Groundwater Level Forecasting Research Articles

Groundwater level forecasting with artificial neural networks: a comparison of long short-term memory (LSTM), convolutional neural networks (CNNs), and non-linear autoregressive networks with exogenous input (NARX)

Abstract. It is now well established to use shallow artificial neural networks (ANNs) to obtain accurate and reliable groundwater level forecasts, which are an important tool for sustainable groundwater management. However, we observe an increasing shift from conventional shallow ANNs to state-of-the-art deep-learning (DL) techniques, but a direct comparison of the performance is often lacking. Although they have already clearly proven their suitability, shallow recurrent networks frequently seem to be excluded from the study design due to the euphoria about new DL techniques and its successes in various disciplines. Therefore, we aim to provide an overview on the predictive ability in terms of groundwater levels of shallow conventional recurrent ANNs, namely non-linear autoregressive networks with exogenous input (NARX) and popular state-of-the-art DL techniques such as long short-term memory (LSTM) and convolutional neural networks (CNNs). We compare the performance on both sequence-to-value (seq2val) and sequence-to-sequence (seq2seq) forecasting on a 4-year period while using only few, widely available and easy to measure meteorological input parameters, which makes our approach widely applicable. Further, we also investigate the data dependency in terms of time series length of the different ANN architectures. For seq2val forecasts, NARX models on average perform best; however, CNNs are much faster and only slightly worse in terms of accuracy. For seq2seq forecasts, mostly NARX outperform both DL models and even almost reach the speed of CNNs. However, NARX are the least robust against initialization effects, which nevertheless can be handled easily using ensemble forecasting. We showed that shallow neural networks, such as NARX, should not be neglected in comparison to DL techniques especially when only small amounts of training data are available, where they can clearly outperform LSTMs and CNNs; however, LSTMs and CNNs might perform substantially better with a larger dataset, where DL really can demonstrate its strengths, which is rarely available in the groundwater domain though.

Deterministic Analysis and Uncertainty Analysis of Ensemble Forecasting Model Based on Variational Mode Decomposition for Estimation of Monthly Groundwater Level

Precise multi-time scales prediction of groundwater level is essential for water resources planning and management. However, credible and reliable predicting results are hard to achieve even to extensively applied artificial intelligence (AI) models considering the uncontrollable error, indefinite inputs and unneglectable uncertainty during the modelling process. The AI model ensembled with the data pretreatment technique, the input selection method, or uncertainty analysis has been successfully used to tackle this issue, whereas studies about the comprehensive deterministic and uncertainty analysis of hybrid models in groundwater level forecast are rarely reported. In this study, a novel hybrid predictive model combining the variational mode decomposition (VMD) data pretreatment technique, Boruta input selection method, bootstrap based uncertainty analysis, and the extreme learning machine (ELM) model named VBELM was developed for 1-, 2- and 3-month ahead groundwater level prediction in a typical arid oasis area of northwestern China. The historical observed monthly groundwater level, precipitation and temperature data were used as inputs to train and test the model. Specifically, the VMD was used to decompose all the input-outputs into a set of intrinsic mode functions (IMFs), the Boruta method was applied to determine input variables, and the ELM was employed to forecast the value of each IMF. In order to ascertain the efficiency of the proposed VBELM model, the performance of the coupled model (VELM) hybridizing VMD with ELM algorithm and the single ELM model were estimated in comparison. The results indicate that the VBELM performed best, while the single ELM model performed the worst among the three models. Furthermore, the VBELM model presented lower uncertainty than the VELM model with more observed groundwater level values falling inside the confidence interval. In summary, the VBELM model demonstrated an excellent performance for both certainty and uncertainty analyses, and can serve as an effective tool for multi-scale groundwater level forecasting.

Combining Group Method of Data Handling with Signal Processing Approaches to Improve Accuracy of Groundwater Level Modeling

Groundwater level forecasting is a paramount necessity for integrated management of a basin. Development of suitable models is an essential step in determining groundwater level fluctuations in the future. The main objective of this study was to provide a powerful hybrid model by combining the group method of data handling (GMDH) and certain signal processing techniques, i.e., ensemble empirical mode decomposition (EEMD), wavelet transform (WT) and wavelet packet transform (WPT) for groundwater level forecasting in monthly time steps. Two different plains were selected to assess the performance of the afore-mentioned methods. The results showed that all of these preprocessing methods improved the capability of the group method of data handling model. The EEMD–GMDH model outperformed the WT–GMDH model. The WPT–GMDH model had a superior performance compared to both of the afore-mentioned hybrid models. However, it was shown that WPT–GMDH model had more computational cost that may affect the feasibility of this modeling approach in some cases. Finally, the EEMD–GMDH model can be introduced as a suitable modeling approach for groundwater level forecasting.

