Evaluating LNAPL-Contaminated Distribution in Urban Underground Areas with Groundwater Fluctuations Using a Large-Scale Soil Tank Experiment

Abstract

Understanding the behavior of light non-aqueous phase liquids (LNAPLs) in urban subsurface environments is essential to developing effective pollution control strategies, designing remediation systems, and managing waste and resources sustainably. Oil leakage from urban industrial facilities, underground pipelines, and fueling systems often leads to contamination that is challenging to characterize due to complex soil structures, limited access beneath densely built infrastructure, and dynamic groundwater conditions. In this study, we integrate a large-scale soil tank experiment with multiphase flow simulations to elucidate LNAPL distribution mechanisms under fluctuating groundwater conditions. A 2.4-m-by-2.4-m-by-0.6-m soil tank was used to visualize oil movement with high-resolution multispectral imaging, enabling a quantitative evaluation of saturation distribution over time. The results showed that a rapid rise in groundwater can trap 60–70% of the high-saturation LNAPL below the water table. In contrast, a subsequent slow rise leaves 10–20% residual saturation within pore spaces. These results suggest that vertical redistribution caused by groundwater oscillation significantly increases residual contamination, which cannot be evaluated using static groundwater assumptions. Comparisons with a commonly used NAPL simulator revealed that conventional models overestimate lateral spreading and underestimate trapped residual oil, thus highlighting the need for improved constitutive models and numerical schemes that can capture sharp saturation fronts. These results emphasize that an accurate assessment of LNAPL contamination in urban settings requires an explicit consideration of groundwater fluctuation and dynamic multiphase interactions. Insights from this study support rational monitoring network design, reduce uncertainty in remediation planning, and contribute to sustainable urban environmental management by improving risk evaluation and preventing the long-term spread of pollution.

1. Introduction

In urban areas and active industrial sites, in situ remediation is often necessary [1]. The applicability of pump-and-treat or bioremediation methods depends on several factors [2], including the soil structure, groundwater flow conditions, and the distribution of contaminants at the site. Detailed site characterization is therefore required for efficient in situ remediation, to a greater extent than for containment or excavation measures [3]. However, the core method of site characterization, boring, is time-consuming and costly when attempting to increase the number of points for detailed investigation [3]. Contamination is often found beneath urban areas or operational structures, which limits the applicability of boring surveys [1]. Understanding the oil permeation mechanism within the soil and the characteristics of contamination distribution is therefore crucial for accurately assessing contamination concentration and selecting rational remediation measures with a limited number of boreholes [2].

Subsurface contamination by light non-aqueous phase liquids (LNAPLs), such as petroleum hydrocarbons, resulting from underground storage tank leakages and surface spillages, represents one of the most challenging global geo-environmental issues [4,5]. LNAPLs are characterized as being less dense than water and immiscible, causing them to migrate vertically through the unsaturated zone under the combined acting forces of gravity, viscosity, and capillarity [4,5]. Upon reaching the water table, they accumulate and spread to form lens-shaped plumes [4,6]. While early conceptual models considered LNAPL to be a uniform “oil pancake” floating atop the water table, modern multiphase flow theory has demonstrated that capillary phenomena actually cause LNAPL to be distributed with non-uniform saturations across a wide range above and below the water table [5,7].

The migration pathways and spatial distribution of LNAPLs are strongly dependent on layered heterogeneity and the pore structure of the soil [4,8,9]. Specifically, the interface between fine- and coarse-grained soil layers acts as a “capillary barrier” depending on their permeability ratio [10,11]. This mechanism can intercept vertical migration and induce horizontal lateral spreading along the interface [8,9,11]. Furthermore, a double-porosity structure—consisting of intra-aggregate and inter-aggregate pores commonly found in agricultural topsoils and fractured rocks—exhibits bimodal pore-size distributions that can accelerate LNAPL infiltration compared to single-porosity media, leading to more complex contamination patterns [4,7].

Dynamic environmental changes in the subsurface also significantly influence LNAPL pollution mechanisms [12,13]. Groundwater table fluctuations lead to the entrapment of LNAPL beneath the water table, creating a “smear zone” that serves as a long-term secondary source of contamination [12,13,14]. Recent research has further highlighted that variations in the “orientation” or tilt of the water table, induced by factors such as rainfall infiltration, can significantly reorient and redistribute LNAPL plumes [13]. In cold regions, freeze–thaw cycles facilitate the remobilization of stable LNAPL through freezing-induced pressure and changes in capillary pressure, promoting further downward breakthrough across otherwise impermeable layered interfaces [11,15]. To accurately characterize these complex transport and retention mechanisms, high-resolution and non-destructive sensing technologies such as Digital Image Analysis and Laser-Induced Fluorescence have been successfully implemented [12,14,16]. Integrating these experimental insights with advanced multiphase numerical simulators, such as STOMP, T2VOC, and FEFLOW, allows for the precise prediction of contamination extents and the development of effective remediation strategies [6,8,10,16,17]. These strategies are essential for assessing monitored natural degradation as a sustainable site closure option [18,19].

The behavior of LNAPLs is governed by the nonlinear relationship between capillary pressure and relative permeability in a three-phase system (air–water–oil). A combination of experimental methods and multiphase flow model simulations is thus essential. Numerous fundamental studies have been conducted using one-dimensional column experiments and small-scale soil troughs. For example, vertical oil penetration and residual behavior near the capillary zone using transmitted light and simple image analysis were examined [20,21,22,23]. The results of these studies demonstrated that both saturation and soil structure can significantly impact oil spread patterns.

In recent years, studies have been published that focus on the dynamic effects of groundwater level fluctuations on oil distribution. For example, the formation of a smear zone and increased BTEX leaching through soil pan experiments involving elevated water levels were demonstrated [24]. Additionally, that periodic water level fluctuations encourage vertical oil diffusion and biodegradation was revealed. These findings suggest that an understanding based solely on static conditions is insufficient [23,25]. Fluctuations in groundwater levels significantly impact the thickness and accessibility of LNAPL at operational oil contamination sites. It cautioned that analyses ignoring such fluctuations may lead to underestimation of contamination distribution [26].

From a small-scale experimental perspective, the effects of walls and capillary forces are inevitably overestimated, which limits the ability to evaluate realistic oil spread and redistribution. In contrast, in medium- to large-scale soil pan experiments, oil migration and residual behavior under more realistic conditions were observed [20,27,28]. These experiments contribute to the calibration and limitation analysis of numerical models. In a study combining experiments and numerical analysis, it was demonstrated that horizontal oil diffusion and retention mechanisms vary significantly due to differences in soil layer structure and flow fields [29]. That study attempted to reproduce the observed oil distribution using multiphase flow numerical simulations and discussed the sensitivity and limitations of model parameters. It was demonstrated that groundwater flow promotes the lateral diffusion of DNAPL, highlighting the limitations of models based on static assumptions [29].

