Ground Water Management – Role of Ground Water Modelling

INTRODUCTION

The ubiquitous availability of ground water, coupled with technological advancements in its extraction and its deemed ownership as easement to land, has led to a quantum leap in development of this resource during the last five decades. The unscientific development of ground water contributes to its increasing scarcity, reflected in steep decline of water level and under certain situations, sharp deterioration in quality of water. A continuous declining trend in ground water levels has been observed in localized pockets through out thecountry. As per the latest ground water estimation of dynamic resources , out of 5723 assessment units nearly 1615 units recorded significant decline in ground water level in either pre or post monsoon period on long term basis and have been categorized as overexploited / critical or semi critical. The phenomenal increase in the overexploited areas are reported globally and India is not an exception. Surprisingly, water is the largestexported commodity in the world, in terms of virtual water. Ground water management on scientific lines is the key for sustainability of this vital resource.

GROUND WATER MANAGEMENT : An Urgent Need

In view of the increasing thrust on groundwater resources and the present scenario of availability vis-a vis demand there is a need to re-orient our approach to ground water management. It is very important that the management of ground water in the country is taken up with a proper planning keeping in mind various social obligations and its requirement for various purposes. Further, emphasis should be given for purpose drivenstudies involving research and development for scientific development of this precious natural resource.

Aquifer management which deals with a complex interaction between human society and physical environment, presents an extremely difficult problem of policy design. Aquifers are exploited by human decisions and over-exploitation can’t be always defined in technical terms, but as a failure to design and implement adequate institutional arrangements to manage people who exploit the groundwater resources. ‘Common pool ‘ resources such as ground water have been typically utilized in an ‘ open access ‘ framework , within which, resource ownership is according to a rule of capture. When no one owns the resources, users have no incentive to conserve for the future, and self interest of individual users leads them to overexploitation.

The various management options available for ameliorating or solving problems related to ground water quantity and quality can be broadly grouped under two major categories. The first category relates to supply side management and are referred as “ Structural techniques” which usually involve some form of water supply augmentation and artificial recharge. For an effective supply side management it is essential to have full knowledge of hydrogeological controls which govern the yield and behaviour of groundwater levels under abstraction stress, the interaction of surface and groundwater in respect of river base flow and changes in flow and recharge rates due to their exploitation. The effects of groundwater development can be short term and reversible or long term and quasi-reversible which require a strong monitoring mechanism for scientificmanagement. The other category encompasses Demand side management which is user targeted and are referred as “ Non – structural techniques”. In demand side management the socio-economic dimension plays an important role involving managing the users of water and land. Mere regulatory interventions like waterrights and permits and economic tools of water pricing etc. cannot be successful unless the different user groups are fully involved to get their cooperation and participation. For effective management of groundwater resources there is a need to create awareness amongst the different water user groups and workout area specific plans for sustainable development. Thus, groundwater management not only requires proper assessment of available resource and understanding of interconnection between surface and groundwatersystem but also actions required for proper resource allocation and prevention of the likely adverse effects of uncontrolled development of groundwater resources. In a nut shell the first category of management options targets policies for ‘managing the water’ and the second category calls for ‘coordinating the people’.

GROUNDWATER MODELS – AS MANAGEMENT TOOL

With the invent of high speed computers and advancement in mathematical techniques, groundwater models have become the most sophisticated tool for the decision makers in planning and management of ground water . Whether, the management options are structural or non structural, ground water models have become inevitable. At the same time , while using the models as a management aid, we must keep in mind that modelsare not an end in itself, they are just one piece of a larger puzzle when we are looking at real-world problems from a systems approach angle. Hence, in true sense Systems analysis is a philosophy of problem solving, most often used with large, complex problems where decision makers collects information and uses everything available to decide what to do. Under the systems analytical approach the groundwater models can be formulated and solved in two frame work, which are well tested in solving real world problems.

