Hydrogeochemical Characteristics and Formation Mechanisms of High-Arsenic Groundwater in the North China Plain: Insights from Hydrogeochemical Analysis and Unsupervised Machine Learning

Abstract

Hydrochemical data were utilized in this study to elucidate the hydrogeochemical characteristics and genesis of high-arsenic groundwater in the North China Plain, employing both traditional hydrogeochemical analysis and unsupervised machine learning techniques. The findings indicate that the predominant hydrochemical types of groundwater in the study area are HCO₃-Ca·Na and SO₄·Cl-Na·Ca. The primary mechanism influencing groundwater chemistry has been identified as rock weathering. The unsupervised machine learning framework incorporates various methods, such as principal component analysis (PCA), non-negative matrix factorization (NMF), machine learning models (gradient boosting trees and random forests), and cluster analysis to explore the characteristics and genesis of groundwater hydrochemical types within the study area. This study demonstrated that the formation mechanism of high-arsenic groundwater results from multiple interacting factors.

1. Introduction

Groundwater arsenic contamination represents a significant environmental challenge worldwide [1]. This issue is particularly pronounced in East Asia. For instance, rural areas surrounding Hanoi, Vietnam, exhibit groundwater arsenic concentrations averaging 430 µg/L in drinking water [2]. In the southwestern region of Bangladesh, located within the Ganges Delta Plain, groundwater arsenic concentrations vary from 43.1 to 1352 µg/L, substantially exceeding the World Health Organization (WHO) guidelines for drinking water (10 µg/L) [3]. In China, the central Datong Basin has recorded groundwater arsenic concentrations reaching up to 366 µg/L [4].

Consumption of high-arsenic groundwater presents considerable health risks to humans, while utilizing such water for crop irrigation can result in arsenic accumulation in agricultural produce, leading to a series of subsequent pollution hazards. Consequently, identifying the factors that influence arsenic concentrations in high-arsenic groundwater and understanding the mechanisms that govern its release are urgent and essential tasks.

The study area is located in Xinxiang City, Henan Province, China, which falls within the arid and semi-arid regions of Northern China. Numerous previous studies have indicated that closed inland basins and aquifer environments under strongly reducing conditions in arid and semi-arid regions are conducive to the accumulation of arsenic in groundwater [5,6,7]. Additionally, the extremely high arsenic concentrations in this area are influenced directly or indirectly by groundwater overextraction and paleoriver channels. Previous researchers, such as Bian Jianmin et al. [8], reported that the spatial distribution of high-arsenic groundwater in Tongyu County shows arsenic enrichment in the low-lying central and northern areas along the horizontal plane, with relatively high arsenic concentrations in aquifers at depths ranging from 30 to 50 m. The total arsenic content is highest in Cl-HCO₃-Na-type water.

As research into groundwater contamination mechanisms has progressed, subsequent studies have concentrated on the enrichment mechanisms and genesis of high-arsenic groundwater. For instance, Zhang Di [9] identified the reductive dissolution of iron oxides as the primary cause of elevated arsenic levels in groundwater in Tongyu County. In contrast, sulfide precipitation was found to inhibit arsenic mobility and enrichment in deeper groundwater, while human agricultural activities emerged as the principal factor contributing to arsenic enrichment in groundwater.

This study aims to analyze the spatial distribution and characteristics of high-arsenic groundwater in the study area presented within this paper through field sampling and testing. By integrating these findings with those from prior research, further investigation into the formation mechanisms and influencing factors of arsenic in the region is pursued. This research offers scientific evidence to enhance the understanding and management of high-arsenic groundwater contamination in similar areas [10].

2. Study Area

Overview of the Study Area

The study area is located in Xinxiang City, Henan Province, within the North China Plain, covering a total area of 4847.8 km². To the west, the region is bordered by the Taihang Mountains, while the Yellow River forms the boundary to the east and south, with Anyang City situated to the north. The terrain generally slopes from west to east, with elevations ranging from 60 to 95 m. The region experiences a warm temperate, semi-humid, and semi-arid continental monsoon climate. Historically, the Yellow River has breached its banks and changed course multiple times in northern Henan, complicating the stratigraphy and depositional environment in areas of the paleoriver channel. This has led to the formation of distinct geomorphological features such as alluvial fans, floodplains, and Yellow River breakout fans across the surface [11] (Figure 1a).

The study area lies within the downstream segment of the paleo-Yellow River, between Wuzhi’s and Neihuang’s groundwater systems. The aquifers in the region primarily consist of loose rock pore aquifers, with lithologies that include loose fine sand, medium-coarse sand, gravel, loess-like silty sand, thin layers of sand, and silty clay with gravel lenses. This configuration results in a pronounced interbedded structure of sand and clay. Furthermore, the region is characterized by relatively low hydraulic conductivity, which contributes to the development of anaerobic or hypoxic conditions, thus affecting the formation of high-arsenic groundwater in the area.

The depth to the aquifer base typically ranges from 60 to 110 m, with some areas near the Yellow River exhibiting highly productive aquifers, where individual wells can yield between 3000 and 5000 m³/day (Figure 1b). Groundwater recharge in the system occurs through atmospheric precipitation, river infiltration, and irrigation return flow. Groundwater extraction, primarily driven by agricultural irrigation and industrial water use, has become the main means of groundwater discharge [12].

3. Methods and Materials

3.1. Sample Collection and Analysis

The hydrochemical data in this study were obtained from field-sampled spring water analyses and collected from the literature. In July 2023, 176 spring water samples were collected from the study area, all of which are located within the Xinxiang city region of Henan Province, China. Sample bottles were prewashed with distilled water and rinsed three times with well water from each target sampling site before collection. On site pH values were measured, and the samples were sealed, labeled, stored at 4 °C, and transported to the laboratory for water quality testing. The analysis was performed in accordance with the “Standards for Drinking Natural Mineral Water” (GB 8537-2018, National Food Safety Standards) and “Standards for Maximum Levels of Contaminants in Food” (GB 2762-2017, National Food Safety Standards).

