Return to the previous page. Paul Fearing's home page.

  Computer graphics is continually striving to emulate the appearance of the real world. Some of the most interesting problems deal with ``natural phenomena'', loosely defined as objects created by Nature. Fournier [Fournier89] points out some of the challenges of modelling natural phenomena, including: 
  • very complex object descriptions that sometimes cannot be easily described using surfaces (e.g. smoke, fire)
  • potential for reuse in large quantities, such as leaves on a tree, or trees in a forest
  • motion that does not often fit easily into a system of rigid transformations, and may need to be built specifically into the model
  • dependence on the natural world  that makes us biologically alert to its appearance, and allows us to be especially good at picking out visual discrepancies
Falling snow and its subsequent accumulation are a prime example of a natural phenomenon. A single snow layer is simultaneously absurdly complex on the local scale and apparently simple on the global level. A smooth, even blanketing of snow consists of a multitude of crystals, each one moving, bonding, changing, and contributing to the overall effect. When snow moves, it can slip, fold, creep, fracture or avalanche - travelling like a single handful of flour or an acre of solid concrete. This complexity is both intrinsically interesting, and important in the way it might give us new insights into how other natural phenomena might eventually be modelled. Snow crosses the conceptual boundary between fluids and solids, smooth and rough surfaces and predictable and unpredictable paths of motion, making it simultaneously applicable to other work, yet distinctly its own problem. 
Automatic snow generation saves animators work.
Click to enlarge. Click to enlarge.
To date, there has been no previous published computer graphics research on thick snow pack modelling, and only two papers on snow pack rendering, which clearly makes it an appropriate and worthy topic of research. Beyond novelty, there are other reasons for learning about snow.  In many countries, snow is a  common fact of life during the winter months. For example, January snow coverage in the Northern Hemisphere has ranged between 41.7 - 49.8 million square kilometres , or nearly half of the hemisphere's total land mass. A phenomenon that is so common and pervasive to most of us is clearly of interest and importance. 
Click to enlarge.
If we move beyond simulating snow for its pure challenge and ubiquity, we can see that there are many practical advantages to a working algorithm. Without an automatic model of snowfall, artists must use natural intuition to produce snow covered surfaces - an extremely time consuming task, depending on how realistic one wants the final result. A single tree might have a hundred branches, each with a complex drapery of snow, and each showering snow down onto branches below. Even if one painstakingly created every surface by hand, the image would still likely be very far from what Nature creates in a few hours.  An automatic snowfall algorithm would simultaneously save time, and open up computer generated effects to the possibilities of a whole new season. 

Besides the practicalities of research and application, there is another reason for investigating snowfall. Snow transforms commonplace scenes into fantastic wonderlands, greatly changing the appearance and mood of the landscape. It is our desire to be able to add snow to some of the classic images produced throughout the development of computer graphics - to see familiar scenes in a fresh, exciting way.