Investigation of the Inﬂuence of Excess Pumping on Groundwater Salinity in the Gaza Coastal Aquifer (Palestine) Using Three Predicted Future Scenarios

: The Gaza coastal aquifer (GCA) is the only source of water for about two million citizens living in Gaza Strip, Palestine. The groundwater quality in GCA has deteriorated rapidly due to many factors. The most crucial factor is the excess pumping due to the high population density. The objective of this article was to evaluate the inﬂuence of excess pumping on GCA’s salinity using 10-year predicted future scenarios based on artiﬁcial neural networks (ANNs). The ANN-based model was generated to predict the GCA’s salinity for three future scenarios that were designed based on di ﬀ erent pumping rates. The results showed that when the pumping rate remains at the present conditions, salinity will increase rapidly in most GCA areas, and the availability of fresh water will decrease in disquieting rates by 2030. Only about 8% of the overall GCA’s area is expected to stay within 500 mg / L of the chloride concentration. Results also indicate that salinity would be improved slightly if the pumping rate is kept at 50% of the current pumping rates while the improvement rate is much faster if the pumping is stopped completely, which is an unfeasible scenario. The results are considered as an urgent call for developing an integrated water management strategy aiming at improving GCA quality by providing other drinking water resources to secure the increasing water demand.


Introduction
The main source of drinking water in the Gaza Strip is a shallow aquifer, which is part of the coastal aquifer. The quality of the groundwater has extremely deteriorated in terms of salinity to determine the annual average system draft that would divert all usable storage volumes of water from the system's reservoir(s). For each specific alternative, configuration, or scenario of a safe yield simulation model, simulated usable storage volumes of water normally vary over the period of the record and the restrictive condition that limits the safe yield estimate is the diversion of all usable storage volumes of water as described in the paragraph above during any one day of the period of the record. The drought during which this restrictive condition occurs is the most severe drought of the record [19,20].
The main source of the elevated concentration levels of chloride concentration Cl − in GCA is the overpumping. During the last decades, GCA has been extensively overexploited due to the increasing population growth, high rates of urbanization, and increased agricultural activity. Intensive exploitation of groundwater in the Gaza Strip in the past years has disturbed the natural equilibrium between fresh and saline water, and has resulted in increasing salinity. This has induced seawater to flow towards the major pumping centers in urban areas leading, to continuous increasing of the levels of Cl − in GCA, particularly in areas close to the coastline [7,21]. It has been found that seawater intrusion extends about 2 km inland in many areas of Gaza Strip [7]. Lateral flow from the eastern Eocene aquifer is another source of CL. Such an aquifer is characterized by high Cl − levels, and its flow regime is normally to the west towards GCA. Consequently, water flux coming from the Eocene aquifer worsens ground water quality by increasing Cl − levels, especially in the eastern parts of Gaza Strip [22,23]. Saline lenses, which are found in many GCA locations particularly in the southern parts, are salty water lenses existing at deeper layers. Hence, if the wells are near these lenses and the screens are deep enough, these lenses participate in feeding these wells and the abstracted water from these wells will have elevated Cl − concentrations [22][23][24]. Many other Cl − sources, such as the recharge of irrigation water and percolation of landfill leachate, contribute to the deterioration of the ground water quality [22,23].
The objective of this study was to evaluate the response of GCA to the excess pumping from groundwater and analyze the future deterioration of groundwater salinity using three future scenarios to predict groundwater salinity in 2030 by developing and generating an empirical model developed using ANNs. Understanding the relationship between groundwater pumping and the deterioration of salinity can contribute to integrated water resources management in the Gaza area.