A. SIMULATION
B. OPTIMIZATIONM

A. GROUNDWATER SIMULATION MODELS

Groundwater simulation models provide a platform to study the problems in broader perspective and resolve solutions for the optimal benefit taking into considerations the simplest and complex aspects along with economic, social and environmental aspects. A model is any device that represents an approximation of a field situation. Physical models such as laboratory sand tanks and R-C analog models simulate groundwater flowdirectly. A mathematical model simulates groundwater flow indirectly by means of governing equations. In the present scenario we shall be discussing mainly the mathematical models which are conceptual descriptions or approximations that describe physical systems using mathematical equations—they are not exact descriptions of physical systems or processes. The applicability or usefulness of a model depends on how closely the mathematical equations approximate the physical system being modelled. In order to evaluate theapplicability or usefulness of a model, it is necessary to have a thorough understanding of the physical system and the assumptions embedded in the derivation of the mathematical equations. These equations are based on certain simplifying assumptions which typically involve: the direction of flow, geometry of the aquifer, the heterogeneity or anisotropy of sediments or bedrock within the aquifer, the contaminant transport mechanisms, and chemical reactions. It is because of these assumptions, and the many uncertainties in thevalues of data required by the model, that a model must be viewed as an approximation and not an exact duplication of field conditions. The equations that describe the groundwater flow and fate of transport processes may be solved using different techniques.

Before one decides to construct a ground water model and use it as a tool for management it is always better to give an eye on the saying of Sherlock Holmes, the great philosopher that “It is a capital mistake to theorize before one has data. Insensibly one begins to twist facts to suit theories, instead of theories to suit facts’.

Groundwater simulation models can be of different type based on the criteria adopted for classification. Based on the objectives and application of model it can of three types. Most common groundwater modeling efforts are aimed at predicting the consequences of a proposed action and are called Predictive models, which essentially requires calibration. There are, however, two other important types of applications. Models can be used in an interpretative sense to gain insight into the controlling parameters and used as a framework for studying system dynamics and or organizing field data called Interpretative Models, does not necessarily involve calibration. The RASA ( Regional aquifer System Analysis) programme of USGS is the ideal example of such models. The third type is the Generic Models , used to study and analyze hypothetical hydrogeological systems, may help in formulating regional regulatory guidelines for a specific region, does not necessarilyrequire calibration.

As we have seen that groundwater models are basically an approximation of field conditions in the form of mathematical equations, hence based upon the approach for solving these equations, it can be classified into three broad categories. The first category includes the Deterministic models (e.g. analytical models and numerical models) which are strictly based on physical laws controlling the flow or transport processes such as conservation of mass, Darcy’s law, Fick’s law etc. Deterministic models describe cause and effect relationshipand generally requires solution of partial differential equations , no chance factor is involved in this. Depending upon the treatment of various aquifer parameters it could be a Lumped parameter model , which precludes the heterogeneous hydraulic properties in the model or a Distributed parameter model which allows the representation of variable system parameters.

Under the Deterministic models , the simplest is Analytical models, attempts exact solutions to equations which describe very simple flow or transport conditions and others (such as numerical models) may be approximations of equations which describe very complex conditions. Because of the simplifications inherent with analytical models, it is not possible to account for field conditions that change with time or space. This includes variations in groundwater flow rate or direction, variations in hydraulic or chemical reactionproperties, changing hydraulic stresses, or complex hydrogeologic or chemical boundary conditions.

The analytical models are best used for:-

- Initial assessments where a high degree of accuracy is not needed,
- Prior to beginning field activities to aid in designing data collection,
- To check results of numerical model simulations, or
- Where field conditions support the simplifying assumptions embedded in analytical models.

The Theiss , Theims , Jacobs , Boultons method for analysis of pumping test data are good examples of analytical solutions of the radial flow equations. Similar analytical solutions are also available for different aquifers in Cartesian coordinates which provides closed form solutions between excitation and response. Under simplifying assumptions these analytical solutions are very useful tool to determine permissible level of pumpage to maintain certain minimum water level / piezometric heads and assist in decision making or to test various other management options .

