Assessing Subsidence and Coastal Inundation in the Yellow River Delta Using TS-InSAR and Active Inundation Algorithm

Abstract

The extensive distribution of quaternary sediments and the extraction of underground resources in the Yellow River Delta (YRD) have resulted in significant land subsidence, which accelerates relative sea level (RSL) rise and heightens the risk of coastal inundation. This study uses Sentinel-1A (S1A) imagery and the time-series synthetic aperture radar interferometry (TS-InSAR) method to obtain subsidence information for the YRD. By integrating data from groundwater level monitoring wells, hydrogeological conditions, extensometer monitoring, and drilling wells, we analyze the causes of subsidence and the deformation response to the groundwater level changes in the corresponding aquifers. For the first time in the YRD, this study introduces the high accuracy CoastalDEM v2.1 digital elevation model, combined with absolute sea level (ASL) data, to construct a coastal inundation simulation. This simulation maps the land inundation caused by RSL rise along the YRD in different scenarios. The results indicate significant subsidence bowls in coastal and inland regions, primarily attributed to shallow brine and deep groundwater extraction, respectively. The main subsidence layers in inland towns have been identified, and residual deformation has been observed. Currently, land subsidence has caused a maximum elevation loss of 141 mm/yr in coastal YRD areas, significantly contributing to RSL rise. Seawater inundation simulations suggest that if subsidence continues unabated, 12.84% of the YRD region will be inundated by 2100, with 8.74% of the built-up areas expected to be inundated. Compared to global warming-induced ASL rise, ongoing subsidence is the primary driver of inundation in the YRD coastal areas.

1. Introduction

Coastal areas are major locations for many of the world’s largest cities, with significant socioeconomic value. Notably, about 12% of the global population resides in areas below 10 m in elevation [1]. With global warming, the global average sea level is rising at a rate of 3.5 mm/yr, increasing the risk of coastal flooding and inundation in low-lying areas while also reducing resilience to extreme weather events [2]. Additionally, delta regions are characterized by large amounts of highly compressible quaternary sediments, and large-scale land subsidence has occurred due to frequent human activities, such as underground resource extraction and urban development [3]. Land subsidence threatens approximately 120 million people living in coastal flood-prone areas, and further subsidence could exacerbate flood risks [4].

Multiple studies worldwide have shown that the combined effects of sea level rise and land subsidence pose significant threats to the geological environment of coastal regions, including Jakarta in Indonesia, the Gulf of Mexico, and the southeastern coast of China. Severe land subsidence has been observed in these regions, exacerbating RSL rise and causing coastal flooding land inundation [5,6]. These studies primarily focus on the spatial distribution, causes, and contributions of land subsidence to RSL rise. Among these regions, Samarang City in Jakarta is one of the most well-known areas experiencing the combined effects of ASL rise and land subsidence. The average elevation of this area is less than 3 m, and excessive groundwater extraction has led to severe land subsidence. Flood inundation models indicate that, under the influence of land subsidence, the risk of coastal inundation in these areas will significantly increase [7,8]. In the Gulf of Mexico, human activities including groundwater and hydrocarbon extraction have accelerated land subsidence. Subsidence has contributed approximately 85% of the RSL rise in the region, which is expected to lead to extensive land loss in the future [9,10]. In coastal residential areas of China, the increase in RSL rise will lead to a significant expansion of areas affected by 100-year extreme coastal flooding events. Under climate change scenarios, the combined effects of SLR and land subsidence in regions such as Hainan and Shanghai have considerably expanded the extent of compound flood inundation [11].

The Yellow River Delta is located on the northern coast of the Shandong Peninsula in China. It was formed by sediment deposition from the river’s continual course changes since 1855, with an average elevation of only 4 m. Against the backdrop of global sea level rise driven by climate change, the YRD is highly vulnerable to coastal flooding and other disasters. Coupled with long-term underground resource extraction, YRD has experienced severe land subsidence, resulting in a rapid rise in RSL. The development of TS-InSAR technology has supplemented traditional deformation monitoring methods, enabling efficient and accurate acquisition of deformation information. It has been widely applied in land subsidence monitoring [12,13].

Previous studies on the YRD have mainly focused on land subsidence monitoring, its causes, and impact assessments [14,15,16,17]. These studies have shown significant land subsidence in both the coastal and inland areas of the YRD, greatly affecting the regional geological environment. Despite these insights, similar to studies on other coastal cities, no further analysis has been conducted on land subsidence in the YRD, especially concerning the response of deformation to groundwater level variations. Additionally, the commonly used Shuttle Radar Topography Mission (SRTM) dataset in seawater inundation simulations has a significant positive elevation bias, leading to inaccurate simulation results.

