The Crucial Role of Data Quality Control in Hydrochemical Studies: Reevaluating Groundwater Evolution in the Jiangsu Coastal Plain, China

Abstract

A vital step for any hydrochemical assessment is properly carrying out quality assurance and quality control (QA/QC) techniques to evaluate data confidence before performing the assessment. Understanding the processes governing groundwater evolution in coastal aquifers is critical for managing freshwater resources under increasing anthropogenic and climatic pressures. This study reassesses the hydrochemical and isotopic data from the Deep Confined Aquifer System (DCAS) in the Jiangsu Coastal Plain, China, by firstly applying QA/QC protocols. Anomalously high Fe and Mn concentrations in several samples were identified and excluded, yielding a refined dataset that enabled a more accurate interpretation of hydrogeochemical processes. Using hierarchical cluster analysis (HCA), principal component analysis (PCA), and stable and radioactive isotope data (δ²H, δ¹⁸O, ³H, and ¹⁴C), we identify three dominant drivers of groundwater evolution: water–rock interaction, evaporation, and seawater intrusion. In contrast to earlier interpretations, we present clear evidence of active seawater intrusion into the DCAS, supported by salinity patterns, isotopic signatures, and local hydrodynamics. Furthermore, inconsistencies between tritium- and radiocarbon-derived residence times—modern recharge indicated by ³H versus Pleistocene ages from ¹⁴C—highlight the unreliability of previous paleoclimatic reconstructions based on unvalidated datasets. These findings underscore the crucial role of robust QA/QC and integrated tracer analysis in groundwater studies.

1. Introduction

The reliability of hydrochemical studies hinges on robust data quality control (QC) throughout both field and laboratory phases. Inadequate quality assurance and quality control (QA/QC) can introduce biases, leading to erroneous interpretations and undermining the credibility of research findings. Despite the critical role of QA/QC, it is often overlooked in hydrochemical research. While detailed laboratory methodologies are frequently described, QA/QC procedures are rarely presented explicitly, possibly because It is assumed that they are inherently followed—a presumption that does not necessarily reflect actual practices. This oversight can result in the acceptance of flawed data, thereby compromising the integrity of the study.

A fundamental QA/QC tool in hydrochemical studies is the Charge Balance Error (CBE) calculation, which serves as an initial indicator of analytical accuracy by assessing the consistency between cations and anions in water samples [1,2,3]. However, relying solely on CBE is insufficient. Fritz mentioned that it is advisable to complement this approach with additional QA/QC measures [2], such the ratio of total dissolved solids (TDS) concentrations to specific conductance, the relationship between field and laboratory TDS, and the ratio between laboratory TDS and the sum of all dissolved ions (calculated TDS), among others, to comprehensively evaluate data quality [4,5,6].

The necessity of rigorous QA/QC becomes even more noticeable when employing statistical methodologies, such as multivariate analysis, to interpret hydrochemical data. These techniques, including principal component analysis (PCA) and hierarchical cluster analysis (HCA), are sensitive to data quality. Inaccurate or biased data can lead to misleading conclusions about the underlying processes affecting water chemistry. Therefore, ensuring data quality through stringent QA/QC protocols is essential in order to derive valid and actionable insights from complex hydrochemical datasets [7,8,9].

Xu et al. [10] conducted a hydrochemical and isotopic evaluation of groundwater in the Jiangsu Coastal Plain. However, in our opinion, their assessment presents QA/QC flaws that are worth mentioning as they could impact the interpretation reliability.

The objective of the present study is to critically reassess the hydrochemical and isotopic data from the DCAS as presented by Xu et al. [10], explicitly addressing the fact that their data quality assurance and quality control (QA/QC) were insufficient. During our QA/QC we identified and excluded outliers, and also anomalous Fe and Mn concentrations, which were discarded before statistical analysis. Our reanalysis leads to distinctly different conclusions, emphasizing the significance of comprehensive data validation and integrated assessment in hydrochemical and isotopic groundwater studies.

2. Study Area

The Jiangsu Coastal Plain is located in the eastern Jiangsu Province, north of the Yangtze River delta in China, with a shoreline of 954 km. The study area is situated at the southern end of the Jiangsu Coastal Plain (Figure 1a). Since the mid-20th century, the Jiangsu coast has been one of the most densely populated areas in China, experiencing continuous population growth [11]. According to the 2010 national census, its population density was about four times higher than the national average [12]. This demographic situation has put constant pressure on groundwater resources, as shallow aquifers typically have poor water quality.