A Shimadzu AA-6000CF atomic absorption spectrophotometer(Shimadzu Corporation, Kyoto, Japan ) and a Dionex ICS-2100 ion chromatograph (Thermo Fisher Scientific, Waltham, MA, USA) were utilized to measure sodium, potassium, calcium, magnesium, bicarbonate, chloride, sulfate, iron, and hydroxide ions. A spectrophotometer was employed to determine the contents of nitrate (NO₃⁻), nitrite (NO₂⁻), and ammonium nitrogen (NH₄⁺), while temperature, odor, pH, and other physical characteristics were assessed in the field.

3.2. Data Processing and Analysis

3.2.1. Traditional Hydrogeochemical Techniques

To analyze the hydrochemical characteristics, Origin 2021 and Python(3.11) were utilized to create Piper trilinear diagrams, Pearson correlation plots, and Gibbs diagrams. These tools qualitatively describe the relationships among various components in water samples and their environmental behaviors, aiding in the identification of water–rock interactions and other processes affecting water chemistry. This methodology was employed to evaluate the water type and ascertain redox processes in the groundwater.

3.2.2. Unsupervised Machine Learning

The unsupervised machine learning module in this study processed the data using Python with applied principal component analysis (PCA) and K-means clustering techniques for the validation and analysis. PCA was utilized to identify and quantify the most significant factors influencing groundwater chemistry, extracting the primary influencing factors and sources of variation. The K-means clustering method was subsequently employed to categorize groundwater data based on the optimal number of clusters indicated by specific metrics [13].

To accurately represent the hydrochemical properties of the groundwater in the study area, 12 variables (pH, Na⁺, K⁺, Ca²⁺, Mg²⁺, Fe²⁺, HCO₃⁻, SO₄²⁻, Cl⁻, NO₃⁻, NH₄⁺, and TDS) were utilized as the input dataset. Following data processing, the self-organizing map (SOM) technique was applied for initial classification, followed by PCA and K-means clustering for additional analysis. Through 10,000 iterations of PCA and K-means clustering, stability was achieved by the random forest and gradient boosting tree models, allowing for the validation of the raw hydrochemical data and the analysis of mechanisms governing hydrochemical formation.

4. Results and Discussion

4.1. Spatial Distribution of High-Arsenic Groundwater

The statistical results indicate that the arsenic (As) concentration in the 176 groundwater samples collected from the study area ranged from 0.01 to 190 μg/L, with an average concentration of 24.88 μg/L (

Table 1. Statistics of groundwater quality parameters.

		As	PH	Ca2+	Mg2+	K+	Na+	Cl−	SO42−	HCO3−	NO3−	NO2−	NH4+	Fe2+	F−	TDS
As < 10	Max	9.8	8.49	351.98	450.97	666.45	1193.9	2120.6	3435.9	946.697	239.35	0.49	2.7	7.6	3.29	7825
	Min	0.01	6.85	7.76	6.17	0.01	7.29	6.35	5.57	265.74	0.05	0	0.02	0.03	0.12	271.29
	Mean	2.86	7.43	98.02	73.38	11.27	168.11	137.89	274.99	541.75	11.15	0.03	0.28	0.95	0.81	1098.37
As ≥ 10	Max	190	7.9	160.48	151.84	10.13	375.79	674.65	320.55	280.02	17.57	0.21	2.82	9.25	1.82	413.94
	Min	10	6.93	24.85	22.17	0.53	42.85	14.11	21.76	771.6	0.045	0	0	0.07	0.34	1990.4
	Mean	35.13	7.48	86.41	56.14	2.73	112.53	99.6	116.57	527.89	0.95	0.02	0.5	2.2	0.62	767.47

and Figure 2). Among these samples, 61 (34.65%) exhibited As concentrations surpassing the World Health Organization’s guideline of 10 μg/L. The data reveal that high-arsenic groundwater (As > 10 μg/L) is generally weakly alkaline, with pH values ranging from 6.93 to 7.9 and an average of 7.748. This suggests that strongly reducing conditions are more conducive to arsenic enrichment. In high-arsenic groundwater, the average concentration of NO₃⁻ was 0.95 mg/L, which was lower than that in low-arsenic groundwater. Additionally, the average concentration of Fe²⁺ in the high-arsenic groundwater was 2.2 mg/L, whereas it was 0.95 mg/L in the low-arsenic samples. The average NH₄⁺ concentration was 0.5 mg/L in high-arsenic groundwater, whereas it was only 0.28 mg/L in low-arsenic groundwater [14,15,16,17,18,19].

For the water chemistry indicators, pH is dimensionless, As is measured in μg·L⁻¹, and the rest are measured in mg·L⁻¹ [20,21].

From a spatial distribution perspective (Figure 3), high-arsenic groundwater is predominantly located in the central interfluvial zone and paleoriver channels of the study area (

Table 2. Proportion of high-arsenic groundwater in geological divisions.

Partition Type	Counting	Proportion
Modern Yellow River Impact Zone	31	50.8%
Neihuang Ancient River Channel	19	31.1%
Wuzhi River Interbank Zone	9	14.8%
Puyang ancient river interzone	2	3.3%

). The highest arsenic concentrations were observed in groundwater from Dongtun township and Zuocheng township in Yanjin County, where the concentrations reached 150 and 190 μg/L, respectively. These areas are situated in low-lying regions at the front of alluvial fans and within the Wuzhi interfluvial zone. The likelihood of high-arsenic groundwater occurring near paleoriver channels has further increased due to historical changes in the Yellow River.

