Looking at the rivers from the previous post it seems that everything is quite pretty. But perhaps, like me, you feel that something is wrong. Oh ye, there are no such pretty smooth rivers in nature!
Let’s open google or other Earth maps and look at the shape of any river banks. Usually, the width of a river does not remain constant and gradually changes, even when the amount of water in it is invariably. Furthermore, the banks do not vary simultaneously. This, as well as the fact that such behavior of the banks does not depends on the width of the river, suggests that the shape of the river is a fractal (do not confuse this fractal with the fractal of the river system). All this is true for majority of small and medium-width rivers, but the widest lowland rivers often have more complex banks and have large well visible islands.
Here is a pair of random Earth rivers of medium width from two continents.
![]() |
![]() |
It is known that in nature there are almost no classical geometric forms, but fractal forms are very common.
So far we consider only the most frequent behavior of the river channel; ducts, deltas, islands, and old channels that have become oblong lakes we are not considering now. Although, the method proposed further, with some changes, can be applied to create models of these nature forms; at least I find it as promising.
Surely, there are already scientific works devoted to the formation of rivers and it would be possible to model river forms based on the strict physical theory. However, the shape of the mountains could be obtained by modeling based on the strict theory of lithospheric plates movement, weathering, and erosion; but, in order to obtain a plausible terrain for virtual worlds, no one does it (correct me if I’m wrong). For this purpose we use one of the algorithms of the Brownian surface generation.
Similarly, with rivers, we should try to get a fractal shape close to the real river forms, and then we could use it to represent procedural generated rivers on maps.
Again, carefully studying the real earth’s rivers, we can notice some special behavior of the river banks. Two banks most often do not diverge or taper simultaneously (it also happens, but less often). Everything looks so that one bank shore departs from the imaginary central line of the river or approaches to it. This phenomenon makes one think to put one more channel (river bad), different from the first but close to it, atop each other. Then we should get the right behavior of the banks.
So it turned out to be valid.
![]() |
![]() |
![]() |
The second channel is constructed by the random walk method. Each point of the second channel is obtained by shifting the corresponding point of the first channel by a certain vector, which is the same in absolute value (for the current river width), but its direction makes a random walk. When passing to each next point, we add a random value to the last angle value, and the random values is distributed according to the normal (Gaussian) distribution. The points thus obtained are also smoothed.
![]() |
![]() |
![]() |
Obtained channels must be merged in the resulting river polygon. How exactly I merge channels I will recount in the next post. For now a few examples from the planet Admete.
Click the images for bigger image size.