LAUSR

Selection of optimal model inputs is a challenge for non-linear dynamic models. The questions as to which inputs should be used for model development have been a challenge in practice. Despite its importance, the literature on comparison of different methods for choosing inputs for estimating evaporation from saline water is limited. In this study, used three methods namely the Gamma test (GT), entropy theory (EnT), and procrustes analysis (PA) for determining suitable variables for estimating saline water evaporation using non-linear models of artificial neural network (ANN). The weather station near Lake Urmia was used for this experiment. At this station, pans of different concentrations (500 g/L, 300 g/L, 100 g/L, 50 g/L, 20 g/L, 10 g/L, 5 g/L, and drinking water) were prepared. In addition to evaporation data, surface water temperature (measured for each pan separately), air temperature, mean cloudiness, sunshine hours, mean relative humidity, mean wind speed, solar radiation, maximum wind speed, station pressure, mean station vapor pressure, maximum and minimum temperatures, and precipitation were also used. Model results were compared with field measurements and model performance was evaluated by the coefficient of correlation, root mean square error, and Nash–Sutcliffe efficiency coefficient. The most important variables identified by GT were surface water temperature, air temperature, mean relative humidity, mean wind speed, mean station pressure, minimum temperature, precipitation, mean station vapor pressure, and solar radiation. Also, as can be seen the most important variables for evaporation from saline water using the EnT method were water surface temperature, wind speed, and precipitation. The three important variables in the estimation of saline water, evaporation selected by the PA method, were air temperature, sunshine hours, and mean wind speed. According to results, as the concentration increased, the mean station vapor pressure and temperature variables had the most influence on saline water evaporation. The uncertainty of model output was determined using the 95 percent prediction uncertainty (95PPU or P-factor) and d-factor. Although ANN-GT and ANN-EnT showed better goodness-of-fit metrics, ANN-PA had the lowest uncertainty among the three models in estimating evaporation from saline water. Generally, the PA method was able to demonstrate acceptable performance over the other two methods, with the least number of input variables.