The analysis of a family of physically based landscape models leads to the
analysis of two stochastic processes that seem to determine the shape and s
tructure of river basins. The partial differential equation determine the s
caling invariances of the landscape through these processes. The models bri
dge the gap between the stochastic and deterministic approach to landscape
evolution because they produce noise by sediment divergences seeded by inst
abilities in the water flow. The first process is a channelization process
corresponding to Brownian motion of the initial slopes. It is driven by whi
te noise and characterized by the spatial roughness coefficient of 0.5. The
second process, driven by colored noise, is a maturation process where the
landscape moves closer to a mature landscape determined by separable solut
ions. This process is characterized by the spatial roughness coefficient of
0.75 and is analogous to an interface driven through random media with que
nched noise. The values of the two scaling exponents, which are interpreted
as reflecting universal, but distinct, physical mechanisms involving diffu
sion driven by noise, correspond well with field measurements from areas fo
r which the advective sediment transport processes of our models are applic
able. Various other scaling laws, such as Hack's law and the law of exceede
nce probabilities, are shown to result from the two scalings, and Horton's
laws for a river network are derived from the first one. (C) 2001 Elsevier
Science Ltd. All rights reserved.