Geology and Hydrogeology of the GCA
The study area of this research was the Gaza Strip, Palestine. It is a narrow strip of land on the Mediterranean coast. It is part of Palestine with an area of about 365 km 2 and it is around 40 km long [25]. Figure 1 shows the regional and location map of the Gaza Strip. The Gaza Strip is an area where the exploitation level of recourses exceeds the carrying capacity of the environment. This is especially true for the water and land resources, which are under high pressure and subject to overexploitation, pollution, and degradation. The salinity of the groundwater is a main difficult issue in the Gaza Strip. GCA is part of the coastal aquifer that extends from GS in the south to Carmel Mountains in the north along the Mediterranean coast line (about 120 km), as shown in Figure 1, which illustrates the layout of GCA and the adjacent aquifers [25]. The width of GCA varies from10 km in the north to about 20 km in the south [22,26]. The GCA thickness varies from about 120 m in the west (at the shoreline) to a few meters in the east [27]. Meanwhile, the depth of the water level of GCA ranges from about 60 m below the ground surface in the east to a few meters near the coastline in the west [25].  Within Gaza Strip, the coastal aquifer overlies the saqiye group, which is considered to be impermeable. The top of the saqiye group therefore marks the base of the coastal aquifer. Approximately 10 to 15 km inland from the coast, the saqiye group pinches out, and the coastal aquifer rests directly on Eocene chalk and clastic sediments of Neogene age. The thickness of the entire coastal aquifer is on average about 120 m. At the eastern Gaza border, the saturated thickness is about 60 m in the north, and only 5-10 m in the south near Khan Younis and Rafah. The sand dunes are the top part of Pleistocene marine sand and sandstone, which is interbedded with clayey layers Within Gaza Strip, the coastal aquifer overlies the saqiye group, which is considered to be impermeable. The top of the saqiye group therefore marks the base of the coastal aquifer. Approximately 10 to 15 km inland from the coast, the saqiye group pinches out, and the coastal aquifer rests directly on Eocene chalk and clastic sediments of Neogene age. The thickness of the entire coastal aquifer is on average about 120 m. At the eastern Gaza border, the saturated thickness is about 60 m in the north, and only 5-10 m in the south near Khan Younis and Rafah. The sand dunes are the top part of Pleistocene marine sand and sandstone, which is interbedded with clayey layers making three or four sub-aquifers as shown in the cross-section of the coastal aquifer in the middle area of GCA in Figure 1. The infiltration rate differs from one type of soil to another. However, due to increasing urban development, the natural soil has been disturbed and covered by impermeable layers, such as paved roads, or occupied by buildings. This, of course, reduced drastically the amount of infiltrated rainfall that replenishes the groundwater. The decrease in infiltrated rainwater appears as surface run-off, or is lost by either evaporation or diverted to the sea. The infiltrated water in Gaza Strip goes through the soil at a rate of one to two meters per day in the areas where fine sand is found, and this rate increases in the coarser formation, e.g., kurkar. However, the percolation rate decreases if it encounters a clayey layer in the subsurface. Water goes horizontally above the non-permeable layer until it encounters a disconnection in this layer and travels vertically downward to the groundwater reservoir [28].
The type of soil (sandy, silty, and clayey) can control the rate of infiltration. For example, a sandy surface soil normally has a higher infiltration rate than a clayey surface soil. The soil texture is important to identify the vulnerability of the artificial recharge basin to surface sealing, where a thin lamina of fine particles covering the surface of the spread basin will decrease the infiltration, with much more clogging [29,30]. The land use and soil type of Gaza Strip are shown in Figure 4. The texture of six different soils in Gaza Strip are shown in Table 1. The model boundary has been defined based on the political boundary of Gaza Strip based on 1948 United nation border lines. The portions of the aquifer system outside of the study area are mainly an open area with some agriculture activities that do not affect the GCA aquifer and scenarios [31].

Groundwater Salinity in Gaza Strip
GCA is extremely exposed to contamination and overpumping. The drinking water has started to become more saline and the chloride concentrations in many areas of the Gaza Strip exceed 2000 mg/L or more. Additionally, most of the water supply wells do not meet the terms of the World Health Organization (WHO) standards of drinking water quality [32][33][34]. GCA is considered as a characteristic case of a highly contaminated aquifer due to hydrogeological stresses in addition to insufficient water resources management [23]. Recent studies reported that no ground water in Gaza Strip meets all WHO drinking water standards; additionally, more than 90% of the ground water in Gaza Strip area is not suitable for drinking due to the elevated concentrations of many chemical parameters, particularly NO 3− and Cl − , in addition to microbiological contamination, which exists in many locations within GS [35][36][37][38].
It is clearly noticed that the chloride concentration increases significantly all over the Gaza Strip, especially in the southeastern and the middle areas. The best water quality is found in the sand dune areas in the north, mainly in the range of 50-250 mg/L. Figure 2 [18,29].

