چكيده به لاتين
The Arvand River is a tidal river located on the border between Iran and Iraq, discharging into the Persian Gulf downstream. Its main water sources are the Tigris, Euphrates, Karun, and Karkheh rivers. In recent years, the salinity of the river has increased due to the reduction in upstream inflow, leading to a higher intrusion of salty seawater from the Persian Gulf. The decrease in the quantity and quality of water poses a threat to the region's water resources and challenges economic, commercial, and other activities in the area.
The aim of this thesis is to examine the effects of various factors on salinity levels and to predict water salinity. For this purpose, data from four stations—Karun, North Arvand, Intersection of the rivers, and South Arvand—were modeled using the MLP neural network, neuro-fuzzy system (ANFIS), and M5P and M5Rules tree models under five different scenarios. The results obtained from each scenario were evaluated separately by station and model.
In the first scenario, which serves as the baseline for the research, the input parameters for the model are flow velocity, water depth, and temperature, while the output parameter is salinity. In the second, third, and fourth scenarios, the standard deviation of temperature, salinity from the previous step, and both of these parameters were added as input parameters to the model, respectively. In the fifth scenario, by merging data from all four stations into a single data file, the primary scenarios one to four were implemented as secondary scenarios one to four.
In a summary of the results, the average RMSE values up to two decimal places for the MLP model in scenarios one to five were 0.07, 0.06, 0.04, 0.03, and 0.09, respectively. The average RMSE values for the M5P model were 0.05, 0.06, 0.04, 0.03, and 0.08, while for the M5Rules model, they were 0.05, 0.05, 0.04, 0.03, and 0.07. Overall, the results indicate that the models responded better in the fourth to first primary and secondary scenarios, respectively. Additionally, the M5Rules tree model performed the best, followed by the M5P and MLP models. The ANFIS model, in contrast to the other models, had the poorest performance and was deemed unsuitable for this study. The results indicated that at upstream, velocity has the greatest impact on salinity levels, while at downstream, the effects of water depth on salinity become predominant.