When the system is complex, anisotropy is to be considered, the only choice left is to go for Numerical Models. Numerical models are capable of solving the more complex equations that describe groundwater flow and solute transport. These equations generally describe multi-dimensional groundwater flow, solute transport, and chemical reactions. In Numerical models we don’t get closed form solutions, here the response is estimated at pre-selected space and time, encompasses large computational burden and numerical errors areinevitable. The commonly used numerical approaches used in practice today for solving ground water flow equations are Finite Difference Method ( FDM) and Finite Element Method ( FEM) , other than that some latest models also uses Method of Characteristics ( MOC) as well. Each model, whether it is a simple analytical model or a complex numerical model, has applicability and usefulness in hydrogeological and remedial investigations, in spite of the simplifications inherent in the model equations. However, the selection and proper use of a model must be based on a thorough understanding of the importance of relevant flow orsolute transport processes at a site, this requires proper site characterization. Proper site characterization involves the collection of site-specific data that accurately describe the movement of groundwater and disposition of solutes at the site. Without proper site characterization, it is not possible to determine whether the model equations are appropriate or to develop a reliably calibrated model. Groundwater flow and transport models are used to predict the migration pathway and concentrations of contaminants in groundwater. The accuracy of model predictions depends upon the degree of successful calibration andverification of the model simulations. Errors in the model used for predictive simulations, even though small, can result in gross errors in solutions projected forward in time. Monitoring of hydraulic heads and groundwater chemistry (performance monitoring) will be required to assess the accuracy of predictive simulations. Numerical modeling requires a complete protocol to be followed starting from establishing the purpose or objectives of modeling to conceptualization to model design to calibration, validation, predictions ,sensitivity analysis and finally the post audit. One of the important aspect is the selection of proper code (modeling software) Depending upon the aquifer to be modeled and objective of the study the model can be two dimensional , fully three dimensional or quasi three dimensional. Availability of data pertaining to geometry of the aquifer, hydraulic parameters as well as prevailing boundary conditions need to be ascertained before a model is conceptualized.

Recent studies have indicated that the variables which are commonly being dealt in ground water hydrology such as hydraulic heads , transmissivity and various other aquifer parameters related in space do not behave truly in deterministic way and many a times it is very difficult to model these parameters strictly within the deterministic framework. There is a gradual shift from deterministic approach to stochastic approach which takes into account the chance factor as well as time dependency and work under probabilistic framework.Statistical or stochastic models are more commonly used in time series analysis in which the deterministic and stochastic components are treated separately. Analysis of deterministic component of time series includes Trend , periodicities, randomness etc. Common questions related to time series are whether the time series is random or not, whether the time series is having a trend, if at all there is a trend whether it is significant or not. To answer some of these questions there are well established techniques which can be applied in theground water as well. Some of the commonly applied techniques are listed below.

- Turning point test for Randomness
- Kedall’s Rank Correlation test for trend
- Linear Regression test for quantification of trend as well as for significance test.
- Harmonic analysis for periodicities.

In addition, there are established techniques used for analysis of the stochastic components of time series, which includes

- Correlogram analysis
- Spectral analysis
- Monte Carlo Simulation
- Auto Regressive ( AR ) Models, ( Markov Chain )
- Moving Average ( MA) Models- Autoregressive Moving Average ( ARMA) Models
- Autoregressive Integrated Moving Average ( ARIMA) Models, etc.

The high frequency data available from the digital water level recorders may be put for time series analysis with an objective to simulate the series and as per the requirement alternate sequence may be generated as well as it can be used for forecasting purposes. It can also be used for data gap filling and record extrapolation.

Hence, we can use ground water simulation models to study the consequences of a proposed action , for predictions / and or management of ground water , we can even have the quantitative aspects of groundwater flow, changes in hydraulic head with respect to time and space, river / canal interactions and impact of various stresses on the ground water regime.

