Temporal and Spatial Modeling of Groundwater Level in Bushehr Plain using Artificial Intelligence and Geostatistics

One of the basic measures to reach the optimal management of water resources is modeling and predicting the level of stagnation of wells. Controlling the level of stability using observation wells is considered as the main source of information to investigate hydrological stresses. By using the daily and monthly data of the wells, it is possible to check the fluctuations of the water level, and these checks are necessary for understanding the behavior of underground water resources in the long term and making any kind of management decisions. Considering the importance of predicting the groundwater level, it is important to find an appropriate method in this regard. In recent years, the use of intelligent systems for predicting the level of underground water is rapidly increasing, which is due to the ease of use and the high accuracy of these models in approximating nonlinear and complex mathematical equations. The aim of this research is to predict the level of groundwater stagnation in Bushehr plain with the help of artificial neural network models, support vector machine model and decision tree.

In the first half of the year, all the support vector machines and the decision tree were made from the hot water of Dashtestan. Also, statistical method was used to study the spatial changes of groundwater. The data of month of 50 wells in the plain aquifer was a ten-year period from 2009-2018. To evaluate the results of the mentioned models were used of RMSE, MBE and R2.

The results showed that all three methods had high accuracy in simulating water level. The artificial neural network method with a higher R2 (0.993), the lowest squared mean error (0.29) and the lowest absolute mean error (0.024) was selected as the superior method for predicting the water table. Kriging zoning showed that the groundwater level in most parts of the plain during the study period has decreased. The maximum drop is equal to 10 meters.

The results of the modeling methods and the results of the model evaluation statistics show the good performance of the models in estimating the groundwater stagnation level, but according to the comparison of the explanation coefficient of the models in this research, it was shown that the neural network method Artificial has a higher accuracy in the distribution of available data and the calculated values were less different than the observed values of the artificial neural network model compared to the other two models. In total, the results of the research showed that the artificial neural network method has a higher accuracy in estimating the level of underground water.