Indo-French Collaborative Project (2013-1) Optimal Development and Management of Groundwater in WeatheredFractured Aquifer

Scientific Report: Volume 3 GEOSTATISTICS, AQUIFER MODELING AND ARTIFICIAL RECHARGE

S. Ahmed, C. Engerrand, E. Ledoux, P.D. Sreedevi, Dewashish Kumar, K. Subrahmanyam and G. de Marsily

Sponsored by

INDO-FRENCH CENTRE FOR THE PROMOTION OF ADVANCED RESEARCH (IFCPAR)

NATIONAL GEOPHYSICAL RESEARCH INSTITUTE Hyderabad, India

CENTRE DINFORMATIQUE GEOLOGIQUE Fontainebleau, France

SIMULATION OF FLOW IN WEATHEREDWEATHERED-FRACTURED AQUIFER, GEOSTATISTICS AND ARTIFICIAL RECHARGE FOR GROUNDWATER MANAGEMENT IN MAHESHWARAM WATERSHED, A.P., INDIA

An Indo-French Collaborative Project (No. 2013-1) on Optimal Development and Management of Groundwater in WeatheredFractured Aquifer

Sponsored by INDO-FRENCH CENTRE FOR THE PROMOTION OF ADVANCED RESEARCH (IFCPAR)

LIST OF PARTICIPANTS

INDIAN TEAM Dr. Shakeel Ahmed Dr. K. Subrahmanyam Dr. P.D. Sreedevi Mr. Dewashish Kumar Dr. P.Babu Rao Dr. B.N. Prasad Mr. S.A.R. Hashimi Mr. B.R. Kishen Mr. Mohd. Idris FRENCH TEAM Dr. E. Ledoux Prof. G. de Marsily Dr. P. Lachassagne Dr. J.C. Marchal Dr. D. Bruel Dr. A. Lavasser Dr. C. Engerrand Prof. P. Davy

NGRI NGRI NGRI NGRI APGWD APGWD APGWD APGWD APGWD

CIG, ENSMP UPMC BRGM BRGM CIG, ENSMP CIG, ENSMP CIG, ENSMP Univ. of Rennes

PRINCIPAL COLLABORATORS India Dr. Shakeel Ahmed, NGRI France Dr. Ledoux Emmanuel, CIG 2

Contents
CHAPTER 2 2.0 INTRODUCTION ON APPLICATION OF GEOSTATISTICS 2.1 Variography of a Hydrogeological Parameter 2.1.1 Calculation of Experimental variogram from scattered data 2.1.2 Modeling an experimental variogram 2.2 Cross-validation test CHAPTER 3 3.0 VARIOGRAPHIC ANALYSES OF WATER LEVELS IN THE AQUIFER AND EVOLVING A COMMON VARIOGRAM 3.1 Variographic Analysis of the water levels 3.2 Cross-Validation Test of the Variogram of the Water Levels 3.3 Evolution of Common Variogram 3.4 Conclusions CHAPTER 4 4.0 DATA COLLECTION NETWORK DESIGN 4.1 Optimal Monitoring Network Design Using Geostatistics and Other Stochastic Methods 4.2 Geostatistical Optimization of the Monitoring Network Developed 4.3 Optimization of water level monitoring network in study area 4.4 Conclusion CHAPTER 5 5.0 MODELLING THE GROUNDWATER FLOW 5.1 Introduction 5.2 Conceptualisation of the hydrogeological system 5.2.1 Time steps 5.2.2 Geometry 5.2.3 Boundary condition 5.2.4 Topography 5.2.5 Aquifer geometry 5.3 Hydraulic conductivity 5.4 Specific yield and storativity coefficient 5.5 Recharge 5.5.1 The Natural recharge 5.5.2 The Preferential recharge 5.5.3 The discharge rates 5.6 Results and Discussions 5.6.1 Fitting Criteria 5.6.2 Balance 5.6.3 Discussion on the simulated head 5.6.4 Discussion on the water reservoir capacities of the watershed and the probable evolution of its groundwater resources. 5.7 Conclusion

CHAPTER 6 6.0 WATER MANAGEMENT AND ARTIFICIAL RECHARGE 6.1 No-Cost Water Conservation Technology For Recharging Over-Exploited Aquifers 6.2 Artificial recharge in the study area and its simulation 6.3 Conclusions and Recommendations

1.0 GENERAL INTRODUCTION

1.1 Background

All life on this planet depends on water, a precious resource. Yet, we are struggling to manage water in ways that are efficient, equitable, and environmentally sound. Many parts of the world are facing increasingly deteriorating conditions as cities expand, populations grow, and sources of clean/fresh water vanish. The situation is even worse in the heavily populated India. The policies of the GOI during 1970s have provided large scale development of groundwater through subsidies in the form of lower levies on power consumption. This has boosted the agricultural production immensely but unfortunately, there were no compelling rules on the extraction of the groundwater that has lead to its exploitation beyond the optimum yielding capacity of the aquifers. Thus the uncontrolled exploitation of groundwater resources specially in the hard rocks has de-saturated the weathered column of about 15-20 meters that formed the dynamic groundwater storage of the aquifer system. Presently, the wells tap the fractured systems at greater depth. It is very well understood that the groundwater in a region is controlled by the climatic and geological conditions. Hard rocks that are devoid of primary porosity, occupy about two third area of the country. Hence over all groundwater scenario calls for understanding the hard rock aquifers for sustainable development and management of this important resource in different geological environs in the absence of major surface water resources. A large amount of similarities exist between India and France as far as the hard rock formations are concerned and hence we have continued research programs in both the countries to tackle such aquifer systems. Thus a collaborative project was taken up aiming at evolution of a cost-effective methodology relevant to fractured rocks to determine: how groundwater is recharged in areas where the fractured basement is covered with weathered rocks, as is common both in India and in many parts of the world; how bore-holes can be best located to tap groundwater in the weathered zone and underlying fractured basement; what is the best management strategy to exploit the aquifer yield and thus minimize the losses, and how can this yield be developed by artificially increasing the recharge.

1.2 Objectives
The objectives of the present study have been to thoroughly understand the functioning of the weathered-fractured aquifer system and simulate the flow to develop methodologies for optimal management of water resources. A carefully selected catchments area in the granite terrene has been surveyed, instrumented and monitored with a view to develop a cost-effective methodologies which could be applied in other areas also for: Assessment of the sustainable yield, Optimal location of wells to exploit groundwater and Cost-effective means to increase the water resources and/or minimize the losses.

1.3 Brief description of the research carried out

The entire project was has been of applied nature and since the study area lied in India most of the experimental work have been carried out in India with some interpretation and modeling work that was carried out in France. The collaborating organizations have been: India: National Geophysical Research Institute & Andhra Pradesh Ground Water Department France: Centre dInformatique Geologique, Ecole des Mines de Paris Bureau des Recherches Geologique et Miniere Sisyphe, Universite Pierre et Marie Curie Institut de Recherche pour les Developments & Geosciences, Univ. of Rennes Various research activities are as follows: Selection of the study area: A study area in the granite terrene (to have weathered and fractured aquifers) has been selected considering a number of logistic and scientific parameters. The study area is a closed watershed of about 60 Km2 so that it could be very well monitored and instrumented. Since we do not intend to study pollution problems, this area has been away from industries. The main activities in the area have been agriculture, of course, with heavy exploitation. A large number of outcrops are available to map the structures and fractures. A number of intrusive like dolerite dyke, pegmatite veins and quartz reefs that are common in such terrene, are present in the area. The area is only 35 Km from the NGRI and very well connected with highway so that any field work could be carried out without loosing much time. The study area was also selected with the Indian and French team together. A complete videography of the area including the main geological features were made and sent to France to show to all the participants from France who could not visit the area due to limited exchange visits. Geological Investigation and well inventories: The entire outcrops were mapped by measuring the dip and strike of the major fractures and joints and a rose diagram is prepared showing the major direction of the fracturation. All the lineaments and intrusive were marked from the areal photographs and ground surveys. Complete inventory of the wells present, were made having about 200 dug wells and more than 800 bore wells. An approximate estimation of water withdrawal was also made. Water level is very crucial parameter and a suitable network has been evolved out of the existing bore wells to start the water level monitoring from November 1999. A number of rock samples were analyzed including their thin sections in BRGM, France to arrive at the thickness and type of the weathering cover. Geophysical investigations soil gas emanometry: The entire area has been surveyed with a variety of geophysical methods depending on the local 6

inhomogeniety and other physical conditions. Several Km lines of electrical imaging were carried out to find out the resistivity contrast along with a large number of Vertical Electrical Sounding (VES) to provide the depth to the bedrock, thickness of the weathered and fractured zones. The survey has been conducted using the equipment from the French collaborative organizations and a team of geophysicists from France has participated in the fieldwork and in the interpretation. The magnetic survey with the electrical profiling has provided the depth till the dykes are weathered as this information has been useful in simulating flow. The new method of gas emanometry mainly by sampling radon gas was applied, of course, in limited areas to correlate the geophysical interpretation. Later this has proved to provide very promising site for groundwater and the same was confirmed by drilling a series of bore wells. Geophysical logs were performed later after the wells were drilled and also a special method of delineating the extent of fracture viz., Mise-a-la-Masse were also employed. Drilling of the observation wells and hydraulic tests: The hydrogeological investigations supported by the geophysical prospecting have provided sites for drilling about 25 bore wells fairly distributed in the area. Well defined drilling records were maintained to arrive at geological logs, drill time logs and depth of water striking etc. These information have extensively been used in interpreting the hydraulic test data and to conceptualize the geometry of the aquifer. A large number of different kind of hydraulic tests have been performed so that a realistic picture of the parameter variability is obtained. Starting with six short duration pumping tests of six hours each have provided aquifer flow and storage parameters. Later a few close observation wells have been drilled and long duration pumping tests (18 to 30 hours) to arrive at the regional picture of the aquifer parameters. Due to logistic difficulties and the fact that water levels have gone below the weathered zone aquifer, it was decided that slug and infiltration tests to be conducted where water is not pumped from the well. Thus all the 25 newly drilled wells in the project were tested with the slug tests and local permeability were determined that were later regionalized. The infiltration tests with flow meter have provided permeability of individual fracture zones and also confirmed the depth and location of water bearing fractures. Water level monitoring and optimal network design: Water level being time variant parameter was monitored every month. Initially about 32 wells from the existing farmers well selected for monitoring then 25 wells drilled in the project were added to the monitoring network. However, the network has become dense and difficult to monitor in the shortest possible time. The water level data were analyzed geostatistically for its stochastic nature and using the variability in space and time, we arrived at (1) a few common variogram to avoid complicated variographic analyses for every time periods and (2) an optimal network of only 40 wells (25 newly drilled wells and 15 from the existing wells) to monitor water level every month. Later these values were used for calibration of the aquifer model. Water quality assessment including profiling of EC: Although analyzing water quality was out of the scope of the study, however, on six monthly basis water 7

samples were collected and analyzed for major ions and study their behaviour with time. In addition using an EC logger all the 25 project wells have been logged with EC and temperature. Although, temperature has not given any significant variation but EC has at times shown significant change confirming mixing of water from fractures at that depth. Water balance study and water budgeting: With the above work on surveying and instrumentation, the aquifer system was conceptualized for its geometries and flows. However, assessment of potential its dynamics, a precise estimation of its most components are important. Various components of water flow have been calculated and a water balance of the entire watershed was prepared. It was obtained that about 1.18 meter depletion of water has been found on average in the area in one year due to over exploitation. This study conceptualizes the different components of water budget and provides the water budgets for the predictive study. Up-scaling of the parameters: Scale effects are very important while studying the hard rock aquifers due to high variability of the parameters. On the other hand most of the hydraulic tests or other measurements provide either point value or very localized value. The mesh of the aquifer model, no matter how small is made, are much bigger than the scale of measurements. Thus statistical approach was applied to up-scale the parameter particularly Transmissivity and Storage coefficient satisfying a few statistical parameters and giving the same variograms. The entire work on Discrete Fracture Network Modeling and up-scaling was performed using FRACAS code at the CIG, France. Public Awareness programs: It was thought useful as part of the water management program to make the local public aware of the groundwater and its availability. A series of meetings were conducted to apprise the villagers how the hydrological cycle works and precautions they should take to avoid water wastages. They were also told to understand water balance and prepare water budgets. The awareness program has been very successful and about 20 volunteers have been assigned the rainfall measurement task using simple buckets. This has provided a very good rainfall data variability in the area that has later used in the aquifer model.

Simulation of flow in the aquifer system: This is the main important part of the project. After the conceptualisation of the aquifer system, its geometry and extent as well as flows, a numerical model has been fabricated. The model simulates two layered system, the first one as porous formation representing the weathered layer the second one as equivalent porous medium representing the fractured layer without any confining layer in between. The entire area was divided into meshes of 100m by 100 m size giving about 5272 meshes in each layer. Integrated finite difference method was used to discretize the groundwater flow equations and simulate the flow. The aquifer parameters and boundary conditions were suitably taken from all the above studies, they were regionalized using theory of 8

regionalized variable and assigned to all the meshes of the model. The model was calibrated for the period of January 2001 till July 31, 2003 against the monthly water level observed in the field. The model responses have been by and large matching with the observed ones. Augmentation of water resources to balance the deficit: The misbalance in the water budget and continuous decline of groundwater levels strongly demands the augmentation of the recharge to the aquifers. Although, the area contains 9 tanks (surface reservoir) but all of them are highly silted making them as evaporation tank. A new and cost-effective (practically no-cost) methodology was developed and applied to recharge the aquifers through defunct dug-wells. A few wells having the maximum or sufficient catchments areas have been selected. The entire run-off water of the rainfall is collected into a pit to settle the silt and transported material and then allowed to flow into a nearby dug-well. The vertical hydraulic conductivity at the bottom of the dug-well was determined before filling the water in it using the double infiltrometer. In addition, a bore well was also drilled near the dug well to monitor the recession of water in the dug-well vis--vis the water level change in the bore-well i.e. the aquifer to establish the effectiveness.

