Visualization design criteria for the communication of weather and weather impact forecast data with explicit uncertainty

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Tuesday, 19 January 2010: 8:30 AM
B213 (GWCC)
Lloyd A. Treinish, IBM Thomas J. Watson Research Center, Yorktown Heights, NY; and E. Novakovskaia and H. Li

Presentation PDF (2.3 MB)

Forecasting specific weather events or combination of conditions that can disrupt business operations in industries as diverse as energy, water, transportation or agriculture, or drive decision making for public good like emergency management, can enable proactive allocation and deployment of resources to minimize time for restoration and recovery. For predictions of such conditions and their impacts to be useful, and to support effective decision making with confidence, the information must be disseminated in a timely fashion. Visualization is an essential means to communicate these predictions, but two challenges must be addressed. Since probabilistic forecasts are generally viewed as being the most appropriate means to capture the range of potential weather and impact scenarios, how can the uncertainty associated with them be depicted? The answer is partially dependent on the second issue. Given the variety of tasks that would be represented by the utilization of the forecasts by an analyst (e.g., an operational forecaster), decision maker (e.g., emergency manager) or a layperson (e.g., the public), what are the most appropriate visualization strategies? In our efforts to primarily support the second class of users, we have determined that appropriate visualization designs need to permit almost pre-attentive interpretation with little time required for training as part of operational planning (e.g., prior to a severe storm event). More generally, however, among the design criteria are appropriate utilization of visualization elements (e.g., geometry, color), consistency with and coupling to weather and impact data, incorporation of annotative spatial and temporal information to provide a familiar reference frame, cartographic reprojection to minimize spatial distortion, etc. We leverage work in the psychophysics, numerical methods and cartography communities to address many of these issues. Further constraints are derived from limitations in web-based deployment, and with the workflow and perception of the intended users.

Traditional means to depict explicit uncertainty in gridded weather data are suboptimal for many user tasks because they remove critical dimensionality in their realizations (e.g., so-called “spaghetti plots” derived from ensemble forecasting systems) and therefore, may not illustrate essential features for prerequisite user tasks. Given the inherently high dimensionality of the probabilistic forecasts compared to the visualization media, restrictions in the design choices were incorporated to minimize the visual clutter associated with overloaded realizations. These include limited color choices with natural ordinality applied to categorical data or imposing categories on interval data (e.g., contour banding). An extension to this approach is applied via categorization of the uncertainty and then mapping the results to various visualization strategies which adds texturing to the original design. The use of textures in such compound presentations that reflect uncertainty have been shown to be effective in a number of user interpretation tasks via studies in the human factors community.

We will discuss several of these ideas and their overall effectiveness in the context of an operational deployment with the emergency management group at a major utility company in the northeastern United States, as well as other types of user tasks. They have been applied to uncertainties derived from physical models (e.g., ensembles of numerical weather prediction results) and stochastic models of the impact of severe weather.