Unraveling Multiscale Spatiotemporal Linkages of Groundwater Storage and Land Deformation in the North China Plain After the South-to-North Water Diversion Project

Highlights

What are the main findings?

• Groundwater storage in the NCP showed a declining trend from 2018 to 2021, with a maximum subsidence rate reached −177 mm/a.
• Downscaled GWSA revealed contrasting groundwater variations across the northwest and southeast sides of the Shunyi fissure.

What is the implication of the main finding?

• A novel GRACE statistical downscaling algorithm integrating a weight allocation strategy and InSAR-derived GWS estimation is proposed.
• The multiscale analysis provides insights into groundwater storage and land deformation patterns in the NCP after SNWDP.

Abstract

Leveraging multi-source remote sensing datasets and dynamic groundwater monitoring well observations, this study explores the multiscale spatiotemporal linkages of groundwater storage changes and land deformation in North China Plain (NCP) after the South-to-North Water Diversion Project (SNWDP). Firstly, we employed Gravity Recovery and Climate Experiment (GRACE) and interferometric synthetic aperture radar (InSAR) technology to estimate groundwater storage (GWS) and land deformation. Secondly and significantly, we proposed a novel GRACE statistical downscaling algorithm that integrates a weight allocation strategy and GWS estimation applied with InSAR technology. Finally, the downscaled results were employed to analyze spatial differences in land deformation across typical ground fissure areas. The results indicate that (1) between 2018 and 2021, groundwater storage in the NCP exhibited a declining trend, with an average reduction of −3.81 ± 0.53 km³/a and a maximum land deformation rate of −177 mm/a; (2) the downscaled groundwater storage anomalies (GWSA) showed high correlation with in situ measurements (R = 0.75, RMSE = 2.91 cm); and (3) in the Shunyi fissure area, groundwater storage on the northern side increased continuously, with a maximum growth rate of 28 mm/a, resulting in surface uplift exceeding 70 mm.

1. Introduction

Groundwater is a dynamic resource whose levels continuously fluctuate under the influence of various natural and anthropogenic factors. Excessive groundwater extraction in many regions disrupts the natural balance between recharge and discharge, leading to significant groundwater level (GWL) declines and ecological degradation [1,2]. The resulting pressure on aquifer structures often causes land subsidence, and groundwater overexploitation has become a major environmental issue [3,4]. It is reported that numerous land subsidence events worldwide have been linked to groundwater depletion [5,6,7,8].

Currently, the Gravity Recovery and Climate Experiment (GRACE) and GRACE-follow on (GRACE-FO) missions indirectly provide large-scale groundwater storage (GWS) estimates, helping to overcome the limitations of sparse groundwater well monitoring [9,10,11]. However, GRACE suffers from the severe underestimation of GWS due to its coarse resolution and signal leakage [9,12]. This limitation hinders effective fine-scale groundwater management.

Interferometric synthetic aperture radar (InSAR) can achieve large-scale precise deformation monitoring. In recent years, InSAR has been used to estimate GWS based on Terzaghi’s (1996) [13] effective stress principle [7,14,15,16]. This method allows for accurate GWS estimation with higher spatial and temporal resolution.

Although previous studies have employed high-resolution land surface models (LSMs) to downscale GRACE data [17,18,19,20,21,22], LSMs like the Catchment Land Surface Model (CLSM) and the Noah LSM primarily focus on shallow soil moisture and do not explicitly simulate variations in GWS [23]. InSAR observations offer a high resolution and strong spatial consistency with groundwater measurement data [24,25]. As a result, many researchers have proposed combining GRACE with InSAR, which partially addresses GRACE’s coarse resolution while also enabling quantitative estimation of GWS [12,15]. In this study, we introduce a novel GRACE statistical downscaling algorithm based on a weight allocation strategy. This methodological innovation provides physically interpretable downscaling results that can be directly linked to aquifer mechanics. Unlike machine learning approaches such as Random Forest (RF), geographically weighted regression (GWR), or evapotranspiration (ET)-based method [26,27,28], our framework explicitly enforces mass conservation, which is essential for water balance studies. Furthermore, it offers clear physical interpretability and lower computational cost, whereas machine learning methods, despite their ability to capture nonlinear relationships, often lack physical meaning, rely heavily on large training datasets, and cannot guarantee water mass balance. This approach leverages InSAR data for high-resolution GWS inversion and applies it to enhance the spatial resolution of GRACE data.