To address these research gaps, this study integrates in-situ observation datasets, including TS-InSAR-derived deformation, optical remote sensing imagery, groundwater levels, geological drilling, and extensometer measurements. These multi-source datasets enable a detailed analysis of the spatial-temporal characteristics and primary factors of land subsidence across both inland and coastal areas of the YRD. This study focuses on revealing the deformation response to groundwater level variations across different aquifers, identifying the main subsidence layers and the presence of residual deformation in multi-aquifer systems within areas of long-term groundwater extraction. Additionally, this study, for the first time, introduces high-accuracy CoastalDEM v2.1 data to the YRD in combination with projections from Sixth Assessment Report of the Intergovernmental Panel on Climate Change (IPCC AR6) to simulate coastal inundation using an active inundation algorithm under different RSL rise scenarios. The deeper understanding of these issues will facilitate the scientific management of land subsidence in multi-aquifer systems. It will also support the development of adaptive strategies for sustainable groundwater extraction and coastal zone management in deltaic environments, helping to mitigate future compound risks associated with RSL rise.

2. Study Area, Datasets, and Methods

2.1. Study Area

The study area includes Dongying (DY) City and Binzhou (BZ) City, located in the northern part of the Shandong Peninsula, and covers the YRD and its surrounding areas. The modern YRD was formed when the mouth of the Yellow River shifted from the Yellow Sea to the Bohai Sea in 1855. Sediment carried by the river during its multiple course changes has continued to accumulate over time. The YRD covers an area of approximately 5400 km², situated between Bohai Bay and Laizhou Bay, making it the youngest river delta in China. Its topography, shaped by ancient river channels, features flat terrain, with primary landforms of plains and low hills, and an average elevation of just 4 m (Figure 1a).

The quaternary sediments in the YRD consist primarily of clay, silty clay, silt, and sand, forming a multi-layer aquifer system [18]. The YRD contains a large number of natural resources such as petroleum, natural gas, and underground brine distributed along the coast. It is an important area for petroleum and brine extraction in China. The general distribution of major petroleum extraction areas, brine extraction zones, and aquaculture areas in the YRD is illustrated in Figure 1b.

2.2. Datasets

In this study, we integrated multi-source datasets, including deformation data, optical satellite imagery, drilling wells, groundwater level observation wells, and extensometer monitoring. Drilling wells were used to characterize the sediment types of the YRD. S1A imagery was employed for TS-InSAR analysis, while Generic Atmospheric Correction Online Service (GACOS) and SRTM data were used to correct for tropospheric and topographic phase delays during processing. Benchmarks and extensometers were utilized to validate the accuracy of the TS-InSAR results. Groundwater level monitoring wells were analyzed to examine the relationship between deformation and groundwater level in different aquifers. The accuracy of CoastalDEM v2.1 was assessed using GNSS, and AVISO data together with sea level rise projections from IPCC AR6 were used to construct the seawater inundation simulation. Optical imagery and Esri Land Cover datasets are employed to analyze land use patterns in the study area [19]. Figure 2 presents the schematic overview of the main processing and analysis workflow in this study.

2.2.1. SAR Images

This study employs synthetic aperture radar images from S1A for TS-InSAR analysis to obtain deformation data for the YRD and its surrounding areas. The S1A satellite, launched by the European Space Agency in 2014, carries a C-band synthetic aperture radar. A total of 392 single look complex images from Path 69, Frame 114, and Frame 109 were selected from the period between 1 July 2015 and 24 August 2023. The coverage area is shown in Figure 1a. To minimize the impact of orbital errors in the TS-InSAR analysis, precise orbit ephemerides for S1A were incorporated during data processing. The specific parameters of the images are provided in

Table 1. Sentinel-1A SAR Image Parameters.

Path	Frame	Flight Direction	Polarization	Range Resolution	Azimuth Resolution	Nr.
69	109	Ascending	VV	2.7~3.5 m	22 m	196
69	114	Ascending	VV	2.7~3.5 m	22 m	196

.

2.2.2. Sea Level Rise Under Different Scenarios

Since the 1940s, climate warming has led to the mass loss of glaciers and the Greenland ice sheet, which is the primary cause of the rapid rise in ASL [20]. A series of IPCC reports indicate that the future variations in sea level will differ based on the intensity of carbon emissions, which will be influenced by international climate policies. In this study, we used the Global Mean Sea Level (GMSL) rise rate from AVISO (Archiving, Validation, and Interpretation of Satellite Oceanographic Data) between 1993 and 2021 at a resolution of 0.25° as the current sea level change trend [21]. The AVISO GMSL integrates measurements from multiple altimetry missions, including TOPEX/Poseidon, Jason-1, Jason-2, and Jason-3. With 29 years of altimetry data, the trend’s stability has reached ±0.3 mm/yr with a 90% confidence level, and the rate of sea level rise in the nearshore area of YRD is approximately 4.0 mm/yr. For projections under different climate scenarios, the study refers to the IPCC AR6 report, using five Shared Socioeconomic Pathways (SSPs) to estimate sea level rise. The median projection data from the NASA Sea Level Projection Tool covering the period 2020 to 2150 were used for these projections, with a baseline period from 1995 to 2014 [22]. The projected sea level rise in YRD under different SSP scenarios is shown in Figure 3b.

