General Remarks on Robust Solutions

Abstract

We summarize here the background and key concepts related to robust solutions in the context of supporting decision-making for problems characterized by deep uncertainties, which also were in the focus of the previous workshops on Coping with uncertainty, see, e.g., [3]. Although such problems are fundamentally different from statistical decision models, yet basic ideas of robust statistics are applicable to methods supporting robust decision-making under uncertainty. The main new issues are concerned with a proper representation of uncertainty, and its interactions with decisions. In particular, a key issue is the sensitivity of robust decisions with respect to low probability catastrophic events, that are of critical importance for analyzing global change problems. Robust decisions for problems exposed to extreme catastrophic events are essentially different from over-simplified decisions that ignore such events. Specifically, a proper treatment of extreme/rare events requires new paradigms of rational decisions, new performance indicators, and new spatio-temporal dimensions of heterogeneous interdependencies including network externalities and risks. This, in particular, needs new approaches to downscaling, upscaling and discounting. Global change processes, in particular climate change, involve inherently unpredictable complex interactions between natural and human-created systems therefore proper modeling of these processes must rely on adequate treatment of uncertainties, and their effects on human’s decisions. Traditional natural science models are based on relations whose validity is estimated from repetitive experiments and observations. If experiments do not affect the underlying relations, then repetitive observations allow to derive them by using the statistical decision theory. Unfortunately, human-created processes do not follow fixed relations. Elements of these processes change their dimensions and structure. For example, introduction of new technologies may increase or reduce uncertainties, risks, critical thresholds and discontinuities. Exact identification of global climate change processes is impossible because such processes are non-stationary, have delayed responses, and human or natural actions may have catastrophic irreversible consequences.