Elizabeth Dirks Fauerbach
Robert M. Edsall
David Barnes
Alan M. MacEachren
Department of Geography
302 Walker
Penn State University
University Park, PA 16802, USA
ABSTRACT
This paper describes a prototype visualization environment designed to facilitate an understanding of, and comparison among, meteorological forecast models. Emphasis is on tools that allow an analyst to compare predictions of three models, both to one another and to the actual weather that they predict. The system allows for surface pressure patterns (an area feature) and storm centers (a point feature) to be examined as they correspond or diverge over time. Our discussion of the prototype includes attention to the role of animation for exploratory analysis of geographically and temporally referenced information, interactive controls for dynamic maps, and methods for representing uncertainty (reliability) of georeferenced forecasts.
INTRODUCTION
A plethora of georeferenced data exist in the earth and atmospheric sciences. These data are derived from both field measurement and models. Typically, they include a temporal as well as a spatial component. A primary goal of data collection or generation in this context is an understanding of change in spatial patterns over time. Visualization (in the form of static maps) has been used throughout the history of science as a primary method for constructing knowledge about change in environmental attributes. Today, dynamic visualization tools offer the potential for new insight into existing questions related to environmental change and for the identification of new questions.
In the context of scientific visualization more generally, georeferenced visualization has been characterized as emphasizing the private interactive exploration of spatial information with a goal of revealing unknowns (MacEachren, 1994). From this perspective, the project presented here might be described as a prototypical visualization application. Its focus is on the exploration of model-derived atmospheric data covering relatively short temporal duration (i.e., meteorological time scales) and continental spatial extent. Specifically, the prototype presented is designed to facilitate an understanding of the spatio-temporal reliability of weather forecast models. This prototype allows an analyst to compare results of three models to one another and to the actual weather that they attempt to forecast. Description of our prototype will be preceded with a brief review of relevant developments in georeferenced visualization.
BACKGROUND
Scientific visualization is a research area that integrates developments in theory, methods, and tools for representation with specific areas of scientific application. Thus, two distinct literatures are usually relevant to any visualization research, one related to representation theory, methods, and tools and the other related to the scientific questions that the methods and tools are designed to explore.
THE PROTOTYPE
As noted above, animation of environmental data has been used to examine change in data reliability over time (MacEachren, et al., 1993 and Mitasova, 1993). Animation of model results have been used to help understand the interaction between vegetation and climate change (Prentice, et al., 1991) and to assess model reliability by comparing several models over time (DiBiase, et al., 1992). Interactive manipulation of a dynamic display has been applied to the study of change in forest structure over time (Buttenfield and Weber, 1994). Our project builds upon all of these efforts to develop a prototype that includes integration of interactive control with animation, exploration of output from spatio-temporal models, and the representation of model reliability. The resulting system enables users to compare individual meteorological models with each other, to explore the relationship between an average of the models and the amount of divergence from that average, and to contrast the models with the actual weather pattern of the same time period. The models explored in this prototype are short-term meteorological forecast models from the National Meteorological Center (NMC).
The Models
Three models, the eta, NGM, and spectral are released simultaneously as the NMC's short-term (two-day) forecast product. Each is a result of a highly complicated computer algorithm which assimilates a fine grid array of observations and, each through a slightly different technique, produces forecasts for North America. The three models differ in resolution, physical parameterizations, and numerics, and use varying lengths of developmental data (Vislocky and Fritsch, 1995). As a result, the models perform differently in different situations; for example, Junker, et al. (1989) found that the NGM (nested grid) model is less reliable than other existing models when predicting winter precipitation when the polar jet stream is in a certain orientation. In light of the large body of literature discussing the variable performance of each model (e.g. Vislocky and Fritsch, 1995; Caplan and White, 1989; Petersen, 1992), we expected that the models would be significantly divergent in their forecasts, especially toward the end of the 48 hour period.
Each of the three models is released by the NMC as a series of static maps, each representing one "time slice" of the 48 hour period. There are eight maps for each model produced every six hours; each map is a forecast of conditions at a time which is an integral multiple of six hours from the time of release. Each forecast is stored as a grid of points, equally spaced on a base map of North America, at a spatial resolution of approximately one point per 50 km, with 30 points in the latitudinal dimension and 40 points in the longitudinal [this is a document with embeded 32K .gif]. Any finer spatial resolution using data exclusively from these models would necessarily involve interpolation. An array of meteorological variables is stored at each point, including surface and upper air winds, temperature, humidity, and pressure. As a result, each time-step of each model consists of 1200 grid points, each with several forecasted attributes.
