Research Hotspots and Trend Analysis in Modeling Groundwater Dense Nonaqueous Phase Liquids Contamination based on Bibliometrics

1. Introduction

Dense nonaqueous phase liquids (DNAPLs) refer to a class of organic contaminants that are denser than water and poorly soluble in water, existing in groundwater as a non-aqueous phase. Typical DNAPLs include chlorinated organic solvents [1], creosote [2], and polychlorinated biphenyls (PCBs) [3]. Once DNAPLs infiltrate the subsurface, they can persist for extended durations due to their resistance to biodegradation. During this time, they slowly dissolve into the flowing groundwater and form an extensive contamination plume downstream. Meanwhile, the dissolved DNAPLs gradually seep into the low-permeability layers and become a potential secondary pollution source after the depletion of the original source, thereby keeping the DNAPL concentrations in the plume above the maximum contaminant level (MCL) [4]. Given the toxicity of DNAPLs and their potential health risks to humans [5], the remediation of groundwater DNAPL contamination has become an urgent issue that needs to be addressed.

Various remediation technologies have been employed to remove DNAPL contamination, such as surfactant-enhanced aquifer remediation (SEAR) [6,7], in-situ chemical oxidation (ISCO) [8,9], and in-situ bioremediation [10]. While remediation technologies are the primary means of pollution control, modeling is also essential to ensure the effectiveness of remediation efforts due to the invisibility of DNAPL distribution. Therefore, modeling DNAPL contamination has consistently been a focal point of interest for scholars in the field of groundwater environmental studies. However, the research on DNAPL models varies in focus, and there is a lack of comprehensive statistical analysis of the research outcomes at a macro scale. Consequently, it is essential to systematically review the hotspots in the field of DNAPL models to accurately grasp the research directions and evolving trends.

Bibliometric analysis, as a method for systematically and quantitatively analyzing scientific literature, provides reliable data support and criteria for scientific research, making it an indispensable tool in the research field. Therefore, this study conducted a statistical analysis of nearly 30 years of literature related to DNAPL models, following a bibliometric process [11]. The study also performed a visual presentation of the research status and trends, aiming to provide insights and references for future study.

2. Methodology

2.1. Database Selection

The WoS Core Collection database was selected as the primary data source for this study. A total of 681 publications were retrieved over the 30-year period from January 1, 1993, to December 31, 2023, taking “DNAPL” and “model” as the search topics. After filtering for document type as “Article” and further screening and deduplication, 614 publications remained for research analysis.

2.2. Bibliometric Indicators and Tools Used

A general analysis of the publication volume of the 614 articles was first conducted, and the data was plotted using Origin software to gain an initial understanding of the overall trends in DNAPL modeling research. The publishing countries, institutions, journals, and keywords over the past 30 years (1993-2023) were then statistically analyzed and mapped using CiteSpace 6.3.R1 [12] as the primary tool for bibliometric analysis.

The CiteSpace parameters were set as follows. The time slice was set to 1 year. For the co-citation analysis of different countries, institutions, and journals, the node types were set to “Country”, “Institution”, and “Cited Journal”, respectively, with the k-value in the g-index parameter set to 25 for each. For the keyword clustering analysis, the node type was set to “Keyword”, with the g-index parameter k value set to 20. This adjustment was made to reduce the number of nodes in the map, preventing it from becoming overly complex and difficult to interpret.

3. Findings of Bibliometric Analysis

3.1. Analysis of The Publication Volume

The annual publication volume and its changes can reveal the development status of a research field and its current level of attention. From 1993 to 2023, the total number of publications related to DNAPL models accumulated each year (Figure 1). During the initial budding stage (1993-1999), the annual publication volume was generally below 10 papers. However, 1998 saw a notable increase, with 20 papers published. During the rapid development stage (2000-2010), the annual publication volume gradually increased and stabilized at 20-30 papers per year from 2002 onward, indicating sustained academic interest. During the continuous breakthrough stage (2011-2023), the annual publication volume fluctuated significantly. Publication numbers were the lowest in 2011 and 2016, with only 12 papers, while 2023 saw a breakthrough with 40 papers published. From the above, it is clear that there is still room for further exploration in the research of DNAPL models.