Predicting groundwater level fluctuations under climate change scenarios for Tasuj plain, Iran

Coastal aquifers world-wide are being confronted with several major challenges, such as overextraction of groundwater, climate change impacts, contamination by wastewater, and saltwater intrusion into water resources. Climate change induced alteration of the hydrological cycle is one of the main threats to future accessibility of water resources. Effective prediction of possible impacts of climate change on groundwater reserves, a crucial water resource, could be of great importance for sustainable water management. In a comparative study, artificial neural network (ANN), least square support vector machine (LSSVM), and nonlinear autoregressive network with exogenous inputs (NARX) models was applied to evaluate possible impacts of three representative concentration pathways (RCP) climate change scenarios (RCP2.6, RCP4.5, RCP8.5) on groundwater levels in Tasuj Plain, Iran. Four general circulation models (GCM) was used to predict temperature and precipitation values for the future period 2022–2050 and found that future temperature increased, while the amount of precipitation decreased. To improve the accuracy of three models in groundwater level prediction, db4 wavelet transform was applied. The results indicated that the Wavelet-NARX approach gave the best accuracy in forecasting groundwater level in the study area. In all cases, prediction indicated that groundwater level in all representative wells would decline in future.

Groundwater level prediction in arid areas using wavelet analysis and Gaussian process regression

Utilizing new approaches to accurately predict groundwater level (GWL) in arid regions is of vital importance. In this study, support vector regression (SVR), Gaussian process regression (GPR), and their combination with wavelet transformation (named wavelet-support vector regression (W-SVR) and wavelet-Gaussian process regression (W-GPR)) are used to forecast groundwater level in Semnan plain (arid area) for the next month. Three different wavelet transformations, namely Haar, db4, and Symlet, are tested. Four statistical metrics, namely root mean square error (RMSE), mean absolute percentage error (MAPE), coefficient of determination (R 2), and Nah-Sutcliffe efficiency (NS), are used to evaluate performance of different methods. The results reveal that SVR with RMSE of 0.04790 (m), MAPE of 0.00199%, R 2 of 0.99995, and NS of 0.99988 significantly outperforms GPR with RMSE of 0.55439 (m), MAPE of 0.04363%, R2 of 0.99264, and NS of 0.98413. Besides, the hybrid W-GPR-1 model (i.e. GPR with Harr wavelet) remarkably improves the accuracy of GWL prediction compared to GPR. Finally, the hybrid W-SVR-3 model (i.e. SVR with Symlet) provides the best GWL prediction with RMSE, MAPE, R2, and NS of 0.01290 (m), 0.00079%, 0.99999, and 0.99999, respectively. Overall, the findings indicate that hybrid models can accurately predict GWL in arid regions.

ON THE EVALUATION OF THE GRADIENT TREE BOOSTING MODEL FOR GROUNDWATER LEVEL FORECASTING

Though groundwater is a replenishable resource, it’s over exploitation has posed greater problem of its depletion. Hence, monitoring and forecasting of groundwater levels has become a primary task of governmental water boards/agencies for sustainable water management. The current study focused on evaluating the performance of Gradient Tree Boosting (GTB) model with that of conventional Adaptive Neuro-Fuzzy Inference System (ANFIS) and Group Method of Data Handling (GMDH) models in forecasting groundwater levels of two coastal aquifers. Data of two groundwater level monitoring wells penetrating into unconfined aquifers located at Shirtadi and Rayee near to Mangalore city of Karnataka state, India was considered in the present study. Monthly groundwater level data of the years 2000 – 2013 were used for model simulation; wherein 70% of data was used for model training and the remaining 30% served as testing data. Comparative result evaluation shows that the proposed GTB approach for one month ahead groundwater level forecasting was giving much accurate results than the other models for the same period of time and same set of data. For Rayee monitoring well, the error statistic, RRMSE of GTB, GMDH and ANFIS models obtained during test phase were 0.473, 0.517 and 0.7522, respectively. The comparison is examined further with different performance metrics.

