Defining and predicting groundwater drought dynamics in a temperate maritime region using Standardised Groundwater Index (SGI) and ensemble machine learning
- 1School of Transport and Civil Engineering, Technological University Dublin, Ireland
- 2Health and Sustainability Research Hub, Technological University Dublin, Ireland; Corresponding author.
Abstract
Groundwater drought events in temperate maritime regions have historically occurred relatively rarely and subsequently received limited research attention despite their significant potential socio-economic and ecological implications. Accordingly, understanding of groundwater drought processes and their hydrological controls remains limited in these regions. However, recent events (2018 European drought, 2025 summer drought) and updated global climate projections indicate that groundwater droughts will likely become more frequent, prolonged and severe. The current study sought to address this gap by developing an ensemble machine learning framework for the detection and prediction of groundwater drought events across a national groundwater monitoring network (100 monitoring stations). The Standardised Groundwater Index (SGI) was used to characterise drought occurrence, severity, and duration, using the 2018 European drought event for validation. Overall, 74 severe (−1.5 ≥ SGI > 2.0) drought episodes were identified across 52 monitoring stations during 2010–2023. Ensemble ML classifiers optimised for class-imbalanced data (e.g., SMOTEBoost) were implemented to optimise drought detection. Model performance varied spatially, reflecting hydrogeological and topographical setting. Consequently, a subset-based modelling approach was implemented, resulting in two groundwater drought types. Subset-1 comprised drought-sensitive sites predominantly underlain by poorly productive aquifers, whereas Subset-2 represented groundwater systems associated with higher-storage unconsolidated aquifers. The RUSBoost model achieved the highest predictive performance, with an overall accuracy of 88%, AUC-ROC of 0.95, and sensitivity of 91% for drought identification. The proposed modelling framework is likely transferable to similar climate regions and provides a robust decision support tool to inform groundwater management, early-warning systems, and long-term water resource planning.
Related
- ORCHIDEE-Hg v1.0 (rev. 9721): A new global land-surface model for mercury dynamics – Part 1: Model description and evaluation in temperate regions
- GROW-Qual: A global observational groundwater quality dataset linked to water table dynamics and Earth system variables
- A Transformer-Based Surrogate for Distributed Groundwater Dynamics with Zero-Shot Cross-Basin Transfer and Precipitation-Regime Data Augmentation
- Summer oxygen dynamics in the bottom waters of a wide, temperate shelf sea
- Monitoring and forecasting groundwater storage dynamics in arid Wadi systems using GRACE/GRACE-FO satellite observations and XGBoost machine learning: Wadi El-Assiuti, Egypt (2003–2030)
BibTeX
@article{mohamed2026defining,
title = {{Defining and predicting groundwater drought dynamics in a temperate maritime region using Standardised Groundwater Index (SGI) and ensemble machine learning}},
author = {Tarig Mohamed and Ahmed Nasr and Paul Hynds},
journal = {Ecological Informatics},
year = {2026},
month = oct,
doi = {10.1016/j.ecoinf.2026.104107},
url = {https://doi.org/10.1016/j.ecoinf.2026.104107},
note = {CC BY},
}
RIS
TY - JOUR
TI - Defining and predicting groundwater drought dynamics in a temperate maritime region using Standardised Groundwater Index (SGI) and ensemble machine learning
AU - Tarig Mohamed
AU - Ahmed Nasr
AU - Paul Hynds
PY - 2026
DA - 2026/10/31
JO - Ecological Informatics
DO - 10.1016/j.ecoinf.2026.104107
UR - https://doi.org/10.1016/j.ecoinf.2026.104107
AB - Groundwater drought events in temperate maritime regions have historically occurred relatively rarely and subsequently received limited research attention despite their significant potential socio-economic and ecological implications. Accordingly, understanding of groundwater drought processes and their hydrological controls remains limited in these regions. However, recent events (2018 European drought, 2025 summer drought) and updated global climate projections indicate that groundwater droughts will likely become more frequent, prolonged and severe. The current study sought to address this gap by developing an ensemble machine learning framework for the detection and prediction of groundwater drought events across a national groundwater monitoring network (100 monitoring stations). The Standardised Groundwater Index (SGI) was used to characterise drought occurrence, severity, and duration, using the 2018 European drought event for validation. Overall, 74 severe (−1.5 ≥ SGI > 2.0) drought episodes were identified across 52 monitoring stations during 2010–2023. Ensemble ML classifiers optimised for class-imbalanced data (e.g., SMOTEBoost) were implemented to optimise drought detection. Model performance varied spatially, reflecting hydrogeological and topographical setting. Consequently, a subset-based modelling approach was implemented, resulting in two groundwater drought types. Subset-1 comprised drought-sensitive sites predominantly underlain by poorly productive aquifers, whereas Subset-2 represented groundwater systems associated with higher-storage unconsolidated aquifers. The RUSBoost model achieved the highest predictive performance, with an overall accuracy of 88%, AUC-ROC of 0.95, and sensitivity of 91% for drought identification. The proposed modelling framework is likely transferable to similar climate regions and provides a robust decision support tool to inform groundwater management, early-warning systems, and long-term water resource planning.
SN - 1574-9541
SN - 1878-0512
ER -