Modelling The Water-Energy-Food-Ecosystem Nexus Using Data-Driven Methods

Attaining resource security in the water, energy, food, and ecosystem (WEFE) sectors is paramount in order to fulfill many of the sustainable development goals. To obtain a holistic understanding of this WEFE nexus and assess the impacts of policy decisions, climate change, and other interventions, a system dynamics approach to modelling has been encouraged. Due to the multiscale, nonlinear nature of this nexus and a recent data deluge in the WEFE sectors, we propose the use of data-driven methods, which rely on dimensionality reduction and machine learning algorithms to find low-rank patterns and parsimonious models in big data sets. As these methods have proven highly successful within other scientific disciplines, we evaluate the prospect of using a data-driven approach to address key issues in nexus research based on analogous case studies from these disciplines. Specifically, we address three key issues with nexus modelling: model discovery, extreme events, and scenario analysis. We first consider how to identify nonlinear dynamical equations from chaotic, noisy, multiscale, and variable-deficient measurements using algorithms like SINDy and HAVOK, which have already been employed on a multitude of physical, biological, and chemical systems. We then investigate how to model the cascading impacts of extreme events on the nexus based on data-driven models of disease outbreak, as well as rapid model recovery after an unprecedented extreme event. Finally, we look into how to employ data-driven control, using case studies in flight control and drug intervention modelling, for the purpose of assessing how policy decisions, climate change, or population growth may be evaluated when modelling the nexus. This can provide tools for stakeholders to see what systemic impacts their decisions might have, and how they can attain synergies between the WEFE sectors. While many studies have conceded that nexus modelling is highly individualized based on the selected region, sectors, and data availability, this overview highlights that a generalized systematic approach to nexus modelling may still be possible, despite these challenges.