Since the 1960s, a significant groundwater depression funnel has developed in the North China Plain (NCP), resulting in the fastest deformation rate in China and expanding the affected area to over 40,000 km² [29]. Since the implementation of the South-to-North Water Diversion Project (SNWDP), groundwater in the NCP was recharged, and land deformation was effectively alleviated [25]. However, due to the complex geological conditions of the NCP and the prolonged historical shortage of water resources, the implementation of the SNWDP has resulted in the coexistence of groundwater recovery and depletion, as well as land subsidence and rebound. These dynamics have introduced more intricate spatiotemporal and regional patterns of groundwater changes and land deformation in the NCP [25,30]. Beyond its regional role, SNWDP can be viewed as a template for similar projects worldwide, such as the California State Water Project [31], underscoring its broader policy relevance for sustainable groundwater management.

Currently, few studies have comprehensively unraveled spatiotemporal linkages of groundwater storage and land deformation on both large and sub-regional scales after the SNWDP. In this study, firstly, we employed a combined GRACE/InSAR methods to estimate GWS and land deformation. Secondly, we developed a statistical downscaling approach that integrates InSAR-derived GWS with weight allocation strategies to reconstruct high-resolution (0.05°) spatiotemporal GRACE data. Finally, the downscaled results were used to analyze spatial differences in land deformation across typical ground fissure areas, providing a basis for elucidating the coupling mechanisms between groundwater and land deformation.

2. Materials and Methods

2.1. Study Area

The NCP (113.0–119.5°E, 34.5–40.5°N) is bordered by the Yanshan Mountains to the north, the Huaihe River to the south, the Bohai Sea and the Yellow Sea to the east, and the Taihang Mountains to the west (Figure 1a). It gently slopes from west to east, with elevations predominantly under 50 m. Situated in a semi-humid climate zone, it receives approximately 500–700 mm of rainfall annually [6]. The groundwater recharge primarily depends on atmospheric precipitation infiltration, accounting for approximately 60% of the total recharge and peaking in summer [28]. The hydrogeological environment in the NCP is complex. From the foothills to the coastal areas, the aquifer system transitions from a single sand and gravel aquifer to multiple water-bearing sand layers, which are separated by silt or clay layers (Figure 1b). The soil particles become progressively finer, and the thickness of the compressible clay layer gradually increases [32].

2.2. Datasets

2.2.1. Gravity Recovery and Climate Experiment (GRACE)

Three versions of GRACE terrestrial water storage (TWS) data, comprising all available observation months from January 2003 to July 2022, were considered: the Mass Concentration (MASCON) solution from the Center for Space Research (CSR) (https://www2.csr.utexas.edu/grace, accessed on 25 September 2025) [33], and the MASCON solution from the Jet Propulsion Laboratory (JPL) (https://grace.jpl.nasa.gov/data/get-data/jpl_global_mascons, accessed on 25 September 2025) [34], and the MASCON solution from the Goddard Space Flight Center (GSFC) (https://earth.gsfc.nasa.gov/geo/data/grace-mascons, accessed on 25 September 2025) [35]. We used the ensemble mean of these three versions as GRACE-TWS dataset. To address the 11-month data gap between the GRACE and GRACE-FO satellites, we use the GRACE/GRACE-FO supplemental dataset from Yang et al. (2023) [36], which accounts for the effects of natural and anthropogenic factors on the TWS to supplement the missing data.

2.2.2. Hydrometeorological Data

The Global Land Data Assimilation System (GLDAS) V2.1 provides global-scale TWS estimates by assimilating ground observations, satellite remote sensing data, and climate models. The soil moisture storage (SMS) and snow water equivalent (SWE) were derived using the means of three surface models: the variable infiltration capacity (VIC), catchment land surface model (CLSM), and NOAH models [23]. Reservoir storage (RSVS) was derived using the WaterGlobal Assessment and Prognosis (WaterGAP) model [37] and in situ measurement (https://nsbd.swj.beijing.gov.cn/dzxsksq.html, accessed on 25 September 2025). In addition, we did not consider canopy water because its contribution to TWS in the study area is negligible compared to groundwater and soil moisture.

