2002 Annual

Monday, 14 January 2002: 3:45 PM
Interactive, web-based three-dimensional visualizations of operational mesoscale weather models
Lloyd A. Treinish, IBM Thomas J. Watson Research Center, Yorktown Heights, NY
Poster PDF (2.8 MB)
Visualization is critical to the effective analysis and assessment of data generated by numerical weather prediction. In that regard, our previous work has discussed methods of appropriate mapping of user goals to the design of pictorial content by considering both the underlying data characteristics and the perception of the visualization. These have been effective in the development of interactive applications utilizing either workstation or PC/game-class 3d graphics systems with sufficient bandwidth for timely access to the model data. When remote access to the model visualizations is required the limited bandwidth becomes the primary bottleneck since desktop systems can support interactive visualization of typical model data. Popular approaches to web-based visualization are problematic as they impose enough compromises in time for access, display fidelity or interactivity to minimize effectiveness. Some utilize a standard workstation-based application and provide the raw data in a severely compressed and/or sampled fashion to enable faster transmission. But when we consider rapid execution of mesocale weather models operating at cloud-scale resolution, such reduction eliminates critical information. Others utilize web browsers directly by providing static images or flip-book animations, which raises two additional problems. The first is providing only a few frames in time to reduce the number of images to be downloaded. For higher-resolution models, this creates a gross mismatch in sampling to the time step used in the simulation, implying that available images are likely to miss important results. As we have presented before, the resolution of the visualization must match that of the scale of the model to build usable products that are preceptually and scientifically coherent. The second is the tendency to provide visualizations of most if not all of the variables computed by or post-processed from the model. This can create a situation where the user may have difficulty finding images of relevance. In addition, many of these visualizations may never be used, resulting in wasted computational resources. Although such renderings are typically two-dimensional, the issues also apply to three-dimensional techniques. To address these problems, an alternative has been developed. First, a set of visualizations are presented as an interactive, three-dimensional image spreadsheet on a web page. The rows and columns are organized at a high-level (e.g., meteorological characteristics vs. model features). Each cell in the spreadsheet provides a simple interactive 3d scene generated from one time step of the model(s) output. These visualizations then become a meta-representation of the model output and serve as an index to simplify finding relevant visualizations. For example, to address the time-sampling problem, one of the choices is a moderate-resolution animation with frames every 10 minutes of forecast time presented as an MPEG-compressed video sequence. Another interaction is applied to the data-sampling problem. A specialized compression has been developed, based upon task-specific abstractions of underlying components of a visualization scene, which are represented by higher-order descriptions of the geometry. Such abstractions are extremely compact (several KB per time step) and thus, can be transmitted and decoded cheaply at a client for interactive viewing.

Supplementary URL: http://www.research.ibm.com/weather