|Title||Characterizing and Visualizing Predictive Uncertainty in Numerical Ensembles Through Bayesian Model Averaging
|in||IEEE Transactions on Visualization and Computer Graphics|
Luke Gosink, Kevin Bensema, Trenton Pulsipher, Harald Obermaier, Michael Henry, Hank Childs, Kenneth I. Joy |
|Keyword(s)||Ensemble visualization, Bayesian Model Averaging, ground truth, outlier detection, predictive uncertainty, accuracy, statistics|
Numerical ensemble forecasting is a powerful tool that drives many risk analysis efforts and decision making tasks. These ensembles are composed of individual simulations that each uniquely model a possible outcome for a common event of interest: e.g. the direction and force of a hurricane, or the path of travel and mortality rate of a pandemic. In this context, numerical ensemble data represent a complex exploration of a hypothesis space where each simulation explores a possible outcome based on distinct set of initial modeling parameters and internal processes. Paramount to ensemble analysis is quantifying and characterizing the ensemble’s predictive uncertainty, i.e. the ability for ensemble constituents to accurately and consistently predict an event of interest based on ground truth observations. This metric is key to defining what aspects of the hypothetical space are valid and how assumptions behind these hypotheses can be tuned to increase predictive accuracy.
This paper directly addresses the challenge of deriving insight from ensemble data and reducing predictive uncertainty in an ensemble through a new visualization strategy that is based on a Bayesian framework. We use the framework to first construct a statistical aggregate from the ensemble. We extend the information obtained from the aggregate with a visualization strategy that characterizes predictive uncertainty at two levels: at a global level, which assesses the ensemble as a whole, as well as a local level, which examines each of the ensemble’s constituents. Through this approach, modelers are able to better test and refine their models and derive greater understanding about the event of interest. We apply our method to two datasets to demonstrate its broad applicability.