Ecological Network Inference for Resilient Food Systems: A Mathematical Approach

Resilient food systems are required to meet the ever-increasing global demand for food, whilst mitigating against the uncertainties wrought by climate change and preventing the further erosion of biodiversity. Essential to meeting these challenges is the appreciation that agriculture takes place as part of a wider ecosystem. Pesticides can have knock-on effects on a range of non-target species, and that it is the interactions between the different elements of the wider agricultural environment that are critical to understand how the whole system behaves.

An important way of understanding these interactions is as a network of species connected with links that represent ecological interactions such as competition, predation, parasitism. These networks can range in size, but might consist of hundreds or thousands of species, all interacting and contributing to the overall state and functioning of the ecosystem.

The aim of this project is to create new network-based ways of studying agro-ecosystems and their response to climate change and other severe disruptions of the Anthropocene. In particular we seek to answer the following key research questions:
* How can we construct agro-ecological networks over the relevant temporal and spatial scales?
* How can we use these complex networks to inform models of the response of agro-ecosystems to disruption and disturbance?

A fundamental question in ecology is how to construct species-interaction networks from observations. Whilst construction can be achieved through directly observing interactions in the field, this method is time-consuming, localised and does not scale to a more global system. Current approaches to network inference take as their input the co-occurence of species in binary "presence-absence" data across space and time (Volkov et al. 2009). Such approaches are inadequate for several reasons: they are unable to distinguish different interaction types; they can conflate interactions with environmental covariance, and do not properly account for indirect interactions.

This project will develop new network-inference techniques based on the statistical-mechanical principle of Maximum Entropy (MaxEnt). MaxEnt has been proposed in the context before (Volkov et al. 2009, Emary et al. 2021), but my goal is to extend this technique to use time-delayed correlations to correctly reconstruct interaction type and not just the magnitude. This is clearly essential for a proper understanding of ecosystem functioning. After a period of method development I will, through computer simulations, explore the quality of inference in complex ecosystems with limited information gathering. Next I shall demonstrate the applicability of the method to real-world biomonitoring data, using the H2020 EcoStack project, of which NCL is a partner. I will thus construct ecosystem services networks relevant to UK food systems.

Understanding and mitigating how agro-ecosystems respond to the rapid and potentially drastic changes wrought by climate change and species loss is a research priority. I contend that the only way to do this is through a network approach that considers the cascading effects of a disturbance throughout different ecosystem components.