A multi-objective simulation-optimisation model for managing saltwater intrusion in a coastal aquifer of the Pacific Island of Vanuatu
Sharan, Ashneel, Roy, Dilip Kumar, Datta, Bithin, and Lal, Alvin (2026) A multi-objective simulation-optimisation model for managing saltwater intrusion in a coastal aquifer of the Pacific Island of Vanuatu. Journal of Hydrology, 669. 135128.
|
PDF (Published Version)
- Published Version
Available under License Creative Commons Attribution. Download (9MB) | Preview |
Abstract
Groundwater resources in Pacific Island countries face threats from climate change, natural disasters, and human activities, with saltwater intrusion (SWI) being a significant issue in coastal aquifers like Vanuatu’s Tagabe coastal aquifer. This study developed numerical and machine learning models to manage SWI in the Tagabe aquifer. The SEAWAT was used to create a numerical groundwater model, calibrated and validated with salt concentration data from various monitoring locations (MLs). The model generated input (pumping values) and output (salt concentrations) datasets, which were used to train and test twelve machine learning models or surrogate models. These models were ranked using performance criteria and decision theory, and the top five were used to form ensemble models to predict salt concentrations at four MLs. The Pearson coefficient, Index of Agreement, Kling-Gupta Efficiency and Nash Sutcliffe Efficiency coefficient of four ensemble models at ML1 = 0.745, 0.738, 0.497 and 0.384; ML2 = 0.946, 0.950, 0.753 and 0.848; ML3 = -0.028, 0.166, −0.369 and −0.070; ML4 = 0.415, 0.369, −0.135 and −4.032, respectively. The ensemble models performed better than individual surrogate models, with higher predictive accuracy across different MLs. These models were integrated with the Controlled Elitist Multiple Objective Genetic Algorithm (CEMOGA) to create a simulation-optimisation (S-O) management model, providing Pareto optimal solutions for managing SWI. Verification using the SEAWAT-based numerical model showed that the ensemble models’ predictive accuracy was within 7% of the complex numerical model. This study is novel in its use of limited datasets to train twelve machine learning models and the application of ensemble surrogate and S-O management models in Vanuatu, demonstrating their global applicability for managing SWI in coastal aquifers and aiding policymakers in formulating effective groundwater use policies.
| Item ID: | 94308 |
|---|---|
| Item Type: | Article (Research - C1) |
| ISSN: | 0022-1694 |
| Related URLs: | |
| Copyright Information: | © 2026 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). |
| Date Deposited: | 15 Oct 2026 00:21 |
| FoR Codes: | 37 EARTH SCIENCES > 3707 Hydrology > 370703 Groundwater hydrology @ 100% |
| SEO Codes: | 18 ENVIRONMENTAL MANAGEMENT > 1803 Fresh, ground and surface water systems and management > 180305 Ground water quantification, allocation and impact of depletion @ 100% |
| More Statistics |