A number of exchange visits have helped working together and provided scientific interaction. The outcome of the project has been development of a numerical model of the aquifer and design of a demonstrative water conservation experiment. About eight research papers have been published from the project findings and the results were presented in about 9 presentations in the International conferences and seminars. Based on the research work in the project two keynote lectures were delivered and two doctoral theses have been prepared.

1.4 Organization of the scientific report

Due to large amount of work the report has been prepared in 3 volumes. Volume 1 describes variety of geophysical investigations (both surface and sub-surface) carried out in the study area to determine the geophysical properties of the formations forming the aquifer system and also determining their geometry as well as extent. The investigation have revealed the information of the variability of the weathered formation in its thickness and resistivity, the information of the fracture density and most important part of the extent of the fractures encountered during the drilling of the bore-wells. Volume 2 deals with all the geological and hydrogeological investigations carried out in the area including the gas emanometry. The work includes the complete well inventory and geological investigations on the outcrops as well as in the dug-wells. Based on the hydrogeological, geophysical information and radon gas surveys, a number of bore-wells have been drilled in the study area to obtain the sub-surface geology and use these wells for frequent water level monitoring. After these investigation a large number of different type of hydraulic tests viz., short and long duration pumping tests, slug tests, specific capacity tests and infiltration tests have been carried out to arrive at an accurate distribution of hydraulic (storing and transmitting) properties of the aquifer. This part of the study has provided a conceptual model of the aquifer system, hydraulic properties and the estimates of the boundary conditions. 9

Volume 3 consists of three distinct studies viz., application of geostatistical methods to analyze the water levels variability in time and space and then designing of adequate/optimal network of the wells for monitoring of the water levels. Then information from all the investigations have been put in and a numerical aquifer model has been prepared. The report also describes the calibration of the model and prediction as well as sensitivity of the model parameters. Finally this model has also simulated an artificial recharge experiment that has been designed and carried out in the area for water conservation and water resource augmentation.

1.5 Acknowledgements
The Indo-French Center for Promotion of the Advanced Research (IFCGR) New Delhi has provided financial support to carry out most of the investigation and we would like to acknowledge the same with thanks particularly the support of Mr. Mony, Director, CEFIPRA. We are also grateful to the Scientific Council of the IFCPAR particularly Profs. R. Sadourny and V. Courtillot who have carried out the reviews and also supported any intermediate approval required particularly providing an extension of one year so that we completed a number of crucial experiments. We are grateful to a number of scientists working in the collaborating organizations whose names are not figured in the project participants for their support particularly of consulting nature. All the organizations involved have supported with their basic facilities and the equipment for completing the project with special mention of the Indo-French Centre for Groundwater Research (IFCGR) set-up at NGRI.

10

2.0

INTRODUCTION ON APPLICATION OF GEOSTATISTICS

The success of any scientific study greatly depends on the quality and the quantity of the basic data. Scarcity of data and their collection on isolated location mainly in the field of hydrogeology makes it necessary to adopt special procedures or an estimation technique to match between field measurements and data requirements. Numerical simulation of flow and transport processes in an aquifer necessitates, dividing and discretizing the natural heterogeneous system into a number of small volumes called mesh which are assumed to be uniform with almost no variation of the aquifer properties over it. To satisfy this condition, it is necessary to discretize the system into much finer and hence more number of grids. Although with the availability of more powerful computers, computation with large number of grids/mesh is not a difficulty but the data preparation that is to assign the aquifer parameters to each grid/mesh becomes cumbersome. Thus an appropriate estimation procedure is required to provide an unbiased, minimum variance estimates with unique value over the entire area of the mesh. Geostatistical techniques in the form of "Theory of Regionalized Variables" were initially developed to apply to mining problems (Matheron, 1963). But soon, hydrogeologists have realized its applications to the groundwater hydrology and a number of studies have been carried out on hydrogeological problems. Works of Delhomme (1978), Mizell (1980) Aboufirassi and Marino (1983), Neuman (1984), Hoeksema and Kitanidis (1984), Ahmed and Marsily (1987, 1993), Ahmed et al. (1988), Roth et al (1996) etc. have shown more applications of Geostatistics in groundwater hydrology. However, multivariate and nonstationary Geostatistics have found comparatively more applications in groundwater hydrology. Also some of them have to be suitably modified as well as some special procedures developed for a meaningful application of Geostatistics in this field. Delhomme (1976) has developed the method of Kriging with Linear Regression, Kriging using erroneous data, Kriging in the presence of a fault etc. Conditional simulation has also been applied in aquifer modeling (Delhomme, 1979). Galli and Meunier (1987) and Ahmed (1987) have worked on Kriging with an External Drift. Ahmed and Marsily (1987) have compared a number of multivariate Geostatistical methods in estimating transmissivity using data on transmissivity and specific capacity. Also Ahmed (1987) has developed a special antisymmetric and anisotropic cross-covariance between residuals of hydraulic head and transmissivity based on the work of Mizell (1980) and used coherent nature of various covariances to cokrige transmissivity and hydraulic head in solving an Inverse Problem (Ahmed and Marsily, 1993). Now Geostatistics has found applications in almost all domain of Hydrogeology from parameter estimation to predictive modeling for Groundwater Management e.g., designing an optimal groundwater monitoring network, estimating parameters at unmeasured locations, groundwater model fabrication (optimal discretization), unbiased model calibration using estimation errors and in deciding the best models for prediction. The work of Hughes and Lettenmair, 1981; Carrera et al., 1984; Rouhani, 1985; Loaiciga, 1989, Gao et al, 1996 etc. are some examples of the application of Geostatistical techniques to the optimal data collection network design. However, Agnihotri and Ahmed, 1997 have brought out some crucial ambiguities in such application and thus a few modifications have been brought out tom make the procedure effective and useful. The present study describes such modified/improved procedures through a few case studies 11

carried out on designing monitoring network optimization in a fractured granitic aquifer from a semi-arid region in south India. Geostatistical estimation variance reduction, cross-validation techniques etc. are a few procedures that could study adequacy of a given monitoring network and could evolve an optimal monitoring network with some given constraints. The advantage of the geostatistical estimation technique is that the variance of the estimation error could be calculated at any point without having the actual measurement on that point (well). Thus the benefits to be accrued from an additional measurement could be studied prior to its measurement. The main steps involved in a geostatistical technique applied to hydrogeological parameters are: Variography i.e., structure analysis, cross-validation, estimation and backward transformation (if any).

2.1 Variography of a Hydrogeological Parameter

The theoretical part of the Geostatistical techniques have already been dealt with earlier workers e.g., Matheron, (1971), Journel and Huijbregts (1978), Marsily (1986), Isaaks and Srivastava (1989), Samper and Carrera (1990), Deutch and Journal (1992), Wackernagel (1995) etc. Most of the hydrogeological parameters are defined and measured at points in a 2D space. Therefore, all the derivations and examples in the chapter are given in 2D space and a point estimation is used. Variography in determining variability of a parameter is an important step and quality of the estimation result depends on it. The procedure involves calculation of experimental variogram, its modeling through theoretical variogram and its validation by reproducing the field data.

2.1.1 Calculation of Experimental variogram from scattered data

Unlike mining or airborne geophysical survey, hydrogeological parameters are measured on scattered locations due to the fact that most of the parameters are collected from the wells which exist in the vicinity of a village and not on a grid pattern. Low cost of the groundwater projects also restrict a systematic and extensive data collection. A generalised formula to calculate experimental variogram from a set of scattered data can be written as follows (Ahmed, 1995). (d, ) = where $ d + d , - $ + d - d d with d = 1 Nd
Nd $ i=1 di , =

1 2 Nd

Nd $ $ )- z( xi , $ ) ]2 [z ( xi + d, i=1

(1)

(2) (3)

1 Nd

Nd $ i=1 i

where d and are the initially chosen lag and direction of the variogram with d and as tolerance on lag and direction respectively. d and are actual lag and direction for the corresponding calculated variogram. Nd is the number of pairs for a particular lag and direction. The additional eqn (3) avoids the rounding off error of pre-decided lags (multiples 12

of the initial lag only are taken in conventional cases) and the direction. It is very important to account for every term carefully while calculating variograms. If the data are collected on a regular grid, and d is taken zero, eqn (2) and eqn (3) will be simplified only for . Often, geohydrological parameters exhibit anisotropy and hence variograms should be calculated at least in 2 to 4 directions to ensure existence or absence of anisotropy. Of course, sufficient number of samples are required in that case.

2.1.2 Modeling an experimental variogram

Calculation of experimental variogram is subjected to many approximation and its plot is mostly irregular. A smooth curve is therefore, fitted to it. This fitting is called modeling and the fitted curve is called a theoretical variogram. A theoretical variogram is defined by a number of parameters e.g., sill, range, nugget effect and model type. These parameters of a variogram could approximately be decided by visualising the experimental variogram as well as by nature of parameter. For example, a variogram of transmissivity posses usually, a nugget effect but that of the water-level does not. Sometimes, an experimental variogram is not satisfactorily fitted by any of the limited available theoretical variograms. Thus a combination of several variograms are fitted and the resultant variogram is called a nested structure. Following resultant variogram is only authorized as a nested structure. (d) = ik=1 a i i (d)
R

(4)

where a i0 > 0 i and i (d) are individual variograms. Mostly the fitting is visual but often an automatic fitting such as least squares etc. is also used. However, a measure of difference between theoretical and experimental variograms is always calculated to decide the best of several fits.

2.2 Cross-validation test

Theoretical variogram obtained by fitting an experimental variogram is usually ambiguous. It is therefore, necessary to validate it against the measured value. The cross-validation is performed by estimating the parameter at the points of measurement and comparing them with the field values in a statistical sense (Ahmed and Gupta, 1989). In this exercise, a measured value is removed from the data set one by one and the same is estimated at that point using the remaining values and the structural model. The process is repeated for the entire data set. Thus we will have, at all the measurement points, measured value (z), estimated value (z*) and variance of the estimation error (2). This leads to computing following statistics. 1 n (5) i=1 ( zi - z*i ) 0 n 1 n 2 ( zi - z* i ) min ..............(6) n i=1 1 n ( zi - z*i )2 1 i=1 2 n i

(7)

13

* z z
i

2i...........(8)

Various parameters of the variogram model are gradually modified to obtain satisfactory values of the eqns. 5 to 8. Therefore, during the cross-validation we test many important points such as: i. ii. iii. iv. Inferring a structural model and removing its ambiguity. Deciding optimum neighbourhood. Selecting suitable combination of additional information particularly in case of multivariate estimation. Sorting out the unreliable data.

3.0 VARIOGRAPHIC ANALYSES OF WATER LEVELS IN THE AQUIFER AND EVOLVING A COMMON VARIOGRAM
Although, water-level is a smooth varying parameter and also maintains continuity, it often shows high variability depending on the variation in the system parameters arising from the heterogeneity of the formation as well as stress applied to it such as rainfall recharge and extraction. Geostatistical techniques, using the theory of regionalized variables, have been applied to study water-level in aquifers by a number of workers (Mizell, 1980; Dagan, 1985; Ahmed, 1985; Rouhani, 1988; Dong et al., 1990; LaVenue & Pickens, 1992; Ahmed & Marsily, 1993; Roth, 1995 etc.). Rouhani (1990) has attempted a multivariate geostatistical model using jointly the time and space variability of water-level but was limited to the theoretical developments. As water-level is time varying and is often monitored using the same network depending on the required monitoring frequency, estimating water-levels for all time periods following complicated steps of geostatistical estimation becomes cumbersome. Also the time variability of water-level is influenced by external stresses that are often periodic in nature (both pumping and recharge due to monsoon rainfall). It is therefore, possible to group water-levels for certain time periods having similar behavior and analyze them geostatistically for its spatial variability. The present study is based on the use of monthly water-level data from a small watershed in an aquifer in semi arid region in India. They have been analyzed geostatistically and an attempt was made to evolve common variogram(s) for the two contrasted periods of the year, affected or not by the monsoon. The results of using these common variograms were compared to the monthly ones.

3.1 Variographic Analysis of the water levels

The presence of drift in a parameter produces an unbounded variogram and it becomes difficult to model it with a finite sill and range. Therefore, drift has to be removed directly or indirectly to obtain a bounded variogram. Initially unbounded variograms were obtained 14

due to presence of a clear drift in water-levels (Fig. 1). Since a typical water-level map shown in fig. 1 depicts a single direction of major flow, it was decided to calculate directional variograms. Since the major drift terms are present in the flow direction, and they have practically no contribution in the perpendicular direction, calculation of experimental variogram in the perpendicular direction of major flow provides a bounded variogram that can be assumed as variogram of residuals. Calculation of experimental variograms for all the 10 months under consideration was performed in a direction perpendicular to the main flow direction and yielded bounded variograms. These variograms were calculated and modeled with theoretical models using an interactive program developed at National Geophysical Research Institute (Ahmed, 2001). A theoretical variogram is fitted graphically by visual trial and error method. The sill ranges from 160 m2 (June 2000) to 100 m2 (April 2000 and January 2001). The range of these variograms varies from 5 000 m (July 2000) to 3000 (January 2001).
9000 IFW-31 8000 IFW-23 7000 IFW-20 6000 IFW-18 IFW-17 IFW-16 IFW-19 IFW-21 IFW-22 5000 IFW-15 IFW-11 IFW-10 IFW-12 IFW-9 IFW-13 4000 IFW-14 IFW-7 IFW-8 3000 IFW-6 2000 IFW-5 IFW-4 IFW-2 IFW-3 Average Flow Direction 1000 Well Location 0 0 1000 2000 3000 4000 5000 6000 7000 8000 IFW-30 IFW-1 IFW-27 IFW-26 IFW-24 IFW-25 IFW-29 IFW-28

IFW-32

630

622

Northing in metres

614

606

598

590

Easting in metres

Fig-1 Water-level Contour map, Jan-2000 (Surfer without kriging)

3.2 Cross-Validation Test of the Variogram of the Water Levels

Due to the number of approximations made during calculation and modeling of the experimental variograms, the fitted models are bound to have ambiguities. Therefore, it was necessary to validate these variograms. Cross validation test was performed for all the months starting from January 2000 to January 2001 (Ahmed and Gupta, 1989) and revised variograms for each time period were then obtained.