Modeling Technique, Artificial Neural Networks (ANNs)
ANNs is a data-driven modeling technique. It is proficient in modeling non-linear and complex relations among input and output data sets without the need of full physical recognition of the natural phenomena. The central theme of ANNs is borrowed from the analogy of highly simplified mathematical models of biological neural networks and it is inspired by the functioning mechanism of the brain. It has the capability to learn and generalize from historical data and previous examples

Modeling Technique, Artificial Neural Networks (ANNs)
ANNs is a data-driven modeling technique. It is proficient in modeling non-linear and complex relations among input and output data sets without the need of full physical recognition of the natural phenomena. The central theme of ANNs is borrowed from the analogy of highly simplified mathematical models of biological neural networks and it is inspired by the functioning mechanism of the brain. It has the capability to learn and generalize from historical data and previous examples to create meaningful explanations of problems [39,40]. The structure of ANNs consists of three or Water 2020, 12, 2218 7 of 18 more layers: One input layer, one or more hidden layer(s), and one output layer. The sole role of the input layer of nodes is to relay the input data to the neurons of the first hidden layer. The output layer is used to produce an output to the given input. The intermediate hidden layers may be only one hidden layer, acting as a collection of feature detectors. The determination of the appropriate network architecture is one of the most important and challenging tasks in the model building process [10,[41][42][43]. Figure 3 demonstrates a schematic description of a general ANN model of three layers. There are many methods to implement ANNs. Finding the best solution is a difficult task, as the modeler should choose an efficient test of a large number of options to keep the model within manageable scales. These assumptions are defined as network topology, training algorithm, and input selection [44]. There are many architectures and kinds of ANNs, such as multilayer perceptron network (MLP), radial basis functions (RBFs), general regression neural network (GRNN), and probabilistic neural networks (PNNs) [4,45,46]. to create meaningful explanations of problems [39,40]. The structure of ANNs consists of three or more layers: One input layer, one or more hidden layer(s), and one output layer. The sole role of the input layer of nodes is to relay the input data to the neurons of the first hidden layer. The output layer is used to produce an output to the given input. The intermediate hidden layers may be only one hidden layer, acting as a collection of feature detectors. The determination of the appropriate network architecture is one of the most important and challenging tasks in the model building process [10,[41][42][43]. Figure 3 demonstrates a schematic description of a general ANN model of three layers. There are many methods to implement ANNs. Finding the best solution is a difficult task, as the modeler should choose an efficient test of a large number of options to keep the model within manageable scales. These assumptions are defined as network topology, training algorithm, and input selection [44]. There are many architectures and kinds of ANNs, such as multilayer perceptron network (MLP), radial basis functions (RBFs), general regression neural network (GRNN), and probabilistic neural networks (PNNs) [4,45,46]. The MLP is the widely used neural network. Each neuron in the hidden or output layers receives weighted inputs, from all neurons in the previous layer. This transformation of the inputs to output within a single neuron is comparatively simple; the complexity of ANNs is eventually achieved by the interaction between neurons in network layers [47][48][49]. ANNs, which belong to black-box models, have been proven to be an effective alternative to conventional groundwater modeling tools. It has been widely and successfully applied in various hydrogeological applications. Compared to ANNs, the classical physically based numerical model requires a wide range of data and parameter values that must be available as well as the time and the expertise needed for model calibration. On the downside, the literature showed that ANNs have some limitations, including some unrealistic results compared with physically based numerical models. In addition to this, a slow learning speed, local minima, human intervention such as the learning rate, and the overfitting problem are also identified. ANN modelers experienced difficulties in input pattern selection and used input optimization or trial and error. Yet, ANNs is profound and characterized by their ability to quickly capture the behavior of a hydrogeological system and provide suitable predictions accurately [50][51][52]. The application of ANNs in engineering modeling has received gradually growing attention during the latest decade or so. There are numerous applications of ANNs in many fields of water quality and hydrology [13,[53][54][55][56]. The MLP is the widely used neural network. Each neuron in the hidden or output layers receives weighted inputs, from all neurons in the previous layer. This transformation of the inputs to output within a single neuron is comparatively simple; the complexity of ANNs is eventually achieved by the interaction between neurons in network layers [47][48][49]. ANNs, which belong to black-box models, have been proven to be an effective alternative to conventional groundwater modeling tools. It has been widely and successfully applied in various hydrogeological applications. Compared to ANNs, the classical physically based numerical model requires a wide range of data and parameter values that must be available as well as the time and the expertise needed for model calibration. On the downside, the literature showed that ANNs have some limitations, including some unrealistic results compared with physically based numerical models. In addition to this, a slow learning speed, local minima, human intervention such as the learning rate, and the overfitting problem are also identified. ANN modelers experienced difficulties in input pattern selection and used input optimization or trial and error. Yet, ANNs is profound and characterized by their ability to quickly capture the behavior of a hydrogeological system and provide suitable predictions accurately [50][51][52]. The application of ANNs in engineering modeling has received gradually growing attention during the latest decade or so. There are numerous applications of ANNs in many fields of water quality and hydrology [13,42,[53][54][55].

