The global observational network is known to be highly inhomogeneous a
nd fractally distributed in space. In this paper the spatial distribut
ion of stations in a network is studied in connection with the type of
dynamics exhibited by the atmospheric fields. The existence of an opt
imal distribution capable of capturing the complexity of this dynamics
with a small Prescribed error, and the role of deviations of the netw
ork distribution from a homogeneous one in this observational error ar
e explored. For sparsely distributed stations like those of the presen
t-day network, it is found that inhomogeneous distributions may genera
te a record that is closer to reality than homogeneous ones.