Uncertainty, as an unavoidable artifact of modeling and observation of physical processes, should be assessed in the context of prediction, analysis, simulation, and decision making. Determining and characterizing uncertainties propagated across scales and domains are critical to design reliable systems.
Two types of uncertainties are recognized. One is the inherent randomness, called aleatory uncertainty, and the other is due to lack of perfect knowledge about the system, called epistemic uncertainty.
Aleatory uncertainty comes from pure randomness and natural fluctuation, whereas epistemic uncertainty in modeling and simulation has several sources, including information conflicts from multiple sources, inconsistent beliefs among experts’ opinions, lack of data, lack of time for introspection, measurement error, lack of dependency information, truncation errors during numerical treatments.The reliability of predictions from models depends on how sensitive the predictions are with respect to model-form uncertainty, the major component of epistemic uncertainty.
The robustness of decision making under uncertainty relies on how sensitive the decision is with respect to risk and ambiguity. Risk is characterized as the perceived chance that something happens, whereas ambiguity represents the lack of knowledge on the chance. Thorough analysis and evaluation of risks provide important information to decision makers.