Robust Resilience: Decision Making with Ensemble of Metrics Improves Response to Novel Shocks

In exploratory model-based decision support for socio-environmental resilience, performance metrics must be specified to value the desirability of alternative outcomes. However, resilience is not a well-defined property, but a multidisciplinary boundary object, for which many different metrics have been proposed. This introduces uncertainty into the decision-making process. The consequences of this are unclear, but may be significant in light of the societal, environmental and economic costs of resilience-building interventions. We study the possible effects of this metric uncertainty by exposing a generic agri-food model to various exogenous shocks, and quantifying its response using a variety of conceptually comparable, yet technically different resilience metrics. We find there is only limited agreement between the metrics regarding the supposed resilience of the system, and the policy interventions useful for improving its response to diverse shocks. We link this lack of agreement to the different aggregation methods in the metrics. However, it is possible to move the studied system into a safe operating space which is limited in performance, but robust across resilience metrics, by means of many-objective optimization. In this operating space, the studied system also performs favorably when experiencing previously unseen shocks, compared to spaces optimized for a single metric. We propose a tentative framework for specifying ensembles of resilience metrics to use in metric-robust decision making. Future work might consider what robustness means across an ensemble of metrics capturing the same boundary object, the burdens placed on many-objective optimization by such ensembles, and whether metric uncertainty could also be addressed in other phases of decision making under deep uncertainty.