The overexploitation of coastal aquifers is a common problem worldwide [13,14,15,16], including in China [17,18,19].

In coastal areas, excessive groundwater abstraction significantly increases the risk of seawater intrusion [19]. Therefore, accurately evaluating seawater intrusion processes and building respective capacities among actors is critical for developing and implementing effective groundwater management strategies according to Scheihing et al. [20].

The subsurface geology of the Jiangsu Coastal Plain consists of a multilayered alluvial system with sediments reaching depths of up to 600 m. Groundwater resources in the Jiangsu Coastal Plain are stored within three aquifer systems. The uppermost level, the Phreatic Aquifer System (PAS), formed during the Holocene and comprises estuarine and coastal sediments, predominantly fine sands and loamy clays, which are laterally discontinuous. The PAS is directly recharged by rainfall and interacts with the Rutai River. It has a thickness of 25–40 m, with a shallow water table ranging from 1 to 4 m below ground level (m bgl). The intermediate level, the Intermediate Confined Aquifer System (ICAS), formed between the Middle and Late Pleistocene, includes two confined aquifers (Aquifer I and Aquifer II) composed of fine sands, silts, and gravels. Aquifer I has an average thickness of about 100 m (40–140 m bsl), while Aquifer II is approximately 50 m thick (220–270 m bsl). Their water tables range from 2 to 3 m bgl and 3 to 5 m bgl, respectively. The lowermost level, the Deep Confined Aquifer System (DCAS), also includes two aquifers (Aquifer III and Aquifer IV) comprising Neogene to Early Pleistocene fluvio-lacustrine sands, gravels, silts, and clays. Aquifer III has an average thickness of 70 m (230–300 m bgl), while Aquifer IV is about 100 m thick (350–450 m bgl). Their water tables vary from 10 to 50 m bgl and 10 to 30 m bgl, respectively [10]. The DCAS is the primary groundwater resource in the area and the focus of this study. A similar aquifer framework has been described for other areas of northern coastal China (e.g., [19,21]).

The general groundwater flow direction within the Jiangsu Coastal Plain is from west to east towards the sea (Figure 1b). However, local groundwater flow is complex due to groundwater extraction in the area. For instance, Figure 1a illustrates a piezometric map for Aquifer III of the DCAS, showing a depression cone centered at Matang, where the water table is around 40 m bgl compared to about 20–25 m bgl near the northern shoreline near Yoco Port. This creates a local eastward groundwater flow from Yoco Port toward Matang.

3. Methodology

The hydrochemical dataset used in this assessment was originally presented in [10] and included measurements of total dissolved solids (TDS), pH, Fe, Mn, Na⁺, Ca²⁺, Mg²⁺, K⁺, HCO₃⁻, Cl⁻, and SO₄²⁻ in 21 groundwater samples from Aquifers III and IV (DCAS). Additionally, δ²H and δ¹⁸O isotopic data were available for 33 samples, which included the same 21 groundwater samples analyzed for hydrochemistry, 5 surface water samples (3 river and 2 ocean samples), 6 samples from the Phreatic Aquifer (PAS), 1 sample from Aquifer I (ICAS), and a final sample that could be attributed to either the Phreatic Aquifer or Aquifer I, as noted by Xu et al. [10].

Furthermore, the dataset also included tritium (³H) and carbon-14 (¹⁴C) measurements for the 21 samples from Aquifers III and IV (DCAS), which were the focus of this study. Hydrochemical data are presented in

Table 1. The hydrochemical data in the study by Xu et al. [10]. We have added columns for Charge Balance Error (CBE) and relative percentage (RP) of Fe and Mn regarding the TDS.