15

3.3 Evolution of Common Variogram

To avoid calculation of experimental and theoretical variograms for each time period, an attempt was made to determine two common variograms for periods of several months. It was observed that during the monsoon affected periods the spatial variability of waterlevels are different than the spatial variability of water-level in the remaining periods of the year (fig. 2). Thus a common variogram by averaging the nugget, sill and range for the monsoon affected months was evolved (called monsoon variogram) and a similar way another common variogram was evolved for non-monsoon period (called non-monsoon variogram). It is important to note that monsoon periods for the rainfall in general are from July to October but for computing of the variograms we considered August to December as during these months the water-levels are influenced by the monsoon rains. The criteria of the limit between these two periods is that the rainfall recharge term is added during the monsoon period and it is absent during the remaining months. The common variograms (Fig. 3) were used for performing a cross-validation test with the observed value from all the months, and the results have been compared with those of the individual month.(Table 1)
Water level Vs Rainfall (IFW-1) 300 250 200 150 100 50 0 Jan-00 Apr-00 Jun-00 Jul-00 Aug-00 Sep-00 Oct-00 Nov-00 Dec-00 Jan-01 625 623 621 619 617 615 Water level (amsl) in m

Rainfall in mm

Rainfall Water level

Time in months

Fig. 2 Variation of water levels with rainfall in the study area

3.4 Conclusions
The present study shows that performing geostatistical analysis of a parameter for different time periods, it is not absolutely necessary to carry out variographic analysis separately for all the time periods, that at times, are quite cumbersome and ambiguous. Although it is not always possible to evolve a single unique variogram for all time ``periods, the study with the help of cross-validation test has shown that if the year cycle is divided into two parts; monsoon and non-monsoon periods, it is possible to evolve a common variogram for each part. The common variograms during the validation test could satisfactorily reproduce the measured values of water-level without losing much on the outcome (mean estimation error & mean reduced error) as compared to cross-validation test using individual mensual variograms. This was possible for water-levels as both the input (rainfall recharge) and the output (groundwater withdrawal for irrigation) more or less follow a cyclic pattern each year. Even in the case of low or high rainfall years, the special variability of water-levels could be assumed to remain same. 16

The evolution of common variogram helps analyzing water-levels for all the time periods that could be used, for e.g. calibration of an aquifer model etc. The study also helps in estimating water-levels at any time period when all the wells could not be monitored due to any reason. This would have, otherwise been extremely difficult because the variographic analysis in the absence of sufficient measurement becomes much more ambiguous.

200

150

(d) in m

Monsoon
100

Non-Monsoon

50

0 0 1000 2000 3000 4000 5000 6000

d in m

Fig-3 Various types of Common variograms studied

Table 1 Comparison of cross-validation with common and individual variograms

Month

Variogram obtained after crossvalidation test

Model type Nugge t (m2) Sill (m2) Range (m)

Month wise

Non-monsoon (d)=0.0+110 sph(4100) SME SMRE

Monsoon (d)=0.0+128.4 sph(4390) SME SMRE

SME*

SMRE

01/2000 04/2000 06/2000 07/2000 08/2000 09/2000 10/2000 11/2000 12/2000 01/2001

Spheric Spheric Spheric Spheric Spheric Spheric Spheric Spheric Spheric Spheric

0 0 0 0 0 0 0 0 0 0

140 100 160 130 140 147 140 110 105 100

4500 4000 4000 5000 3750 4200 5000 4500 4500 3000

38.45 24.78 63.49 35.75 44.67 36.44 29.62 29.32 27.91 33.50

1.03 0.69 1.02 0.97 0.72 0.80 0.84 0.83 0.86 0.66

38.37 24.99 64.74 36.13

1.20 0.65 1.54 0.93 47.93 36.36 29.91 29.27 27.80 0.99 0.96 0.80 0.69 0.68

28.87

0.75

* where SME and SMRE are square mean error and square mean reduced error respectively

4.

DATA COLLECTION NETWORK DESIGN

Data collection from a finite number of observation/monitoring points randomly or systematically distributed is necessary to infer the spatial variability of any parameter under study. The number and distribution of such stations are constrained by numerous factors of which cost and feasibility are quite common to consider. Therefore, it is imperative that an optimal monitoring network be evolved using minimum number of 17

observations stations that can provide maximum information. At the same time configuration of a network also depends on the objectives and the end use of the project. One of the important and obvious end use of the data collection is to infer or estimate the parameter at the intermediate and/or unmeasured locations. Obviously even using the best available interpolation/estimation techniques, there would certainly be an estimation error and the further objective should be to improve upon this error in the form of minimization of variance of the estimation error. Based on this criterion a procedure of optimizing a temperature measurement network using geostatistical technique is developed.

4.1 Optimal Monitoring Network Design Using Geostatistics and Other Stochastic Methods
A number of workers from several years have been applying the geostatistical and statistical techniques for optimization of groundwater monitoring network. However, mostly the developments have been limited to theoretical studies and insignificant application made to the real field study. Rouhani and Hall (1988) used composite programming with a combination of Geostatistics and multicriterion decision making in optimizing network of measuring thickness and porosity of a two layer aquifer system. Dillon (1988) used the method of estimation variance reduction calculated at the center of a set of descritized blocks in the area. Hudak and Loaiciga (1993) describe a number of crucial features considered in monitoring network design methods. Das, 1995 developed an analytical method integrating the practical implementing aspects and applied to a multilayered groundwater flow system for contaminant detection. Hughes and Lettenmaier (1981) have suggested an algorithm to optimize the location of data collection points by minimizing the variance of the error in estimating the parameter over the entire area of the aquifer. Sophocleous et al. (1982) have applied the technique of Universal Kriging in analyzing the network of wells for water-level measurement in Kansas, USA with respect to cost of the network and the accuracy obtained. However, it has been limited to the data regularly spaced along a square grid. Virdee and Kottegoda (1984) have proposed a map of kriging estimation error (k) and located new measurement points at places where a high value of k was found. This procedure has two drawbacks; (1) it is difficult to decide a limit to compare the k values to, and (2) an additional point improves the estimation variance not only at that point but also at the neighbouring points forcing the procedure to work in an iterative way only. Bogardi et al. (1985) used composite programming with a combination of geostatistics and multicriterion decision making in optimizing a network of measuring thickness and porosity of a two layer aquifer system. Gao et al. (1996) presented a simple algorithm to rapidly compute the revised kriging estimation variance when new sample locations are added. Moreover, we believe that this algorithm is useful only if a network has to be improved from a sparse one by adding additional measurement points. This algorithm may not be useful for optimizing a dense network either by deleting the measurement points or by shifting them. Bras and Rodriguez-Iturbe (1976) applied multivariate estimation theory to obtain the arial mean precipitation of an event over a fixed area considering the spatial uncertainty and correlation of process, correlation in measurement errors and non homogeneous sampling costs. A direct optimization of the monitoring network through geostatistical technique is not feasible as the objective function in the form of the variance of the estimation error does not directly contain the locations of the measurement point. This is why it is necessary to optimize in an indirect and iterative way.

18

There has been a large amount of work using different statistical and geostatistical procedures in monitoring network design. Langbein (1979) and Loaiciga et al (1992) have presented overview for such applications. However, Agnihotri and Ahmed (1997) have made a short review and highlighted ambiguities in the methods with suitable examples. In the present work a procedure is developed to optimize a given network for measurement of an air temperature for a given accuracy in the form of variance of the estimation error and the advantages of the additional measurement points are shown.

4.2

Geostatistical Optimization of the Monitoring Network Developed

It is difficult to define or generalize the necessary and/or sufficient data for a particular study but availability of adequate measurements to capture the variability of the parameter is the key for a successful scientific study. Large amount of measurements will make the study easy but the project extremely ill-favoured or uneconomic but less number of data will make the study gloomy. It is difficult but important to determine the optimal requirement of data for any study. Often it depends on the scientific objective of the study also. The main two objectives on which the optimization was based have been that the optimized network should be able to: represent the true variability of the parameter under study and provide its estimates on fairly finer grid with a desired accuracy in the form of the variance of the estimation error. Thus the entire area is usually divided into reasonably finer grid and the variance of the estimation error are calculated through a suitable kriging technique and the same are compared with the pre-decided or desired limit of the variance of the estimation error. Thus depending on the outcome of the comparison a network is categorized into dense, sparse or near optimal. The network is iteratively optimized either by discarding, adding or shifting the measurement points. Following norms are calculated by comparing the c values with i values all over the area. Mean value of i as < i > (9) (10) (11)

No. of grids (M) where i > c Sum of the Squared Difference (SSD) = ( i c)2 where i > c

If the network is dense, the value of M and SSD will both be zero and measurement points can be discarded starting from the location with the least i . If the network is sparse, M and SSD will be non zero and positive and their magnitude directly relates to the sparseness of the network. New measurement points can be added at the highest I areas. The monitoring network need not be dense or sparse but it may not be still optimal as per the desired accuracy/constraints. In that case values of M and SSD will be moderately positive and the network could be optimized by shifting the measurement points.

19

4.3

Optimization of water level monitoring network in study area

In a small watershed of 60 km2 area (Maheshwaram watershed) near Hyderabad, India (Fig. 4) groundwater is mainly found in a coupled system of weathered and fractured granitic rocks. However, due to over exploitation and successive reduction in the rainfall recharge, the water table has declined and the saturated flow is mainly confined to aquifer consisting of highly fractured rocks only. Crystalline rocks of Archaean age, comprising gray and pink granites cover a major portion of the study area; porphyritic granites intruded by dolerite dykes and quartz reefs. The granites have undergone variable degree of weathering and fracturing. Large scale fracturing and jointing has resulted in formation of huge boulders of granite, which are also scattered randomly in the area. The water levels are being monitored through a network of about 56 bore wells out of which 25 have been specially drilled to observe comparatively undisturbed water table and the other 31 bore wells are selected based on the drainage pattern and intervals etc. from the existing private wells used for irrigation (Fig. 4). The water level measurements have been carried out on monthly basis for a period of almost one hydrological cycle. About 31 wells (indicated as IFW) out of the 600 irrigation wells existing in the area have been selected for water level measurement based on the drainage pattern present, variation in rock formation covering the study area. Later 25 wells (indicated as IFP) taping the fractured aquifer have been drilled up to a depth of 45 m uniquely to monitor the water levels. These wells have been drilled based on the recommendation from geophysical investigations. Thus it was thought to reduce the number of wells so that; all the wells are monitored in a shortest possible time say one single day, discard some of the irrigation wells fitted with pumps as it was difficult to monitor static levels in these wells and reduce the cost of monitoring also

without loosing the monitoring benefits. The objectives for geostatistical optimization of the monitoring network has been that the monitored water levels should (i) represent the true variability of the parameter and (ii) provide its estimate on unmeasured locations with a desired accuracy. Thus to obtain an optimal monitoring network having 25 IFP wells and minimize the IFW wells such that the kriging estimation of water levels provide standard deviation of the estimation error not more than 8 m (against the average standard deviation of 12 m of the water level data) in the entire area. Through a special procedure as described through equations 9-11 above the IFW wells were removed one by one and the impact with the above constraints were studied. Finally a network with 25 IFP wells and 15 IFW wells have been evolved for monitoring the water levels every month. The contour map of the standard deviation of the estimation errors (k) from a network of 56 wells as well as from a network of 40 wells are shown in Figs. 5 and 6 respectively using a suitable kriging method. It is very clear that using the optimized monitoring network it is still possible to maintain the magnitudes of k. Fig. 7 shows the location of wells with the wells (o) for optimal monitoring network. The constrained optimization of the monitoring network with only 40 wells will ensure that all the wells are measured in the shortest possible time every month. Also that the revised network consists all the 25 wells without pumping and one has to be only careful for monitoring the 15 private wells fitted with pumps for irrigation. This provides a hydrogeologist much ease for an accurate water level measurement. The revised network 20

Fig-4 Maheshwaram Watershed

21

WELL LOCATION Counter Interval = 0.5 m

Fig-5 Contour map of sigma (in m) with 56 measurements points (500m*500m)

IFW LOCATION IFP LOCATION Contour Interval = 0.5 m

Fig-6 Contour map of sigma (in m) with 40 measurement points (500m*500m)

22

IFP-1

IFP-23 IFP-22 IFP-7 IFP-2 IFP-21

IFP-8
Northing in metres

IFP-9

IFP-24 IFP-3 IFP-6 IFP-4 IFP-20

IFP-10

IFP-5 IFP-13 IFP-18 IFP-14 IFP-15 IFP-11 IFP-17 IFP-12 IFP-16 IFP-19 IFP-25

Easting in metres

Fig-7 Optimized well location

will also provide almost same accuracy in estimation that would have been obtained from the network of 56 measurement wells.

4.5

Conclusion

The field heterogeneity of groundwater basins is often inextricable and very difficult to analyse with deterministic methods. Another advantage of using stochastic approach is that it provides the variance of the estimation error together with the estimated values. Of course, there are many advantages of these methods particularly in Groundwater modelling: The closer the values of the aquifer parameters to reality, the faster will be the model calibration. Better estimated values (with lower estimation variance) are initially assigned to the nodes of aquifer model using geostatistical estimation. 23

An assumption is made in Aquifer Modelling that a single value of system parameter represents the entire mesh (Of course, very small). Averaging over a block in two or three dimension can be obtained through block estimation. An optimal mesh size and number of nodes in discretizing aquifer system, can be obtained and best location of new control points can be predicted. A confidence interval given by the standard deviation of the estimation error provides a useful guide to T modification at each mesh and to check that the calculated heads fall inside the confidence interval of the observed heads. A performance analysis of the calibrated model can be achieved to decide the best calibrated model using variance of the estimation error which can be used for prediction.

