Page 212 - Hydrogeology Principles and Practice

HYDC05 12/5/05 5:35 PM Page 195

Groundwater investigation techniques 195

BO X
Continued
5.6

Constant inflows on the southern and eastern boundaries were used years after the start of pumping from either wellfield, with max-
to represent system recharge, with a range of possible total inflows imum impacts likely to occur 75–300 years after abstraction ceases.
simulated. The model was refined within the bounds established in The location of the Monturaqui wellfield upgradient of approx-
the conceptual model so that it reproduced as closely as possible imately 60% of the aquifer recharge was found to significantly
the observed steady-state (pre-abstraction) head distribution in the reduce its relative impacts on groundwater flows at Tilopozo. From
aquifer (Fig. 1b). Following model calibration, a time-variant model an interpretation of the results of sensitivity analyses, a ‘worst case’
for a period of up to 500 years was developed for predictive prediction model was developed which maximized the magnitude
modelling of impacts. Groundwater abstractions with a range of of predicted impacts, and model runs executed until a pumping
durations from both the Monturaqui and Negrillar wellfields were duration of between 10 and 20 years and a gross abstracted
8
3
represented in the model with the model simulations used to volume of 7.38 × 10 m (Fig. 1c) were found that limited through-
predict outflow changes with time across the northern boundary flow reduction at the Tilopozo wetland to 6% (Anderson et al.
representing the Tilopozo wetland. Significant uncertainty in the 2002). The resulting model now forms the basis for the sustainable
value of specific yield (varied between 0.05 and 0.2) and recharge groundwater development of the MNT aquifer which, fortunately,
−1
rate (between 450 and 1800 L s ) was addressed at all stages of is supported by the extremely large volume of groundwater storage
10
3
the modelling through sensitivity analyses. in the aquifer (approximately 10 m ).
Analysis of the model output (Fig. 1c) showed that the impacts of
groundwater abstraction will reach the wetland between 20 and 40
matrix methods for solving the large number of (1982), Spitz and Moreno (1996) and Rushton (2003).
unknowns. Successful completion of a model run is A popular finite-difference model for application in
obtained when convergence of the head solution two- and three-dimensional groundwater flow prob-
is reached, usually determined by a model error lems is the United States Geological Survey’s code
criterion set at the beginning of the run. MODFLOW (McDonald & Harbaugh 1988), with
In all groundwater modelling investigations, for demonstrations of this model presented by Anderson
example the case study presented in Box 5.6, the and Woessner (1992) and Chiang and Kinzelbach
process of deriving a model for predictive purposes (2001).
involves the following common steps: conceptualiza- Other groundwater modelling approaches in-
tion of the flow mechanisms in the model area based clude more specialist applications such as solute
on existing knowledge; acquisition of available field transport modelling using the method of character-
data on groundwater heads, aquifer properties and istics (Konikow & Bredehoeft 1978) or the ‘random
river flows; discretization of the model domain and walk’ method (Prickett et al. 1981) for solving the
construction of the model input file; calibration of the advection-dispersion equation (see eq. 6.7). In the
model by comparing simulated steady-state (equilib- method of characteristics, advective transport is
rium) and transient (time-variant) heads and flows (flow simulated by particles distributed in a geometrically
vectors and water balance) against field-measured uniform pattern over the entire model area with each
values; sensitivity analysis of aquifer property values particle assigned an initial concentration associated
and recharge and boundary conditions; validation with the concentration in the cell containing the
of the model against an independent set of data (for particle. Dispersive transport is simulated by a finite-
example, groundwater head and river flow data not difference calculation on the rectangular grid after
used during model calibration, or hydrochemical which the particle concentrations are updated accord-
data such as salinity); and, lastly, prediction of aquifer ing to the changes in the grid concentrations, and
response under changed groundwater conditions. advective transport in the next time step is calculated.
For further discussion of finite-difference and In a similar way, the random walk method com-
finite-element numerical modelling techniques, the bines a flow submodel, usually based on the finite-
reader is referred to the texts by Wang and Anderson difference method, with the use of random variables