Policy attribution based on data-driven control

The Water-Energy-Food (WEF) Nexus is a highly complex system that is difficult and time-consuming to construct models for. Yet these models are often necessary to assess how impacts from different control variables—such as policy-, population-, climate-, and land use changes—propagate through the WEF Nexus. Here we look at this problem through the lens of control systems theory, where control refers to how we manipulate or disturb the system.

Dynamic Mode Decomposition (DMD) and its many variants have shown promise in modelling ever-more complex dynamical systems based entirely on measurement data. In this study, we specifically look at DMD with control (DMDc) and evaluate how it may be used to assess policy impacts within the WEF nexus based entirely on data. DMDc takes state data and control input data and constructs a linear dynamical system. The resulting model allows for easy interpretation and manipulation of the control variables to investigate how the system evolves under e.g., different policies.

To assess the performance of DMDc for WEF Nexus purposes, we use an existing high-dimensional System Dynamics Model (SDM) of the WEF nexus to generate different policy-driven data scenarios, which are used to benchmark the DMDc method. DMDc shows promise at reconstructing and attributing the impacts of policies based on data generated by the SDM, despite the data being extremely underdetermined. However, there are several caveats and questions that remain, which we believe newer DMD variants may address.