Joint estimation of states and parameters of two-layer coastal aquifers based on ENKF

Abstract Management of groundwater resources has become a source of heated discussion in coastal hydrogeology. Thus, we introduced an Ensemble Kalman Filter (ENKF) into a two-layer confined groundwater model based on the interactive operation between the MATLAB and GMS to investigate the capability of ENKF under complex conditions and obtain a relatively new forecasting method. ENKF was employed to assimilate and forecast groundwater levels, and invert the hydraulic conductivity (K) of the heterogeneous study area, where the initial values of K were obtained by using trial-and-error based on the two-period groundwater levels. After comparing the efficiencies in forecasting groundwater levels among ENKF, the modified model, and the initial model, four major conclusions could be drawn. ENKF converged fast when forecasting groundwater levels and the accuracy was high. Various convergent results would be represented by ENKF when K in different layers was observed in the same error. ENKF performed better than the initial simulation when monitored data subjected to a certain range of interferences. Forecasting accuracy in the middle of the study area could be enhanced by the large improvement degree of K through ENKF. Therefore, this analytical method could be a theoretical reference for groundwater resources management in coastal areas.

Comparative study of groundwater level forecasts using hybrid neural network models

Groundwater is the world's central supply of fresh water. Water supply policies, particularly in dry seasons, thus need to be based on accurate modelling of groundwater level (GWL) fluctuations. In the work reported in this paper, a hybrid wavelet-transform-based extreme learning machine (ELM) model was investigated for predicting GWL. Two other popular models – a wavelet-transform based artificial neural network and a wavelet-transform-based adaptive neuro-fuzzy interference system – were used to evaluate the model. GWL data and mean temperatures of observation wells in an Iranian watershed between 1981 and 2017 were used in the study. The performance of the models was assessed be evaluating their root mean square error, correlation coefficient and mean absolute error. The wavelet-transform-based ELM model outperformed the other two models with a correlation coefficient of 0.983 during a 1 month period. The model was also superior to the others in terms of training and testing speeds.

Eigenmodels to forecast groundwater levels in unconfined river-fed aquifers during flow recession

Low-land alluvial gravel aquifers are formed from, and tend to be recharged, by rivers. These interconnected river - groundwater systems can be highly dynamic with groundwater levels following the seasonality of the hydrological regime of the river. The associated groundwater resources are regularly under stress during summer periods when abstractive demand is high and recharge is low. Predicting lead-times for critical groundwater levels allows for a more flexible and adaptive groundwater management. An eigenmodel approach is proposed here as a way of making such predictions, fast and efficiently. The eigenmodel is a mathematical concept that represents the hydraulic function of a groundwater aquifer as a set of conceptual linear reservoirs, arranged in-series. River recharge, land surface recharge, and groundwater abstraction for irrigation are considered as model forcings. The eigenmodel approach is demonstrated on three wells of the unconfined Wairau Aquifer in the Marlborough District of New Zealand, which are used for water resources management. Individual eigenmodels were calibrated to historic data and predictive uncertainty bounds were determined by Markov chain Monte Carlo sampling. Hindcasting of past recession periods showed a low predictive error of the models and a good coverage of the predictive uncertainty bounds. The main advantage of the approach is a 4-orders of magnitude higher computational efficiency compared to a numerical benchmark model. This allows for probabilistic simulation in operational forecasting of groundwater levels. The framework is implemented as a web application for 30-day operational forecasts that comprises automatic data downloads and model input generation, stochastic simulation, uncertainty estimation, visualization, and daily updates on a website.

Multiscale groundwater level forecasting: Coupling new machine learning approaches with wavelet transforms

Groundwater level (GWL) forecasting is crucial for irrigation scheduling, water supply and land development. Machine learning (ML) (e.g., artificial neural networks) has been increasingly adopted to forecast GWL due to its ability to model nonlinearities between GWL and its drivers (e.g., rainfall). Although ML approaches have been successful at forecasting GWL, they are often inaccurate when GWL exhibits multiscale changes (e.g., due to urbanization). To address this shortcoming, wavelet transforms (WT) are routinely coupled with ML methods. Unfortunately, researchers frequently neglect key issues associated with WT that render such forecasts useless for real-world scenarios. This study demonstrates how new ML methods, such as eXtreme Gradient Boosting and Random Forests, can be properly coupled with WT to generate accurate GWL forecasts (1-–3 months ahead) for 7 wells in Kumamoto City in Southern Japan that can be used to help address current pressing issues such as groundwater quality and land subsidence.