As shown in

Table 1. Previous studies on NAPL infiltration in unsaturated media.

Author	NAPL	Scale	Method and Observations	Model
Kechavarzi et al. (2005) [20]	Diesel	$2\mathrm{D} \mathrm{tank} (1.8 \mathrm{m} \mathrm{H} \times$$1.2 \mathrm{m} \mathrm{L} \times$ 0.08 m D)	Light transmission imaging was used to map saturations in homogeneous sand during a laboratory release experiment, revealing the deep vertical penetration and lateral spreading of LNAPL in the vadose zone. No fluctuation in the water table was observed.	No
Dobson et al. (2007) [24]	Gasoline	$2\mathrm{D} \mathrm{tank} (0.8 \mathrm{m} \mathrm{H} \times$$0.5 \mathrm{m} \mathrm{L} \times$ 0.03 m D)	Fluctuations caused NAPL and air to become trapped below the water table, expanding the smear zone by a factor of ~6.7× and increasing water throughput by a factor of ~18×. Under fluctuation, dissolution and biodegradation rates spiked (owing to increased oxygen ingress and nitrate loss), and total dissolved hydrocarbon mass output rose by 10–20×.	No
Zheng et al. (2015) [29]	PCE	2D tank	Laboratory infiltration of dense-NAPL (DNAPL) in heterogeneous sand under different horizontal groundwater flow rates. Lateral spreading was documented above low-permeability lenses, and down-gradient migration was enhanced by water flow. A multiphase flow simulation that replicated the DNAPL migration patterns and dissolution was coupled with the above, and it was appropriately calibrated.	Yes
Gupta et al. (2019) [28]	Toluene	2D tank	Two-dimensional sand tank experiments were conducted with a fluctuating water table. The migration and fate of toluene under dynamic groundwater levels were investigated. After multiple water table raise/lower cycles, a pronounced smear zone with residual LNAPL saturations formed. The relationship between the rate at which the water table is lowered and NAPL migration was emphasized.	No
Gupta and Yadav (2020) [27]	Toluene	3D tank	In three-dimensional laboratory experiments, the release of LNAPL under varying groundwater flow conditions was examined. We monitored the distribution and migration of LNAPL down-gradient under the influence of flow. The findings revealed that higher groundwater velocity increases the sweep of LNAPL downstream but also leaves more residual material trapped in the source zone. A 3D demonstration was provided, demonstrating that some LNAPL may bypass observation wells when carried by lateral groundwater movement.	No
Alazaiza et al. (2020) [23]	Light oil	Columns	One-dimensional column tests involving repeated sequences of water table fluctuations were monitored using a simplified image analysis method to track the volume of LNAPL. The results demonstrated that water table oscillations progressively redistribute LNAPL; each cycle traps some LNAPL as residuals in previously saturated zones. After multiple cycles, the volume of mobile LNAPL decreased, and a wider smear zone of residuals formed.	No
Teramoto and Chang (2017) [26]	Jet fuel	Field site	Field monitoring of an LNAPL-contaminated site over time was combined with a numerical simulation of BTEX transport under a fluctuating water table. Observations revealed an inverse correlation between the thickness of the LNAPL in the wells and the water table, suggesting smear zone entrapment. The model reproduced seasonal concentration trends and variations in free product thickness.	Yes
Yang et al. (2017) [25]	BTEX	Columns	Laboratory simulation of the effects of groundwater fluctuation on biodegradation. Small soil columns contaminated with BTEX were subjected to controlled drops and rises in the water table. The results showed that water table fluctuation significantly enhanced the aerobic biodegradation of BTEX due to improved oxygen delivery, thus resulting in faster contaminant mass loss than in static controls.	No

, major research findings have emerged from various perspectives, such as experimental scale, NAPL species, groundwater conditions, visualization techniques, and numerical model applications. However, each study has clear limitations, with many studies facing the following challenges: (1) insufficient characterization of the saturated distribution, (2) an inability to simulate continuous and controlled fluctuations in the groundwater level, and (3) numerical model assumptions (e.g., saturation curves and boundary conditions) that are inconsistent with the experimental data.

Against this backdrop, this study aims to elucidate the distribution and retention mechanisms of LNAPL under dynamic groundwater conditions. This will be accomplished by explicitly resolving spatially continuous saturation distributions and linking them to controlled groundwater rise processes through large-scale experimentation and numerical analysis. The objectives of this study are as follows:

• The goal is to visualize and quantify planar and vertical LNAPL saturation fields during infiltration and redistribution processes. This can be achieved by using a large-scale soil tank combined with multispectral image analysis rather than relying on plume geometry or bulk indicators alone.
• The goal is to experimentally isolate the effects of contrasting groundwater rise rates (rapid and slow) on the trapping of LNAPL and residual saturation beneath the groundwater table.
• The goal is to evaluate the applicability and limitations of a commonly used NAPL simulator by directly comparing experimentally measured and simulated LNAPL saturation fields under identical dynamic groundwater conditions.

Addressing these objectives provides evidence from experiments that can be used to resolve saturation and insight from models that can be used to understand rate-dependent LNAPL retention processes in urban subsurface environments. This evidence and insight can be used to support more reliable contamination assessment and remediation planning.

2. Theoretical Background

2.1. Image Analysis

2.1.1. Principle

In order to understand the distribution patterns of oil infiltrating the soil, an appropriate method must be selected to detect the oil content therein. In large-scale soil tank experiments, point-based evaluation using measurement sensors is limited in terms of both moisture and oil content; therefore, determining a method that evaluates the entire surface is desirable. Based on the methodology employed in previous studies [20,21,22], we chose an image analysis method that involves the use of a digital camera to measure the distribution of moisture and oil content.

This method exploits the difference in light reflectance between water and oil. From the captured images, we calculate the color density of the target object and record the change in color density before and after the event. By establishing the relationship between changes in color density and saturation in advance, changes in liquid saturation within the soil can be derived from the measured changes in color density. The reflectance of the experimental soil surface changes when liquid is present in the soil pores. The relationship between light transmittance and wavelength bands for distilled water and the A-grade heavy oils used in the test is shown in Figure 1.

As shown in Figure 1, water has a transmission rate close to 100% across nearly the entire wavelength range. In contrast, the A-grade heavy oil used in this study shows a decrease in transmission rate beginning at around 500 nm, transmitting no light in the 400 nm range. Therefore, when comparing water and oil at 400 nm, the amount of light reflected increases as the oil content increases relative to water, even when the same amount of light is shone on both. However, comparing them at 800 nm reveals a different result. The amount of light reflected from water and oil is nearly identical, and the total light intensity increases proportionally to the combined volume of water and oil. By measuring the change in reflectance accompanying changes in saturation levels of two liquids across two wavelength bands, we can evaluate changes in saturation levels of unknown liquids. Specifically, coloring the oil red makes the difference between water and oil more pronounced. In this study, we quantitatively evaluated water and oil by attaching a bandpass filter (Figure 2) in front of the camera that transmits light with wavelengths of 500 or 760 nm and capturing images.