Methodology
Modeling the response of coastal aquifers to the overpumping on groundwater and operating future scenarios by using ANNs requires dealing with regular time series data to construct data modeling and a training data matrix.

Data Preprocessing and Variables Calculation
There are approximately 4000 wells in the Gaza Strip. Nearly all of these wells are privately owned and used for agriculture. About 100 wells owned and operated by municipalities are used for domestic supply. The data were collected from 56 wells, most of which are municipal wells and they almost cover the total area of the Gaza Strip as presented in Figure 4. The choice of these wells depends only on the availability of required data. Hydrogeologically, and depending on the case study conditions, the change of the chloride concentration (salinity) was assumed to be affected by many variables, such as infiltration, pumping, time duration of pumping from aquifer, aquifer depth, and aquifer thickness [2]. The modeling data were extracted mainly from 56 domestic wells in the Gaza Strip since they usually have records of the chloride concentration twice a year, in February and October periodically. Many data sources were neglected because of the deficiency of complete required data. Therefore, detailed pumping records were not obtained for years prior to 1996, the period of the model that includes the modeling and calibration, starting from 1997 to 2006 [42].
Water 2020, 12, x FOR PEER REVIEW 9 of 18 standard deviation, and ranges of different variables used to train the model are shown in Table 2. It is worth mentioning that the settings of the model were set as a 95% confidence interval.      The selection of appropriate model inputs is extremely important. The problem definition was achieved by specifying the input (independent) and the output (dependent) variables for the ANN model. There are six input variables: Initial chloride concentration (Clo); recharge rate (R); pumping rate (Q), which represents the total pumping quantities for each study well for 6 months, collected from Palestinian Water Authority's data bank, and then calculated for each time phase of 6 months; and the hourly rate was considered as input variables. The pumping average rate of area (Qr) was calculated for each governate in Gaza Strip separately by summation of all municipal abstraction quantities and dividing it to each governate area for each time phase of six months. Lifetime (Lt) was the total operation duration in years for each well. The aquifer thickness (Th) represents the saturated aquifer thickness. It was calculated for each study well separately using the groundwater tables and aquifers' cross-sections shown in Figure 1 deepening on the depth from the land surface to the well screen. The output was the final chloride concentration (Clf). Monthly pumping rates for the study well were collected from Palestinian Water Authority's data bank. The pumping average rate was calculated for each governate in Gaza Strip by summation of all municipal abstraction quantities and dividing it to the each governate area for each time phase [2,31].