The correlation between total arsenic concentration and well depth (Figure 4) indicates that high-arsenic groundwater is predominantly found at depths ranging from 30 to 75 m. Groundwater samples with arsenic levels surpassing 50 μg/L are primarily located in aquifers at depths of 30–45 m. Hydrogeological information demonstrates that the thickness of quaternary loose sediments in the piedmont plain ranges from 2 to 50 m. The active hydrogeochemical processes influenced by groundwater flow in this area promote the mobilization of arsenic from sedimentary environments [14].

4.2. Hydrochemical Characteristics

Based on the analysis of the Piper trilinear diagram (Figure 5), the majority of the groundwater sample points are located in the central area of the cation triangle, suggesting that Na⁺, K⁺, and Ca²⁺ are the predominant cations in the groundwater. In contrast, the distribution of groundwater samples in the anion triangle exhibits greater variability. A considerable number of samples group in the region characterized by HCO₃⁻.

Figure 5 indicates that the predominant hydrochemical type of groundwater in the study area is HCO₃-Ca·Na, suggesting that groundwater composition is primarily influenced by rock weathering and recharge processes. In some areas, SO₄·Cl-Na·Ca-type water is also observed, indicating potential influences from sulfate, chloride, and evaporation processes.

The distribution of cations in the groundwater displays a distinct trend of increased Na⁺ and K⁺ concentrations, likely associated with cation exchange processes and extended water–rock interactions. The relatively high levels of Ca²⁺ and Mg²⁺ imply that the weathering of carbonate rocks significantly affects groundwater composition. Furthermore, the location of the study area within a plain region results in slower flow velocities, which contribute to prolonged evaporation—leading to elevated concentrations of Cl⁻ and SO₄²⁻ in the groundwater and the formation of SO₄·Cl-Na·Ca-type water [15,16,17,18,19,22].

According to the Schoeller classification system, the ionic composition of groundwater is mainly governed by cation exchange, evaporation, and carbonate rock weathering. The presence of ions exceeding 25% milliequivalent percentages for both cations and anions reflects a varied hydrochemical landscape within the study area. High-arsenic groundwater is primarily located in regions characterized by high total dissolved solids (TDS), indicating that these samples exhibit relatively high mineralization, potentially influenced by a combination of evaporation, rock weathering, and anthropogenic activities.

In summary, the primary groundwater types in the study area are HCO₃-Ca·Na and SO₄·Cl-Na·Ca, reflecting the complex hydrogeological conditions across different regions [23].

Another useful index for water classification is Ionic Salinity or Total Ionic Salinity (TIS), showing the sum of anion and cation total contents (expressed in meq/L). Iso-TIS lines are reported in Figure 6 (SO₄ ²⁻ vs. HCO₃⁻ + Cl⁻), where low-arsenic groundwater falls in the 4 and 56 meq/L range of the iso-TIS lines and high-arsenic groundwater show TIS (from 12 to 44 meq/L).

4.3. Mechanism of High-Arsenic Groundwater Formation

Through comprehensive analysis of surface water hydrochemical data, Gibson suggested that the primary mechanisms influencing natural water systems can be categorized into three types: evaporation concentration, rock weathering, and atmospheric precipitation. The Gibbs diagram employs Na⁺/(Na⁺ + Ca²⁺) and Cl⁻/(Cl⁻ + HCO₃⁻) as the horizontal axes, with the logarithm of total dissolved solids (TDS) as the vertical axis. Characteristic regions corresponding to these mechanisms are illustrated in Figure 5 [24,25,26,27].

In the Gibbs diagram, the TDS exceeds 1000 mg/L and ion ratios are above 0.5. The region dominated by rock weathering is found in the central-left area of the diagram, characterized by TDS values ranging from 100 to 1000 mg/L and ion ratios below 0.5. The area dominated by atmospheric precipitation is located in the lower left, where TDS is under 100 mg/L and ion concentration ratios are greater than 0.5. Numerous studies have established the applicability of the Gibbs diagram to groundwater, and it has been extensively utilized in research concerning groundwater formation mechanisms; however, its inability to account for human activities introduces certain limitations.

The results (Figure 7) indicate that the majority of groundwater samples exhibit TDS values between 100 and 1000 mg/L, positioning them within the central region of the diagram. These findings suggest that rock weathering (water–rock interaction) is the primary process influencing groundwater geochemistry in the study area. It is inferred that the chemical composition of high-arsenic groundwater mainly results from the weathering of carbonate and silicate rocks. Previous research has indicated that Ca²⁺ is the predominant cation in groundwater, implying that groundwater composition may be influenced primarily by carbonate weathering.

The region dominated by evaporation concentration is positioned in the upper right of both figures. The Cl⁻/(Cl⁻ + HCO₃⁻) ratio represents the relative amounts of chloride and bicarbonate ions among the total anions. A ratio near 0 signifies HCO₃⁻ dominance, while a ratio close to 1 reflects Cl⁻ dominance. Data reveal that most groundwater samples possess Cl⁻/(Cl⁻ + HCO₃⁻) ratios ranging from 0.2 to 0.6, indicating significant contributions from both bicarbonate and chloride ions in these waters.

The total dissolved solids (TDS) values indicate the overall mineral content dissolved in the water and serve as crucial indicators of water hardness and salinity. TDS values in the study area range from about 10 mg/L to over 1000 mg/L, demonstrating considerable variations in groundwater mineralization. In the TDS versus Cl⁻/(Cl⁻ + HCO₃⁻) ratio plot, most high-arsenic groundwater samples (represented by red dots) show relatively high TDS values. This suggests a correlation between high-arsenic groundwater and increased mineralization, often associated with deeper groundwater flow or prolonged contact between groundwater and minerals in the slow-flowing waters of the plain region.