2.1.2. Photography Environment

A digital SLR camera was used in this study. However, the color and low-pass filters attached to the camera body were removed and replaced with clear filters. Additionally, a bandpass filter that transmits light with wavelengths of 500 or 760 nm was attached to the front of the camera to photograph the experimental soil tank. The tank has a glass front that enables visualization of its contents. The photography environment is shown in Figure 3. The camera is fixed to enable front-facing photography of the experimental soil chamber. A light source is provided to clearly capture the presence of water and oil inside the chamber. Two types of light sources were prepared: a metal halide light source (Unifocus 575W, Toshiba Lighting & Technology Corporation, Yokosuka, Japan), which emits a bright, slightly bluish light, and a halogen light source (JQ Spotlight 2 kW, Toshiba Lighting & Technology Corporation), which emits a light with a stronger reddish hue. Two units of each light source were prepared and positioned to uniformly illuminate the subject. The illuminance characteristics of each light source are shown in Figure 4. The metal halide light source was incorporated to increase the light intensity at 500 nm and render the oil more discernible. Since the photography process must be conducted while blocking out unstable external light, it is essential to conduct the experiment under darkroom conditions.

2.1.3. Method for Quantifying Saturation

A detailed account of the theoretical development is provided in previous study [22]; in this paper, we therefore offer only an outline. Treating the water and oil saturations as unknown variables enables them to be quantified:

$$S_w={\displaystyle \frac{{\lambda }_{500}^o\Delta D_{760}-{\lambda }_{760}^o\Delta D_{500}}{{\lambda }_{500}^o{\lambda }_{760}^w-{\lambda }_{760}^o{\lambda }_{500}^w}}+S_{w,init}$$

(1)

$$S_o={\displaystyle \frac{{\lambda }_{760}^w\Delta D_{500}-{\lambda }_{500}^w\Delta D_{760}}{{\lambda }_{500}^o{\lambda }_{760}^w-{\lambda }_{760}^o{\lambda }_{500}^w}}+S_{o,init}$$

(2)

where ${\lambda }_{500}(\equiv \partial D_{500}/\partial S_w)$: sensitivity of 500 nm light to water saturation, ${\lambda }_{760}(\equiv \partial D_{760}/\partial S_w)$: sensitivity of 760 nm light to water saturation, ${\lambda }_{500}(\equiv \partial D_{500}/\partial S_o)$: sensitivity of 500 nm light to oil saturation, and ${\lambda }_{760}(\equiv \partial D_{760}/\partial S_o)$: sensitivity of 760 nm light to water saturation. Furthermore, $\Delta D_{500}=D_{500}-D_{500,init}$ and $\Delta D_{760}=D_{760}-D_{760,init}$. Therefore, by obtaining calibration coefficients under specified shooting conditions, changes in the subject’s luminance distribution due to water or oil movement during the soil tank experiment can be measured using a digital camera, thus enabling the calculation of $S_w$ and $S_o$, which are water and oil saturations, respectively.

This study employs an image analysis method that follows previously established approaches. This method assumes a linear relationship between optical density and liquid saturation under controlled illumination conditions. Although factors such as light scattering, surface roughness, and local oil redistribution can affect absolute saturation values, these effects are minimized through consistent calibration, homogeneous soil properties, and controlled lighting. Therefore, the image analysis results are interpreted in terms of relative saturation distributions and temporal evolution rather than exact absolute values.

2.2. Multiphase Flow Analysis in Porous Media

2.2.1. Governing Equations

Multiphase flow analysis in porous media has been extensively covered in published books and papers [30,31]. In this study, we therefore focus solely on describing the fundamental equations that must be solved numerically. The behavior of non-aqueous fluids in soil is represented by the following equations:

$$\varphi {\displaystyle \frac{\partial \left({\rho }_w{S}_w\right)}{\partial t}}+\nabla ·\left[-{\rho }_w{\displaystyle \frac{k_{rw}K}{{\mu }_w}}\left(\nabla p_w+{\rho }_{w}g\nabla z\right)\right]=0$$

(3)

$$\varphi {\displaystyle \frac{\partial \left({\rho }_o{S}_o\right)}{\partial t}}+\nabla ·\left[-{\rho }_o{\displaystyle \frac{k_{ro}K}{{\mu }_o}}\left(\nabla p_o+{\rho }_{o}g\nabla z\right)\right]=0$$

(4)

$$\varphi {\displaystyle \frac{\partial \left({\rho }_a{S}_a\right)}{\partial t}}+\nabla ·\left[-{\rho }_a{\displaystyle \frac{k_{ra}K}{{\mu }_a}}\left(\nabla p_a+{\rho }_{a}g\nabla z\right)\right]=0$$

(5)

where $S_w$: water saturation, $S_o$: oil saturation, $S_a$: air saturation, ${\rho }_w$: fluid density of freshwater (kg/m³), ${\rho }_o$: fluid density of oil (kg/m³), ${\rho }_a$: fluid density of air (kg/m³), ${\mu }_w$: viscosity coefficient of freshwater (Pa$·$s), ${\mu }_o$: viscosity coefficient of oil (Pa$·$s), ${\mu }_a$: viscosity coefficient of air (Pa$·$s), $\varphi$: porosity, $p_w$: water pressure (Pa), $p_o$: oil pressure (Pa), $p_a$: air pressure (Pa), $K$: intrinsic permeability (m²), $k_{rw}$: relative permeability coefficient ratio, $k_{ro}$: relative oil permeability coefficient ratio, $k_{ra}$: relative gas permeability coefficient ratio, and $g$: gravitational acceleration (=9.8 m/s²). However, each saturation is subject to the following constraint equation:

$$S_w+S_o+S_a=1$$

(6)

There are six unknown variables: $S_w$, $S_o$, $S_a$, $p_w$, $p_o$, and $p_a$. The constraint equations are given by Equations (3)–(6), and the SP curve is shown in Equations (7) and (8), which will be described in subsequent sections. The SK curve representing fluid permeability is given by Equations (9)–(11), which will be described in subsequent sections. The above ensures that there is an equal number of unknown variables and equations, which enables us to obtain a numerical solution under the given initial and boundary conditions.

In this study, the numerical solution was obtained using the NAPL simulator [32]. The NAPL simulator’s key features are its ability to account for the hysteresis of the SP curve and the residual saturation trapped within the pore space, both of which will be described hereafter. The NAPL simulator is a suitable code for simulating how the contamination zone expands due to groundwater fluctuations when oil is present above the groundwater table.

