Modeling the Effects of Underground Brine Extraction on Shallow Groundwater Flow and Oilfield Fluid Leakage Pathways in the Yellow River Delta

Abstract

The distribution of fresh and salty groundwater is a critical factor affecting the coastal wetlands. However, the dynamics of groundwater flow and salinity in river deltas remain unclear due to complex hydrological settings and impacts of human activities. The uniqueness of the Yellow River Delta (YRD) lies in its relatively short formation time, the frequent salinization and freshening alternation associated with changes in the course of the Yellow River, and the extensive impacts of oil production and underground brine extraction. This study employed a detailed hydrogeological modeling approach to investigate groundwater flow and the impacts of oil field brine leakage in the YRD. To characterize the heterogeneity of the aquifer, a sediment texture model was constructed based on a geotechnical borehole database for the top 30 m of the YRD. A detailed variable-density groundwater model was then constructed to simulate the salinity distribution in the predevelopment period and disturbance by brine extraction in the past decades. Probabilistic particle tracking simulation was implemented to assess the alterations in groundwater flow resulting from brine resource development and evaluate the potential risk of salinity contamination from oil well fields. Simulations show that the limited extraction of brine groundwater has significantly altered the hydraulic gradient and groundwater flow pattern accounting for the less permeable sediments in the delta. The vertical gradient increased by brine pumping has mitigated the salinization process of the shallow groundwater which supports the coastal wetlands. The low groundwater velocity and long travel time suggest that the peak salinity concentration would be greatly reduced, reaching the deep aquifers accounting for dispersion and dilution. Further detailed investigation of the complex groundwater salinization process in the YRD is necessary, as well as its association with alternations in the hydraulic gradient by brine extraction and water injection/production in the oilfield.

1. Introduction

The Yellow River Delta (YRD), among the world’s youngest river deltas and critical coastal wetland ecosystems [1,2], exemplifies environmental pressures facing river deltas that sustain over 500 million people globally. Its low elevation makes the YRD particularly vulnerable to storm surges and salty water intrusion. Soil salinization has always been a critical issue in the YRD [3,4]. Frequent drought, rising sea levels, and declining groundwater levels due to oilfield activities have exacerbated the threat of sea water intrusion [5]. Since the 1980s, wetland degradation has accelerated due to reduced runoff and storm surge-induced groundwater salinization [6]. The area of coastal wetland has shrunk from approximately 2000 km² in 1995 to 800 km² in 2015. To protect the Shengli Oilfield (China’s second largest petroleum reservoir) from storm surge, a 16.88 km seawall was constructed in 1987. However, the rapid development of land subsidence in the YRD has intensified the risk of salty water intrusion. Moreover, environmental investigations have confirmed groundwater contamination by polycyclic aromatic hydrocarbons (PAHs), heavy metals, and nitrates in the YRD [7,8,9,10]. The shallow aquifer supporting the wetlands is at risk of pollutants discharged through well leakage, drilling mud pond seepage, and pipeline failure [11]. By the end of 2013, remediation efforts have safely treated a total of 4000 drilling mud ponds. All produced water has been injected back into formations to enhance oil recovery and waste disposal.

Oil and produced water can leak from wells if the casing is damaged, particularly in abandoned wells. After nearly 50 years of exploration and development in Shengli Oilfield, casing damage has become a widespread issue, affecting approximately 30% of wells [12]. Most of these wells were abandoned due to a decline in oil production resulting from insufficient formation pressure after prolonged exploitation. The injection of plugging polymer to reduce water infiltration loss may also affect groundwater quality. Because of the highly saline groundwater in the YRD, the effects of brine water disposal on groundwater salinity may not be evident. Assessing the pollution risk to wetlands requires a clear understanding of the potential migration pathways and the hydrodynamic conditions of the shallow groundwater system [13]. Sufficient information regarding the hydraulic relation between oil production or water injection wells and shallow aquifers is crucial for the evaluation of the effects of oil production on wetlands.