The main handicap in applying geostatistical techniques to hydrogeological problems is the scarcity and scarcity of observed data. However, a few modifications and additions to the existing techniques permit to utilise hydrogeological data successfully in prediction of aquifer parameters. One such modification in kriging techniques is called "kriging with an external drift". This technique has been found quite useful in arriving at the estimation of hydrogeologic parameters. Cross-validation, though very cumbersome and not useful when data are numerous as in case of mining, it is much more useful and almost necessary when there are less data as in case of hydrogeological parameters. A large number of works have been reported particularly using geostatistical methods in optimal data collection network design. However, very few have found application in real field. It is therefore, useful to analyse and discuss the problems of their application. A number of ambiguities have been found in the methods so far applied; some of them are quite severe. Since most of the network design are based on the reduction of kriging variance which does not depend on the measured value of the parameter at a newly decided location, a common ambiguity is about the maximum value allowed of the variance or the standard deviation of the estimation error (say a threshold). In the absence of an objective function directly involving the location of measurement points, it is difficult to minimize the variance of the estimation error ( 2k) Either this value is arbitrarily chosen or optimization of a data collection network may be terminated if the corresponding change in the 2k is neglegible. The second point is where to calculate the 2k either on a point in the considered area or on the area itself. One can take central point of the area to calculate and reduce the variance of the estimation error. However this central point has generally no significance with the objective of the study. Also the method of calculating estimation variance at the central point of the area has the ambiguity that any worked out network can be valid for a much larger area provided the centre remains unchanged. In addition k (x0) will obviously be smallest if all the measurement points are located closest to the central point. These are quite irrational. The method of calculating estimation variance over an area of the system does not have the above ambiguity. The studied examples show that as the area increases, k increases provided the centre of the area remains unchanged. However, when the area was changed with a new centre and a different origin, sometimes a reduction in the k was obtained depending upon distribution of the density of the network. Therefore, it is concluded that a reduction in k calculated over the initial area may not be sufficient and in addition a 24

few more areas by shifting the size and origin must be checked to obtain an adequate network. In all the three examples when a few measurement points were added, the revised k has shown reduction as expected but the nature of ambiguities were retained. It is finally recommended that the method of estimating kriging variance on a block/area may be used in data collection network design but with the care as described above. Of course, the better way of analysing and designing a network is to discretize the area into a number of blocks and design a network by reducing the estimation variance on an average basis. This procedure could be repeated by reducing the size of discretized blocks until there is no change in the average statistics of the estimation variance. The water levels are being monitored through a network of about 56 bore wells out of which 25 have been specially drilled to observe comparatively undisturbed water table and the other 31 bore wells are selected based on the drainage pattern and intervals etc. from the existing private irrigation wells. The water level measurements have been carried out on monthly basis for a period of almost one hydrological cycle. In addition, to investigate the water quality of the aquifer, samples from about 60 wells were taken and water were analyzed mainly for fluoride content. It was necessary to use the wells fitted with pump for sampling water for chemical analysis. Also to study the variation of fluoride in time, it was decided to optimize the monitoring network and if possible reduce it so that frequent monitoring could feasibily be done.

25

5.0 MODELING THE GROUNDWATER FLOW 5.1 Introduction

Maheshwaram watershed is modelled with MARTHE software developed at BRGM (Thiery, 1993a and 1993b). MARTHE is a hydrodynamic modelling code, in transient regime with three-dimensional and/or multi-layer flow in aquifer. The resolution method uses finite differences with, in the current project, a regular rectangular grid and offers the possibility of having a free surface in a mesh of any layer. More details on this software are given in Appendix A. Aquifer in Maheshwaram watershed is taken as a two-layer aquifer: the upper part is consist of weathered rocks (porous media) and overlays the lower weathered-fissured granite layer (equivalent porous media). The aquifer is unconfined and the code allows the weathered layer to dry out. MARTHE code is used with its coupled climatic balance model (GARDENIA ; Thiry et al., 1993 ; Thiry, 2001, 1991 and 1988). The groundwater flow in Maheshwaram watershed is simulated in transient regime in order to represent the piezometric variations observed in the wells in the studied area from January 2001 to July 2003. The first part of this chapter describes the various input parameters and the fitting parameters of the model. Most of the parameters were subjected to sensitivity tests in order to ascertain their relative influence in the model. The second part of the chapter presents the results of the final simulation.

5.2

Conceptualisation of the hydogeological system

5.2.1 Time-steps
The time-steps of the hydrodynamic calculation in transient flow are of 14 days during the dry season and one week during the rainy season (they are daily when it rains). MARTHE code uses GARDENIA to calculate the hydroclimatic balance; this balance is achieved with a weekly time-step (during the dry season) and a daily one (during rainy days) throughout the year. 5.2.2 Geometry The studied aquifer is represented by a two-layer aquifer flow system. The upper layer is located in the weathered rocks, the lower one in the weathered-fissured granite represented as an equivalent porous media. Each layer was divided into 5272 square meshes of 100 metres side (Fig 8). Layer 1 is unconfined and layer 2 may become so when layer 1 is dried.

26

Figure 8. Grid of the watershed, layer 1 is similar to layer 2. Since all the locations were mapped and have coordinates in WGS84-UTM projection system, a shift of the origin was applied as follows to avoid the large number on axis scale. Xmodel = XMNT 223000 Ymodel = YMNT 1892000 where the values are in metres. 5.2.3 Boundary conditions The topographical limits of the watershed are taken as the groundwater divides (no-flow boundaries), except at the northern limit were a non perennial stream can evacuate the surface water of the watershed. At this location, the hydraulic heads are prescribed and have been set at the average of 2000-2002 field values. The streams are not perennial, they are filled only during high rainfall events due to runoff. That is why, in a first approximation, the hydrographic network is assume to be negligible for recharging the aquifer. However, the groundwater is allowed to overflow at the river location during high flood as base flow.

5.2.4 Topography The DEM values have been averaged to fit the grid 100 metres by 100 metres (Fig. 9). For the meshes containing the IFP wells where some precise elevation indications have been estimated through DGPS (Lebert, 2001), the topographic values have been taken as the 27

DGPS ones. For the meshes containing the other IFP wells, the topography values have been taken as the nearest local values of the DEM (interpolated every 10 metres).

Figure 9. Topography of Maheshwaram watershed.

5.2.5 Aquifer geometry The geometry of weathered layer (Fig. 10) has been input from a linear kriging of the layer thickness obtained through the 25 existing lithologs and the VES interpretations (Krishnamurthy et al., 2000). The geometry of the weathered-fissured granite layer (Fig. 10) has been deduced jointly by the total depth of the 900 inventoried farmer wells of the watershed after removing the wells with a total depth of more than 70 metres as well as the result of VES.

28

Figure 10. Aquifer bottom of layer 1 and 2 (ASL, in metres).

5.3

Hydraulic conductivity

No hydraulic tests could be carried out in the first layer as during the project period the rainfall were constantly less than normal and there was no water in the weathered zone. Thus the hydraulic the hydraulic parameters for that layer was assumed from the literatures. The hydraulic conductivities chosen for the weathered layer have been selected following two assumptions :

29

 The hydraulic conductivities values have to be in the range of the ones found in the literature for the same type of geology; The hydraulic conductivity variability in the weathered rocks is based on the conceptual model of the hydrogeological functioning of weathered-rock/hard rock aquifers in Africa proposed by Chilton et al. (1995). This model assumes that, in the general case, the weathered rocks have a weak hydraulic conductivity which increases with depth. In Maheshwaram watershed, if it is assumed that the thickness of the weathered rock layer is linked to its degree of erosion, the thin, eroded weathered rock layer is, in reality, the base of the weathered horizon and, consequently, it is composed of coarser grains than the thick layers where the weathered rocks show the entire profile. Thus, the thicker the weathered-rock layer, the weaker their average hydraulic conductivity. Figure 11 shows the distribution of the calibrated hydraulic conductivities for the weathered layer. They are ranging between 8.10-7 m.s-1 and 5.10 -6 m.s-1.

Figure 11. Hydraulic conductivities in the weathered layer, after calibration. In the weathered-fissured granite layer, the transmissivities calculated based on the results of a number of aquifer tests carried out. But the high variability observed in the field and the difficulties to match the simulated heads with the field ones lead to further calibrate the hydraulic conductivities manually. Fig. 12 shows the comparison of the hydraulic tests results and the values of hydraulic conductivities used for simulations.

30

1.00E+00 Hydraulic Conductivities, m/s 1.00E-01 1.00E-02 1.00E-03 1.00E-04 1.00E-05 1.00E-06 1.00E-07 1.00E-08 Slugtests Long Duration Pumping Tests K Marthe Layer 1 K Marthe Layer 2 ifp 1 ifp 2 ifp 3 ifp 4 ifp 5 ifp 6 ifp 7 ifp 8 ifp 9 ifp 10 ifp 11 ifp 12 ifp 13 ifp 14 ifp 15 ifp 16 ifp 17 ifp 18 ifp 19 ifp 20 ifp 21 ifp 22 ifp 23 ifp 24 ifp 25

Figure 12. Hydraulic conductivities (m/s) : Comparison between interpretation of hydraulic tests (Charlier, 2002) and values used in the simulations. Fig 12 shows that most of the data are not matching with the interpretation of hydraulic tests. However, the average hydraulic conductivity obtained by the simulations has been compared with the equivalent horizontal hydraulic conductivity calculated from FRACAS model (Bruel et al., 2002). Table 2 shows the comparison and suggests that the model fits well with the discrete fracture network study results obtained for the weathered-fractured granite. Equivalent horizontal Mean hydraulic conductivity hydraulic conductivity value, calibrated by Marthe from Fracas (m/s) (m/s) Layer 1 2.61.10 -6 No interpretation -6 Layer 2 7.67.10 7.70.10-6 Table 2. Hydraulic conductivities : comparison with the discrete fracture network study interpretations. The hydraulic conductivities in the weathered-fissured granite layer are shown by Fig. 13. They vary greatly from one place to another because of the heterogeneity of the rocks : 1.10 -7 to 3.10 -5 m.s-1.

31

Figure 13. Hydraulic conductivities in the weathered-fissured granite layer, after calibration.

Figure 14. Impervious vertical boundaries input in the model. In the weathered-fissured granite layer, linear heterogeneities (impervious vertical barriers) had to be introduced into the model because this was the only way to take into account the observed piezometric data. These heterogeneities were attributed to the dykes that are cropping out in the studied area (Wyns, personal communication, 2002), to a South-North quartz reef crossing the watershed and to some assumed extension of dykes (Fig. 14). 32

5.4

Specific yield and storativity coefficients

The specific yield coefficient is taken to 2.4 % in the weathered rocks. This value is an average of the specific yields observed in the weathered rocks of similar watersheds (Rangarajan et al., 2001). The specific yield of the weathered rocks has also been used for the calculation of the preferential recharge (see section 2.5.2.). This value lies within the orders of magnitude given by the PMR measurements carried out in the watershed (Legchenko et al., 1999). In the fissured granite, this coefficient is assumed constant at a value of 0.01 which is near from the specific yield value deduced from the discrete fracture network interpretation (0.008, Bruel et al.; 2002). Some sensitivity tests have been realised by decreasing the specific yield in layer 2 of one order of magnitude. This has lead to lower the simulated water levels because: The storage became lower in layer 2. This effect was amplified by the fact that layer 1 is almost desaturated and can not play its role of reservoir which could have hidden the influence of the second layer specific yield as it has been seen in other watersheds (Engerrand, 2002). Storativity of the weathered layer plays a role only for the calculation of flows when this layer gets saturated in pseuso-ZNS scheme (see section mah100.ecp, Appendix B). Its value has been taken at 8.10-5. For layer 2, the storativity has been taken at 1.10-5.

5.5

Recharge

5.5.1 The natural recharge

The natural recharge has been calculated on the basis of the tritium injection tracer tests that have been carried out in 1999 and in 2000. The values have been estimated by assuming that between end of July and November 1999, the recharge by piston flow was 22.2 mm (Rangarajan et al., 2001) and that during all monsoon 2000, the recharge by piston flow was 42 mm (Rangarajan, personal communication, 2002). GARDENIA code elaborated in BRGM (Thiry et al., 1991, 1993, 2001 and Thiry, 1988) has been used at a daily time-step to calculate the balance between: The rainfall (daily data from a raingauge located in Maheshwaram village). The potential evapotranspiration, calculated from the evaporation in a Pan A and multiplied by a factor 0.9 (Monteith et al., 1989). The data are daily when they are existing at Maheshwaram village or weekly and coming from an average of six years of survey in CRIDA farm (CRIDA, 1991-2000) when they are not available for Maheshwaram. The run-off. For the Musi basin, Central Water Commission calculates a runoff/Rainfall ratio of 11.2 % for the period 1989-1994 (Marechal, personal communication, 2001). The available water capacity of the soil. The most common soils in the area have a water available capacity ranging from 75 to 110 mm (CRIDA, 1990). The parameters (run-off and water available capacity) have been adjusted for 1999 and 2000 to get 22.2 mm and 42 mm of recharge respectively for the respective periods. The same parameters have, then, been used to estimate the recharge for the years 2001 and 2002. Table 3 presents the selected model, Rumax represents the available water capacity parameter of the Gardenia code and Ruiper controls the runoff (Thiery et al., 1991). 33

Year
1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999

Rainfall (mm)
515 1020 770 772 1017 831 463 466 758 873 791 521 1134 582

Recharge (mm)
0 112 113 89 69 58 3 0 76 59 63 3 173 26

Runoff (mm)
0 118 167 156 70 47 0 0 110 39 42 0 312 6

21 mm for the rainy season tritium* = 22.2 mm

2000 630 39 mm for the rainy season tritium * = 42 mm 83 22 60 26

2001 2002 Mean

886 584 742

188 5 76

CWC Mean rainfall 1989-1994 816 Maheswaram Mean Rainfall 1986-2002 742

Runoff 91 Mean Runoff 76

Runoff/Rainfall 11.2 % Runoff/Rainfall 10.2 %

Table 3. Annual water balance calculated by GARDENIA code, Maheshwaram watershed, Rumax = 90 mm, Ruiper = 40 mm, Tperc=30 days.
* recharge estimated from the tritium injection experiments.