2.2.2. SP Curve

The SP curve is an equation that relates saturation to fluid pressure. For two-phase flow problems involving gas and liquid, such as those involving water and air, the van Genuchten equation [33] is widely used. For gas–liquid–liquid three-phase flow problems, the van Genuchten equation is extended as follows:

$$S_{ew}\equiv {\displaystyle \frac{S_w-S_{rw}}{1-S_{rw}}}={\left[1+{\left(\alpha {\beta }_{ow}{p}_{ow}\right)}^n\right]}^{-m}$$

(7)

$$S_{et}\equiv S_{ew}+S_{eo}={\displaystyle \frac{S_w+S_o-S_{rw}}{1-S_{rw}}}={\left[1+{\left(\alpha {\beta }_{ao}{p}_{ao}\right)}^n\right]}^{-m}$$

(8)

where $S_{ew}$: the effective water saturation, $S_{et}$: the effective liquid saturation, $S_{rw}$: the residual water saturation, $\alpha$: van Genuchten (VG) parameter (1/Pa), $n$: VG parameter, $m=1-1/n$, ${\sigma }_{aw}$: the surface tension between air and water (N/m), ${\sigma }_{ow}$: the surface tension between oil and water (N/m), and ${\sigma }_{ao}$: the surface tension between air and oil (N/m), where ${\beta }_{ow}$ = ${\sigma }_{aw}$/${\sigma }_{ow}$ and ${\beta }_{ao}$ = ${\sigma }_{aw}$/${\sigma }_{ao}$. Capillary pressures are $p_{ow}=p_o-p_{w }$ and $p_{ao}=p_a-p_o$. VG parameters control the shape of the soil–water retention curve. $\alpha$ represents the inverse of the air-entry pressure and $n$ describes the pore-size distribution. By conducting water retention tests on the target soil and identifying the VG parameters, the SP curve can be extended to a three-phase flow problem using the surface tensions and interfacial tensions.

2.2.3. SK Curve

The relative permeability coefficients in a three-phase system of water, oil, and air are given by the following equation as a function of the effective saturation of water and the effective saturation of all liquids [34,35].

$$k_{rw}=S_{ew}^{{\displaystyle \frac{1}{2}}}{\left[1-{\left(1-S_{ew}^{{\displaystyle \frac{1}{m}}}\right)}^m\right]}^2$$

(9)

$$k_{ro}={\left(S_{et}-S_{ew}\right)}^{{\displaystyle \frac{1}{2}}}{\left[{\left(1-S_{ew}^{{\displaystyle \frac{1}{m}}}\right)}^m-{\left(1-S_{et}^{{\displaystyle \frac{1}{m}}}\right)}^m\right]}^2$$

(10)

$$k_{ra}={\left(1-S_{et}\right)}^{{\displaystyle \frac{1}{2}}}{\left(1-S_{et}^{{\displaystyle \frac{1}{m}}}\right)}^{2m}$$

(11)

3. Methods

3.1. Camera Calibration

To establish a relationship between color intensity changes in digital images and changes in soil liquid saturation, the saturation of the soil was varied, and the corresponding changes in color intensity were evaluated. Using the obtained relationship as a calibration formula for changes in color intensity before and after the soil pan experiment enables saturation changes to be derived from color intensity changes.

Silica sand #5 was mixed with a predetermined amount of water or oil and packed into a container with an inner diameter of 60 mm and a height of 60 mm to achieve a dry density of 1500 kg/m³. Images were captured using bandpass filters at 500 nm or 760 nm (see Figure 5).

The water or oil content within the soil was determined using the following average optical density of arbitrary regions in the resulting digital images:

$$D={\displaystyle \frac{1}{n}}\sum _{j=1}^n\left[-{\log }_{10}\left({\displaystyle \frac{I_{r,j}}{I_{o,j}}}\right)\right]$$

(12)

where $D$: average optical density, $I_{r }$: intensity of the reflected light given by the individual pixel values, $I_{o }$: intensity of the light that would be reflected by an ideal white surface, and $n$: total number of pixels in the region. $I_{o }$ was measured for every pixel of the nominal white patch (MCPET, Furukawa Electric Co., Ltd., Tokyo, Japan) for each spectral band. $I_{o }$ was on average close to 255 at 8-bit resolution.

3.2. Large-Scale Soil Tank Experiment

3.2.1. Preparation of Experimental Soil Tank

The experimental soil tank possesses the following internal dimensions: 2.4 m width, 2.4 m height, and 0.6 m depth. The rear, side, and bottom surfaces of the soil layer comprise FRP panels. The front surface of the tank comprises 38 mm thick laminated glass to enable visualization of the internal conditions. However, since steel reinforcement frames are installed around the tank’s perimeter, visualization is limited to areas not covered by these frames. Consequently, several regions in the front-view images could not be analyzed using image-based saturation evaluation. These regions were defined as non-observable areas and masked prior to quantitative image analysis. Therefore, all image-derived saturation distributions and subsequent comparisons with numerical simulation results were restricted to observable regions only.

Silica sand #5 was selected as a homogeneous reference medium to isolate the fundamental mechanisms of LNAPL redistribution under dynamic groundwater fluctuations. Using a well-characterized, uniform sand minimized the influence of soil heterogeneity, enabling repeatable visualization and quantitative comparison with numerical simulations. Additionally, the same sand and packing conditions were used in the image-based saturation calibration tests to ensure internal consistency between the calibration and the tank-scale observations. The adopted dry density of 1500 kg/m³ represents medium-density sand, which is typical of permeable sandy backfill and engineered fill layers commonly found in urban subsurface environments, such as utility trenches and reclaimed grounds. This density provides stable packing and reproducible hydraulic behavior in a large-scale tank.

Although the present study used homogeneous sand to isolate fundamental redistribution mechanisms, it is recognized that natural urban subsurface soils often have layered, heterogeneous structures. In such settings, contrasts in pore size distribution and permeability between layers can induce capillary barrier effects. These effects lead to enhanced lateral spreading, preferential flow paths, and localized light nonaqueous phase liquid (LNAPL) accumulation above low-permeability strata. Thus, the sharp vertical redistribution and residual trapping mechanisms observed in this study are expected to be modified by soil heterogeneity, which could increase spatial variability and uncertainty in contamination distributions.

Soil (silica sand #5) was introduced into the tank using the underwater drop method. The soil tank was filled in ten layers, each achieving a dry density of 1500 kg/m³, using the following steps: (1) Fill the soil tank to a water level of approximately GL + 50 cm. (2) Place porous permeable partition boards (BPF-5, Daisen Membrane Systems Co., Ltd., Tokyo, Japan) on both sides of the tank and gently pack washed C30 crushed stone along their exterior. (3) Gently introduce a specified amount of silica sand #5 into the space inside the partition boards using the water drop method. (4) Compact the soil to the specified height using a portable vibrator to achieve the targeted density. After lightly loosening the surface of the prepared soil layer, proceed to prepare the next soil layer. That is, return to step (1). The completed experimental soil tank possessed a dry density range of 1450–1550 kg/m³, with an average dry density of 1500 kg/m³ and a standard deviation of 26 kg/m³. The physical conditions of the soil tank are shown in

Table 2. Experimental conditions for the soil tank test.

