Development of Groundwater Level Model Using Artificial Intelligence Approach

Abstract

Water is a vital source supporting all forms of developments all over the world. In Gaza Strip (GS), Gaza coastal aquifer (GCA) is the most precious natural source where it is the only source of water for different uses. Groundwater crisis in Gaza includes two major folds: shortage of water supply and contamination. The extraction of groundwater currently exceeds the aquifer recharge rate. As a result, the Groundwater level (GWL) is falling continuously and by contamination of many pollutants mainly nitrate and salinity. Therefore, one important requirement for effective management of groundwater is forecasting the GWL fluctuations. The undertaken research is concerned with the development of GWL model using one of Artificial Intelligence techniques namely Artificial Neural Networks (ANNs).The applicability of ANNs models in simulating groundwater level was investigated in Khanyounis Governorate (KYG). ANNs are being used increasingly as alternative tools to physical based groundwater modeling approaches to predict and forecast water resources variables in complex groundwater systems with limited data as the case of GCA. In order to model GWL using ANNs it is necessary to gather data for training purposes. Physically, the GWL influenced by many variables such as: recharge from different sources, abstraction, precipitation, return flow. In this study, dependent variable used in the developed ANNs model were the initial (GWL), recharge from rainfall (R), recharge from return flow from water networks system (RRFW), recharge from return flow from wastewater system (RRFWW), recharge from return flow from irrigation water (RRFIW), abstraction from municipal wells (QM), abstraction from agricultural wells (QA). The aforementioned input variables were used to predict the final (GWL) at 17 monitoring wells which distributed over all study area. After a number of trials, the best neural network was determined to be Radial Basis Function networks with three layers: an input layer of 7 neurons, one hidden layer with 9 neurons and the output layer with 1 neuron. The developed model generated very good results which were clearly appeared through high correlation between the observed and predicted values of GWL. The correlation coefficient (r) between the predicted and the observed output values of the ANNs model was 0.993. The high value of r showed that the simulated GWL values using the ANNs model were in very good agreement with the observed GWL which means that ANNs model is a useful and applicable. The developed model was utilized as an analytical tool to study influence of the input variables on GWL. It was utilized as simulation and prediction tool of GWL in monitoring wells in KYG. Moreover, ANNs model was utilized as decision support tool by considering two future scenarios based on management of overall abstraction from the aquifer. Two selected management scenarios were tested; (1) work as usual (zero scenario), (2) reduction of overall abstraction by 50%. In the first scenario, the GWL in the cone of depression area will be expected to decrease from -15 m below MSL at year 2015 to -24 m below MSL at year 2025. In addition, it was noticed that seawater intrusion phenomena will increase and its impact will cover more than 60% of total area of KYG in year 2025. For the second scenario, the GWL in the cone of depression area will be expected to increase from -8 m below MSL at year 2015 to -3 m below MSL at year 2025. It was noticed that the effect of seawater intrusion will be reduced where the influenced area will be 25 % of total area of KYG in year 2025. Thus, this study has shown that ANNs are effective tools in forecasting GWL fluctuations in the GCA in spite of GWL model is very difficult since it is affected by many interconnected variables.