Agenda

Aoibheann Brady, University of Bath, UK

Abstract:
Water-related natural hazards can have tremendous impacts on the well-being of communities; water levels severely below average can bring periods of drought and water scarcity, while those severely above average can be connected to floods. These types of hazards are typically spatial in nature, and it is often helpful to incorporate spatial methods in the analysis of their trends. In this talk, we discuss two spatial methods for the detection of trends in rivers in Great Britain.
The first of these methods proposes the use of a Bayesian methods to investigate the evidence for changes in the magnitude of peak river flows in Great Britain. We model all stations together in a Bayesian multilevel framework to be better able to detect any signal which is present in the data by pooling information across several stations. We include a spatial random effect within the model to account for similarity between nearby stations. A Gaussian process is used to model this spatial variation between gauging stations. This model is then used to estimate trends in peak river flows in the UK.
In the case of a single river with multiple gauging stations, the system can be considered as a network. Learning how river flows evolve throughout a network is beneficial when it comes to accurately estimating the probability at each location of exceeding some threshold for flooding. We investigate an approach to modelling trends in the daily mean flow at each gauging station in a network, which exploits the network structure of rivers in to model the covariance between stations, making use of conditional independence and directed graphs to map out these relationships. As the river is represented as a network, this allows us to fit a conditional autoregressive (CAR) model, which allows for faster inference through its Gaussian Markov random field (GMRF) representation. This in turn will help to accurately estimate trends on the network. This method discussed will be showcased using the network of stations in the river Eden catchments in the northwest of England, which has experienced a series of devastating floods in the last 15 years.

Bio Sketch:
Aoibheann Brady is a PhD Candidate in Statistics at the University of Bath, with a focus on causal methods for environmental studies, and the use of Bayesian multilevel models for the detection and attribution of long-term trends in peak river flows in the UK.