Underground brine exploitation in the YRD is a major factor influencing local hydraulic conditions [14]. Currently, brine development efforts in the YRD focus on formations at depths of 120–150 m, while deep brine resources remain unexploited [15]. By 2011, underground brine extraction had reached 0.12 billion m³/yr for salt production and aquaculture. The groundwater level in the brine layers has decreased approximately 60 m. Intensive underground brine exploitation has caused significant land subsidence, particularly in the northern YRD [16]. The InSAR-derived maximum subsiding rate can reach up to 250 mm/yr at aquaculture facilities [15]. Although oil and gas mining, geothermal water withdrawal, and natural compaction of the sediments may also contribute to surface deformation [16,17,18], underground brine extraction remains the primary contributor to land subsidence in the YRD [2]. However, due to the lack of multi-level groundwater monitoring in the YRD, data on the vertical flow of groundwater across the shallow aquifer, brine formations, and oil-bearing layers remain unavailable.

The aim of this study is to investigate the spatial distribution of potential oilfield fluid migration pathways and to address groundwater monitoring issues in regions affected by underground brine extraction. The analysis focuses on the YRD, which has a long history of oil production and underground brine resource development. A variable-density numerical groundwater flow model is developed and calibrated using monitored groundwater levels and salinity distribution patterns. Particle tracking is used to evaluate potential fluid migration pathways. The uncertainty arising from aquifer heterogeneity is examined through multiple runs of a sediment texture model, with pathway interception probabilities using a particle counting method.

2. Material and Methods

2.1. Regional Geology and Hydrogeology

This study focuses on the modern YRD (Figure 1a), which has developed since 1855, when the Yellow River’s course migrated from the Yellow Sea back to the Bohai sea [4]. The YRD is dominated by a monsoon climate with an annual precipitation of approximately 590 mm. Due to the high sediment content of the Yellow River, a large amount of sediment enters the estuary every year, resulting in the long-term and ceaseless changes in accumulation, extension, oscillation, and channel shifting in the estuary [19,20]. The terrain of the YRD is lower in the northeast and higher in the southwest, with the abandoned channels of the Yellow River forming high ridges and relatively low-lying areas between the old channels. The current flow channel of the Yellow River, Qingshui Gou, flows into the Bohai Sea from the southwest to the northeast, creating a fan-shaped terrain [21]. This landform slopes from the middle to the sides along the southwest–northeast direction. The delta terrain is relatively gentle, with elevations ranging from 2.0 to 9.0 m, and the natural gradient of the delta is between 1/8000 and 1/12,000 [5].

The quaternary geology of the YRD is mainly influenced by the sea-level changes during the glacial–interglacial periods. It has been identified that around the Bohai Sea, there have been three transgressions since the late Pleistocene. The first transgression occurred 75,000 to 127,000 years ago and deposited marine sediments with a thickness of about 20 to 30 m, known as the third marine facies layer. According to the borehole data of the YRD, the top interface of this marine facies layer is located 55.4 m below the surface. The sediments are yellow, bluish-gray fine sand and sandy fine sand, containing foraminifera and marine ostracods, indicating tidal flat and shallow marine environmental deposition. The second transgression occurred 23,000 to 31,000 years ago, with a similar extent to the first transgression. In the YRD, this layer is 6.19 m thick, buried at a depth of 37 to 43 m, composed of dark gray to grayish-blue clayey fine sand and yellow sandy fine sand, containing foraminifera, marine ostracods, and mollusk fossils, with biogenic burrows developed, representing the deposition of shallow seas and intertidal flats, known as the second marine facies layer. The third transgression reached the YRD region around 8800 years ago, with the highest sea level in this area occurring around 6000 years ago. In the YRD, the base of this marine facies layer is generally buried at a depth of 20 m, consisting of sandy and clayey fine sand, interlayered with clay, known as the first marine facies layer. Both the upper and lower boundaries of each marine facies layer are terrestrial deposits, creating conditions for the unique distribution of the Yellow River groundwater due to the alternating appearance of terrestrial and marine facies.

The total dissolved solid (TDS) concentration of shallow groundwater, typically buried at depths less than 20 m in the YRD, ranges from 1 to 38 g/L [23]. The TDS concentration of salty groundwater at depths of 50–200 m ranges from 24 to 113 g/L [24]. Based on buried depth, underground brine in the YRD is classified into shallow, mid-deep, and deep brine. Shallow brine is mainly located in the southeast of YRD at a depth of 18–120 m. Mid-deep brine is primarily in the northeast at a depth of 102–195 m. Deep brine is mainly located in the south of YRD at a depth of 2400–3000 m. The estimated total brine resource volume is 5.84 billion m³, with a potentially exploitable resource volume of 0.86 billion cubic m³. The brine extraction rate is estimated at 600,000 m³/yr. Since 2022, the extraction rate of the brine has been reduced to mitigate the land subsidence issue in the YRD.