Parameter	Unit	Value
Soil type		Silica sand #5
Soil tank preparation method		Underwater drop method
Soil particle density	kg/m3	2620
Dry bulk density	kg/m3	1500
Porosity		0.44
Void ratio		0.79
Relative density	%	60

.

3.2.2. Experimental Procedure

The soil tank experiment for oil permeation consists of four processes: drainage, oil injection, rapid groundwater rise, and slow groundwater rise. The time schedule for the experiment is summarized in

Table 3. Time schedule for the soil tank experiment: The hatch area shows when each process was executed.

Date	November 18				November 19				November 20				November 21				November 22				November 23
O’clock	0	6	12	18	0	6	12	18	0	6	12	18	0	6	12	18	0	6	12	18	0	6	12	18
Drainage 1)
Oil injection 2)
Rapid GW rise 3)
Slow GW rise 4)
Date	November24				November25				November26				November27				November28				November29
O’clock	0	6	12	18	0	6	12	18	0	6	12	18	0	6	12	18	0	6	12	18	0	6	12	18
Drainage 1)
Oil injection 2)
Rapid GW rise 3)
Slow GW rise 4)

¹⁾ November 18, 01:00 start, groundwater level GL-0.00 m to GL-2.05 m. ²⁾ November 22, 18:00 start, heavy oil A injection 30 L, constant pressure 1.5 cm-oil. ³⁾ November 24, 00:00 start, groundwater level GL-2.05 m to GL-1.78 m, water level rise rate 5.5 cm/h. ⁴⁾ November 26, 18:00 start, groundwater level GL-1.78 m to GL-1.31 m, water level rise rate 1.0 cm/h.

.

• Drainage process in experiment

A soil tank prepared using the submerged drop method is shown in Figure 6. After capturing the initial images for image analysis (water saturation $S_{w,init}$ = 1 and oil saturation $S_{o,init}$ = 0), the drain valve at the bottom of the tank was opened. Water was drained from the fully saturated soil tank until the groundwater level reached GL-2.05 m (0.35 m above the tank’s bottom). Thereafter, the drain valve was closed. Once the moisture behavior within the tank reached a steady state, the soil was considered unsaturated for the subsequent oil injection process.

2.

Oil injection process in experiment

Next, during the oil injection process, 30 L of colored heavy oil A (with Oil-Red O, from Sigma-Aldrich) was continuously injected from an oil injection box set at ground level (GL-0 m). The oil was allowed to permeate into the ground while maintaining a constant injection pressure in order to produce a 15 mm thick oil layer within the injection tank. During this process, we observed the oil permeation phenomenon occurring within the unsaturated zone and verified the numerical analysis results.

3.

Groundwater rise process in experiment

Thirty hours after commencement of the oil injection process, the groundwater transitioned into the ascent phase. The ascent phase consists of two stages: the rapid ascent stage and the slow ascent stage. During the first stage, rapid ascent, the groundwater table rose from GL-2.05 m to GL-1.78 m at a constant rate of 5.5 cm/hr over five hours. The oil contamination distribution at the end of the oil injection process (see Figure 7) was used as the initial condition, resulting in a +0.27 m rise in the groundwater table during this stage. Water was supplied to the soil tank via a drain valve installed at the bottom of the tank. After setting the groundwater level at GL-1.78 m, the system was left undisturbed for 61 h (66 h after the start of the groundwater rise process).

In the second stage, the groundwater level was raised at a constant rate of 1.0 cm/h for 47 h, from GL-1.78 m to GL-1.31 m. The initial condition was the oil contamination distribution at the end of the first stage of the groundwater rise process (second stage groundwater rise height: +0.47 m). Thereafter, changes in the oil contamination distribution were monitored for approximately one month (see Figure 8). Using this process, we aimed to investigate how the oil contamination distribution changes as the groundwater level rises and disturbs the contamination pattern.

These two rise rates were selected to represent contrasting scenarios of groundwater fluctuation relevant to urban contaminated sites. The rapid rise rate (5.5 cm/h) corresponds to event-driven conditions, such as heavy rainfall, sudden recharge, or drainage malfunction. Under these conditions, buoyant oil migration cannot fully follow the rising water table. In contrast, the slower rise rate (1.0 cm/h) represents quasi-static increases in groundwater associated with seasonal variation or prolonged precipitation. This allows for gradual oil recovery while promoting residual trapping within pore spaces. The experiment was designed to contrast these two conditions and distinguish between transient, high-saturation entrapment during rapid groundwater rise and residual oil trapping under slower, near-equilibrium conditions.

3.3. Numerical Simulation

• Drainage process in simulation

Numerical analysis of the transition from draining the saturated experimental soil tank to preparing the unsaturated soil tank was omitted.

2.

Oil injection process in simulation

The analysis domain for the oil injection process is shown in Figure 9a. It measures 2.4 m in the $x$-direction and 2.4 m in the $z$-direction, which matches the size of the soil tank used in the oil permeation experiment. The oil injection point is designated as ($x$,$z$) = (1.07 m, 2.25 m). Oil is continuously injected at a rate of 2 L/min/m, which means the normalized rate per unit depth, for 25 min. After injection terminates, the permeation behavior of the oil within the soil column is simulated. In this calculation, the cumulative oil volume injected into the soil tank is 30 L (oil volume per 0.6 m depth), which is identical to the volume injected in the soil tank experiment. The boundary conditions are as follows: the sides of the tank are in non-draining conditions; the top is fixed at atmospheric pressure; and the bottom is fixed at a hydrostatic pressure of 3.43 kPa. These conditions imply a groundwater table at a height of 0.35 m from the bottom of the tank. The measured values of the moisture characteristic curve of the experimental soil tank are shown in Figure 9b. The parameters used in the numerical analysis are summarized in

Table 4. Analysis conditions.

Parameter	Unit	Value
Fluid density of freshwater, ${\rho }_w$	kg/m3	1000
Viscosity coefficient of freshwater, ${\mu }_w$	Pa$·$s	1.00 $\times$ 10−3
Fluid density of oil, ${\rho }_o$	kg/m3	861
Viscosity coefficient of oil, ${\mu }_o$	Pa$·$s	4.44 $\times$ 10−3
Fluid density of air, ${\rho }_a$	kg/m3	1.18
Viscosity coefficient of air, ${\mu }_a$	Pa$·$s	1.82 $\times$ 10−5
Surface tension between air and water, ${\sigma }_{aw}$	mN/m	72.8
Surface tension between air and oil, ${\sigma }_{ao}$	mN/m	28.2
Interfacial tension between oil and water, ${\sigma }_{ow}$	mN/m	44.6
VG parameter, $\alpha$	1/kPa	63.7
VG parameter, $n$		3.88
Residual water saturation, $S_{rw}$		0.10
Residual oil saturation, $S_{ro}$		0.02
Residual air saturation, $S_{ra}$		0.10
Saturated hydraulic conductivity, $k_{w,sat}$	m/s	2.00 $\times$ 10−3
Intrinsic permeability, $K$	m2	2.04 $\times$ 10−10
Porosity, $\varphi$		0.44

.

