Kernel-based predictive framework for groundwater level forecast in semiarid areas : a CEEMDAN-based GWO-KELM hybrid mode
Shahnazi, Saman and Hashemi, Hossein LU
(2026)
p.85-98
- Abstract
Accurate prediction of groundwater levels (GWL) is essential for effective and sustainable water resource management, particularly in arid and semiarid regions where excessive extraction and climate change significantly accelerate aquifer depletion. The nonstationary characteristics of groundwater data pose significant challenges to the performance and reliability of conventional machine learning methods. To address this challenge and improve GWL prediction, this chapter introduces a hybrid framework that integrates complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), the gray wolf optimizer (GWO), and the kernel extreme learning machine (KELM). The proposed CEEMDAN-GWO-KELM model was evaluated using 25 years of... (More)
Accurate prediction of groundwater levels (GWL) is essential for effective and sustainable water resource management, particularly in arid and semiarid regions where excessive extraction and climate change significantly accelerate aquifer depletion. The nonstationary characteristics of groundwater data pose significant challenges to the performance and reliability of conventional machine learning methods. To address this challenge and improve GWL prediction, this chapter introduces a hybrid framework that integrates complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), the gray wolf optimizer (GWO), and the kernel extreme learning machine (KELM). The proposed CEEMDAN-GWO-KELM model was evaluated using 25 years of monthly GWL data (1996-2021), collected from 20 observation wells in the Marand Plain in Iran, a semiarid region facing severe aquifer depletion. The results indicated that incorporating CEEMDAN as a preprocessing step significantly enhances the robustness of predictions, achieving up to a 52% reduction in errors compared to the standalone GWO-KELM model. The optimal input configuration, comprising GWLₜ₋₁, GWLₜ₋₂, and GWLₜ₋₃, yielded strong predictive performance during the test phase, with a correlation coefficient (R) of 0.982, a Nash-Sutcliffe Efficiency of 0.942, and a root mean square error of 0.016. Among the evaluated kernel functions, the Polynomial kernel demonstrated superior performance compared to the others.
(Less)
- author
- Shahnazi, Saman
and Hashemi, Hossein
LU
- organization
- publishing date
- 2026-01-01
- type
- Chapter in Book/Report/Conference proceeding
- publication status
- published
- subject
- keywords
- environmental management tool, groundwater, groundwater level prediction, Hybrid predictive model, iran, nonstationarity
- host publication
- Hydrological Insights : Synergizing Groundwater Models, Remote Sensing, and AI for Water Sustainability - Synergizing Groundwater Models, Remote Sensing, and AI for Water Sustainability
- pages
- 14 pages
- publisher
- Elsevier
- external identifiers
-
- scopus:105032949084
- ISBN
- 9780443363955
- 9780443363948
- DOI
- 10.1016/B978-0-443-36394-8.00014-5
- project
- The United Nations University Hub: Water in a Changing Environment (WICE)
- language
- English
- LU publication?
- yes
- additional info
- Publisher Copyright: © 2026 Elsevier Inc. All rights reserved.
- id
- 95988113-6875-4678-8223-f85579d5bf03
- date added to LUP
- 2026-05-19 14:49:56
- date last changed
- 2026-09-24 06:32:09
@inbook{95988113-6875-4678-8223-f85579d5bf03,
abstract = {{<p>Accurate prediction of groundwater levels (GWL) is essential for effective and sustainable water resource management, particularly in arid and semiarid regions where excessive extraction and climate change significantly accelerate aquifer depletion. The nonstationary characteristics of groundwater data pose significant challenges to the performance and reliability of conventional machine learning methods. To address this challenge and improve GWL prediction, this chapter introduces a hybrid framework that integrates complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), the gray wolf optimizer (GWO), and the kernel extreme learning machine (KELM). The proposed CEEMDAN-GWO-KELM model was evaluated using 25 years of monthly GWL data (1996-2021), collected from 20 observation wells in the Marand Plain in Iran, a semiarid region facing severe aquifer depletion. The results indicated that incorporating CEEMDAN as a preprocessing step significantly enhances the robustness of predictions, achieving up to a 52% reduction in errors compared to the standalone GWO-KELM model. The optimal input configuration, comprising GWLₜ₋₁, GWLₜ₋₂, and GWLₜ₋₃, yielded strong predictive performance during the test phase, with a correlation coefficient (R) of 0.982, a Nash-Sutcliffe Efficiency of 0.942, and a root mean square error of 0.016. Among the evaluated kernel functions, the Polynomial kernel demonstrated superior performance compared to the others.</p>}},
author = {{Shahnazi, Saman and Hashemi, Hossein}},
booktitle = {{Hydrological Insights : Synergizing Groundwater Models, Remote Sensing, and AI for Water Sustainability}},
isbn = {{9780443363955}},
keywords = {{environmental management tool, groundwater; groundwater level prediction; Hybrid predictive model; iran; nonstationarity}},
language = {{eng}},
month = {{01}},
pages = {{85--98}},
publisher = {{Elsevier}},
title = {{Kernel-based predictive framework for groundwater level forecast in semiarid areas : a CEEMDAN-based GWO-KELM hybrid mode}},
url = {{http://dx.doi.org/10.1016/B978-0-443-36394-8.00014-5}},
doi = {{10.1016/B978-0-443-36394-8.00014-5}},
year = {{2026}},
}