IBM Research

3D Visualization for Weather Forecasting at the Olympic Games THREE-DIMENSIONAL VISUALIZATION FOR SUPPORT OF OPERATIONAL FORECASTING AT THE 1996 CENTENNIAL OLYMPIC GAMES Lloyd A. Treinish  IBM T. J. Watson Research Center, Yorktown Heights, NY Lans P. Rothfusz National Weather Service Forecast Office, Tulsa, OK Introduction To support precision forecasting at the 1996 Centennial Olympic Games in Atlanta, a parallelized version of the Regional Atmospheric Modeling System (RAMS) was installed on a 30-node distributed memory supercomputer (IBM RS/6000 SP) at the National Weather Service in Peachtree City, GA (Zaphiris, Edwards and Snook [1997]; Snook, Christidis and Edwards [1997]; and Edwards, Snook and Christidis [1997]). As an aid in the presentation and analysis of weather simulations at different resolutions (e.g., eight km over the full domain and two km over specific Olympic venues) from this system, interactive three-dimensional visualization methods were introduced. This new capability of producing high-resolution, model data required a change in how operational meteorologists utilized the results so that they might quickly assess whatever data were necessary for the forecast formulation, and still keep the media informed about potential effects of the weather on Olympics events. Since large volumes of complex data for each RAMS run are quickly produced, the use of traditional graphical representations of data for forecasters can be burdensome. Instead of static or simple flip-book animations of two-dimensional techniques like contour maps (as shown in Figure 1), novel three-dimensional visualization strategies were employed. These capabilities were implemented with a general-purpose, portable, data analysis and visualization toolkit IBM Visualization Data Explorer, Abram and Treinish [1995]). Figure 1. 24-hour total precipitation from the remnants of Tropical Storm Jerry as forecast by RAMS. Approach These methods have been developed within a natural coordinate system to provide a context for three-dimensional analysis, viewing and interaction. They provided representations of the atmosphere, as derived from RAMS output, and are registered with relevant terrain and political boundary maps. The focus was placed on surface conditions and precipitation, which were of greatest interest to Olympic organizers and participants. This approach is derived from the notion of correlative visualization, where each data set to be examined is processed independently and merging takes places only at render time (Treinish [1994]). Initially, two classes of visualization were developed under this strategy, both of which provided the forecasters interactive capabilities with three-dimensional representations of the state of the atmosphere (e.g., thermodynamics, moisture, clouds) derived from the data. The data were often realized with redundant encoding schemes (e.g., color and size) and always registered with relevant maps. The first class was a set of quantitative techniques to support model analysis and development of more precise forecasts by providing the forecasters with new tools to interact with and interrogate model output graphically, track the progress of the model as it executes, and quickly and accurately assess results. The second class was intended more for media demonstrations and briefings. An example of output from the first class is shown in Figure 2, which shows the visualizations and a prototype user interface presented to the forecaster as a full screen capture from a workstation. Figure 2. Screen capture of a prototype user interface and interactive visualization of RAMS. The main image in Figure 2 (upper left) shows an isosurface of humidity at 65%, which is pseudo-colored by temperature. This effectively segments the surface into the cold, but moist upper air from the warmer air near the ground. The user has the ability to probe the atmospheric volume with the mouse as a virtual met-station. The dialog box near the bottom center shows the interrogated coordinates [latitude, longitude, pressure] and humidity value. That same probe can be used to define a virtual sounding, a graphical analogue to a meteorologist placing a collection of instruments at a specific location to observe the real atmosphere. In this case, one can derive the same information from the simulated atmosphere (i.e., measurements of specific physical quantities at locations of interest). The probe is extruded into a profile tube, which is pseudo-colored by humidity in this image. Along this profile, the computed winds are shown as vector arrows of velocity, which are pseudo-colored and sized by speed. In addition, there are streamlines of wind velocity, also pseudo-colored by speed. The seed points for the streamline calculations are the same points where the vector arrows are shown. This information is analogous to what might be garnered by a set of balloon-borne instruments launched from that location in a real atmosphere. Based upon the three-dimensional representations, the forecaster can identify features of potential importance, such as upward vertical motion, mesoscale convergence, etc. and place this virtual station accordingly. The data selected by the probe are further displayed by a conventional humidity-pressure profile plot in the lower right of the image. Given the very public nature of the Olympics, the second class of visualizations was composed of a set of simplified qualitative techniques that served two purposes: 1) browse products for gross assessment of the data and 2) source material for dissemination of forecast information potentially suitable for the non-meteorologist (e.g., media, World-Wide-Web, etc.). Figure 3 illustrates a typical presentation of this class. It shows a three-dimensional representation of a cloud derived from a calculation of humidity. The data are shown as a translucent, white isosurface of 95% humidity registered with a pseudo-colored terrain map overlaid with coastline, river and state boundary maps. This familiar representation can effectively show gross atmospheric motion and potential distribution of moisture. Figure 3. 