2.2.3. Digital Elevation Model

To remove the topographic phase component in the interferograms during InSAR processing, the 30 m resolution SRTM DEM was used to simulate the terrain phase. However, due to the vertical accuracy limitations of the SRTM DEM, it is not ideal for providing accurate elevation information in coastal lowlands [23]. For better analysis of the impacts of RSL rise on the Yellow River Delta, we utilized the 90 m resolution CoastalDEM v2.1 dataset from Climate Central. CoastalDEM v2.1 uses a multi-layer perceptron neural network with various reference variables to vertically calibrate the global SRTM DEM, significantly reducing elevation biases in coastal areas. Using ICESat-2 data as a reference, the mean bias of CoastalDEM v2.1 compared to SRTM DEM in areas with populations under 5 m of elevation was reduced from 1.59 m in NASADEM to −0.03 m, while the global elevation RMSE decreased by about 50% [24]. CoastalDEM v2.1 greatly improves the vertical elevation accuracy in coastal lowlands and can effectively serve as a data source for active inundation simulations. The vertical accuracy of CoastalDEM v2.1 in the YRD will be discussed in Section 3.2.

2.2.4. In-Situ Observations

Groundwater level monitoring data from eight unconfined and four confined wells in the study area (2018–2023) were obtained and compared with the temporal deformation characteristics derived from TS-InSAR. In addition, long-term records of groundwater table depth in the study area (2000–2020) were utilized to analyze the spatial correlation between land subsidence and groundwater extraction.

Benchmark and extensometer measurements can provide deformation monitoring results with millimeter-level precision. To validate the accuracy of the TS-InSAR results and analyze the soil layers deformation characteristics of the study area, 75 benchmarks and two extensometers with deformation data from 2016 to 2023 were acquired. The distribution of the benchmarks and extensometers is shown in Figure 1a. To evaluate the accuracy of CoastalDEM v2.1 elevation data in the YRD, elevation values from 14 GNSS stations along the YRD coastline were compared with those from CoastalDEM v2.1 and SRTM DEM.

2.3. Methods

2.3.1. Time-Series InSAR (TS-InSAR) Analysis

The YRD region has relatively sparse built-up areas and dense vegetation, leading to low coherence in interferograms. Therefore, the small baseline subset (SBAS) method was selected to obtain deformation information. SBAS is one of the TS-InSAR techniques; it utilizes short baselines for interferograms and can consider partially distributed scatterers to improve the spatial coverage of deformation results [25]. This effectively mitigates the impact of spatiotemporal decorrelation noise on the results in the interferograms. Moreover, based on the increased spatiotemporal sampling density of observational data, SBAS can better retrieve the average deformation rate and time-series results of nonlinear deformation characteristics [26,27,28]. In this study, TS-InSAR analysis was performed using GAMMA (v202407).The S1A images information is provided in Table 1. The enhanced spectral diversity method was applied to co-register the S1A images, and azimuth phase deramping was performed to remove the linear frequency modulation [29]. A total of 582 interferograms were generated using a combination of three adjacent scenes, with maximum temporal and spatial baselines of 192 days and 308 m, respectively (Figure 4). To improve interferometric phase coherence and processing efficiency, a multi-looking operation of 20 × 5 in range and azimuth was applied. The 30 m resolution SRTM was used to remove the topographic phase, and adaptive filtering was applied to reduce random noise. Minimum cost flow was selected for phase unwrapping. GACOS was used to correct the tropospheric delay phase. GACOS separates turbulent and elevation-related atmospheric signals from the zenith total delay (ZTD) and generates ZTD maps for correcting differential interferograms [30,31]. In the coastal region of the YRD, the standard deviation of interferometric phase in 84.6% of interferograms was reduced by an average of 36.4% [32].

The time-averaged coherence and standard deviation of the unwrapped interferogram were computed and normalized. High-coherence pixels with an average coherence greater than 0.6 and a standard deviation of less than 0.3 were selected for SBAS-InSAR processing. The unwrapped phase for each high-coherence pixel can be expressed as:

$$\phi ={\phi }_{disp}+{\phi }_{topo}+{\phi }_{orb}+{\phi }_{atm}+{\phi }_{noi}$$

(1)

where ${\phi }_{disp}$ is the phase caused by deformation, defined as ${\phi }_{disp}={{\displaystyle \frac{4\pi }{\lambda }}d}_{los}$, where $d_{los}$ is the line-of-sight (LOS) deformation and $\lambda$ is the radar wavelength; ${\phi }_{topo}$ is the topographic residual phase, which is related to DEM accuracy and the vertical baseline, and ${\phi }_{orb}$, ${\phi }_{atm}$, and ${\phi }_{noi}$ represent the orbital error, atmospheric delay phase, and other noise phases, respectively [33].

