Major Iraqi water resources are Tigris and Euphrates rivers. Adhaim River is one of Tigris tributaries. Artificial Neural Networks (ANNs) was selected to simulate flow of Adhaim River. GIS-technique was used to delineate area of catchment. The six built models based on time steps (1, 5, 10, 15, 30, and 365 days). Statistical analysis applied to estimate main statistical parameters. Five methods used to estimate the average rainfall. Five scaling factor formula are used for data scaling. ANNs technique is formulated to simulate flows at Naros flow station. based on average rainfall readings at the upstream. Different procedures were applied to predict stream flow. Variable number of ANNs technique parameters was tested such as neurons, layers, ten types of transfer functions, different learning rate and epoch values are tested. The most appropriate simulation is selected based on two approaches (graphical and statistical). Six models are built based on time step, which can be used by Adhaim’s engineers according to the purpose.
Finally the best building model for Adhaim river defined as, (ANNs), trained with Levenberg-Marquradt back propagation algorithm, with two layers, three transfer functions mainly of ‘tansig’,’logsig’,’trainlm’ was able to provide a best generalization of the complex, non-linear for Adhaim rainfall-runoff process
KEYWORDS: ANNs, Levenberg-Marquradt, Scaling Factor, performance criteria, Adhaim, Iraq.
Iraq faces serious water shortages. In order to mitigate the impact of this problem, it is necessary to find an effective way to manage these water resources using different strategies and techniques. One of these techniques, Artificial Neural Networks (ANNs), was applied on Adhaim River, one of the Tigris tributaries, for simulation hydrological rainfall-runoff relationship.
Several studies were carried out on the Adhaim River Basin, but most of these studies were either geological or morphological or tourism purposes. 1,2, a few of which were for hydrological purposes, and the following is a summary of the most important hydrological research.
Daily Surface inFiltration Baseflow (SFB) conceptual rainfall-runoff model was applied to simulate stream flow for Adhaim river basin. Three versions of the model were tested: the original Australian three-parameters (SFB), the modified five parameters version (SFB-5) and modified six parameters (SFB-6) model, which developed by the researcher for the original Australian model. The data enter was precipitation, evapotranspiration and observed runoff while the output was simulated runoff. The five parameters version (SFB-5) provided better performance runoff simulation than others models 3.
Dynamic Regression model (DR) was used for forecasting the discharge of Adhaim river. The Auto Correlation Function (ACF) was used to determine the stationary level of the time series., also the Partial Auto Correlation Function (PACF) was used to identify a suitable Auto Regression Integrated Moving Average (ARIMA) model for time series of rainfall and discharges for river and the factors of Transfer Function models (TF) were determined. The model passed the tests and can forecasting the discharges successfully. 4.
A monthly modified Stanford rainfall-runoff model was used to develop Adhaim river runoff curves through simulating runoff processes. The runoff curves are developed by inserting various equations related to runoff calculations and coefficients, runoff curves accordingly provide a better and more accurate estimate for runoff coefficients 6.
Database was built for morphometric properties of Adhaim basin, using geographical information system, which was conducted to understand its hydrologic characteristic’s and consequently understanding the flow regime for all Adhaim tributaries’ 11, other researchers studied, hydro-geographical analyses, geology, topography, climate, morphology, soil, land use, natural vegetation, sedimentation and soil erosion for Adhaim river basin to estimated how these factors influence on spatial and temporal distributions of Adhaim river flow 7.
The Soil and Water Assessment Tool (SWAT) model, was used to evaluate the impacts of climate change on water resources in Adhaim Basin.in have been manifesting increasing variability contributing to more severe floods and droughts due to climate change. The results showed worsening water resources regime into the future 8.
In view of research’s that carried out on Adhaim river basin, it was found that most of the models used require a lot of data, the access to the required data is often costly, difficult, requires long time for collection, In addition to the difficulty of printing hardcopy data without errors, so we need a model that could simulate runoff with minimum requirement of data. The most appropriate model which is recommended in most recent research is, Artificial Neural Networks (ANNs) which is used in this study. ANNs have been applied for rainfall-runoff modeling since 1980, 20.
Most recent researches recommends to use artificial neural networks because of their high efficiency to simulate the flow of rivers, plus they can be used when we have only one input such as rainfall and one output as observed runoff for training stage, but unfortunately there is no specific rule for determining its parameters and variables, so the idea beyond this paper is to suggest a recommended way to find the best solution way for building model for Adhaim river basin.