Additionally, further investigations have indicated that carbonate weathering has a more substantial influence on groundwater chemistry than silicate weathering in the study area. In arid to semi-arid inland basins, evaporation concentration also impacts the chemical characteristics of high-arsenic groundwater, alongside weathering. Figure 5 illustrates an upward trend of sample points toward the upper right, with some samples located within the evaporation concentration-dominated region, indicating a secondary role of evaporation in the groundwater geochemistry of the study area. Nevertheless, the effect of evaporation on arsenic concentration in groundwater remains limited.

In the analysis of groundwater sources (Figure 8), correlations among various ions can be utilized to investigate groundwater formation mechanisms and hydrogeochemical processes. Pearson correlation matrices were applied to examine the relationships between key ions that may influence arsenic enrichment in the groundwater of Zhouzhi and its surrounding areas. The results are presented in Figure 7. The correlation coefficients range within [−1, 1], where positive values indicate a positive correlation and negative values signify a negative correlation.

As illustrated in Figure 7, arsenic enrichment is primarily governed by reducing conditions. Correlation analysis indicated that arsenic is significantly positively correlated with Fe²⁺ and NH₄⁺ (0.621 and 0.563, respectively), suggesting that under reducing conditions, the reductive dissolution of iron and the decomposition of organic matter—which generate ammonium—may enhance the release and migration of arsenic. In such environments, arsenic predominantly exists as As(III), which is more mobile and soluble, leading to increased arsenic concentrations.

Conversely, arsenic displays a negative correlation with oxidative indicators such as SO₄²⁻ and NO₃⁻ (−0.173 and −0.102, respectively), implying that oxidizing conditions may promote arsenic immobilization, thereby inhibiting its dissolution. The presence of sulfate and nitrate generally indicates oxidizing conditions, where As(V) is more stable and tends to adsorb onto mineral surfaces, such as iron oxides, limiting arsenic migration. Therefore, dynamic variations in redox conditions are crucial in influencing arsenic transport in groundwater [27,28,29,30,31].

Additionally, the correlation between arsenic and the TDS (total dissolved solids) is relatively low (−0.148), indicating no significant linear relationship between arsenic enrichment and groundwater mineralization. This suggests that while mineral dissolution increases the overall mineral content in groundwater, it does not directly contribute to higher arsenic concentrations. Similarly, the correlation between arsenic and pH is weak (−0.017), suggesting that pH has a limited direct effect on arsenic concentration, although it may indirectly influence arsenic migration by affecting its speciation and adsorption behavior.

In summary, the mechanism behind high arsenic concentrations in groundwater in this area is closely associated with the stable presence of As(III) in localized reducing environments. Specifically, processes such as the reductive dissolution of iron minerals and the degradation of organic matter facilitate the release of arsenic from the solid phase into the aqueous phase. Moreover, changes in redox conditions are critical for arsenic mobilization and immobilization. Therefore, future strategies to mitigate high arsenic levels in groundwater should focus on regulating redox conditions to minimize arsenic migration and accumulation.

4.4. Redox Reactions

Indicators such as Fe²⁺ and NH₄⁺ serve as sensitive components reflecting the redox environment of groundwater. Analyzing the correlations among nitrogen, iron, and arsenic in groundwater allows for the qualitative identification of environmental characteristics associated with arsenic occurrence. Iron oxides act as the primary carriers of arsenic within aquifers. Arsenic generally exists on the surface of iron oxides in either adsorbed or bonded forms, and nitrogen can affect arsenic enrichment through dynamic interactions with Fe(II) present in iron oxides [32,33].

Figure 9 depicts the relationships among the concentrations of Fe²⁺ (ferrous iron), NH₄⁺ (ammonium), and arsenic in groundwater, contrasting high-arsenic groundwater with groundwater with normal arsenic levels. This emphasizes the influence of reductive dissolution of iron oxides on arsenic enrichment in reducing environments. Examining the variations in Fe²⁺ and NH₄⁺ concentrations aids in understanding the mechanisms of arsenic release under reducing conditions. The figure indicates that as the Fe²⁺ concentration rises, NH₄⁺ concentration also increases, particularly in high-arsenic groundwater samples. This trend suggests an interaction between the reductive dissolution of iron oxides and the degradation of organic matter under reducing conditions. The rise in Fe²⁺ signifies the dissolution of iron oxides, while NH₄⁺ results from the anaerobic decomposition of organic matter. The concurrent increase in these two components reflects the intensity of the reducing environment.

The elevation of Fe²⁺ and NH₄⁺ concentrations in groundwater is often linked with the release of arsenic, particularly As(III), which is adsorbed onto iron oxide surfaces. As illustrated in the circled region of the figure, Fe²⁺ and NH₄⁺ concentrations between 1 and 10 mg/L are densely clustered, especially in high-arsenic groundwater. This phenomenon indicates that when reducing conditions are sufficiently strong, the dissolution of iron oxides releases both Fe²⁺ and arsenic into the groundwater. Under such conditions, Fe²⁺ acts as an indicator of reducing environments and is closely associated with arsenic enrichment. High-arsenic groundwater exhibits Fe²⁺ concentrations ranging from 1 to 10 mg/L and NH₄⁺ concentrations from 0.1 to 10 mg/L, suggesting that arsenic dissolution and migration predominantly occur under strongly reducing conditions.

In contrast, groundwater samples with normal values of arsenic (blue squares) present lower concentrations of Fe²⁺ and NH₄⁺, indicating weaker reducing conditions that limit arsenic dissolution and migration. These samples are primarily located in regions where Fe²⁺ concentrations fall below 1 mg/L and NH₄⁺ concentrations are below 0.1 mg/L, suggesting that the reductive dissolution of iron oxides is relatively weak, thereby restricting the release of arsenic adsorbed onto iron oxide surfaces. Typically, more oxidizing environments hinder the accumulation of Fe²⁺ and NH₄⁺, consequently restricting arsenic migration.

