Testing the structure of a hydrological model using Genetic Programming

JOURNAL OF HYDROLOGY

卷:397

Article

Selle, Benny¹  Muttil, Nitin^2,3 

[1] Dept Primary Ind, Tatura, Vic 3616, Australia

[2] Victoria Univ, Sch Sci & Engn, Melbourne, Vic 8001, Australia

[3] Victoria Univ, Inst Sustainabil & Innovat, Melbourne, Vic 8001, Australia

关键词: Data mining;    Machine learning;    Diagnostic model evaluation;    Model structure uncertainty;    Parsimonious inductive model;    Data-based modelling;    Dominant process concept;   

DOI  :  10.1016/j.jhydrol.2010.11.009

来源: Elsevier

【 摘 要 】

Genetic Programming is able to systematically explore many alternative model structures of different complexity from available input and response data. We hypothesised that Genetic Programming can be used to test the structure of hydrological models and to identify dominant processes in hydrological systems. To test this, Genetic Programming was used to analyse a data set from a lysimeter experiment in southeastern Australia. The lysimeter experiment was conducted to quantify the deep percolation response under surface irrigated pasture to different soil types, watertable depths and water ponding times during surface irrigation. Using Genetic Programming, a simple model of deep percolation was recurrently evolved in multiple Genetic Programming runs. This simple and interpretable model supported the dominant process contributing to deep percolation represented in a conceptual model that was published earlier. Thus, this study shows that Genetic Programming can be used to evaluate the structure of hydrological models and to gain insight about the dominant processes in hydrological systems. (C) 2010 Elsevier B.V. All rights reserved.

【 授权许可】

Free