Scholarlink

<p>Groundwater resource management in regions facing increasing water demand and climate variability requires accurate modeling of spatiotemporal variation of groundwater level. While physics-based models like transient MODFLOW codes are valuable, their computational cost and accuracy limitations hinder their application in complex aquifer systems. This study aims to investigate the spatiotemporal dynamics of groundwater levels and examine how seasonal and geological factors influence flow patterns across three groundwater fields (WF1, WF2, and WF3) within the Quaternary aquifer system in Debrecen, Hungary. This system with its complex geological structure is critical for supplying water for drinking and irrigation purposes. The study employed analytical and data-driven techniques that combine self-organizing maps (SOMs), cross-wavelet transforms (XWT), and deep learning (DL) models. Initially, SOMs and XWT analysis were employed to identify the spatiotemporal relationships between different hydrological and meteorological variables and reveal the synchronization of the groundwater level between different fields. The results showed a general decline in groundwater levels, leading to shifts in groundwater flow direction and intensity. Seasonal changes in groundwater levels were found to be driven by natural (Climate and geology) and anthropogenic (Extraction rates) causes. Subsequently, deep learning models, including bi-directional long short-term memory (Bi-LSTM) and gated recurrent unit (GRU) neural networks optimized through random search, are trained using the temporal hydrological and meteorological data to predict groundwater level fluctuations. The models are evaluated using various performance metrics (MSE, MAE, and R<sup>2</sup>). These models demonstrated significant potential for forecasting future groundwater levels, when adequate input data are available, making them ideal for supporting sustainable groundwater management. This integrated methodology offered a practical alternative to complex physics-based models that can be useful for sustaining groundwater management in the region.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Exploring spatiotemporal groundwater flow patterns in heterogeneous systems: a comprehensive workflow combining multiple machine learning models

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年8月2日

查看全文
  • Musaab A. A. Mohammed,
  • Norbert P. Szabó,
  • Péter Szűcs

摘要

Groundwater resource management in regions facing increasing water demand and climate variability requires accurate modeling of spatiotemporal variation of groundwater level. While physics-based models like transient MODFLOW codes are valuable, their computational cost and accuracy limitations hinder their application in complex aquifer systems. This study aims to investigate the spatiotemporal dynamics of groundwater levels and examine how seasonal and geological factors influence flow patterns across three groundwater fields (WF1, WF2, and WF3) within the Quaternary aquifer system in Debrecen, Hungary. This system with its complex geological structure is critical for supplying water for drinking and irrigation purposes. The study employed analytical and data-driven techniques that combine self-organizing maps (SOMs), cross-wavelet transforms (XWT), and deep learning (DL) models. Initially, SOMs and XWT analysis were employed to identify the spatiotemporal relationships between different hydrological and meteorological variables and reveal the synchronization of the groundwater level between different fields. The results showed a general decline in groundwater levels, leading to shifts in groundwater flow direction and intensity. Seasonal changes in groundwater levels were found to be driven by natural (Climate and geology) and anthropogenic (Extraction rates) causes. Subsequently, deep learning models, including bi-directional long short-term memory (Bi-LSTM) and gated recurrent unit (GRU) neural networks optimized through random search, are trained using the temporal hydrological and meteorological data to predict groundwater level fluctuations. The models are evaluated using various performance metrics (MSE, MAE, and R2). These models demonstrated significant potential for forecasting future groundwater levels, when adequate input data are available, making them ideal for supporting sustainable groundwater management. This integrated methodology offered a practical alternative to complex physics-based models that can be useful for sustaining groundwater management in the region.