Using Data-Driven Uncertainty Quantification to Support Decision Making

As data collection and analysis methods become increasingly sophisticated, interpretation and use of results by end users become increasingly challenging. In this paper, we discuss the role of data-driven uncertainty quantification in supporting and improving decision making. We illustrate our argument with a case study in seismic onset detection, comparing statistically computed distributions over possible signal onset times to the onset times chosen by a set of domain analysts. Importantly, the uncertainty distributions sometimes identify subtle changes in the seismic waveform that are missed by both point estimate calculations and by domain analysts.