Uncertainty Visualization by Representative Sampling from Prediction Ensembles.

Area of Science:

• Data visualization
• Scientific communication
• Geospatial analysis

Background:

• Traditional data ensembles for uncertain predictions often use spatial spread, leading to glyphs that increase in size with uncertainty.
• This size increase can be misinterpreted as an increase in the phenomenon's actual size, strength, or other attributes.
• Existing methods lack an efficient way to encode additional information alongside uncertainty.

Purpose of the Study:

• To propose and validate a novel visualization technique for uncertain predictions using a carefully selected subset of a prediction ensemble.
• To demonstrate how this method can implicitly convey uncertainty, avoiding misinterpretations common in spatial spread visualizations.
• To explore the potential for encoding additional information within the visualization.

Main Methods:

• Developing a method to select a representative subset of a prediction ensemble.
• Applying this method to hurricane prediction data and creating visualizations.
• Conducting a cognitive experiment to compare the proposed visualization with traditional methods.

Main Results:

• The selected subset display effectively conveys uncertainty without increasing glyph size.
• The method avoids the confounding of uncertainty with phenomenon size.
• Visualizations successfully preserved spatial statistics of the original ensemble.
• An explicit encoding of uncertainty was also constructible from the selection.
• Cognitive experiment results showed significantly reduced confounding of uncertainty with storm size.

Conclusions:

• Displaying a carefully chosen subset of prediction ensembles is an effective strategy for visualizing uncertainty implicitly.
• This approach mitigates misinterpretations of prediction uncertainty in spatial contexts, particularly in hurricane forecasting.
• The method enhances viewers' ability to accurately assess potential storm damage by improving the estimation of storm size and uncertainty.