5.5.2 The preferential recharge

The preferential recharge has been calculated from the groundwater level rise in the weathered rocks during the tritium experiments of 1999. In 2000, most of the dugwells located near the tritium injection points were dried. Now, the preferential recharge is calculated with the specific yield value of the weathered layer that is known (Rangarajan et al., 2001) and the groundwater level fluctuations within this layer. That is why the preferential recharge has been calculated only for the year 1999. In 1999, the water level rose of 1.5 m in average, during the time of the experiment, near the injection points. The specific yield of the weathered layer is assumed to be 2.4 % (Rangarajan et al., 2001). That means that the recharge was around 36 mm in 1999 and not 22.2 mm. The extra recharge that takes into account whatever is not measured by a piston flow recharge model is attributed to the preferential recharge. The recharge measured by tritium injection or calculated by Gardenia code is called natural recharge. If the ratios preferential recharge on total recharge is assumed constant every year, (3622.2)/350=4 % of the total rainfall recharge the aquifer as preferential recharge. Table 4 shows the calculated natural recharge and preferential recharge for the years 2000, 2001 and 2002. 34

2000 2001 2002

Natural recharge in mm Preferential recharge in mm Total recharge in mm 42 25 67 83 35 118 22 23 45 Table 4. Annual recharge calculated for 2000, 2001 and 2002

5.5.3 The discharge rates

The discharge rates due to pumping are evaluated from the well inventory given by IFCGR. For the years 2000, 2001 and 2002, all the wells aged upto 2000, 2001 and 2002 respectively have been taken into account. The wells are assumed to be pumped in layer 2 only (Fig. 15).

Figure 15. Farmer well locations The return flow from the withdrawals is taking place in layer 1. It has been deduced from the pumping from : APGWD (personal communication, 2000) report on Maheshwaram landuse, APGWD (1977) report on hydrologic parameters of groundwater recharge in AP. In the ideal case, 1200 mm (Vmmrice) of water is required for one culture of rice and 350 mm (Vmmother) of water per culture is required for other crops (APGWD, personal communication, 2000). There are in average two crops for the rice and two crops for the vegetables in one year. It is assumed that it is raining 700 mm per year. In 2000, the cultivated areas for rice and other vegetables are respectively known for winter and monsoon (Sricewinter, Sricemonsoon, Sotherwinter, Sothermonsoon). We have: Vtot (m3) = Vrice (m3) + Vother (m3) Vtot (m3) = Vmmrice (mm) / 1000 * Srice (m2) + Vmmother (mm) / 1000 * Sother (m2) Vtot (m3) = [(Vmmricemonsoon (mm) 700) * Sricemonsoon (m2) + (Vmmricewinter (mm) * Sricewinter (m2)] / 1000 + [(Vmmothermonsoon (mm) 700) * Sothermonsoon (m2) + Vmmotherwinter (mm) * Sotherwinter (m2)] / 1000 If Vmmricemonsoon (mm) 700 < 0, we take Vmmricemonsoon (mm) 700 = 0 and if Vmmothermonsoon (mm) 700 < 0, we take Vmmothermonsoon (mm) 700 = 0 Vtot (m3) = (1200 700) * 362.56 * 10 4 + 1200 * 217.38 * 104) / 1000 + (0 + 350 * 198.10 * 104) / 1000 Vtot (m3) = 5 114 650 m3/yr 35

With : Vrice (m3) = 4 421 300 m3/yr and Vother (m3) = 693 350 m3/yr i.e. 86 % of the water is used for the rice and 14 % of the water is used for the other crops. In APGWD report (1977), experiments on the return flow from the rice in a nearby watershed showed that 55 to 88 % of the irrigation water was going back to the groundwater. After calibration, 60 % of return flow from the rice and 20 % for the other crops have been taken. In the model we will not distinguish the difference in location between the rice and the other crops, because it is not known. The average return flow of one crop (rice/other) is then taken as : 0.86* 60 + 0.14 * 20 = 54.4 55 % of return flow from the withdrawals.

5.6

Results and discussion

5.6.1 Fitting criteria

The final model is presented in Appendix C, section C10. The fitting is done: - on the hydraulic conductivities of the two layers; - on the return flow from withdrawals. The fitting criteria is based on: - probable orders of magnitude of the fitting parameters; - the similarity between the heads observed in the IFP wells in the field and the simulated ones. 5.6.2 Balance

Parameters 1/1/2001-31/12/2001 1/1/2002-31/12/2002 Recharge (m3) 4455.498 1265.701 Return flow (m3) 5003.662 3365.154 Outlet from prescribed -41.5303 -59.7632 heads (m3) Withdrawals (m3) (% -7088.47 -6899.92 of the water that can not be pumped due to desaturation) Groundwater overflow 20.20823 -9.83368 (m3) Storage variation (m3) 6537.535 1626.574 Balance deviation (%) 0.15936 0.115841

1/1/2003-31/7/2003 110.5803 1856.686 -22.5355 -4758.7

-2.28657 90.81401 0.044672

Table 5a. Yearly water balance

36

Balance Month Recharge Return Outlet from Withdrawals Groundwater Storage (m3) overflow (m3) variation deviation (m3) flow prescribed (%) (m3 ) (m3 ) heads (m3)
Jan-01 Feb-01 Mar-01 Apr-01 May-01 Jun-01 Jul-01 Aug-01 Sep-01 Oct-01 Nov-01 Dec-01 Jan-02 Feb-02 Mar-02 Apr-02 May-02 Jun-02 Jul-02 Aug-02 Sep-02 Oct-02 Nov-02 Dec-02 Jan-03 Feb-03 Mar-03 Apr-03 May-03 Jun-03 Jul-03 0 0 0 0 0 0 0 0 3276.77 45442.54 1690.70 710.68 160.91 177.18 97.24 54.29 35.09 33.71 17.96 8390.91 4629.59 1285.11 160.11 81.98 42.44 22.09 35.64 18.72 2.56 5.60 1086.57 0.00 0.00 850.62 5449.65 223.74 6426.29 1382.11 10168.65 17683.90 16381.91 0.00 0.00 0.00 0.00 791.08 351.59 1616.18 5488.72 5372.74 14545.64 3227.60 6812.06 0.00 0.00 0.00 0.00 2855.16 3595.81 0.00 3676.19 11493.90 -10.50 -5.61 -11.25 -27.75 -5.65 -33.60 -14.71 -55.05 -69.12 -197.13 -32.71 28.51 -23.76 -23.25 -32.82 -26.38 -38.61 -80.03 -77.66 -173.84 -109.90 -77.25 -17.48 -15.61 -14.01 -12.44 -46.81 -43.62 -9.10 -48.56 -100.30 -1993.84 -2066.72 -4676.13 -9672.36 -2642.77 -11857.79 -7036.56 -18220.69 -13661.12 -9038.52 -2063.45 2064.73 -2414.34 -2299.07 -3716.54 -3222.66 -5207.38 -10781.04 -11133.78 -18782.19 -11638.80 -8091.21 -2316.69 -2321.84 -2306.43 -2296.01 -8911.57 -8857.92 -2339.99 -12617.02 -20371.97 0 0 0 0 0 0 0 0 11.02 231.52 -7.92 5.83 -3.58 -3.72 -3.93 -2.16 -2.80 -6.99 -6.82 -30.77 -21.78 -22.26 -2.10 -1.23 -0.91 -0.75 -4.15 -4.44 -0.54 -4.44 -12.13 -892.79 -921.80 -1219.41 1260.26 -935.57 1325.97 -1673.72 2312.96 15636.45 59169.91 730.70 241.85 -933.07 -865.94 -780.25 -1041.25 -665.60 778.25 598.50 14990.03 2659.63 4658.42 -857.98 -928.22 -954.09 -955.93 -806.42 14.79 -927.93 -1164.82 4763.52 1.64E+03 4.68E-05 2.78E-03 1.45E-01 6.12E-05 1.99E-01 1.06E-02 1.81E-01 7.85E-01 5.23E-01 3.66E-03 -4.45E-03 2.00E-03 2.38E-03 3.47E-02 1.56E-03 1.50E-02 8.92E-02 2.33E-01 5.25E-01 2.20E-02 2.78E-01 3.84E-05 4.43E-05 4.00E-05 4.55E-05 8.85E-02 1.04E-01 4.62E-05 9.90E-02 2.38E-01

Table 5b. monthly water balance In 2000, the balance (storage variation) is negative (Table 5). The withdrawals are greater than the recharge and the stock is positive. The pumping rates are operational at 99.9 %. Outlet from prescribed heads are 1.3 % of the recharge which is negligible. The overflows are nil. In 2001, the balance is positive. The recharge is greater than the withdrawals and the stock is positive. The pumping rate are operational at 99.7 %. Outlet discharge from the prescribed heads are 1.8 % of the recharge which is negligible. The overflows are 0.2 % of the recharge which is also negligible. The aquifer overflows are located near the river meshes that seems logic. In 2002, the balance is highly negative. The recharge is less compare to the other years and the withdrawals are increasing. The pumping rates are operational at 97.6 %. But the return flow is calculated on the basis of the maximum demand and not on the basis of the real withdrawals; in that case, because the quantity of water that cannot be pumped is not negligible in regards to the water that is demanded, the prescribed return flow is slightly 37

over-estimated and the balance is optimistic. In fact the return flows range between : 6 720 000 m3 and 6 560 000 m3. Outlet discharge from prescribed heads are 3.1 % of the recharge. The overflows are 0.2 % of the recharge which is negligible and are still located near the river meshes.

5.6.3 Discussion on the simulated heads

January 2001 Simulated water levels in m (asl) 660 640 620 600 580 580

600

620

640

660

Observed water levels in m (asl) January 2002 Simulated water levels in m (asl) 660 640 620 600 580 580

600

620

640

660

Observed water levels in m (asl)

Figure 16. Comparison between simulated heads and calculated one for Jamuary 2001 and January 2002 Fig. 16 shows the comparison between the simulated heads and the field one for January 2001 and January 2002. Most of the water levels are well simulated. Some wells show lower simulated water levels than the observations. Figure 10 presents the comparison between the simulated groundwater levels and the field data for all the IFP well located within the watershed.

38

Hydraulic heads in m (asl)

IFP1
Hydraulic heads in m (asl)

IFP9 618 613 608 603 598 oct-00

Hydraulic heads in m (asl)

IFP20 620 615 610 605 600 oct-00

600 595 590 585 580 oct-00 avr-01 nov-01 mai-02 dc-02

avr-01 nov-01 mai-02 dc-02

avr-01 nov-01 mai-02 dc-02

Hydraulic heads in m (asl)

Hydraulic heads in m (asl)

Hydraulic heads in m (asl)

IFP2 620 615 610 605 600 oct-00 avr-01 nov-01 mai-02 dc-02 636 631 626 621 616 oct-00

IFP11

IFP21 620 615 610 605 600 oct-00 avr-01 nov-01 mai-02 dc-02

avr-01 nov-01 mai-02 dc-02

Hydraulic heads in m (asl)

IFP3 626 621 616 611 606 oct-00

Hydraulic heads in m (asl)

IFP12 652 647 642 637 632 oct-00

Hydraulic heads in m (asl)

IFP22 620 615 610 605 600 oct-00

avr-01 nov-01 mai-02 dc-02

avr-01 nov-01 mai-02 dc-02

avr-01 nov-01 mai-02 dc-02

Hydraulic heads in m (asl)

Hydraulic heads in m (asl)

Hydraulic heads in m (asl)

IFP4 622 617 612 607 602 oct-00 avr-01 nov-01 mai-02 dc-02 630 625 620 615 610 oct-00

IFP13

IFP24 630 625 620 615 610 oct-00

avr-01 nov-01 mai-02 dc-02

avr-01 nov-01 mai-02 dc-02

Hydraulic heads in m (asl)

620 615 610 605 600 oct-00 avr-01 nov-01 mai-02 dc-02

Hydraulic heads in m (asl)

Hydraulic heads in m (asl)

IFP5 632 627 622 617 612 oct-00

IFP14

IFP25 640 635 630 625 620 oct-00 avr-01 nov-01 mai-02 dc-02

avr-01 nov-01 mai-02 dc-02

Hydraulic heads in m (asl)

Hydraulic heads in m (asl)

IFP6 620 615 610 605 600 oct-00 avr-01 nov-01 mai-02 dc-02

IFP15 640 635 630 625 620 oct-00 avr-01 nov-01 mai-02 dc-02
Hydraulic heads in m (asl)

IFP8 610 605 600 595 590 oct-00 avr-01 nov-01 mai-02 dc-02

Hydraulic heads in m (asl)

IFP7
Hydraulic heads in m (asl)

IFP18 640 635 630 625 620 oct-00 avr-01 nov-01 mai-02 dc-02
Hydraulic heads in m (asl)

IFP19 641 636 631 626 621 oct-00 avr-01 nov-01 mai-02 dc-02

620 615 610 605 600 oct-00 avr-01 nov-01 mai-02 dc-02

Figure 17. Comparison between the simulated groundwater levels (squared points) and the observed ones (cross points). Figure 18 shows that most of the wells are correctly simulated. The wells located at the western part of the watershed show lower groundwater levels during the simulations compared to the field observations. Some tests have been carried out by reducing the pumping rate values, i.e. assuming that the farmers were not pumping during summer; the results, show that the water levels are then, better simulated in this area. Varying the 39

pumping rates with more accurate data may lead to a better fitting. In the North-Eastern part of the watershed, the water levels are showing a more complicate figure: within a few hundred of metres, the simulated water levels are either too high, or too low or else fitting with the field observations. This is an illustration of the complexity of the system; these variations can come from a lack of accuracy in the pumping or from geological local heterogeneities; this area is highly pumped and the fitting problem can also come from the proximity of the farmer well from the observation wells: the observation wells can be influenced by the pumping rates at a very small scale that is not adapted with the mesh size of the model. In this case, a smaller mesh size at this location could lead to a better result and can be an orientation of development for the model. Even after varying the hydraulic conductivities and storativities near the wells IFP1 and IFP19, the simulated water levels in these wells show a smaller amplitude in the simulations than in the reality. Now, they are the only wells in the watershed to show so high amplitude in the water level fluctuations and these two wells are also aligning with the quartz reef that is crossing the watershed from North to South. This high amplitude in the water level fluctuation may come from a low storativity induced by the presence of the quartz reef that is only few ten of metres wide and cannot be represented at the model mesh scale. Here also a smaller mesh size near these wells could lead to a better result.

