Visually accurate multi-field weather visualization

Weather visualization is a difficult problem because it comprises volumetric multi-field data and traditional surface-based approaches obscure details of the complex three-dimensional structure of cloud dynamics. Therefore, visually accurate volumetric multi-field visualization of storm scale and cloud scale data is needed to effectively and efficiently communicate vital information to weather forecasters, improving storm forecasting, atmospheric dynamics models, and weather spotter training. We have developed a new approach to multi-field visualization that uses field specific, physically-based opacity, transmission, and lighting calculations per-field for the accurate visualization of storm and cloud scale weather data. Our approach extends traditional transfer function approaches to multi-field data and to volumetric illumination and scattering.

Visually Accurate Multi-Field Weather Visualization Kirk Riley Purdue University [e-mail] David Ebert Purdue University ebertd @ purdue.edu Charles Hansen University of Utah hansen @ cs. utah.edu Jason Levit University of Oklahoma [e-mail] A Figure 1: Time Series of a Cloud Scale Visualization Abstract Weather visualization is a difficult problem because it com­prises volumetric multi-field data and traditional surface-based ap­proaches obscure details of the complex three-dimensional struc­ture of cloud dynamics. Therefore, visually accurate volumet­ric multi-field visualization of storm scale and cloud scale data is needed to effectively and efficiently communicate vital information to weather forecasters, improving storm forecasting, atmospheric dynamics models, and weather spotter training. We have developed a new approach to multi-field visualization that uses field specific, physically-based opacity, transmission, and lighting calculations per-field for the accurate visualization of storm and cloud scale weather data. Our approach extends traditional transfer function approaches to multi-field data and to volumetric illumination and scattering. CR Categories: 1.3.7 [Computer Graphics]: Three-Dimensional Graphics and Realism-color, shading, and texture Keywords: Multi-Field Visualization, Visually Accurate Visual­ization, Weather Visualization 1 Introduction One of the most common applications of visualization techniques is weather visualization. The majority of these images are created for representing weather data at the global, or synoptic, scale. At this scale, atmospheric interactions are approximately two-dimensional, IEEE Visualization 2003, October 19-24, 2003, Seattle, Washington, USA 0-7803-8120-3/03/S 17.00 ©2003 IEEE and involve global scale phenomena. Less commonly, but no less importantly, weather events are examined and predictions are made at the storm, or even cloud scale, where observation and visual­ization are particularly useful for the evaluation and prediction of severe storms. Weather visualizations at the synoptic scale represent storms and fronts as they traverse the globe. At this scale, the data represents coarse scale quantities such as mean precipitation over large ar­eas; therefore, visualization techniques conveying particular values are popular. Isosurfaces, and two-dimensional colored representa­tions of weather fields convey particular data values to the observer. While these are effective when predicting large-scale weather pat­terns, they are ineffective when examining storm scale weather phe­nomena. Two-dimensional techniques cannot represent the highly three-dimensional event that is a forming severe storm. Although isosurfaces can represent exact data values in three-dimensions, they obscure the subtle details present in a storm. We present a new visually accurate visualization system based on the particle scattering properties of the constituent fields. Through the use of these properties, we provide essential visual cues that are lost in standard weather representations. This new visually ac­curate weather visualization system improves the evaluation and prediction of cloud and storm behaviors from both simulation and measured weather data. This system may also be applied to cloud microphysics model evaluations and weather observer training. 