Application of the Self-Organizing Map (SOM) Algorithm to Identify Hydrochemical Characteristics and Genetic Mechanism of High-Nitrate Groundwater in Baoding Area, North China Plain

Abstract

Nitrate pollution poses a pervasive environmental issue for groundwater systems worldwide, which is particularly pronounced in the agricultural heartlands of the North China Plain. Combining hydrochemical analysis, the Self-Organizing Map algorithm, and Human Health Risk Assessment, 91 shallow groundwater samples were collected to identify the hydrochemical characteristic and the genetic mechanisms of high NO₃⁻ concentration groundwater. The SOM analysis identified three distinct hydrochemical clusters. Cluster 1, with the hydrochemical characteristic of HCO₃-Ca and HCO₃-Mg, is severely contaminated, showing the highest NO₃⁻, Ca²⁺, and TDS. In contrast, the majority of samples fell into Cluster 3, which is characterized by the lowest ion concentrations and an HCO₃-Ca type. Cluster 2, characterized by HCO₃-Ca/Mg, exhibits an intermediate chemical signature with elevated Na⁺, Mg²⁺, and HCO₃⁻. Nitrate concentrations varied widely, with 30.43% of collected samples exceeding the anthropogenic pollution threshold. Agricultural activities are identified as the primary nitrate source, with domestic sewage as a secondary contributor. The Human Health Risk Assessment further reveals that long-term exposure poses non-carcinogenic health risks, particularly for children, who are found to be the most vulnerable group. This study provides a hydrogeochemical perspective on nitrogen pollution in shallow groundwater and offers scientific support for sustainable groundwater management in typical agricultural regions worldwide.

1. Introduction

Groundwater serves as a vital freshwater resource, essential for drinking water supply, agricultural irrigation, and industrial processes [1]. With rapid socioeconomic growth and urban expansion, anthropogenic activities have exerted increasing pressure on both groundwater quantity and quality [2,3,4], giving rise to a range of hydro-environmental issues such as groundwater resources depletion, land subsidence, and nitrate contamination [5,6,7]. Groundwater nitrate pollution has emerged as a widespread environmental problem globally. Elevated nitrate concentrations pose considerable threats to ecosystem integrity and public health. Long-term consumption of nitrate-contaminated groundwater is associated with severe health issues, including methemoglobinemia and gastrointestinal disorders [8,9]. Therefore, elucidating the factors controlling nitrate enrichment is critical for developing scientifically sound land-use strategies, implementing effective pollution mitigation measures, and safeguarding groundwater quality.

Groundwater nitrate contamination is strongly linked to intensive agricultural and economic development [10,11]. Due to agricultural over-fertilization and infiltration of untreated wastewater, the Baoding area of the central North China Plain faces severe groundwater nitrate pollution [12]. These anthropogenic inputs have caused a continuous increase in groundwater nitrate concentrations. Furthermore, the regional hydrology is characterized by converging surface and subsurface flows that drain eastward into Baiyangdian Lake, a vital wetland system in the region. While previous studies in the Baoding area have largely addressed general water quality assessment and groundwater recharge–runoff–discharge processes, limited attention has been given to elucidating the specific genetic mechanisms of high-nitrate concentrations and conducting associated Human Health Risk Assessment [13,14]. Therefore, identifying the genetic mechanisms of nitrate pollution is essential for guiding sustainable development and utilization of regional groundwater resources, as well as for protecting ecological security.

Hydrogeochemical indicators serve as important technical tools for addressing scientific challenges in the fields of hydrogeology and groundwater pollution. Cluster analysis is a commonly used method for investigating the hydrochemical characteristics of groundwater [15,16]. However, traditional algorithms have significant limitations in high-dimensional data processing, making it difficult to accurately analyze the core characteristics of data. Currently, the Self-Organizing Map (SOM) method, a type of machine learning algorithm, has been widely applied in water quality modeling due to its superior performance in handling non-linear and high-dimensional hydrochemical data [17,18], and relevant studies provide valuable references for this research.

This study reveals the factors controlling groundwater nitrate distribution in Baoding City, a representative agricultural area in the North China Plain. Based on hydrochemical data from 91 shallow groundwater samples, the SOM algorithm is employed to identify the key hydrochemical processes controlling the spatial variability of elevated nitrate concentrations. The specific objectives areas follows: (1) delineate the spatial patterns of high-nitrate groundwater in the study area; (2) elucidate the hydrochemical mechanisms influencing nitrate enrichment; and (3) assess the associated human health risks posed by nitrate contamination. The findings are expected to provide scientific support for sustainable groundwater management and protection strategies.

