Development and Implementation of Support Vector Machine Regression Surrogate Models for Predicting Groundwater Pumping-Induced Saltwater Intrusion into Coastal Aquifers

Predicting the extent of saltwater intrusion (SWI) into coastal aquifers in response to changing pumping patterns is a prerequisite of any groundwater management framework. This study investigates the feasibility of using support vector machine regression (SVMr), an innovative artificial intelligence-based machine learning algorithm for predicting salinity concentrations at selected monitoring wells in an illustrative aquifer under variable groundwater pumping conditions. For evaluation purpose, the prediction results of SVMr are compared with well-established genetic programming (GP) based surrogate models. SVMr and GP models are trained and validated using identical sets of input (pumping) and output (salinity concentration) datasets. The trained and validated models are then used to predict salinity concentrations at specified monitoring wells in response to new pumping datasets. Prediction capabilities of the two learning machines are evaluated using different proficiency measures to ensure their practicality and generalisation ability. The performance evaluation results suggest that the prediction capability of SVMr is superior to GP models. Also, a sensitivity analysis methodology is proposed for assessing the impact of pumping rates on salt concentrations at monitoring locations. This sensitivity analysis provides a subset of most influential pumping rates, which is used to construct new SVMr surrogate models with improved predictive capabilities. The improved prediction capability and the generalisation ability of the SVMr models together with the ability to improve the accuracy of prediction by refining the input set for training makes the use of proposed SVMr models more attractive. Prediction models with more accurate prediction capability makes it potentially very useful for designing large scale coastal aquifer management strategies.     

Coupled Sharp-interface and Density-dependent Model for Simultaneous Optimization of Production Well Locations and Pumping in Coastal Aquifer

The main management challenge in coastal aquifers is to prevent saltwater intrusion, ensuring ample freshwater supply. Saltwater intrusion happens due to unregulated pumping from production wells. Therefore, it is essential to have an effective management policy, which ensures the requisite amount of freshwater to be withdrawn from coastal aquifers without causing saltwater intrusion. A methodology for optimizing production well locations and maximizing pumping from production wells is presented to achieve these conflicting objectives. The location of production wells directly affects the amount of freshwater pumped out of the coastal aquifer. Simultaneous optimization of production well locations and pumping from the same is achieved by linking mathematical simulation models with the optimization algorithm. A new methodology using coupled sharp-interface and density-dependent simulation models is developed to find optimal well locations and optimize the amount of freshwater pumped from the coastal aquifer. The performance of the developed methodology is evaluated for saltwater intrusion in the coastal city of Puri, India. The performance evaluation results show the developed methodology's applicability for managing saltwater intrusion while maximizing freshwater pumping in coastal aquifers under constraints of well location.

     

Artificial Neural Network and Grey Wolf Optimizer Based Surrogate Simulation-Optimization Model for Groundwater Remediation

We herein propose a simulation-optimization model for groundwater remediation, using PAT (pump and treat), by coupling artificial neural network (ANN) with the grey wolf optimizer (GWO). The input and output datasets to train and validate the ANN model are generated by repetitively simulating the groundwater flow and solute transport processes using the analytic element method (AEM) and random walk particle tracking (RWPT). The input dataset is the different realization of the pumping strategy and output dataset are hydraulic head and contaminant concentration at predefined locations. The ANN model is used to approximate the flow and transport processes of two unconfined aquifer case studies. The performance evaluation of the ANN model showed that the value of mean squared error (MSE) is close to zero and the value of the correlation coefficient (R) is close to 0.99. These results certainly depict high accuracy of the ANN model in approximating the AEM-RWPT model. Further, the ANN model is coupled with the GWO and it is used for remediation design using PAT. A comparison of the results of the ANN-GWO model with solutions of ANN-PSO (ANN-Particle Swarm Optimization) and ANN-DE (ANN-Differential Evolution) models illustrates the better stability and convergence behaviour of the proposed methodology for groundwater remediation.

     

Regional Analysis of Flow Duration Curves through Support Vector Regression

A flow-duration curve (FDC) shows the relationship between magnitude and frequency of daily streamflows over a specific time period. Artificial intelligence methods e.g. Support Vector Machines for Regression (SVR) and Artificial Neural Network (ANN) are useful techniques in the prediction of FDCs in ungagged basins. Regional analysis of FDCs were performed through SVR, ANN and Nonlinear Regression (NLR) using streamflow with durations of 0.02, 0.10, 0.20, 0.50 and 0.90% as dependent variables and six watershed characteristics chosen as effective independent variables on 33 selected watersheds in the Namak-Lake basin located in central zone of Iran. The results shows that the most important watershed characteristics are weighted average height, area, rangeland area, drainage density, permeable formation, and average stream slope. SVR has higher accuracy with relative root mean squared error (RMSE_r) of 9.37 to 1.45 and Nash-Sutcliff criterion (NSE) of 0.54 to 0.91 than ANN with RMSE_r with 9.42 to 3.79 and NSE of 0.39 to 0.86 and NLR with RMSE_r with 18.04 to 3.38 and NSE of 0.53 to 0.79. In general, SVR is proposed to be used to estimate FDCs.

