Genetic Programming and Its Application in Rainfall-Runoff Modeling

The role and importance of rainfall-runoff process in water resources studies has led this process to be considered by many researchers. Different methods such as artificial neural networks, fuzzy systems, neurofuzzy, wavelet analysis, genetic algorithm, genetic programming and stochastic differential equations have been developed for rainfall-runoff modeling. Furthermore, genetic programming which involves a mathematical model relating output and input variables, is able to select input variables that effectively contribute to the model. In this research, genetic programming (GP) was applied to modeling of daily basis rainfall-runoff process in Lighvan watershed with area of 76.19 km2. According to the ability of GP in selecting the best variables, the significant variables were selected after 10 times running of GP. Modeling process was carried out using selected variables as well as two sets of mathematical operators. Comparing the results obtained for both models indicated that correlation coefficients and mean square errors using training data set were equal for both of them i.e. 0.85 and 0.06, respectively. For the test data the coefficients became 0.93, 0.2 for set (1) and 0.97 and 0.08 for set (2), respectively. The model obtained from set (2) of the mathematical operators, was selected as the desirable one for the rainfall-runoff analysis in the watershed.