monthly river flow prediction using adaptive neuro-fuzzy inference system (a case study: gharasu watershed, ardabil province-iran)

there is different methods for simulating river flow. some of thesemethods such as the process based hydrological models need multiple input data and high expertise about the hydrologic process. but some of the methods such as the regression based and artificial inteligens modelsare applicable even in data scarce conditions. this capability can improve efficiency of the hydrologic modeling in ungauged watersheds in developing countries. this study attempted to investigate the capability of the adaptive neuro-fuzzy inference system (anfis) for simulating the monthly river flow in three hydrometric stations of pole-almas, nir, and lai; which have different rate of river flow. the simulations are conducted using three input data including the precipitation, temperature, and the average monthly hydrograph (amh). the study area islocated in the gharasu watershed, ardabil province, iran. for this aim, six groupsof input data (m1, m2, … m6) were defined based on different combinations of the above-mentioned input data. theconducted simulations in pole-almas and nir stations have presented an acceptable results; but in lai station it was very poor. this different behavoirs was referred to the lower volume of flow and consequently irregularity and variability of flow in lai station, which cause the decrease of accuracy in the simulation. the amh parameter had an important role in increasing the accuracy of the simulations in pole-almas and nir stations. the findings of this study showed that anfis is an efficient tool for river flow simulation; but in application of anfis, the selection and utilization of relevant and efficient input data will have a determinativerole in achieving to a successful modeling.