USP Electronic Research Repository

Robust Ensemble Modeling Paradigm for Groundwater Salinity Predictions in Complex Aquifer Systems

Lal, Alvin and Datta, Bithin (2021) Robust Ensemble Modeling Paradigm for Groundwater Salinity Predictions in Complex Aquifer Systems. In: Groundwater Resources Development and Planning in the Semi-Arid Region. Springer Nature, Switzerland, pp. 53-72. ISBN 978-3-030-68123-4

Full text not available from this repository.


Accurate groundwater salinity monitoring and prediction is an essential component of a groundwater management framework. In hydrological investigations, complex 3D numerical simulation seawater intrusion models are used to monitor groundwater salinity levels in aquifers. However, using a complex numerical simulation model to monitor salinity levels and predict future groundwater conditions is computationally demanding. Also, integrating complex numerical simulation models to develop aquifer management frameworks are deemed to be computationally expensive, inefficient, and time consuming. One of the user-friendly solutions to these two problems is the use of data-driven prediction models developed using machine learning algorithms. This study presents a methodology of using homogeneous ensemble model to monitor and predict groundwater salinity concentrations in an aquifer system instead of using a more complex FEMWATER-based groundwater numerical simulation model. The FEMWATER computer package is used to simulate seawater intrusion processes in the aquifer. Different transient input pumping patterns are implemented into the numerical simulation model to obtain corresponding salinity concentration data at monitoring bores. These input (transient pumping patterns) and output (salinity concentration at monitoring bores) are used to train and test individual prediction model in the ensemble. The homogeneous ensemble prediction model development and evaluation is conducted in two phases. In Phase 1, five different machine learning algorithms namely, Artificial Neural Networks (ANN), Genetic Programming (GP), Support vector Machine Regression (SVMR), Gaussian Process Regression (GPR), and Group Method of Data Handling (GMDH) are separately used to develop individual groundwater salinity prediction models. The performances of these five prediction models are evaluated using various statistical indices. In Phase 2, to counter groundwater pumping uncertainties, different realizations of the initial input-output dataset and the best performing prediction model from models are then used to predict groundwater salinity concentrations in the aquifer. The proposed approach, are applied to an illustrative complex coastal aquifer system. The results show that all Phase 1 are used to construct homogeneous ensemble models. These homogeneous ensembles five machine learning algorithm had reasonably accurate prediction capabilities. However, the GPR-based prediction model outperformed all other models. The results also provide insights on the ability of the suggested best performing GPR model-based homogeneous ensembles to produce improved prediction performance with available data. This approach also allowed incorporation of groundwater pumping uncertainty when developing prediction models. Overall, the study demonstrates the conceptual benefit of ensemble modeling paradigm for improved groundwater salinity prediction, and also its use as a platform for incorporating groundwater pumping uncertainties. This research work has further applications potential in the field of hydrology, as robust ensemble prediction models can be integrated into aquifer management models, which are viable tools for developing robust management strategies for contaminated aquifer systems.

Item Type: Book Chapter
Subjects: T Technology > TA Engineering (General). Civil engineering (General)
Divisions: School of Information Technology, Engineering, Mathematics and Physics (STEMP)
Depositing User: Ms Shalni Sanjana
Date Deposited: 11 Jun 2021 01:36
Last Modified: 11 Jun 2021 02:58

Actions (login required)

View Item View Item