Groundwater dynamics clustering and prediction based on grey relational analysis and LSTM model: A case study in Beijing Plain, China

Journal of Hydrology Regional Studies, Год журнала: 2024, Номер 56, С. 102011 - 102011

Опубликована: Окт. 15, 2024

Язык: Английский

---

Training deep learning models with a multi-station approach and static aquifer attributes for groundwater level simulation: what is the best way to leverage regionalised information?

Hydrology and earth system sciences, Год журнала: 2025, Номер 29(4), С. 841 - 861

Опубликована: Фев. 18, 2025

Abstract. In this study, we use deep learning models with advanced variants of recurrent neural networks, specifically long short-term memory (LSTM), gated unit (GRU), and bidirectional LSTM (BiLSTM), to simulate large-scale groundwater level (GWL) fluctuations in northern France. We develop multi-station collective training for GWL simulations, using dynamic variables (i.e. climatic) static basin characteristics. This approach can incorporate features cover more reservoir heterogeneities the study area. Further, investigated performance relevant feature extraction techniques such as clustering wavelet transform decomposition simplify network regionalised information. Several modelling tests were conducted. Models trained on different types GWL, clustered based spectral properties, performed significantly better than whole dataset. Clustering-based reduces complexity data targets information efficiently. Applying without prior lead preferentially learn dominant behaviour, ignoring unique local variations. respect, pre-processing was found partially compensate clustering, bringing out common temporal characteristics shared by all available time series even when these are “hidden” (e.g. if their amplitude is too small). When employed along a technique improves model performances, particularly GWLs dominated low-frequency interannual decadal advances our understanding simulation learning, highlighting importance approaches, potential pre-processing, value incorporating attributes.

Язык: Английский

---

Direct impact of climate change on groundwater levels in the Iberian Peninsula

The Science of The Total Environment, Год журнала: 2025, Номер 970, С. 179009 - 179009

Опубликована: Март 1, 2025

The Iberian Peninsula is a water-scarce region that increasingly reliant on groundwater. Climate change expected to exacerbate this situation due projected irregular precipitation patterns and frequent droughts. Here, we utilised convolutional neural networks (CNNs) assess the direct effect of climate groundwater levels, using monthly meteorological data historical levels from 3829 wells. We considered temperature antecedent cumulative over 3, 6, 12, 18, 24, 36 months account for recharge time lag between level changes. Based CNNs performance, 92 location-specific models were retained further analysis, representing wells spatially distributed throughout peninsula. used influence future considering an ensemble eight combinations general regional under RCP4.5 RCP8.5 scenarios. Under RCP4.5, average annual increase 1.7 °C 5.2 % decrease in will result approximately 15 experiencing >1-m decline reference period [1986-2005] long-term [2080-2100]. RCP8.5, with 3.8 20.2 same periods, 40 are experience water drop >1 m. Notably, 72 wells, main driver, implying evaporation has greater impact levels. Effective management strategies should be implemented limit overexploitation reserves improve resilience

Язык: Английский

---

Impact of Climate Change on Groundwater Level Changes: An Evaluation Based on Deep Neural Networks

Applied Computational Intelligence and Soft Computing, Год журнала: 2025, Номер 2025(1)

Опубликована: Янв. 1, 2025

Climate change has a substantial influence on groundwater levels (GWLs), which are critical for agriculture, safe drinking water, and ecosystem health, essential to successful water resource management adaptation strategies. Recently, there been an increase in the use of machine learning (ML) deep (DL) models hydrogeology estimate GWL monitoring wells. This study presents novel technique predicting changes that uses three independent datasets: historical climatic variables (CVs) data such as rainfall temperature influencing dynamics. In our experimental research, models’ prediction output real‐world datasets ensures model’s significant patterns recorded while taking into account noise data, resulting perfect balance bias variance. The DL results show score root mean square error (RMSE) between 2.20 12.40 coefficient determination ( R ‐squared 0.84–0.99), showing improvement RMSE absolute (MAE) testing validation categories, when compared current state‐of‐the‐art methods. improves understanding modeling provides decision‐makers with reliable tool controlling change. advances environmental by exhibiting methodological complexity emphasizes importance comprehensive analysis management.

Язык: Английский

---

Are Deep Learning Models in Hydrology Entity Aware?