2. Geological Setting

The study area is located in the central part of the North China Plain (Figure 1). Tectonically, it belongs to the piedmont sloping plain along the northern section of the eastern flank of the Taihang Mountains, with the terrain generally sloping eastward and northeastward. The region is primarily covered by Quaternary deposits, which gradually thickens from west to east and transition from a single layer to multiple layers, with a thickness of 200~600 m. Lithologically, the piedmont plain consists mainly of sand–gravel and pebble layers, whereas the central–eastern plain is dominated by interbedded sand and cohesive soil.

The study area is characterized by a continental monsoon climate. The multi-year average temperature and annual mean precipitation are 13.4 °C and 498.9 mm, respectively. Over 50% of the annual precipitation is concentrated in July and August. Nine rivers are distributed within the study area, including the Baigou River, Southern Juma River, Tang River, Pu River, Cao River, Ping River, Fu River, Xiaoyi River, and Zhulong River. However, the construction of upstream reservoirs has led to perennial drying of most river channels in the area. As a result, surface water resources are limited, rendering groundwater an indispensable source for sustaining local water supply and socioeconomic activities [13].

The groundwater in the study area is mainly Quaternary unconsolidated porous water. Generally, the aquifers are divided into four aquifer groups, designated as Groups I, II, III, and IV from shallow to deep. Aquifer Groups I and II have good hydraulic connectivity and are regarded as a single aquifer system, i.e., shallow aquifers, with a burial depth of 20~212 m. Aquifer Groups III and IV belong to deep aquifers. Groundwater recharge in the study area mainly includes atmospheric precipitation, agricultural irrigation infiltration, and surface water seepage. The primary discharge pathway is artificial pumping. At the regional scale, the general groundwater flow direction is from northwest to southeast. The hydraulic connectivity between shallow and deep aquifers is weak, and the research focus of groundwater nitrate in this study is mainly on shallow groundwater.

3. Samples and Method

3.1. Samples

A total of 91 groundwater samples were collected in Baoding City in November 2023. This sampling timeframe was intentionally selected to minimize the impact of seasonal variations on hydrochemical test data, thereby enhancing the rigor of the data analysis. All samples were stored in sterile polyethylene bottles. Prior to sampling, the sampling bottles were rinsed three times with the target water body. During sampling, the bottles were filled completely without leaving any air bubbles. After sampling, all sample bottles were sealed with parafilm, thereby ensuring the authenticity of the samples and the reliability of subsequent testing. To maintain the stability of ion components in the samples, pretreatment was conducted simultaneously during the sampling process. For the cation analysis, concentrated nitric acid was added dropwise to adjust the pH to < 2.

The sample analysis included on-site testing and laboratory testing. On-site testing parameters comprised pH and total dissolved solids (TDSs). These parameters were measured with high precision using an SX-620 pH meter (Shanghai Sanxin Instrument Co., Ltd., Shanghai, China) and an SX-650 conductivity/TDS meter (Shanghai Sanxin Instrument Co., Ltd., Shanghai, China), respectively. Laboratory testing primarily focused on the determination of cation and anion concentrations in groundwater. Major cations (Na⁺, K⁺, Ca²⁺, Mg²⁺) were analyzed using an inductively coupled plasma optical emission spectrometer (ICP-OES; Perkin-Elmer Optima 5300DV, Waltham, Massachusetts, USA). Major anions (Cl⁻, SO₄²⁻, NO₃⁻) were determined via ion chromatography (ICS-2500, Dionex, Sunnyvale, California, USA). The HCO₃⁻ concentration was measured through acid–base titration using 0.02 mol/L H₂SO₄.