2.2. Stratigraphy Simulation Based on Transition Probability

The impact of aquifer heterogeneity is generally quantified through Monte Carlo-based simulations of lithology facies distributions, e.g., [25,26]. Lithology records of boreholes were used to determine primary sediment types and construct the stratigraphy model. A total of 179 boreholes were collected from the engineering geological investigations in the YRD carried out by the China Geological Survey. Most boreholes have drilling depths of less than 30 m, with core stratigraphic thicknesses generally less than 3 m. The spatial density of boreholes is approximately 5 per 100 km². The limited number of boreholes and low spatial density is the most important constraint for stratigraphy model construction, making it difficult to accurately simulate the sediments distribution particularly in oilfields and brine pumping zones. Although great limitations exist in the current model, it can be continuously improved when more boreholes or geophysical information are available.

The sediment texture model was constructed using the T-PROGS code [27], which performs transition probability geostatistics to generate multiple stratigraphy models. The sediment types were reclassified according to lithology descriptions in the borehole logs into three categories: silty sand, silt, and clay. The three sediment types account for 43%, 16% and 41% of the total borehole core thickness, respectively. Although the silt is generally incorporated into the clay in modeling works for alluvial plains, e.g., [25,26], using it as a separate major sediment type to characterize the spatial continuity may benefit the salinity transport simulation in this study. The transition probability between sediment types is used to represent the spatial structure of the sediments and can be calculated as the following conditional probability:

$$t_{i,k}(h)=\mathrm{Pr}\left\{k\mathrm{occurs}\mathrm{at}x+h)\left|i\mathrm{occurs}\mathrm{at}x\right.\right\}$$

(1)

where t_i,k is the transition probability from sediment types i to k; x is a spatial location; and h is the lag distance. The occurrence of sediment type k at location x + h is only dependent on the occurrence of sediment type i at location x. All the generated models can be conditioned to the borehole data and thus can be utilized for model uncertainty analyses with respect to aquifer heterogeneity [28,29].

The transition probabilities for the vertical directions were first calculated, and a Markov chain model was developed to fit the observed data (Figure 2). The embedded transition probability matrix was adjusted to achieve a fit between the Markov chain and the transition probabilities in the borehole data (

Table 1. Embedded transition probability matrix used in the T-PROGS simulations. The diagonal elements are the observed thicknesses and mean lengths of the sediment types in the vertical and horizontal directions, respectively. The other elements are the transition probabilities between sediment types.

Sediment Type	Vertical Direction			Horizontal Direction
	Silty Sand	Silt	Clay	Silty Sand	Silt	Clay
Silty sand	5.137 m	0.244	0.756	500.0 m	0.244	0.756
Silt	0.375	4.290 m	0.625	0.375	400.0 m	0.625
Clay	0.804	0.195	4.734 m	0.789	0.211	455.4 m

). Because the number of boreholes was small and the borehole data were not sufficiently dense in the horizontal direction, the Markov chain models for the horizontal directions were developed assuming that the material proportions observed in the vertical directions hold true in the horizontal direction. The diagonal elements in Table 1 are the observed thicknesses and mean lengths of the sediment types in the vertical and horizontal directions. The mean lengths of the sediments can be adjusted to represent different degree horizontal continuity. Because the horizontal length scale of sediment types may be smaller than the spacing between boreholes, the estimation of mean lengths in horizontal directions from transition probabilities may be problematic [30]. Mean lengths of 500.0 m and 400.0 m in the horizontal direction were used for the silty sand and silt, respectively. They are reasonable and consistent with the spatial continuity assessed from cross sections in the YRD [31]. A horizontal length scale of hundreds of meters is usually used in transition probability depositional model construction for alluvial fan aquifer [32,33,34]. Isotropic behavior in the horizontal plane was assumed, and clay was used as the background sediment. After setting the horizontal lengths for the silty sand and silt, the program would automatically calculate the horizontal length values for the background lithology (Table 1). The Markov chains were converted into a continuous three-dimensional Markov chain. The three-dimensional cell size of 200 m, 200 m, and 3 m was used in the depositional strike, dip, and vertical directions, respectively, which are less than the mean lengths of each sediment type. Sequential indicator simulations were then carried out to generate the distributions of sediment types.