2 Motivation While representing particular data values with two-dimensional and isosurface coloring techniques in weather data sets is essential at the synoptic scale, it is often not the optimal approach to cloud and storm scale visualization. Meteorologists are trained to extract large amounts of information about a forming storm through observation in the field. Therefore, comprehension of storm scale data is max­imized by presenting it in a visually accurate fashion. For exam­ple, a thunderstorm undergoes significant visual changes through­out its lifecycle, as it grows from a cumulus cloud, develops a cir­rus anvil, and eventually develops rain, hail, and possibly severe weather. Rendering of numerical clouds to make convective clouds look "puffy," or cirrus clouds look "wispy," is important to help 279 Proceedings of the 14th IEEE Visualization Conference (VIS'03) 0-7695-2030-8/03 $ 17.00 © 2003 IEEE Authorized licensed use limited to: The University of Utah. Downloaded on July 06,2010 at 21:35:50 UTC from IEEE Xplore. Restrictions apply. Figure 2: Isosurface Rendering of Storm Scale Data rate interpretation of multi-field weather data based on the individ­ual particle properties. Volume rendering has been an important tool in rendering and visualization, stalling with [Drebin et al. 1988J. [Ebert and Parent 1990J extended these techniques to gaseous volume rendering. [En­gel et al. 2001J have concentrated on hardware accelerated volume rendering, while [Kniss et al. 2003J have focused on light trans­port models in volumetric systems. [Jensen and Christensen 1998J discuss realistic image generation with photon maps. Procedural models have also been applied to rendering atmo­spheric phenomena. [Stam and Fiume 1995; Stam 1995J have dis­cussed the modeling of gaseous phenomena with warped blobs, and the importance of diffusion in cloud rendering. [Ebert et al. 2003J describe many procedural techniques for cloud rendering. a meteorologist determine the character of particular atmospheric features and structures. While isosurface methods can convey par­ticular values of these multiple fields, they cannot represent the tur­bulent inhomogeneous mass of multiple-field weather. Inspection of a hai'd isosurface rendering of a severe storm from Vis5d [Hi­bbard and Sante 1986J in Figure 2 does not clearly communicate the structure of the storm because visual relationships among the various fields are obscured. To achieve visual accuracy, we present a system based on the light scattering properties of the atmospheric particles. The vari­ous component particles in storms and clouds scatter light differ­ently. Images that do not account for the different interactions of the particles with light do not provide believable visual cues to the observer. The scattering properties of these particles form the basis of the transfer function mapping from data values to the observed color and opacity. Realistic weather visualization has many applications. Visually accurate representations will increase the effectiveness of storm scale atmospheric analysis and severe weather forecasting, improve the training of weather observers and students, and enhance the for­mulation, parameterizations, and physics of numerical weather pre­diction models. 3 Previous Work Weather visualization has been an active area of research for many years [Papathomas et al. 1988; McCaslin et al. 2000; Trembil- ski 2002; Kniss et al. 2002J. One of the most widely used tools for weather visualization is the Vis5d system [Hibbard and Sante 1986J. Other important weather visualization systems are being de­veloped by researchers at Georgia Tech [Tian-yue et al. 2001J and at IBM [Treinish 1997J. While these tools are useful in weather prediction and visualization, they do not provide visually accurate images needed for cloud scale and storm scale analysis. Atmospheric rendering has also been an active research area, stalling with the early low-albedo illumination model [Blinn 1982J. Extensions to this work in [Kajiya and Von Herzen 1984J, showed the importance of multiple scattering in volumes. Unfortunately, fully realized multiple scattering systems are veiy computationally expensive. Advanced scattering and illumination methods have been developed [Max 1995; Klassen 1987; Nishita et al. 1996J. [Preetham et al. 1999J described an atmospheric scattering model for rendering daylight. A high-performance hardware accelerated forward scattering model, using impostors for the cloud render­ing, and simulating cloud formation was presented in [Harris et al. 2003; Harris and Lastra 2001J. [Dobashi et al. 2002J developed atmospheric models for viewing the earth from space. While this research has developed techniques for rendering and illuminating atmospheric bodies, they do not provide a system for visually accu- 4 Visualization Methodology Multi-Field Data Values Volume Particle Concentration Volume Rendering Rendering Equation Illumination Image Figure 3: Weather Visualization System Flow This system renders weather particle fields based on the particle properties. A diagram describing the structure of our rendering sys­tem is shown in Figure 3. To visualize these particles in an accurate manner, we first translate the input data fields into particle concen­trations. We discuss meteorological data and how to translate it into particle concentrations in Section 4.1. Volume rendering techniques allow