Surface water resource and effect of weather parameters in estimating the annual rainfall: A case study in Lebanon

The quality and quantity of freshwater resources are continually decreasing in the world. The objective of this paper is to review the literature on the water resource with a focus on the surface water, quality of surface water in terms of physical and chemical properties in different locations in Lebanon. Moreover, one of the most important sources influencing the surface water is rainfall. Forecasting rainfall is one of the most essential issues in the hydrological cycle. It is very challenging because is still not possible to develop an ideal model given the uncertainty and unexpected variation. In the present study, prediction models using artificial neural networks (ANN) and multiple linear regressions (MLR) are developed to estimate the annual rainfall as a function of weather parameters and geographical coordinates. The annual data used in this study are recorded in 1942 locations in Lebanon. The latitude, longitude, and altitude of the location, global solar radiation, average temperature, wind speed, and relative humidity are used as the input variables and annual rainfall is estimated as the output variable. The measured values are compared versus those predicted by the ANN and MLR models by evaluating R-squared and Root mean squared error.


Introduction
Water availability and use depend on several factors including increased population, energy demand, and related environmental problems [1,2]. Climate change significantly affects the environment and natural resources [2]. Air temperature and precipitation (rainfall or snow) are the major parameters of climate that influence human activities such as urban water resources [3] and agricultural production [4,5]. Precipitation is one of the most important factors in the Earth's water cycle, affecting several human activities, like agriculture, with significant impacts on the economy [6,7].
Lebanon is a small Mediterranean country (surface area 10,452 km 2 , and average width 45 km) located in South-West Asia, between N latitude 34°42′ and 33°3′and E longitudes 35°6′ and 36°37′ [8]. Lebanon's physiography is unique, dominated by two mountain ranges which run parallel to the sea (NNE-SSW) and are separated by the Bekaa valley. Lebanon has mild, dry summers and cold, wet winters. The heaviest rainfall occurs between November and April, with relatively minimal precipitation, if any, between July and August [9].Lebanon is a Middle Eastern country that is fortunate to have significant water resources, unlike its neighbours. However, rain is mainly concentrated in the winter months. While water is abundant in winter, significant water shortages are still experienced around the country for the rest of the year. Besides, water quality in many areas is questionable. Recently, various models are used to estimate the monthly or annual rainfall such as the mathematical model, Neuro-Fuzzy Inference System (ANFIS), and Artificial Neural Network (ANN).
The review aims to assess the water resources in Lebanon focused on surface water and physicochemical of the surface water. One of the primary sources surface water is rainfall, thus in the present study, four meteorological parameters including average temperature (Tav), global solar radiation (GSR); relative humidity (RH) and wind speed (WS) are selected for predicting the annual rainfall. Also, geographical coordinates in terms of latitude, longitudes, and elevations are utilized as input variables for the model. For this purpose, Artificial Neural Networks (ANN) and Multiple Linear Regressions (MLR) are developed and tested for predicting annual rainfall in Lebanon. The present study aims to estimate the rainfall at any location in Lebanon where there are no measurements.