Comparison of a physical model and phenomenological model to forecast groundwater levels in a rainfall-induced deep-seated landslide

Increases in groundwater levels play an important role in triggering deep-seated landslides. As such, it is important to be able to accurately predict variations in groundwater levels. In order to select a more suitable method for the prediction of groundwater levels in relation to deep-seated landslides, a comparative study was conducted between a physical model and phenomenological model. The physical model is a finite-element seepage code named Slide by Rocscience. The phenomenological model is a Particle Swarm Optimization Support Vector Machine (PSO-SVM). In order to obtain more accurate calculated results from the physical seepage model, the input parameters of the physical seepage model were calibrated by using a trial-error method to compare the computed results with actual monitoring data. The input data of the phenomenological model were also processed in order to obtain more accurate calculated results. The results showed that the physical seepage model was difficult to well calibrate in the condition of less data and hence performed poorly. The validation results showed that the phenomenological model performed better. The Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) of phenomenological model were 0.052 m and 0.043 m, respectively, compared with the values of physical seepage model were 0.92 m and 0.81 m, respectively. It means if constructing a satisfactory physical seepage model was difficult or understanding of physical process could not be considered, the phenomenological model might be sufficient.

A Nonlinear Autoregressive Modeling Approach for Forecasting Groundwater Level Fluctuation in Urban Aquifers

The application of a nonlinear autoregressive modeling approach with exogenous input (NARX) neural networks for modeling groundwater level fluctuation has been examined by several researchers. However, the suitability of NARX in modeling groundwater level dynamics in urbanized and arid aquifer systems has not been comprehensively investigated. In this study, a NARX-based modeling approach is presented to establish a robust water management tool to aid urban water managers in controlling the development of shallow water tables induced by artificial recharge activity. Temperature data series are used as exogenous inputs for the NARX network, as they better reflect the intensity of artificial recharge activities, such as excessive lawns irrigation. Input delays and feedback delays for the NARX networks are determined based on the autocorrelation and cross-correlation analyses of detrended groundwater levels and monthly temperature averages. The validation of the proposed approach is assessed through a rolling validation procedure. Four observation wells in Kuwait City are selected to test the applicability of the proposed approach. The results showed the superiority of the NARX-based approach in modeling groundwater levels in such an urbanized and arid aquifer system, with coefficient of determination (R2) values ranging between 0.762 and 0.994 in the validation period. Comparison with other statistical models applied to the same study area shows that NARX models presented here reduced the mean absolute error (MAE) of groundwater levels forecasts by 50%. The findings of this paper are promising and provide a valuable tool for the urban city planner to assist in controlling the problem of shallow water tables for similar climatic and aquifer systems.

Lion Algorithm-Optimized Long Short-Term Memory Network for Groundwater Level Forecasting in Udupi District, India

Groundwater is a precious natural resource. Groundwater level (GWL) forecasting is crucial in the field of water resource management. Measurement of GWL from observation-wells is the principle source of information about the aquifer and is critical to its evaluation. Most part of the Udupi district of Karnataka State in India consists of geological formations: lateritic terrain and gneissic complex. Due to the topographical ruggedness and inconsistency in rainfall, the GWL in Udupi region is declining continually and most of the open wells are drying-up during the summer. Hence, the current research aimed at developing a groundwater level forecasting model by using hybrid long short-term memory-lion algorithm (LSTM-LA). The historical GWL and rainfall data from an observation well from Udupi district, located in Karnataka state, India, were used to develop the model. The prediction accuracy of the hybrid LSTM-LA model was better than that of the feedforward neural network (FFNN) and the isolated LSTM models. The hybrid LSTM-LA-based forecasting model is promising for a larger dataset.