The reliability of laboratory-derived hydrochemical analysis results was validated using the charge balance error (CBE), as defined in Equation (1). If the CBE < 10%, the hydrochemical analysis results are considered reliable; if the CBE ≥ 10%, their reliability is deemed questionable [17,19]. The CBE values are calculated for all collected groundwater samples. The results show that the CBE values of all samples range from 0.96% to 3.85%, and the absolute value of CBE for all samples is below 10%. This confirms the authenticity and reliability of the data, which meets the accuracy requirements for subsequent analysis and research and provides a solid data foundation for further hydrochemical analysis.

$$CBE(\%)=\left|{\displaystyle \frac{{\displaystyle \sum cations-{\displaystyle \sum anions}}}{{\displaystyle \sum cations+{\displaystyle \sum anions}}}}\right|\times 100\%$$

(1)

3.2. Self-Organizing Map

Traditional clustering methods, such as K-means clustering and principal component analysis, are unable to directly identify the intrinsic structure of high-dimensional hydrochemical data. They tend to overlook the spatial relationships between clusters and lack the ability to capture the hierarchical structure of data, making it difficult to identify the dominant factors influencing hydrochemical characteristics. The SOM algorithm, a machine learning algorithm, provides a robust solution to address the aforementioned challenges [20].

The SOM algorithm, developed by Kohonen [21], is a clustering technique that facilitates the analysis of complexity and variability in hydrochemical data. A key advantage of the SOM algorithm is its ability to project and visualize high-dimensional data into a low-dimensional space while retaining the topological structure of the original data [15]. Within the clustering matrix, a smaller topological distance between neurons corresponds to a higher degree of similarity between them. If groundwater samples fall into the same neuron, this implies they share analogous hydrochemical characteristic. Differences and correlations in hydrogeochemical properties among samples can be visualized via the weight values of variables and the distribution patterns of samples within the SOM matrix [16,18].

The learning process of the SOM algorithm comprises two core stages, including self-organizing training and feature map generation. Furthermore, the computational workflow of the SOM could be divided into three specific steps: (1) the number of neurons is computed using the formula $5\sqrt{n}$, where n denotes the total number of groundwater samples [17]; (2) during the partitioning of the neural matrix, the minimum Quantization Error (QE) and Topographic Error (TE) are used to identify the optimal number of neurons [20]; and (3) the minimum Davies–Bouldin Index (DBI) is applied to determine the optimal number of clusters [15].

In the present study, a total of 91 groundwater samples were analyzed by the SOM approach. Ten hydrogeochemical parameters were incorporated into the SOM computations, including K⁺, Ca²⁺, Na⁺, Mg²⁺, HCO₃⁻, SO₄²⁻, Cl⁻, NO₃⁻, pH, and TDS. The SOM algorithm was implemented using MATLAB software R2022a [21].

3.3. Human Health Risk Assessment

Human Health Risk Assessment (HHRA) serves as a methodological framework for establishing the causal nexus between groundwater contamination and adverse human health outcomes, thereby facilitating the quantitative characterization of potential exposure risks [15]. This assessment framework fully accounts for populations across different age groups and genders, encompassing children, adult females, and adult males. Among its key analytical steps, the hazard quotients for oral exposure (HQ_Oral) and dermal exposure (HQ_Dermal) can be derived using the following equations [12]:

$$H{Q}_{Oral}={\displaystyle \frac{I_{Oral}}{Rf{D}_{Oral}}},I_{Oral}={\displaystyle \frac{\left(c_i\times IR\times EF\times ED\right)}{BW\times AT}}$$

(2)

$$H{Q}_{Dermal}={\displaystyle \frac{I_{Dermal}}{Rf{D}_{Oral}\times AB{S}_{gi}}},I_{Dermal}={\displaystyle \frac{\left(c_i\times K_{sp}\times S_a\times T\times EV\times CF\times EF\times ED\right)}{BW\times AT}}$$

(3)

The empirical parameters involved in the above equation are presented in

Table 1. The empirical parameters of Human Health Risk Assessment.

Parameter	Unit	Children	Female	Male
Oral reference dose for NO3− (RfDoral)	mg/(kg × day)	0.04	0.04	0.04
Gastrointestinal absorption factor (ABSgi)	-	1	1	1
Drinking rate (IR)	L/day	0.7	1.5	1.5
Exposure frequency (EF)	days/year	365	365	365
Exposure duration (ED)	years	6	30	30
Average body weight (BW)	kg	15	55	75
Average time (AT)	days	2190	10,950	10,950
Skin permeability (Ksp)	cm/h	0.001	0.001	0.001
Contact duration (T)	h/d	0.4	0.4	0.4
Exposure frequency of daily dermal contact (EV)	-	1	1	1
Unit conversion factor (CF)	L/cm3	0.001	0.001	0.001
Skin surface area (Sa)	-	6597.01	15,475.85	18,742.36
Average body height (H)	cm	99.4	153.4	165.3

Note: The empirical parameters are referenced by [22,23,24,25,26].