us to represent a continu­ous field through transfer function mapping of data values to colors and opacities that correctly reveal the spatial relationships between thick, opaque structures and thin, wispy structures. Therefore, vol­umetric techniques are necessary to render multi-field data in a vi­sually accurate manner. The volume model for weather data is pre­sented in Section 4.2. Additionally, by visualizing the volume using the scattering of the individual particles, this system provides subtle visual cues crucial to understanding the composition of the storm as presented in Section 4.3 and Section 4.4.2. Illumination is an essential cue for understanding the structure of three-dimensional storm fields. The high-albedo nature of cloud particles makes accurate illumination models difficult to apply at reasonable frame rates. Fortunately, phase function calculations show that large cloud particles heavily favor forward scattering, and hence a tractable translucency model [Kniss et al. 2003J is imple­mented. We present our illumination model in Section 4.4. Storm models vary in their resolution and detail. Some applica­tions of this system, such as the training of weather observers and 280 Proceedings of the 14th IEEE Visualization Conference (VIS'03) 0-7695-2030-8/03 $ 17.00 © 2003 IEEE Authorized licensed use limited to: The University of Utah. Downloaded on July 06,2010 at 21:35:50 UTC from IEEE Xplore. Restrictions apply. Variable Definition Hydrometeor Color Pair______________°f :dr _____________________ Piieid Density, v"}^e, of individual particles in field ^ volume* Volume of region of air under consideration m3 7]lldd Particle concentration, pJvo^|ld‘l T (t. fj Light attenuation between points s and I L| (?. Q) Light contribution at point s in Q direction r Optical depth Pex Extinction coefficient i 8,!;eld Extinction coefficient for aiven field -J- i ca m a!'eld Extinction cross-section for aiven field -ir tfA_____________________________________1X1" P(f2) Phase function CP(Q) Cumulative phase function L]t Intensity of the light at source Table 1: Variables List even the evaluation of weather models, benefit from the capability for the user to introduce additional details. Therefore, we provide the user with the ability to add procedural details to storm and cloud scale weather visualization as described in Section 4.5. Cloud Red Ice Magenta Rain Blue Snow Yellow Graupel Green Table 2: Equivalent Spherical Particle Radii Given the hydrometeor mass ratio and some properties of the particles, we determine the particle concentration. We define the particle concentration j]lleld as the particles per unit volume of that field. We assume the volume of air is much larger than the total volume of the particles. Therefore, the volume of the mixture is approximately the volume of the air. We calculate the total mass of all particles of a given particle field (tmiidd) as the following: tmiidd = V0lumeuirpuirHnt;|d (1) The total particles per volume is then expressed in terms of the mass of a single particle of that field, pm,le|ct. rjneldvolumeuir = (2) Pmiieid Pmiieid is defined as: Pmiieid = volume^ltc!cleplidd (3) 4.1 Meteorological Data A storm is comprised of many water particles of various states (ice/water), sizes, and habits (or shapes). Hydrometeor particles are the large (compared to air) water particles that create a storm cloud. The hydrometeor fields we currently consider in this model are cloud, ice, rain, snow, and graupel (soft hail). To properly ren­der the multiple hydrometeor field data, we must determine how these particles interact with light. This relationship is dependent on the geometry of the particles and the concentration of those parti­cles at a point in space. In this section, we focus on the issue of particle concentration. Hydrometeor fields are often stored as mass ratios (Hfleid = "nms dr° where field specifies the hydrometeor field (cloud, rain, ice, snow, graupel, or vapor). In order to simplify our discussion, a list of variables is given in Table 1. An image showing the hydrom­eteor composition of a storm cloud is given in Figure 4. The color mapping is given in Table 2. Figure 4: Cloud with Colored Hydrometeor Fields The individual particle volume, volume^1}-^, is a function of the atmospheric conditions and the particle geometry of the given field. We now solve for the particle concentration of the given field. jj field _ Paii-Hfield ^ VOiumc^ePfield Now that we can translate the particle mass ratios into the concen­trations of the various particles, we discuss the volume rendering model used in the system. 