B. OPTIMIZATION MODELS

Though simulation models are very robust mathematical tool , it has certain limitations in the sense that it can basically checks for feasibility of a management strategy , it can not provide the optimal management strategy. However , an optimization model identifies an optimal management strategy from a set of feasible alternative strategies. In order to ensure that the optimal strategy is physically / hydraulically feasible , a simulation model is necessary to simulate the system behaviour. Because of mathematical complexities of involving the nonlinearity the present approach in ground water domain is to have separate optimization model, and testing the optimal strategy through simulation models. The same has been adopted in all the Conjunctive use projects completed by CGWB , in which the optimal conjunctive use plan was derived from Linear Programming ( an optimization model) by way of optimizing the benefits or minimizing the cost. In this approach generally we get two to three cost effective scenarios , which are then tested through simulation model to arrive at the decision which scenario is most viable or optimal. Hence only simulation model do not provide sufficient alternatives for the decision makers to arrive at best solution.

In the water resources sector various optimization models are used very frequently to solve the water allocation problems. Some of the important optimization techniques used are as below

- Linear Programming
- Dynamic programming
- Integer Programming
- Non Linear programming etc.

The above models are commonly used in surface water domain, the use of optimization models alone in ground water domain is limited. In coming years, we have marched a step ahead and the combined use of simulation and optimization techniques have been demonstrated to be most powerful and useful method in determining and planning management strategies for optimal development and operation of ground watersystem. Simulation models can answer the question “ What if “ and it can simulate the responses of the system to a specified management strategy. However, an optimization model identifies an optimal management strategy from a set of feasible alternative strategies. In order to ensure that the optimal strategy is physically acceptable, a simulation model is necessary to simulate the system behaviour. The best option is to go for a combined Simulation / Optimization approach, where simulation models can be combined with the management model either by using the governing differential equations as binding constraints in the optimization model or by using a response matrix ( Gorelick, 1983) or an external simulation model. In the following paragraph an attempt has been made to explain various techniques.

The goal of a formal mathematical optimization-based groundwater management model is to achieve a specified objective in the best possible manner within the various limiting restrictions. The limiting restrictions are derived from managerial considerations and physical behaviour of the system. In order to ensure that the final solution does not violate the physical laws of the system, a model simulating the behaviour and responseof the system is incorporated within the management model. Once the optimization model is formulated, a suitable mathematical programming technique is applied to obtain the solution. Some of the important optimization techniques for the solution of groundwater quantity and quality management problems are discussed in the following paragraphs.

Embedding technique and response matrix approach are the two methods generally used to incorporate the simulation model within a management model (Gorelick 1983). In embedding technique, the finite difference or finite element form of the governing groundwater flow and solute transport equations are directly incorporated as part of the constraint set in a formal mathematical programming-based management model. Other physical and managerial constraints on heads, gradients, velocities or pumping/injection rates can be incorporated easily. Some of the unknown groundwater variables, i.e. hydraulic heads, source/sink rates,existing solute concentrations, solute concentrations of the source/sink at each node may become decision variables in the optimization problem. When large numbers of pumping cells are used and steady state management policies are desired, the embedding technique requires less computer memory and processing time than the response matrix approach (Peralta & Datta 1990). This conclusion may not be valid for transient cases. For nonlinear systems, the response matrix approach is not applicable and use of embedding techniquebecomes necessary. However, the time step used in the embedding approach for transient problems may require a larger number of variables and constraints for accuracy of the solution. In the response matrix approach, the time step used may not be an important consideration except from the management view point. In highly nonlinear problems such as those involving density dependent transport models, where the response matrix approach is not strictly applicable, a management model even for a small study area may become dimensionally large. Solving such management models using nonlinear optimization techniques, therefore, becomes difficult. Computational difficulties in using standard optimization packages for large scale problems are reported by Elango & Rouve (1980), Gorelick (1983), Tung & Koltermann (1985), Yazicigil & Rasheeduddin (1987).