[12]. For groundwater-related health risk assessment, the total non-carcinogenic health risk associated with groundwater utilization may be calculated using the following equation [12],

$$H{I}_{Total}={\displaystyle \sum _{i=1}^{n}H{I}_i,}H{I}_i=H{Q}_{Oral}+H{Q}_{Dermal}$$

(4)

4. Results

4.1. SOM Clustering Results

The ten different variables of 91 groundwater samples are selected to conduct the SOM analysis, including K⁺, Ca²⁺, Na⁺, Mg²⁺, HCO₃⁻, SO₄²⁻, Cl⁻, NO₃⁻, pH, and TDS. An SOM with 7 × 7 topology is developed for sample classification. As presented in Figure 2, yellow and blue colors correspond to the high and low values of neuron parameters, respectively.

Color gradient variations across the ten variables offer an effective approach to analyzing hydrochemical processes. By comparing color gradients of different hydrochemical indicators, four distinct relationships are readily identifiable. Firstly, Na⁺ and Cl⁻ exhibit analogous color gradients, decreasing from the upper-right corner to the lower-left corner (Figure 2a), suggesting that they share a common source [15]. During water–rock interactions, halite dissolution serves as their primary source. Secondly, Na⁺ and Mg²⁺ both show a decreasing trend from the upper-right to the lower-left corner of the map (Figure 2a), indicating that they are regulated by the same geochemical process. This process is mainly driven by the weathering and hydrolysis of silicate minerals (e.g., albite) [17]. It may also be associated with the retention of Mg²⁺ released via dolomite dissolution and the subsequent combination of this retained Mg²⁺ with exogenous Na⁺ [17]. Thirdly, the color gradients of Na⁺ vs. SO₄²⁻ and Ca²⁺ vs. HCO₃⁻ display obvious differences, suggesting the potential occurrence of cation exchange or carbonate precipitation. Fourthly, when Ca²⁺ and SO₄²⁻ share similar color gradients, this indicates the occurrence of gypsum dissolution or precipitation. This further implies an additional source for these two ions.

As presented in

Table 2. The Davies–Bouldin indices (DBIs) of different clustering.

Clusters	1	2	3	4	5
DBI	NaN *	0.9898	0.8418	0.8946	0.9126

Note: * the null value.

, the calculated Davies–Bouldin Index (DBI) values indicate that all groundwater samples collected from the study area are effectively classified into three distinct clusters, with significant differences in physicochemical parameters between clusters. Specifically, six samples (6.59% of the total) are classified into Cluster 1, which is characterized by high Ca²⁺ concentrations. This cluster also exhibits the highest TDS among the three clusters, and its NO₃⁻ concentration exceeds the permissible limits. Thus, Cluster 1 is categorized as “high NO₃⁻ and Ca²⁺ concentrations with the highest TDS”. Cluster 2 comprises 21 samples (23.08% of the total samples) and is defined by “high Na⁺, Mg²⁺, and HCO₃⁻ concentrations”. The largest group, Cluster 3, is characterized by low Ca²⁺, low HCO₃⁻, and low TDS, including 64 samples (70.33% of the total samples). It also has the lowest NO₃⁻ concentration among the three clusters, leading to its classification as “low concentrations of Ca²⁺, HCO₃⁻, and NO₃⁻ with low TDS”.

4.2. Hydrochemical Results

The statistical summary of groundwater physicochemical parameters is provided in

Table 3. Statistical results of physicochemical parameters of different SOM’s clusters.