3.2. Analysis of The Leading Countries

The national collaboration network reveals the importance of individual countries and the interconnections between multiple countries in a specific research field. The co-citation network of collaborative countries (Figure 2), based on WoS Core Collection data, revealed that 48 countries have published research on groundwater DNAPL contamination modeling. The countries with 10 or more publications are listed in Table 1.

In terms of publication volume, the United States leads the field of groundwater DNAPL contamination modeling with 307 publications, followed by Canada with 112. This is related to the fact that most DNAPL-contaminated sites are located in the United States and Canada [13]. China ranks third with 82 publications but started later than the United States and Canada, with relevant research beginning in 2003. Among the top ten countries by publication volume, the United States, Canada, Germany, and the Netherlands started research early, while most of the other countries began their work after 2000.

Centrality is an indicator that reflects the importance of a country within a collaboration network. A centrality greater than 0.1 indicates that the country plays a crucial role in advancing the research field. In terms of centrality, the United States has a centrality of 0.95 in the national collaboration network, significantly higher than any other country. England follows with a centrality of 0.23. Other countries with a centrality greater than 0.1 include China (0.18), France (0.16), Turkey (0.16), Greece (0.15), and Scotland (0.11). The above countries with high centrality are marked with a purple outer ring in Figure 2. Although Canada has a high publication volume, its influence is relatively weaker, with a centrality of 0.09.

Overall, the United States leads in DNAPL contamination modeling. This is due to its advanced chemical industry, which results in more contaminated sites and draws significant scholarly attention, along with its strong economy providing substantial support for related research. Therefore, referencing American scholars’ research can offer valuable guidance for mastering and applying DNAPL models.

3.3. Analysis of The Leading Institutions

A total of 365 institutions have published research related to DNAPL models, according to the analysis of selected publications. The top 25 institutions with more than 10 publications were selected to create a co-citation network map (Figure 3) for visual analysis.

In terms of timeline and publication volume, the United States Department of Energy (DOE), Queen’s University, and the University of Waterloo are the top three institutions with earlier research and higher publication numbers, with 38, 29, and 27 papers, respectively. Moreover, they continue to produce relevant research in recent years. China and France started later but have recently focused more on DNAPL model research, achieving notable results. In DNAPL contamination modeling, China is represented by Nanjing University and Jilin University, with 24 and 22 publications, respectively.

Spatially, it is clear that most research institutions with higher output are located in the United States, followed by Canada, which aligns with the earlier analysis of publication by country. In contrast, only Nanjing University and Jilin University stand out in DNAPL model research in China, indicating that DNAPL models are not yet widely adopted. Additionally, there is significant collaboration between U.S. institutions, but international collaboration remains noticeably limited. Therefore, future research should focus on two key areas: (1) increasing global awareness of groundwater DNAPL contamination and promoting the application of DNAPL models; (2) enhancing international collaboration to leverage research experience and drive innovation in DNAPL modeling technologies.

3.4. Analysis of The Dominant Journals

The top 10 journals ranked by citation counts in DNAPL model research from 1993 to 2023 are listed in Table 2. The top three journals by citation count are Journal of Contaminant Hydrology, Water Resources Research, and Environmental Science & Technology, which also lead in publication volume in this field. The publication journals indicate that DNAPL model research can be divided into two categories. One focuses on areas like “environment”, “chemistry”, and “materials”, with representative journals such as Environmental Science & Technology, Journal of Hazardous Materials, and Chemosphere, emphasizing the exploration of DNAPL reaction mechanisms and generally having higher impact factors. The other category leans toward “groundwater”, “hydrogeology”, and “pollution distribution and remediation”, represented by journals like Journal of Contaminant Hydrology, Water Resources Research, and Groundwater, focusing on the simulation and prediction of DNAPL pollution evolution. In the top ten journals, most publications are related to the latter category, showing that DNAPL contamination modeling with a focus on hydrogeology is the primary area of research.