2.3. Variable-Density Groundwater Flow Model

To reveal the effects of salinity on groundwater flow in the YRD, the variable-density flow simulation code SEAWAT [35] was implemented to construct the flow model. The modeling domain extends from the south to the Yellow River, west to the ancient course of the Diaokou River, and southeast to the coastline, covering an area of 1238 km² (Figure 1). The modeling domain was divided into 290 rows, 315 columns, using a uniform grid size of 200 × 200 m, and 12 layers vertically. The depths of most groundwater monitoring wells in the study region are less than 30 m. Considering the depths of monitoring wells and boreholes, the groundwater flow field within 30 m from the ground surface is of interest in this study. The ground surface elevation serves as the top elevation. The bottom elevation of model layer 1 was assigned a uniform value of −3 m above the mean sea level (a.m.s.l). The thickness of model layers 2 to 10 was 3 m. A model layer with a thickness of 30 m (layer 12) was used to represent the brine water. To reduce the number of grid cells, a thick single layer with a thickness of 90 m was used to represent the shallow aquifer below −30 m and above the brine water layer. There are a total of 371,400 effective cells.

The primary recharge sources of groundwater in the study region include the infiltration of precipitation and leakage from rivers [14,31]. The current flowing course and two ancient channels of the Yellow River were simulated as head-dependent flux boundaries in the model. The boundary head values were determined by linear interpolation of observed river stages along the channels. The coastline was simulated as a fixed-head boundary with the constant salinity concentration of the sea. The recharge from precipitation was determined by multiplying an infiltration coefficient, which was derived using the water table fluctuation method [36]. Groundwater evapotranspiration was simulated by the EVT package, assuming a linear evaporation model. The potential evapotranspiration rate was calculated using the Penman–Monteith equation [37], and an extinction depth of 4.5 m was used.

The measured groundwater TDS concentration in boreholes with depths less than 50 m was interpolated using the Ordinary Kriging method and was used as the initial salinity concentration. Frequent changes in routine locations of the Yellow River had caused freshening of the salty water near the rivers [5]. The model was run under transient mode for a 150-year preheat period to simulate the effects of river leakage freshening during the formation of the modern YRD. The following 20-years period was used to simulate brine groundwater pumping, assuming a constant pumping rate. Observed annual precipitation and river stages from 1970 to 2017 were averaged and used as the model input, assuming the multiyear-averaged hydraulic conditions remained constant during the simulation period.

Model parameters include the hydraulic conductivity, specific storage, and specific yield of the sediment facies. The hydraulic conductivities of silty sand, silt, and clay were estimated at 0.83, 0.23, and 0.1, respectively, using pedotransfer functions (PTFs) via the Rosetta code [38]. They were used as the initial values in the parameter optimizing process. The specific yields for the silty sand, silt, and clay were estimated at 0.10, 0.02, and 0.01, respectively, using the water table fluctuation method [39]. A uniform value of 0.0001 was assigned for specific storage. The vertical hydraulic conductivity of the Diaokou River’s riverbed was measured at 0.03 m/d using the standpipe test [40]. The sediment thickness of the Diaokou riverbed is set to 0.3 m. The riverbed sediment thickness of the current flow path of the Yellow River was set to 0.5 m. The hydraulic conductance of the riverbed sediments was then estimated based on the width of the river channel intersecting the model grid cell.

Model calibration was carried out based on observed groundwater levels from 12 piezometers with depths less than 2 m and 9 monitoring wells with depths of 30 m. The vertical conductivities of the three lithologies were adjusted to obtain an agreement between the observed and simulated vertical hydraulic gradient. Considering the varied TDS concentration, the simulated equivalent freshwater heads were converted to native groundwater head for comparison with the field measured data [35]. The model was calibrated by using a manual trial-and-error method and by employing the PEST [41], a universal code for model sensitivity analysis and calibration. The PEST software uses the Levenberg–Marquardt optimization algorithm to minimize the objective function, i.e., the weighted sum of the squared differences between model predictions and observed data. PEST version 11.3 was using in this study for model calibration.