The response matrix approach (Gorelick 1983) uses an external groundwater simulation model to develop unit responses. The unit response describes the influence of a unit change in an independent decision variable/design variable (such as sink/source rates) at pre-selected well locations, upon a variety of dependent variables/other design variables (such as hydraulic head, velocity, solute concentration) at specified observation points. The assembled unit responses are used to construct the response matrix, which is includedin the management model. In order to generate the unit response matrix, a simulation model is solved several times each with a unit stress (pumping/recharge) or concentration loads at a single node. The response matrix approach works on the principles of superposition. It is applicable when the system is linear or approximately linear and the boundary conditions are homogeneous. For highly nonlinear systems, the performance of response matrix Optimization techniques in groundwater management approach is reported to be unsatisfactory (Rosenwald & Green 1974). Any change in boundary condition, location of the source/sink, and observation wells requires several simulations to generate the responses and also requires recalculation of the response matrix. Many researchers have reported the use of embedding technique and/or response matrix approach in conjunction with mathematical programming methods to find a solution of the groundwater management model. Application of some of the optimization models alone as well as coupled Simulation /optimization models in the ground water domain has been discussed in following paragraphs.

Application of linear programming

Linear programming (LP) techniques can be utilised for solving groundwater quantity and quality management problems when the imposed physical and managerial constraints and the objective function are linear. The capability of LP techniques to solve large-scale problems and to guarantee global optimal solutions has attracted the widespread attention of many researchers in the groundwater management field. Some of the aquifer management problems formulated and solved by using LP technique are discussed here. Alley et al(1976) applied LP formulation to two-dimensional transient situations in a confined heterogeneous anisotropic aquifer. In the governing differential equation, the source/sink term was expressed as the sum of specified net source/sink and unknown source/sink terms. Their objective was to maximise the hydraulic heads for a portionof the management period such that a fixed total pumping was maintained with a certain minimum pumping from a specified location, while maintaining a certain minimum head during the specified period. For the remaining portion of the management period the objective was to maximise the pumping subject to maintaining of some lower limits for the heads at the interior nodes, with restrictions on pumping at some fixed nodes. This as a different objective and thus could not be formulated as a single problem with theprevious one, though these two objectives were applicable to constituent parts in the total management period. The management period was divided into small periods of intervals and solved for each of these small periods separately by using LP. The solution from the previous LP formulation was used as initial condition for the next period. Further they extended the methodology for steady state cases to study the feasibility of disposing of waste water by injection into an aquifer system. The objective was to minimise total pumpingfrom two lines of wells subject to: (i) a reversal of hydraulic gradient towards the pumping well, (ii) maintenance of monotonicity of head values to prevent the recharged waste product from reaching a particular area, and (iii) to meet certain water demands for irrigation.

For optimal management of a coastal aquifer in southern Turkey, Hallaji & Yazicigil (1996) used LP technique. They proposed six LP models for steady state and transient state, and one quadratic optimization model for steady state management of the aquifer system. The general constraints were (i) water demand constraints, (ii) drawdown limitations, (iii) maximum pumping rate constraints, and (iv) minimum pumping rate constraints.The response matrix approach was used to obtain the drawdown limitations. However, the hydraulics of saltwater intrusion was not considered in the response matrix. The objectives considered for the steady state management were (i) maximisation of steady state water withdrawals from the existing wells, (ii) maximisation of withdrawals without any maximum limit on withdrawals from the wells, (iii) minimisation of the sum of drawdowns at pumping wells and saltwater-control nodes, and (iv) minimisation of the sum of the drawdownsat the saltwater-control nodes along the coast. The objectives for transient state management were (i) maximisation of the sum of monthly withdrawals, and (ii) minimisation of the sum of drawdowns at the pumping wells and saltwater-control nodes for all pumping periods.