The process of the reductive dissolution of iron oxides plays a key role in arsenic release under reducing conditions. When the Fe²⁺ concentration reaches about 10 mg/L, the arsenic concentration in high-arsenic groundwater increases significantly, highlighting the close association between arsenic enrichment and the reductive dissolution of iron oxides. During this process, iron serves as a major carrier of arsenic in reducing environments, and arsenic adsorbed on iron oxide surfaces is released into groundwater as the iron dissolves. In strongly reducing environments, As(III) is the more stable and soluble form, further enhancing arsenic migration and accumulation in groundwater.

The increase in ammonium (NH₄⁺) concentrations reflects the degradation and decomposition of organic matter in groundwater. The correlation between NH₄⁺ and Fe²⁺ indicates that in reducing environments, the decomposition of organic matter not only enhances the accumulation of ammonium but also indirectly intensifies reductive processes, which further accelerates the dissolution of iron oxides. This process is accompanied by the migration of arsenic from the solid phase to the aqueous phase. Consequently, the degradation of organic matter and the presence of anaerobic conditions establish favorable conditions for arsenic release.

The components or elements sensitive to oxidation–reduction in groundwater, such as NO₃⁻ and the SO₄²⁻/Cl⁻ molar ratio, serve as indicators of the surrounding redox environment, as evidenced by the correlation between these components and arsenic concentrations (Figure 10). In instances where the NO₃⁻ concentration exceeds 15 mg/L, the aquifer is characterized by a (weak) oxidizing state and the arsenic concentration remains relatively low, below 10 μg/L. This phenomenon occurs because the elevated redox potential of the NO₃⁻/NO₂⁻ pair inhibits the reduction of As(V) to As(III) and hinders the reductive dissolution of Fe/Mn (hydr)oxides, thereby preventing the mobilization of arsenic into the aqueous phase.

Conversely, samples exhibiting high arsenic concentrations are associated with extremely low SO₄²⁻/Cl⁻ ratios, suggesting that the groundwater environment is in a pronounced sulfate-reducing state. The reduced sulfur species (HS⁻ and S2⁻) generated during sulfate reduction can form thioarsenic complexes that exhibit high mobility and facilitate the migration and release of arsenic, which results in arsenic accumulation in these samples. Furthermore, HS⁻ and S2⁻ react with Fe²⁺ to produce insoluble iron sulfides, leading to reduced Fe²⁺ concentrations in these samples. Some studies have indicated that arsenic coprecipitates with iron sulfides under conditions of strong reduction, thus decreasing its mobility. However, this conclusion appears inconsistent with the present findings, which is potentially attributable to the specific types of iron sulfide minerals that form within the groundwater environment. Laboratory simulations have demonstrated that amorphous FeS has a limited impact on arsenic coprecipitation, while FeS₂ plays a more significant role [5,6,7,31,34,35,36,37,38,39,40,41,42].

Compared with oxidized As(V), reduced As(III) is more mobile and more easily activated in the aqueous phase. In oxidizing environments, arsenic typically coexists with Fe oxides or hydroxides through adsorption or coprecipitation; however, in reducing environments, the reductive dissolution of Fe oxides results in the release of coexisting arsenic into the aqueous phase. In strongly reducing environments, reduced sulfur species (S°, HS⁻, and S2⁻) formed during sulfate reduction can either react with Fe²⁺ to form iron sulfides (e.g., FeS and FeS₂) that partially immobilize arsenic or form thioarsenic complexes with arsenic, increasing its mobility. Additionally, sulfur species such as S° and HS⁻ can act as electron shuttles, facilitating the reductive dissolution of iron oxides, especially under microbial mediation. This is in line with previous studies on the influence of iron and sulfur on arsenic, and it can also be confirmed that iron and sulfur affect the morphology and content of arsenic by affecting the soil redox potential.

4.5. Verification and Analysis of Factors Influencing as Formation in Shallow Groundwater

In this study, traditional hydrogeochemical methods were initially utilized to analyze the water chemistry characteristics and genesis of the study area. To facilitate a more comprehensive and precise analysis, machine learning techniques—including principal component analysis (PCA), K-means clustering, non-negative matrix factorization (NMF), random forest analysis, and gradient boosting tree analysis—were employed to systematically assess the hydrochemical data. The mechanisms underlying the formation of high-arsenic groundwater in the study area were further discussed and validated.

PCA was employed to extract key hydrochemical information from the groundwater samples, with the first five principal components explaining 74.04% of the total variance, underscoring their importance in characterizing the chemical properties of the water samples. Analysis of the principal components (

Table 3. Dataset of parameters after training the PCA model.

Methodology
PCA	Statistical Index	PC1	PC2	PC3
	Max	0.41	0.58	0.57
	Min	−0.13	−0.30	−0.42
	Mean	0.18	0.03	0.01
	Contribution rate	35%	12%	10%

) indicated that PC1 accounted for 35.46% of the variance, with significant contributing factors including TDS (loading 0.4128), Ca²⁺ (loading 0.2922), and Mg²⁺ (loading 0.3933). However, while previous correlation coefficient analyses suggested a weak relationship between TDS and As, the PCA findings imply that elevated TDS levels reflect high groundwater mineralization, which may be linked to the dissolution of arsenic-bearing minerals. This aligns with the mechanisms of arsenic formation in high-arsenic groundwater as increased mineralization is often associated with the aquifer depth, water–rock interactions, and evaporation processes. Thus, PC1 signifies the primary role of mineralization in the release and migration of arsenic.

