A discrete state space model for a multiple-reach river system is formulated using the dynamics of biochemical oxygen demand(BOD) and dissolved oxygen(DO). A hierarchical optimization technique, which is applicable to large-scale systems with time-delays in states, is also described to control stream quality in a river as an optimal manner based on the interaction prediction method. The steady state tracking error of the proposed method is determined analytically and a necessary and sufficient condition on which a constant target tracking problem has zero steady-state error is derived. Computer simulations for the river pollution model illustrate the algorithm.