		pH	TDS	K+	Ca2+	Na+	Mg2+	HCO3−	SO42−	Cl−	NO3−
Unit			mg/L	mg/L	mg/L	mg/L	mg/L	mg/L	mg/L	mg/L	mg/L
Cluster 1	Max	7.3	1172	5.9	218.0	140.0	88.3	559.0	197.0	203.0	68.0
	Min	7.0	527	0.6	116.0	14.7	30.7	337.0	61.4	27.5	9.7
	Ave	7.2	719	3.2	149.0	52.5	58.3	412.3	116.5	98.9	34.8
	SD	0.1	220	2.1	35.5	42.4	20.0	74.0	56.1	61.4	21.3
Cluster 2	Max	7.7	1128	1.8	159.0	162.0	88.0	567.0	135.0	358.0	36.8
	Min	7.1	457	0.2	40.7	13.9	47.6	349.0	15.2	19.2	2.1
	Ave	7.3	609	0.7	95.9	60.0	65.5	469.2	62.5	77.7	12.6
	SD	0.2	151	0.4	30.2	33.9	14.0	51.1	34.7	67.5	8.9
Cluster 3	Max	8.1	487	6.6	132.0	99.2	52.5	475.0	103.0	94.8	20.0
	Min	7.2	219	0.2	23.3	4.4	9.6	161.0	4.6	5.7	0.0
	Ave	7.5	351	1.4	80.4	18.2	29.7	312.7	33.6	25.3	6.2
	SD	0.2	64	1.0	19.5	12.7	9.7	62.0	24.0	15.2	4.8

, which serves as a key indicator of the overall hydrochemical characteristics. To offer a comprehensive depiction of physicochemical characteristic across the study area, Table 1 and Figure 3 and Figure 4 present an integrated analysis combining statistical and visual representations. Cluster 1 exhibits a narrow pH range from 7.0 to 7.3, averaging 7.2. The mean value of TDS is 719 mg/L, ranging from 527 mg/L to 1172 mg/L. Notably, this cluster shows the highest NO₃⁻ concentration among all clusters, with the concentrations varying from 9.7 mg/L to 68.0 mg/L and a mean value of 34.8 mg/L. The hydrochemical characteristics are predominantly HCO₃-Ca and HCO₃-Mg. In Cluster 2, pH values range from 7.1 to 7.7, with an average of 7.3. TDS varies between 457 mg/L and 1128 mg/L, with an average value of 609 mg/L. Similar to the Cluster 1 samples, the hydrochemical characteristic in this cluster is also characterized by HCO₃-Ca and HCO₃-Mg types. Cluster 3 is distinguished by pH values between 7.2 and 8.1, averaging 7.5. It has the lowest TDS levels among all clusters, ranging from 219 mg/L to 487 mg/L with a mean of 351 mg/L. The dominant hydrochemical characteristic is HCO₃-Ca. Correspondingly, this cluster also exhibits the lowest NO₃⁻ concentrations, ranging from below the detection limit to 20.0 mg/L, with an average of 6.2 mg/L.

Figure 4 displays the Pearson correlation coefficients among various physicochemical parameters measured in groundwater samples. Conducting a quantitative analysis of inter-ionic correlations aids in deciphering key hydrogeochemical processes, tracing ion origins, and understanding solute transport mechanisms within groundwater systems. When two ions show a significant positive correlation in concentration, it often implies a shared source or co-migration behavior. Conversely, a negative correlation typically reflects competitive interactions or inverse geochemical pathways.

5. Discussion

5.1. Hydrochemical Driven Factors

Ionic ratios are widely employed as an effective method for identifying hydrochemical processes [27,28]. To distinguish the dominant hydrochemical process controlling hydrochemical characteristics, the Gibbs diagram is used to evaluate the effect of evaporation, water–rock interaction, and atmospheric precipitation on hydrochemical characteristics. As shown in Figure 5, the Na⁺/(Na⁺ + Ca²⁺) ratios of groundwater samples in Cluster 1 range from 0.10 to 0.36, while their Cl⁻/(Cl⁻ + HCO₃⁻) ratios are 0.12~0.39. For Cluster 2 and Cluster 3, the Na⁺/(Na⁺ + Ca²⁺) ratios are 0.10~0.65 and 0.06~0.79, respectively, and the Cl⁻/(Cl⁻ + HCO₃⁻) ratios are 0.07~0.59 and 0.03~0.41, respectively. All groundwater samples are predominantly clustered in the central region of the Gibbs diagram, which indicates that water–rock interaction is the primary process controlling the hydrochemical characteristics of groundwater.

To further elucidate the processes of water–rock interaction, this end-member plotting is introduced to identify the specific rock types involved in the hydrochemical processes [29,30], including evaporites, silicate rocks, and carbonate rocks. The diagnostic binary diagrams (Mg²⁺/Na⁺ vs. Ca²⁺/Na⁺ and HCO₃⁻/Na⁺ vs. Ca²⁺/Na⁺) are analyzed to identify the dominant rock types controlling groundwater hydrochemical characteristics. As presented in Figure 6, all samples in the three clusters are distributed within the transitional zone between silicate rocks and carbonate rocks. Specifically, the Cluster 2 samples are closer to the silicate rock-dominated area, while the Cluster 3 samples are nearer to the carbonate rock-dominated area. This distribution pattern indicates that silicate weathering and carbonate dissolution constitute the primary sources of ionic constituents in the groundwater.