Subsequently, we inspected the unwrapped phase of each interferogram individually and removed those with obvious errors to prevent them from affecting the deformation results. Singular value decomposition was applied to the differential interferometric phases of the small baseline subset for each high-coherence pixel, transforming the multi-master image interferometric phases into a time series of interferometric phases. Based on the spatiotemporal characteristics of different phase components, spatiotemporal filtering was employed to separate the atmospheric delay phase and nonlinear deformation phase from the time series of interferometric phases. A phase regression model was then used to estimate the linear deformation phase, and the nonlinear deformation phase was combined. Ultimately, the deformation time series was obtained [34].

The InSAR observations were on the LOS [35]. Given the absence of significant active faults and tectonic movements in the YRD, regional deformation is predominantly vertical [36]. Therefore, horizontal deformation components were neglected, and the incidence angle ($\theta$) of each pixel was calculated to convert LOS deformation ($d_{los}$) into the final vertical deformation ($d_v$) result using the following equation:

$$d_v\approx d_{los}/cos\theta$$

(2)

2.3.2. Relative Sea Level Rise-Driven Inundation Mapping Under Different Scenarios

The current subsidence rate in the nearshore YRD significantly exceeds the rate of ASL rise. The combined effect of these processes leads to a rapid increase in RSL, resulting in land inundation. Since seawater gradually advances from the coastline during inundation, the process of land inundation due to sea level rise is considered an active process. To simulate this effect, an active inundation algorithm was developed [37,38].

First, the ASL rise rates from AVISO and the IPCC AR6 were converted to cumulative sea level changes. These datasets and the deformation from TS-InSAR were interpolated to match the pixel resolution of the CoastalDEM v2.1 grid and ensure consistency across all datasets.

Next, the mean intensity of SAR images from 2023 was used to extract the coastline, which served as the seed point for active inundation analysis. If the predicted elevation at a seed point was below zero, it was classified as inundated. The algorithm then examined the eight neighboring pixels, marking them as inundated if their predicted elevations were also below zero. This process continued until no additional qualifying pixels were found. To avoid connectivity errors, water areas were masked. The predicted elevation values for different years were computed using the following function:

$$H_{i,j,t}=H_{i,j,0}-H_{i,j,msl}+H_{i,j,def}$$

(3)

where $H_{i,j,t}$ is the elevation at pixel ($i,j$) at time $t$; $H_{i,j,0}$ is the initial elevation; $H_{i,j,msl}$ is the sea level rise at that pixel, and $H_{i,j,def}$ is the deformation due to land subsidence or uplift. Sign conventions are defined as follows: sea level change is positive for a rise and negative for a fall, whereas deformation is negative for subsidence and positive for uplift.

Finally, the predicted $H_{i,j,t}$ values that are less than 0 and connected to the seed points are marked as inundated areas. The predicted inundation areas for the years 2030, 2050, and 2100 were then generated.

3. Results

3.1. Land Deformation Results from TS-InSAR

3.1.1. Validation of TS-InSAR Results with Benchmark and Extensometer

A total of 75 benchmarks and two extensometers were used to validate TS-InSAR results. A 200 m buffer zone around each point was used to calculate the mean deformation during the same period, accounting for the spatial differences between the observed points and the TS-InSAR results. As shown in Figure 5, the TS-InSAR deformation rate had an RMSE of 0.7 mm/yr and an R² of 0.97 compared to the benchmarks data. The deformation trends at the extensometers were similar. For the BX extensometer, the average error was 3.8 mm/yr. For the GR extensometer, the average error was 5.0 mm/yr. These results show strong agreement between SBAS and the benchmarks and extensometers observation.

3.1.2. Spatial and Temporal Characteristics of Land Subsidence in YRD

The vertical deformation rate of the YRD (July 2015 to August 2023) is shown in Figure 6a. Significant subsidence bowls were detected in the eastern part of HK, the HK–KL junction, the coastal KL area, the BX–GR urban area, and the adjacent Weifang city, with the maximum deformation rate reaching −232.8 mm/yr, located in the eastern part of HK. The optical imagery indicates large-scale brine extraction and aquaculture areas along the YRD coast (Figure 1b). As shown in Figure 6f, the subsidence rate along the profile III-III’ reveals that the most subsided areas are mainly located near the salt pans and aquaculture zones. In contrast, the deformation rate is lower at the Gudong Oilfield in the southeast, with an average rate of about −10 mm/yr. Subsidence at the Gudong Oilfield is mainly influenced by the expansion of the subsidence bowl from the salt pans and aquaculture areas in the northwest.