     

Artificial Neural Network Modeling for Groundwater Level Forecasting in a River Island of Eastern India

Forecasting of groundwater levels is very useful for planning integrated management of groundwater and surface water resources in a basin. In the present study, artificial neural network models have been developed for groundwater level forecasting in a river island of tropical humid region, eastern India. ANN modeling was carried out to predict groundwater levels 1 week ahead at 18 sites over the study area. The inputs to the ANN models consisted of weekly rainfall, pan evaporation, river stage, water level in the drain, pumping rate and groundwater level in the previous week, which led to 40 input nodes and 18 output nodes. Three different ANN training algorithms, viz., gradient descent with momentum and adaptive learning rate backpropagation (GDX) algorithm, Levenberg–Marquardt (LM) algorithm and Bayesian regularization (BR) algorithm were employed and their performance was evaluated. As the neural network became very large with 40 input nodes and 18 output nodes, the LM and BR algorithms took too much time to complete a single iteration. Consequently, the study area was divided into three clusters and the performance evaluation of the three ANN training algorithms was done separately for all the clusters. The performance of all the three ANN training algorithms in predicting groundwater levels over the study area was found to be almost equally good. However, the performance of the BR algorithm was found slightly superior to that of the GDX and LM algorithms. The ANN model trained with BR algorithm was further used for predicting groundwater levels 2, 3 and 4 weeks ahead in the tubewells of one cluster using the same inputs. It was found that though the accuracy of predicted groundwater levels generally decreases with an increase in the lead time, the predicted groundwater levels are reasonable for the larger lead times as well.     

Planning Groundwater Development in Coastal Deltas with Paleo Channels

In this study, a management model is presented for planning groundwater development in costal deltas with paleo channels. It is demonstrated that paleo channels are the best locations for locating the wells for large-scale pumping. Groundwater flow in these aquifers is simulated using a three-dimensional (3-D) density-dependent flow and transport model SEAWAT, which is suitable for a coastal and deltaic environment. A simulation-optimization model is used to determine the optimal locations and pumpages for groundwater development for a group of production wells, while limiting the salinity below desired levels. The mixed integer problem is solved using the Simulated Annealing algorithm and the SEAWAT simulation model. A trained Artificial Neural Network (ANN) is used as the virtual SEAWAT model to perform the simulations, in order to reduce the computational burden for application of the model on desktop computers. The applicability of the model is demonstrated on a hypothetical, but near-real, delta system.     

Comparative Study of Hybrid-Wavelet Artificial Intelligence Models for Monthly Groundwater Depth Forecasting in Extreme Arid Regions,Northwest China

Prediction of groundwater depth (GWD) is a critical task in water resources management. In this study, the practicability of predicting GWD for lead times of 1, 2 and 3 months for 3 observation wells in the Ejina Basin using the wavelet-artificial neural network (WA-ANN) and wavelet-support vector regression (WA-SVR) is demonstrated. Discrete wavelet transform was applied to decompose groundwater depth and meteorological inputs into approximations and detail with predictive features embedded in high frequency and low frequency. WA-ANN and WA-SVR relative of ANN and SVR were evaluated with coefficient of correlation (R), Nash-Sutcliffe efficiency (NS), mean absolute error (MAE), and root mean squared error (RMSE). Results showed that WA-ANN and WA-SVR have better performance than ANN and SVR models. WA-SVR yielded better results than WA-ANN model for 1, 2 and 3-month lead times. The study indicates that WA-SVR could be applied for groundwater forecasting under ecological water conveyance conditions.     

