Attribution of large-scale drivers for environmental change

Abstract

In this thesis, spatial multilevel models are developed within a causal framework for the attribution of long-term changes in environmental studies to large-scale drivers of interest. In particular, these methods are applied to river flows in this thesis. Two common themes are apparent throughout the thesis. The first involves the accurate detection of long-term changes in river flows in Great Britain. The development of spatial multilevel methods for the accurate detection of trends is a major focus of the thesis. We firstly focus on the detection of countrywide trends in which the correlation between stations is modelled through a Gaussian process. Using a multilevel approach allows for the pooling of gauging station data to accurately estimate trends in peak river flows that may not be possible using traditional at-site analyses. This approach resulted in a first detection of trends in peak river flows at a countrywide level in Great Britain. We then switch in focus towards the analysis of a single river network, exploiting this network structure in order to understand how flows in a given region will evolve. This method considers the river gauging station measurements on the network as a graph, which is comprised of nodes (the river gauging stations themselves) and edges between nodes (the direction of ow between each station). The network structure of this river is encoded using a first-order conditional autoregressive (CAR) model. This method allows for the use of fast inference methods through the construction of a sparse precision matrix, and also respects the physical structure of the network.

The second theme focuses on the development of a causal framework for the attribution of long-term, large-scale changes in environmental studies to some climate drivers of interest. We first perform a preliminary attempt at attribution, where a clear association is seen between the East Atlantic (EA) index and peak river flows, even when a multivariate approach is used to account for temporal confounding. We then provide a more rigorous approach towards the attribution of long-term, large-scale drivers of environmental change. A systematic checklist is developed to provide a thorough causal assessment of drivers of environmental changes, demonstrating through this checklist that changes in peak river flows can be attributed to the EA index.

Date of Award	19 Feb 2020
Original language	English
Awarding Institution	University of Bath
Supervisor	Ilaria Prosdocimi (Supervisor) & Julian Faraway (Supervisor)