SCIENTIA GEOGRAPHICA SINICA ›› 2012, Vol. 32 ›› Issue (1): 74-80.doi: 10.13249/j.cnki.sgs.2012.01.74

• Orginal Article • Previous Articles     Next Articles

Rainfall-runoff Simulation Based on Runoff Classification Using Dynamic Artificial Neural Networks

Yue-hong SHAO1(), Bing-zhang LIN1, Yong-he LIU2   

  1. 1. Applied Hydrometeorological Research Institute, Nanjing University of Information Science and Technology, Nanjing, Jiangsu 210044, China
    2. Institute of Resources and Environment, Henan Porytechnic university, Jiaozuo, Henan 454000, China
  • Received:2011-01-17 Revised:2011-06-02 Online:2012-01-20 Published:2012-01-20

Abstract:

A runoff sequence is classified into several sub-runoff sequences with cluster analysis and practical application, and local dynamic neural network for each sub-runoff sequence is performed separately. An Elman recurrent neural network model (ENN) is constructed and applied for the daily runoff forecast in the Linyi sub-catchment of the upper Yishu river basin in this paper. In order to further evaluate the performance of local Elman neural network (LENN), global Elman neural network (GENN) is applied as a comparison at the same time in the study region. Based on analysis indexes such as Nash-Sutcliffe coefficient, correlation coefficient, mean relative error and root mean relative square error, the results of daily runoff and flooding processes are attained. It suggested that Elman recurrent neural network model has a high accuracy of forecast on the rainfall-runoff dynamic process and the maximum peak flow and peak occurrence time, whether it is configuration samples or evaluation samples. However, local Elman neural network based on runoff classification is more suitable and efficient to daily runoff forecasting than global Elman neural network, especially during the period of arid season and semi-humid season. The simulation precision of different periods decreases in the order of arid season, semi-humid season and humid season. Elman recurrent neural network model is feasible to effectively simulate the hydrological dynamic characteristic of the daily runoff and to reflect the complex nonlinear runoff regular pattern in the basin as a promising and efficient method. In order to further evaluate superiority of this method, longer series of data, more regions and hydrology model are need to study and analysis because of the highly nonlinear, spatial and temporal heterogeneity and dynamic uncertainty of rainfall-runoff process.

Key words: runoff classification, cluster analysis, Elman neural network, rainfall-runoff, local neural network, global neural network

CLC Number: 

  • S273