Scholarlink

<p>In this study, two time-series statistical models, autoregressive integrated moving average (ARIMA) and generalized autoregressive conditional heteroskedasticity (GARCH), and one deep-learning model, a stacked long short-term memory (LSTM) neural network, were applied to evaluate groundwater chloride concentration predictability in Guam, a western Pacific island. The analysis used 42&#xa0;years (1980–2021) of chloride concentration data from two coastal wells: low (F11) and high (F04) variance, to compare data characteristics and model performance. LSTM model showed strong predictive performance with 1,000 epochs, 32 blocks, two hidden layers, and 0.7 training ratio. Since sea-level rise began around 1993, three chloride dataset scenarios were analyzed using the LSTM model to assess pre- and post-onset data variability. Performance was evaluated using relative mean squared error (MSE) and the training-to-test relative MSE ratio, where values closer to 1 indicate lower variability. The F04 dataset showed a clear post-1993 shift with a ratio of 0.9262 (k-fold, <i>k</i> = <i>1</i>) indicating a strong influence of sea-level rise, whereas F11 showed a lower ratio (0.5192), suggesting weaker temporal change. This result indicates that high-variance datasets better reflect the effects of sea-level rise, while low-variance datasets may be more controlled by hydrogeological properties, such as hydraulic conductivity, or rainfall-driven recharge. This study highlights that incorporating hydrogeological context and advanced error metrics with physically interpretable variables improves the reliability of statistical and machine learning analyses for complex field-based groundwater quality datasets. Relative MSE ratio approach enables interpretation of complex environmental stressors by analyzing variability differences through deep learning analysis.</p> Graphical Abstract <p></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

Performance Assessment of Statistical and Deep Learning Models for Predicting Chloride Trends in Island Groundwater Systems Influenced by Saltwater Intrusion and Sea Level Rise

  • 期刊论文

  • 发表日期：

    2026年6月20日

查看全文
  • Yong Sang Kim,
  • Meejoung Kim,
  • Ujwalkumar D. Patil,
  • Byoungyong Lee

摘要

In this study, two time-series statistical models, autoregressive integrated moving average (ARIMA) and generalized autoregressive conditional heteroskedasticity (GARCH), and one deep-learning model, a stacked long short-term memory (LSTM) neural network, were applied to evaluate groundwater chloride concentration predictability in Guam, a western Pacific island. The analysis used 42 years (1980–2021) of chloride concentration data from two coastal wells: low (F11) and high (F04) variance, to compare data characteristics and model performance. LSTM model showed strong predictive performance with 1,000 epochs, 32 blocks, two hidden layers, and 0.7 training ratio. Since sea-level rise began around 1993, three chloride dataset scenarios were analyzed using the LSTM model to assess pre- and post-onset data variability. Performance was evaluated using relative mean squared error (MSE) and the training-to-test relative MSE ratio, where values closer to 1 indicate lower variability. The F04 dataset showed a clear post-1993 shift with a ratio of 0.9262 (k-fold, k = 1) indicating a strong influence of sea-level rise, whereas F11 showed a lower ratio (0.5192), suggesting weaker temporal change. This result indicates that high-variance datasets better reflect the effects of sea-level rise, while low-variance datasets may be more controlled by hydrogeological properties, such as hydraulic conductivity, or rainfall-driven recharge. This study highlights that incorporating hydrogeological context and advanced error metrics with physically interpretable variables improves the reliability of statistical and machine learning analyses for complex field-based groundwater quality datasets. Relative MSE ratio approach enables interpretation of complex environmental stressors by analyzing variability differences through deep learning analysis.

Graphical Abstract