Geophysical Research Letters, Год журнала: 2025, Номер 52(6)

Опубликована: Март 22, 2025

Abstract Hydrology is experiencing a shift from process‐based toward deep learning (DL) models. Entity‐aware (EA) DL models with static features (predominantly physiographic proxies) merged to dynamic forcing show significant performance improvements. However, recent studies challenge the notion that combining forcings attributes make such entity aware, suggesting are not effectively leveraged for generalization. We examine awareness using state‐of‐the‐art Long‐Short Term Memory (LSTM) networks and CAMELS‐US data set. compare EA provided ablated variants inputs. Findings suggest superior of primarily driven by information meteorological data, limited contributions features, particularly when tested out‐of‐sample. These results previously held assumptions regarding how proxies contribute generalization ability in Models, highlighting need new approaches robust

Язык: Английский

---

Training deep learning models with a multi-station approach and static aquifer attributes for groundwater level simulation: what’s the best way to leverage regionalised information?

Опубликована: Май 13, 2024

Abstract. In this study, we used deep learning models with recurrent structure neural networks to simulate large-scale groundwater level (GWL) fluctuations in northern France. We developed a multi-station collective training for GWL simulations, using both “dynamic” variables (i.e. climatic) and static aquifer characteristics. This approach offers the possibility of incorporating dynamic features cover more reservoir heterogeneities study area. Further, investigated performance relevant feature extraction techniques such as clustering wavelet transform decomposition, intending simplify network regionalised information. Several modelling tests were conducted. Models specifically trained on different types GWL, clustered based spectral properties data, performed significantly better than whole dataset. Clustering-based reduces complexity data targets information efficiently. Applying without prior can lead learn dominant station behavior preferentially, ignoring unique local variations. respect, pre-processing was found partially compensate clustering, bringing out common temporal characteristics shared by all available time series even when these are “hidden” because too small amplitude. When employed along thanks its capability capturing essential across scales (high low), decomposition technique provided significant improvement model performance, particularly GWLs dominated low-frequency advances our understanding simulation learning, highlighting importance approaches, potential preprocessing, value attributes.

Язык: Английский

---

Assessing groundwater level modelling using a 1-D convolutional neural network (CNN): linking model performances to geospatial and time series features

Hydrology and earth system sciences, Год журнала: 2024, Номер 28(19), С. 4407 - 4425

Опубликована: Окт. 7, 2024

Abstract. Groundwater level (GWL) forecasting with machine learning has been widely studied due to its generally accurate results and low input data requirements. Furthermore, models for this purpose can be set up trained quickly compared the effort required process-based numerical models. Despite demonstrating high performance at specific locations, applying same model architecture multiple sites across a regional area lead varying accuracies. The reasons behind discrepancy in have scarcely examined previous studies. Here, we explore relationship between geospatial time series features of sites. Using precipitation (P) temperature (T) as predictors, monthly groundwater levels approximately 500 observation wells Lower Saxony, Germany, 1-D convolutional neural network (CNN) fixed hyperparameters tuned each individually. GWL observations range from 21 71 years, resulting variable test training dataset ranges. performances are evaluated against selected characteristics (e.g. land cover, distance waterworks, leaf index) autocorrelation, flat spots, number peaks) using Pearson correlation coefficients. Results indicate that is negatively influenced near waterworks densely vegetated areas. Longer subsequences measurements above or below mean impact accuracy. Besides, containing more irregular patterns higher peaks might performances, possibly closer link dynamics. As deep known black-box missing understanding physical processes, our work provides new insights into how input–output model.

Язык: Английский

---

Associations between deep learning runoff predictions and hydrogeological conditions in Australia

Journal of Hydrology, Год журнала: 2024, Номер unknown, С. 132569 - 132569

Опубликована: Дек. 1, 2024

Язык: Английский

---

Comment on egusphere-2024-794

Опубликована: Июнь 11, 2024

Abstract. In this study, we used deep learning models with recurrent structure neural networks to simulate large-scale groundwater level (GWL) fluctuations in northern France. We developed a multi-station collective training for GWL simulations, using both “dynamic” variables (i.e. climatic) and static aquifer characteristics. This approach offers the possibility of incorporating dynamic features cover more reservoir heterogeneities study area. Further, investigated performance relevant feature extraction techniques such as clustering wavelet transform decomposition, intending simplify network regionalised information. Several modelling tests were conducted. Models specifically trained on different types GWL, clustered based spectral properties data, performed significantly better than whole dataset. Clustering-based reduces complexity data targets information efficiently. Applying without prior can lead learn dominant station behavior preferentially, ignoring unique local variations. respect, pre-processing was found partially compensate clustering, bringing out common temporal characteristics shared by all available time series even when these are “hidden” because too small amplitude. When employed along thanks its capability capturing essential across scales (high low), decomposition technique provided significant improvement model performance, particularly GWLs dominated low-frequency advances our understanding simulation learning, highlighting importance approaches, potential preprocessing, value attributes.

Язык: Английский

---

Testing the Possibilities and Limits of Groundwater Level Prediction Using Global Model Frameworks with Deep Learning on a Worldwide Scale

Опубликована: Янв. 1, 2024

Язык: Английский

---