Figure 18. Results of the simulation.

40

10000 9000
South-Northern direction in metres

10000 9000 8000 7000 6000 5000 4000 3000 2000

8000 7000 6000 5000 4000 3000 2000 1000

1000 10000 9000

South-Northern direction in metres

1 November 2001, layer 1

3000 5000 7000

1000

655 645 635 625 615 605 595 585 575 565 555

1 November 2001, layer 2

1000 3000 5000 7000 9000

9000 10000 9000 8000 7000 6000 5000 4000 3000 2000

West-Eastern direction in metres

8000 7000 6000 5000 4000 3000 2000 1000

1000 10000 9000

South-Northern direction in metres

3 Mars 2002, layer 1

3000 5000

1000 9000 10000 9000 8000 7000 6000 5000 4000 3000 2000 1000

655 645 635 625 615 605 595 585 575 565 555

7000

3 Mars 2002, layer 2

3000 5000 7000

9000

West-Eastern direction in metres

655 645 635 625 615 605 595 585 575 565 555
1000

8000 7000 6000 5000 4000 3000 2000 1000

1000

15 October 2002, layer 1

3000 5000 7000

1000

9000

15 October 2002, layer 2

3000 5000 7000

9000

West-Eastern direction in metres

Figure 19. Simulated water levels for layer 1 and 2. Fig. 19 presents some simulated water level maps for both the layers. There are not much differences in the water levels from layer 1 to layer 2 except in the zone of high groundwater withdrawals where the layer 2 shows lower hydraulic heads than layer one.

41

5.6.4 Discussion on the water reservoir capacities of the watershed and the probable evolution of its groundwater resources.
The storage capacity of each layer has been calculated (Table 6) as follow:

H
i =1

S Sy

with, n , the total number of meshes in the layer, Hi, the height of the mesh, in m, S , the surface of the mesh (100 m 100 m), Sy, the specific yield of the layer (0.024 for layer 1 and 0.01 for layer 2). In the same way, the water storage of each layer has also been calculated for each year, with the same formula by replacing Hi by the hydraulic head of January calculated in the layer for each mesh. Layer 1 Layer 2 Water storage capacity (m3) 14 491 474 19 803 453 Minimal water storage (m3) 724 574 1 980 345 Storage in 01/2001 (m3) 1 182 864 (8%) 16 561 160 (84 %) Storage in 01/2002 (m3) 1 896 799 (13%) 16 944 939 (86%) 3 Storage in 01/2003 (m ) 967 094 (7%) 14 837 332 (75%) Table 6. Water storage in both layers. The minimal water storage is a parameter that the user can prescribe in the model to help the convergence when the layers get desaturated to saturated. It is also input through a limitation for the withdrawals in the layer 2. The term expressed in % in Table 5 is the real storage of the layer compared to the total water storage capacity. It appears that the layer 1 is almost totally desaturated. Around 80 % of layer 2 is filled in the month of January. In 2002, the water demand reach 12 200 000 m3 which represents around 65 % of the total water storage available at the beginning of the year in both layers. Table 7 presents the main parameters that controls the balance. Rainfall (mm) 2000 630 2001 886 2002 584 Water Water that can not Recharge (m3) Return Flow Groundwater demand (m3) be pumped (m3) (m3) storage (m3) -9 479 540 10 399 3 521 310 5 213 780 17 744 024 -11 324 400 30 901 6 212 840 6 228 460 18 841 738 -12 218 600 291 152 2 372 280 6 720 260 15 804 426 Table 7. Main parameters controlling the balance.

Table 7 shows that even if the water demand is much higher than the recharge (upto 5 times higher in 2002), the balance is not alarming because of the high value of return flow (55% of the withdrawals). Another parameter that is controlling the state of the aquifer is the water that can not be pumped due to the desaturation of the second layer in some meshes of the watershed. This value may regulate the balance at a long term. Some predictions tests have been carried out until the year 2023 assuming: - that the recharge was set at a constant value equivalent to the average of the recharge calculated between 1986 and 2002; - that the recharge was varying as it was varying between the years 1986 and 2002. In both scenarios, the water demand is set constant and equal to the 2002 water demand. 42

Groundwater storage
Volume of water in m3 2.0E+07 1.5E+07 1.0E+07 5.0E+06 0.0E+00 2000

2005

2010

2015

2020

2025

Water Demand that can not been withdrawn

Volume of water in m3 2.5E+06 2.0E+06 1.5E+06 1.0E+06 5.0E+05 0.0E+00 1995 2000 2005 2010 2015 2020 2025

Figure 20. Prediction : evolution of the groundwater storage and the water demand that can not be pumped. Squared points, the recharge varies; line, the recharge is set constant. Figure 20 shows that the groundwater flow reaches a steady state regime. This regime is reach when the Recharge = (Water demand Water demand that can not be pumped) Return flow. In 2023, 69 farmers wells will not be able to provide the water demand due to the desaturation of the aquifer. If the Water demand increases, the scheme is different. Fig 21 shows an extrapolation of the water demand over the next 20 years. This figure is just an example. It proposes a logarithmic increase in the water demand justified by the idea that the yearly increase of the well implantation should decrease due to a saturation of the land use. With this figure, the number of farmer wells located in the watershed should increase from 709 to 1004 wells in January 2023.
Number of farmer wells extrapolation real water demand

Groundwater demand, m3/yr

2.0E+07 1.6E+07 1.2E+07 8.0E+06 4.0E+06

1150 950 750 550 350 150

0.0E+00 -50 01/2001 01/2004 01/2007 01/2010 01/2013 01/2016 01/2019 01/2022

Figure 21. Estimation of the increase in the water demand for the next 20 years. Under this new condition, predictions have been carried out. Fig 22 shows that the global groundwater storage decrease rapidly and as a consequence, the water demand that can not be pumped due to the desaturation of the aquifer increased (Fig 23) and touch 149 wells in 2023. At a more long term (few years more), the groundwater storage will be 43

directly equivalent to the recharge. The monsoon will allow the aquifer to recharge but the irrigation will, within one year or less empty the aquifer till the next season. Under this condition, the farmers will be more highly dependant of dry years than nowadays and may not be able to carry on two nor even one cultivation in these years. This will lead to an economic instability and a difficult situation to handle for the farmers.
Groundwater storage Volume of water in m3
2.0E+07 1.5E+07 1.0E+07 5.0E+06 0.0E+00 1999

2009

2019

2029

Water demand that can not be withdrawn

3.0E+06 2.5E+06 2.0E+06 1.5E+06 1.0E+06 5.0E+05 0.0E+00 1999 2004 2009 2014 2019 2024 2029

Figure 22. Prediction : evolution of the groundwater storage and the water demand that can not be pumped. Recharge is set constant and the water demand increase.

Volume of water in m3

Figure 23. State of the groundwater in 2023.

44

5.7

Conclusion

Groundwater flow in Maheshwaram watershed has been simulated taking into account the withdrawals for the agriculture. A two-layer aquifer model has been taken: the upper layer being the weathered rocks and the lower layer being the weathered-fissured rocks. For a majority of wells, hydraulic head simulations are in accordance with the water levels observed in IFP wells for the years 2001 and 2002. Hydraulic heads that are not in accordance with the field observations are located in areas where the withdrawals have to be checked. The average hydraulic conductivity and the specific yield of the weatheredfissured layer are in accordance with the one found with the DFN model (Bruel et al., 2002). The other hydraulic parameters are in accordance with the one found in the literature. The model has to be further improved with : - the input of a more accurate geometry of the layer; - a more accurate estimation of the recharge; - the verification of some withdrawal values. The Return Flow from the groundwater withdrawals may stay as an approximate parameter and can further be changed till its value stay in the range of values found in the literature. Some prediction tests have also been carried out showing that: - If the water demand stays equal to the one of 2002, the regime will reach a steady state flow where the mean water level in the wells will be as nowadays to 30 metres deeper. - If the water demand increases, the water levels in the watershed will decrease drastically everywhere. More and more wells will get dried and the groundwater storage will be entirely dependant on the recharge. This will lead to a difficult situation to handle for the farmers and an economic instability.

The later part of the simulation and calibration of the aquifer model till July 2003 is shown in Figure 24.

45

IFP1

IFP2 626

IFP3

600 595 590 585 580 575 570 Mar- Oct-00 Apr-01 Nov-01 May00 02 Dec- Jun-03 Jan-04 02

620 615 610 605 600 Mar-00 Oct-00 Apr-01 Nov-01 May02 Dec02 Jun-03 Jan-04

621 616 611 606 Mar-00 Oct-00 Apr-01 Nov-01 May-02 Dec-02 Jun-03 Jan-04

IFP4 622 617 612 607 602 Mar-00 Oct-00 Apr-01 Nov-01 May-02 Dec-02 Jun-03 Jan-04 620 610 600 590

IFP5

IFP6

620 610 600 590 Dec- Jun-03 Jan-04 02 580 Mar-00 Oct-00 Apr-01 Nov-01 May02 Dec02 Jun-03 Jan-04

580 Mar-00 Oct-00 Apr-01 Nov-01 May02

IFP7 610 620 615 610 605 600 595 590 Mar-00 Oct-00 Apr-01 Nov-01 May02 605 600 595 590 585 Dec02 Jun-03 Jan-04

IFP8

IFP9 620 615 610 605 600 595 590 585 580 Mar-00 Oct-00 Apr-01 Nov-01 May-02 Dec-02 Jun-03 Jan-04

580 Mar-00 Oct-00 Apr-01 Nov-01 May-02 Dec-02 Jun-03 Jan-04

IFP11 635 630 625 620 615 610 605 600 Mar-00 Oct-00 Apr-01 Nov-01 May-02 Dec-02 Jun-03 Jan-04 652 647 642 637

IFP12 630 625 620 615 632 Mar-00 Oct-00 Apr-01 Nov-01 May-02 Dec-02 Jun-03 Jan-04

IFP13

610 Mar-00 Oct-00 Apr-01 Nov-01 May-02 Dec-02 Jun-03 Jan-04

IFP14 632 640 627 622 617 612 Mar-00 Oct-00 Apr-01 Nov-01 May-02 Dec-02 Jun-03 Jan-04 635 630 625

IFP15

IFP18

620 Mar-00 Oct-00 Apr-01 Nov-01 May02

Dec02

Jun-03 Jan-04

640 635 630 625 620 615 610 Mar- Oct-00 Apr-01 Nov-01 May00 02

Dec- Jun-03 Jan-04 02

IFP19 640 635 630 625 620 615 Mar-00 Oct-00 Apr-01 Nov-01 May02 Dec- Jun-03 Jan-04 02

IFP20 640 635 630 625 620 615 610 605 600 Mar-00 Oct-00 Apr-01 Nov-01 May-02 Dec-02 Jun-03 Jan-04

IFP21 620 615 610 605 600 595 590 585 580 Mar-00 Oct-00 Apr-01 Nov-01 May-02 Dec-02 Jun-03 Jan-04

IFP22 620 615 610 605 600 Mar-00 Oct-00 Apr-01 Nov-01 May02
630 625 620 615

IFP24

IFP25 640 635 630 625 620 Mar- Oct-00 Apr-01 Nov00 01

Dec- Jun-03 Jan-04 02

610 Mar-00 Oct-00 Apr-01 Nov-01 May-02 Dec-02 Jun-03 Jan-04

May02

Dec02

Jun03

Jan04

Figure 24 Simulated and observed water levels for the simulation period

46

6.0 WATER CONSERVATION AND ARTIFICIAL RECHARGE

6.1 No-Cost Water Conservation Technology For Recharging Over-

Exploited Aquifers
A demonstrative experiment by the NGRI through the dry dug wells at Maheshwaram Watershed under the Indo-French Collaborative project

47

WATER CONSERVATION AND RECHARGE ENHANCEMENT ARE ESSENTIAL STEPS IN MANAGING WATER SOURCES IN THE GROWING DEMANDS Situations are already alarming: 1. Most of the dug wells remain dry through out the year. 2. The pumps in the bore-wells get and pump water for 45 seconds and run without water for the next 60 seconds. 3. Climate change is taking place. Both time and space variability of rainfall is changed and becoming unfavorable. 4. Ponds and tanks have become evaporation tanks. 5. Fluoride levels in groundwater are much higher than the WHO safe limit for drinking for the whole year. Corrective measures often do not work and not foolproof so it is necessary to think of preventive measures. Could groundwater be managed without rainfall harvesting in the form of artificial recharge? The answer is no as there is a significant gap between dynamic resource availability and the demand. The concept of water/groundwater banking has to be introduced.