Figure 1. (a) Description of datasets coverage in the study area, in which the gray rectangle indicates the range of the GRACE 0.25° grid, the red frame indicates the coverage of the Sentinel-1A data used, and the blue dots indicate the water level observation locations. (b) The map of A-A’ profile line hydrogeological [38].

Figure 1. (a) Description of datasets coverage in the study area, in which the gray rectangle indicates the range of the GRACE 0.25° grid, the red frame indicates the coverage of the Sentinel-1A data used, and the blue dots indicate the water level observation locations. (b) The map of A-A’ profile line hydrogeological [38].

2.2.3. Synthetic Aperture Radar (SAR) and Auxiliary Data

The SAR data used in the study were acquired from ascending data obtained by the Sentinel-1A Satellite from January 2018 to October 2021. The locations and details of the Sentinel-1A images are provided in Figure 1a and

Table 1. Sentinel-1A data acquisition.

Track	Frame	Swath	Period	Master Image Date	Number of SAR Data
40	112	1, 2, 3	8 January 2018–7 October 2021	27 February 2020	113
40	117	2, 3	8 January 2018–7 October 2021	27 February 2020	113
69	119	1, 2	10 January 2018–21 October 2021	17 February 2020	112
69	124	1, 2, 3	10 January 2018–21 October 2021	17 February 2020	112
142	116	1, 2, 3	3 January 2018–14 October 2021	22 February 2020	113
142	121	1, 2, 3	3 January 2018–14 October 2021	22 February 2020	113
142	126	2, 3	3 January 2018–14 October 2021	22 February 2020	113

. Additionally, we collected 30 m resolution Shuttle Radar Topography Mission (SRTM) data for use as an external digital elevation model (DEM).

2.2.4. In Situ Groundwater Level (GWL) Measurements and Borehole Data

We selected 1005 sets of GWL data for the confined aquifer in the NCP obtained from the China Institute of Geo-Environment Monitoring (https://geocloud.cgs.gov.cn/, accessed on 25 September 2025). Their spatial distribution is illustrated in Figure 1a.

Additionally, we collected data from two boreholes in the Shunyi ground fissure area to generalize the properties of the soil layers. The borehole data were obtained from the China Institute of Geo-Environment Monitoring (https://geocloud.cgs.gov.cn/, accessed on 25 September 2025), and their details are provided in

Table 2. Borehole data information.

Borehole	Clay Layer Thickness (m)	Position
B1	40.20	Northwest of the Shunyi ground fissure
B2	87.18	Southeast of the Shunyi ground fissure

. Datasets used in this study are listed in Table S1 of the Supplementary Materials.

2.3. Methodology

Figure 2 summarizes the main framework and methodology of this study. First, GRACE and GLDAS data were used to estimate groundwater storage anomalies in the NCP. Next, Sentinel-1A images were processed using the Persistent Scatterer InSAR (PSI) method to obtain the spatiotemporal dynamics of land deformation in the region. Then, a statistical downscaling approach combining InSAR-derived groundwater storage and a weight allocation strategy was applied to reconstruct high-resolution groundwater storage anomalies (GWSA), and the results were validated. Finally, the downscaled results were used to analyze spatial differences in land deformation across typical ground fissure areas.

2.3.1. Deriving Groundwater Storage (GWS)

Based on the water balance principle [39], we calculated the change in the GWS:

$$GWS=TWS-(SWE+SMS+RSVS),$$

(1)

where $GWS$ is the groundwater storage; $TWS$ is the terrestrial water storage measured by the GRACE; $SWE$ is the snow water equivalent; $SMS$ is the soil water storage; and $RSVS$ is the reservoir storage. In this paper, all of the variables, including the water storage anomalies estimated by GRACE and in situ GWL measurements, are expressed as the equivalent water thickness (EWT). There were differences in the spatial resolution of GRACE/GLDAS data. We did not use any interpolation while extracting the values at the 0.25° grid, rather the value at the 0.5°/1° grid closest to the 0.25° grid is adopted. For robustness, canal leakage, irrigation return flows, and post-SNWDP pumping restrictions were not incorporated into the GWS estimates, as these processes are difficult to quantify and highly uncertain.

