HR: 1340h
AN: H53F-0541 [Abstracts]
TI: Efficiency of ANN Model in Approximating Saltwater Intrusion Process in Coastal Aquifers with Noisy
Data
AU: * Bhattacharjya, R K
EM: rajibkb@nits.ac.in
AF: Department of Civil Engineering, National Institute of Technology Silchar, Silchar, AS 788010
India
AB:
The over exploitation of groundwater resources has become unavoidable in many parts of the world due to the increasing
demands for freshwater. The unplanned exploitation of freshwater from coastal aquifers may cause saltwater intrusion into
coastal aquifers. The flow and transport equations are to be solved in order to model saltwater intrusion process in coastal
aquifers. The density variations in transition zone make the flow and transport equations highly complex and nonlinear.
Therefore, simulation of these highly nonlinear processes is also complex and costly in terms of computer memory and
computational time requirements. An approximate simulation of these complex processes would be useful especially when
repetitive simulations are necessary. There are various possible applications for such an approximator, if the simulations
are reasonably accurate. One possible application would be the use of such an approximator for simulation, linked to an
optimization model for decision making.
This paper evaluates the efficiency of an artificial neural networks model (ANN) in approximating three dimensional density
depended saltwater intrusion process in coastal aquifer and also evaluates its performance with noisy data. The data required
for training and testing the ANN model is generated using a numerical simulation model. The trained ANN predicts the
concentration at specified observation locations at different times. The performance of the ANN model is evaluated using an
illustrative study area. These evaluation results show that the performance of the ANN model is quite satisfactory and does
not degrade much with noisy data. For example, the average absolute relative error (AARE) value during testing is 4.86 when
the training patters are free form error. The AARE value increases to 5.674 when the training patterns are perturbed with
normally distributed error of standard deviation (SD) 0.05, and mean zero. Similarly, the AARE value increases up to 6.701,
7.489 and 9.194 when the training patterns are perturbed with normally distributed error of SD 0.10, 0.15, and 0.20
respectively and mean zero.
The performance of the model is also evaluated with respect to threshold statistics (TSx) for an absolute relative error
(ARE). The ANN model is very sensitive with respective to TS5 statistics. The average value of TS5 is 70.910% when the ANN
model is trained with error free patters. This indicates that 70.910% of predicted concentration values had an ARE value
less than 5%. The TS5 value decreases to 64.518% when the training patterns are perturbed with normally distributed error
of SD 0.05 and mean zero. The TS5 values further decrease to 56.426%, 50.301%, and 38.883% when the training patterns are
perturbed with normally distributed error of SD 0.10, 0.15, 0.20 respectively and mean zero. The ANN model is less sensitive
for TS40 and TS50 statistics. For example the TS40 value is 99.693% when the ANN model is trained with error free patterns
and 98.816% when the ANN model is trained with the patterns perturbed with normally distributed error of SD 0.20 and mean
zero.
The performance evaluations of the developed ANN model show that this approach is potentially useful for approximate
simulation of the transient density dependent three dimensional flow and transport processes in coastal aquifers. The
performance of the model does not degrade much with noisy data. However, more rigorous evaluations using larger and more
complex study areas are necessary before the applicability of this approach can be fully established.
DE: 1828 Groundwater hydraulics
DE: 1829 Groundwater hydrology
DE: 1831 Groundwater quality
DE: 1832 Groundwater transport
SC: Hydrology [H]
MN: Fall Meeting 2005