Each model by itself may look fascinating, but to really provide a realistic scene, a complete scenery must be created. It is most often the color that betrays the illusion of a real image. 24 bit per color representation is therefore a must when combining all of the previous techniques. High resolution is of course also a plus. However if one needs to make a tradeoff between color and resolution, color must be favored.
The creation of a complete scenery is pretty straight forward when all the tools are available. It is important to create everything using the same viewpoint, view window and scale. One must also remember to define the sun at the same spot for all algorithms. The easiest way is to just create everything in the order they appear from the horizon to the viewpoint. For every triangle that builds up the mountain can a set of plants be "planted".
All the methods described in this thesis provide a powerful set of tools for creating images of the nature. The most useful property of the generative algorithms is the power of the low bandwidth. One implementation could allow the user to simply state: "Sunny afternoon, beach, light wind" or "Cloudy evening, mountain forest". The user could also define the surface of the water, the shape of the mountain, plants, trees and the color-shade of the sky, and still only use a few thousand bytes to describe the scene down to the last pine needle.
As mentioned before, the compactness of the input parameters can be very useful when sending images over the internet. It would be interesting to see an implementation of these techniques and how fast a whole scene can be transmitted and displayed in comparison to sending the image as a normal graphics file. It would also be beneficial to add other natural phenomena: for example sun and rain modeling.
The low bandwidth needed to create highly detailed images is the most important aspect of this thesis. The most interesting is that this thesis has also shown that it is possible to use L-systems for more than just modeling plants. The L-system clouds can not replace any of the other techniques for modeling clouds, since they all model different types of clouds. These techniques could be used to complement each other to generate a variety of clouds. The L-system mountains display a much higher flexibility than the fractal mountains in terms of roughness of the surface, but is at the same time more limited in size. These techniques are also complementary to each other. Further research is suggested for generating different types of clouds and mountains using L-systems.
This thesis has mainly discussed the algorithmic aspects of modeling natural phenomena. The artistic aspects are just as significant. The most important step is to create a graphical user interface which is easy to use and understand and hides the complexity of the algorithms. When this is done will the real power behind the techniques in modeling natural phenomena be discovered.