The ionic ratios and Pearson’s correlation analysis are used to identify the hydrochemical processes [31,32]. As shown in Figure 7a, most samples are distributed close to the 1:1 line for Na⁺/Cl⁻, and a strong positive correlation was observed between Na⁺ and Cl⁻ (r = 0.51), implying a common origin dominated by halite dissolution. Between Ca²⁺ and SO₄²⁻, a moderate correlation (r = 0.38) is identified. With most samples exhibiting Ca²⁺/SO₄²⁻ ratios above 2 (Figure 7b), it can be inferred that carbonate weathering contributes supplementary Ca²⁺ to the groundwater. The combined relationship of Ca²⁺ + Mg²⁺ versus HCO₃⁻ + SO₄²⁻ is used to assess the influence of dolomite, calcite, and gypsum solution. In Figure 7c, the majority of samples plot near the 1:1 line, indicating that the dissolution of these minerals predominantly regulates Ca²⁺ and Mg²⁺ levels. Further examination of HCO₃⁻ against Ca²⁺ + Mg²⁺ (Figure 7d) indicates that most data points plot below the 1:2 equiline, suggesting the release of HCO₃⁻, Ca²⁺, and Mg²⁺ through carbonate weathering, while gypsum dissolution may contribute to excess Ca²⁺. Additionally, HCO₃⁻ correlates moderately with Ca²⁺ (r = 0.49) and strongly with Mg²⁺ (r = 0.86), supporting that carbonate minerals (e.g., calcite and dolomite) serve as the principal sources of HCO₃⁻, Ca²⁺, and Mg²⁺ in the groundwater system.

5.2. Identifying Nitrate Sources

In natural environments, the NO₃⁻ concentration typically remain below 10 mg/L, which is indicated that the threshold widely recognized as an indicator for detecting anthropogenic pollution [33,34]. In the study area, the NO₃⁻ concentration of collected groundwater samples ranged from 0.02 mg/L to 68.0 mg/L. Specifically, 83.8% of samples in Cluster 1, 52.4% in Cluster 2, and 17.2% in Cluster 3 exceeded this 10 mg/L benchmark. This widespread exceedance indicates that groundwater quality across the study area is significantly impacted by anthropogenic activities. Accordingly, a source apportionment study is conducted for the aforementioned high-nitrate groundwater samples to elucidate the mechanisms underlying nitrate enrichment in groundwater.

Based on the molar ratios of Cl⁻/Na⁺ vs. NO₃⁻/Na⁺ and Cl⁻ vs. NO₃⁻/Cl⁻, the sources of NO₃⁻ can be effectively identified by comparing these ratios with the ranges of known end-members [35,36], such as those from agricultural activities, industrial production, and domestic sewage. As shown in Figure 8a, most samples in Cluster 1, Cluster 2, and Cluster 3 are characterized by high NO₃⁻/Na⁺ ratios and low Cl⁻/Na⁺ ratios. This is attributed to the application of nitrogen fertilizers in agricultural activities within the study area, which introduces a large amount of NO₃⁻. Thus, this proportional characteristic indicates that agricultural activities are the main source of NO₃⁻. In Figure 8a, a few samples from the three clusters are distributed between the agricultural and sewage end-members. This placement is due to the concurrent increase in NO₃⁻ and Cl⁻ from sewage, as Cl⁻ primarily originates from human waste and household detergents. Therefore, the NO₃⁻ in the study area likely comes from a mixture of agricultural and domestic sewage sources.

