| Title |
Forecasting change in groundwater storage: A multilayer perceptron model leveraging system dynamics simulation data |
| Authors |
Eslamifar, Gholamreza ; Zaslavsky, Ilya ; Balali, Hamid ; Samalavičius, Vytautas ; Fernald, Alexander |
| DOI |
10.1016/j.ejrh.2026.103398 |
| Full Text |
|
| Is Part of |
Journal of hydrology: regional studies.. Amsterdam : Elsevier. 2026, vol. 65, art. no. 103398, p. [1-17].. ISSN 2214-5818 |
| Keywords [eng] |
groundwater modeling ; machine learning ; artificial neural network ; system dynamics ; lower Rio Grande, water ; agriculture ; water resources management ; MLP |
| Abstract [eng] |
Study region This study focuses on the Lower Rio Grande (LRG) region of Southern New Mexico, a semi-arid area facing significant water scarcity challenges. Groundwater serves as a critical resource for agriculture, domestic use, and industry in the region, necessitating robust management and predictive tools. Study focus The research develops a novel groundwater modeling approach by integrating System Dynamics (SD) and Machine Learning (ML) techniques. A validated SD model of the LRG provided simulation data, which was used to train an ML model employing a feedforward Multilayer Perceptron (MLP) as an artificial neural network (ANN) technique. This approach simplifies groundwater dynamics by concentrating on key variables identified through correlation analysis and simulation results. The model achieved a training RMSLE of 0.031 and R-squared of 0.77, with testing results showing RMSLE of 0.036 and R-squared of 0.71, demonstrating predictive reliability. New hydrological insights for the region The proposed model enables accurate forecasting of groundwater storage changes while reducing reliance on extensive data inputs. It pioneers using SD simulation data as a data augmentation method for MLP model training, enhancing predictive capabilities. This integrated methodology supports informed groundwater management and policy-making, offering a transferable framework for other hydrologically similar regions. |
| Published |
Amsterdam : Elsevier |
| Type |
Journal article |
| Language |
English |
| Publication date |
2026 |
| CC license |
|
