Hydrologic Network Design Methods and Shannon's Information Theory

Abstract

Various existing methods of hydrologic network design are reviewed. These methods are grouped into mean-square error, inter-station correlation and iso-correlation, regionalization, simulation, systems analysis, rational and the simplified form of Fisher's information approach. The applicabilities and limitations of each of the above approaches are discussed in context wvith the hydrologic network design and estimation objectives. The relative performance of each of the above approaches is then evaluated in selecting the optimum station locations and estimating hydrologic events at ungauged locations using simulated rainfall data. It is found that due to normality and linearity assumptions in above approaches, the existing methods are not very effective in identifying true optimum station locations. A broader and more universal methodology based on Shannon's information measure is proposed which is able to select the optimum station locations without the assumptions of normality and linearity. It also treats the network in true multivariate form. The methodology is then extended to estimate the events at ungauged locations on the basis of multivariate optimum information transmission criterion.