4.2 Volumetric Representation of Hydrometeor Particles To properly render the various hydrometeor elements, a transfer function is applied to translate these ratios into opacities. Our volume rendering equation then determines the final color of each pixel, L(w) using the following equation [Nishita et al. 19961: L(tf') = T (0. w) Lbg + /" Ti.v.ir>/3,c;1i.v'> / P(v/(Q))L,(?.Q)dQd? (5) JO jAk T (s.w) is the light attenuation between points s and w. fiica (?) is the scattering coefficient. P {y/ {O)) is the scattering phase func­tion as a function of the angle between the incident light and the viewpoint, iequal to zero means the light is directly be­hind the viewpoint. L(,g is the background light intensity. L| (s. SI) is the light at the point s, in the direction fl Section 4.3 describes the calculation of T (s. w) for multi-field hydrometeor elements, and Section 4.4 discusses our translucency model used for calculation of L| (?.Q). 281 Proceedings of the 14th IEEE Visualization Conference (VIS'03) 0-7695-2030-8/03 $ 17.00 © 2003 IEEE Authorized licensed use limited to: The University of Utah. Downloaded on July 06,2010 at 21:35:50 UTC from IEEE Xplore. Restrictions apply. Hydromctcor Equivalent Radius mm Cloud 0.01 Icc 1 Rain 1 Snow 2 Graupcl 2.5 Tabic 3: Equivalent Sphcrical Particlc Cross Sections 4.3 An Optical Model for Multiple Hydrometeor Fields Wc now calculate the attenuation term of Equation 5 using the par­ticlc concentration calculated in Scction 4.1. The intensity attenu­ation as wc traverse the cloud. T (s, w), for a given sample is often described in terms of the optical depth r [Fu and Liou 1993; Blinn 1982; Nishita ct al. 1996; Max 19951. T(?,tf) = e-r (6) For an inhomogeneous volume, this optical depth is the integral of the extinction cocfticicnt at that point in spacc with rcspcct to distance. T (?, tf) = -Is (?/)df/ (7) Numerically, wc approximate the optical depth of a sample in terms of the extinction cocfticicnts A!jf,c,(?) of the hydromctcor fields and the width of the sample As ([Key ct al. 20021). Sample- ( X \AJJ Fields / As is the length of the current sample. The extinction cocfticicnt for a given field j8jje,d is a function of the average particlc extinction cross scction for the given field (crt!ie,ci^ and the particlc concentra­tion of that field r),le,d [Nishita ct al. 19961. P^ki = (9) For icc and clouds, wc assume the singlc-scattcring albedo to be approximately 1 [Key ct al. 20021 [Fu and Liou 19931. Thus, by Equation 10 and Equation 11. the extinction cross-scction is domi­nated by the scattering cross scction. Albedo = \-- (10) Oext = Cabs + ^sca HI) The scattering cross scction. for hydromctcor particlcs substan­tially larger than the wavelength of light, is approximately twicc the average geometric cross scction [Liou ct al. 1991; Bohrcn and Huff­man 19831 of the particlcs. Cloud and rain particlcs arc approx­imately sphcrical; therefore, the extinction cross scction estimate for these particlcs is twicc the area of a circlc with radius equal to the particlc radius. Although many hydromctcor particlc shapes ex­ist. particularly in icc fields [Key ct al. 20021. wc use an equivalent sphcrical cross scction for all the particlcs. A tabic of approximate cross sections is given in Table 3 [Pruppachcr and Klctt 20001. Rendering the various fields according to their combined extinc­tion is essential to understanding the structure of the cloud in a vi­sually accurate system. Comparing images with different j8ex for the various fields in Figure 6 and equal extinctions for all fields in Figure 5. wc see a dramatic difference. In summary, wc have developed the calculation of the extinction for each of the fields This enables the calculation of the Figure 5: Improperly Sealed Hydromctcors Low-Albcdo Figure 6: Properly Sealed Hydromctcors Low-Albcdo overall optical depth by Equation 8. which is then used to calculate the overall attenuation factor T(s',w) in Equation 5. Now. with the attenuation cocfticicnt available, wc focus our attention on the model for L| (s',£1). 4.4 A Physically Inspired Illumination Model for Cloud Rendering In this model, there arc two primary aspccts of illumination. The first is the light transport model, i.e.. how wc calculate L/ls). The sccond is the overall phase function at s. Wc first describe our light transport approximation, then wc discuss how wc apply the phase function for the combination of hydromctcor fields. 