PC2 accounted for 12.52% of the variance, with a pH loading of −0.2991. Groundwater pH directly influences arsenic speciation: at elevated pH levels, arsenic is more likely to exist as As(V), which is less mobile, while under acidic conditions, As(III) is more mobile, facilitating arsenic entry into water. In high-arsenic groundwater, the carbonate system and acid–base balance regulate arsenic behavior. Carbonates act as buffers for pH levels and affect arsenic speciation and solubility. PC2 indicates that arsenic mobility is enhanced in environments characterized by high alkalinity and elevated HCO₃⁻ concentrations. Carbonate equilibrium may decrease arsenic adsorption by forming calcium carbonate precipitates, consequently increasing dissolved arsenic levels.

PC3 explained 10.95% of the variance, primarily associated with Fe²⁺ and Mg²⁺. Redox reactions involving Fe²⁺ are critical in arsenic release and migration. Under reducing conditions, the reductive dissolution of the iron oxides releases adsorbed arsenic, thereby increasing arsenic concentrations in the groundwater. This process typically occurs in anaerobic environments enriched with organic matter. PC3 suggests that metal ions, especially iron, significantly contribute to arsenic enrichment, aligning with the observations depicted in Figure 8.

Subsequent cluster analysis categorized the water samples into three distinct groups.

Table 4. Cluster analysis.

	Cluster 1 Mean	Cluster 2 Mean	Cluster 3 Mean
PC1	4.893218453	27.21070411	−0.435959501
PC2	0.163452669	−1.665889574	−0.003922947
PC3	−1.141352754	4.820688752	0.055541138
PC4	1.423497914	−5.429592234	−0.071679233
PC5	0.37764131	−0.392501558	−0.023346113
PC6	0.201097174	3.27863246	−0.026843749
PC7	−1.17724032	7.16416938	0.048230065
PC8	0.106600213	1.312202174	−0.012470271
PC9	−0.05767884	0.533911558	0.001607231
PC10	0.311394195	−0.215691252	−0.01969676
PC11	−0.05150223	−0.000866872	0.003408688
PC12	0.522860824	−1.438479223	−0.028625182
PC13	−0.006931872	0.179685313	−0.000284196
PC14	0.176629866	−0.31228099	−0.01038759
PC15	−0.012850346	0.02618517	0.000741406
PC16	0.023103539	0.019183984	−0.001606779

outlines the hydrochemical characteristics among these clusters. The results reveal that Cluster 2 possesses a markedly higher mean PC1 value (27.21) compared to the other clusters, indicating extremely high mineralization and potentially elevated arsenic concentrations. Further examination of cluster centers indicated that Cluster 1, with a mean PC1 value of 4.89, represents groundwater with moderate TDS and metal ion concentrations, likely corresponding to typical mineralized groundwater without particularly high arsenic levels. Cluster 2, exhibiting the highest mineralization, is likely influenced by enhanced mineral dissolution and water–rock interactions, leading to arsenic enrichment. Cluster 3 displays a lower mean PC1 value (−0.43), suggesting reduced mineralization, potentially correlating with shallow groundwater characterized by lower arsenic mobility.

Through additional factor decomposition of the PCA results using non-negative matrix factorization (NMF), three primary factors were identified: Factor 1 (pH factor), Factor 2 (salinity factor), and Factor 3 (metal content factor) (Figure 11c).

Factor 1 demonstrates the greatest characteristic weight for pH, recorded at 10.61, signifying the importance of pH in regulating arsenic behavior in groundwater. In mildly alkaline conditions (pH ≈ 7.4), As(V) exhibits increased stability in oxidizing environments and tends to be significantly adsorbed, reducing its likelihood of dissolution in water. Conversely, in reducing conditions, the presence of As(III) promotes arsenic mobility and enhances its solubility. Consequently, fluctuations in pH directly influence the chemical speciation of arsenic and its bioavailability.

Factor 2 is strongly linked to salinity, displaying high weights for Na⁺ (10.43) and Cl⁻ (5.43). This indicates a direct connection between this factor and water salinity. Elevated salinity is frequently associated with phenomena such as seawater intrusion or evaporation, which may result in arsenic release through mineral dissolution. For instance, in areas impacted by seawater intrusion, increased salinity in groundwater can facilitate the dissolution of arsenic-bearing minerals, consequently elevating arsenic concentrations [42,43,44,45,46,47,48,49,50,51].

Factor 3 is primarily associated with metal ions, particularly Fe²⁺ and Mg²⁺, which have weights of 7.36 and 7.90, respectively. Iron fulfills a dual function in groundwater systems: under oxidizing conditions, iron oxides can adsorb arsenic, while under reducing conditions, the dissolution of iron oxides can lead to arsenic release into the water. This underscores the significant role of metal ions, especially iron, in the migration and accumulation of arsenic in groundwater.

In research on high-arsenic groundwater, employing machine learning models like gradient boosting trees (GBTs) and random forests (RFs) has been effective in pinpointing crucial hydrochemical features that impact arsenic levels. By analyzing the importance of these features after training the models, not only can arsenic concentrations in groundwater be predicted accurately but the primary factors influencing arsenic activity can also be identified. This section examines in detail the outcomes of both models (refer to

Table 5. Random forest feature importance and gradient boosting tree feature importance.