In order to better analyze the deformation time characteristics of different human activity areas, six feature points from P1 to P6 were selected (Figure 6h). Points P1, P2, P4, and P5 are near salt pans in different subsidence bowls. Point P1 in the largest subsidence bowl in HK. It had an average deformation rate of −194.5 mm/yr, with rapid subsidence since late 2017. From August of 2019 to March of 2020, subsidence paused and shifted to a slow rebound of about 50 mm, before resuming the subsidence trend. Similar behavior was observed at the other points near salt pans. Point P2 at the HK–KL subsidence bowl showed −128.2 mm/yr with rapid subsidence, a slight rebound (September 2019 to July 2020), and gradual stabilization. Point P4, which is near salt pans and aquaculture ponds in KL, had −104.5 mm/yr with continuous slow subsidence. Point P5, near brine extraction sites in eastern GR, had −115.5 mm/yr, similar to Point P2. Point P3 in the Gudong Oilfield remained stable with slight subsidence. Point P6 in the GR urban area showed rapid subsidence from January 2016 to December 2017, followed by stabilization after September 2021.

InSAR deformation results show that severe land subsidence has occurred in the YRD and surrounding areas, with subsidence bowls concentrated along the YRD coast and BX–GR urban area. Time-series deformation in typical areas reveals that the Gudong Oilfield area is currently relatively stable, while there is still a significant subsidence trend in brine extraction and urban areas. In some brine extraction regions, a period of rebound has been observed, and the specific causes will be discussed in Section 4.1.

3.2. Coastal Inundation Simulation Under Different Scenarios

To assess CoastalDEM v2.1’s usability in the YRD region, elevation data from 14 GNSS stations along the coastline were compared with CoastalDEM v2.1 and SRTM. The results indicate that CoastalDEM v2.1 aligns more closely with actual elevations at 12 of the 14 stations (Figure 7a). Compared to SRTM, CoastalDEM v2.1 improved the R² from 0.35 to 0.84 and reduced the RMSE from 0.65 m to 0.21 m, demonstrating superior vertical precision and enhancing seawater inundation simulation accuracy (Figure 7b).

The ASL rise rate from AVISO represents the current sea level change trend along the YRD coastline, while the ASL rise rate from IPCC AR6 represents future global warming scenarios for sea level change. Additionally, we simulated inundation scenarios under the current ASL rise rate while reducing deformation rates to 25% and 50% of their original values to assess the impact of subsidence mitigation. The estimated inundation areas under different conditions based on the active inundation analysis are shown in Figure 7.

As shown in Figure 7a, the RSL along the YRD coast exhibits a rising trend near the boundary between HK and KL, particularly along the KL–DY–GR coastal zone, consistent with the current spatial distribution of subsidence bowls in the YRD. In the Gudong Oilfield area, an extensive coastal seawall has been constructed, which provides partial protection against RSL rise, especially during high tide, preventing seawater inundation of the oilfield region. However, with the continued development of land subsidence in the nearby brine extraction zones, the northern section of the coastal seawall in the Gudong Oilfield is also subsiding, which will ultimately compromise its protective capacity (Figure 6b).

The inundation simulation results in

Table 2. Inundation area and proportion under different scenarios.

Scenarios (Sea Level & Deformation Rate)	Years	Inundation Area (km2)	Inundation Area (km2) and Percentage in YRD	Build-Up Area Inundation Percentage in YRD
ASLcurrent&LS0%	2030	205.42	127.9 (2.60%)	2.33%
	2050	210.22	131.5 (2.67%)	2.39%
	2100	485.12	162.8 (3.31%)	3.15%
ASLcurrent&LS25%	2030	479.75	188.0 (3.82%)	2.74%
	2050	551.35	233.5 (4.75%)	3.10%
	2100	1014.67	320.7 (6.52%)	4.48%
ASLcurrent&LS50%	2030	558.05	238.8 (4.85%)	4.85%
	2050	949.15	302.4 (6.15%)	6.15%
	2100	1280.90	387.9 (7.88%)	7.88%
ASLcurrent&LS100%	2030	975.13	309.9 (6.30%)	4.22%
	2050	1223.54	368.8 (7.50%)	5.10%
	2100	2179.87	631.8 (12.84%)	8.74%
ASLSSP1-1.9&LS100%	2030	987.66	314.2 (6.39%)	4.27%
	2050	1220.20	367.3 (7.47%)	5.07%
	2100	2150.76	619.4 (12.59%)	8.49%
ASLSSP1-2.6&LS100%	2030	975.37	310.0 (6.30%)	4.22%
	2050	1222.55	368.4 (7.49%)	5.09%
	2100	2179.86	631.8 (12.84%)	8.74%
ASLSSP2-4.5&LS100%	2030	975.60	310.1 (6.30%)	4.22%
	2050	1226.88	370.1 (7.52%)	5.13%
	2100	2234.13	649.7 (13.20%)	9.09%
ASLSSP3-7.0&LS100%	2030	975.82	310.2 (6.31%)	4.22%
	2050	1229.31	370.2 (7.54%)	5.16%
	2100	2294.93	674.2 (13.70%)	9.63%
ASLSSP5-8.5&LS100%	2030	975.99	310.3 (6.31%)	4.23%
	2050	1237.03	373.4 (7.59%)	5.24%
	2100	2385.02	694.8 (14.12%)	10.09%