3.

Groundwater rise process in simulation

During the groundwater rise process, the calculation results from the end of the oil injection process are used as the initial conditions. The analysis space and boundary conditions during the groundwater rise process are shown in Figure 10a. The rise in the groundwater level is represented by gradually increasing the boundary pressure applied to the bottom of the soil tank. During the rapid groundwater rise process, the hydraulic pressure applied to the bottom of the soil tank increases as follows, as shown in Figure 10b: $t$ < 5 h = 0.54 kPa/h and 5 < $t$ < 66 h = 6.09 kPa. During the subsequent slow rise process, the hydraulic pressure increases at a rate of 0.10 kPa/h for 66 < $t$ < 113 h and then remains at 10.70 kPa for $t$ > 113 h.

4. Results

4.1. Camera Calibration Results

As shown in Figure 11, moisture content or oil content exhibits a linear relationship with luminance density in digital images. The slope of the linear form is summarized in

Table 5. Calibration coefficients obtained from image analysis.

Bandpass Filter Used	Water	Oil
500 nm	${\lambda }_{500}^w$ = 0.101	${\lambda }_{500}^o$ = 0.197
760 nm	${\lambda }_{760}^w$ = 0.027	${\lambda }_{760}^o$ = 0.129

. Here, $\alpha$: index representing water (=w) or oil (=o) and $i$: index representing wavelength 500 nm (=500) or 760 nm (=760). Compared to colorless water, colored oil exhibits a larger calibration coefficient, indicating higher analytical sensitivity for oil saturation.

4.2. Results of Oil Injection Process

The results of visualizing the behavior of oil permeation by photographing the soil tank with a digital camera equipped with a bandpass filter are shown in the left-hand column of

Table 6. Oil-contaminated distributions evaluated using image and numerical analyses: The semitransparent orange areas indicate regions that were masked due to the presence of steel frames and excluded from the quantitative analysis.

	Image Analysis Results	Numerical Analysis Results (NAPL Simulator)
Elapsed time = 0.25 h
Elapsed time = 1 h
Elapsed time = 12 h
Elapsed time = 24 h

. For the soil tank, the color map only represents areas where oil saturation distribution could be calculated via image analysis. The two horizontal and three vertical lines visible on the color map correspond to the steel frames installed at the front of the tank. These masked regions interrupt the continuous visualization locally, but they do not affect the overall interpretation of plume geometry or migration trends because the main plume axis and saturation gradients are fully captured within the observable area. Initially, oil infiltrating the soil spreads radially from the injection point as soon as injection begins. However, once the oil supply to the soil ceases, the plume tip accumulates high concentrations of oil. Over time, this plume tip progresses deeper into the soil, gradually decreasing in concentration (see Elapsed time = 0.25 or 1 h). This demonstrates oil’s characteristic of migrating while leaving small residual amounts within the soil voids of the permeable pathway. When the leading edge of the plume reached the highly saturated zone near the groundwater table, the concentration at the leading edge increased, and oil began accumulating on the surface of the groundwater (see Elapsed time = 12 h). This finding is thought to be due to oil remaining in the unsaturated zone that infiltrates vertically with a time delay. Subsequently, as oil accumulated above the groundwater table, the plume expanded horizontally. Ultimately, most of the injected oil was found floating above the groundwater table, though a small amount remained within the infiltration path (see Elapsed time = 24 h).

The results of the numerical analysis are shown in the right-hand column of Table 6. The temporal evolution of oil saturation within the soil column is depicted through comparison of the analysis results with the image analysis results or photographs. The contrast in the figure represents oil saturation, indicating the volume fraction of oil present in the soil pores. The analysis results show that, by the end of the oil contamination process, the oil reached the groundwater table and accumulated above it. However, in the simulation, the oil did not reach the groundwater table. Although the volume of oil injected into the analysis space was confirmed to be the same as in the experiment, significant differences appeared in the penetration velocity of the oil in the depth direction. Our experimental results (Elapsed time = 1–24 h) show that the wetted surface of the oil injected into the unsaturated zone possessed a sharp shape in the depth direction. In contrast, the analysis showed a broad shape, as shown in the numerical analysis results (Elapsed time = 1–24 h). The experimental results suggest that, within the soil tank, oil expansion was minimal perpendicular to the depth direction, with behavior dominated solely by the depth direction. However, the analysis showed significant expansion perpendicular to the depth direction. It is thought that the oil volume that should have infiltrated diffused laterally instead, delaying oil infiltration in the depth direction.

The oil saturation distribution in the depth direction from the oil injection point, derived from the image results, is shown in Figure 12a. The white plot represents the distribution after 1 h, the hatched plot represents the distribution after twelve hours, and the black plot represents the distribution after 24 h. Note that oil saturation data at 0.9 and 1.6 m are missing because image analysis could not be applied due to the influence of the steel frame installed in front of the soil tank. One hour after oil injection, the oil had penetrated approximately 1.2 m in the depth direction, with higher oil saturation toward the leading edge of the plume. Oil saturation at the plume tip is estimated to be 40–50%. Subsequently, the oil continued to permeate the soil over time. Twelve hours after injection, oil began accumulating in the capillary zone above the groundwater table. The oil saturation of the accumulated oil layer above the groundwater table was approximately 50%, with a thickness of roughly 20 cm. Furthermore, after 24 h, the volume of oil above the groundwater table had increased significantly, and the influence of the oil extended to the position adjacent to the groundwater table. Along the oil migration path, residual oil saturation of roughly 10% was observed. Figure 12b shows the water saturation distribution organized by depth before and after oil injection, which can be interpreted as the difference in water saturation between clean soil and oil-contaminated soil. Before oil injection, the silica sand #5 soil contained a capillary zone approximately 20 cm thick above the groundwater table with nearly 100% water saturation. After heavy oil permeated the soil, however, and sufficient time elapsed for the oil to accumulate above the groundwater table, the water saturation in the capillary zone decreased significantly. In one region, the water saturation dropped from nearly 100% to around 30%, which is believed to be due to the expulsion of water from the voids of the capillary zone as the oil accumulated above the groundwater table. It is recognized that the water saturation distribution near the groundwater table differs between clean and oil-contaminated soil.

4.3. Results of the Oil Groundwater Rise Process

The oil saturation distribution 13 h after the start of the groundwater level rise (eight hours after raising the level to GL-1.78 m) is shown in the left-hand column of

Table 7. Effects of groundwater fluctuations on oil-contaminated distributions: The semitransparent orange areas indicate regions that were masked due to the presence of steel frames and excluded from the quantitative analysis.