Surface water resources in Lebanon
Generally, Lebanon's water resources are classified into groundwater and surface water. The sources of surface water resources in Lebanon are from 16 river systems, as shown in Figure 1. The main water resources in Lebanon are groundwater (51%), and surface water (49%). The surface water resources are principally sourced from rivers (46%) and surface storage like dams and lakes (3%) [10].
Lebanon has 40 rivers, 16 of which are permanent ( Figure 1). The combined annual flow of rivers is estimated to be about 3,900 million cubic meters, where the majority of the flow (75%) occurs between January and May [11]. Surface water resources are mainly sourced from 16 river mainstream discharges in Lebanon, particularly: El Kabir, Ostuene, El Bared, Abou Ali, El Kjaouz, Ibrahim, EL Kalb, Beirut, Damour, EL Awali, El Zahrani, El Assi, Al Qasmieh, Litani, Wazzani and Hasbani ( Figure 1). The Kusba and Abu Samra comprise the Abou Ali river systems. Al-Janin and Al Khodaira constitute the Ibrahim river systems. In addition, El Yamouneh, Qaraoun heights, Qaraoun El Khardali, and El Khardali Sea form the Litani river systems. Table 1 presents the short time-series discharge data from 1971 to 1975 and from 2005-2009 for the 16 perennial rivers in Lebanon. According to the State of the Environment Report of Lebanon, El Assi and Hasbani are the only two rivers that do not discharge into the Mediterranean Sea. Additionally, the highest river flows in Lebanon are associated with Nahr el Litani, Nahr Ibrahim, and Nahr el Assi. Rivers are mainly replenished from springs that are fed from melting snow. Over 2,000 regular springs feed into different streams in Lebanon, creating an aggregate of 1,150-1,200 Mcm/yr of water that is not completely misused [12]. Most of the surface water resources come from springs, where 637 Mcm/yr is currently used [13]. The surface water supply also comes from storage dams, namely the Qaroun dam and the Shibroh Dam, which currently provide about 45 Mcm/yr of water [13]. Table 2 shows the water resources available annually in Lebanon. In addition, exploited resources according to water source type in Mm 3 /y are listed in Table 3.
Water resources in Lebanon are heavily polluted, as domestic and industrial wastewater is largely dne ttetuer tnu r intolerable agricultural practices exacerbate the situation. The main rivers in t tngnretftrft areaaeretftevrgir tbet aterbgnetnanteagnrudtregreetriegorgir torvtotateroeaberposes a real threat to public health [16].
Random disposal of solid waste also results in water contamination due to leakage of chemicals.
nevger betnabter bgn etnanteagnr gir vd itbtr tnur a gdnuotet r i gnr anudve ater uavbet atvr avr common. Untreated wastewater containing heavy metals is often disposed [17], while leaks from underground gasoline tanks and uncontrolled pumping of oil and petroleum by-s gudbevr t tr tev gr bgnngnc Coastal waters in Lebanon are also heavily polluted by wastewater flowing from domestic sewage discharges, industrial wastewater discharges, coastal agricultural waste, and huge wasted dumps ( in Tripoli, Burj hammoud, Beirut, Saida, and Tyre).Oil spills especially the large leakage that occurred after the 2006 war with Israel, and coastal power stations ( in Beddawi, Zouk, Jiyeh, and Al-zahrani) are the other sources of pollution.. Several studies gnreetrbgtvetertnurnt antrotet vranr t t ngnretftr igdnurft areaaeretftevrgirsgeedetneverst eabdet earanrareas close to the three main cities of Tripoli, Beirut, and Sidon, including heavy metals, which t treglabregr geernt antrtnuredntnrsgsdeteagnv [18].
The intensive use of fertilizers and pesticides in agricultural practices, especially during the dry vttvgnveretvreturegreetriaeet anargirnae tetvranvautreetrvgaertnurbgnetnanteagnrgira gdnuwater vavetnvr oaeereaaerbgnbtne teagnvrigdnuranrbgtvetersetan r [19], and the ytlttr teeta r [20]. As farmers rely on wells for irrigation, health and tnfa gnntneter bgnbt nvr t tr anb ttvanacr gnr tuuaeagner aer avr tsg etur eeter tor vtotatravrdvturig rirrigation in many areas, including Akkar and Bekaa and farmers ee gdaegdereetrbgd ne art trealtearegr tvg eregreeavrs tbeabtroetnri tveotet ravrngertftaet et r [21].   In recent years, scientific researchers have been examined the quality of surface water in different locations in Lebanon. For instant, Najjar et al. [22] measured twenty-three the physicochemical parameters of Ibrahim River. The results indicated that the quality water was considered as medium to good water quality with an average value of 69.0± 1.9. Haydar et al. [23] measured the physicochemical parameters including pH, T°, TDS, EC, Na+, Ca2+, Mg2+, Clí, SO2í4, NH3+, NOí3, PO2í4, K+ and Heavy metals of Upper Litani River Basin for different seasons. The results demonstrated that the degree of pollution depends on the location and the season. Also, they concluded that the presence of pollution in terms of mineral and anthropogenic came from municipal wastewater and agricultural purposes  [24] evaluated the water quality in the coastal area of Tripoli using Landsat 7 ETM+ data. They found that the coastal area of Tripoli is described as moderate eutrophic conditions with fluvial and wastewater runoff sources. Massoud [25] evaluated the water quality index of the small Mediterranean river in Southern Lebanon (Damour River). The results show that the index of water quality is classified as good and anthropogenic activities are mainly source of pollution affect the river. Saadeh et al. [26] assessed the groundwater quality of the Upper Litani River Basin with nearly 300,000 persons depends on it for their daily domestic needs. The authors found that Total Dissolved Solids (TDS) of the river was above the normal range of 100-500 mg/L in the summer season, which indicated that the river received runoff from all other tributaries including the heavily polluted Berdawni. Daou et al. [27] measured the physicochemical and microbiological parameters of the surface water samples taken from the Arka River located in the Akkar District, north of Lebanon. The analysis showed that the most polluted sources in the river come from flatland and the surrounding villages. Therefore, an effective surface water quality management system could be established enabling the proper use of water for irrigation purposes. Fadel and Slim [28,29] analyzed the physicochemical parameters in terms of pH, electrical conductivity, TDS, turbidity, alkalinity, Ca, Mg, TH, Cl−, SO2 4−, NH3, NO− 3, PO3 4−, Fe, Al, Na, Zn, Cr, Cu and As of Qaraaoun reservoir and the results are compared with water standards. The results showed that suboptimal quality would probably be remediated through customary water treatment processes. Korfali et al. [30] evaluated the water quality of the Qaraaoun reservoir that considered as having three water quality zones and measured the parameters (pH, Eh, DO and temperature) of 15 samples collected from different site. The authors found that the sediment data showed higher metal contents where the river entered the reservoir which matched higher concentrations of water parameters at the influx site. Korfali and Davies [31] investigate the variation of the total metal content (Fe, Mn, Zn, Cu, Pb, and Cd) in bed sediments and water of River Nahr-Ibrahim. The results indicated that the decreasing in water pH caused by the decrease in precipitation rate, lowering the level of water and the dilution of industrial discharges. Korfali and Jurdi [32] examined the quality of domestic water in Beirut city. The results showed deterioration patterns in domestic water quality. Korfali and Jurdi [33] measured the physicochemical and bacteriological parameters of domestic water collected from three household water sources. They found a high frequency of water-borne diseases in the collected water. Korfali and Jurdi [34] determined metal speciation sediment chemical fractions and metal speciation in reservoir water. The measured data indicated that the highest percentages of total metal content in sediment fractions were for Fe in residual followed by reducible, Cr and Ni in residual and is reducible, Cu in organic followed by exchangeable, Zn in residual and inorganic, Pb in organic and carbonate, Cd was mainly in carbonate. Kouzayha et al. [35] analyzed chemically the water quality of drinking water collected from the major cities in Lebanon. The authors found that high pesticide ecotoxicological risk determined by diazinon, chlorpyrifos, fenpropathrin and bifenthrin insecticides in many surface waters. Semerjian [36] analyzed physicochemical and bacterial water quality parameters of domestic bottled water collected from shops and supermarkets throughout Lebanon. The results showed negative growth for fecal Coliforms and positive results for total coliforms. Houri et al. [37] studied the chemical and microbiological properties of Lebanese perennial coastal rivers during the dry season. They found that the most polluted rivers in most categories were Abu Ali and Antelias. Table 4 summarizes the physical parameters for selected rivers in the dry season (July, August, and September). It is found that pH values for all rivers with exception Damour river are within the limit range of limit drinking water. In addition, it is observed that the lowest value of saturation was recorded in Abu Ali and Antelias rivers and the highest value was obtained in the Damour river. Increase algal growth produced oxygen from photosynthesis led to an increase in the level of dissolved oxygen. In the dry season, the highest average TDS (Total dissolved solids) was found in the Awali river (1863 g/s) and the lowest was found in the Damour river (44 g/s) as shown in Table 5.