Notably, most journals publishing DNAPL model literature have relatively low impact factors, likely due to the specialized and independent nature of the research. As shown in Figure 4, DNAPL model research is mainly concentrated in fields such as “Ecology, Earth, Marine”, “Physics, Materials, Chemistry”, and “Mathematics, Systems, Mathematical”, with limited interdisciplinary integration, resulting in restricted attention and impact.

3.5. Analysis of Research Hotspots and Trends Based on Keyword Clustering

3.5.1 Research Hotspots in modeling DNAPL contamination

The DNAPL model framework can be broadly divided into three parts: multiphase flow model, mass transfer model, and dissolved phase transport model. The model can be integrated with experimental findings to clarify DNAPL migration in groundwater and predict contamination distribution, aiding site remediation efforts. To better understand the research status in DNAPL contamination modeling, a keyword co-occurrence analysis was conducted, followed by clustering using the log-likelihood ratio (LLR) algorithm to reveal current research hotspots. The keyword clustering yielded a Q value of 0.4169 and an S value of 0.7499 (Q > 0.3 indicates significant structure; S ≥ 0.7 indicates full reliability). The 10 clusters related to DNAPL models (Figure 5) highlight five prominent keywords each and exhibit interconnections between clusters.

Cluster 1 (#0 multiphase flow): The multiphase flow model describes DNAPL multiphase migration and is the foundation of DNAPL modeling research. Due to their high density and low viscosity, DNAPLs infiltrate deep into the subsurface, pass through the unsaturated zone, contaminate aquifers, and accumulate on the aquitard, forming pools [14]. In this process, DNAPLs remain as a NAPL phase and slowly dissolve into the groundwater, becoming a source of dissolved DNAPLs. Many researchers use multiphase flow models to simulate this process and determine DNAPL distribution in the source zone. Research indicates that the key factors to consider when setting up multiphase models include pollutant release situations [15,16,17], aquifer heterogeneity [18,19,20], constitutive relations representing the permeability-saturation-capillary pressure ($K_{r,N}-S_w-P_c$) correlation [21,22,23], and groundwater flow velocity [14,24].

Cluster 2 (#1 reductive dichlorination): On one hand, researchers focus on chlorinated organic solvents to enhance reductive dechlorination techniques. F. Fagerlund et al. [25]used experiments and modeling to study the coupled process of PCE dissolution and dechlorination by nanoscale zero-valent iron in DNAPL source zones. On the other hand, given the common occurrence of PCE and TCE contamination sites, most studies focus on simulating and predicting these pollutants [26,27,28].

Cluster 3 (#2 mass transfer): Understanding the mass transfer mechanism of DNAPLs from the NAPL to the dissolved phase and establishing an appropriate expression is crucial for source-sink terms in plume modeling. A typical empirical rate-limited expression based on dissolution kinetics is:

$$J=\overline{k}{a}_{nw}(C_s-C)$$

(1)

where $J$ is the mass flux of dissolution from the NAPL phase to the aqueous phase, [ML⁻³T⁻¹]; $\overline{k}$ is the average mass transfer coefficient at the NAPL-water interface, [LT⁻¹]; $a_{nw}$ is the effective specific interfacial area between the NAPL phase and the aqueous phase, [L⁻¹]; $C_s$ is the equilibrium aqueous phase concentration, also known as the effective solubility, [ML⁻³]; and $C$ is the aqueous phase concentration, [ML⁻³].

Numerous studies have focused on optimizing mass transfer coefficients to improve the analysis and mathematical representation of the DNAPL dissolution process [29,30,31,32]. These coefficients are then incorporated into solute transport models to achieve accurate estimates of contaminant concentrations near source zones. To enable site-scale simulations, Parker and Park [33] developed an empirical expression for effective mass transfer coefficients under pseudo-steady-state conditions, providing a valuable reference for subsequent research [34,35,36,37,38].