indicate that under current trends of sea level rise and land subsidence, the most affected areas in the YRD are the northeastern HK area and along Laizhou Bay. No significant differences in inundation area or distribution were observed between SSP1-1.9 and SSP5-8.5 under the current sea level rise rate. Reducing the deformation rate to 50% and 25% of the current rate while maintaining the same sea level rise led to substantial decreases in inundated areas. Specifically, with the deformation rate reduced to 50% of the current value, inundated areas in 2030, 2050, and 2100 would decrease by 43%, 22%, and 41%, respectively, and the KL coastal area would no longer be inundated by 2100. With the deformation rate reduced to 25% of the current value, the inundated areas would decrease by 51%, 54%, and 55%, respectively, and by 2050 the GR coastal area would be free from inundation.

The analysis of inundation under different deformation rates and ASL rise scenarios indicates that land subsidence is the dominant driver of seawater inundation in the YRD and surrounding regions. Without effective regulations for the sustainable development and use of underground resources, such as brine extraction, continued land subsidence will further increase the risk of seawater inundation in these areas in the future.

4. Discussion

4.1. Differences in the Causes of Land Subsidence

As shown in Figure 8a, shallow brine and deep groundwater are distributed along the coastal and inland areas of the YRD, respectively [39]. Currently, shallow brine (<100 m) is the primary target for extraction, mainly distributed along the coastal areas of WD, HK, DY, and GR [40]. Due to the widespread distribution of saline water, groundwater extraction activities are primarily conducted in inland cities, with concentrated extraction zones in the deep groundwater regions such as BX and GR [41].

The shallow brine in the YRD is primarily used for chemical industry production and aquaculture. By 2013, many areas had reached over-extraction status [42]. Prolonged extraction has reduced the mineralization of shallow brine aquifer, requiring deeper wells to maintain extraction quality. The brine pumping wells are about 40 m deep in the ZH area, 70 m deep in the DY area, and as deep as 94 m [43]. The occurrence of shallow brine is mainly in thin layers of silt and fine sand, with aquitard made of compressible fine-grained materials like clay and silty clay (Figure 1c). By 2019, the GR saltworks area had seven brine extraction companies, around 1600 pumping wells, and an annual extraction volume of approximately 3.2×10⁷ m³. Those pumping wells are distributed along salt field divisions and roads, about 100 m apart, leading to concentrated extraction [44].

The historical optical imagery shows that the HK subsidence bowls have long been associated with large salt ponds and aquaculture areas (Figure 8a). The deformation rate along profile III-III’ reveals a strong spatial correlation between subsidence and brine extraction industries (Figure 6f). The brine groundwater level and deformation exhibit same trends, where a slowdown in groundwater level decline coincides with a reduction or even reversal of subsidence. For example, the ZH area experienced a rebound of 75 mm. In the HK area, deformation rates decreased by approximately 33% (Figure 9 (#p1, #p2)). Brine extraction follows a distinct seasonal pattern, peaking from April to June and October to November. Precipitation affects both brine industrial production activities and the brine groundwater level [45]. During the rainy season, the brine groundwater level in DY rises by approximately 20 m. Additionally, subsidence near salt ponds slowed or even reversed after Super Typhoon Lekima struck the Shandong coast. It is indicated that brine extraction activities are temporarily halted due to typhoons, resulting in a reduction in subsidence (Figure 6h). Due to the rapid response of deformation in the YRD coastal area to variations in shallow groundwater level (Figure 9 (#p2)), the high-intensity precipitation brought by the typhoon temporarily recharged the shallow groundwater, resulting in a slight uplift.

The distribution and extraction of groundwater in the study area are primarily concentrated in BX and GR. The unconfined groundwater, with a burial depth of 0–60 m, is mainly utilized for agricultural irrigation and rural domestic water supply. Outside the heavily exploited areas, groundwater level dynamics closely correspond to precipitation patterns and evaporation intensity (Figure 9 (#p3–#p8)) [46].

