Perceptual Effects of Multiple Forecast vs. Ensemble Interval Visualizations Exp 2
Year: 2026
Type: preprint
Abstract: Advances in computational power and growing interest in data visualizations have led to a rapid increase in the use of forecast visualizations. Although interpreting such forecasts has traditionally been the domain of experts, recent global events have heightened public engagement with forecast data and its role in decision-making. In response, a substantial body of research has emerged to establish best practices for forecast visualization across a wide range of domains. One type of forecast visualization that has received comparatively less attention is multiple forecast visualization (MFV), in which independent forecasts from different groups are displayed within a single plot. This technique is designed to convey the range of agreement or disagreement among forecasts, each generated using distinct modeling procedures, input data, and assumptions. MFVs preserve the identity of individual forecasts, allowing viewers to observe structural variability across models. However, MFVs are not the current standard approach in real-world applications. Instead, public-facing forecasts tend to be ensemble forecast visualizations, in which independent models are aggregated into a statistical overview, usually consisting of a mean or median trend line and one or more confidence intervals (CIs). Ensemble forecasts are widely used in weather, climate, sports, and finance domains and are generally found to outperform any individual model. The most common approach to visualizing ensemble forecasts is to display the median prediction along with a confidence interval, such as a 50% or 95% range. Drawing on visualization theories of expressiveness and affordances, this study examines whether MFVs offer perceptual advantages over traditional ensemble summaries. According to the principle of expressiveness, an effective visualization should convey all and only the relevant relationships present in the data. Additionally, the concept of affordances highlights that certain visualization techniques may implicitly suggest specific information, including interpretive strategies, to viewers. We investigate whether visualizing the complete set of independent forecasts provides a more expressive representation than ensemble summary statistics, such as median trajectories with confidence intervals. Visualization theory suggests that exposing more of the underlying data structure may enable viewers to perceive the shape and variability of the forecasts more accurately. We empirically test this theory in the context of multiple forecast visualizations, using longitudinal climate forecasts as a use case. In addition, we will collect participant responses to assess the interpretive affordances of confidence intervals in longitudinal uncertainty visualizations. Understanding the inferences that confidence intervals elicit may clarify when they are appropriate for public communication and when alternative representations should be considered.
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7169877887