We investigate a sequence of low-dimensional models of turbulent channel fl
ows. These models are based on the extraction of the Karhunen-Loeve (KL) ei
genfunctions from a large-scale simulation in a wide channel with R-* = 180
(based on the friction velocity). KL eigenfunctions provide an optimal coo
rdinate system in which to represent the dynamics of the turbulent flow. Th
e hierarchy of KL modes identifies the most energetic independent phenomena
in the system. A series of Galerkin projections is then used to derive a d
ynamical approximation to flows. In order to capture essential aspects of t
he how in a low-dimensional system, a careful selection of modes is carried
out. The resulting models satisfy momentum and energy conservation as well
as yielding the amount of viscous dissipation found in the exact system. T
heir dynamics includes modes which characterize the flux, rolls, and propag
ating waves. Unlike previous treatments the instantaneous streamwise flow i
s time dependent and represented by KL flux modes. The rolls correspond to
the streaks observed in experiments in the viscous sublayer. Propagating wa
ves which first appear in the model flow at low Reynolds number (R-* simila
r to 60) persist through the chaotic regime that appears as the Reynolds nu
mber is increased. Statistical measures of the chaotic flows which have bee
n generated by the models compare favorably with those found in full-scale
simulations.