2.3.2. Land Deformation Derived from PSI

PSI processing was conducted using SARPROZ [40]. First, one SAR image was selected as the master, and all slave images were co-registered by calculating the coordinate mapping relationships in both azimuth and range directions, followed by resampling. Differential interferograms were then generated, and persistent scatterer (PS) candidates were identified using both the coherence coefficient and the amplitude stability index (ASI). PSI processing was conducted using SARPROZ, and an ASI threshold of 0.8 was applied, effectively reducing noise and errors while yielding an optimal number of PSs. The unwrapped phases were filtered in both the temporal (high-pass) and spatial (low-pass) domains to estimate orbital errors of the master and slave images as well as atmospheric effects. The topographic fringes were corrected using the 30 m resolution SRTM DEM as the reference terrain [41], and residual orbital errors were further compensated through polynomial fitting. The atmospheric phase screen (APS) module was employed to mitigate the delayed atmospheric phases. Finally, after removing residual topographic and atmospheric phase errors, the deformation rates and cumulative displacements of each PS point were recalculated, providing detailed ground deformation information over the study area. By assuming that the horizontal deformation of the surface is negligible [42], the deformations along the line-of-sight ($LOS$) direction were converted into vertical deformations using triangulation formula:

$$d_v=d_{LOS}/\cos \theta ,$$

(2)

where $d_v$ represents the vertical deformation, $d_{LOS}$ represents the vertical deformation along the satellite line of sight, and $\theta$ is the incident angle.

2.3.3. Groundwater Storage Anomaly (GWSA) Derived from InSAR Technology

The skeleton storage coefficient represents the capacity of an aquifer system to respond to hydraulic changes. It is defined as the volume of water released (or stored) per unit area of the aquifer system in response to a one-unit decrease (or increase) in the GWL [14]. In this study, we assumed that all land deformation was attributed to changes in the thickness of the clay layer resulting from variations in GWL within the confined aquifer. We excluded the effects of hydrologic loading and considered the compression of the aquifer skeleton to be elastic, responding to seasonal water level fluctuations during the study period. Therefore, the elastic skeleton storage coefficient ($S_k$) can be expressed as follows:

$$S_k=\mathsf{\Delta }b/\mathsf{\Delta }h,$$

(3)

where $\mathsf{\Delta }b$ is the monthly land deformation observed by InSAR technology, and $\mathsf{\Delta }h$ is the in situ monthly GWL change. We calculated the monthly elastic skeletal storage coefficients ($S_k$) at the monitoring sites of the 1005 wells. The $S_k$ was interpolated to a 0.05° resolution ($S_k^{\prime }$) using the kriging method and was applied across the entire NCP [43]. We selected kriging interpolation because it accounts for spatial autocorrelation via the variogram, providing optimal linear unbiased estimates and enabling quantification of interpolation uncertainty. Deterministic methods, such as inverse distance weighting (IDW) or spline interpolation, do not consider spatial correlation and may underestimate local heterogeneity. Cross-validation of the kriging results yielded R² = 0.82, confirming the robustness of this approach.

We then calculated the change in the GWS (${\mathsf{\Delta }h}^{\prime }$) at a 0.05° resolution in the NCP:

$${\mathsf{\Delta }h}^{\prime }={\mathsf{\Delta }b}^{\prime }\times S_k^{\prime },$$

(4)

where ${\mathsf{\Delta }b}^{\prime }$ is the mean land deformation value for each 0.05° grid across the NCP. The change in the GWS (${\mathsf{\Delta }h}^{\prime }$) was multiplied by the specific yield ($S_y$) (provided by the China Institute of Geo-Environment Monitoring, https://geocloud.cgs.gov.cn/, accessed on 25 September 2025) to derive the GWS at a 0.05° resolution [44]. The specific yield across the NCP varied widely, ranging from 0 to 0.2. Finally, these GWS values were converted to GWSA.

$$\mathsf{\Delta }GWS={\mathsf{\Delta }h}^{\prime }{·S}_y,$$

(5)

To assess robustness, a sensitivity test was performed by perturbing $S_y$ values by ±20%. The resulting downscaled GWSA estimates showed variations within ±8% of the baseline, with negligible impact on spatial patterns. This confirms that the downscaling framework is relatively insensitive to moderate uncertainties in $S_y$.

2.3.4. Statistical Downscaling Approach with Weight Allocation Strategies

In this study, we built upon Wan’s approach [21] by utilizing the InSAR-derived GWSA at a 0.05° resolution to reconstruct the high-resolution GWSA data (Figure 3).