2.4. Delineating Flowpaths Through Particle Tracking

Particle tracking was implemented using the MODPATH code [42], which uses simulated volumetric fluxes from the groundwater flow model to calculate linear velocity. Particle tracking has been used to analyze groundwater flow paths in many regions [43,44,45]. Probabilistic particle tracking can be used to investigate model uncertainty and identify potential contamination sources [46,47]. Forward and backward particle tracking was conducted in this study to reveal the alternations of flow path by brine groundwater pumping. In the forward particle tracking simulations, particles were released from selected 38 oil wells and water injection wells. Particles were tracked to the end position to obtain the longest traveling distance. In the backward particle tracking simulations, twenty particles from each of the five virtual brine pumping fields were released and tracked backward to their starting locations, resulting in a capture zone of the brine wells. Groundwater levels monitored in brine pumping wells showed small temporal variations, indicating that the groundwater had reached a near-equilibrium state for such a low-permeability aquifer system. Moreover, the variable-density flow model can only be run in transient mode, and the simulated time period may not be long enough to track the entire flow path. Therefore, a steady-state flow field was assumed, and the simulated groundwater budgets in the last time step were used to provide the fluxes for particle tracking. In the steady-state model, changes in porosity values will not change the simulated flow path but will change calculated travel time. A uniform porosity of 0.3 was used in the particle tracking simulations.

Forward particle tracking was used to provide a conservative estimate of potential contaminant migration. Particles were released from 38 oil extraction and brine disposal water injection well fields and were tracked forward to demonstrate the flow path affected by brine pumping. These well fields were selected from the oil well datasets delineated by image identification [48] to represent regions with high-density oil activity. To simulate brine pumping effects on the flow field in three-dimensional space, a particle was placed each 3 m along the vertical directions. Stochastic simulations based on multiple reasonable geological statistics of aquifer structure were used in flowing path uncertainty analysis, associated with the uncertainty of hydrogeological structure. Ideally, this method can characterize the spatial variability of aquifer lithology based on interpretable geological parameters (such as lens geometry parameters, lithology proportions, and anisotropy). The Monte Carlo particle tracking simulation conducted with the transition probability stratigraphy model provided 50 model runs, resulting in a total of 95,000 particle tracks.

Probabilistic flow solutions drive particle tracking to obtain the probability of a path passing the grid cells. Spatial analysis was conducted to identify the regions with a high risk of brine leakage contamination. The spatial analysis identified the number of particle tracks that intercepted each grid cell, and the number of particle tracks were summarized for all layers to calculate the interception probability. A scripted process was executed to convert the simulated path lines to interception probabilities for analysis in GIS.

3. Results and Discussion

3.1. Sediment Stratigraphy

A series of 50 equally probable realizations were generated in this study. Figure 3 shows one realization of the sediment spatial distribution in the YDR used in the groundwater flow and particle tracking simulations. The dominant sediments in the YRD are silty sand and clay, with volumetric proportions of 43% and 42%, respectively. The simulated volumetric proportions of the three sediments are consistent with their measured proportions of the total sediment thickness in the boreholes. Because of the relatively small number of boreholes, the horizontal continuity of the sediments depends on the values of specified mean length and may bring a certain degree of uncertainty.

3.2. Groundwater Salinity in Predevelopment Conditions

The simulated groundwater TDS concentration using one realization of sediment distribution is shown in Figure 4. Most of the YRD area is occupied by salty groundwater with TDS concentrations greater than 10 g/L, which represents the highest level of TDS concentration of water that could be considered as a potential source of drinking water. The simulated inland boundary of this TDS threshold is in agreement with the derived position, based on the groundwater salinity observed in 2005. Most fresh and brackish groundwater is distributed in a narrow zone along the current and past flow paths of the Yellow River, as well as the floodplains in the southeast part with a relatively high elevation. This indicates that salinity freshening by the leakage of freshwater from river channels and the infiltration of precipitation is limited to some local regions, and salinization is the primary process affecting groundwater salinity in the whole YRD.