Number	Random Forest Model Importance	Gradient Boosting Tree Model Importance
Feature 1	9.59 × 10−6	0
Feature 2	1.09 × 10−5	1.74 × 10−9
Feature 3	2.39 × 10−2	2.45 × 10−2
Feature 4	9.18 × 10−6	2.70 × 10−9
Feature 5	1.20 × 10−5	5.55 × 10−5
Feature 6	4.36 × 10−6	3.57 × 10−4
Feature 7	6.93 × 10−6	0
Feature 8	1.58 × 10−5	0
Feature 9	2.54 × 10−6	2.06 × 10−9
Feature 10	1.83 × 10−8	0
Feature 11	2.87 × 10−2	2.44 × 10−2
Feature 12	3.09 × 10−6	1.73 × 10−5
Feature 13	2.31 × 10−5	1.78 × 10−5
Feature 14	3.43 × 10−4	2.85 × 10−5
Feature 15	1.83 × 10−5	2.30 × 10−5
Feature 16	9.46 × 10−1	9.51 × 10−1
Feature 17	0	0
Feature 18	0	0
Feature 19	0	0
Feature 20	0	0
Feature 21	0	0
Feature 22	0	0
Feature 23	0	0
Feature 24	0	0
Feature 25	0	0
Feature 26	0	0
Feature 27	0	0
Feature 28	4.68 × 10−5	2.05 × 10−6
Feature 29	9.41 × 10−6	0
Feature 30	2.15 × 10−5	0
Feature 31	0	0
Feature 32	3.36 × 10−7	0
Feature 33	1.70 × 10−7	6.33 × 10−8
Feature 34	1.35 × 10−7	1.89 × 10−8
Feature 35	3.86 × 10−8	4.48 × 10−8
Feature 36	4.77 × 10−6	0
Feature 37	4.69 × 10−6	0
Feature 38	2.32 × 10−6	0
Feature 39	6.74 × 10−7	0
Feature 40	9.75 × 10−4	3.62 × 10−5
Feature 41	3.10 × 10−7	0
Feature 42	7.61 × 10−6	9.65 × 10−9
Feature 43	2.86 × 10−6	3.04 × 10−8
Feature 44	5.27 × 10−8	0
Feature 45	5.88 × 10−7	0
Feature 46	2.97 × 10−7	0
Feature 47	0	0
Feature 48	9.00 × 10−7	0
Feature 49	1.22 × 10−6	0
Feature 50	1.54 × 10−7	0
Feature 51	1.79 × 10−6	0
Feature 52	6.15 × 10−7	0
Feature 53	3.00 × 10−6	4.12 × 10−8
Feature 54	9.89 × 10−7	0
Feature 55	1.68 × 10−5	1.14 × 10−7
Feature 56	7.56 × 10−6	8.41 × 10−7
Feature 57	2.27 × 10−6	7.55 × 10−7
Feature 58	6.11 × 10−6	0
Feature 59	2.28 × 10−5	5.70 × 10−8
Feature 60	1.91 × 10−6	7.48 × 10−8
Feature 61	1.68 × 10−6	0
Feature 62	0	0

) and investigates the influence of significant features on predicting arsenic concentration.

The gradient boosting tree model reported an R² value of 0.9111 and a mean squared error (MSE) of 12,784.6562, demonstrating substantial accuracy in forecasting arsenic concentrations. Such precision indicates that the model effectively represents the complex nonlinear interactions between the arsenic concentration and various hydrochemical parameters.

During the analysis of feature importance, Feature_16 was found to have an importance score of 95.06%, significantly higher than other features, which suggests its major role in arsenic concentration predictions. Assuming Feature_16 relates to total dissolved solids (TDS) or specific key metal ions (like Fe²⁺ or Mg²⁺), the findings imply that mineralization (TDS) is a predominant factor affecting arsenic levels.

TDS denotes the overall quantity of dissolved ions in water and serves as a crucial marker of mineralization. Elevated TDS levels often suggest extensive interactions between water and rock, potentially leading to significant dissolution of arsenic-containing minerals (e.g., arsenopyrite), resulting in arsenic’s presence in the groundwater. Waters with higher mineralization levels exhibit stronger dissolving capabilities, enhancing arsenic release from minerals. This underlines the exceptional importance of TDS in the gradient boosting tree model. Furthermore, TDS might also relate to groundwater salinity and concentrations of various metal ions. In environments with high salinity, the competitive ionic conditions may hinder the adsorption of arsenic, maintaining its solubility and thus elevating its groundwater concentration [52].

In addition to TDS, metal ions such as Fe²⁺ and Mg²⁺ are pivotal in redox conditions. Particularly, the solubilization of Fe²⁺ under reductive conditions can mobilize arsenic previously adsorbed onto iron oxides into the water. Considering the elevated levels of Fe²⁺ and Mg²⁺ in some water samples, these ions could affect arsenic levels by altering the redox state. Therefore, the prominent importance of Feature_16 might also indicate its link to the redox conditions of the groundwater.

The random forest model recorded an R² value of 0.8716 and an MSE of 18,470.7957, indicating a slightly less accurate prediction capability compared to the gradient boosting tree model, yet it remains effective in illustrating the intricate relationships between arsenic concentrations and hydrochemical variables.

Similar to the gradient boosting tree model, Feature_16 also held a high importance score in the random forest model, further corroborating the critical role of mineralization (TDS) or significant metal ions (like Fe²⁺) in influencing arsenic mobility. Additionally, Feature_3 and Feature_11 were also identified as highly important in the random forest model, indicating their significant roles in arsenic prediction. Feature_3 is likely associated with specific metal content in the water, such as Ca²⁺ or Mg²⁺. These metals are intricately linked to arsenic’s chemical behavior in water, especially in the dissolution of aquifer minerals that can liberate arsenic. Furthermore, these metals might affect arsenic adsorption mechanisms, particularly in environments with higher pH levels, influencing arsenic’s precipitation or dissolution.

Feature_11 is associated with the redox environment of groundwater, potentially tied to the oxidation–reduction processes of Fe²⁺. Iron not only serves as an adsorption carrier for arsenic but also directly influences its speciation and mobility through redox reactions. In oxidizing environments, iron oxides adsorb arsenic, reducing its mobility; under reducing conditions, iron oxides dissolve, releasing arsenic into the water. Therefore, the redox environment plays a significant role in arsenic migration and enrichment.