Material and method
Climate change affects water resources through changes in temperature, rainfall, runoff, and groundwater recharge. Furthermore, the changes in air temperature and rainfall could affect river flows, and hence, the mobility and dilution of contaminants [38]. Consequently, this study aims to study is to

Study region and data
In the current study, the used data consist of the annual rainfall, average temperature, relative humidity, global solar radiation and wind speed. The data are recorded in 1942 locations in Lebanon. The geographical area of the selected location is bounded by the latitude of 34.691° and 33.060° and by the longitudes of 36.575° and 35.179° (see Figure 2), whose elevations range from 1 to 2300 m above mean sea level.

Artificial neural networks
The most widespread technique used in calculating outputs of many systems is the artificial neural networks (ANN) model. A large number of academicians in many different fields have used ANN in their studies [39][40][41][42]. The artificial neural networks (ANN) model, also known as the black-box model, is composed of interconnected processing units called artificial neurons or nodes [43]. Generally, the multilayer feed-forward neural network is widely used for solving engineering problems. It consists of three layers, namely the input layer(s), Hidden layer(s) and output layer(s). Besides, the number of these layers depends on the nature of the problem.
In this study, the feed forward architecture with the three layers is used. TRAINLM is used as a training function that updates the weight and bias values of the neuron connections according to Levenberg-Marquardt (LM) optimization. The back-propagation algorithm is used as a learning algorithm and it is a gradient descent algorithm. The activation function for the neurons can be linear or non-linear. The logistic-sigmoid (logsig) and tangent-sigmoid (tansig) were used as an activation function whose output lies between 0 and 1. By trial and error, the optimum number of nodes in the hidden layers, the most suitable transfer function and the number of neurons are determined. To obtain the best performance results, various ANN models are designed.
In this research, a conventional data division technique was used to divide the data, whereby the sets were divided on an arbitrary basis and the statistical properties of eetr tvstbeaftr utetr vtevr ot tr nger bgnvaut turiiiacrnss glantetearnedrgireetrutetrotvrdvturig re tananaeroeaetreetr tntananarredrotvr tvt fturig retveanacreetre tining data was used to train the ANN models with the LM algorithm. The testing data do not affect training and provide an independent measure of network performance during and after training. Moreover, normalization of the data is required for improving the performance of the ANN model. The minimum (min) and maximum (max) values of the inputs and output parameters are shown in Table 6. In general, the number of hidden layers and the number of neurons are the most factors that can affect the performance of the ANN model. Figure 3 shows the structure of the ANN model used in this study. In this study, the number of epochs and performance goal were 100,000 and 0.001, respectively. In addition, the number of the hidden layers varied between 1 and 10, while the number of neurons varied between 5-50 neurons.