3.3. Groundwater Flow and Salinity Affected by Brine Extraction

The simulated and observed water levels of the 21 observation wells after parameter optimization, using one realization of the sediment texture model, are shown in Figure 5. The horizontal hydraulic conductivities for fine sand, silt, and clay are 0.70 m/d, 0.54 m/d, and 0.10 m/d, respectively. The Root Mean Square Error (RMSE) and correlation coefficient (R) are 2.66 m and 0.92, respectively. The comparison of the average groundwater levels from the 50 realizations with the observed groundwater level is shown in Figure 5. It is noteworthy that, due to the heterogeneous structure of the aquifer controlled by the lithology distribution generated by the hydrogeological structural model, there may still be spatial differences in the permeability of the same lithology within the structural model. Adjusting lithological hydrogeological parameters alone makes it difficult to effectively reduce the optimization objective function. Moreover, the output of the SEAWAT code is the equivalent freshwater head, which cannot be used in comparison with the groundwater level observed in wells. The model’s simulated equivalent head was converted to the native groundwater level using the simulated TDS concentrations. Therefore, the error in the simulated TDS concentration may also cause errors in the simulated groundwater level.

The simulated groundwater head in model layer 1 with a depth of 3 m gradually decreases from the southwest to northeast coast (Figure 6a), which can represent the groundwater level distribution pattern under natural conditions. The groundwater level in model layer 10 with a depth of 30 m shows an overall decline compared to layer 1 (Figure 6b), indicating that the groundwater level has been pulled down by leakage into the brine aquifer. The groundwater drawdown funnels and converging areas formed by several major brine extraction areas cover most of the study area, indicating that the hydraulic gradient and groundwater flow have been significantly altered by brine extraction. Exploitation of middle and deep groundwater brine has formed several groundwater-level depression cones centered around the brine extraction wells, with the lowest groundwater level in the center of the funnel reaching approximately −40 m (Figure 6c). This is consistent with the observed groundwater levels at the brine pumping wells near the Gudao Oilfield. Several depression cones have interconnected, leading to widespread groundwater level decline in the northwest of Gudao Town and the area north of Gudong Oilfield, bounded by a groundwater head contour line of −10 m (Figure 6c).

Due to the low vertical permeability of the aquifer, the influence of deep-level extraction on shallow groundwater levels is minimal, with an overall trend of decreasing groundwater levels from southwest to northeast. From shallow to deep layers, the groundwater level gradually decreases, resulting in a maximum vertical gradient of 0.10~0.38 m/m around the brine extraction well fields (Figure 6d). This indicates that the shallow groundwater flows towards the deep groundwater through leakage. Brine extraction has had a significant impact on the flow direction of shallow groundwater, but at a depth of 30 m, the flow direction has been noticeably altered by deep-level extraction. The groundwater level depression cone in the northwest of Gudong Oil Field has become the final discharge zone, leading to the convergence of groundwater towards the center of the depression cone.

Given that detailed information regarding brine abstraction wells, including their locations and pumping volumes, has not been comprehensively investigated, it is challenging to accurately represent the location and shape of the depression cones through the simulated groundwater levels. In consideration of the computational efficiency of the variable-density flow model, uniform year-length stress periods were adopted. Consequently, the impacts of tides, seasonal fluctuations in precipitation, and river stage variations cannot be incorporated into the model. Seasonal or daily groundwater dynamics in the YRD were not simulated in the current model.

3.4. Potential Impacts on Brine Disposal Water Transport

Contaminant migration paths in the YRD have been significantly influenced by brine pumping in the plan view of those simulated particle tracks (Figure 7a). The particle path lines simulated using one realization of sediment stratigraphy ultimately terminate in the brine pumping fields, indicating that the final discharge path of groundwater has shifted from Bohai Bay to the brine extraction fields. The groundwater from well fields or oil production plants at the Gudao Oilfield is collected in the brine extraction area north of Gudao Town. Groundwater from the east part of the Gudao Oilfield, as well as from well fields or oil production plants at Gudong Oil Field, converges towards the brine extraction areas to the west and northwest of Gudong. Groundwater from the Zhuangxi Oilfield well fields converges towards the brine extraction area near Pile 5. The actual groundwater flow direction has deviated significantly from that inferred from the shallow groundwater gradient. The brine extraction areas form groundwater receptors. The maximum particle velocities are 1.5~2.0 m/year, indicating that the impacts of contaminant leakage are likely to be restricted in the immediate vicinity of the wells.