From the feature importance analyses of both the gradient boosting tree and random forest models, several key conclusions can be drawn:

Mineralization, particularly TDS, is identified as a critical factor for predicting arsenic concentrations in groundwater. Highly mineralized waters, characterized by stronger dissolving capabilities, frequently exhibit elevated arsenic levels. This phenomenon is attributed to the concentration of dissolved solids, such as metal ions, in the water. As TDS increases, the release and migration of arsenic become more pronounced.

Fe²⁺ and other metal ions significantly influence arsenic behavior. The redox reactions involving iron ions directly affect the release or adsorption of arsenic, while other metal ions, such as magnesium, may indirectly impact arsenic mobility by altering groundwater hardness or pH levels.

Through the gradient boosting tree and random forest models, key hydrochemical factors affecting arsenic concentrations were successfully identified. These factors, including TDS, metal ions (such as Fe²⁺ and Mg²⁺), and redox conditions, play crucial roles in the migration and enrichment of arsenic. The insights provided by these findings are vital for understanding the mechanisms behind the formation of high-arsenic groundwater, emphasizing the roles of mineralization and redox conditions in arsenic mobility.

In summary, arsenic behavior in groundwater is influenced by multiple hydrochemical factors, with mineralization (TDS) being particularly significant. Higher mineralization implies greater dissolving power, facilitating the release of more arsenic-bearing minerals. Redox conditions, closely associated with Fe²⁺ concentrations, are also vital as the dissolution and precipitation of iron oxides are key to arsenic fixation and migration. In oxidizing environments, iron oxides effectively adsorb arsenic, while under reducing conditions, their dissolution releases arsenic into the water, explaining the correlation between high iron concentrations and elevated arsenic levels. In addition, pH variations significantly impact arsenic speciation and mobility. Under neutral to mildly alkaline conditions, As(V) tends to bind with iron oxides, reducing its mobility; conversely, under acidic or strongly alkaline conditions, As(III) becomes more soluble and mobile, facilitating arsenic’s entry into water.

PCA and NMF analyses indicate that the formation of high-arsenic groundwater is closely related to the dissolution of arsenic-bearing minerals. As minerals like arsenopyrite dissolve due to changes in the hydrochemical environment (e.g., increased salinity or mineralization), arsenic is progressively released. In water samples with high TDS, the enhanced solubility of mineralized water leads to greater arsenic release and enrichment. Iron oxides serve as primary arsenic carriers in groundwater. In oxidizing conditions, they can effectively adsorb arsenic; however, under reducing conditions, the reduction of iron oxides liberates arsenic, increasing its concentration in groundwater. NMF analysis suggests that metal contents of Fe²⁺ and Mn are important for arsenic release. This mechanism explains the higher arsenic concentrations found in reducing environments, especially in deeper aquifers or groundwater influenced by reducing conditions. Hydrogeological factors, including aquifer types, groundwater flow paths, recharge sources, and geological structures also significantly influence arsenic distribution. According to clustering analysis, different hydrochemical groups may represent distinct hydrogeological environments. Samples in Cluster 2 (high-mineralization clusters) are likely influenced by deeper hydrological conditions, where mineral dissolution and reducing environments lead to arsenic release and migration. In contrast, samples in Cluster 3, characterized by lower mineralization, likely originate from shallow aquifers with less mineral dissolution, resulting in relatively lower arsenic concentrations [51,53,54].

5. Conclusions

• The groundwater in the study area is predominantly weakly alkaline with low mineralization. Anions are primarily dominated by HCO₃⁻, whereas cations are mainly Ca²⁺ and Mg²⁺. The primary hydrochemical types are HCO₃-Ca·Na and SO₄·Cl-Na·Ca;
• The Gibbs diagram indicates that the dominant mechanisms influencing groundwater chemistry are rock weathering and evaporation, with a significant influence from carbonate dissolution. Correlation analysis of the groundwater chemical parameters revealed strong positive relationships among Fe²⁺, NH₄⁺, and As. Subsequent redox analysis confirmed that Fe²⁺ and NH₄⁺ are key factors contributing to the formation of high-arsenic groundwater in the region;
• This study comprehensively investigated the mechanisms behind the formation of high-arsenic groundwater using multiple analytical methods, including principal component analysis (PCA), non-negative matrix factorization (NMF), machine learning models (gradient boosting trees and random forest), and cluster analysis. The PCA results indicated that the first principal component (PC1) accounted for 35.46% of the total variance, primarily associated with TDS, Ca²⁺, and Mg²⁺ ions. Samples with high TDS levels exhibited strong correlations with elevated arsenic concentrations, underscoring the pivotal role of mineral dissolution in arsenic release. The second principal component (PC2), which explained 12.52% of the variance, underscored the substantial influence of pH on arsenic speciation. The third principal component (PC3), capturing 10.95% of the variance, highlighted the significant roles of Fe²⁺ and Mg²⁺ in the fixation and release of arsenic;
• NMF analysis identified three influential factors that further clarified the interactions between hydrochemical parameters and arsenic behavior. Factor 1 (pH factor) demonstrated the profound impact of pH on the dissolution and mobility of arsenic. Factor 2 (salinity factor) emphasized the effects of Na⁺ and Cl⁻ on arsenic solubility, particularly in environments with high mineralization where increased salinity promotes mineral dissolution and subsequent arsenic release. Factor 3 (metal content factor) was closely linked to the concentrations of Fe²⁺ and Mg²⁺;
• The gradient boosting tree model reported an R² value of 0.9111, reflecting a robust predictive capability. Feature importance analysis revealed that Feature_16 (likely representing TDS or a crucial metal ion) held the highest importance, with a score of 95.06%, indicating that mineralization is a key factor in arsenic migration. The random forest model, achieving an R² value of 0.8716, similarly highlighted the strong connections between mineralization (TDS) and metal ion concentrations with arsenic enrichment.