As shown in Figure 8b, the agricultural activity end-member exhibits the characteristics of high NO₃⁻/Cl⁻ ratios and low Cl⁻ concentrations, which is due to the increase in NO₃⁻ caused by fertilization and the relatively low content of Cl⁻. The domestic sewage and fecal source end-member shows the characteristics of low NO₃⁻/Cl⁻ ratios and high Cl⁻ concentrations, as domestic sewage and feces contain a large amount of Cl⁻ but relatively little NO₃⁻. The soil nitrogen source end-member presents the characteristics of medium NO₃⁻/Cl⁻ ratios and medium Cl⁻ concentrations: the mineralization of soil organic nitrogen produces NO₃⁻, while Cl⁻ is mainly from the background, so the ratio falls between the agricultural end-member and the sewage end-member. The rainfall end-member exhibits the characteristics of low Cl⁻ concentrations and medium NO₃⁻/Cl⁻ ratios; however, no samples from the study area are distributed in this region, indicating that rainfall is not the main source of NO₃⁻ in the study area. As shown in Figure 8b, most samples are distributed in the area between the agricultural activity end-member and the sewage end-member, indicating that NO₃⁻ pollution in the groundwater of the study area is the result of multi-source mixing. In conclusion, high NO₃⁻ pollution in the groundwater of the study area is mainly dominated by agricultural activities and is also affected by the superimposition of domestic sewage.

5.3. HHRA Results

Groundwater resources in the study area serve as an important water source for human livelihood and agricultural production. Quantitative assessment of the potential health risks associated with groundwater is a prerequisite for its development and utilization. HHRA is a widely used evaluation method that establishes a non-linear relationship between human health conditions and potential health risks posed by major contaminants. This study quantitatively estimates the potential health risks of groundwater with high-nitrate concentrations in the study area.

The total hazard index (HI_Total) for children, adult females, and adult males is shown in Figure 9. Generally, health risks based on the hazard quotient (HQ) values are classified into three categories as follows: negligible health risk when HQ < 1, moderate health risk when HQ ranges between 1 and 4, and high health risk when HQ > 4. For children, HITotal values range from 0.21 to 79.63, with a mean value of 11.17. In the children group, 72% of samples are categorized as high health risk, 21% as moderate health risk, and 8% as negligible health risk. For adult females, HI_Total values vary from 0.12 to 46.55, with a mean of 6.53. In this group, 62% of samples indicated high health risk, 28% moderate health risk, and 10% negligible health risk. For adult males, HI_Total values range from 0.09 to 34.17, with an average of 4.79. Among the adult males, 12%, 38%, and 50% of samples are classified as negligible, moderate, and high health risk, respectively. The HHRA results indicate that the non-carcinogenic health risks posed by high-nitrate groundwater decrease in the order of children > adult females > adult males. Consistent with previous studies, children exhibited greater susceptibility than adults, which can be attributed to their lower body weight and immature physiological systems [11].

6. Conclusions

The primary objective of this research is to investigate the hydrochemical characteristics and genetic mechanisms of high-nitrate groundwater in the shallow aquifer of the Baoding area, North China Plain. Based on the analytical results from 91 groundwater samples, the study employs a combined approach of hydrochemical analysis, the SOM algorithm, and Human Health Risk Assessment, alongside a quantitative evaluation of the associated human health risks. The key conclusions are as follows:

(1)

The SOM analysis effectively categorized 91 samples into three distinct clusters with markedly different physicochemical characteristics. Cluster 1 (6.59% of samples) is characterized by the highest concentrations of NO₃⁻ (exceeding permissible limits), Ca²⁺, and TDS, suggesting a high degree of mineralization and nitrate pollution. In contrast, Cluster 3 (70.33% of samples) represents the largest group and exhibits the lowest levels of NO₃⁻, Ca²⁺, HCO₃⁻, and TDS. Cluster 2 (23.08% of samples) presents an intermediate signature with elevated Na⁺, Mg²⁺, and HCO₃⁻. The statistical data and supporting figures collectively affirm these hydrochemical characteristics, which are predominantly HCO₃-Ca and HCO₃-Mg types for Clusters 1 and 2, and HCO₃-Ca for Cluster 3.

(2)

The NO₃⁻ concentrations in the 91 shallow groundwater samples range from below the detection limit to 68.0 mg/L, with a mean value of 9.59 mg/L. Notably, 30.43% of the samples surpassed the anthropogenic pollution threshold of 10 mg/L, indicating significant anthropogenic influence. Molar ratio analyses further reveal that agricultural activities are the primary source of NO₃⁻, supplemented by contributions from domestic sewage.

(3)

Utilizing high-nitrate groundwater for long-term consumption presents a potential health hazard. The assessment reveals a discernible pattern in non-carcinogenic risk: children are at the highest risk, followed by adult females and then adult males. The elevated risk for children is directly linked to their lower body weight and heightened sensitivity of their developing metabolic processes.