Application of mixed-integer programming

Mixed-integer programming (MIP) can be used to solve optimization problems with linear objective function and linear constraints in which some of the variables can take only integer values. These types of requirements arise when dealing with groundwater management problems in which decision variables seek answers of the type yes or no, or the decision variables decide the number of installations, locations etc. In special cases allthe variable may take only integer values, where the problem reduces to one of integer programming. Some example applications of MIP in groundwater management are presented in Rosenwald & Green (1974), Willis (1976). MIP has also been applied successfully in developing optimal monitoring network for groundwater quality by Meyer & Brill (1988), Datta & Dhiman (1995), Loaiciga et al (1992). Some of these applications ofMIP for general management of groundwater quality and quantity are presented here. Rosenwald & Green (1974) developed a methodology to find the best locations for a specified number of wells. They used the branch and bound method to solve the mixed-integer programming problem and utilised a transient responsematrix. Willis (1976) presented a planning model for the optimal conjunctive use of groundwater and surface water resources. As the pumping, recharges and boundary conditions were known a priori, the flow equations did not need to appear in the constraints. However, the flow equations needed to be solved externally to supply the velocities as fixed coefficients in the transport equation. The steady state solute transport simulation model was first formulated as a finite difference coefficient matrix. The inverse of this matrix wasthen computed and relevant portions were then included in the management model as constraints. The management model optimised the assimilative waste capacity of the aquifer in waste water treatment. The sum of annual cost of removal and the cost of incorporation of dilution water for all constituents were minimised. The conjunctive use model minimised the costs associated with (i) surface waste water treatment, (ii) dilution water, and (iii) waste water treatment plant. The model considered several unit processes for the waste treatment plant that involved primary, secondary, and various forms of advanced waste treatment. Thesolution determined the optimal unit treatment process and the most cost effective volume of imported dilution water. The resultant mixed integer programming problem was solved readily by decomposing the overall problem into individual sub-problems involving decisions on the amount of dilution water and each unit process combination. Each sub-problem minimised a concave objective function subject to a convex linear constraint set.

Recently the MIP has been successfully used in the joint study of CGWB and NIH to develop an operational model based on Simulation – Optimization model for Palla well field in the Yamuna flood plain area of , Delhi.

Application of nonlinear programming

Many groundwater planning and management models involve nonlinearities in the objective function and constraints. These nonlinearities may arise due to various causes such as (i) nonlinear cost functions, (ii) nonlinear equations governing the flow particularly for unconfined aquifers, (iii) nonlinearities in the governing equations for solute transport in groundwater, (iv) other types of nonlinear physical and managerial objective functions and constraints. These nonlinear management problems can be solved using nonlinear programming (NLP) algorithms.

Maddock(1972b) used quadratic programming with nonlinear objective function and linear constraints for managing an unconfined aquifer. He developed a nonlinear technological function for the unconfined aquifer which was used in the management model. He used mixed-integer quadratic programming to minimise the pumping costs plus fixed costs for well and pipeline construction. The quadratic portion of the objective function was made separable by a suitable transformation that facilitates the solution by a combination ofmixed-integer and separable programming. Further, postoptimality sensitivity and error analysis was performed to evaluate the effects of uncertainties in economic and hydrologic factors, on the planning activities.

Gorelick et al (1984) presented a general modelling approach to determine the optimal design of reclamation schemes for contaminated groundwater systems. The planning model combined a nonlinear, distributed parameter groundwater flow and solute transport simulation model (SUTRA) with a nonlinear optimisation method (MINOS). They used the embedding technique. The planning model was applied for two systems. Thefirst system involved steady-state aquifer reclamation. Contaminant withdrawal, in-ground dilution, and combined pumping/recharge strategies were considered. The second system involved transient flow and transport. Capturing a migrating contaminant plume and in situ dilution were the two management strategies considered.

Willis & Finney (1988) presented the planning and management model for the control of seawater intrusion in the Yun Lin regional groundwater basin. The aquifer was unconfined. The management model was formulated as a problem in optimal control. The optimal control problem was solved using (i) the influence coefficient method and quadratic programming, and (ii) the reduced gradient methods in conjunction with a quasi-Newton algorithm (MINOS). The simulation model developed by Mercer et al (1980a, b) was used to simulatethe response of the aquifer system within the planning model. The simulation model was based on the assumptions that a sharp interface separates freshwater from seawater. Also, Dupuit’s approximations were assumed to be valid. The control variables of the optimization model were the locations and magnitude of groundwater pumping/recharge. The state variables of the aquifer system were the freshwater head, saltwater head and the location of the interface toe, at the end of the planning period. The objective function minimised a weighted cost function of saltwater intrusion, water supply and recharge volume.

