Analysis and simulation to the relation between rainfall and runoff in karst basin with artificial neural network model

Rainfall-runoff response in karst basin is a non-linear process. Determination of the major factors influencing underground river flow by proper non-linear analysis method is very important for simulating karst hydrological processes. In this study, the observed rainfall and flow discharge data in the Houzhai catchement of Puding County was used. The BP model structure of two hidden layers and three inputs in the study catchment was determined by the neural network weight analysis. This structure is able to keep stability of the rainfall-runoff simulation. Cross training and validation results of the BP model show that efficient coefficient (NSC) is over 0.9 during the training period, and NSC is over 0.88 during the validation period. Therefore, the neural network weight analysis can be used to determine the relationship between forecast object and its influencing factors. The artificial neural network model offers an efficient way for rainfall-runoff simulation in karst basin.