Cassava Brown Streak disease (CBSD) is a viral disease of cassava that causes necrosis of the edible root tissue, which reduces both consumable and marketable yield. In 2004, CBSD emerged in Uganda and has since been spreading rapidly through previously unaffected regions of East Africa and into Central Africa. Preventing spread to West Africa is a major food security and development priority, along with mitigating the impact of CBSD in endemic regions. This thesis focuses on the development of a landscape-scale spatial model of the CBSD epidemic to inform management.

Currently, there is disparate information on the epidemiology of CBSD and significant associated uncertainty. We begin with a review of CBSD from an epidemiological perspective. The review focuses on: mechanisms and rates of pathogen dispersal, surveillance, disease impact and management efficacy to inform the structure of the CBSD model.

Prior to model development, it was necessary to aggregate all available data on the historic spread of the epidemic. Minimal surveillance data were available in the literature. Therefore, it was necessary to work extensively with East African collaborators to acquire and digitise over 10 years of previously unavailable surveillance records from Uganda and surrounding countries. Extensive post-processing was performed to minimise errors in the data. In parallel with digitisation of the surveillance data, we describe work to enable digital data collection via the creation of a cassava disease surveillance app, along with extensive training. The goal was to minimise errors in data collection and reduce the time lag between disease surveillance and reporting in surveillance programmes.

The second section of the thesis describes the development, parameterisation, and validation of a stochastic, spatio-temporal epidemic model for CBSD. Using digitised Ugandan surveillance data from 2005-2010, and estimates of cassava density throughout Uganda and immediately surrounding regions, we apply Approximate Bayesian Computation (ABC) to estimate dispersal parameters, providing methodological details on the development and validation of summary statistics. The model fitting also takes account of empirical data for vector density across Uganda and surrounding regions. The model fits the data well for the training set for 2005-2010. Survey data from Uganda and the surrounding region from 2011-2017 are then used as a rigorous independent test to validate model predictions.

The third section of the thesis describes the application of the model to address questions concerning historic, current and future epidemic spread. We use the model to identify reasons why, although there were historically high levels of CBSD infection in Malawi, negligible epidemic spread occurred into Zambia from Malawi showing that low density of cassava cultivation in south east Zambia could account for the inhibition of spread. The model does successfully predict the incursion of the epidemic into north east Zambia from the Democratic Republic of Congo (DRC). We run cross-continent simulations to predict the spatiotemporal spread of the epidemic through central Africa, including DRC and the Central African Republic, where there is very little disease surveillance and reporting for CBSD. The simulations allow us to compute the likely distributions of arrival times of the epidemic in West African countries. We also simulate rates of spread of the disease in West African countries following direct introduction for example by importation and by natural spread from adjoint countries. Finally, we simulate management interventions in Nigeria, to identify the scale and speed at which management programmes would need to be deployed to contain the epidemic.

The thesis concludes with a review of the principal results and critical assumptions underlying the results. Some proposals are presented for future work in epidemiological modelling to address practical problems of the management of CBSD.