3D humidity cloud registered with base maps. Given that Data Explorer would not be accessible on all forecaster workstations at the time, the ability to provide the qualitative class of visualizations on one workstation was deemed a higher priority for the Olympics weather support. Hence, subsequent efforts focused on the development of an operational version of these capabilities. Operational Implementation A set of applications invoked through c-shell scripts on a UNIX workstation (IBM RS/6000 39H with a GXT1000-2 three-dimensional graphics accelerator) was developed to enable the forecaster to view and interact graphically with RAMS output. (The suite has also been ported to Silicon Graphics workstations, and is parallelized for symmetric multi- processor systems manufactured by both IBM and SGI.) Static and animated three-dimensional visualizations could be created, stored, played back locally and converted for world-wide-web distribution. Most of the tools were interactive and presented a user interface based upon XWindow/Motif for indirect interaction and OpenGL for direct three-dimensional interaction via graphics hardware (i.e., typically provided via the aforementioned Data Explorer software). There was a primary application that enabled one to view, interact with and create time-based output products (animations, snapshot images, geometry) of RAMS data. The forecaster could choose the model run of interest at eight km over the full domain or at two km over a specific venue. Special output from RAMS for this application at 10-minute intervals of simulated time were utilized through a set of custom file readers optimized for network access between the 39H workstation and the SP server running RAMS. The capability to visualize clouds (water density and/or reflectivity) as isosurfaces or direct volume renderings was provided. Surface wind velocity could be shown via a number of techniques (vector arrows, directional streamlines, waving flags) and one surface scalar field could be selected (temperature, humidity, dew point, total precipitation or heat index). All of these fields were shown in a three-dimensional stereographic coordinate system vertically deformed by local topography with vector map overlays and several choices for annotation. Options associated with these selections were presented via various control panels. This application is illustrated in Figure 4 via a full-resolution workstation screen capture. It shows surface humidity as pseudo-colored topography. Surface winds are visualized as vector arrows indicating direction, which are warped onto the local terrain and pseudo-colored by speed. A cloud boundary is represented by a translucent white isosurface of total cloud water density (both liquid and ice) at 10-4 kg/kg. Figure 4. Screen capture of an application for interaction with RAMS output. Inside the cloud isosurface is a translucent cyan isosurface of RAMS-derived reflectivity (at 30 dBz). The reflectivity corresponds to internal rain shafts. Portions of some of the aforementioned control panels and the information and options that they present are also shown in the figure. Among the other facilities provided in this suite were: An application to interactively define control points/camera views for a flight path through the local geography and Olympic venues for key-frame fly-over animations. Given a flight path, and geometry saved from an interactive session with RAMS output, a fly-over animation could be created. Either time-based or fly-over animations could be played back to the workstation and edited. The animations were stored at workstation resolution, 24-bits deep and losslessly compressed to enable full fidelity playback. Stored animation could be converted to the MPEG format at video resolution (352x240), typically for distribution on the world-wide-web. Stored geometry could be converted to the VRML format for distribution on the world-wide-web. Results The suite of tools provided the facilities to interact graphically with data generated by RAMS, and to create a large number of visualization products. To demonstrate that capability a few examples are shown herein. Additional examples, including animations, are available on the World-Wide-Web (http://www.research.ibm.com/OlympicsWeather). RAMS output at 8 km resolution over the entire Olympics domain is illustrated in Figure 5. Over a map similar to those used previously, cloud water density is shown as white isosurfaces with surface precipitation shown as a continuous pseudo-color terrain. Olympic venue sites are marked by white pins. Cloud reflectivity corresponding to rain shafts is shown as interior cyan isosurfaces. Clusters of thunderstorms can be seen where the rain shafts are visible and dark blue puddles of rainfall appear on the surface. The correspondence between the rain shaft and the region of relatively heavy precipitation is quite clear, especially in animation. Figure 5. Visualization of clusters of thunderstorms as predicted by RAMS. The model predicted thunderstorm activity through the greater Atlanta area several hours before it actually occurred, which is illustrated via animation. RAMS output at two km resolution is illustrated in Figure 6. It covers northern Georgia, southeastern Tennessee and southwestern North Carolina showing the region around the northern Olympic venue sites, which are marked with three-dimensional icons. Cloud water density is shown via direct volume rendering -- higher data values are mapped to increased opacity and darker shades of white. Surface heat index is realized as a continuous pseudo-colored terrain. Surface winds are shown as streamlines with directional arrows, pseudo-colored by speed and draped over the topography. Cloud reflectivity is shown as cyan isosurfaces, corresponding to the formation of rain storms. Figure 6. Visualization of two-km-resolution data from RAMS. Orographically influenced clouds over the southern Appalachian mountains that formed in the late afternoon as well as the beginning of thunderstorm activity are visible, both of which can be seen via animation. Utilization Operations of RAMS included several runs each day either over the entire Olympics domain (eight km) or subsets thereof (two km). In general, the first run of each day was a 15- to 18-hour simulation (depending upon how far it ran before