The Case
We have compiled the 24 forecast data sets (eight data sets from each of the three models) from one release time, 7:00 am on February 2, 1995, for our project. During the 48 hour period following the forecast, an important storm developed over the Tennessee valley and rapidly intensified as it moved over the warm waters off the U.S. east coast. The storm brought blizzard-like conditions to the major cities of the northeast, including Philadelphia, New York, Boston, and Buffalo. Such a storm is called a "Nor'easter" by meteorologists because of its northeasterly track as it deepens off the Atlantic coast.
The parameter we deemed most useful for representing this event was surface pressure. Isobars (lines of equal pressure) reveal a great deal about weather conditions at a particular point; in general, surface winds circulate clockwise around an area of high pressure, counterclockwise around a low, and parallel to the surface isobars. Closely packed isobars are indicative of high winds, widely spaced isobars are indicative of light and variable winds. The location of the "low" or "high" can also have a bearing on the moisture content of the atmosphere; for example, a low positioned off the Carolina coast will be likely to bring moist easterly winds to the middle Atlantic and Chesapeake regions. If, as in the case of a classic "Nor'easter," a wintertime cold high is positioned over Labrador, the clockwise winds associated with it will bring cold air to the northeast U.S., and any moisture transported into this cold air (from the Atlantic low, for example) will fall as snow.
A focus on isobars and the pressure patterns they represent is also of practical significance for us as cartographers, since a time series of isarithmic depictions allows us to track the movement of the systems in a fluid way in our animation. Other parameters such as temperature and humidity are not as indicative of air mass (e.g. storm) movement, and thus would be less informative in an animation of the development of a Nor'easter.
Procedures
As mentioned above, the data from the models is formatted in an array of 40 columns and 30 rows. Each array represents a grid of pressure values at a particular time, the total area of which corresponds to the area of the continental United States and some of southern Canada. We received data for each model in six-hour increments, as well as the actual pressure values for 7 a.m., February 2 (time zero) and for every subsequent 12 hours within the forty-eight hour time period. Details of procedures followed to construct the application are available here
Results
[To access the application, click here. Requires one of the following browsers: Netscape 2 (or higher), Websurfer 5, Emissary 2, or Internet Explorer and the Shockwave plug-in . To acquire the bowser and/or the plug-in (available for Windows and Mac platforms) go to the Shockwave distribution site.]
The completed interactive animation application allows the user to view several variables simultaneously or separately, which can reveal otherwise hidden characteristics of the data. For example, when the uncertainty information is displayed with the average isobars, it is clear that the models disagree widely in the extreme northwestern corner of the map, off the coast of Victoria Island in British Columbia. Upon investigation, this disagreement was found to be a result of inherent differences in each model's methods of resolving the intense deep low pressure systems which form regularly in the Gulf of Alaska. Also supported by the animation is the supposition that maritime climates are not handled as well by the models as continental climates, as evidenced by the significant uncertainty of the Nor'easter track as it reaches the warm waters of the Atlantic.
The animations of the models' storm tracks, displayed simultaneously, reveal that a key component of uncertainty surrounding the storm is temporal discrepancy among the models. For instance, as the storm moves across the Appalachians, the uncertainty of the prediction is attributable not to differences in the predicted location of the storm center but rather to differences in the rate at which the storm will move along the predicted path. Here, the display shows the lack of synchronization in the model predictions. The spectral model carries the storm toward the Atlantic much more rapidly than do the other two models, but the storm centers predicted by the NGM and eta models eventually catch up to that of the spectral.
A feature of the system that provides further insight into the data is the interactive control over the animation provided by Director. The ability to change the pace of the movie or to step through the animation one frame at a time reveals details about the data that would be difficult if not impossible to detect otherwise. This was clearly illustrated in the early stages of the system's development, when initial animations of the fluid isolines appeared jumpy, indicating data processing errors. Stepping through the animation allowed us to easily pinpoint the aberrant frame, which would have been difficult to locate in a table or in a series of static maps.
DISCUSSION
Since we made the decision to represent the model output as an average of the three short-term forecast models, meteorologists Vislocky and Fritsch (1995) have investigated the forecasting performance of an average of two of these three models and compared this average to the performance of each model individually. They have found that such an average (termed consensus) in fact outperforms each model alone. They call for a "strategy for statistically combining available forecast products rather than relying upon the single most superior product" (p. 1157). Though we had been unaware of their findings when making our initial data processing and representation choices, we are of course pleased that our project by its nature employs such a strategy.