4.4.1 Translucent Light Transport Model Light transport calculations determine the extinction of light as the combination of absorption and outscattering of light, based on the absorption and scattering cross sections of the particlcs. Because of the high-albcdo nature of clouds, the absorption cross scction is negligible, but the outscattering is quite high. If wc consider all the photons that strike a cloud particlc as lost to outscattering (the low-albedo model [Blinn 19821). wc sec in Figure 6 that the clouds become dark too quickly. To account for multiple scattering wc need to use an appropriate phase function model. If wc consider one phase function model, the Comette and Shanks model [Cornctte and Shanks 19921. and its cumulative phase function shown in Figure 7. wc sec that for­ward scattering is overwhelmingly dominant in small cloud parti­clcs. In fact. 90 % of the light is scattered within about 10 degrees. Although the phase function varies from field to field, all particlcs arc large compared to wavelength, and. therefore, scatter predom­inantly in the forward direction. Therefore, a model similar to the transluccncy model described in [Kniss ct al. 20031 is appropri­ate for clouds. This model creates sampling slices with a normal 282 Proceedings of the 14th IEEE Visualization Conference (VIS'03) 0-7695-2030-8/03 $ 17.00 © 2003 IEEE Authorized licensed use limited to: The University of Utah. Downloaded on July 06,2010 at 21:35:50 UTC from IEEE Xplore. Restrictions apply. halfway between the eye and light vectors, thus taking samples that arc well conditioned to both the eye and light viewpoints. Two two­dimensional buffers arc used, one for eye compositing, and one to store the attenuation of light. The eye-pass samples the light buffer through use of the render-to-texture OpenGL extension at the frag­ment level to determine correct liahtina for the fraament. Cornette and Shanks Phase and Cumulative Phase Figure 7: The Cornette and Shanks Phase (normalized) and Cumu­lative Phase for Clouds Knowing that scattering in the cloud is predominantly forward scattering, we develop an approximation to calculate L|(.y). For our illumination model, we break the light into three categories: unscattered light, forward scattered light, and outscattered (scat­tered in any non-forward direction) light. Unscattered light is the unimpeded light in the current sample. The forward scattered light consists of light that strikes a particle in the given region but still forward scatters. The outscattered light is the light that strikes the particle, scatters anywhere else, and is effectively extinguished. In our model, we define a forward scattering angle 0 such that all light scattered within that arc towards the destination is considered for­ward scattered. This arc is shown for the per-pixel light buffer cal­culations in Figure 8. The light that contributes to the eye direction lighting as presented in Equation 5 is given in Equation 12 [Nishita et al. 19961. L^v.O,,) = L„T (?,!() + T (s'Jf pXn (s') / P(0)L| (AOit+Q) d£Mv" (12) J 0 The first term represents the unscattered light and the second term represents the contribution of light due to forward scattering over the 0 arc angle, where L|, is the intensity of the light source. Note that the assumption of high-albedo implies j3scu = fieyL. Despite the indirect peripheral contribution, we assume 0 is small enough that if/(SI) is approximately constant over the arc. This allows us to factor the P(y/(S2)) term in Equation 5, treating it as a directional light. For efficiency, we quantize this arc into two regions: a dj/4 center region and a 0j / 4 to Of/ 2 outer region. We approximate the light in the center region as constant and equal to the center light contribution, and the light in the outer band as constant and equal to the average of four equally spaced samples around the band. In Figure 8, we show the calculation regions. The blue line represents unscattered light, the green lines the 0/4 region, and the red lines Figure 8: Per Pixel Calculation Regions for Translucency the 0/2 region. We are thus integrating the product of the phase and the light over a small forward angular region. The cumulative phase function, CP(0), is the integral of the phase function. We calculate the light contribution to the light buffer in terms of this cumulative phase, the transparency in the center region.Tctr , the transparency in the outer region, Tper. and the light contributions in the inner and outer arcs, L™r and Lper. CP{0)= / P(0)d0 J 0 The unscattered propagating light is: Lunsc - L^rTctr The center region forward scattering is: LfOTsc = L^ti-( 1 - Tctr)CP (13) (14) (15) The peripheral region forward scattering in terms of the mean pe­ripheral light: Lper is then: Mbrsc Lper( 1 Tper) cpit)-cp(t (16) Equations 15 and 16 combine to form the second term in Equa­tion 12. The final light propagating to the next pixel is then given in Equation 17 below. L| ^It) - Ltmsc + LfJsc + Lj?"c (17) To maintain calculation tractability, we reduce volume texture reads by assuming that the extinction for center and for peripheral light is approximately the same: Tctr « Tper. This equation is applied to the system using techniques similar to [Kniss et al. 2003]. An image with translucency illumination is given in Figure 9. Notice the softness caused by the blurring and the prevention of artificial low-albedo darkening in the cloud. 