Multiple linear regressions (MLR)
Multiple linear regressions are described as the relationship dependent (y) and independent variables (x). It can be expressed as where denotes the dependent variable (rainfall) and where i=1,2,..,n, denotes the explanatory or independent variables and β is called the intercept. Minitab 17 was used for the regression and testing of the data.

Results and discussions
The annual values of rainfall and the other parameters (Tav, RH, GSR, WS) at 1942 locations for 2016 are recorded and La, Lo and El of each location are used as explanatory input variables. First, the data are randomly split into 80% training data (1554 data) and 20% testing data (388). The result of the ANN model is compared with the MLR model. The statistics of the data are summarized in Table 7.Different statistical measures, including the mean, standard deviation (SD),coefficient of variation(CV), minimum, maximum, skewness and kurtosis are calculated for each variable. It is found that the annual mean rainfall varied from 90 mm to 165 mm. The maximum and minimum annual rainfall recorded in choueifat and El Khraïbé locations, respectively. The CV values are high, ranging from 0.76 to 68.32.
During the investigation period, the Skewness value for rainfall is negative, which indicates that all distributions are right-skewed. In addition, Lebanon has a maximum and minimum average temperature of 23.561℃ and 12.487℃, respectively. Moreover, it is found that the annual mean wind speed in the country was 2.04m/s. additionally, the global solar radiation and relative humidity values ranged between 184.93-230.62 kWh/m 2 and 55.326-76.501%, respectively. Consequently, it can be established that this country has considerable solar potential. Generally, the mean and standard deviation values suggest that there is good consistency in the meteorological parameter behaviour. The La, Lo, El, Tav, RH, GSR, WS are utilized in the input layer as input data for the feed forward architecture with a back propagation algorithm (Figure 3). The annual rainfall is the outcome of the output layer. In this study, various ANN configurations were designed and the number of hidden layers and neurons was determined by trial and error for obtaining the best performance results. The logisticsigmoid and the tangent-sigmoid functions are tried in the hidden layer and output layer. The best performance of the network was obtained by training the developed ANN architecture several times until the MSE showed the minimum value. The same-trained network was tested with the new datasets to check the performance of the network. Table 8 shows the best number of hidden layers and neurons, training rule, activation function, epochs, R-squared and mean squared error (MSE) that were chosen for each ANN model. Also, the R-squared and RMSE for testing data are tabulated in Table 8. It is found that the best ANN model has two hidden layers with 5 neurons and having TANSIG as the activation function.      In addition, it is found that the scatter diagrams, the noises of the predicted values around the bestfit lines are wider for MLR models. Furthermore, the R 2 for annual rainfall estimated by the ANN and MLR for test phases are found to be 0.7436 and 0.5805, respectively (see Table 9). It is concluded that the ANN model is a capable model to define a non-linear relationship as output variable and weather parameters and geographical coordinates of any location in Lebanon as input variables without needing a priori information and without having to make preliminary assumptions.

Conclusions
The paper has highlighted that water resources in Lebanon focused on surface water and the quality of the water. Since the main surface water sources are rainfall thus, annual climate data in terms of rainfall, average temperature relative humidity, global solar radiation, and wind speed were analyzed statistically to show the effect of climate parameters on the rainfall. This study has shown the power of ANN to evaluate the most influencing input parameters in the prediction of monthly rainfall. ANN model using a back-propagation algorithm was developed. Out of the ANN and MLR models, the ANN model has given the best prediction with the highest R-squared and minimum RMSE. These models can be used for determining the level of groundwater based on the amount of rainfall and rainfall distribution at any site in Lebanon. Therefore, it can be used for the assessment of trends in groundwater levels across Lebanon.