SEPARATION OF STREAM FLOW INTO COMPONENTS THROUGH THE USE OF A CO-ACTIVE NEURO FUZZY INFERENCE SYSTEM (CANFIS)

Abstract

In this study, the usability of a Co-Active Neuro-Fuzzy Inference System (CANFIS) as an alternative to the Digital Filtering (DFM) and United Kingdom Institute of Hydrology (UKIH) mathematical methods, which are frequently used for separating total stream flow into surface and base flow, was examined. Surface flow and base flow values determined from the daily average flow data of the Aksu Stream in the Melen Basin of Turkey's Northern Black Sea Region through the use of DFM (alpha = 0,830) and UKIH (N = 5) methods were used as the training and test data of CANFIS. The applications trained through DFM and UKIH were, respectively, titled as CANFIS(DFM) and CANFIS(UKIH). Performances of all of the methods used were compared by error analysis and the examination of base flow indexes (BFI). Obtained flow values and BFI results showed that the surface flow and base flow estimations of all methods are significantly similar, and that the base flow values provided by the UKIH and CANFIS(UKIH) methods were bigger than those obtained from the DFM and CANFIS(DFM) methods as reported in the studies included in the literature. In addition, it was understood that in both CANFIS(DFM) and CANFIS(UKIH) methods, the effects of the methods used for training were fairly limited on the surface flow (R-2=0,9709) and base flow (R-2=0,9765) test values. In conclusion, the study demonstrated that CANFIS can be used in the determination of surface flow and base flow without needing parameters required by the DFM and UKIH methods, namely, recession coefficient and number of members in minimum groups.