The confined groundwater with a burial depth exceeding 200 m has been extensively exploited since 1960 (Figure 10a). From 2000 to 2020, multiple deep confined groundwater depression cones formed in the BZ-GR region, and the maximum decline of groundwater table depth reached 2–3 m/yr (Figure 10b). After 2015, the groundwater table depth in the northwestern BC area showed no significant downward trend during 2015 to 2020, indicating that groundwater extraction had essentially ceased. In the western BX and GR, particularly in the latter, the groundwater table depth continued to decline at rates of up to approximately 2 m/yr (Figure 10c,d). The BX–GR subsidence bowls correlate spatially with deep groundwater depression cones. For the groundwater level observation well #c1, during the period of deep groundwater level decline (January 2018 to December 2020), the deformation rate was about −91 mm/yr. After 2021, deep confined groundwater rapidly recovered, and the deformation rate decreased to −50 mm/yr, indicating a clear slowdown in subsidence. The groundwater level observation wells #c2, #c3, and #c4, located at the edges of the subsidence bowls, exhibit seasonal fluctuations in deep groundwater level. The InSAR time-series deformation also shows a temporary deceleration or rebound corresponding to the seasonal recovery of the groundwater level, confirming a clear link between the InSAR deformation time series and deep groundwater level variations [47].

From the distribution of subsidence bowls and groundwater extraction activities, it can be inferred that coastal subsidence is mainly caused by shallow brine extraction, with deformation characteristics influenced by extraction intensity and timing. In contrast, inland subsidence is primarily due to the extraction of deep groundwater.

4.2. Response of Deformation to Groundwater Level Variations in Corresponding Aquifers

The extensometer monitoring results for BX and GR from May 2022 to August 2023 indicate total monitoring depths of 320 m and 400 m, with cumulative deformations of −23.9 mm and −48.0 mm, respectively (Figure 11a,b). The regions exhibited similar deformation patterns, including slight rebounds during the summer months.

The shallow soil layers (<60 m) in both BX and GR are primarily composed of silt and cohesive soil (Figure 1c) and exhibit deformation patterns that closely correspond to variations in the unconfined groundwater level (Figure 11b,f). In GR, the seasonal pattern is more pronounced, likely influenced by irrigation and precipitation, which leads to periodic rises and declines in groundwater level and consequently results in alternating subsidence and rebound phases.

Soil layers deeper than 60 m form a multi-layered aquifer-aquitard system consisting of silt, cohesive soils, and sand. In BX, the 60–140 m layer exhibited noticeable rebound from May to August 2022, coinciding with a rising groundwater level in the corresponding aquifer (Figure 11c). The 260–320 m layers experienced both subsidence and rebound, but the deformation lagged behind groundwater level variations, likely due to the delayed response of deeper soils (Figure 11d). In GR, soil layers deeper than 60 m displayed fluctuating subsidence, with the 60–260 m layers exhibiting a more pronounced rebound than the 260–400 m layers during the same period. Unlike BX, the 260–400 m layers in GR continued to subside despite a continuous recovery of the groundwater level. This phenomenon can be explained by the long-term history of deep groundwater extraction in the GR region (Figure 11g).

Both BX and GR have undergone prolonged deep groundwater extraction, leading to long-term declines in groundwater level. In BX, groundwater recovery began around 2015, gradually restoring hydraulic gradients and reducing subsidence. In GR, groundwater levels continued to decline until 2021, when recovery commenced. Due to the long-term drawdown, pore water pressure in GR aquifers remains in a dissipation phase, sustaining residual deformation [48,49]. Consequently, subsidence persisted in GR even after groundwater recovery, consistent with InSAR observations.

The deformation of different soil layers in the BX and GR regions shows that the main subsidence layer in the 140–260 m and 260–400 m, respectively. Shallow soil (<60 m) deformation corresponds to unconfined groundwater level variations, whereas middle and deep layer deformation lags behind groundwater level changes. Notably, residual deformation is more evident in the GR region. The differences in the response of deep soil deformation to groundwater level variations between BX and GR are due to the distinct aquifer structures and groundwater extraction histories. In BX, groundwater extraction had nearly ceased by 2015, and the groundwater level began to rise gradually, restoring the hydraulic gradient balance within the aquifer system. As a result, subsidence has become negligible. In contrast, groundwater extraction in GR continued until 2020, with recovery beginning only in 2021, leading to ongoing residual deformation.

4.3. The Impact of Land Subsidence on RSL Rise

RSL rise is the combined effect of ASL rise and vertical land motion, particularly land subsidence. In the YRD, land subsidence dominates the RSL increase, with magnitudes several times greater than the climate-induced ASL rise of approximately 4 mm/yr. InSAR measurements along the current coastline indicate deformation ranging from −141 mm/yr (subsidence) to 22 mm/yr (uplift). Subsidence is spatially concentrated in densely populated and industrialized areas, while uplift primarily occurs along the paleochannels and at the estuary. Although sediment accumulation contributes to localized uplift, this compensatory effect is both limited and temporary. Over the long term, sedimentary layers are prone to compaction and renewed subsidence, further amplifying RSL rise [50]. Consequently, the superposition of land subsidence and ASL rise accelerates RSL increase, thereby expanding the spatial extent and severity of future coastal inundation risks in the YRD (Table 2).