The spatial analysis identified the number of particle tracks that intercepted each grid cell, and the number of particle tracks were summarized for all layers to calculate the interception probability in the plan view (Figure 7b). The uncertainty associated with the sediment stratigraphy simulation causes a much wider region with brine contamination risk compared with that identified by single model run. Generally, the area with high interception probability is restricted to the vicinity of the well. However, the overlap of particle paths from multiple wells causes higher interception probabilities in regions with high well densities. The interception probability is the highest in the Zhuangxi Oil Field region because of the short distance between the oil well field and brine pumping field. The interception probability map for model cells in layers 1~10 (Figure 8a) shows that the interception region of the particle paths is limited near wells along the coastline. This indicates that vertical particle movement is dominant in the shallow aquifer. However, a longer particle path flowing through the shallow aquifer can still be found for the Gudong Oilfield in the southeast floodplains. There are no groundwater pumping wells in the shallow aquifer in this region, and the coastal wetlands, the most important receptors, may be influenced by brine and contaminant leakage. The shorter horizontally traveling distance in the shallow aquifer of particles starting at the Gudao and Zhuangxi oilfields suggests that their influence on wetlands is weaker than the Gudong Oilfield. The interception probability map for model cells in layers 11~12 (Figure 8b) suggests that the deep aquifer and those brine pumping well fields are the primary receptors. However, given the low groundwater velocities and long travel time to reach those brine well fields, the peak concentrations would be greatly reduced, accounting for dilution and degradation.

3.5. Implication for Groundwater Monitoring

The placement of new monitoring wells for the early detection of groundwater quality impacts can be informed by simulation results. According to technical guidelines issued by the Ministry of Ecology and Environment of China, at least three monitoring wells must be installed downstream of the pollutant migration path. Currently, 62 monitoring wells are located in the shallow aquifer of the study area, corresponding to an average density of 5 wells per 100 km². However, monitoring wells are sparse in the Gudao and Zhuangxi oilfields, where the interception probability is high. The current layout of groundwater monitoring wells at oil well fields and production facilities is primarily based on the natural groundwater flow direction in the YRD, which generally moves from southeast to northwest. Simulated particle paths (Figure 7a) reveal that the direction of groundwater flow has shifted seaward to converging towards the neighboring brine extraction zones. The Gudong Oilfield is influenced by brine extraction to its northwest, while the Zhuangxi Oilfield is affected by brine extraction to its south. As a result, existing monitoring wells are often misaligned with actual flow paths, limiting their effectiveness in detecting potential groundwater contamination. Therefore, accurately identifying the locations of brine pumping wells and the resulting drawdown distribution within the oilfield is essential for guiding the construction of new monitoring networks. Analyzing the uncertainty of oilfield leakage migration pathways attributed to aquifer heterogeneity can assist in evaluating the effectiveness of monitoring strategies and guiding the installation of new monitoring wells. Reporting on the casing failure of oil production and water injection wells in oilfields indicates a relatively high risk of contamination in overlying aquifers [9]. Regular monitoring of well integrity, groundwater levels in both aquifers and oil wells, and the geochemical properties of groundwater and injection water is essential for managing the risks of oil contamination in the shallow aquifer supporting coastal wetlands in the YRD.

4. Conclusions

This study examined the impact of underground brine exploitation on groundwater dynamics and the migration pathways of oilfield fluid leakage in the YRD, using probabilistic groundwater flow modeling and particle tracking. A probabilistic model combing sediment texture distribution simulation and particle tracking was developed to evaluate the uncertainty in groundwater flow pathways. Simulation results reveal substantial variations in groundwater level distributions and flow patterns caused by brine extraction. Groundwater flow directions vary with depth, and shift from the original seaward pattern to a converging flow towards the brine extraction well fields. The horizontal migration distance and area of oilfield fluid leakage in the shallow aquifer has been significantly reduced by the depression cones resulting from brine extraction. Discrepancies between flow direction inferred from groundwater table distributions and derived from particle tracking highlight the importance of understanding the connectivity between shallow and deep groundwater in assessing fluid migration in the YRD. However, most existing groundwater monitoring wells in the oilfields only observe the water table or groundwater level in a single shallow layer, making it difficult to characterize vertical groundwater movement. The current groundwater monitoring networks, designed according to groundwater table distributions in the YRD, may be insufficient to capture fluids released from oilfields. Establishing a multi-level groundwater monitoring network spanning shallow to deep layers to capture groundwater level variations across different depths is recommended. Given the severe groundwater overexploitation in the YRD, the probabilistic particle tracking approach offers a valuable tool for designing and optimizing groundwater monitoring networks to assess impacts of oilfield activities on wetlands.