First, the values from the InSAR-derived GWSA were normalized to obtain standardized values ($S_i$). These 0.05° grids were then aggregated into a 0.25° grid, serving as an area-weighting factor to match the 0.25° GRACE-derived GWSA:

$$S_M={\displaystyle \frac{\sum {(S}_i\times a_i)}{\sum a_i}}={\displaystyle \frac{\sum {(S}_i\times a_i)}{A}},$$

(6)

where $S_M$ is the normalized groundwater equivalent thickness data at a 0.25° resolution (mm); $a_i$ is the area of a 0.05° grid (m²); and $A$ is the total area of all of the 0.05° grids within a 0.25° grid (m²).

The GRACE-derived GWSA was taken as the true value. The difference between the aggregated 0.25° normalized groundwater equivalent thickness data ${(S}_M$) and the 0.25° GRACE normalized equivalent water thickness ($S_G$) was defined as the bias (B) of the modeled equivalent water thickness:

$$B=S_M-S_G,$$

(7)

The groundwater offset ($B\times A$) between the model and GRACE data was further distributed across the 0.05° grids using water as the weighting factor:

$$b_i={\displaystyle \frac{B\times A\times \frac{S_{oi}\times a_i}{\sum (S_{oi}\times a_i)}}{a_i}}={\displaystyle \frac{B\times A\times S_{oi}}{\sum (S_{oi}\times a_i)}},$$

(8)

where $b_i$ is the bias of the 0.01° model equivalent water thickness, and $S_{oi}$ is the pre-standardized 0.05° model equivalent water thickness. By subtracting the bias $b_i$ from $S_i$, the bias-corrected 0.05° monthly equivalent water thickness $S_i^{\prime }$ can be obtained:

$$S_i^{\prime }=S_i-b_i.$$

(9)

This downscaling method preserves the accuracy of the GRACE data while ensuring that the sum of the equivalent water thicknesses of all of the 0.05° grids within a 0.25° grid matches the original 0.25° GRACE value.

2.3.5. Uncertainty Analysis

When evaluating the contributions of our study, it is important to consider its limitations, which can be addressed in future research to further enhance the ability to identify groundwater depletion hotspots. First, there is inconsistency in the spatial resolution of the hydrological datasets used for GWS estimation. We addressed this by directly extracting coarser-resolution data onto a higher-resolution grid for calculations. The uncertainties of hydrological models can be dealt with by using the ensemble mean of several models [45]. Second, to reduce the limitations introduced by spatial interpolation during the downscaling process, we employed a denser network of observation wells in the InSAR-based GWS estimation approach.

To provide a quantitative error budget, we explicitly propagated uncertainties from multiple sources into the final downscaled GWSA estimates. Specifically: (1) the spread among the three GRACE mascon solutions (CSR, JPL, GSFC) was computed as the grid-wise standard deviation (${\mathsf{\sigma }}_{GRACE}$); (2) ensemble variance across GLDAS models (Noah, VIC, CLSM) was included as ${\mathsf{\sigma }}_{GLDAS}$; (3) InSAR phase noise was quantified from redundant interferograms and converted into land deformation uncertainty, which was then propagated through the $S_k$ inversion to obtain ${\mathsf{\sigma }}_{InSAR}$; (4) the kriging interpolation error of $S_k$ was assessed via leave-one-out cross-validation (R² = 0.82), yielding ${\mathsf{\sigma }}_{S_k}$; and (5) the sensitivity of the downscaled GWSA to ±20% variations in $S_y$ was treated as ${\mathsf{\sigma }}_{S_y}$.The total uncertainty was calculated as:

$${\mathsf{\sigma }}_{GWSA}=\sqrt{{\mathsf{\sigma }}_{GRACE}^2+{\mathsf{\sigma }}_{GLDAS}^2+{\mathsf{\sigma }}_{InSAR}^2+{\mathsf{\sigma }}_{S_k}^2+{\mathsf{\sigma }}_{S_y}^2},$$

(10)

The resulting ${\mathsf{\sigma }}_{GWSA}$ values were mapped at 0.05° resolution (Figure 4), providing spatially explicit confidence intervals for the downscaled groundwater storage anomalies. Higher uncertainties are concentrated in the piedmont zone near Shijiazhuang and the central NCP.