The ASL and deformation data in the YRD coastal area indicate that the greatest current impact on RSL rise comes from land subsidence due to shallow brine extraction (Figure 12). Formulating reasonable policies and measures for brine extraction is crucial to slowing down the development of land subsidence and RSL rise. It should also be noted that shallow brine aquifer groundwater in the YRD is easily recharged (Figure 8b), resulting in no significant long-term drop in groundwater level. However, long-term exploitation leads to a decrease in the mineralization degree of brine aquifers, which further drives the deepening of brine extraction depths. As pump depths increase, residual deformation in deep soil layers may occur, leading to sustained elevation loss and further exacerbating RSL rise, thereby increasing the potential for more extensive future coastal inundation.

4.4. Coastal Inundation Risk and Countermeasures

The inundated area in this simulation primarily consists of coastal lowlands, which contain numerous salt ponds and aquaculture pools. These areas are fragmented into several coastal low-lying zones, making them more susceptible to large-scale inundation. Based on ESRI’s LULC data for 2023, it is estimated that by 2030, 2050, and 2100, 4.22%, 5.10%, and 8.74% of the built-up area in the YRD will be inundated (Figure 7c). The current inundation estimates at the current ASL rise rate show no significant difference in the inundated area and distribution compared to the most severe scenarios (SSP5-8.5) except for a general inland expansion of the inundated regions. The predicted inundated areas show that, regardless of the SSP scenario, the salt pans and aquaculture ponds along the YRD coast are highly likely to be inundated, with a trend of expanding toward inland residential areas and the Gudong Oil Field. This will directly impact the lives of coastal residents and the normal operation of related industries, causing significant socioeconomic losses. Additionally, these areas face continuous elevation loss, increasing the risk of extreme precipitation events. For example, in 2019, Super Typhoon Lekima caused 48 h of intense rainfall on the YRD, with a total precipitation of over 250 mm, leading to major losses in coastal industries [51,52].

It is important to note that the results of this coastal inundation simulation do not account for human intervention. Considering socioeconomic and practical factors, adaptation and protection strategies have been widely implemented. Initial responses to sea level rise typically include inundation risk management, seawall reinforcement, and measures to address localized threats from rising sea levels [53,54] The simulation results further indicate that seawalls near the Gudong Oil Field have successfully prevented inundation, suggesting that similar measures should be considered for other potentially inundated areas, such as the eastern HK and DY–GR coastal lines, to mitigate future inundation risks. Additionally, the distribution of inundated areas is strongly correlated with subsidence bowls, emphasizing the importance of limiting or adjusting underground resource extraction as a key strategy for reducing future coastal inundation risks.

5. Conclusions

This study employs TS-InSAR and an active inundation algorithm, integrated with multi-source datasets, to comprehensively assess land subsidence and RSL rise-driven inundation in the YRD. This study identifies the main subsidence layers and quantifies the deformation responses of different soil layers to groundwater level fluctuations in a multi-aquifer system. The high vertical accuracy CoastalDEM v2.1 is applied with an active inundation algorithm to simulate land inundation under different combinations of ASL rise and land subsidence scenarios in the YRD. The main conclusions are as follows:

• Multiple subsidence bowls are observed in both inland and coastal zones, with a maximum deformation rate of −232.8 mm/yr. Inland subsidence (BX–GR) is primarily driven by deep groundwater extraction, while coastal subsidence (HK–KL–GR) is linked to shallow brine extraction. Typhoon Lekima disrupted brine extraction and recharged groundwater via heavy rainfall, resulting in temporary rebound in coastal areas.
• The main subsidence layers are located at depths of 140–260 m in BX and 260–400 m in GR. Persistent deep groundwater extraction has caused long-term residual deformation, especially in GR. With declining mineralization of coastal aquifers due to brine extraction, future pumping is expected to shift to greater depths, potentially increasing residual deformation and underestimating long-term subsidence risk.
• CoastalDEM v2.1 exhibits higher elevation accuracy compared to SRTM, which improves the reliability of seawater inundation analysis. The active inundation algorithm reveals that land subsidence, rather than ASL rise (e.g., SSP5-8.5), is the dominant driver of future coastal inundation. Under current subsidence and ASL rise trends, 12.84% of the land and 8.74% of built-up areas may be inundated by 2100. Mitigating subsidence particularly by regulating shallow brine extraction can significantly reduce inundation risk.

In summary, this study integrates TS-InSAR, an active inundation algorithm, and in-situ observation datasets to systematically assess the spatiotemporal evolution of land subsidence in the YRD and its impact on coastal inundation risk induced by future RSL rise. The results demonstrate that subsidence caused by coastal brine exploitation is the primary factor exacerbating future inundation risk. The study underscores the urgent need for integrated groundwater resource management and adaptive coastal zone strategies to mitigate the compound disaster risks associated with land subsidence and sea level rise, ensuring long-term sustainability in vulnerable deltaic regions.