A relatively new method of addressing different hydrological problems is the use of artificial neural networks (ANN). In groundwater management ANNs are usually used to predict the hydraulic head at a well location. ANNs can prove to be very useful because, unlike numerical groundwater models, they are very easy to implement in karstic regions without the need of explicit knowledge of the exact flow conduit geometry and they avoid the creation of extremely complex models in the rare cases when all the necessary information is available. With hydrological parameters like rainfall and temperature, as well as with hydrogeological parameters like pumping rates from nearby wells as input, the ANN applies a black box approach and yields the simulated hydraulic head. During the calibration process the network is trained using a set of available field data and its performance is evaluated with a different set. Available measured data from Edward's aquifer in Texas, USA are used in this work to train and evaluate the proposed ANN. The Edwards Aquifer is a unique groundwater system and one of the most prolific artesian aquifers in the world. The present work focuses on simulation of hydraulic head change at an observation well in the area. The adopted ANN is a classic fully connected multilayer perceptron, with two hidden layers. All input parameters are directly or indirectly connected to the aquatic equilibrium and the ANN is treated as a sophisticated analogue to empirical models of the past. A correlation analysis of the measured data 1144 I.C. Trichakis et al. is used to determine the time lag between the current day and the day used for input of the measured rainfall levels. After the calibration process the testing data were used in order to check the ability of the ANN to interpolate or extrapolate in other regions, not used in the training procedure. The results show that there is a need for exact knowledge of pumping from each well in karstic aquifers as it is difficult to simulate the sudden drops and rises, which in this case can be more than 6 ft (approx. 2 m). That aside, the ANN is still a useful way to simulate karstic aquifers that are difficult to be simulated by numerical groundwater models.

Abstract:The simulation of karstic aquifers is difficult using traditional groundwater numerical simulators, as the exact knowledge of the hydraulic characteristics of the physical system in small scale is rarely available and the numerical simulators produce results of limited reliability. In the present work, artificial neural networks (ANNs) are utilized to predict the response of a karstic aquifer, using the hydraulic head change per time step rather than the hydraulic head itself as output parameter of the network. As it will be demonstrated, in the first case a better approximation of the physical system's response is achieved as the change of the hydraulic head is more naturally connected to the input parameters of the network, which model the aquatic equilibrium of the system. The correlation of rainfall and hydraulic head change per time step was initially used to determine the time lag of the rainfall input data, which represents the time needed by the rainfall to percolate and reach the water table. In a second step, a differential evolution (DE) algorithm is utilized for the optimal selection of rainfall time lag as well as ANN's architecture and training parameters. Although a time consuming procedure, the improvement obtained suggests that the empirical determination of the ANN parameters and structure is not always sufficient and an optimization procedure, which minimizes the training and evaluation errors of the ANN, may provide substantially better simulation results. The optimized networks were finally used for midterm predictions (30 to 90 days ahead) of the hydraulic head, showing the ability of the ANN with hydraulic head change as output parameter to provide predictions with high accuracy at the end of the considered time period.

No abstract

No abstract

Abstract:Following many applications artificial neural networks (ANNs) have found in hydrology, a question has been rising for quantification of the output uncertainty. A pre-optimized ANN simulated the hydraulic head change at two observation wells, having as input hydrological and meteorological parameters. In order to calculate confidence intervals (CI) for the ANN output two bootstrap methods were examined namely bootstrap percentile and BCa (Bias-Corrected and accelerated). The actual coverage of the CI was compared to the theoretical coverage for different certainty levels as a means of examining the method's reliability. The results of this work support the idea that the bootstrap methods provide a simple tool for confidence interval computation of ANNs. Comparing the two methods, the percentile requires fewer calculations and yields narrower intervals with similar actual coverage to that of BCa. Overall, the actual coverage was proved lower than desired when not modeled points were present in the data subset.