3. Results

3.1. Validation

We compared the average displacement rate of PS points within a 100 m buffer zone around the leveling points with the land deformation measurements derived from InSAR and leveling observations (Figure 5a). The InSAR-derived results showed excellent agreement with leveling data, with a correlation coefficient of 0.95, an RMSE of 3.36 mm/a, and an MAE of 2.87 mm/a.

We obtained three types of GWSA data: GRACE-derived GWSA, downscaled GWSA, and in situ GWSA (Figure 5b and

Table 3. Groundwater depletion rate and groundwater depletion in the NCP (2018–2021) and their R and RMSE values compared to the in situ measurements.

Data	Change Rate (mm/a)	Change Rate (km3/a)	Volume Change (km3)	R	RMSE (cm)
GRACE-GLDAS	−22.50 ± 3.12	−3.81 ± 0.53	−13.64 ± 2.02	0.19	10.85
InSAR	−7.91 ± 2.32	−1.34 ± 0.39	−5.11 ± 1.49	0.97	7.31
Downscaling	−9.10 ± 4.68	−1.54 ± 0.79	−5.87 ± 3.01	0.75	2.91

). The correlations between these datasets and the in situ measurements were 0.97, 0.75, and 0.19, respectively, with corresponding RMSE values of 7.31 cm, 2.91 cm, and 10.85 cm. These results indicate that inverting GWL changes using land deformation data is both effective and feasible, and the downscaling method significantly improves the accuracy of the original GRACE data.

3.2. Analysis of GWSA and Land Subsidence Variations in the NCP

3.2.1. Characteristics of GWSA in the NCP

Figure 6 illustrates the monthly precipitation, GRACE-derived Terrestrial Water Storage Change (TWSC), and GLDAS-derived Terrestrial Water Storage Anomaly (TWSA) in the NCP from January 2003 to August 2021. From 2003 to 2004, the TWSA increased sharply at a rate of 38.25 mm/a. However, due to heavy groundwater extraction, there was a sudden decline in the TWSA, with a depletion rate of −18.99 mm/a from 2005 to 2009. Between 2010 and 2013, the TWSA stabilized due to increased precipitation. Despite the successful implementation of the SWDP in 2014, the TWSA continued to decrease at a rate of −35.89 mm/a from 2014 to 2015, primarily due to severe drought and low rainfall. From 2016 to 2020, the TWSA remained stable due to increased precipitation and recharge from the SWDP. Subsequently, the TWSA increased sharply again until the middle of 2021. Overall, the terrestrial water storage in the NCP exhibited a decreasing trend, with a rate of decrease of −11.57 ± 2.33 mm/a (or −3.85 ± 0.75 km³/a) as retrieved by GRACE.

Additionally, we analyzed the changes in GWS in the NCP from 2018 to 2021 and generated a graph of the rate of change in the GWS (Figure 7a). The results indicate that the GWS exhibited a general decreasing trend across the NCP during the study period. The severe groundwater depletion was concentrated in the southwestern part of the NCP, and Anyang City in Henan Province experienced a rate of decrease of −47.86 ± 5.89 mm/a. In contrast, the northern part of the NCP exhibited less significant groundwater loss, and some areas even exhibited increasing trends. The highest increases occurred in Beijing and Tianjin, the major cities of the SWDP, with rates of increase of up to 8.71 ± 3.99 mm/a.

3.2.2. Characteristics of Land Deformation in the NCP

We obtained vertical displacement data for the NCP from 2018 to 2021 and generated corresponding deformation rate maps (Figure 7b). The research results show that land subsidence continues to occur in the central and southern parts of the NCP, with significant expansion of subsidence centers in Xingtai and Hengshui, where the maximum subsidence rate reaches up to 177 mm/a. In contrast, the subsidence trend has slowed in some areas of Beijing. Notably, the Chaobai River Basin has exhibited ground uplift, with a cumulative rise of 78.37 mm, which is closely related to groundwater recovery in the region.