Study wells
Recharge from rainfall is perhaps the most difficult parameter to quantify in the Gaza Strip due to all of the potential factors that affect the infiltration of rainwater. This includes land use, soil type, as well as other influencing factors. The groundwater is recharged from different sources, including rainwater, runoff, and return flow, which includes leakage from the municipal water distribution system, sewage infiltration, and irrigation return flow. Moreover, recharge comes from inflow from occupied areas and Egypt and from the Mediterranean Sea as a seawater intrusion. In this study, the recharge rate was calculated based on the Coastal Aquifer Management Program (CAMP) model [16].
The time distribution divides the year into two phases, A and B. Phase A starts from April until September, while phase B starts from October until March in the following year. For example, time phase 1996-A extends from April 1996 to September 1996, time phase 1996-B extends from October 1997 to March 1997, and time phase 1997-A extends from April 1997 to September 1997, etc. Therefore, all other factors were prearranged according to this time distribution. Considering only those cases that had complete numeric values for all variables without any missing data, only 499 cases satisfied the above-mentioned criteria from 1997 to 2004, and processed 56 municipal wells. These cases were the dataset, which was split into 3 sets: Training, calibration, and test. These sets were selected randomly by ANN model before starting the training process. The test set extraction was about 50% of cases for training, 25% for calibration, and 25% for testing [56]. The test set provided a means by which the network knew when to stop training and start using calibration and testing. The mean, standard deviation, and ranges of different variables used to train the model are shown in Table 2. It is worth mentioning that the settings of the model were set as a 95% confidence interval.

Scenarios Development
An evaluation of the effect of the overpumping on groundwater salinity was performed considering three future scenarios. These scenarios were mainly designed to study the influence of pumping on the chloride concentration of groundwater in the Gaza Strip and predict the chloride concertation after 10 years [32].
After the network was trained and tested, it was then examined against a set of cases withheld from it during its training session. The ANNs model was then ready to be applied to any other cases. Evaluation of the effect of the overpumping on groundwater salinity was performed for 10 years considering three future scenarios. These scenarios were mainly designed based on the pumping rate to study the influence of pumping and the pumping average rate on the chloride concentration of groundwater in the Gaza Strip. The pumping rate remained as it is in the present conditions (no change scenario), and the pumping rate was reduced to half and zero pumping conditions.

Performance Evaluation
The developed model's performances were initially evaluated for training, validating, testing, and overall datasets. The developed model performance was assessed based on different statistical criteria. The coefficient of determination (R 2 ) is a statistical measure used to indicate the strength of a linear relationship between two groups of data representing two variables or observed and predicted data [57,58]. R 2 can be estimated using the following Equation (1): where n is the number of data records, t is the actual/target output, and y is the modeled output. However, t and y are the averages. The R 2 value falls between 0 and +1; R 2 = 0 signifies no linear relationship between the variables, and R is +1 in a perfectly increasing linear relationship circumstance. Although, in practice, the correlation coefficient is a good indicator of performance, it also provides a simple way to evaluate the model performance by comparing it to advanced indicators, such as the error mean and average error (residual between target and actual output values) of the output variable; the average absolute error (difference between target and actual output values) of the output variable; and standard deviation of errors for the output variable. S.D. Ratio: The error/data standard deviation ratio.

