evaporation predict using artificial neural network and climat signals in dez basin

introduction evaporation is a phenomenon of hidrological cycle that is particularly important in studies of water. much of the annual rainfall in arid and semi-arid weather characteristics that iran immediately returns to the atmosphere. so it will be important to estimate and predict the amount of evaporated to estimate of water. dez basin in south-west iran, where the country's main centers of permanent ponds and water supply for the probe is located at the river country. dez basin in south-west iran, where most centers are located in its permanent pond.thus, based on the predicted evaporation as one of the most important hydraulic phenomena essential role of water-related programs in the basin. evaporation is a function of several factors due to different climatic variables and the interaction of these variables on each other is a complex nonlinear phenomena and should be carefully studied and the methods used for simulations. investigated the relationship between climatic signals and the amount of evaporation is something that can make it possible to predict. climatic signals are signs that climate variables in different areas are affected these signals can be noted from nino, nao, sst and enso.there are several models to predict the evaporation rate based on these signals. researchers have used various methods to achieve satisfactory results for different parts of the world. among these methods, artificial neural network because it has a similar behavior is a biological neuron system has many fans.this network has the power to govern the replication of data mining and forecasting climatic parameters make possible. in this study, evaporation changes in dez basin associated with large-scale climate signals are studied and while determining the performance of artificial neural networks and climate signals to evaporation predict, the most important climate signals associated with evaporation in the basin is characterized.â methodology the study area includes part of the catchment basin of dez that is part of the persian gulf catchment.the basin, parts of the provinces of lorestan, chahar mahal va bakhtiari, khuzestan and markaz.to conduct research related statistics within and around the basin evaporation in four synoptic stations including khorramabad, arak, dezful and koohrang is applied. during the period of 29 years, from 1983 to 2011 for stations in khorram abad and arak, and 19 years from 1992 to 2011 for dezful and is koohrang. facts about climate signals obtained from the noaa site is evaporation during the same time period. in this study, to simulate evaporation in the basin made to fit the artificial networks whichconsists of the determination of the number of neurons in each layer of the network, network training and testing the network. after training the neural network to determine the proportion of data values between zero and one attempt and the number of neurons in the hidden layer through trial and error will be determined. each entry is multiplied by the corresponding weight and then come together to say that the summation function.the resulting number is the sum of the excitation function is sent to the output formats the network.â  discussion using the correlation matrix of the 24 climate signal, the signal for each station 4 have been identified as most relevant to evaporation.to determine the optimal structure of the network, the number of neurons in the middle layer of 5 to 15 than learning ratio from 0.1 to 0.3 were changed where the lowest root mean square error and coefficient of determination showed the highest value was determined as the optimal structure that optimal structure for the station khorramabad 7 neurons in the middle layer and learning ratio 0.3, dezful 10 neurons in the middle layer and learning ratio 0.2, koohrang 9 neurons in the middle layer and t learning ratio 0.2, arak 8 neurons in the middle layer and learning ratio is 0.3.to determine the optimal neural network, the output network evaporate observational data are compared with the results of the comparison indicate a high correlation between stations in the basin. amount of correlation in khorramabad station is79%, dezful 96%, koohrang 72% and arak72%. in order to predict the evaporation, the output data networks as dependent variables and data related to climate signals as independent variables, the correlation is derived that in all four stations are highly correlated, so the station khorramabad 99.5% , dezful 98.3%, koohrang 99.2% and arak is 99%.â â â â â  â conclusionâ â â the best neural network structure which has a higher coefficient and root mean square is less the network of stations khorramabad has7 neurons in the middle layer with learning ratio 0.3, dezful 10 neurons in the middle layer with learning ratio 0.2, kuhrang 9 neurons in the middle layer with learning ratio 0.2, and arak 8 neurons in the middle layer with learning ratio 0.3. comparison to evaporation observational data and output of artificial neural network shows the high correlation between these data so that the extent of this correlation on the khorramabad station is 79%, dezful 94%, kuhrang 80% and arak 72%. in order to predict the evaporation, the output data networks and data related to climate signals, the correlation is derived so the station khorramabad is 99.5%, dezful 98.3%, koohrang 99.2% and arak is 99%. therefore, given the high correlation coefficient, using the correlation equation can be predicted with high accuracy of 98% compared to evaporate for months without data should be taken.