evaluation of artificial intelligent and empirical models in estimation of annual runoff

abstract from longley, the various equations for determining the runoff to water management are presented by the researchers that are widely used in hydrologic sciences. in this study by using observational data, was evaluated empirical, artificial neural network (ann) and ca-active neuro-fuzzy inference system (canfis) models in estimation of runoff. for this purpose, by using climatic and physiographic information in three stations of pole zamankhan, ghale shahrokh and sade zayandeh rood, runoff values were estimated from empirical models and intelligent models were compared to annual runoff values. input parameters include rain, mean temperature, mininmum temperature and maximum temperature. the results showed that the artificial intelligent models had good accuracy in estimating runoff. among the empirical methods, method of di souza was appropriate. comparison statistical parameters between methods was showed that mean percent error (mpe) in ann, canfis and empirical method was 7, 12 and 43 percent respectively that confirmed differences of between the methods is significant. also, canfis model did not artificial improve ann results. the results showed, with reduction of input variables from 4 parameters to one parameter of precipitation, modeling error reaches its maximum value (from mpe=7% to mpe=16%). versus, the optimal structure of ann had less sensitivity to remove the mean air temperature parameter (from mpe=7% to mpe=10%(. therefore, according to empirical models required information limitations and high accuracy of artificial intelligent models, intelligent models application is recommended.