Neural network modeling for groundwater-level forecasting in coastal aquifers

Advances in the artificial intelligence-based models can act as robust tools for modeling hydrological processes. Neural network architectures coupled with learning algorithms are considered as useful modeling tools for groundwater-level fluctuations. Emotional artificial neural network coupled with genetic algorithm (EANN-GA) is one such novel hybrid neural network which has been used in the present study for the forecasting of groundwater levels at three sites (Site H3, Site H4.5, and Site H9) in a coastal aquifer system. This study was conceived to address and investigate the efficiency of the ensemble model (EANN-GA) for forecasting one-month ahead groundwater level and to compare its performance with emotional artificial neural network (EANN), generalized regression neural network (GRNN), and the conventional feedforward neural network (FFNN). Variations in the rainfall, tidal levels, and groundwater levels are selected as inputs for the development of EANN-GA, EANN, GRNN, and FFNN models. Suitable goodness-of-fit criteria such as Nash–Sutcliffe efficiency (NSE), bias, root mean squared error (RMSE), and graphical indicators are used for assessing the efficiency of the developed models. The improvement in the performance of the EANN-GA model over the developed EANN, GRNN, and FFNN models in terms of NSE is 0.81, 6.02, and 9.56% at Site H3; 4.35, 5.50, and 22.68% at Site H4.5; and 1.05, 7.18, and 21.75% at Site H9. Thus, it can be inferred that the EANN-GA model outperforms the developed EANN model, GRNN model, and FFNN model. Further, this paper examines the predictive capability of extreme events by the EANN-GA, EANN, GRNN, and FFNN models. The RMSE values of the EANN-GA model at all peak points are found as 0.27, 0.23, and 0.10 m at sites H3, H4.5, and H9, respectively, and the results indicate superior performance of EANN-GA model. To check the generalization ability of the developed EANN-GA models, they are validated with the data of another site (Site I2) located in the same coastal aquifer. Superior prediction capability and generalization ability make the EANN-GA model a better alternative for predicting groundwater levels. Overall, this study demonstrates the effectiveness of EANN-GA in modeling spatio-temporal fluctuations of groundwater levels. It is also concluded that the EANN-GA model yields remarkably better predictions of extreme events, and hence, it could be a promising technique for developing alarm systems for real-world water problems.

Prediction of groundwater level fluctuations by artificial neural network models : A case study

In the last few years, due to excessive exploitation of groundwater resources, the water level has gone down in many parts of India. According to a data from the Central Ground Water Board (CGWB), the average groundwater level in India has decreased by 61 per cent between 2007 and 2017, due to which the availability and quality of water are taking serious problems day-by-day. According to the CGWB (2017), 313 blocks out of a total of 6881 blocks are Critical, 1186 blocks are over exploited, and 100 blocks fall into the category of saline. And there are 94 such blocks where the availability of groundwater is significantly less. The overall development and management of groundwater resources are critical because of the water crisis arising from the declining level of groundwater resources and its immense importance in the social and economic growth of the country. Accurate forecasting of groundwater levels is essential for efficient management of groundwater resources and sustainable development of ecosystems. In this study, Multi-Layer Perceptron (MLP) artificial neural network (ANN) was used to predict groundwater levels in a selected aquifer system. The development of the ANN model included data on rainfall, temperature and river water level and groundwater level. The ANN model was trained using the gradient decant algorithm (GDM). The said model was applied and validated at seven different observation locations, and the estimated capacity of the ANN model was developed for each location. This was assessed using four statistical indicators [Bias, RMSE, NSE and MSE] as well as visual testing. Based on the results of this study, the Neural Network Model (ANN) was found to be efficient for forecasting monthly groundwater levels at almost all selected observation locations. The study concluded that artificial neural network techniques (ANNs) could be used efficiently to predict groundwater level fluctuations, especially in data-scarce situations.

Comparative Study between Wavelet Artificial Neural Network (WANN) and Artificial Neural Network (ANN) Models for Groundwater Level Forecasting

Groundwater level fluctuation modeling is a prime need for effective utilization and planning the conjunctive use in any basin.The application of Artificial Neural Network (ANN) and hybrid Wavelet ANN (WANN) models was investigated in predicting Groundwater level fluctuations. The RMSE of ANN model during calibration and validation were found to be 0.2868 and 0.3648 respectively, whereas for the WANN model the respective values were 0.1946 and 0.1695. Efficiencies during calibration and validation for ANN model were 0.8862 per cent and 0.8465 per cent respectively, whereas for WANN model were found to be much higher with the respective values of 0.9436 per cent and 0.9568 per cent indicating substantial improvement in the model performance. Hence hybrid ANN model is the promising tool to predict water table fluctuation as compared to ANN model.