Finney et al (1992) presented the development and application of a quasi three-dimensional optimal control model for groundwater management in the Jakarta coastal aquifer basin. The movement of the freshwater–seawater interface was again based on the sharp interface assumption. The finite difference simulation model of Essaid (1990) was used to simulate the aquifer system response within the control model. The objective function of the model was a function of freshwater and seawater heads, and locations and magnitudes ofgroundwater pumping, or artificial recharge. The management problem was mathematically a nonlinear nonconvex programming problem with a flat response surface. They reported that MINOS was unable to differentiate between stationary points and local solutions and thus terminated with unusually large reduced gradients. Box’s algorithm which is a sequential search algorithm, improved the solution generated by MINOSby approximately 20%.

Application of combinatorial search algorithms

In the category of combinatorial search algorithms, two algorithms, viz. the genetic algorithm (GA) and simulated annealing (SA), have been used for groundwater management. Some recent works report the application of GA and SA for solving groundwater management problems. The genetic algorithm imitates some of the salient features of natural selection and natural genetics in order to find near-optimal solutions in a search space. The genetic algorithm operates on a population of decision variable sets. Three genetic operations, namely selection, cross over, and mutation, are applied on the initialised population to obtain an optimal solution. Simulated annealing uses the analogy between (i) the cooling and annealing process of solids, and (ii) optimisation of a multivariable function. However, this is an imperfect analogy. The five major steps of simulated annealing are: (1) representation of the possible system configuration in a concise form, (2) specification of the penalty type objective function, (3) rearrangement of the system, (4) control parameter and annealing schedule and (5) criteria for terminating the algorithm. More details on genetic algorithm andsimulated annealing can be found in Goldberg (1989), Holland (1975), Kirkpatrick et al (1983), Press et al (1986), Van Laarhoven & Aarts (1987), and Aarts & Korst (1989).

Application of the Artificial Neural Networks technique

Artificial neural networks (ANN) are intended for modelling the organisational principles of the central nervous system in the hope that the biologically inspired computing capabilities of ANN will allow the cognitive and sensory tasks to be performed more easily and more satisfactorily than with conventional methods. The network architecture has three basic components, namely: (i) a weighted summer which accumulates the weighted sum of the incoming signals to a neuron from other interconnected neurons; (ii) a linear dynamicsystem; and a (iii) a non-dynamic nonlinear function i.e. the transfer function defining the output responses of a neuron for a given input signal. Formulation of the network is the crux in the ANN technique. Training of the network is the Next phase in the ANN technology. For training of a network in a groundwater system, several groundwater responses corresponding to the aquifer stress scenarios are used. Once an ANN is trained toimitate a particular aquifer system, it can be suitably applied for further use in optimal management of the system also. Rogers & Dowla (1994) report the use of the ANN technique for optimal groundwater remediation design.

In the joint study of CGWB with NIH which has been recently completed with an objective to develop an operational management model to optimize the pumping at Palla well field to exclude the possibility of any upcoming of brackish water underlying the fresh water aquifer in the area a Simulation / optimization ( S / O) approach has been adopted. In the present approach the nonlinear , non –convex problem involving discrete (pumping locations) and continuous decision variables ( pumpage) has been solved within S/O framework. Itprovides an accurate representation of the aquifer responses but involve high computational burden. Therefore, in the present study Artificial Neural Network ( ANN ) is used as a virtual simulator of a variable density driven numerical flow model aquifer simulation. Simulated Annealing , non gradient based algorithm has been used as an optimizer in this study.

DECISION SUPPORT SYSTEM (DSS) FOR PLANNING AND MANAGEMENT OF GROUND WATER RESOURCES

Decision Support Systems (DSS) are most advanced technical tool to support information needs of ground water resources management processes. DSS application may pertain to groundwater resources planning or management depending on the scope of the decisions they intend to support.