	Image Analysis Results	Numerical Analysis Results (NAPL Simulator)
Elapsed time = 13 h from GW rise stage
Elapsed time = 47 h from GW rise stage

, using the oil contamination distribution at the end of the injection process as the initial condition. The groundwater level increased at a constant rate of 5.5 cm/h from GL-2.05 m to GL-1.78 m, and it is evident that some oil remains below the groundwater table even after it is raised. The residual saturation of liquid trapped in soil pores is generally considered to be around 20% saturation at most. In this experiment, however, the oil saturation below the groundwater table reached 60–70% in high-concentration areas, far exceeding typical residual saturation levels. This finding is attributed to the rapid rise in the groundwater table, with which the upward movement of oil could not keep pace. It should be noted that the high oil saturations (60–70%) observed below the water table following a rapid rise in groundwater represent a transient, non-equilibrium state rather than an equilibrium residual saturation state. With sufficient stabilization time, the buoyant oil will gradually redistribute upward. The remaining saturation will then converge toward the residual values reported in the literature.

Next, the groundwater level was raised at a constant rate of 1.0 cm/h from GL-1.78 m to GL-1.31 m. The oil saturation distribution 47 h after the groundwater level began to rise again, when it reached GL-1.31 m, is also shown in the left-hand column of Table 7. When the groundwater level was raised slowly, oil with a saturation of approximately 10–20% remained below the groundwater level. The portion of high concentration (oil saturation of 60–70%) present below the groundwater level, when it was raised rapidly, disappeared. Rapidly raising the groundwater level can sometimes result in residual high-concentration oil below the water table. However, this state is temporary. Over time, the oil’s density causes it to rise above the groundwater level. Nevertheless, not all of the oil present below the groundwater level rises to the surface. Oil with a saturation level of approximately 10–20% becomes trapped within the soil pores. The final contamination distribution resulted in oil with a saturation level of 10–20% below the water table; in comparison, the high-concentration oil (with a saturation level of approximately 60%) floated in the capillary zone above the water table.

The results of simulations for rapid (0 < $t$ < 66 h) and slow (66 < $t$ < 133 h) rises in the groundwater level, using the distribution of oil saturation at the end of the oil injection process as the initial condition, are shown in the right-hand column of Table 7. Image analysis results clearly show the distribution of remaining oil in the groundwater; however, the simulation results show almost no oil distribution therein. This simulation model appears to underestimate residual oil retention under groundwater level fluctuations compared to reality. Furthermore, the oil floating on the groundwater surface spreads more widely in the simulation than in the experiment, thus indicating that the simulation model has difficulty evaluating the steep oil contamination distribution observed in the experiments.

4.4. Considerations

Both the large-scale soil tank experiment and the numerical simulations were designed under quasi-two-dimensional conditions. The tank was 0.6 m deep, and impermeable boundaries were applied in the out-of-plane direction to suppress lateral spreading in the depth direction in both approaches. Thus, the focus of the comparison between the experiments and simulations is on vertical and in-plane horizontal redistribution under identical boundary constraints rather than on three-dimensional spreading effects.

Table 8. Area of oil contamination determined by soil tank experiments and numerical analysis: The portion of the oil contamination area hidden by steel frames is not counted as part of the area in either the soil tank experiments or the numerical analysis.

		Contaminated Area (Pixel)		Ratio
Stage	Elapsed Time	Experiment	Simulation	=Sim./Exp.
Infiltration stage	0.25 h from Infiltration stage	2009	2290	1.14
	1.00 h	2967	8500	2.87
	12 h	3445	11,491	3.34
	24 h	3792	11,897	3.14
GW rapid rise stage	13 h from GW rise stage	5768	11,138	1.93
GW slow rise stage	47 h	6559	5565	0.85

compares the sizes of the contaminated areas to determine if numerical analysis can accurately assess the extent of the oil contamination observed in the soil tank experiment. The contamination concentration distributions in Table 6 and Table 7 were converted to grayscale, and Image J (developed by National Institutes of Health) was used to calculate the number of pixels occupying the contaminated area. However, since the observable contaminated area in the soil tank experiment is limited to the portion not covered by steel frames, the contaminated area obtained from the numerical analysis was assumed to be covered by equivalent steel frames as well. Pixels occupying the positions of the steel frames were excluded from the evaluation. The results show that, during the oil infiltration process, numerical analysis estimated a larger oil contamination area than the soil tank experiment. Conversely, once the groundwater rise process began, the numerical analysis indicated limited oil contamination redistribution, whereas the soil tank experiment showed extensive redistribution.

Simulations of the oil injection and groundwater rise processes revealed that, in both cases, the calculated distribution of oil contamination tended to expand laterally more than the effects observed in reality. Two factors were identified as responsible for this lateral expansion of oil in the analysis. First, the capillary pressure of silica sand #5 was calculated to be higher than necessary. Fluid intrusion into soil pores requires a certain pressure threshold, known as the intrusion pressure. Typically, liquid does not infiltrate soil pores unless liquid pressure exceeding this threshold is exerted on them. In our analysis, however, the relationship between liquid volume and pressure within soil pores was expressed as a continuous function, similar to the VG model. Consequently, liquid could intrude into soil pores with relatively low liquid pressures. When the liquid pressure acting on the soil exceeds the threshold, or intrusion value, the relationship between liquid pressure and volume should be described using a moisture characteristic curve. Conversely, if the liquid pressure is below the intrusion value, the liquid volume within the pores should be treated as zero. Brooks and Corey’s formula [36] can be used for this calculation; however, it presents disadvantages in that it complicates the process because the moisture characteristic curve cannot be described as a continuous function. At present, van Genuchten’s equation is predominantly used for multiphase flow problems.

The second issue concerns numerical methods for solving multiphase flow equations. As advection dominates in these problems, standard Eulerian methods (e.g., finite difference or finite element methods) tend to diverge. To suppress numerical divergence, the upwind method, which incorporates artificial diffusion, is typically employed. While this process improves convergence, it also broadens the sharp front surface of the fluid’s wetted boundary. The upwind method is used in general-purpose multiphase flow analysis codes, such as MOFAT [37] and the NAPL simulator [32]. However, the field is also increasing development toward multiphase flow analysis codes with superior adaptability. These codes use the Petrov–Galerkin method [30] or the high-resolution shock-capturing conservative method [38,39,40,41] instead of the upwind method. The authors of future studies should incorporate threshold-based capillary models and high-resolution numerical schemes to enable sharp front tracking, which is essential for reliable prediction and risk-based decision-making in urban remediation planning.

Additionally, the present study focused on single-rise groundwater scenarios with contrasting rates and did not address the repeated rise-fall cycles typical of urban subsurface environments. Previous column-scale studies have suggested that cyclic groundwater fluctuations can enhance residual light non-aqueous phase liquid (LNAPL) trapping over time. However, verification of such cumulative effects at the present experimental scale remains a topic for future investigation. Therefore, further large-scale experiments and numerical studies that incorporate cyclic and multi-rate groundwater fluctuations are required to assess the long-term persistence and evolution of trapped LNAPL.

