Title: Performance of Geno-Fuzzy Model on rainfall-runoff predictions in claypan watersheds Authors
|Senaviratne, G.M.M.M. -|
|Udawatta, Ranjith -|
|Anderson, Stephen -|
|Thompson, Allen -|
Submitted to: Journal of Hydrology
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: June 16, 2014
Publication Date: September 19, 2014
Citation: Senaviratne, G.A., Udawatta, R.P., Anderson, S.H., Baffaut, C., Thompson, A. 2014. Performance of Geno-Fuzzy Model on rainfall-runoff predictions in claypan watersheds. Journal of Hydrology. 517:1008-1018. DOI: 10.1016/j.jhydrol.2014.06.023. Interpretive Summary: Fuzzy logic is an artificial intelligence technique used for modeling non-linear relationships such as that between rainfall and runoff. Fuzzy membership functions describe the membership of input and output variables to fuzzy sets, e.g., low rainfall. Fuzzy rules define the relationships between these sets. A genetic algorithm is an optimizing method inspired by biological evolution. The objective of this study was to develop rules and membership functions for runoff prediction based on rainfall for three small row crop watersheds prior and after establishment of upland grass and grass+tree buffers. Measured rainfall and runoff data were used to develop rules and optimize membership functions corresponding to conditions before and after buffer establishment. Additional data were used to validate the models. Results showed that the models performed well in estimating runoff based only on rainfall. Results from these fuzzy inference systems were also close to the outputs of a physically based rainfall-runoff simulation model for the same watersheds. Scientists in research institutions and agencies can use this technology when rainfall and runoff data are available but little is known about the physical attributes of a watershed.
Technical Abstract: Despite increased interest in watershed scale model simulations, literature lacks application of long-term data in fuzzy logic simulations and comparing outputs with physically based models such as APEX (Agricultural Policy Environmental eXtender). The objective of this study was to develop a fuzzy inference system (FIS) with genetic algorithm (GA) optimization for membership functions (MFs), for rainfall-runoff prediction for three adjacent row crop watersheds prior to and after establishment of upland contour grass and agroforestry (tree + grass) buffers. One watershed remained as the control. A Mamdani type FIS with 5 MFs (with and without GA optimization) and 5 fuzzy rules was created using the Fuzzy Logic Toolbox of MATLAB 7.10.0. Measured rainfall-runoff data from 1993 to 1997 for the pre-buffer period of the agroforestry watershed were used for GA optimization of MFs. The GA optimized FIS simulated event-based runoff with r2 and NSC (Nash-Sutcliffe Coefficient) values of 0.84 and 0.80, respectively. The FIS was validated with r2 and NSC values of 0.69 and 0.68, respectively, using control watershed data from 1998 to 2008. The MFs were modified and optimized with GA to suit post-buffer agroforestry and contour grass watersheds. The FISs were calibrated (1998-2001) and validated (2002-2008) with r2 and NSC values all between 0.7 and 0.8. Genetic algorithm optimization of MFs improved all r2 and NSC values. The model performance coefficients for FIS and APEX were similar and thus FIS offers an alternate modeling tool for runoff estimation in the absence of detailed watershed data.