Results and Future Scenarios Outcomes
After many training trials, the best neural network was determined to be the multilayer perceptron network (MLP). Figure 5 presents the observed versus the simulated chloride concentration using ANNs. A high correlation between the observed and the predicted values of the chloride concentration exists, where the correlation coefficient (r) between the predicted and the observed output values of the ANN model was 0.98. The values of regression statistics for the ANN model for all modeling data, training data set, calibration data set, and test data set are presented in Table 3. The high value of the correlation coefficient (r) showed that the modelled chloride concentration values using the ANN model are in good agreement with the observed chloride concentration. While, the low value of the error mean, Abs E. mean, and S.D. ratio showed that the error between the observed and simulated chloride concentration values using the ANN model are small. Table 3. The high value of the correlation coefficient (r) showed that the modelled chloride concentration values using the ANN model are in good agreement with the observed chloride concentration. While, the low value of the error mean, Abs E. mean, and S.D. ratio showed that the error between the observed and simulated chloride concentration values using the ANN model are small.  The developed model was utilized to predict the chloride concentration for October 2007. These data were not modeled or trained by the model. The spatial distribution of the measured chloride concentration is shown in Figure 6. The spatial distribution of the predicted chloride concentration is shown in Figure 7. A visual comparison between the measured and predicted values of the chloride concentration show a high degree of similarity between the predicted and the measured contour maps, indicating the high prediction efficiency of the models and opening the door to applying it to the evaluation of the effect of over pumping on groundwater salinity considering three future scenarios. The developed model was utilized to predict the chloride concentration for October 2007. These data were not modeled or trained by the model. The spatial distribution of the measured chloride concentration is shown in Figure 6. The spatial distribution of the predicted chloride concentration is shown in Figure 7. A visual comparison between the measured and predicted values of the chloride concentration show a high degree of similarity between the predicted and the measured contour maps, indicating the high prediction efficiency of the models and opening the door to applying it to the evaluation of the effect of over pumping on groundwater salinity considering three future scenarios. Evaluation of the effect of the over pumping on groundwater salinity was performed considering three future scenarios. These scenarios were mainly designed to study the influence of pumping on the chloride concentration of groundwater in the Gaza Strip for 10 years ahead.

Scenario 1: No Change of Pumping Condition
This scenario assumed that the pumping quantity and pumping average rates remain the same as in the current conditions. The ANN model was utilized to predict the chloride concentration in the Evaluation of the effect of the over pumping on groundwater salinity was performed considering three future scenarios. These scenarios were mainly designed to study the influence of pumping on the chloride concentration of groundwater in the Gaza Strip for 10 years ahead.

Scenario 1: No Change of Pumping Condition
This scenario assumed that the pumping quantity and pumping average rates remain the same as in the current conditions. The ANN model was utilized to predict the chloride concentration in the groundwater domestic wells in 2030. Figure 8   Evaluation of the effect of the over pumping on groundwater salinity was performed considering three future scenarios. These scenarios were mainly designed to study the influence of pumping on the chloride concentration of groundwater in the Gaza Strip for 10 years ahead.

Scenario 1: No Change of Pumping Condition
This scenario assumed that the pumping quantity and pumping average rates remain the same as in the current conditions. The ANN model was utilized to predict the chloride concentration in the groundwater domestic wells in 2030. Figure 8

Scenario 2: The Total Pumping to Be Reduced by Half
Scenario 2 assumes that the pumping quantity and pumping average rates will decrease to half of 50% of the current pumping rates for each well. This scenario can be achieved by utilizing other sources of water, such as treated wastewater and stormwater harvesting. In addition, to control illegal connections and decrease water consumption by public campaigns, Figure 9 depicts the predicted chloride concentration of pumped groundwater in the Gaza Strip for scenario 2 in 2030. The figures show that the chloride concentration decreased slowly in most areas of the Gaza Strip except for the Khan Younis area, where the chloride concentration almost remained stable and will not improve like other areas. It could be due to the relatively shallow aquifer thickness in this area compared with other areas in the Gaza Strip. The thickness of the entire coastal aquifer is on average about 120 m, but it is only 5-10 m in the south near the Younis area and Rafah. However, the worst appeared in the east area of khan Younis area due to the soil profile in this area, which is sandy loam soil (17.5% clay, 16.5% silt, 66% sand), which have very small recharge rates as stated in the CAMP project outcomes [16,29]. like other areas. It could be due to the relatively shallow aquifer thickness in this area compared with other areas in the Gaza Strip. The thickness of the entire coastal aquifer is on average about 120 m, but it is only 5-10 m in the south near the Younis area and Rafah. However, the worst appeared in the east area of khan Younis area due to the soil profile in this area, which is sandy loam soil (17.5% clay, 16.5% silt, 66% sand), which have very small recharge rates as stated in the CAMP project outcomes [16,29].