This study focuses on a single representative LNAPL (A-grade heavy oil) and does not aim to investigate oil-specific properties parametrically. The primary objective is to elucidate the redistribution mechanisms of LNAPL under dynamic groundwater fluctuations and evaluate the limitations of conventional multiphase flow models in reproducing sharp saturation fronts rather than quantifying sensitivity to fluid properties. LNAPLs encompass various substances, including benzene, toluene, ethylbenzene, and xylene; however, their densities, viscosities, and interfacial tensions typically fall within a relatively narrow range compared to the dominant contrasts between the LNAPL, water, and air phases. Therefore, the qualitative mechanisms governing vertical trapping, smear zone formation, and residual saturation under fluctuating groundwater conditions are expected to be similar among common petroleum-derived LNAPLs. While quantitative differences may arise for specific oil types, this study’s findings are considered to capture fundamental, transferable processes relevant to a wide range of LNAPL-contaminated sites.

4.5. Implications for Site Remediation

The implications discussed below should be interpreted in light of the experimental scope of this study, which focused on single-event groundwater rises rather than long-term cyclic groundwater fluctuations.

The experimental and numerical findings of this study have several important implications for the remediation of urban sites contaminated by light non-aqueous phase liquids (LNAPLs). First, the results show that characterizing a site based on static groundwater conditions can substantially underestimate the extent of residual contamination. Experiments showed that fluctuations in the groundwater table lead to the formation of a persistent smear zone in which 10–20% LNAPL saturation remains trapped below the current groundwater level. This residual contamination is not transient and cannot be recovered by conventional free-product removal techniques. Therefore, remediation strategies that rely solely on current groundwater levels and apparent free-phase thickness in monitoring wells may fail to identify significant secondary contamination sources.

Second, the difference observed between rapid and slow rises in the groundwater level highlights the importance of groundwater dynamics in remediation planning. A rapid rise in the groundwater level temporarily traps high-saturation LNAPL (60–70%) below the water table; subsequent slow rises reduce this level to residual amounts. In urban areas, such rapid fluctuations can occur during heavy rainfall, flooding, or changes in subsurface drainage systems. These findings suggest that remediation designs should explicitly consider historical and potential future groundwater fluctuations instead of relying on snapshot observations.

Third, the persistence of residual LNAPL below fluctuating groundwater tables has direct implications for remediation method selection. Pump-and-treat systems and free-product recovery wells may be ineffective at removing this trapped oil, even when monitoring wells indicate minimal or no free product. In such cases, remediation approaches that focus on reducing mass flux, monitored natural attenuation, or in situ treatment technologies may be more appropriate than aggressive, recovery-based methods.

Furthermore, the results emphasize the need for improved monitoring network design in urban areas, where access constraints often limit the number and placement of boreholes. Monitoring wells should be designed with screen intervals that account for historical groundwater level ranges rather than the present water table alone.

Finally, discrepancies between experimental observations and numerical simulations suggest that commonly used multiphase flow models may underestimate residual LNAPL retention under fluctuating groundwater conditions. This has important implications for risk-based remediation planning because numerical models are often used to inform decisions and obtain regulatory approval. Therefore, incorporating threshold-based capillary models and high-resolution numerical schemes is essential for improving the reliability of remediation design and long-term environmental risk assessment in urban subsurface environments.

4.6. Limitations and Future Work

This study has several limitations that should be acknowledged. First, the experimental results are derived from a single, large-scale, long-duration soil tank experiment that was not directly replicated. Due to the substantial size of the apparatus (2.4 m $\times$ 2.4 m $\times$ 0.6 m), the lengthy experimental period, and the large quantity of materials used, it was not feasible to conduct multiple experiments under identical conditions.

Consequently, statistical reproducibility in the conventional sense could not be evaluated. However, this study’s objective was not to provide statistically averaged results but rather to elucidate the dominant physical mechanisms governing LNAPL redistribution under dynamic groundwater fluctuations. This study aimed to do so at a scale closer to real subsurface conditions than typical laboratory column tests. Additionally, this study investigated the accuracy of general simulation models in large-scale tests.

The observed trends, such as enhanced residual trapping under rapid groundwater rise and persistent low-level residual saturation under slow fluctuations, are consistent with established multiphase flow theory and previous smaller-scale studies. However, the amount of LNAPL trapped in porous media cannot be accurately predicted using the NAPL simulator, which considers residual saturation and SP curve hysteresis. The predicted amount was much lower than the measured amount.

Future work should include systematic replication under varying soil properties, oil types, and groundwater fluctuation patterns, as well as ensemble-based numerical simulations calibrated against the present experimental dataset. This will help to further evaluate the generality and uncertainty of the results. Addressing these uncertainties through systematic large-scale experiments and ensemble numerical simulations is a critical direction for future research aimed at supporting robust risk-based remediation frameworks.

5. Conclusions

In this study, we conducted oil permeation experiments using soil tanks that simulated actual ground conditions. We observed the oil permeation mechanism and contamination distribution within the ground and discussed the effectiveness of numerical simulation technology in evaluating oil contamination distribution. The following findings were obtained:

• When the leading edge of the plume contacts the highly saturated zone near the groundwater table, oil accumulates as if floating above the table. Water saturation in the capillary zone above the groundwater table decreased significantly with oil intrusion. When 30 L of oil was injected into a soil tank containing silica sand #5, the water saturation in the capillary zone penetrated by the oil was approximately 30%.
• In soils with a history of groundwater level fluctuations, oil diffuses into zones where the water level has fluctuated. When the current groundwater level is higher than past levels, approximately 10–20% of the oil remains as residual saturation below the current groundwater surface.

Furthermore, numerical simulations based on the soil tank experiments yielded the following findings:

• Simulations of the oil injection and groundwater rise processes showed that the calculated oil contamination distribution tended to expand laterally more than was observed in reality. The analysis tends to represent the oil contamination zone as a low-concentration, wide-area distribution.
• When subjected to groundwater level fluctuations, the oil contamination distribution in the experiment clearly showed the distribution of residual oil in the groundwater. In contrast, the analysis showed almost no oil distribution in the groundwater. This analysis model calculates the residual oil under groundwater level fluctuations to be less than the actual amount.
• To simulate the sharp shape of the oil contamination distribution demonstrated in the experiment, it is necessary to use the Brooks and Corey equation as the moisture characteristic curve model and adopt a new calculation method instead of using artificial diffusion as the numerical analysis method.

The findings offer critical implications for urban environmental governance, including the rational placement of monitoring wells, improved uncertainty evaluation in remediation design, and prevention of long-term pollution spreading beneath densely developed infrastructure.

Future work should expand the current methodology to include layered and heterogeneous soils, which are commonly found in urban subsurface environments. Incorporating soil heterogeneity into large-scale experiments and comparing the results to numerical simulations is essential for evaluating preferential flow paths and localized accumulation above low-permeability layers. This is also necessary to account for uncertainty in risk-based remediation design.