3.3. Downscaling of GRACE-Derived GWSA

According to the method described in Section 3.2.1, we first inverted the GWSA at a 0.05° resolution using InSAR deformation time series, as shown in Figure 7c. The inversion results exhibit a high consistency with in situ groundwater level measurements (R = 0.97, RMSE = 7.31 cm), confirming the reliability of this method. In the Beijing–Tianjin region, the InSAR-inverted GWSA remained generally stable and showed an increasing trend, with a maximum growth rate of 9.84 mm/a; in contrast, the central NCP, particularly the Shijiazhuang–Hengshui–Dezhou area, displayed a pronounced decreasing trend, with the maximum annual decline in Jingxian County, Hengshui, reaching −22 mm/a.

Subsequently, the high-resolution GWSA derived from InSAR was used to downscale the GRACE data (Figure 7d). From January 2018 to July 2021, the original GRACE results at 0.25° resolution indicated an overall groundwater deficit across the NCP, with an average annual change rate of −12.36 ± 1.65 mm/a. After downscaling, the groundwater storage variations revealed much clearer spatial heterogeneity. For instance, some areas of the Beijing Plain exhibited a noticeable recovery in groundwater storage (shown in blue), which was not clearly captured in the original GRACE data.

Based on the downscaled GWSA data, we observed pronounced differences in groundwater variations on either side of the Shunyi ground fissure in Beijing (Figure 8a). The results reveal a clear north–south contrast: groundwater storage on the northern side continued to increase, reaching a maximum growth rate of 28 mm/a, whereas the southern side remained in deficit, with a maximum decline rate of −29.5 mm/a. Furthermore, groundwater storage in the upper Chaobai River region also exhibited recovery, which closely aligns with the surface rebound trends detected by InSAR.

4. Discussion

4.1. Advantages and Limitations of the GRACE Downscaling

To evaluate the effectiveness of the GRACE downscaling method proposed in this study, we conducted a comparative experiment using the Random Forest (RF) method to downscale GWSA [46] (see Supplementary Materials Sections S1–S3 for details). The results indicate that the RF-generated groundwater storage distribution is relatively smooth (Figure 9), but its accuracy is lower compared with in situ groundwater measurements, and it struggles to resolve fine-scale spatial variations, limiting its applicability for local-scale groundwater assessments. In contrast, the method proposed in this study not only demonstrates a significant advantage in accuracy (correlation coefficient R = 0.75, RMSE = 2.91 cm) but, more importantly, effectively preserves fine-scale spatial heterogeneity that is critical for local-scale investigations. For instance, although GRACE observations indicate overall groundwater depletion in the NCP, our approach reveals localized groundwater recovery in the Beijing Plain, highlighting its enhanced spatial resolution.

The 0.05° downscaled GWSA maps provide actionable insights for groundwater management, including (1) identifying zones with relatively stable storage for safer well drilling, (2) highlighting aquifer sections with greater storage potential for artificial recharge, and (3) integrating GWSA with InSAR-derived deformation to prioritize subsidence-prone regions for mitigation. These applications demonstrate the method’s practical value in data-rich regions.

Nevertheless, several limitations should be noted. The validation dataset covers only 2018–2021, providing a relatively short observation window. Although the high correlation with well data highlights the method’s value, this limited period may not capture longer-term variability. In particular, extreme hydroclimatic events, such as the severe 2014–2015 drought, could drive groundwater storage changes independent of SNWDP inflows [47], complicating attribution. Moreover, the recovery from long-term groundwater overexploitation is inherently gradual [48]. Under extreme drought conditions, water demand increases substantially, and surface water inputs from diversion projects and precipitation alone may not be sufficient to offset deficits. Future research should therefore incorporate longer well records or reconstructed time series to disentangle the combined effects of climate extremes and inter-basin water transfers.

Furthermore, while the method performs well in unconsolidated sedimentary basins, its applicability to karst or fractured aquifers is constrained. InSAR coherence is often poor in these environments due to vegetation and rough terrain, and aquifer responses are dominated by preferential flow and localized storage processes that may not be adequately captured by the current bias allocation scheme. Future extensions could integrate complementary geophysical and hydrological data (e.g., spring discharge, dense well networks) to enhance applicability in such complex hydrogeological settings.

4.2. Analysis of Surface Deformation Differences on Both Sides of Ground Fissures Based on GRACE Downscaling Results

In the Shunyi ground fissure area, the downscaled GRACE-derived groundwater storage anomaly (GWSA) results reveal pronounced spatial heterogeneity in groundwater changes, providing crucial evidence for explaining the differential surface deformation observed on both sides of the fissure (Figure 8).