283 Proceedings of the 14th IEEE Visualization Conference (VIS'03) 0-7695-2030-8/03 $ 17.00 © 2003 IEEE Authorized licensed use limited to: The University of Utah. Downloaded on July 06,2010 at 21:35:50 UTC from IEEE Xplore. Restrictions apply. Figure 9: Cloud with Translucent Illumination 4.4.2 Particle Dependent Phase Functions The scattering of a particle is largely dependent on its shape and size. As water particles increase in size, accurate particle scatter­ing calculations change from a predominantly Rayleigh scattering model to a Mie scattering model [Boliren and Huffman 1983]. Mie scattering favors greater forward scattering as the particle size in­creases significantly above the wavelength of the light. Addition­ally, the scattering becomes less wavelength dependent. Exami­nation of cloud water phase functions and ice water phase func­tions [Wendling et al. 1979] reveals that cloud water has higher for­ward scattering, while ice particles have more side and back scat­tering. To model this effect, we apply different phase functions to water (cloud and rain) and to ice fields (ice, snow, graupel). Phase is applied to the illumination of the system as given in Equation 12. These phase functions are based on the calculated values in Wendling, Wendling, and Weickman [Wendling et al. 1979]. Each phase function is calculated based on the relative concentration of each field, weighted by the field's extinction coefficient [Key et al. 2002], Phasevoxe|(0) = )fe'd *18> Note that the dominance of forward scattering, and the limitations imposed by modeling the sunlight as an RGB color, requires the phase function to be normalized over its dynamic range and the peak at zero degrees truncated. A plot of the normalized phase functions used in this simulation is given in Figure 10. For compar­ison, a rendering of the system with particle based phase is given in Figure 11(b). Compared with Figure 11(a), we can see subtle differences between the ice fields in the center and top of the cloud, and the water fields to the outside and bottom of the cloud. Phase Function For Water and Ice Normalized by Dynamic Range Phase Angle (Degrees) Figure 10: The Normalized by Dynamic Range Phase Functions for Ice and Water Figure 11: Illumination with (a) Same Phase for All Particles and (b) Particle Specific Phase with a Perl in noise texture [Perl in and Hoffert 1989], higher lev­els of detail are realized. Comparison of Figures 12 and 13 with Figure 4 shows how various noise levels in the different fields pro­vide insight to the composition of the cloud. Figure 14 shows more natural viewing conditions of the storm simulation, where we are looking up at the storm cloud. 5 Results 4.5 Procedural Detail for Visual Accuracy Cloud and weather modeling systems are limited in their ability to cover all appropriate scales of detail. Large storm scale models (WRF, ARPS) effectively model the large structures of the cloud, but cannot calculate the behavior at smaller scales. Smaller cloud and microscale models can determine particle behavior, but cannot tractably model larger weather phenomena. Large phenomenolog­ical rendering can be enhanced by adding procedural details based on known cloud behavior to emphasize the different fields of the cloud. For example, we know that the cloud portion of a form­ing storm should be billowing, while the ice portion, as the cloud reaches the stratosphere, should be thin and wispy. Adding proce­dural details that differentiate fields based on these behaviors can be useful for model evaluation to determine what detail is missing, and to make more convincing images for the training of weather spotters. By multiplying a user adjustable portion of the data field For interactive visualization, it must be possible for the meteorolo­gist to adjust the relative mixing ratios of the various fields at will. This system has been applied both to storm scale simulations, Fig­ures 9, 11, 12, and to smaller cloud scale simulations, Figure 1. The storm level data was generated from the weather research and fore­casting (WRF, http://www.wrf-model.org) model. This storm is ap­proximately 20,000 meters tall and 100,000 meters long. This par­ticular simulation is a splitting supercell storm. Through visually accurate visualization we determined, from the unrealistic smooth­ness of the cloud in the image, that the data generated from that simulation lacked medium scale turbulence that should have been present. The smaller cloud model represents a small cumulus cloud capable of light precipitation that is approximately 6500 meters tall and 6500 meters wide. It was created with a large-eddy simulation from the Straka Atmospheric Model [Straka and Anderson 1993], as modified in [Carpenter et al. 1998], and initialized with the pa­rameters described in [Lasher Trapp et al. 2001], Figure 1 shows 