Application of Artificial Neural Networks and Particle Swarm Optimization for the Management of Groundwater Resources

Ground management problems are typically solved by the simulation-optimization approach where complex numerical models are used to simulate the groundwater flow and/or contamination transport. These numerical models take a lot of time to solve the management problems and hence become computationally expensive. In this study, Artificial Neural Network (ANN) and Particle Swarm Optimization (PSO) models were developed and coupled for the management of groundwater of Dore river basin in France. The Analytic Element Method (AEM) based flow model was developed and used to generate the dataset for the training and testing of the ANN model. This developed ANN-PSO model was applied to minimize the pumping cost of the wells, including cost of the pipe line. The discharge and location of the pumping wells were taken as the decision variable and the ANN-PSO model was applied to find out the optimal location of the wells. The results of the ANN-PSO model are found similar to the results obtained by AEM-PSO model. The results show that the ANN model can reduce the computational burden significantly as it is able to analyze different scenarios, and the ANN-PSO model is capable of identifying the optimal location of wells efficiently.     

Application of Genetic Algorithms and Artificial Neural Networks in Conjunctive Use of Surface and Groundwater Resources

Abstract

In this paper, a methodology for conjunctive use of surface and groundwater resources is developed using the combination of the Genetic Algorithms (GAs) and the Artificial Neural Networks (ANN). Water supply to agricultural demands, reduction of pumping costs and control of groundwater table fluctuations are considered in the objective function of the model. In the proposed model, the results of MODFLOW groundwater simulation model are used to train an ANN. The ANN as groundwater response functions is then linked to the GA based optimization model to develop the monthly conjunctive use operating policies. The model is applied to the surface and groundwater allocation for irrigation purposes in the southern part of Tehran. A new ANN is also trained and checked for developing the real-time conjunctive use operating rules.

The results show the significance of an integrated approach to surface and groundwater allocation in the study area. A simulation of the optimal policies shows that the cumulative groundwater table variation can be reduced to less than 4 meters from the current devastating condition. The results also show that the proposed model can effectively reduce the run time of the conjunctive use models through the composition of a GA-based optimization and a ANN-based simulation model.     

Artificial neural network modeling for identification of unknown pollution sources in groundwater with partially missing concentration observation data

Groundwater pollution sources are characterized by spatially and temporally varying source locations, injection rates, and duration of activity. Concentration measurement data at specified observation locations are generally utilized to identify these sources characteristics. Identification of unknown groundwater pollution sources in terms of these source characteristics becomes more difficult in the absence of complete breakthrough curves of concentration history at all the time steps. If concentration observations are missing over a length of time after an unknown source has become active, it is even more difficult to correctly identify the unknown sources. An artificial neural network (ANN) based methodology is developed to identify these source characteristics for such a missing data scenario, when concentration measurement data over an initial length of time is not available. The source characteristics and the corresponding concentration measurements at time steps for which it is not missing, constitute a pattern for training the ANN. A groundwater flow and transport numerical simulation model is utilized to generate the necessary patterns for training the ANN. Performance evaluation results show that the back-propagation based ANN model is essentially capable of extracting hidden relationship between patterns of available concentration measurement values, and the corresponding sources characteristics, resulting in identification of unknown groundwater pollution sources. The performance of the methodology is also evaluated for different levels of noise (or measurement errors) in concentration measurement data at available time steps.     

Assessment of Water Resources and Salinity Intrusion in the Mekong Delta

Abstract

A methodology is presented for the assessment of water resources and salinity intrusion in the Mekong Delta in Vietnam. The flow and water salinity at different locations in the river network have been dynamically assessed by the developed three-step approach in which the hydrodynamic, advection-dispersion models, harmonic analysis, and regression techniques have been employed in the development of the relationships between the boundary conditions (upstream inflows and tides at river mouths) and the harmonic constituents of flow and water salinity. These relationships were subsequently used to determine the parameter values of harmonic constituents of flow and water salinity at specific locations under various hydrologic conditions and water allocation alternatives as needed for water management purpose. With known harmonic parameters, the flow and water salinity at the locations can be predicted by a harmonic analysis method and the river water available for agricultural use can then be determined. The advantage of this methodology is that the river water can be dynamically assessed without performing the hydrodynamic and advection-dispersion simulations in the water resources management process. Due to the limitation of data availability, a preliminary assessment of groundwater resources has been included. The results indicate that the use of groundwater as supplemental resources for agricultural production is possible     

Dynamic Optimal Monitoring Network Design for Transient Transport of Pollutants in Groundwater Aquifers

