فهرست مطالب مرضیه داداش بابا

Increasing agricultural, industrial, and domestic demands as well as over abstraction of groundwater especially in semiarid areas like Iran is very important. Groundwater modeling in aquifers like Tabriz aquifer is vital for achieving the correct understanding of groundwater resources and also managing demands (agriculture, domestic and industrial) and engineering projects, as well as avoiding of groundwater declination. This research presents a hybrid model compose of self-organizing map (SOM) and artificial neural network (ANN) (SOM-ANN), for predicting groundwater level in the 10 observation wells which are located in unconfined Tabriz aquifer. The prevailing climate over Tabriz plain is semi-arid and cold, according to Emberger classification (Emberger, 1930). The average annual temperature is 13 °C (based on data from meteorological stations in Tabriz, 1986-2015) and the annual average rainfall is 251 mm as per data for the years of 1986-2015, recorded by the East Azerbaijan Regional Water Authority. Tabriz plain aquifer is composed of unconfined and confined aquifers with high heterogeneity and complexity. The unconfined aquifers overlay the confined aquifer and their composition comprises silt, clay, sand and gravel, which is repeated frequently, with varying thicknesses and grain size in different places. The bedrock of the confined aquifers is marl composed of clay and salt form Pliocene and Miocene. The unconfined aquifer in south of Tabriz is flanked to the hillside of Sahand mountains and the groundwater within this aquifer is freshwater; but the unconfined aquifer at the northern and western part of the plain is shallow (5 up to 58 m) and suffers from varying degree of salinity due to sedimentation originating from the Miocene Formation. The plain is drained by the Aji Chay River (the Aji River). The confined aquifer (58-120 m) has better quality (Nadiri, 2008) and is confined by marl. This study has focused on the unconfined aquifer component, which serves as the supply source of a significant proportion of drinking water in Tabriz. Based on data of 22 logs of drilled piezometers in the unconfined aquifer (prepared by East Azerbaijan Regional Water Authority, 2016), the saturation thickness in the central part of the plain is thicker than its margins. Unconfined aquifer of the plain has 42 piezometers and the general direction of flow is from the Northeast to the Southwest. In order to examine the effectiveness of the models in predicting groundwater levels, the performance measures were quantified for all models using Root Mean Squared Error (RMSE), Pearson correlation coefficient (R), Coefficient of determination (R^2). RMSE values range from 0 (the ‘perfect model’) to higher real numbers and measure global goodness-of-fit between the observed values and modelled value. The closer the value of RMSE to 0, the more accurate the prediction is. Pearson correlation coefficient and Coefficient of determination (0≤r^2≤1) describes the proportion of the total variance in the observed data that can be explained by the model. The amount of r^2 and r, with higher values, indicates better agreement between observed and predicted values. The SOM method was adopted to classify aquifer to homogeneous area, the ANN model utilized to predict groundwater level in each cluster. The 10 observation wells with monthly groundwater level data during 14 years were selected for this research. The observation wells were classified by SOM method into 4 classes. Effective parameters on groundwater variation were select as input data sets. The normalized five input parameters with one-month lag (t0-1) including precipitation (P), temperature (T), evaporation (E), withdrawal (Q) and the groundwater level (GWL) were applied to run the hybrid model. The SOM-ANN model is implemented for each class. According to the results of the models for Classes 1, 2 and 4, the average RMSE of the training and testing stages were obtained 0.10, 0.20, and 0.22 meters respectively, and for observation wells in Class 3 that are located in the center of the plain the average error was equal to 0.23 meter. The Higher error in model 3 or class 3 could be originated from increasing saturated thickness and complexity of hydrogeological in this area. The hybrid model error was acceptable to prediction of groundwater level variation. The aquifers are often complex like Tabriz aquifer, so, the SOM-ANN model is suggested as an applicable method to simplify other aquifers to decrease the errors of model.

Increasing development of engineering projects construction such as city subway needs appropriate investigation, management and control of groundwater. Therefore, precise estimation of hydrogeological parameters such as hydraulic conductivity is the most important factor in studies and modeling of groundwater and geotechnical issues. In recent decades, various laboratory and field methods exist for estimating this parameter, but estimation of hydraulic conductivity using these methods due to the heterogeneity and anisotropy hydrogeological environments is costly, time-consuming and inherent uncertain. In this study, three fuzzy inference methods, Sugeno (SFIS), Mamdani fuzzy inference model (MFIS) and Larsen Fuzzy Inference System (LFIS) that is suitable for handling the uncertain data, was adopted to estimate the hydraulic conductivity in Tabriz city subway area. After, the hybrid SOM-FL model was presented to improve efficiency of individual model and solve heterogeneity problem of the Tabriz city aquifer. Based the evaluation criterions R2 and RMSE the results of individual models are acceptable but the proposed hybrid model improved, R2 in train and test stage, 18% and 15.4 percent, respectively.