@article{WEI2025, 
author = {Yaqiang WEI and Yiran CHEN and Yuling CHEN and Zi ZHAN and Jiao ZHANG and Jian LIANG and Jialun XIE},
title = {Research progress of AI-driven groundwater numerical modeling},
year = {2025},
journal = {Journal of Northwest University (Natural Science Edition)},
volume = {55},
number = {3},
pages = {647-657},
keywords = {artificial intelligence, groundwater, numerical modeling, contaminants},
url = {https://www.sciopen.com/article/10.16152/j.cnki.xdxbzr.2025-03-012},
doi = {10.16152/j.cnki.xdxbzr.2025-03-012},
abstract = {Groundwater is a critical resource for maintaining ecological security and sustainable development, yet it faces dual challenges of fluctuant quantity and deteriorating quality. While process-based models can describe groundwater flow and contaminant transport, they are highly dependent on precise parameter inputs and computationally intensive, making them less suited for dynamic simulations in complex, heterogeneous environments. Artificial Intelligence (AI) technologies, with their strengths in nonlinear modeling, predictive optimization, and high-dimensional feature extraction, offer novel solutions to overcome bottlenecks in complex system modeling. This article provides a comprehensive review of recent advancements in AI applications for groundwater modeling, covering key areas such as water level prediction, contaminant transport simulation, and remediation optimization. The results indicate that AI models perform well in dynamic forecasting, pollutant identification, and optimization of remediation strategies. Hybrid modeling approaches demonstrate strong robustness in modeling complex variable interactions, while deep learning frameworks show significant advantages in spatiotemporal feature extraction.However, AI models still face challenges such as limited generalization capabilities and a lack of physical consistency. Future research should focus on the following aspects: ① developing multi-scale data fusion and scale-transfer mechanisms to enhance model stability and adaptability; ② Improving the transferability and reusability of same-scale models, with reduced reliance on data from the target site; ③ shifting the paradigm from "big data" to "effective data" to strengthen modeling capabilities under small-sample conditions; ④ embedding physical constraints to improve the reliability and physical consistency of surrogate models; ⑤ constructing intelligent systems that integrate the Internet of Things and edge computing to enable efficient groundwater sensing, modeling, and real-time decision-making, thereby advancing groundwater management into a new era of intelligent operation.}
}