Optimal groundwater pollution monitoring network design models are developed to prescribe optimal and efficient sampling locations for detecting pollution in groundwater aquifers. The developed methodology incorporates a two dimensional flow and transport simulation model to simulate the pollutant concentrations in the study area. Different realizations of the pollutant plume are randomly generated by incorporating the uncertainty in both source and aquifer parameters. These concentration realizations are incorporated in the optimal monitoring network design models. Two different objectives are considered separately. The first objective function minimizes the summation of unmonitored concentrations at different potential monitoring locations. This objective function in effect minimizes the probability of not monitoring the pollutant concentrations at those locations where the probable concentration value is large. Although this probability is not explicitly incorporated in the model, a surrogate form of this objective is included as the objective function. The second objective function considered is the minimization of estimation variances of pollutant concentrations at various unmonitored locations. This objective results in a design that chooses optimal monitoring locations where the uncertainties in simulated concentrations are large. The developed optimization models are solved using Genetic Algorithm. The variances of estimated concentrations at potential monitoring locations are computed using the geostatistical tool, kriging. The designed monitoring network is dynamic in nature, as it provides time varying network designs for different management periods, to account for the transient pollutant plumes. Such a design can eliminate temporal redundancy and is therefore, economically more efficient. The optimal design incorporates budgetary constraints in the form of limits on the number of monitoring wells installed in any particular management period. The solution results are evaluated for an illustrative study area comprising of a hypothetical aquifer. The performance evaluation results establish the potential applicability of the proposed methodology for optimal design of the dynamic monitoring network for detection and monitoring of pollutant plumes in contaminated aquifers.     

堤坝滑坡灾害的探地雷达应用研究

通过太浦河泵站滑坡工程实例,尝试应用探地雷达进行滑动面的探测,探讨了探地雷达辨别滑坡区、滑动面、有机质夹层的探测机理,分析推断了软土层分布,有机质夹层位置,滑坡区域,滑动面的起点、走向和出溢处.探测结果表明,在一定条件下结合地质勘察资料,利用探地雷达对堤坝滑动面进行探测是有效的.     

Artificial Neural Network (ANN) Based Modeling for Karstic Groundwater Level Simulation

A relatively new method of addressing different hydrological problems is the use of artificial neural networks (ANN). In groundwater management ANNs are usually used to predict the hydraulic head at a well location. ANNs can prove to be very useful because, unlike numerical groundwater models, they are very easy to implement in karstic regions without the need of explicit knowledge of the exact flow conduit geometry and they avoid the creation of extremely complex models in the rare cases when all the necessary information is available. With hydrological parameters like rainfall and temperature, as well as with hydrogeological parameters like pumping rates from nearby wells as input, the ANN applies a black box approach and yields the simulated hydraulic head. During the calibration process the network is trained using a set of available field data and its performance is evaluated with a different set. Available measured data from Edward??s aquifer in Texas, USA are used in this work to train and evaluate the proposed ANN. The Edwards Aquifer is a unique groundwater system and one of the most prolific artesian aquifers in the world. The present work focuses on simulation of hydraulic head change at an observation well in the area. The adopted ANN is a classic fully connected multilayer perceptron, with two hidden layers. All input parameters are directly or indirectly connected to the aquatic equilibrium and the ANN is treated as a sophisticated analogue to empirical models of the past. A correlation analysis of the measured data is used to determine the time lag between the current day and the day used for input of the measured rainfall levels. After the calibration process the testing data were used in order to check the ability of the ANN to interpolate or extrapolate in other regions, not used in the training procedure. The results show that there is a need for exact knowledge of pumping from each well in karstic aquifers as it is difficult to simulate the sudden drops and rises, which in this case can be more than 6 ft (approx. 2 m). That aside, the ANN is still a useful way to simulate karstic aquifers that are difficult to be simulated by numerical groundwater models.     

Minimization of Transient Groundwater Pumping Cost - Analytical and Practical Solutions

In this paper, we study pumping cost minimization for any number and layout of wells under transient groundwater flow conditions in infinite confined aquifers and semi-infinite ones, to which the method of images applies. Moreover, we take into account additional steady-state flow, which is independent of the well system and results in non-horizontal initial hydraulic head level distribution. We prove analytically that, at any time, the instant pumping cost is minimum, when the following condition holds: the observed at that instant differences between hydraulic head values at the locations of the wells are equal to the half of the initial ones, which are due to the additional steady-state flow. Based on this proof, an analytical calculation procedure of the time-dependent optimal distribution of the required total flow rate to the individual wells is also presented. Moreover, as well flow rates usually remain constant over the pumping period, an approximate calculation of the optimal constant well flow rate distribution is outlined, based again on an analytical procedure.     

