Scholarlink

This study evaluates the performance of Artificial Neural Networks (ANN) and Emotional Artificial Neural Networks (EANN) for simulating groundwater level (GWL) fluctuations in the Tabriz Plain aquifer, northwestern Iran, using K-means clustering. Observation wells were clustered into five hydrogeological groups using K-means clustering, with key inputs including precipitation, temperature, evapotranspiration (ET), normalized difference vegetation index (NDVI), groundwater extraction, and lagged GWL data. Both models exhibited strong performance in stable clusters (1, 2, and 4), with Determination Coefficient (DC) ≥ 0.91, root mean square error (RMSE) ≤ 0.060(m), and correlation coefficients (CC) ≥ 0.95 during testing. EANN outperformed ANN in complex, data-scarce clusters (3: DC = 0.87, RMSE = 0.044(m), CC = 0.93; 5: DC = 0.84, RMSE = 0.044(m), CC = 0.94), owing to its hormonal mechanism that better handles nonlinearity and uncertainty. Clustering enhanced localized accuracy, identifying lagged GWL and precipitation as dominant predictors. These findings underscore superiority of EANN for GWL simulations. Future directions include incorporating additional variables (e.g., recharge), hybrid models (e.g., Wavelet-EANN), and expanded monitoring networks for improved reliability.

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

Clustering-Enhanced ANN and Emotional ANN Models for Groundwater Level Prediction in the Tabriz Plain Aquifer

  • 章节

  • 发表日期：

    2026年

查看全文
  • Vahid Nourani,
  • Elnaz Bayat Khajeh,
  • Soheil Emamalipour,
  • Nardin Jabbarian Paknezhad

摘要

This study evaluates the performance of Artificial Neural Networks (ANN) and Emotional Artificial Neural Networks (EANN) for simulating groundwater level (GWL) fluctuations in the Tabriz Plain aquifer, northwestern Iran, using K-means clustering. Observation wells were clustered into five hydrogeological groups using K-means clustering, with key inputs including precipitation, temperature, evapotranspiration (ET), normalized difference vegetation index (NDVI), groundwater extraction, and lagged GWL data. Both models exhibited strong performance in stable clusters (1, 2, and 4), with Determination Coefficient (DC) ≥ 0.91, root mean square error (RMSE) ≤ 0.060(m), and correlation coefficients (CC) ≥ 0.95 during testing. EANN outperformed ANN in complex, data-scarce clusters (3: DC = 0.87, RMSE = 0.044(m), CC = 0.93; 5: DC = 0.84, RMSE = 0.044(m), CC = 0.94), owing to its hormonal mechanism that better handles nonlinearity and uncertainty. Clustering enhanced localized accuracy, identifying lagged GWL and precipitation as dominant predictors. These findings underscore superiority of EANN for GWL simulations. Future directions include incorporating additional variables (e.g., recharge), hybrid models (e.g., Wavelet-EANN), and expanded monitoring networks for improved reliability.