DETAILS


COLUMNS


CONTRIBUTIONS


ARCHIVE



New Visualization Techniques

Vol.34 No.1 February 2000
ACM SIGGRAPH


3D Visualization Development at NOAA Forecast Systems Laboratory



Paula T. McCaslin, Philip A. McDonald and Edward J. Szoke
National Oceanic and Atmospheric Administration



Figure 1:†An example of hourly three-dimensional LAPS analyses created automatically and available at the following website.

Introduction

Visualization transforms numeric data into a visual form that enables users to conceptualize and understand the information. Three-dimensional (3D) visualization is the ability to display, analyze, manipulate and interact with 3D data in 3 space. New visualization tools, 3D in nature, are being designed to display meteorological datasets for use in operational forecasting.

Forecast Systems Laboratory (FSL) has been supporting the development of 3D visualization software and applications since 1990. Until recently, the emphasis has been on research application. Using commercial visualization software called the Application Visualization System (AVS5), both analysis and forecast 3D data were displayed and investigated. The software was used for visual analysis and scanning of data for the presence of desired features. For example, plotting of the station observations with forecast-generated data allows visual comparison of the two. Figure 1 shows 3D images of weather parameters from the Local Analysis and Prediction System (LAPS) analysis output, which is created automatically on an hourly basis for the World Wide Web.

Operational meteorologists who issue forecasts and warnings for the nation from the various Weather Forecast Offices (WFOs; see the website for a display of WFOs across the nation) currently utilize two-dimensional displays of analyses and numerical model output, combined with images from satellites and radar. They do this on a system developed at FSL call the Advanced Weather Interactive Processing System (AWIPS). Visualization has been used successfully in the research context for some time. The next logical step was to determine if 3D visualization can offer added value in an operational setting. This required developing a 3D visualization workstation application, conducting a forecast exercise using the system and evaluating feedback from the exercise. An experimental workstation application, Display 3D (D3D), was developed at FSL to investigate the complexities and 3D structure of atmospheric parameters, and potential value added of 3D displays in an operational forecast setting. The D3D system was designed to be used with the two-dimensional AWIPS operational system known as D2D.

The scope of this article covers the development of the D3D application, plus a brief description of the D3D real-time exercise (RT98).

Building D3D

Early in 1997 a decision was made to use a software product known as Vis5D, developed at the University of Wisconsin, as the core of D3D. Version 4.2 of Vis5D included the development of an application programmerís interface (API). This was an attractive choice because the API is a division between the systems user interface and its main core, enabling system developers to include Vis5D as a visualization subsystem of other systems. The Vis5D software also had superior animation capabilities, and was designed specifically for atmospheric science (eliminating any overhead that may have been inherent in a multidisciplinary application). Animation and performance were weighted heavily because of the sensitivity of the human visual perception to motion, and the ease with which Vis5D could handle these tasks. Other reasons for choosing Vis5D were that it had an established a user base in the meteorological community, was free (with no licensing issues) and came with an optimized data structure for the fast rendering of objects.

Vis5D is being developed at FSL in collaboration with the University of Wisconsin. General purpose enhancements developed at FSL are integrated into the standard releases of Vis5D. This process minimizes the divergence between the two versions: theirs and ours.

Graphical Software Development

Because D3D is intended to be used as a component of AWIPS in WFOs, it must coexist with the other AWIPS applications, principally D2D, with a minimum impact on workstation resources. The first step in meeting this requirement was to modify Vis5D for accessing AWIPS database. This modification provides access to the D2D basic and derived gridded datasets without having to create intermediate Vis5D files.

In a research environment, Vis5D is typically used to read one data file at a time, but in D3D it would need to be reloaded with model data without being restarted. Although this was a capability of Vis5D, its memory management and reinitialization functions needed to be refined. Since the D3D user interface was designed to operate as a process separate from Vis5D, the problem of communication between the two processes had to be addressed. The solution we chose was to incorporate an event-driven interprocess communication scheme in both the interface and Vis5D.

