Home >

news Help

Publication Information


Title
Japanese: 
English:Integrating MODFLOW and machine learning for detecting optimum groundwater abstraction considering sustainable drawdown and climate changes 
Author
Japanese: Ahmed Makhlouf, Mustafa El-Rawy, 鼎 信次郎, Mona G Ibrahim, Mahmoud Sharaan.  
English: Ahmed Makhlouf, Mustafa El-Rawy, Shinjiro Kanae, Mona G Ibrahim, Mahmoud Sharaan.  
Language English 
Journal/Book name
Japanese: 
English:Journal of Hydrology 
Volume, Number, Page Volume 637        131428
Published date May 25, 2024 
Publisher
Japanese: 
English:Elsevier 
Conference name
Japanese: 
English: 
Conference site
Japanese: 
English: 
Official URL https://www.sciencedirect.com/science/article/pii/S0022169424008230?via%3Dihub
 
DOI https://doi.org/10.1016/j.jhydrol.2024.131428
Abstract Detecting optimum groundwater abstraction is a challenging objective. This study proposes a new approach for identifying the optimum pumping rates by integrating MODFLOW with machine learning (ML). The adopted strategy has been applied to the Oligocene and Eocene aquifers in the western desert of Minia, Egypt. MODFLOW-2005 model has been utilized to simulate various management scenarios, including current and expected climate scenarios of two Representative Concentration Pathways (RCPs): RCP 4.5 and RCP 8.5. The steady-state and transient MODFLOW models were calibrated using the groundwater levels of 37 observation wells with a coefficient of correlation (R2) of 0.986 and 0.988, respectively. The developed model has been run for 80 years, from 2020 to 2100. The results of MODFLOW demonstrated a drawdown in the groundwater level in the two aquifers, about 40, 41, and 42 m, owing to current climate conditions and scenarios of RCP 4.5 and RCP 8.5, respectively, in 2100. Generated data based on the calibrated model has been used to develop the ML models. ML models have been applied to predict the pumping rates using the well coordinates, aquifer type, evapotranspiration, the operating period in days, and drawdown as inputs. For this purpose, four ML models, Gaussian Process Regression (GPR), Support Vector Regression (SVR), Tree, and Artificial Neural Network (ANN)) were developed and evaluated. All ML models demonstrated an effective performance during the training and validation processes, particularly the GPR model, which has the best ability to predict the pumping rates with a correlation coefficient, RMSE, and Mean Absolute Error (MAE) equal to 1.0, 6.53, and 8.04, respectively. Therefore, the GPR model has been utilized to identify the optimum pumping rates. The findings of the GPR model revealed that the optimum abstraction of the Oligocene aquifer is 602266, 625329, and 635061 m3/day, while the Eocene aquifer’s optimal groundwater abstraction is 10229801, 1027906, and 102936 m3/day, considering the current climate, RCP 4.5, and RCP8.5, respectively. Anticipated findings are supposed to be instrumental for decision-makers, offering valuable insights into sustainable groundwater management. The novel integration methodology promotes the efficient use of groundwater, contributing to achieving sustainable development goals (SDGs).

©2007 Tokyo Institute of Technology All rights reserved.