed models, to improve prediction of when and where extreme rainfall will occur. These models incorporate environmental and geographical factors to adapt dynamically to local conditions, providing more precise and actionable forecasts. Yet, critical gaps remain in how thresholds for “extreme” events are set and how predictions can adapt to changing climate conditions. This research will push the boundaries of climate modelling by developing adaptive, covariate-driven threshold methods, and improving forecast reliability.
PhD opportunity
Extreme rainfall events are among the most damaging climate-related hazards, driving floods, infrastructure losses, economic disruption, and human displacement. Yet, near-term projections of their occurrence and intensity remain highly uncertain, particularly under climate change. Current statistical models often fail to incorporate both the spatio-temporal structure of rainfall and the dynamic nature of extreme-event thresholds, limiting their predictive value. This PhD tackles these challenges by developing novel spatio-temporal Generalised Linear/Additive Mixed Models with covariate-dependent, dynamically estimated thresholds for extreme event prediction.
Within the UNRISK framework, this project directly addresses the pressing need for rare-event modelling under deep uncertainty, as identified in UK Climate Resilience priorities. It will integrate environmental covariates, climate drivers, and socio-geographical features to produce location-specific, time-varying thresholds, allowing more accurate forecasts for regions most at risk. Methodologically, it will advance computational algorithms for threshold estimation, enabling scalable application to petabyte-scale climate datasets.
The research will combine statistical modelling, machine learning, and high-performance computing, aligning with UNRISK’s emphasis on advanced data science to handle both “too much” and “too little” data in climate prediction. Outputs will include a computational package for potential integration into UKCP18.
This project offers the candidate the opportunity to work with leading academics in the University of Leeds and the UK Met Office. Outcomes will not only advance methodology but also deliver actionable insights for climate adaptation, disaster risk reduction, and infrastructure resilience, critical to mitigating the multi-billion-pound risks associated with climate uncertainty.
Applicant Profile
Students with a strong background in mathematics or statistics. Some degree of familiarity with computational packages in R is required. Understanding of weather or climate systems is not required as this can be learned during the project.
Other information
Davison, A.C. & Smith, R.L. (1990). Models for exceedances over high thresholds. Journal of the Royal Statistical Society: Series B, 52(3), 393–442.
Davison, A.C., Padoan, S.A. & Ribatet, M. (2012). Statistical modeling of spatial extremes. Statistical Science, 27(2), 161–186.
Cooley, D., Nychka, D. & Naveau, P. (2007). Bayesian spatial modeling of extreme precipitation return levels. Journal of the American Statistical Association, 102(479), 824–840.
