1. A Simple, Efficient Method for Realistic Animation of Clouds
Yoshinori Dobashi* Kazufumi Kaneda** Hideo Yamashita**
Tsuyoshi Okita* Tomoyuki Nishita***
*Hiroshima City University **Hiroshima University ***University of Tokyo

2. Contents Introduction and Motivation Simulation Rendering Results
Conclusion

3. Introduction

4. Problem Overview
Realistic modeling and animation of (cumulus-type) clouds
Two sub-problems:
Simulation of cloud formation, extinction and advection by wind
Rendering of the clouds, shadows and shafts of light

5. Previous Work - Simulation
Two categories of simulation methods:
Physical process of fluid dynamics
Very accurate
Computationally expensive
Heuristic approach (procedural modeling)
Computationally inexpensive
Easier to implement
Parameters needed

6. Previous Work - Rendering
Accounting for multiple scattering of light
Computationally expensive
Using 3-D textures for volume density
Does not handle atmospheric effects such as shafts of light
Rendering shafts of light using ray-tracing or a similar method

7. Goals Simple and efficient simulation method
Support of effects such as ...
Cloud color by single scattering of light
Shadows of clouds cast on the ground
Shafts of light through clouds
Hardware-accelerated rendering
Visually convincing result

8. Simulation – Basic Idea
Cellular automaton with binary states

9. Nagel‘s Method Water vapor turns to water to form clouds
Use Nagel‘s method to simulate cloud formation:
Divide 3-D space evenly into 3-D cells
Assign boolean variables to each cell:
cld indicates whether cell contain clouds
hum indicates whether cell has enough water vapor to form clouds
act indicates whether phase transition is ready to occur

10. Nagel‘s Method (cont‘d)
Cell properties in the current animation frame ti are used to compute the cell properties in the next frame ti+1:
hum(x, y, z, ti+1) = hum(x, y, z, ti) Ù Øact(x, y, z, ti)
cld(x, y, z, ti+1) = cld(x, y, z, ti) Ú act(x, y, z, ti)
act(x, y, z, ti+1) = Øact(x, y, z, ti) Ù hum(x, y, z, ti) Ù ¦act(x, y, z , ti)
¦act is a boolean function and its value is calculated by the status of act in the surrounding cells.

11. Cloud Extinction Extension to Nagel‘s method:
cld(x, y, z, ti+1) = cld(x, y, z, ti) Ù IS(rnd > pext(x, y, z, ti))
hum(x, y, z, ti+1) = hum(x, y, z, ti) Ú IS(rnd < phum(x, y, z, ti))
act(x, y, z, ti+1) = act(x, y, z, ti) Ú IS(rnd < pact(x, y, z, ti))
rnd : uniform random number
pext : probability of cloud extinction
phum : probability of vapor forming
pact : probability of phase transition occurence

12. Advection by Wind Clouds move, blown by winds
Wind velocity is different depending on the height from the ground
cld(x, y, z, ti+1) = cld(x – v(z), y, z, ti)
hum(x, y, z, ti+1) = hum(x – v(z), y, z, ti)
act(x, y, z, ti+1) = act(x – v(z), y, z, ti)
v(z) : wind velocity, piecewise linear function
Assumption: wind blows towards the direction of x-axis

13. Controlling Cloud Motion
Ellipsoids simulate air parcels
Vapor and phase transition probability:
higher at center / lower at edge
Cloud extinction probability:
Lower at center / higher at edge
Ellipsoids move in direction of wind
Different kinds of clouds by controlling ellipsoid parameters (sizes and position)

14. Fast Simulation using Bitfields
Each cell state (cld, act, hum) can be stored in a single bit
Low memory requirements
Fast computation of simulation process
Problem: Random numbers
Solution: Precalculated look-up tables

15. Rendering – Basic Idea Smoothing and volume rendering
Splatting method for clouds
Spherical shells for shafts of light

16. Continuous Density Distribution Calculation
Simulation output is a binary distribution
Continuous density distribution results from smoothing the binary distribution
Cloud density of a cell is the weighted average of the surrounding cells
Each cell contributes a density distribution over an effective radius (Metaballs)
Cloud density of an arbitrary point is therefore a weighted sum of a simple basis function

17. Metaball Billboards
Generate 2-D texture of metaballs

18. Rendering - Step 1
Set up parallel projection with sun at viewpoint and initialize framebuffer to 1.0
Place billboards at centers of metaballs with their normals toward the sun
Starting with billboard closest to the sun, project and blend billboards to framebuffer
Read back value at projected billboard center to get attenuation ratio between sun and metaball

19. Rendering – Step 1 (cont‘d)
After all metaballs have been projected, framebuffer contains shadow texture

20. Rendering – Step 2 Render all world objects except clouds
Place billboards at centers of metaballs with their normals toward the observer
Project and blend billboards to framebuffer starting with those farthest from viewpoint

21. Rendering – Step 2 (cont‘d)

22. Shafts of Light Render using spherical shells (made of polygons)
Modify Rendering – Step 2 to:
Calculate colors of vertices of shell polygons (atmospheric conditions)
Repeat for all shells (back-to-front):
Render shell k with additive blending function and shadow texture mapping
Render billboards between shell k-1 and shell k

23. Shafts of Light (cont‘d)

24. Results

25. Results (cont‘d)

26. Conclusion Advantages: Possible improvements:
Simulation requires little computation
Memory requirements are small
Rendering is fast by making use of graphics hardware
Shadows of clouds and shafts of light can also be rendered
Possible improvements:
Effects of terrain under clouds
Level of detail

27. The End
Questions?