Although emphasis in developing our prototype has been on particular spatial and temporal scales and on a particular kind of model derived data, the animation principles, interactive interface tools, and reliability representation methods implemented in the prototype are relevant to a broader range of environmental change applications. Since the interactive controls on our animations are similar to those provided by many hypermedia tools (including some available for use with the World Wide Web), results of building and assessing our prototype are relevant to a range of visualization applications well beyond those build within the Director environment.
As with most prototype visualization tools, the one described here is probably more valuable as a device for identifying unanswered questions than it is as an application tool that will facilitate scientific thinking. At this point, the prototype (due to the Macromedia Director multimedia development environment it was designed within) is "hardwired" to a particular data set. To be put into practice as a tool for scientific research, the basic operations described above would need to be implemented in a datadriven environment in which analysts can easily apply the tool to any data set of interest (including non meteorological data). In addition, more flexible interactive controls are needed that will allow analysts to change symbolization used to depict each model, their combination, and the variation among them. We anticipate making these and other improvements as the project evolves.
REFERENCES
Beard, M. K., B. P. Buttenfield, 1991. NCGIA Research Initiative 7: Visualization of Spatial Data Quality. NCGIA.
Buja, A., McDonald, J. A., Michalak, J. and W. Stuetzle, 1991. Interactive data visualization using focusing and linking. Proceedings, Visualization '91, IEEE Conference on Visualization San Diego, CA, pp. 156-163.
Buttenfield, B. P.,1993. Representing Data quality. Cartographica, special issues on Mapping Data Quality, 30(2&3): 1-7.
Buttenfield, B. P. and M. K. Beard, 1994. Graphical and geographical components of data quality. Visualization in Geographic Information Systems. London, John Wiley & Sons. D. Unwin and H. Hearnshaw, (ed.) 150-157.
Buttenfield, B. P. and C. R. Weber, 1994. Proactive graphics for exploratory visualization of biogeographical data. Cartographic Perspectives 19: 8-18.
Caplan, P. M. and G. H. White, 1989. Performance of the National Meteorological Center's Medium Range Models. Weather and Forecasting 4: 391-400.
Cleveland, W. S. and M. E. McGill, 1988. Dynamic Graphics for Statistics, Belmont, CA: Wadsworth & Brooks/Cole.
DiBiase, D., 1990. Visualization in the Earth Sciences. Earth and Mineral Sciences (Bulletin of the College of Earth and Mineral Sciences, the Pennsylvania State University), 59(2): 13-18.
DiBiase, D., J. Krygier, C. Reeves, A. M. MacEachren and A. Brenner, 1991. Animated Cartographic Visualization in Earth System Science. Proceedings, 15th International Cartographic Conference, Bournemouth, England, 23 Sept. - 1 Oct., 1991, ICA. 223-232.
DiBiase, D., MacEachren, A. M., Krygier, J. B. and C. Reeves, 1992. Animation and the role of map design in scientific visualization. Cartography and Geographic Information Systems 19(4): 201-214.
DiBiase, D., C. Reeves, J. Krygier, A. M. MacEachren, M. von Weiss, J. Sloan and M. Detweiller, 1994. Multivariate display of geographic data: Applications in earth system science. Visualization in Modern Cartography. Oxford, UK, Pergamon. A. M. MacEachren and D. R. F. Taylor ed. 287-312.
Dorling, D., 1992. Stretching Space and Splicing Time: From Cartographic Animation to Interactive Visualization. Cartography and Geographic Information Systems 19(4): 215-227, 267-270.
Dorling, D. and S. Openshaw, 1992. Using Computer Animation to Visualize Space-Time Patterns. Environment and Planning B: Planning and Design 19: 639-650.
Egbert, S. L. and T. A. Slocum, 1992. EXPLOREMAP: An exploration system for choropleth maps. Annals of the Association of American Geographers 82(2): 275-288.
Fegeas, R. G., J. L. Cascio and R. A. Lazar, 1992. An overview of FIPS 173, The spatial data transfer standard. Cartography and Geographic Information Systems 19(5): 278-293.
Fisher, P., 1994. Hearing the Reliability in Classified Remotely Sensed Images. Cartography and Geographic Information Systems 21(1): 31-36.
Grotjahn, R. and R.M. Chervin, 1984. Animated graphics in meteorological research and presentations. Bulletin of the American Meteorological Society 65(11):1201-1208.
Hibbard, W. and D. Santek, 1989. Visualizing large data sets in the Earth Sciences. IEEE Computer 22:53-57.
Hibbard, W., L. Uccellini, D. Santek and K. Brill, 1989. Application of the 4D McIDAS to a model diagnostic study of the Presidents' Day cyclone. Bulletin of the American Meteorological Society 70(11): 1394-1403.