After addressing the issue of compatibility, attention was directed toward refining and enhancing Vis5Dís display capabilities. Every effort was made to accomplish the following list of enhancements using Vis5Dís API whenever possible, and minimizing changes to Vis5Dís core:

  • †Improved graphics, transparency, font support and volume visualization for Hewlett-Packard computers (the selected AWIPS hardware) that use the PEX graphics library
  • †Improved fitting of maps to the topography
  • †Improved contour labeling
  • †A base around the topography to hide confusing subterranean meteorological features
  • †Slice linking so that multiple cross sections can be moved in unison and toggled on/off simultaneously
  • †Changes to the sounding plots so that they more closely resemble those produced by D2D.

Figure 2:†A WFO-advanced D3D application showing the narrow menu window, the large display pane and the D3D volume browser. The image shows a 0-h Eta model analysis valid 1200 UTC 2 October 1998, as viewed from the south. Displayed is a 100-kt wind speed isosurface (dark gray), a vertical cross section of wind barbs centered through Denver, a 290-K potential temperature isosurface (white),†l2.0 g/kg specific humidity isosurface (red) and a contoured and colored vertical†cross section displaying heights (km).
Figure 3: D2D volume browser showing redundancy in the customized list of graphics.
Figure 4: D2D volume browser showing no redundancy in the customized list of graphics.
Figure 5: D3D volume browser showing an opened isosurface property editor and the widgets used to change the isosurface attributes.
Figure 6:†An example of model predicted cloud ice (white colored isosurface), radar reflectivity 40 dBz (orange isosurface), radar reflectivity 20 dBz (transparent yellow isosurface), for spring storms in eastern Colorado, viewed from the east.

User Interface Software Development

It was a major undertaking to replace the original Vis5D user interface designed for research meteorologists with a more suitable one for forecast meteorologists.

The Tool Command Language/Tool Kit (Tcl/Tk) was used to create a new D3D user interface similar to the one already familiar to forecasters using D2D. Even though D2D and D3D applications focus on 2D and 3D displays, respectively, it was suggested that, where possible, the applications have the same look, feel and function. This design approach required significant prototype development, which became the foundation software for further development. An efficient result of writing both graphical user interfaces in Tcl/Tk was reusability. Some of the source code for the D2D user interface could then be directly used in D3D. To illustrate, a tool used to control the attributes of, say, the animation function (called loop properties) is invoked from the toolbar in both D2D and D3D. The loop properties graphical interface has the same look, feel and function in both applications. This is because the source code that creates the user interface differs only in the resource call, where D3D sends API messages to Vis5D to effect changes in the looping speed, direction and delay intervals. By reusing source code where possible, we made faster progress in the software development, and it was easier for users already familiar with D2D to learn how to use the system.

The D3D user interface (Figure 2) consists of a narrow window containing menu options at the top of the screen, one large display pane and the D3D volume browser. The menu and icon buttons displayed along the second row of the menu options (the toolbar) have the same function and appearance as the D2D toolbar, with the addition of several view positioning buttons. The D2D and D3D menu bars may appear the same, but the options in the D3D system pull-down menus vary from those available in D2D. The volume browser, invoked from the menu bar, provides access to numerical models and gridded data sources (currently limited to volume radar). Through the browser interface, the user can select the data source, fields and rendering techniques to generate a customized list of graphics to display.

The choice of rendering techniques sets the D2D and D3D browsers apart. The rendering techniques available in D3D include isosurfaces, vertical and horizontal cross section contours and images, surface contours and images, volume visualization, a vertical sounding plot and a virtual data probe. The rendering techniques available in D2D, loaded through the D2D browser, are horizontal and vertical cross-sections displayed as contours and images.