the 09 UTC run began) starting at 06 UTC at eight km (Christidis, Z., J. Edwards and J. S. Snook [1997]). The interactive application was used to evaluate the model output and create browse products. Typically, an animation with 10-minute resolution over the full model run was created after the forecaster selected the variables, orientation and geographic view of the domain to display. These would remain invariant throughout the animation. To aid in this selection process the forecaster would interactively move through the geographic scene, experiment with different displays and do limited animation. One or more of the animations from this run (and later RAMS simulations) would be generated for local playback at full fidelity (e.g., 1100 x 800 pixels, 24-bits deep) to support media briefings and news conferences. For animations well-suited to support a particular day's forecast, an MPEG file would be created, as well as a corresponding higher-resolution 8-bit image for distribution on the Olympics weather home page. The interface that illustrates this capability is shown in Figure 7. Figure 7. Olympics/Paralymics weather home page. Evaluation While the first class of applications (quantitative) never reached sufficient maturity for use during the Olympics weather support, the qualitative class was used heavily and, to some extent, made up for the absence of the quantitative class. For example, one could easily infer vertical motion based on a three-dimensional display of clouds forming. Although RAMS may not have indicated precipitation occurring in a specific location, the existence of clouds gave forecasters a forewarning that precipitation may be possible in that vicinity. The single, most useful aspect of the Data Explorer applications was that they virtually eliminated the need to laboriously evaluate numerous two-dimensional images (i.e., similar to Figure 1) by presenting all the relevant information to the forecaster in an easy-to-interpret, four-dimensional display. Conceptual models that would normally require inference from a significant amount of two-dimensional data (e.g., the horizontal extent of cloud dissipation in the lee of the Appalachian mountains) were immediately obvious in three-dimensional animations. Originally intended for media displays, these applications quickly gained favor by forecasters as valuable operational forecasting tools. The success of these tools was due to the impressive synergy between RAMS' excellent mesoscale forecasting skill and the visualization power of Data Explorer. Although the end results (images) were quite impressive, several improvements must be made to these applications for them to gain wider acceptance among forecasters. Combining separate windows would increase the user-friendliness. Integrating the distinct applications cited above would also help. Fly-over animations, for example, were seldom used because they required stopping one application and starting another. (Futhermore, the utility of fly-over animations for operational forecasting is debatable.) Most importantly, however, the speed at which animations are created must be increased significantly. Generating a full-fidelity animation for local playback of an entire run typically required up to 30 minutes of wall clock time, depending on content and options (e.g., 91 frames for a 15-hour simulation). This is a difficult challenge since each RAMS run generated about three MB of data per 10-minute time step (e.g., almost 250 MB for a 15-hour, 8-km run). At six to nine runs per day, this volume exceeded the ability of available workstation hardware to provide highly-interactive utilization. The use of advanced, multi-processor graphics systems should alleviate this bottleneck. The suggested areas of improvements are all technically feasible. Given the need to have an operational capability by the time the Olympics began, and the fact that this effort really only began in earnest a few months prior to that deadline, many of these requirements could not be addressed. On the other hand, placing a prototype facility into a demanding, operational environment provided invaluable feedback for further improvement of these tools. Conclusion and Future Work While this approach to operational weather forecasting has been of immediate value at this year's Olympics, enabling athletes, spectators and officials to plan around adverse weather conditions, these technologies could be applied in other areas where precision forecasting shows promise. Such potential applications include travel, agriculture, broadcast media, pollution monitoring, and fire control and management. Additional work is required to provide a more focused user interface for the techniques already implemented, migrate the prototype quantitative facilities to an operational status, implement tighter coupling between the model and the visualization interface, utilize workstation hardware better matched to the RAMS/SP capabilities, and generalize it to accept arbitrary model output. Acknowledgements This effort was part of the Olympic Games weather forecasting system, a joint project between IBM, and the National Weather Service and the Forecast Systems Laboratory of the National Oceanic and Atmospheric Administration. References Abram, G. and L. Treinish. An Extended Data-Flow Architecture for Data Analysis and Visualization. Proceedings of the IEEE Visualization 1995 Conference, October 29-November 3, 1995, Atlanta, GA, pp. 263-270. Christidis, Z., J. Edwards and J. Snook. Regional Weather Forecasting in the 1996 Summer Olympic Games Using an IBM SP2. Proceedings of the AMS 13th IIPS, February 2-8, 1997, Long Beach, CA Edwards, J., J. Snook and Z. Christidis. Forecasting for the 1996 Summer Olympic Games with the NNT-RAMS Parallel Model. Proceedings of the AMS 13th IIPS, February 2-8, 1997, Long Beach, CA. Snook, J., Z. Christidis and J. Edwards. Forecast Results from the Local-Domain Mesoscale Model Supporting the 1996 Summer Olympic Games. Proceedings of the AMS 13th IIPS, February 2-8, 1997, Long Beach, CA. Treinish, L. Visualization of Disparate Data in the Earth Sciences. Computers in Physics, 8, n.6, November/December 1994, pp. 664-671. lloydt@watson.ibm.com [ DX Home Page | Contact DX ]