284 Proceedings of the 14th IEEE Visualization Conference (VIS'03) 0-7695-2030-8/03 $ 17.00 © 2003 IEEE Authorized licensed use limited to: The University of Utah. Downloaded on July 06,2010 at 21:35:50 UTC from IEEE Xplore. Restrictions apply. Figure 12: Large Scale Model with Additional Detail Figure 13: Large Scale Model with Additional Detail three steps in the time series of the formation of a cumulus cloud. To make adjustment of individual field (phase and extinction) properties available, without the need for costly recalculation of the data set, we use modem graphics hardware with per fragment cal­culation capability. The system presented here was implemented on an nVidia GeForceFX 5800 Ultra, using their Cg (C for graph­ics) compiler for vertex and fragment programs. Both the eye and light buffers require one vertex and one fragment program to eval­uate Equations 5, 12 and 18 per fragment. This system provides great flexibility, but higher complexity options require highly com­plex fragment programs, which reduce performance. A table of the frame rates is given in Table 4 for a 300 x 300 image with 128 sampling planes (1 per voxel). Interactive rates are maintained for the less complex modes of the system, allowing the user to adjust settings in a fast mode, then examine the data with more advanced options. 6 Conclusions Figure 14: Bottom View of Large Scale Model with Additional De­tail systems are useful for weather data because of their capability to show varying degrees of opacity in inhomogeneous cloud systems. Our rendering system utilizes the individual extinction and scatter­ing of these particles to produce a realistic representation that also provides insight into the structure of the cloud. Already, this system has been useful in determining missing turbulence components in one of the simulated models. These particle properties also reveal that forward scattering is dominant in hydrometeor particles, thus a translucency model is appropriate to the illumination of clouds. To improve differentiation and photorealism of cloud quantities, noise details can be added that provide additional cues to the user. Mod­ern graphics hardware programs make flexible calculations possible at the fragment level, with varying costs in frame rate. Acknowledgements The authors wish to thank Professor Sonia Lasher-Trapp for her aid with the micro-scale cloud data set. Special thanks to Dr. Jerry Tessendorf for his many insightful suggestions. The authors appre­ciate Nikolai Svakhine's volume renderer contribution. We would also like to thank Martin Kraus for his feedback and the reviewers for their suggestions and helpful comments. This work has been supported by the US National Science Foundation under grants: NSF ACI-0222675, NSF ACI-0081581. NSF ACI-0121288. NSF IIS-0098443, NSF ACI-9978032. NSF MRI-9977218. NSF ACR- 9978099 and the US Department of Energy's VIEWS program. We have developed a new visually accurate multi-field weather vi­sualization system that effectively conveys fine volumetricly vary­ing atmospheric data detail, improves the assessment of weather models, aids the training of weather observers, and presents more complete information in an intuitive style. Volumetric rendering Mode Frame Rate Uniform Phase Low-Albedo Light 5.1 fps Uniform Phase Translucent Light 4.28 fps Per Field Phase, Translucent Light 1.68 fps Per Field Phase, Noise Details, Translucent Light 1.18 fps Table 4: Rendering Speeds (in frames/second) for various modes References Bunn, J. 1982. Light reflection functions for simulation of clouds and dusty surfaces. Computer Graphics 8, 3 (July), 21-28. Bohren, C. F., and Huffman, D. R. 1983. Absorption and Scattering of Light by Small Particles. John Wiley and Sons. Carpenter, R. L. J., Droegemeier, K. K., and Bi.yth, A. M. 1998. Entrainment and detrainment in numerically simu­lated cumulus congestus clouds, part i: General results. J. Atmos. Sci. 55. 3417-3432. CORNETTE, W. M., AND Shanks, J. G. 1992. Physically reason­able analytic expression for the single-scattering phase function. Applied Optics 32. 16 (June), 3152-3160. Proceedings of the 14th IEEE Visualization Conference (VIS'03) 0-7695-2030-8/03 $ 17.00 © 2003 IEEE 285 Authorized licensed use limited to: The University of Utah. Downloaded on July 06,2010 at 21:35:50 UTC from IEEE Xplore. Restrictions apply. Dobashi, Y., Yamamoto, T„ and Nishita, T. 2002. Inter­active rendering of atmospheric scattering effects using graphics hardware. Proc. Graphics Hardware. 99-108. Drebin, R., Carpenter, L., and Hanrahan, P. 1988. Vol­ume rendering. Computer Graphics (SIGGRAPH '88 Pore.) 22 (Aug.), 65-74. Ebert, D. S., and Parent, R. E. 1990. Rendering and an­imation of gaseous phenomena by combining fast volume and scanline a-buffer techniques. Computer Graphics 24. 4 (Aug.). 