Adopting a systems approach it is always advisable to go for an integration of all relevant data and knowledge under one platform which may include an open relational databases, an interactive tool to administrate and display geographical data, an interactive graphically supported user interface and a reliable knowledge base including, for instance, different groundwater models, statistical data about recent water utilization and long term trends in the region, area wise maximum acceptable groundwater levels, etc. Such integrated systemshould provide tools :

- For two or three dimensional grid design
- For establishing links between the data base and the specific grid and ground water model
- To analyze model simulations,
- To display the results in comprehensive form and ,
- To control the logical sequence in ground water modeling and optimization.

Such systems which should be implemented on a graphical work station are summarized by terms such as :

- Decision Support System
- Knowledge Based System
- Expert System

Expert system has been defined as an intelligent computer programme that uses knowledge and inference procedures to solve problems that are difficult enough to require significant human expertise for their solution. That is, an expert system is a computer system which emulates the decision making ability of human expert. The term ‘ emulate’ means that the expert system is intended to act in all respects like a human expert. The terms expert system and knowledge based system are often used synonymously.

Decision Support System ( DSS) support the decision making process in a complex environment by providing flexible models which can be interactively adapted to a specific problem and by providing a reliable data base which should be user friendly combined with the models. A typical DSS includes five main components: data acquisition system, user-data model interface, databases, data analysis tools, and a set of inter-linked models.The components of a typical DSS is shown in figure below.

Case Study :

A ground water Operational Management Model was developed for the northern part of Yamuna flood plain area in Delhi (Fig.1). The area is underlain by alluvial formation, in order to augment the water supply in Delhi area , to Central Ground Water Board constructed 95tube wells in Palla Sector in the depth range of 38-50 m for Delhi Jal Board, which is the domestic water supply agency of the Delhi area. The ground water pumping in the Pall well field offer a complex situation as the area is underlain by geologically occurring saline ground water. The amount of pumping in this case is mostly guided from water quality considerations rather than water quantity. Though the flood plain area offers goodalternative for ground water development but because of underlying saline water , any overwithdrawl or excess pumping may result in upconing of saline water leading to deterioration of water quality especially for drinking water needs. This may further complicate the overall saline –fresh water interface down below and cautious development of such aquifers arerecommended.

Keeping this in view, an attempt has been made to address the problem and develop anoperational ground water management model so that optimal ground water development plan along with withdrawl schedule in space and time may be recommended to useragency. The work has been taken up jointly by CGWB and NIH. The nonlinear, non-convex problem involving discrete (pumping locations) and continuous decision variables (pumpage) has been solved within the Simulation – Optimization (S/O) framework. S/O approach provides an accurate representation of the aquifer responses but involve high computational burden. Therefore in the present study Artificial Neural Network (ANN) is used as virtual simulator of a variable density driven numerical flow model for aquifer simulation.Simulated Annealing ( SA) a non – gradient based algorithm is used as an optimizer in this study.

On the basis of a Simulation Optimization model developed during the study it was concluded that nearly 30 MGD of water can be safely drawn from these tube wells during monsoon and non-monsoon seasons to augment the water supply to meet the ever increasing drinking water requirements of National Capital Territory, Delhi. The Palla well field is situated in the Yamuna river bank and during simulation, the river Yamuna has been considered as constant head boundary. After calibration of the model, the volumetricbudget indicates that out of the total ground water withdrawl a part of flood water (rejected recharge) is utilized to augment sub-surface storage during monsoon. This also provides the scope for induced recharge from the river Yamuna. Hence, the storage space created during the non monsoon pumpage gets replenished during the flood season and hence the overall ground water regime situation remains under control and provides sustainability.

The experience of Yamuna flood plains in Delhi has shown the scope of enhancing ground water recharge by pumping to lower the water table ahead of the rainy season and thus creating more space for the flood water to percolate. The concept can be implemented in similar situations in different parts of the country after carrying out detailed study on the hydrodynamics of the flood plain zones involving stream-aquifer interaction.

ACKNOWLEDGMENT

The author is thankful to Shri Sushil Gupta,Chairman, Central Ground Water Board for kindly permission to publish this paper.

Sujit Kumar Sinha
Central Ground Water Board, Faridabad