A COMPARATIVE EVALUATION OF ARTIFICIAL NEURAL NETWORK MODELS AND EMPIRICAL METHODS IN ACTUAL EVAPOTRANSPIRATION ESTIMATION (CASE STUDY: AMMAMEH REPRESENTATIVE CATCHMENT)

摘要

Actual evapotranspiration (ET) is a basic term in the hydrology cycle and water balance ofrncatchment. In this research an artificial neural network, with multilayer perception structure, andrnempirical equations Advection–Aridity of Granger and Gray and Combination Equations wasrnadapted and evaluated to model ET by means of conventional climatic data. The estimates ofrnthe models are also compared with those of the monthly actual ET of Ammameh catchmentrnobtained through water balance equation. Combination Equation with R2 of 0.84 and RMSErnof 0.46 performed better than the other empirical models.rnThe best ANN model between 13 different combination and empirical model is ANN5 withrnminimum, maximum temperature and pan evaporation in the model input and with R~2 of 0.88rnand RMSE of 0.32 mm per day.rnThe ANN1 with Minimum and maximum temperature as model inputs, with R~2 of 0.83 wasrna poor performer, but the R~2 differed by only 1% when compared to the performance of the.