A key decision in the design of the D3D browserís user interface was to pair field selection with rendering technique selection. To add a product to the list of graphics to display, specifically, the user presses the pointer over the desired rendering technique, then selects the desired field from the list of fields that appears. Coupling of the rendering technique and field selection eliminated the implicit redundancy in the customized list of graphics available to display (Figure 3). With the amount of information present in 3D visualization, it is important to avoid automatically adding products to be generated that are not explicitly requested by the user (Figure 4). Also, as each selected item is added to the customized list, a property editor (Figure 5) is created that can be accessed by a click of the right mouse button on the field name in the list. Opening the property editor for an item allows a user to modify the default data attributes, including the value, the value ranges and the colors, and then generate a new object once Load is selected.

Real-Time Exercise

A limited real-time forecast exercise, RT98, was conducted during the summer of 1998 at FSL to evaluate D3D. The RT98 exercise involved 20 participants, with broad participation attendance from FSL and other NOAA laboratories, each conducting five, three-hour shifts scheduled morning and afternoon five days a week for a period of six weeks. Prior to the plans for a formal exercise, an effort began in the fall of 1997 to incorporate D3D into the regular FSL daily weather briefings, a half-hour discussion of the weather. Volunteers from the briefing staff worked closely with the D3D meteorological staff to create displays for the briefing. This interaction helped introduce the D3D capabilities to FSL meteorologists, as well as visitors and other attendees at the briefings. This effort was useful in the development of new products to display, but it was difficult to get other meteorologists to spend enough time with D3D to adequately evaluate the products and the interface. This led to the decision that an exercise would be the next productive step in D3D development.

Although development of the D3D Tcl/Tk interface had just begun, the time was right for meteorological evaluation. Developers worked closely with meteorologists who suggested specific areas of improvement and evaluated the changes. FSLís evaluation team was brought in to develop detailed evaluation plans, evaluation metrics and schedules for the exercise.

Training of the evaluators was an important aspect of the exercise. There was an initial meeting to introduce the application, followed by personalized training sessions for each participant. The D3D developers made sure that D3D capabilities were reliable throughout the exercise. They were called upon during the exercise to provide follow-up training and to give detailed explanations of the inner workings of D3D. Since the main focus of RT98 was to evaluate D3D, there were new 3D datasets for participants to work with, such as products from a high-resolution local model and a 3D radar volume product.

RT98 was useful in providing considerable feedback on both the user interfaces for controlling D3D as well as the meteorological value of 3D displays. There was a great deal of enthusiasm displayed for the application, although there was still scepticism concerning its potential operational significance. We did discover that considerable training was necessary to effectively use D3D, even among those who had some prior experience through the daily weather briefings.

Future Work

The development of D3D has been driven by the ideas and questions posed by meteorologists interested in spatial observations and investigation of numerical data. This has been true since LAPS modelers first looked at their output on FSLís Stardent 2000, circa 1990. Historically, the motivation for the development of 3-space visualization tools has been the enormous amount of data that one can analyze in a relatively short amount of time, especially when verifying model accuracy and diagnosing model problems. It is also desirable to examine meteorological fields in three-dimensions since so many of them are naturally 3D (clouds, for example). This contrasts with the 2D sampling of a time series of several variables on both horizontal and vertical cross sections, at many pressure levels and spatial locations, respectively, to diagnose the model output.

The principal development of D3D is, not surprisingly, still driven by the ideas and questions posed by meteorologists interested in spatial observations and investigation of numerical data. The D3D developers are often cued to a disguised request for a new 3D tool or feature that starts with phrases such as: "It would be really neat if you could do...", or "Is it possible to look at the data in a way that shows...?".

The response to the question, "is value added by the use of 3D visualization in an operational forecast setting?," is not a simple yes or no. The issue is more complex. Certain atmospheric variables lend themselves very well to 3D visualization, such as ice and moisture variables (as shown in Figure 1). Additionally, surface variables displayed on relief topography are also informative, especially when the influence of the topography on the variable(s) can be observed (also shown in Figure 1). In contrast, there are variables that do not lend themselves very well for presentation in 3D. There are also variables that when displayed convey information at certain threshold values, yet do not convey information at other threshold values. The graphical representation of a 10% relative humidity isosurface (a single valued contour in 3 space), for example, would create such a large object that it would almost completely obscure the whole domain, adding more noise than information. Perhaps a more plausible and well-grounded question could be phrased: "Is there value added by the use of displaying 3D visualization of certain candidate variables and techniques in an operational forecast setting?"