Artificial Neural Network and Support Vector Machine Models for Inflow Prediction of Dam Reservoir (Case Study: Zayandehroud Dam Reservoir)

Inflow prediction of reservoirs is of considerable importance due to its application in water resources management related to downstream water release planning and flood protection. Therefore, in this research, different new input patterns for predicting inflow to Zayandehroud dam reservoir is proposed employing artificial neural network (ANN) and support vector machine (SVM) models. Nine different models with different patterns of input data such as inflow to the dam reservoir considering time duration lags, time index, and monthly rainfall of Ghaleh-Shahrokh station have been proposed to predict the inflow to the dam reservoir. Comparison of the results indicates that the ninth proposed model has the least error for inflow prediction in which the results of SVM model outperform those of ANN model. That is, the least error has been obtained using the ninth SVM (ANN) model with correlation coefficient (R) values of 0.8962 (0.89296), 0.9303 (0.92983) and 0.9622 (0.95333) and root mean squared error (RMSE) values of 47.9346 (48.5441), 42.69093 (43.748) and 23.56193 (28.5125) for training, validation and test data, respectively.

     

Water table dynamics of Dhaka city and its long-term trend analysis using the “MAKESENS” model

The water table (WT) data of Dhaka city for 17 years (1988–2004) was analysed for assessing WT fluctuation and predicting its trend using a computer model, “MAKESENS”. The analysis revealed that the WT declined drastically at different locations of the city. Moreover, the model also predicted that WT would further decline 9–25 m by the year 2015 and 18–40 m by the year 2025, rendering most wells inoperative, if the pumping rate was not reduced.     

A sensitivity analysis and parametric study for the evaluation of the optimal management of a contaminated aquifer

A typical groundwater remedation problem is studied by using a combined simulation-optimization model. The management procedure employs groundwater flow and contaminant transport simulation models in conjunction with linear and quadratic programming techniques. The methodology is applied to the hydrodynamic control of a contaminant plume that has to be stabilized and removed by a system of pumping wells. The paper focuses mainly upon a sensitivity analysis to the aquifer transmissivity. The effect of changes in the transmissivities of a zoned aquifer upon the optimal solutions of the management problem is examined by considering the optimal pumping rates, the time to remediation and the pumped groundwater volume as the key output variables of the remediation strategies. In addition, the influence of the dispersivities and the imposed hydraulic gradient upon the same output variables is critically evaluated. The results of the study illustrate the need for uncertainty reduction in the knowledge of the hydrogeologic parameters.     

Comparative Evaluation of Genetic Programming and Neural Network as Potential Surrogate Models for Coastal Aquifer Management

Determining the optimal rates of groundwater extraction for the sustainable use of coastal aquifers is a complex water resources management problem. It necessitates the application of a 3D simulation model for coupled flow and transport simulation together with an optimization algorithm in a linked simulation-optimization framework. The use of numerical models for aquifer simulation within optimization models is constrained by the huge computational burden involved. Approximation surrogates are widely used to replace the numerical simulation model, the widely used surrogate model being Artificial Neural Networks (ANN). This study evaluates genetic programming (GP) as a potential surrogate modeling tool and compares the advantages and disadvantages with the neural network based surrogate modeling approach. Two linked simulation optimization models based on ANN and GP surrogate models are developed to determine the optimal groundwater extraction rates for an illustrative coastal aquifer. The surrogate models are linked to a genetic algorithm for optimization. The optimal solutions obtained using the two approaches are compared and the advantages of GP over the ANN surrogates evaluated.     

Predictive Temporal Data-Mining Approach for Evolving Knowledge Based Reservoir Operation Rules

The persistent problem in reservoir operation is that the derived optimal releases fail to incorporate the decision maker or reservoir operators’ knowledge into reservoir operation models. The reservoir operators’ knowledge is specific to that particular reservoir and incorporating such an experienced knowledge will help to derive field reality based operation rules. The available historical reservoir operation databases are the representative samples of reservoir operators’ knowledge or experience. Thus, an attempt has been made that deals with the development of a methodological framework to recover or explore the historical reservoir operation database to derive the reservoir operators’ knowledge as operational rules. The developed methodological framework utilizes the strength and capability of recently developed predictive datamining algorithms to recover the knowledge from large historical database. Predictive data-mining algorithms such as a) classifier: Artificial Neural Network (ANN), and b) regression: Support Vector Regression (SVR) have been used for single reservoir operation data-mining (SROD) modelling framework to explore the temporal dependence between different variables of reservoir operation. The rules of operation or knowledge learned from the training database have been used as guiding rules for predicting the future reservoir operators’ decision on operating the reservoir for the given condition on the inflow, initial storage, and demand requirements. The developed SROD model was found to be efficient in exploring the hidden relationships that exist in a single reservoir system.