Abstract
The objective of this dissertation is a critical review of all the required stages for the development of a sound groundwater mathematical model. Towards this objective a groundwater model is developed and implemented for the management of the Pieria aquifer, Greece. The stages of the modeling process which are analyzed in this dissertation are the following: - development of the conceptual model, - construction of the numerical model, which includes the definition of the space and time discretization, the initial and boundary conditions, the recharge and discharge of the system, the representation of hydrogeologic features of the system, the spatial structure of the aquifer’s hydrogeologic parameters, the selection of the method for the solution of the groundwater flow equation and the convergence criteria for this method, - calibration of the model, - evaluation of the model, - sensitivity analysis, - validation of the calibrated model and – application of the model to predict the sy ...
The objective of this dissertation is a critical review of all the required stages for the development of a sound groundwater mathematical model. Towards this objective a groundwater model is developed and implemented for the management of the Pieria aquifer, Greece. The stages of the modeling process which are analyzed in this dissertation are the following: - development of the conceptual model, - construction of the numerical model, which includes the definition of the space and time discretization, the initial and boundary conditions, the recharge and discharge of the system, the representation of hydrogeologic features of the system, the spatial structure of the aquifer’s hydrogeologic parameters, the selection of the method for the solution of the groundwater flow equation and the convergence criteria for this method, - calibration of the model, - evaluation of the model, - sensitivity analysis, - validation of the calibrated model and – application of the model to predict the system’s response under various conditions. The model which was developed for the Pieria aquifer is a two-dimensional transient flow model. It covers an area of 256 km² and it has 1759 active cells, with a size of 350 m x 350 m each. Software packages which were used for the construction of the model are: MODFLOW 2000, Argus ONE, ArcGIS and PEST. The zonation method was used for the description of the hydrogeologic parameters’ spatial variability. The grid was divided in 11 zones of uniform hydraulic conductivity in the x-direction, 6 zones of uniform horizontal anisotropy, 4 zones of uniform specific storage and 4 zones of uniform specific yield. The values of the 25 hydrogeologic parameters resulted from the calibration process. The model was calibrated against 1173 piezometric head observations and 26 spring flow observations with the Gauss-Levenberg-Marquardt method. The piezometric head and the spring flow observations were collected from September 1992 to April 1995, on a monthly basis. Both qualitative and quantitative criteria were used for the evaluation of the calibrated model. The results which we obtained show that the model represents adequately the simulated system. The effect of each one of the 38 model parameters (25 hydrogeologic and 13 recharge parameters) was examined with the sensitivity analysis process. A second set of monthly data (118 head observations and 4 spring discharge observations, from May 1995 to August 1995 were used for the validation of the calibrated model. The validation process verified that the calibrated model describes accurately the water dynamics of the Pieria aquifer. Finally, the calibrated model was applied for a period of 3 years to assess the consequences of two pumping scenarios concerning the location of irrigation wells. The proper time discretization was examined by using a number of uniform and geometrically increased time steps. The comparison of total execution time and computed heads revealed that increasing the length of the time step geometrically, can reduce execution time significantly, but the corresponding heads diverge from those computed with uniform time steps. It is concluded that the time discretization depends on the special characteristics of the model and must be investigated thoroughly in any groundwater mathematical model. The initial condition of the computations must be also examined very carefully. An arbitrarily defined initial condition (i.e. interpolation from field measurements), sometimes can influence the results until the end of the whole simulation period. The known head boundary condition constitutes an inexhaustible water source for the model, so its contribution to the model’s water balance should also be examined and taken into account during the calibration of the model. Sensitivity analysis of this boundary condition, in order to assess the magnitude and the spatial distribution of its influence, is an essential prerequisite for the model construction. The particularity of this boundary condition should also be taken into consideration during predictive simulations. Plots of the computed recharge/discharge from every known head boundary cell with respect to time can be a valuable tool for the estimation of the known head cell values in predictive simulations. It is concluded that the integration of this boundary condition in the calibration process leads to a more accurate representation of the real system. In our case this was accomplished using artificial observations of zero inflow from the known head boundary cells. The head convergence criterion is very important and depends on the solution method. It must always be examined in conjunction with the computed water balance error. In our case this error never exceeded 0.1%. It was found that the use of different kind of observations in the calibration process decreased the correlation between the estimated parameters and thus it led to a unique solution of the inverse problem. Model efficacy in predicting the response of the system depends on the quality and quantity of available data. A transient model should be calibrated against a considerable number of observations with satisfactory spatial and temporal distribution, representing the system response under various conditions. The evaluation of a calibrated model must be based on statistical criteria referring to the residuals between observations and corresponding computed values, to the absolute residuals and to the final parameter values resulted from the calibration process. The water balance of the model should also be considered. Most of the commonly used statistical criteria come from time series analyses where large data sets are usually available. In groundwater models, however, availability of such data sets is rare. It is thus important to use several criteria to get a better evaluation of the model. Software’s automated procedures for data handling and parameter default values often mislead the development of a groundwater model. Appropriate interpolation techniques and parameter values are problem depended and should be decided after investigation.
show more