357-366. Ebert, D., Musgrave, K., Peachey, D., Perun, K., and WORLEY, S. 2003. Texturing and Modeling A Procedural Ap­proach. 3 ed. Morgan Kaufman Publishers. Engel, K., Kraus, M., and Ertl, T. 2001. High-quality pre­integrated volume rendering using hardware-accelerated pixel shading. Proceedings of the ACM Siggraph/Eurographics Work­shop on Graphics Hardware 2001. 9-16. Fu, Q., and LlOU, K. N. 1993. Parameterization of the radiative properties of cirrus clouds. Journal of the Atmospheric Sciences 50, 13, 2008-2025. Harris, M. J., and Lastra, A. 2001. Real-time cloud render­ing. Eurographics 2001 Proceedings 20. 3 (Sept.), 76-84. Harris, M. J., Baxter III, W. V., Scheuermann, T., and Lastra, A. 2003. Simulation of cloud dynamics on graphics hardware. Proceedings of Graphics Hardware 2003 (to appear). Hibbard, W., and Sante, D. 1986. 4-d display of meteoro­logical data. Proceedings of 1986 Workshop on Interactive 3D Graphics. 129-134. Jensen, H., and Christensen, P. 1998. Efficient simulation of light transport in scenes with participating media using photon maps. Proceedings SIGGRAPH '98, Computer Graphics Proc, Ann Conf Series, 311-320. Kajiya, J. T., and Von Herzen, B. P. 1984. Ray tracing vol­ume densities. Computer Graphics 18. 3 (July), 165-174. Key, J. R., Yang, P., Baum, B. A., and Nasiri, S. L. 2002. Parameterization of shortwave ice cloud optical properties for various particle habits. Journal of Geophysical Research 107. D13 (July), 4181. KLASSEN, V. R. 1987. Modeling the effect of the atmosphere on light. ACM Transactions on Graphics 6, 3 (July), 215-237. Kniss, J., Hansen, C., Grenier, M., and Robinson, T. 2002. Volume rendering multivariate data to visualize meteo­rological simulations: A case study. IEEE Visualization Sympo­sium. }89~m. Kniss, J., Premoze, S., Hansen, C., Shirley, P., and McPherson, A. 2003. A model for volume lighting and mod­eling. IEEE Transactions on Visualization and Computer Graph­ics 2003 9. 2 (April-June), 109-116. Lasher Trapp, S., Knight, C., and Straka, J. 2001. Echoes from ultragiant aerosol in a cumulus congestus: Modeling and observations. J. Atmos. Sci. 58. 3545-3562. Liou, K„ Lee, J., Ou, S., Fu, W„ and Takano, Y. 1991. Ice cloud microphysics, radiative transfer and large-scale cloud processes. Meteorology and Atmospheric Physics 46. 41-50. Max, N. 1995. Optical models for direct volume rendering. IEEE Transactions on Visualization and Computer Graphics /, 2 (June), 99-108. McCaslin, P. T., McDonald, P. A., and Szoke, E. J. 2000. 3d visualization development at NOAA forecast systems labora­tory. Computer Graphics 34. 1 (Feb.), 41-44. Nishita, T., Dobashi, Y., and Nakamae, E. 1996. Display of clouds taking into account multiple anisotropica scattering and sky light. Proceedings of the 23rd Annual Conference on Com­puter Graphics and Interactive Techniques. 379-386. Papathomas, T., Schiavone, J., and Julesz, B. 1988. Ap­plications of computer graphics to the visualization of meteoro­logical phenomena. Computer Graphics 22. 4, 327-334. PERLIN. K.. AND HOFFERT. E. 1989. Hypertexture. Computer Graphics 23. 3 (July), 253-262. Preetham, A. J., Shirley, P., and Smits., B. E. 1999. A practical analytic model for daylight. In Siggraph 1999, Com­puter Grahics Procedings. Addison Wesley Longman, Los An­geles, A. Rockwood, Ed., ACM Siggraph, 91-100. Pruppacher, H. R., AND Klett, J. D. 2000. Microphysics of Clouds and Precipitation. 2 ed., vol. 18. Kluwer Academic Publishers. Stam, J., and Fiume, E. 1995. Depiction of tire and other gaseous phenomena using diffusion processes. SIGGRAPH 95 Conference Proceedings, Annual Conference Series (Aug.). 129-136. Stam, J. 1995. Multiple scattering as a diffusion process. Proceed­ings of the 6th Eurographics Workshop on Rendering, Dublin, Ireland (June), 51-58. Straka, J. M., and Anderson, J. R. 1993. Numerical simula­tions of microburst-producing storms: Some results from storms observed during cohmex. J. Atmos. Sci. 50. 1329-1348. Tian-yue, J., Ribarsky, W., Wasilewski, T., Faust, N., Hannigan, B., and Parry, M. 2001. Acquisition and display of real-time atmospheric data on terrain. Eurographics-IEEE Vi­sualization Symposium 2001. 15-24. Treinish, L. A. 1997. Regional weather forecasting in the 1996 summer Olympic games using an IBM SP2. Proceedings of the AMS 13th UPS (Feb.). Trembilski, A. 2002. Transparency for polygon based cloud rendering. Proceedings of the 2002 ACM Symposium on Applied Computing. 785-790. Wendling, P., Wendling, R., and Weickman, H. K. 1979. Scattering of solar radiation by hexagonal ice crystals. Applied Optics 18. 15 (Aug.), 2663-2671. 286 Proceedings of the 14th IEEE Visualization Conference (VIS'03) 0-7695-2030-8/03 $ 17.00 © 2003 IEEE Authorized licensed use limited to: The University of Utah. Downloaded on July 06,2010 at 21:35:50 UTC from IEEE Xplore. Restrictions apply.