Components of D3D that are actually more 2D in nature but are more powerful in their capabilities within D3D have generated much interest and positive response. These include the capabilities of horizontal and vertical cross sections that were expanded through the use of an easy to access (and very popular) slider bar, which enables one to quickly move the cross sections through the data, stopping at any arbitrary location. When used with an isosurface, such functionality adds a level of quantitativeness to examining the data (that is, one would know more precisely at what level the isosurface existed, as well as its value).

Another powerful feature is the sounding/parameter plot. In D3D a sounding (temperature and dewpoint plot with wind barbs) can be chosen, with up to three additional meteorological parameters plotted. The sounding site can then be easily moved to any location, with an instant update to the readout, quickly revealing (in a very quantitative manner) a great deal of data. The model data can also be looped through time when stopped at a particular location. Again, it is presumed that maximum benefit might arise from using such features in conjunction with more purely 3D displays like an isosurface.

It is noteworthy to mention that the value added by 3D visualization of atmospheric data is directly related to the resolution of the data. As the temporal and spatial resolution of the datasets increase so does the value of the ability to more quickly make observations in the datasets that have the three recursive phases: detection, identification and measurement and analysis.

As long as support and funding are available, the development of D3D will continue. It has been proven through demonstration (i.e., at the last U.S. Olympics) that 3D visualization offers a good deal of potential for its use in an operational setting.

The development efforts will involve the technical aspect of adding new features - for example, the ability to display multiple models in the same display context, and a controllable cropping tool that would allow the viewer to peek behind objects that may obscure a perspective view at, say, Denver, CO, or elsewhere. The key development efforts also involve the implementation of input generated during RT98, addressing how best to use 3D capabilities, training issues, potential display variables and techniques, etc.

While RT98 provided valuable feedback, we realized that further exposure to a broader group of mainly operational meteorologists would be required before attempting to implement D3D in an operational setting. Forecasters from WFOs across the nation, as well as from the National Centers (which are responsible for more nationally oriented products, for example the National Hurricane Center (website)), participated in a more formal exercise, RT99, from late October through December 1999. Each forecaster spent two weeks at FSL working with D3D. This exercise provided the next level of feedback that could pave the way for implementation of D3D in operational forecast settings throughout the nation.



Paula McCaslin is a Computer Scientist developing 3D visualization capabilities to investigate meteorological data in Boulder, CO. She is looking for art amidst the technology. She has a degree in mathematics from the University of Colorado-Boulder.

Phil McDonald is a Research Associate at the Cooperative Institute for Research in the Atmosphere (CIRA) at Colorado State University, who actually works as a 3D visualization code jockey at the Forecast Systems Laboratory. He has an M.S. in environmental design and meteorology from the University of Oklahoma and a B.S. in forestry from the University of California - Berkeley. His passions include the restoration and preservation of narrow gauge steam railroads.

Ed Szoke is a meteorologist working with the D3D group and a long-time weather fanatic, having kept weather records since 3rd grade. His areas of interest include forecasting, severe weather and winter weather, and a FSL assists in determining ways to make new technology viable for operational weather forecasting. He has degrees in meteorology from the University of Wisconsin - Madison and Penn State University.

Paula McCaslin
NOAA Forecast Systems Laboratory
325 Broadway R/FS1
Boulder, CO 80303

Tel: +1-303-497-3187
Fax: +1-303-497-7262
Website

Phil McDonald

Tel: +1-303-497-6055

Ed Szoke

Tel: +1-303-497-7395

The copyright of articles and images printed remains with the author unless otherwise indicated.