Abstract: Most of the medium- and small-size rivers in China are located in hilly areas where hydrological stations are sparse and limited monitoring data are available, but these rivers can generate floods characterized by sudden, strong and great destructiveness. Thus, how to predict the floods in such rivers attracts much attention. This paper presents a flood prediction method for medium and small rivers for better consideration of the characteristics of these rivers in Liaoning Province. We collect and extract the basic data of underlying surface, hydrology, meteorology, etc. , and extract a new set of irrelevant indexes using the principal component analysis that can reduce the dimension and influence of related indexes. Then, a variable fuzzy model is used to classify and identify the hydrological similarity of 30 reference stations and 84 river stations. Finally, we construct a Topmodel model for prediction of distributed floods, and verify its effectiveness and practicability through comparison of its accuracies in flood predictions for the reference stations before and after parameter transplantation. Our results lay a basis for flood prediction of the medium- and small-size rivers in Liaoning Province, and are also useful to flood forecasting of other valleys short of hydrological data.

Key words: flood forecast, ungauged region, medium- and small-size rivers, parameter transplantation, regionalization method, Topmodel model, principal component analysis, variable fuzzy models