Cluster 4 (#3 dense non-aqueous phase liquids): DNAPLs naturally form a cluster as a research focus. However, it is noteworthy that keywords such as “tomography”, “spectral induced polarization”, and “conductivity” appear under this cluster. This highlights that coupling geophysical multi-source data for DNAPL contamination modeling has become a key area of interest for scholars. Power et al. [39] developed a DNAPL-ERT numerical model by integrating Electrical Resistivity Tomography (ERT). This model calculates the resistivity response to key hydrogeological parameters (hydraulic permeability, porosity, clay content, groundwater salinity and temperature, and air, water, and DNAPL contents evolving with time), which enhances the sensitivity to heterogeneity in DNAPL distribution and soil structure. Kang et al. [40,41,42] coupled geophysics with DNAPL models and integrated them into various inversion frameworks, which improved source zone characterization.

Cluster 5 (#4 partial mass depletion): Mass depletion refers to the gradual dissolution and eventual depletion of the NAPL phase in the source zone over time. Similar to mass transfer, this cluster describes the conversion of the NAPL phase to the dissolved phase. However, the difference is that the mass transfer model links the multiphase flow model with the dissolved phase transport model, whereas the source strength function, which characterizes NAPL mass depletion, acts as a source term in the dissolved phase transport model. This simplifies the dissolution process in the source zone and reduces model complexity. The source strength function is typically a power-law relationship between the effluent concentration and the remaining DNAPL mass. The source strength function proposed by Falta et al. [43]has been widely adopted:

$${\displaystyle \frac{C_s\left(\mathrm{t}\right)}{C_0}}={\left({\displaystyle \frac{M\left(t\right)}{M_0}}\right)}^{\mathsf{\Gamma }}$$

(2)

where $C_s\left(\mathrm{t}\right)$ and $M\left(t\right)$ correspond to the DNAPL concentration in the source zone and the residual DNAPL mass at time t, respectively; $C_0$ and $M_0$ are the DNAPL concentrations in the source zone and the residual DNAPL mass at the initial time; and $\mathsf{\Gamma }$ is a model parameter.

Cluster 6 (#5 source zone): This cluster focuses on the inversion and identification of DNAPL source zones in saturated aquifers. China has conducted extensive research in this area, with most efforts focused on improving the accuracy of DNAPL source zone inversion. For example, Kang [40] proposed a joint inversion framework (CVAE-ESMDA) combining a convolutional variational autoencoder (CVAE) with the ensemble smoother with multiple data assimilation (ESMDA). This approach integrates multiple data sources (OHT, downstream DNAPL concentrations, and ERT) to more accurately estimate DNAPL saturation in the source zone. Wang et al. [44] combined the ensemble Kalman filter with an improved butterfly optimization algorithm, improving inversion accuracy and effectiveness.

Cluster 7 (#6 uncertainty analysis): Inversion of DNAPL source or optimization of remediation strategies based on simulation-optimization methods often involves uncertainty, requiring repeated model runs and high computational costs. Therefore, many studies develop surrogate models to reduce computational load and conduct uncertainty analysis. Hou et al. [45] developed an integrated surrogate model based on support vector regression (SVR), kriging, and kernel extreme learning machine (KELM). The homotopy-differential evolution (DE) algorithm was then combined with the surrogate for source inversion and uncertainty analysis, significantly improving identification accuracy. Du et al. [46] developed a fast-running convolutional neural network (CNN) surrogate model to identify the optimal SEAR scheme under uncertainty, improving optimization speed by 99.8% in 3D numerical experiments.

Cluster 8 (#7 immiscible displacement): In DNAPL contamination scenarios, immiscible displacement describes the relative movement between the NAPL and water phases at a small-scale heterogeneous pore level, driven by differences in gravity and viscosity, leading to NAPL displacing water and migrating downward. At the macro level, this corresponds to the multiphase flow described in Cluster 1. However, while multiphase flow simulation based on continuous models can statistically characterize heterogeneity at the macro level, it is difficult to capture the displacement behavior between the NAPL and groundwater phases at the pore scale. Therefore, some researchers have developed models specifically for immiscible displacement at the pore scale. Trantham et al. [47] developed a Stochastic Aggregation Model (SAM) using an improved DLA algorithm to simulate the displacement of groundwater by DNAPLs with both higher and lower viscosities than groundwater. Nsir et al. [48] developed a numerical simulator based on a discrete network model, using pore body and throat size parameters from the particle size distribution of real porous media. The simulated NAPL-water immiscible two-phase flow results matched experimental data.