Scenario 3: Zero Pumping Condition
This scenario assumed that there is no pumping from groundwater and therefore pumping is completely stopped. Despite this, this scenario is hypothetical and impossible to achieve; modeling of groundwater salinity based on this scenario was performed to illustrate the differences among scenarios. Figure 10 presents the predicted chloride concentration in 2030. It was observed that the predicted chloride concentration of pumped groundwater in Gaza Strip in 2030 will reach very acceptable levels and there will be almost no groundwater salinity problem in most areas of Gaza Strip.
As the developed model shows a high accuracy, as described in Table 3, any future scenario can be applied and modelled with accordingly accurate predictions. By the selection of the three applied scenarios, it was intended to show the suitability of the model and to show the differences of the future salinity among them. We recognize that a zero-pumping condition is a hypothetical scenario, but we mention it just to prove that the aquifer can be recovered and rehabilitated by very strict measures, such as a zero-pumping situation. Scenario 3 is a hypothetical scenario that is proposed to stop abstraction from the aquifer since 2007. This scenario leads to complete aquifer recovery in 24 years (2007-2030), so the datum was the 2006 chloride concentration.

Scenario 3: Zero Pumping Condition
This scenario assumed that there is no pumping from groundwater and therefore pumping is completely stopped. Despite this, this scenario is hypothetical and impossible to achieve; modeling of groundwater salinity based on this scenario was performed to illustrate the differences among scenarios. Figure 10 presents the predicted chloride concentration in 2030. It was observed that the predicted chloride concentration of pumped groundwater in Gaza Strip in 2030 will reach very acceptable levels and there will be almost no groundwater salinity problem in most areas of Gaza Strip.

Conclusions
Based on the results obtained from this study, a new empirical model for groundwater salinity in the Gaza Strip utilizing ANNs was successfully developed and applied. This empirical model provided very accurate prediction, where the correlation coefficient (r) between the observed and As the developed model shows a high accuracy, as described in Table 3, any future scenario can be applied and modelled with accordingly accurate predictions. By the selection of the three applied scenarios, it was intended to show the suitability of the model and to show the differences of the future salinity among them. We recognize that a zero-pumping condition is a hypothetical scenario, but we mention it just to prove that the aquifer can be recovered and rehabilitated by very strict measures, such as a zero-pumping situation. Scenario 3 is a hypothetical scenario that is proposed to stop abstraction from the aquifer since 2007. This scenario leads to complete aquifer recovery in 24 years (2007-2030), so the datum was the 2006 chloride concentration.

Conclusions
Based on the results obtained from this study, a new empirical model for groundwater salinity in the Gaza Strip utilizing ANNs was successfully developed and applied. This empirical model provided very accurate prediction, where the correlation coefficient (r) between the observed and predicted output values of the model was 0.985. It was proven that the chloride concentration in groundwater was proportional to the pumping rate. By studying the effect of overpumping on groundwater salinity, it was proven that strong remedial actions to solve the groundwater deterioration problem in the aquifer of Gaza Strip (salinity) have to mainly be based on reducing the pumping rate from the groundwater at least by half.
The results showed that if the pumping rate is kept as the current pumping rates, the chloride concentration will increase rapidly in most areas of Gaza Strip and the availability of fresh water will decrease in disquieting rates by the year 2030. If the pumping rate is decreased by 50% of the current pumping rates, the chloride concentration of groundwater will decrease slowly in most areas of the Gaza Strip. This study is not only about the application of ANNs in the hydrogeological study to model the groundwater quality. The most important contribution and the novelty of the study is related to the use of the model outputs to predict future salinity levels based on selected scenarios related to groundwater pumping. Given that the modeling process and future prediction were done using a cost-effective approach, i.e., minimum input data, this is essential for any decision-maker. Further detailed studies must be performed with more accurate physical models to address all related variables that affect the ground water quality in GCA. The results are considered as an urgent call for developing an integrated water management strategy aiming at improving GCA quality by providing other drinking water resources in order to secure the increasing water demand.