InSAR monitoring results along Profile I and Profile II indicate significant deformation on the northwest side of the fissure (Figure 8b and Figure 10): Profile I shows cumulative subsidence reaching −99.32 mm, while Profile II displays surface uplift with a maximum value of 78.37 mm. In contrast, the southeast side exhibits smaller and more stable surface deformation, with an average of only −15.39 mm along Profile I and just 2.38 mm along Profile II.

The observed differences in surface deformation are closely linked to groundwater dynamics. The downscaled GWSA results suggest a continuous increase in groundwater storage on the northwest side, primarily due to recharge from the upstream Chaobai River, resulting in surface rebound in the northern area. Meanwhile, the southeast side, located downstream with limited recharge, shows little variation in groundwater storage and correspondingly stable surface conditions.

Moreover, previous studies have indicated that since the implementation of the Middle Route of the South-to-North Water Diversion Project, significant groundwater recharge has occurred in the northern part of Shunyi, alleviating regional land subsidence trends to some extent [25]. The downscaled GWSA results in this study further clarify the spatial distribution of this localized recharge effect.

It is noteworthy that the response time of land deformation to groundwater variations exhibits clear differences across the fissure (Figure 8a). To quantify these lags, we first extracted the seasonal components of both groundwater levels and surface deformation using STL decomposition. The time-lagged cross-correlation (TLCC) method [49] was then applied, with the lag corresponding to the maximum correlation between the seasonal signals taken as the response time. Based on this procedure, the northwest side shows a lag of approximately 2–30 days, whereas the southeast side exhibits a lag of 30–60 days, indicating that, under fault control, the hanging wall (northwest side) responds more slowly to groundwater fluctuations than the footwall (southeast side). Further evidence from the five groundwater monitoring wells (W1–W5) along Profile II shows that the southeast wells (W4 and W5) display more pronounced seasonal fluctuations, while the northwest wells (W1–W3) exhibit smaller variations (Figure 8c). These differences are largely controlled by subsurface lithology: borehole data (Figure 8c) indicate that the clay layer on the southeast side of the fissure is 87.18 m thick, 2.17 times that of the northwest side (40.20 m), and the thicker clay layer induces a greater lag effect, reducing the propagation rate of groundwater changes to the land surface.

Comparable fault-controlled hydrogeological responses have been reported in other regions [50,51,52], including Mexico, the California Central Valley, and Utah, USA, where aquifer deformation is strongly modulated by variations in lithostratigraphy across fault boundaries, highlighting that the mechanism observed at the Shunyi fissure is consistent with global observations of fault-controlled aquifer behavior.

5. Conclusions

This study systematically analyzed the multiscale spatiotemporal responses of groundwater storage (GWS) and land deformation in the NCP following the implementation of the South-to-North Water Diversion Project (SNWDP), based on GRACE satellite data, Persistent Scatterer InSAR (PSI) technology, and dynamic groundwater monitoring well observations. A GRACE statistical downscaling algorithm, integrating InSAR-derived GWS and a weight allocation strategy, was developed to reconstruct high-resolution (0.05°) groundwater storage anomalies (GWSA), which were then used to investigate spatial differences in surface deformation in typical ground fissure areas. The main conclusions are as follows:

(1) Between 2018 and 2021, groundwater storage in the NCP exhibited a declining trend, with an average reduction of −3.81 ± 0.53 km³/a, and the maximum vertical deformation rate reached −177 mm/a. Although the SNWDP supplied water to some areas, causing groundwater recovery in the northern region, the overall groundwater deficit persisted, and land subsidence continued.

(2) The proposed GRACE downscaling algorithm enhanced the spatial resolution from 0.25° to 0.05° and significantly improved the accuracy of GWSA inversion. Validation indicates a high correlation with in situ measurements (R = 0.79, RMSE = 2.91 cm). The method is applicable to regions with confined or semi-confined aquifers where deformation can be reliably monitored using InSAR, providing a feasible solution for GWSA estimation in areas lacking dense in situ monitoring data.

(3) Downscaled GRACE data revealed pronounced spatial heterogeneity in groundwater variations across the Shunyi ground fissure area, explaining the asymmetric surface deformation: the northwest side experienced groundwater recovery accompanied by surface rebound, whereas the southeast side remained in deficit with relatively stable surface conditions.