Cluster 9 (#8 back diffusion): Back diffusion is the process where dissolved DNAPLs migrate into low-permeability zones, accumulate, and then diffuse back into the aquifer after a concentration reversal [49,50,51]. After source zone DNAPLs are depleted or isolated, back diffusion can occur once the contaminant concentration in the aquifer drops to a certain level, becoming a secondary pollution source and keeping plume concentrations above the MCL over time [4,52]. In recent years, back diffusion has gained attention due to its role in prolonging contamination persistence. Most studies simulate and explore factors influencing its occurrence, such as DNAPL solubility [53,54], soil heterogeneity [49], adsorption-desorption [55,56], and biodegradation [50,57,58,59,60]. Simulations of back diffusion are typically divided into two stages, marked by the removal or isolation of the source zone [49,52,61]. In the first stage, contaminants accumulate in low-permeability zones through forward diffusion. In the second stage, after source removal, the simulation continues to study back diffusion by observing plume tailing.

Cluster 10 (#9 contaminant mass discharge): The study of contaminant mass discharge is often closely linked to the mass transfer models in Cluster 3. Considering the challenges and costs of simulating DNAPL dissolution at the field scale, researchers have developed upscaled mass transfer models with domain-averaged coefficients to approximate real-site dissolution processes [34]. Simplified upscaled models, linked to mass discharge flux, can serve as an effective screening tool for evaluating source zone management strategies [36].

In summary, by reviewing and analyzing the current state of research, the distribution of key research hotspots is presented (Figure 6).

3.5.2 Assessment of Future Research Trends

The top 20 keywords with the highest burst strength from 1993 to 2023 are identified (Table 3). Based on the analysis of publication volume, the 30-year period can be divided into three stages: the initial budding stage (1993-1999), the rapid development stage (2000-2010), and the continuous breakthrough stage (2011-2023). The main keywords during the initial budding stage were “two-phase flow”, “multiphase flow”, and “contaminant transport”, with research focusing on the simulation of DNAPL multiphase migration. During the rapid development stage, more detailed descriptions were considered, including shifts in contaminant types (from TCE to PCE), applications in heterogeneous sites, deeper exploration of mass transfer processes, and increased emphasis on NAPL depletion in the source zone. In the continuous breakthrough stage, the focus shifted toward contaminant removal and remediation, the impact of permeability, back diffusion in low-permeability zones, and the integration of geophysical techniques for source zone identification.

From a research trend perspective, DNAPL modeling has developed into a well-established system, with further studies primarily focused on optimizing model details based on experimental findings. Future research directions include: (1) investigating back diffusion mechanisms and exploring methods to reduce plume persistence; (2) applying models to real sites for source zone characterization and pollution distribution; (3) optimizing remediation strategies to enhance effectiveness and reduce costs.

4. Summary and Outlook

This study conducted a bibliometric analysis using CiteSpace on 614 DNAPL modeling-related publications from the WoS core database (1993-2023). The findings are as follows:

(1) DNAPL models remain a focus of scholarly attention, with research outputs continuing to grow steadily. The United States is leading the international research on DNAPL models, followed by China and Canada. However, research priorities vary across countries, and international collaboration and exchange need to be strengthened.

(2) The core of DNAPL model research focuses on the simulation of DNAPL migration, transformation, and pollution distribution. In terms of published journals, the field is highly specialized with limited broader impact. To raise awareness and increase research impact, developing a health risk assessment model based on DNAPL contamination and strengthening cross-disciplinary connections could be beneficial.

(3) Based on keyword clustering analysis, the key research hotspots related to DNAPL models focus on multiphase flow models, mass transfer models, back diffusion, and practical applications of the models.

(4) Based on keyword burst analysis, the research trends in DNAPL modeling are centered on back diffusion mechanisms, characterization of contamination source zones and prediction of contaminant distribution in real-world sites, as well as the optimization of remediation strategies.