Junker, N. W., J. E. Hoke and R. H. Grumm, 1989. Performance of NMC's Regional Models. Weather and Forecasting 4: 368-390.
Karl, D., 1992. Cartographic Animation: Potential Research Issues. Cartographic Perspectives 13(Fall): 3-9.
Koussoulakou, A., 1994. Spatial-temporal analysis of urban air pollution. in Visualization in Modern Cartography MacEachren, A.M. and D. R. F. Taylor (ed). London: Pergamon, pp. 243-269.
Kraak, M.-J., 1994. Interactive modeling environment for three-dimensions maps: Functionality and interface issues. Visualization in Modern Cartography. A. M. MacEachren and D. R. F. Taylor (ed). London: Pergamon. 269-287.
Kraak, M.-J., A.M. MacEachren, 1994. Visualization of spatial data's temporal component. SDH94 (Sixth International Symposium on Spatial Data Handling), Edinburgh, Scotland, 5-9 September, 1994, IGU, pp. 391-409
Leidner, M., 1995. Pennsylvania State University, personal communication.
MacEachren, A. M., 1992. Visualizing Uncertain Information. Cartographic Perspectives 13(Fall): 10-19.
MacEachren, A. M., 1994. Visualization in modern cartography: setting the agenda. in Visualization in Modern Cartography MacEachren, Alan and D. R. F. Taylor (ed). London: Pergamon, pp. 1-13.
MacEachren, A. M., 1995. How Maps Work. New York: Guilford Press.
MacEachren, A. M. and J. H. Ganter, 1990. A pattern identification approach to cartographic visualization. Cartographica 27(2): 64-81.
MacEachren, A. M., Howard, D., von Wyss, M., Askov, D. and T. Taormino, 1993. Visualizing the health of Chesapeake Bay: An uncertain endeavor. Proceedings, GIS/LIS '93, Minneapolis, MN, 2-4 Nov., 1993, pp. 449-458.
MacDougall, E. B., 1992. Exploratory analysis, dynamic statistical visualization, and geographic information systems. Cartography and Geographic Information Systems 19(4): 237-246.
McGranaghan, M., 1993. A Cartographic View of Spatial Data Quality. Cartographica 30 (2&3): 8-19
Max, N., R. Crawfis and D. Williams, 1993. Visualization for climate modeling. IEEE Computer Graphics and Applications 13(4): 34-40.
Mitasova, H., 1993. Multidimensional interpolation, analysis and visualization for environmental modeling. Proceedings, GIS/LIS '93 Minneapolis, MN, 2-4 Nov., 1993, pp. 550-556.
Moellering, H., 1980. The real-time animation of three dimensional maps. American Cartographer 7: 67-75.
Monmonier, M., 1989. Geographic brushing: Enhancing exploratory analysis of the scatterplot matrix. Geographical Analysis 21(1): 81-84.
Monmonier, M., 1990. Strategies for the visualization of geographic time-series data. Cartographica 27(1): 30-45.
Neeman, B.U. and P. Alpert, 1990. Visualizing atmospheric fields on a personal computer: application to potential vorticity analysis. Bulletin of the American Meteorological Society 71(2): 154-160.
Petersen, R., 1992. Comparisons of CFM and NGM lifted index calculations. Weather and Forecasting 7: 536-541.
Peterson, M. P., 1995. Interactive and Animated Cartography. Englewood Cliffs, NJ: Prentice Hall.
Prentice, I. C., Bartlein, P. J. and T. I. Webb, 1991. Vegetation and climate change in eastern North America since the last glacial maximum. Ecology 72(6): 2038-2056.
Schiavone, J.A. and T. V. Papathomas, 1990. Visualizing meteorological data. Bulletin of the American Meteorological Society 71(7):1012-1020.
Treinish, L.A., 1992. Correlative data analysis in the Earth Sciences. IEEE Computer Graphics and Applications May: 10-12.
van der Wel, Franz, R. M. Hootsmans and F. J. Ormeling, 1994. Visualization of data quality. in Visualization in Modern Cartography MacEachren, Alan and D. R. F. Taylor (ed). London: Pergamon, pp. 313-333.
Vislocky, R. and M. Fritsch, 1995. Improved model output statistics forecasts through model consensus, Bulletin of the American Meteorological Society 76(7): 1157-1163.
Wilhelmson, R., B.F. Jewett, et al., 1990. A study of the evolution of a numerically modeled severe storm. The International Journal of Supercomputer Applications 4(2):20-36.
Go back to working papers
or to Commission home page
or to MacEachren