6.2 Artificial recharge in the study area and its simulation

There are no prominent rivers in the area, though streams fill the depressions (forming tanks) during the wet period of about 4 months. The weathered column is distributed unevenly but is in the range of 3 to 15 meters. The over-exploitation of the groundwater resources has eventually resulted in the decline of water levels beyond the weathered zone. There are more than 600 bore wells presently in the area and the yearly draft is about 1.0 million cubic meters (mcm). Natural recharge to aquifers of Maheswaram watershed was measured for 1999 monsoon using tritium injection method. The mean recharge value obtained from measurement at 14 sites was 22.2 mm for the effective seasonal rainfall of 350 mm occurred during July to October, 1999 (Rangarajan & Prasad Rao 2001). The total recharge quantum over the effective infiltration area of 55 sq km is calculated as 1.22 mcm. The soil retension properties in terms of its rate of infiltration were determined at 38 places in the Maheswaram watershed. The infiltration rate varied from 848.8 to 1.7 mm/hr indicating non-uniform properties of the soil profile. The values of transmissivity (T) on single well tests and well field tests varied from about 35m2 /day to about 245 m2/day (Ledoux et al. 2002). The general elevation of the area ranges from 620 m amsl to 600 m amsl, although the hillocks in the southern portions (south of Maheswaram) occupy higher elevations around 640 to 660 m amsl. Geomorphologically areas of higher elevation have shallow pediplain features. In these the weathered zone thickness is less than 10 m below 48

ground level (bgl). Tor complexes are often observed in the area. In general, these complexes are also spheroidically weathered and form poor aquifers. The undulating topography of the granite can be advantageously utilized for furrowing and contour bunding, thus increasing the residence time of the surface water. A fairly good correlation exists between mapped lineaments and well capacities. Since there is sufficient recharge near tanks or discharge areas, the well yields have been observed to be substantial. The existing geologic, topographic, geophysical and borehole information have been integrated; minimizing the risk of poorly located well sites. Groundwater exploration in hard rocks is impaired by the high uncertainty of hydro geological conditions. The uncertainty is mainly due to the geometry of the fractured system. Though a good well inventory data base of about 600 wells with regard to the yields and depth is available, it has been observed that it cannot be applied in the highly heterogeneous hard rock terrain on a local to sub-regional scale for well siting purposes.

Construction of tanks has been quite common in the agricultural rural areas. However due to poor or no maintenance, tank beds are silted and allows negligible infiltration. Thus a new method is proposed utilizing the defunct dug wells that are numerous in the area. Figure 5, shows that a natural runoff from the local catchments could be stored either directly to a dug well or allowing to seep through an artificially designed gravel pit. The method requires the minimum financial implications with maximum benefits. A simple but practical calculation is presented below.

Figure 25 . Run-off diversion to a defunct dug well through a pit so as to minimize silt and other waste materials 0.5 sq km 0.004m (4mm) 10% of rainfall = .0004 x 0.5 x 1000 x 1000 = 200 m3 Run-off: = .0004 10% = 0.0036 m Volume of runoff = 0.0036m = 0.0036 x 0.5 x 1000 x 1000 = 1800 m3 Assuming the total runoff is diverted into a dug well of 10 x 10 x 10 m dimensions: Even if 60% water is captured in the dug well, we can expect a water column equal to = 1800 x 0.6 /10 x 10 = 10.8 meters Recharge area: One day rain: Infiltration:

49

Even if there is considerable percolation from the bottom of the well, the rise in water column will be such that the well will overflow. Since there are several sub-horizontal joints observed in several dug wells in the area, the water eventually would try to saturate the aquifer and also flows through the preferential paths of fractures and joints. The results of the simulation of this artificial recharge experiment in the model is shown in figures 26 to 28 showing the hydrographs of the meshes after the dug-well recharge.

Recharge Dug well

630 625 620 615 610 605 600 Oct-00 Apr-01 Nov-01 May-02 Dec-02 Jun-03 Jan-04

50

Surrounding meshes
1st_1 660 650 640 630 620 610 600 Oct-00 630 625 620 615 610 605 600 Jan-04 Oct-00 1st_2 630 625 620 615 610 605 600 Oct-00 1st_3

Apr-01

Nov-01

May-02

Dec-02

Jun-03

Apr-01

Nov-01

May-02

Dec-02

Jun-03

Jan-04

Apr-01

Nov-01

May-02

Dec-02

Jun-03

Jan-04

1st_4 630 625 620 615 610 605 600 Oct-00

2nd_1 630 625 620 615 610 605 600 Oct-00 Apr-01 Nov-01 May-02 Dec-02 Jun-03 Jan-04 630 625 620 615 610 605 600 Oct-00

2nd_2

Apr-01

Nov-01

May-02

Dec-02

Jun-03

Jan-04

Apr-01

Nov-01

May-02

Dec-02

Jun-03

Jan-04

2nd_3 630 625 620 615 610 605 600 Oct-00 630 625 620 615 610 605 600 Oct-00 Apr-01 Nov-01

2nd_4
630 625 620 615 610 605

3rd_1

Apr-01

Nov-01

May-02

Dec-02

Jun-03

Jan-04

May-02

Dec-02

Jun-03

Jan-04

600 Oct-00

Apr-01

Nov-01

May-02

Dec-02

Jun-03

Jan-04

3rd_2 630 625 620 615 610 605 600 Oct-00 630 625 620 615 610 605 600 Oct-00

3rd_3 630 625 620 615 610 605 600 Oct-00

3rd_4

Apr-01

Nov-01

May-02

Dec-02

Jun-03

Jan-04

Apr-01

Nov-01

May-02

Dec-02

Jun-03

Jan-04

Apr-01

Nov-01

May-02

Dec-02

Jun-03

Jan-04

51

Mesh representing Dug well

630 625 620 615 610 605 600 Oct-00 Apr-01 Nov-01 May-02 Dec-02 Jun-03 Jan-04

Without Recharge

630 625 620 615 610 605 600 Oct-00 Apr-01 Nov-01 May-02 Dec-02 Jun-03 Jan-04

With Recharge

52

Regional Effect of Artificial Recharge

Artificial Recharge Dug well

630 625 620 615 610 605 600 Oct-00 Apr-01 Nov-01 May-02 Dec-02 Jun-03 Jan-04

IFP7 620 615 610 605 600 595 590 Oct-00 Apr-01 Nov-01 May-02 Dec-02 Jun-03 Jan-04 635 630 625 620 615 610 605 600 Oct-00

IFP11

Apr-01

Nov-01

May-02 Dec-02

Jun-03

Jan-04

IFP24 630 625 620 615 610 Oct-00

Apr-01

Nov-01 May-02 Dec-02

Jun-03

Jan-04

53

6.3 Conclusions And Recommendations

Reduction in abstraction through improved irrigation practices and the enactment of a licensing policy are possible approaches to help alleviate the problem Diverting run off to the defunct open wells through an infiltration pit so as to recharge the deeper fracture rocks

It is suggested that the recharge in the rural sector could be enhanced through contour bunding etc., the recharge to groundwater in the urban areas could be increased through check dams constructed across 1st and 2nd order streams in the hilly tracts. The excessive run off and soil losses in the rural areas can be utilized properly through the implementation of the following: protection of non-agricultural lands from biotic interference by human beings, so that they can be used with a natural cover of grasses and trees contour bunding and contour terracing of rain-fed agricultural lands and construction of small storage pits in each of the mini watersheds.

Some of the protection measures that could be adopted in the urban areas for improving the groundwater situation are: The water level data is the only measurable parameter that can be utilized properly for the sustainability of this resource on a long-term basis. Since, the aquifers of the hard rock areas are discontinuous and directional inhomogenieties are rampant, it is desirable to divide the entire area into possible mini-watersheds to implement the water monitoring and quality aspects. Thus, spatial variations based on geomorphology, geology, and land use pattern within the watersheds would enhance accurate demarcations between over-exploited and under-developed areas. This would pave a way for the overall management of the surface and subsurface water sources.

54

REFERENCES
Aboufirrasi, M. and M.A. Marino (1983): Kriging of water levels in Souss aquifer, Morocco. Math. Geol., 15(4), 537-551. Agnihotri, V. and S. Ahmed (1997): Analyzing ambiguities in the data collection network design by Geostatistical estimation variance reduction method, Jour. Environ. Hydrology, Vol. 5, paper 5, (http://www.hydrweb.com). Ahmed, S. and G. de Marsily (1989): Cokriged estimates of transmissivity using jointly water-level data. In Armstrong (ed.) Geostatistics, Kluwer Academic Publishers, 615-628. Ahmed, S. and G. de Marsily (1993): Cokriged estimation of aquifer transmissivity as an indirect solution of inverse problem - A practical approach. Water Resour. Res., 29(2), 521-530. Ahmed, S. & Gupta, C.P. (1989) Stochastic spatial prediction of hydrogeologic parameters: role of cross-validation in krigings. International Workshop on Appropriate Methodologies for Development and Management of Groundwater Resources in Developing Countries February 28-March 4, 1989, India, Oxford & IBH Publishing Co. Pvt. Ltd., IGW, Vol. III, 77-90. Ahmed, S. & Marsily, G de. (1993) Cokriged estimation of aquifer transmissivity as an indirect solution of the inverse problem: A practical approach. Water Resour. Res. 29,(2), 521-530. Ahmed, S. (1987): Estimation des transmissivits des aquifres par methodes Gostatistique Multivariables et resolution du Problme Inverse. Doctoral thesis, Paris School of Mines, France, 148pp Ahmed, S. (1989): Geostatistical estimation of aquifer parameters: some case studies, (Invited Paper). In Dillon and Ahmed (eds.), Proc. of the Australian Workshop on "Geostatistics in Water Resources", Glen Osmond, Australia, Nov. 13-17., Vol. 2, E1-E20, Ahmed, S. (1993): Geostatistical solution to the inverse problem in groundwater hydrology using prior information. In Proc. of the workshop on "Inverse Problem in Science and Engineering" held at NAL/CMMACS, Banglore, Feb. 12-16, 15 p. Ahmed, S. (1995): An interactive software for computing and modeling a variograms. In Mousavi and Karamooz (eds.) Proc. of a conference on "Water Resources Management (WRM'95)", August 28-30. Isfahan University of Technology, Iran, 797-808. Ahmed, S. (2001) Program PLAYG, version 3.0, users guide, NGRI report. Ahmed, S. and G. Murali (1992): Regionalization of fluoride content in an aquifer. Jour. of Environmental Hydrology, 1(1), 35-39. Ahmed, S. and C.P. Gupta (1989) Stochastic Spatial Prediction of Hydrogeologic Parameters: Use of cross-validation in Krigings, In Proc. of Internat. Groundwater Workshop , Hyderabad, India, Feb. - March, 1989, (Gupta et al. eds.), Oxford and IBH Pub. Co., Vol III, p. 77-90. Ahmed, S. and C.P. Gupta (1996): Optimal data collection network having a pre-set kriging variance. submitted to the 30th International Geological Congress, Beijing, China. Ahmed, S. and G. de Marsily (1987): Comparison of geostatistical methods for estimating transmissivity using data on transmissivity and specific capacity. Water Resour. Res., 23(9), 1717-1737. Ahmed, S. and G. de Marsily (1987b): Geostatistical estimation of transmissivity using different sources of information. In Abstract volumes of IAHS Workshop (HW-2) during XIX IUGG Assembly at Vancouver, Canada, Aug. 9-22, Vol. 3, p. 956. Ahmed, S. and G. de Marsily (1987c): Multivariate Geostatistical approach in estimating aquifer parameters. In Awadalla & Noor (eds.) Proc. of an Internat. Groundwater 55

Conference, "Groundwater and the Environment", Malaysia, June 22-26, Universiti Kebangsaan Malaysia, C1-C10. Ahmed, S. and G. de Marsily (1988): Some applications of multivariate kriging in groundwater hydrology. Science de la Terre, Serie Informatique, 28, 1-25. Ahmed, S. and G. de Marsily (1989a): Cokriging hydrogeologic parameters without having common measurement points. Presented at the 28th Internat. Geol.Congress, Washington D.C., U.S.A., July 9-19. Ahmed, S. and V. Agnihotri, (1998) Fluoride pollution: a geostatistical method of desiging an optimal monitoring network, In Proc. of workshop on Water pollutionAssessment and Management, published by the IGNA, Hyderabad, p. 49-53. Ahmed, S., 2002: Groundwater Monitoring Network Design: Applications of Geostatistics With A Few Case Studies From A Granitic Aquifer from semi-arid region, In A Semi-Arid Region, In Sherif M.M. et al (eds.) Groundwater Hydrology, Volume 2, p. 37-57, A.A. Balkema Publishers. Ahmed, S., and Murthy, P.S.N., Could Radial Basis Function replace Ordinary Kriging?, In proc. of V International Geostatistical Congrss, Wollongong, Australia, 1996. Ahmed, S., G. de Marsily and Alain Talbot (1988): Combined use of hydraulic and electrical properties of an aquifer in a geostatistical estimation of transmissivity, Ground Water, 26(1), 78-86. Ahmed, S., G. de Marsily and C.P. Gupta (1990): Coherent structural models in cokriging aquifer parameters - transmissivity and water-levels. In Proc. of International Conf. on "Water Resources in Mountainous Regions", Lausanne, Switzerland, Aug. 27 - Sept. 1, Internat. Asso. of Hydrogeol. Pub. Vol. XXII Part 1, 173-183. Ahmed, S., S. Sankaran and C.P. Gupta (1995): Variographic analysis of some hydrogeological parameters: use of geological soft data. Jour. of Environ. Hydrology, Vol. 3(2), p. 28-35. (http://www.hydroweb.com.) APGWD, 1977. Studies on hydrologic parameters of groundwater recharge in water balance computations, Andhra Pradesh. Government of Andhra Pradesh Ground Water Department, Hyderabad; Research series no.6, 151 pp. Aspie, D., and Barnes, R. J., Infill sampling design and the cost of classification errors: Math. Geology, v. 22, no. 8, p. 915-932, 1990. Bardossy, A., I. Bogardi and W.E. Kelly (1986): Geostatistical analysis of geolectric estimates for specific capacity. J. Hydrol. 84, 81-95. Barnes, R. J., Sample design for geologic site characterization, in Armstrong, M., ed., Geostatistics, v. 2, Kluwer Academic Publ. Dordrecht, p. 809-822, 1989. Bogardi I., Bardossy A., Duckstein, L., 1985: Multicriterion Network Design Using Geostatsitics. Water Resour. Research 21(2): 199-208. Bras R.L., Rodriguez-Iturbe I., 1976: Network design for the estimation of areal mean of rainfall events. Water Resour. Research 12(6): 1185-1195. Bruel D., Engerrand C., Ledoux E., Marechal J-C., Touchard F., Subrahmanyam K., Ahmed S.; 2002. The evaluation of aquifer parameters in Maheshwaram mandal, RR dist., AP., India. Ecole des Mines de Paris, CIG, technical report LHM/RD/02/28, July 2002, updated November 2002, 27 p. Carrera, J., E. Usunoff and F. Szidarovsky (1984): A method for optimal observation network design for groundwater management. J. Hydrol., 73, 147-163. Charlier J-B., 2002. Contribution to hydrogeology in monsoon ares. Study cases in Andhra Pradesh, India. ??, 54 p. Chilton P.J., Foster S.S.D.; 1995. Hydrogeological characteristics and water-supply potential of basement aquifers in tropical Africa. Hydrogeology Journal, vol. 3, pp. 3-49.

