Wang et al. (2026) Regional transitions in factors associated with groundwater recovery: Insights from fused monitoring-well and GRACE-FO observations coupled with interpretable deep learning in the North China Plain
Identification
- Journal: Ecological Indicators
- Year: 2026
- Date: 2026-11-13
- Authors: Yongkang Wang, Haigang Wang, Beibei Chen, Chaofan zhou, Lin Zhu, Huili Gong
- DOI: 10.1016/j.ecolind.2026.115842
Research Groups
- College of Resource Environment and Tourism, Capital Normal University, Beijing, China
- Beijing Laboratory of Water Resources Security, Capital Normal University, Beijing, China
- Key Laboratory of the Ministry of Education Land Subsidence Mechanism and Prevention, Capital Normal University, Beijing, China
- China Institute of Geo-Environment Monitoring (Guide Center of Prevention Technology for Geo-hazards), Beijing, China
Short Summary
This study assessed groundwater storage anomalies (GWSA) in the North China Plain from 2018 to 2025 by fusing monitoring-well and GRACE-FO data with interpretable deep learning, revealing a widespread transition to groundwater recovery around 2021 driven by varied contributions from precipitation, water diversion, and reclaimed water use across different regions.
Objective
- To quantify spatiotemporal variations in Groundwater Storage Anomalies (GWSA) across the North China Plain (NCP) from 2018 to March 2025 at multiple scales (well, city, subplain, regional) by fusing monitoring-well and GRACE-FO observations.
- To identify regional transitions in the predictive contributions of influencing factors (natural recharge, human regulation, water use structure) to GWSA changes using interpretable deep learning.
Study Configuration
- Spatial Scale: North China Plain (NCP), approximately 140,000 square kilometers (km²). Subdivided into Piedmont Plain (~54,000 km²), East-Central Plain (~66,000 km²), and Coastal Plain (~20,000 km²). Assessments also conducted at individual well, city, and regional scales.
- Temporal Scale: January 2018 to March 2025, with monthly resolution.
Methodology and Data
- Models used:
- Temporal Convolutional Network (TCN) with Integrated Gradients (IG) for attribution analysis.
- Global Land Data Assimilation System (GLDAS) models: Catchment Land Surface Model (CLSM) (selected as best-performing), Variable Infiltration Capacity model (VIC), and Noah Land Surface Model (NOAH) for Soil Moisture Storage Anomalies (SMSA).
- Seasonal-Trend decomposition using Loess (STL) for trend extraction.
- Two-sided Mann-Kendall (MK) tests for statistical significance.
- Piecewise Cubic Hermite Interpolating Polynomial (PCHIP) for gap filling in well data.
- Monte Carlo simulations and robust residual Bootstrap method for uncertainty estimation.
- Data sources:
- Groundwater monitoring wells (1267 wells) observations.
- GRACE-FO satellite gravimetry data (CSR, JPL, GSFC mascon products; CSR selected).
- Aquifer hydrogeological parameters (specific yield and specific storage).
- GPM IMERG Final Precipitation L3 V07.
- Terrestrial evapotranspiration dataset across China (1982–2024).
- South-to-North Water Diversion (SNWD) statistics by city.
- Crop planting data (Provincial Statistical Yearbook, Chinese Academy of Science Discipline Data Center for Ecosystem).
- Water supply and use data (Groundwater Use (GWU), Agricultural Water Use (AWU), Reclaimed Water Use (RWU)) by city.
Main Results
- The North China Plain groundwater system underwent a marked phased transition around 2021, shifting from widespread decline to recovery. Ground-based GWSA change rates transitioned from -13.35 millimeters per year (mm/yr) (2018-2020) to 51.39 mm/yr (2021-March 2025), while GFO-based rates shifted from -41.01 mm/yr to 43.15 mm/yr.
- Groundwater recovery exhibited pronounced spatial heterogeneity, with Fusion-GWSA identifying the strongest recovery in the Piedmont Plain (PP) and generally weaker recovery toward the East-Central Plain (ECP) and Coastal Plain (CP). Over the full study period (2018-March 2025), Ground-based and GFO-based GWSA change rates were 78.92 mm/yr and 16.80 mm/yr in the PP, 25.60 mm/yr and 19.00 mm/yr in the ECP, and 3.99 mm/yr and 5.74 mm/yr in the CP, respectively.
- Multi-source evidence linked the groundwater storage transition to enhanced natural recharge (increased precipitation minus actual evapotranspiration (PRE - ET) during the extremely wet year of 2021) and continued adjustments in water use (increases in SNWD transfers, RWU, and surface-water use, alongside declining GWU and AWU).
- Region-specific TCN–IG analyses revealed distinct changes in predictive contributions:
- Precipitation (PRE) showed robust increases in importance in both high water-receiving region (R1) and low water-receiving region (R2), with ΔIG values of 6.34 mm in R1 and 5.43 mm in R2.
- South-to-North Water Diversion (SNWD) contributions increased more in R1 (ΔIG = 7.83 mm) than in R2 (ΔIG = 3.25 mm).
- Reclaimed Water Use (RWU) contributions showed a larger increase in R1 (ΔIG = 33.55 mm) than in R2 (ΔIG = 6.09 mm).
- Agricultural Water Use (AWU) showed a robust increase in predictive importance only in R2 (ΔIG = 5.10 mm).
- Evapotranspiration (ET) and Groundwater Use (GWU) showed no clear shifts in predictive contribution.
Contributions
- Developed a dynamic data fusion approach that integrates local monitoring-well observations with regional GRACE-FO satellite constraints, improving the spatial continuity and local representation of Groundwater Storage Anomaly (GWSA) estimates, particularly in sparsely monitored areas.
- Quantitatively attributed the changes in influencing factors to GWSA transitions using an interpretable deep learning framework (Temporal Convolutional Network with Integrated Gradients), revealing regional differences in the predictive contributions of natural recharge and human water management strategies.
- Provided a comprehensive multiscale assessment of GWSA across the North China Plain, consistently identifying a widespread phased transition to groundwater recovery around 2021 and characterizing its significant spatial heterogeneity.
- Highlighted the ecohydrological implications and potential management risks of groundwater recovery, advocating for integrated management strategies that consider ecological water requirements, land subsidence, and the safety of urban underground infrastructure.
Funding
- National Natural Science Foundation of China (NO. U24A20433, No. 42371081/D0104, NO. 41930109/D0104)
- Technology Innovation Demonstration Special Project of the Inner Mongolia Autonomous Region (2025ZDSF001702)
Citation
@article{Wang2026Regional,
author = {Wang, Yongkang and Wang, Haigang and Chen, Beibei and zhou, Chaofan and Zhu, Lin and Gong, Huili},
title = {Regional transitions in factors associated with groundwater recovery: Insights from fused monitoring-well and GRACE-FO observations coupled with interpretable deep learning in the North China Plain},
journal = {Ecological Indicators},
year = {2026},
doi = {10.1016/j.ecolind.2026.115842},
url = {https://doi.org/10.1016/j.ecolind.2026.115842}
}
Original Source: https://doi.org/10.1016/j.ecolind.2026.115842