Enhancing groundwater predictions by incorporating response lag effects in machine learning models

Groundwater is essential for water resources but faces over-extraction and supply-demand imbalance. Precisely comprehending alterations in groundwater is crucial for sustainable development. Groundwater levels demonstrate a delayed reaction to meteorological circumstances, frequently neglected in cu...

Full description

Bibliographic Details
Main Authors: Yang, F., Zhang, X., Yang, J., Zhang, J., Dai, Q., Zhu, S.
Format: Journal Article
Language:Inglés
Published: IWA Publishing 2025
Online Access:https://hdl.handle.net/10568/173583
_version_ 1855519264442679296
author Yang, F.
Zhang, X.
Yang, J.
Zhang, J.
Dai, Q.
Zhu, S.
author_browse Dai, Q.
Yang, F.
Yang, J.
Zhang, J.
Zhang, X.
Zhu, S.
author_facet Yang, F.
Zhang, X.
Yang, J.
Zhang, J.
Dai, Q.
Zhu, S.
author_sort Yang, F.
collection Repository of Agricultural Research Outputs (CGSpace)
description Groundwater is essential for water resources but faces over-extraction and supply-demand imbalance. Precisely comprehending alterations in groundwater is crucial for sustainable development. Groundwater levels demonstrate a delayed reaction to meteorological circumstances, frequently neglected in current research, diminishing predictive accuracy. This study investigates the lag effect of precipitation and evapotranspiration on groundwater forecasting in Hebei Province, China. We performed a lag analysis utilizing long-term data to ascertain correlations between groundwater levels and climatic variables. Two groundwater prediction models using the random forest algorithm were developed, one incorporating the lag effect and the other excluding it. The Shapley Additive exPlanations (SHAP) method assessed the significance of each element and its influence on model variations. Research reveals a notable lag in groundwater response: shallow groundwater reacts to precipitation after 4.55 months and to evapotranspiration after 9.21 months; deep groundwater responds after 5.91 and 9.63 months, respectively. The inclusion of the lag effect resulted in higher accuracy of the model, with an average reduction of 35.7% in MAE and 18.20% in RMSE. The improved model more accurately captured the influence of meteorological factors on groundwater levels, potentially providing more scientific decision support for the rational allocation and sustainable use of water resources.
format Journal Article
id CGSpace173583
institution CGIAR Consortium
language Inglés
publishDate 2025
publishDateRange 2025
publishDateSort 2025
publisher IWA Publishing
publisherStr IWA Publishing
record_format dspace
spelling CGSpace1735832025-10-26T13:02:00Z Enhancing groundwater predictions by incorporating response lag effects in machine learning models Yang, F. Zhang, X. Yang, J. Zhang, J. Dai, Q. Zhu, S. Groundwater is essential for water resources but faces over-extraction and supply-demand imbalance. Precisely comprehending alterations in groundwater is crucial for sustainable development. Groundwater levels demonstrate a delayed reaction to meteorological circumstances, frequently neglected in current research, diminishing predictive accuracy. This study investigates the lag effect of precipitation and evapotranspiration on groundwater forecasting in Hebei Province, China. We performed a lag analysis utilizing long-term data to ascertain correlations between groundwater levels and climatic variables. Two groundwater prediction models using the random forest algorithm were developed, one incorporating the lag effect and the other excluding it. The Shapley Additive exPlanations (SHAP) method assessed the significance of each element and its influence on model variations. Research reveals a notable lag in groundwater response: shallow groundwater reacts to precipitation after 4.55 months and to evapotranspiration after 9.21 months; deep groundwater responds after 5.91 and 9.63 months, respectively. The inclusion of the lag effect resulted in higher accuracy of the model, with an average reduction of 35.7% in MAE and 18.20% in RMSE. The improved model more accurately captured the influence of meteorological factors on groundwater levels, potentially providing more scientific decision support for the rational allocation and sustainable use of water resources. 2025-02-01 2025-03-12T06:29:37Z 2025-03-12T06:29:37Z Journal Article https://hdl.handle.net/10568/173583 en Open Access IWA Publishing Yang, F.; Zhang, X.; Yang, J.; Zhang, J.; Dai, Q.; Zhu, S. 2025. Enhancing groundwater predictions by incorporating response lag effects in machine learning models. Journal of Hydroinformatics, 27(2):338-356. [doi:https://doi.org/10.2166/hydro.2025.295]
spellingShingle Yang, F.
Zhang, X.
Yang, J.
Zhang, J.
Dai, Q.
Zhu, S.
Enhancing groundwater predictions by incorporating response lag effects in machine learning models
title Enhancing groundwater predictions by incorporating response lag effects in machine learning models
title_full Enhancing groundwater predictions by incorporating response lag effects in machine learning models
title_fullStr Enhancing groundwater predictions by incorporating response lag effects in machine learning models
title_full_unstemmed Enhancing groundwater predictions by incorporating response lag effects in machine learning models
title_short Enhancing groundwater predictions by incorporating response lag effects in machine learning models
title_sort enhancing groundwater predictions by incorporating response lag effects in machine learning models
url https://hdl.handle.net/10568/173583
work_keys_str_mv AT yangf enhancinggroundwaterpredictionsbyincorporatingresponselageffectsinmachinelearningmodels
AT zhangx enhancinggroundwaterpredictionsbyincorporatingresponselageffectsinmachinelearningmodels
AT yangj enhancinggroundwaterpredictionsbyincorporatingresponselageffectsinmachinelearningmodels
AT zhangj enhancinggroundwaterpredictionsbyincorporatingresponselageffectsinmachinelearningmodels
AT daiq enhancinggroundwaterpredictionsbyincorporatingresponselageffectsinmachinelearningmodels
AT zhus enhancinggroundwaterpredictionsbyincorporatingresponselageffectsinmachinelearningmodels