56

CRIDA, 1990. Soil Map of Hayathnagar farm and Appendix. Center for Research In Dryland Agriculture, Ranga Reddy District, Andhra Pradesh, India. CRIDA, 1991-92; 1993-94; 1994-95; 1997-98; 1998-99; 1999-2000. Annual Report. Center for Research In Dry-land Agriculture, Ranga Reddy District, Andhra Pradesh, India. Dagan, G. (1985) Stochastic modelling of groundwater flow by unconditional and conditional probabilities. Wat. Resour. Res. 21(1), 65-72. Darricau-Beucher, H. (1981): Approche Gostatistique du passage des donnes de terrain aux paramtres des modles en Hydrogologie. Doctoral thesis, Paris School of Mines, France, 119pp. Das, S., et al., Manual: Hydrograph Network Stations, Central Groundwater Board, 1995. De Marsily, G. 1986: Quantitative Hydrogeology, Groundwater Hydrology for Engineers, Academic Press. 440pp. De Smedt, F., M. Belhadi and I.K. Nurul (1985): Study of spetial variability of hydraulic conductivity with different measurement techniques. In Marsily (ed.) Proc. 'The stochastic approach to sub-surface flow', Motvillargen, France, June 3-6. Delhomme, J.P. (1974): La cartographie d'une grandeur physique partir des donnes de diffrentes qualits. In proc. of IAH congress, Montpelier, France, Tome X, Part-1, 185-194. Delhomme, J.P. (1976): Application de la thorie des variables rgionalises dans les sciences de l'eau. Doctoral thesis, Paris School of Mines, France, 130pp. Delhomme, J.P. (1978): Kriging in Hyrosciences. Adv. in Water Resour., 1(5), 251-266. Delhomme, J.P. (1979): Spatial variability and uncertainty in groundwater flow parameters: a Geostatistical approach. Water Resour. Res., 15(2), 269-280. Deutsch, C.V. and A. G. Journel (1992): GSLIB, Geostatistical software library and User's guide. Oxford Univ. Press, New York, 340pp. Dillon P.J., 1988: Quantitative methods for monitoring network design - Future directions, Hydrology and Water Resources Symposium, ANU, Canberra, 1-3 Feb. Dong, A., Ahmed, S. & Marsily, G de. (1990) Development of geostatistical methods dealing with boundary condition problems encountered in fluid mechanics, in Proceedings, 2nd European Conference on the Mathematics of Oil Recovery, Arles, France, September 11-14, 1990, edited by Guerillot and Guillon, 21-30, Technip, Paris. Engerrand C., 2002. Hydrogeology of the weathered fissured hard rock aquifers located in monsoon areas : hydrogeological study of two watersheds in Andhra Pradesh (India). PhD Thesis, Universit Pierre et Marie Curie, Paris, France, 203 pp. Galli A. and G. Meunier (1987): Study of a gas reservoir using the external drift method. In Matheron and Armstrong (eds.) Geostatistical case studies, D. Reidel Hingham, 105120. Gambolati, G. and G. Volpi (1979): A conceptual deterministic analysis of Kriging technique in Hydrology. Water Resour. Res. Vol. 15(3), 625-629. Gao,H., J.Wang and P. Zhao(1996): The updated Kriging Variance and Optimal sample design.Math.Geol.,28(3), 295-313. Handa, B.K (1975), Geochemistry and genesis of fluoride containing groundwater in India, Ground Water: 13:275-281. Handa, B.K (1988), Fluoride occurrences in natural water in India and its significances, Bhujal: 4-6:31-37. Hoeksema, R.J. and P.K. Kitanidis(1984):An application of the geostastical approach to the inverse problem in two-dimentional ground water modeling. Water Resour,Res.,20(7),1003-1020. Hudak, P.F., and Loaiciga, H.A., An optimization method for monitoring network design 57

in multilayered Groundwater flow system: Water Resour. Res., v. 29, no. 8, p. 28352845, 1993. Hughes, J.P. and Lattemair (1981): Data requirements for Kriging: Estimation and Network design. Water Resour. Res., 17(6), 1641-1650. Isaaks, E.H. and R.M. Srivastava (1989): An introduction to Applied Geostatistics, Oxford Univ. Press, 561pp. Jha, L.B; and Jha, M (1982), Fluoride pollution in India: Int. J. Env. Studies; 19:225-230. Journel, A.G. and C. Huijbregts (1978): Mining Geostatistics. Academic Press, 600pp. Kitanidis, P.K. (1997) Introduction to Geostatistics: Application to Hydrogeology, Cambridge University Press. Krisnamurthy N.S., Kumar D., Sankaran S., Ananda Rao V., Negi B.C., Engerrand C., Jain S.C.; 2000. Magnetic Investigations across some lineaments in Maheshwaram watershed, Andhra Pradesh, India. Technical Report National, Geophysical Research Institute of India, NGRI-2000-GW-288, September 2000, 5 pp. Kupfersberger, H. and G. Bloschl (1995): Estimating aquifer transmissivities - On value of auxilary data. J. Hydrol., 165, 85-99. La Venue, A.M. & Pickens, J.F. (1992) Application of a coupled adjoint sensitivity and kriging approach to calibrate a groundwater flow model. Water Resour. Res. 28 (6), 1543-1569. Lebert F ; 2001. Mesure de topographie par DGPS sur le bassin de Maheshwaram, (Andra Pradesh, Inde). tude ralise dans le cadre des oprations de Service public du BRGM 2001-EAU-R01 ; rapport BRGM/RP-50728-FR ; Fvrier 2001 ; 25 p. Legchenko A., Baltassat J-M.; 1999. Application of the Numis proton magnetic resonance equipment for groundwater exploration in a fractured granite environment 30 km south of Hyderabad, India. BRGM- rapport interne Dcembre 1999 R40925, 44 p. Loaiciga, H.A. (1989): An optimization approach for Groundwater quality monitoring network design. Water Resour. Res. 25(8), 1771-1782. Marsily, G de. (1986) Quantitative Hydrogeology: Groundwater Hydrology for Engineers: Academic Press, Inc., London. Marsily, G. de and S. Ahmed (1987): Application of kriging techniques in groundwater hydrology. Invited paper, Jour. of Geol. Soc. of India, 29(1), 47-69. Matheron G., 1963: Trait de Gostatistique applique. Mmoires du BRGM, No. 14, Edition Technip. Matheron, G. (1963): Trait de Gostatistique applique. Mmoires du BRGM, No. 14, Edition Technip. Matheron, G. (1971): Theory of Regionalized Variables and their Applications. Cahier du C.G.M.M., Fontainebleau, France, 211pp. Matheron, G., Principles of Geostatistics, Econ. Geol., v. 58, p. 1246-1266, 1963. Mizell, S.A.(1980): Stochastic analysis of spatial variability in two -dimentional Groundwater flow with implications for observation-well network design. Doctoral thesis, New maxico institute of mining and technology, U.S.A.,133 pp. Mizell, S.A.(1980): Stochastic analysis of spatial variability in two -dimentional Groundwater flow with implications for observation-well network design. Doctoral thesis, New maxico institute of mining and technology, U.S.A.,133 pp. Monteith J.L., Huda A.K.S., Midya D., 1989. Modelling Sorghum and Pearl Millet Rescap : A resource capture model for Sorghum and Pearl Millet. In Modelling the growth and development of Sorghum and Millet, Research Bulletin n12, International Crops Research Institute for the Semi-Arid Tropics, pp. 30-34. Murthy, P.S.N. and S. Ahmed (1994): Cokriging under constrained condition as applied to mineral deposits. In Fabbri & Royer (eds.), 3rd CODATA conference on 58

"Geomathematics and Geostatistics" held during June 7-11, 1993 at Netherlands, Science de la Terre, Serie Informatique, 32, 63-79. Myers, D.E. (1982): Matrix formulation of Cokriging. Math. Geol., 14(3), 249-257. Myers, D.E. (1983): Estimation of linear combination and cokriging. Math. Geol. 15(5), 633637. Myers, D.E. (1984): Cokriging - New developments. In Verly et al. (eds.) Geostatistics for natural resources characterization. D. Reidel Pub. Co., 295-305. Neuman, S.P. (1984): Role of Geostatistics in subsurface hydrology. In Verly et al. (eds.) Geostatistics for natural resources characterization, Proc. NATO-ASI, D. Reidel Pub. Co., 787-816. Rangarajan R., Prasada Rao N.T.V.; 2001. Natural recharge measurements in Maheshwaram granitic watershed, Ranga Reddy district, Andhra Pradesh, India 1999 Monsoon. Technical Report No. NGRI-20001-GW-298, National Geophysical Research Institute, Hyderabad, India, 14 p. Ravi Prakash, M., S. Ahmed and V.V.S. Gurunadha Rao (1990) Stochastic Analysis of Transmissivity and Specific Yield obtained from Dug Well Pumping Tests, In Proc. of "Groundwater Investigation, Management and Geophysical Technique", Lucknow, India, Dec, 1990. Roth,C.(1995): Contribution de la Geostatistique a la resolution du probleme inverse en hydrogeologie. Doctoral thesis,Paris school of mines,France,195 pp. Roth,C.,J.P. Chiles and C.de Fouquet (1996): Adapting geostatistical transmissivity simulations to finite differece flow simulators. Water Resour. Res. Vol. 32 (10), 32373242. Rouhani S., 1990: Multivariate geostatistical approach to space-time data analysis. Water Resour. Research 26(4): 585-591. Rouhani S., Hall, T.J., 1988: Geostatistical Schemes for Groundwater sampling. Jr. Hydrology 103: 85-102. Rouhani, S. (1985): Variance reduction analysis. Water Resour. Res., 21(6), 837-840. Rouhani, S. (1990) Multivariate geostatistical approach to space-time data analysis. Wat. Resour. Res. 26(4), 585-591. Rouhani, S., and Hall, T.J. (1988) Geostatistical Schemes for Groundwater sampling: Jr. Hydrology v. 103, p. 85-102. Samper F.J. and J. Carrera (1990): Geoestadstica - Aplicaciones la hidrologa subterrnea. Bercelona Univ., 484pp. Saxena, V.K and Ahmed, S. (2001) Dissolution of fluoride in groundwater: a water-rock interaction study, Environmental Geology, Vol. 40(9):1084-1087, 2001. Sen, Z. (1989): Cumulative Semivariogram model of Regionalized variables. Math. Geol., 21(8), 891-903. Sen, Z. (1992): Standard Cumulative Semivariograms of Stationary Stochastic Processes and Regional correlation. Math. Geol., 24(4), 417-435. Sirayonone, S., Comparative studies of kriging, multiquadric-biharmonic and other methods for solving mineral resources problems, Doctoral thesis, IowA State University, USA, 1988. Sophocleous M., Paschetto E., Olea R.A., 1982: Groundwater Network design for Northwest Kansas, Using the Theory of Regionalized Variables. Groundwater 20(1): 48-58. Sophocleous, M., Paschetto, E., and Olea, R.A., Groundwater Network design for Northwest Kansas, Using the Theory of Regionalized Variables: Groundwater v. 20, no. 1, p. 48-58, 1982.

59

Subrahmanyam K., Ahmed S., Dhar R.L., 2000: Geological and Hydrogeological Investigations in the Maheshwaram Watershed, R.R. Dist., Andhra Pradesh, India, Technical Report No. NGRI-2000-GW-292. Thangarajan, M. and S. Ahmed (1989): Kriged estimates of water-levels from the sparse measurements in a hard rock aquifer. In Gupta et al. (eds.) Proc. of Internat. Groundwater Workshop (IGW-89), Hyderabad, India, Feb. 28 to March 4, Oxford and IBH Pub. Co., Vol. I, 287-302. Thiry D. ; 1993a. Logiciel Marthe : Modlisation dAquifres par un maillage Rectangulaire en rgime Transitoire pour le calcul Hydrodynamique des Ecoulements Version 4.3. Rapport BRGM 4S/EAU n. R32210. Thiry D. ; 1993b. Modlisation des aquifres complexes Prise en compte de la zone non sature et de la salinit. Calcul des intervalles de confiance. Revue Hydrogologie, Vol. 4, pp. 325-336. Thiry D., Boisson M.; 1991. Logiciel GARDENIA modle Global A Rservoirs pour la simulation des DEbits et des Niveaux Aquifres ; guide dutilisation (version 3.2). BRGM, R32209, janvier 1991. Thiry D., Dilucas C., Diagana B. ; 1993. Modelling the aquifer recovery after a long duration drought in Burkina Faso. Proceedings of the IAHS/IAMAR International Symposium on Extreme hydrological events : precipitation, floods and drought , Yokohama, Japon, Juillet 1993. Publication AIHS 213, pp. 43-50. Thiry, D., Amraoui, N., 2001. Hydrological modelling of the Sane basin. Sensitivity to the soil model. Physics and Chemistry of the Earth Journal, Part B, 26, 467-472. Thiry, D.; 1988. Analysis of long-duration piezometric records from Burkina Faso used to determine aquifer recharge. In Estimation of Natural Groundwater Recharge, edited by Simmers, Netherlands; pp. 477-489. U.S.P.H.S (1987) Drinking water standard public health services, Washington DC; p. 333. Virdee T.S., Kottegoda, N.T., 1984: A brief review of kriging and its application to optimal interpolation and observation well selection. Hydrological Sciences Jr. 29(4): 367-387. Wackernagal,H.(1995): Multivariate Geostatistics: An introduction with applications. Springer, 256 pp. WHO (1984) Guidelines for drinking water quality, Values.3; Drinking water quality control in small community supplies, WHO Geneva: p. 212.

60