Forecasting Groundwater Level for Soil Landslide Based on a Dynamic Model and Landslide Evolution Pattern

The occurrence and migration of groundwater is a key natural factor that directly affects the stability of landslides, and rainfall has a large effect on the groundwater level in soil landslides. This study used the Baijiabao landslide in the Three Georges area of China as the research subject and used a combination of more than seven years of rainfall and GPS (Global Positioning System) monitoring data from 2007 to 2013. We applied the K-means clustering method to classify one cycle of the evolution stage into three classes based on the relative displacement of the main sliding surface of the landslide. To illustrate the relationship between the three landslide evolution divisions and the dynamic indicators, we identified rainfall factors that correspond to the actual change in the landslide using the minimal description length principle method. Based on the relationship between the actual deformation stage of the landslide and the rainfall factor from historical monitoring, the mean absolute error of the dynamic exponential smoothing model was 0.053, and the correlation coefficient was 0.929. The size of the smoothness index could be modified in real-time to achieve dynamic correction, which indicates that the model exhibited high reliability and confirmed the usefulness of the proposed model for forecasting groundwater level changes based on deep-seated soil landslide type.

Choosing between linear and nonlinear models and avoiding overfitting for short and long term groundwater level forecasting in a linear system

Groundwater level forecasting is a useful tool for a more efficient and sustainable groundwater resource management. Developing models that can accurately reproduce groundwater level response to meteorological conditions can lead to a better understanding of the groundwater resource availability. Here an autoregressive neural network (NNARx) approach is proposed and compared with autoregressive linear models with exogenous input (ARx) in order to forecast groundwater level in an aquifer system where a linear groundwater level response to recharge by rainfall is observed. A well known problem regarding neural networks consists in the high risk of overfitting. Here, three NNARx model were trained using different methods to avoid overfitting: Early stopping, Bayesian regularization and a combination of both. The results show that on the short term forecasting (up to 15 days) the performance of NNARx and ARx are comparable but the ARx model generalizes better, while the NNARx trained with Bayesian regularization outperforms the linear models and the other NNARx models on longer scenarios on the test set. As linear models are less time demanding and do not require high computational power, they can be considered as suitable tools for short term groundwater level forecasting in linear systems while when longer scenarios are needed neural networks can be considered more reliable, and training them with Bayesian regularization allows to minimize the risk of overfitting.

Evaluation of data-driven models (SVR and ANN) for groundwater-level prediction in confined and unconfined systems

Modeling the behavior of groundwater levels is necessary to implement sustainable groundwater resource management. Groundwater is a non-linear and complex system, which can be modeled by data-driven models. This study evaluates the performances of data-driven models, support vector machine regression (SVR) and artificial neural network (ANN), for forecasting groundwater levels of confined and unconfined systems at 1-, 2-, and 3-month ahead. This is the first time that confined and unconfined aquifers have been compared using data-driven models. In addition, to identify the optimal input combination, a hybrid gamma test (GT) and genetic algorithm (GA) was used. The coefficient of correlation (R), Mean Absolute Error (MAE), root mean squared error (RMSE), Nash–Sutcliffe efficiency (NSE), and developed discrepancy ratio (DDR) were applied to evaluate the SVR and ANN models. Results showed that the SVR and ANN models were more accurate for the unconfined system than the confined system for forecasts up to 3-month ahead. In both hydrogeological systems for 1-month ahead, the models performed better than for 2- and 3-month ahead forecasts, and the accuracy of the models decreased as the months ahead increased. The SVR model performed better than the ANN model for 1-, 2-, and 3-month ahead groundwater-level forecasting. The SVR model could be successfully used in predicting monthly groundwater in confined and unconfined systems.

Developing comparative mathematic models, BN and ANN for forecasting of groundwater levels

An accurate prediction of groundwater level is a key element in appropriate planning and managing benefits driven from aquifers in watershed. In this study, the conceptual model (MODFLOW), the Bayesian network (BN) and the artificial neural network (ANN) models are employed to forecast monthly groundwater level in Birjand Aquifer, South Khorasan, Iran. In this regard, total monthly evaporation, average temperature, aquifer recharge and discharge, and water table of previous months from 13 observation piezometers and in a 12-year period are used as input variables in order to forecast the groundwater level in the forthcoming months. The results of different models revealed that the Bayesian network models are superior to the ANN and mathematical models. The Bayesian network, ANN and mathematical models developed for the 13 piezometers have an average coefficient of determination of 0.9, 0.76 and 0.72, respectively. The results of this study demonstrate that the BNs are efficient tools for forecasting groundwater level.