Document Type : Research Paper

• منابع آب

EXTENDED ABSTRACT

Introduction:

Effective groundwater resource management in arid and semi-arid regions is paramount, especially in the context of climate change. The depletion of groundwater levels (GWL) due to excessive extraction, declining precipitation, and rising temperatures poses severe challenges for agricultural sustainability and potable water availability. Predicting GWL variations is crucial for ensuring long-term water security. Recent advancements in machine learning and climate modeling have significantly enhanced prediction accuracy, facilitating informed decision-making in water resource management. This study employs the Random Forest (RF) algorithm in conjunction with CMIP6 climate data to predict GWL variations in the Eslamabad-e-Gharb plain, incorporating the K-Means clustering method to enhance model precision.

Materials and Methods:

The study area, Eslamabad-e-Gharb plain, situated in western Iran, experiences a semi-humid cold climate with an average annual precipitation of 463 mm. Data were collected from 20 observation wells spanning the period 1997–2014. Historical GWL data were sourced from the national water data repository, while soil moisture data from MERRA2 and future climate projections from CMIP6 models were integrated into the analysis.

The K-Means clustering technique was utilized to classify observation wells into three distinct clusters: Jangeh, Barfabab, and Bureg. The RF algorithm was employed to identify the most relevant CMIP6 soil moisture models for predicting GWL under three future climate scenarios: SSP1-2.6, SSP2-4.5, and SSP5-8.5. The models were trained using 80% of the dataset, with the remaining 20% allocated for validation. Model performance was evaluated using Nash-Sutcliffe Efficiency (NSE), correlation coefficient (R), mean absolute error (MAE), and normalized root mean square error (NRMSE).

Results and Discussion:

During the observational period (1997–2014), all study regions exhibited a pronounced downward trend in GWL. The most significant decline was recorded in Barfabab, with an annual reduction of 1 meter, followed by Bureg (0.715 m/year) and Jangeh (0.59 m/year).

Under the SSP1-2.6 scenario, GWL is projected to continue its decline in the near future, with Bureg experiencing a substantial reduction (-0.21 m/year). In the mid-term period (2051–2075), Jangeh is anticipated to undergo the most pronounced decrease (-0.14 m/year). The SSP2-4.5 scenario suggests a significant long-term decline across all regions, with Bureg exhibiting the steepest decrease (-0.16 m/year).

Future projections under SSP5-8.5 indicate a notable GWL increase in Jangeh (3 to 3.5 meters) in the late 21st century (2076–2100). Conversely, under SSP1-2.6, Bureg is expected to experience the most substantial decline (3.5 to 4 meters) in spring and summer during the mid-21st century. The low-emission scenarios (SSP1-2.6 and SSP2-4.5) suggest a more stabilized GWL condition in the distant future.

The RF model exhibited high predictive accuracy, with an R-value of 0.97 and NSE ranging from 0.89 to 0.98. The NorESM2-MM model emerged as the most reliable climate model for GWL prediction, enhancing forecasting precision by up to 99.5%. The findings underscore the pivotal role of climate change in GWL depletion within the Eslamabad-e-Gharb plain. Even under the most optimistic climate scenario (SSP1-2.6), GWL is projected to decline further, highlighting the necessity of implementing sustainable water management strategies. The study reinforces that machine learning models, particularly RF, surpass traditional numerical models in GWL prediction.

The integration of K-Means clustering with RF enhanced model precision by selecting the most relevant climate predictors for each region. These results align with previous research, indicating that hybrid machine learning approaches significantly improve groundwater modeling. The exceptional predictive capability of RF supports its potential for future groundwater management applications.

Conclusion:

This study demonstrates that RF, when integrated with CMIP6 climate data, serves as an effective tool for predicting future GWL variations in the Eslamabad-e-Gharb plain. The findings indicate that during the observational period, Barfabab exhibited the most favorable GWL conditions (8.7–10.2 meters), whereas Bureg experienced the least favorable levels (15.5–17.3 meters). Water availability peaked in spring and reached its lowest levels in fall. Projections suggest that under SSP5-8.5, Jangeh will experience the highest GWL increase (3–3.5 meters) in fall (2076–2100), whereas SSP1-2.6 predicts the most severe decline in Bureg (3.5–4 meters) during spring and summer. The SSP1-2.6 and SSP2-4.5 scenarios suggest more stable GWL conditions over the long term.

Overall, GWL displayed a declining trend, with Barfabab experiencing the most substantial drop (1 meter per year). Under SSP1-2.6, GWL is projected to decline in Bureg (-0.21 m/year) in the near future and in Jangeh (-0.14 m/year) in the mid-future. Under SSP2-4.5, a significant long-term decline is anticipated across all regions, with Bureg experiencing the steepest reduction (-0.16 m/year). The RF model demonstrated outstanding performance (R = 0.97, NSE = 0.89–0.98), while the NorESM2-MM model significantly enhanced prediction accuracy up to 99.5%. These findings emphasize the critical role of advanced machine learning methodologies in groundwater resource management, providing a robust framework for future hydrological assessments and policy-making initiatives.

Author Contributions

Conceptualization, Kobra Soltani, Seyed Ehsan Fatemi and Jafar Masoompour Samakosh; methodology, Seyed Ehsan Fatemi and Jafar Masoompour Samakosh; software, Seyed Ehsan Fatemi and Kobra Soltani; validation, Jafar Masoompour Samakosh and Maryam Hafezparast Mavadat; analysis, Kobra Soltani and Seyed Ehsan Fatemi; investigation, Kobra Soltani and Seyed Ehsan Fatemi; resources, Kobra Soltani and Maryam Hafezparast Mavadat; writing—original draft preparation, Kobra Soltani;writing—review and editing, Jafar Masoompour Samakosh, Seyed Ehsan Fatemi and Maryam Hafezparast.

All authors have read and agreed to the published version of the manuscript.

Data Availability Statement

Data can be sent from the corresponding author by email upon request.

Acknowledgements

The authors would like to thank all participants of the present study.

Ethical considerations

The authors avoided data fabrication, falsification, plagiarism, and misconduct.

Conflict of interest

The author declares no conflict of interest.