ID	Aquifer	pH	T	TDS	Cl−	HCO3−	SO42−	Ca2+	Mg2+	Na+	K+	H2SiO3	Fe	Mn	CBE	TDScalc	TDScalc *
			(°C)	mg/L	mg/L	mg/L	mg/L	mg/L	mg/L	mg/L	mg/L	mg/L	mg/L	mg/L	%	mg/L	mg/L
1	III	7.9	19	723.3	129.8	541	3.9	83.9	37.5	112.7	2.3	35.2	1.8	0.1	−0.4	673	671
2	III	7.9	17.8	2479.3	1113.3	526.3	94.4	236.3	94.7	462	5.8	24.2	13.2	0.2	−2.2	2303	2290
3	III	7.4	24.5	462	25.2	478	4	64.1	26.1	82.2	1.5	33.8	1.5	0	0.3	473	472
4	III	7.7	23.8	464	18.9	459	3.7	61	26.7	78.1	1.6	31.1	0.5	0	0.5	447	447
5	III	7.9	25.1	534	22.1	466	3.7	64	26.3	78.7	1.4	33.4	0.9	0	0.5	460	459
6	IV	7.7	23.8	542	63	490	3.8	48.1	20.6	140	1.3	28	0.9	0.1	0.3	547	546
7	IV	7.7	24.8	2262	1236	301	76.3	170	61.4	587	10.1	22.1	1.4	0.2	−2.1	2313	2311
8	III	7.8	22.9	942	357	407	15.1	50	23.6	278	2.6	26.5	1.8	0.1	−0.5	955	953
9	III	8.1	20.3	852	372	286	19.7	31.9	20.8	286	2.4	15.2	3.6	0.1	0.2	892	889
10	III	7.9	23	558	51.3	502	7.2	50.9	24	135	1.6	13.4	0.2	0.2	0.6	531	530
11	III	7.8	24.5	1156	375	313	209	115	49.3	236	5	11.4	2	0.1	0.1	1157	1155
12	III	7.7	24.9	1106	458	295	7.1	72.1	35.1	263	3.4	10.6	0	0.1	0.1	994	994
13	IV	8.3	21.1	1066	433	301	11.2	69.6	33.2	255	5.2	12.1	0	0	0.1	967	967
14	IV	7.7	22	992	338	419	12.7	63.1	24	250	2.5	18.1	775	16.4	−0.6	1706	914
15	IV	8	19.6	622	68.1	475	38.8	20.9	11.7	190	2.6	19.5	725	15.6	−0.2	1326	585
16	IV	7.8	29	1432	606	304	18.3	102	49.8	287	4.2	23.8	2010	61.3	−0.7	3312	1241
17	IV	8.1	34.3	1238	418	413	58.5	55.6	33.7	316	4.1	21.3	2120	66.5	−0.4	3297	1110
18	III	7.8	23.7	688	114	493	8.4	53.5	26.3	155	1.6	21.1	3120	89.5	0.1	3832	622
19	III	7.8	24.3	608	29.8	621	8.8	45.7	21.6	166	1.5	22.2	730	55.7	0.1	1387	601
20	III	8	18.2	19,962	9608	401	1190	887	687	4438	37.4	23.3	49	0.5	−7.6	17,117	17,068
21	III	7.9	18.8	19,718	9740	382	1203	533	534	5454	44.7	23.7	130	0.3	2.9	17,851	17,720

* TDScalc without considering Fe and Mn concentrations.

, while isotopic data are compiled in

Table 2. The isotopic data in the study by Xu et al. [10]. We have added deuterium excess (D_ex) values.

ID	Aquifer System	Membership	δ2H	δ18O	Dex	3H	14C Based Age	14C Activity
			(‰)	(‰)	(‰)	TU	Years BP	pmc
1	DCAS	Aquifer III	−42.94	−5.25	−0.94	<2	17,940 ± 140	9.70
2	DCAS	Aquifer III	−36.58	−3.97	−4.82	<2	11,270 ± 110	21.75
3	DCAS	Aquifer III	−41.84	−5.25	0.16	15.8	18,490 ± 340	9.08
4	DCAS	Aquifer III	−43.8	−5.52	0.36	<2	16,100 ± 100	12.12
5	DCAS	Aquifer III	−39.92	−5.17	1.44	<2	17,260 ± 100	10.54
6	DCAS	Aquifer IV	−45.13	−5.72	0.63	<2	18,200 ± 130	9.40
7	DCAS	Aquifer IV	−39.92	−5.13	1.12	<2	16,140 ± 130	12.07
8	DCAS	Aquifer III	−44.9	−5.97	2.86	<2	26,140 ± 300	3.60
9	DCAS	Aquifer III	−46.25	−6.61	6.63	2.47	22,950 ± 510	5.29
10	DCAS	Aquifer III	−40.79	−4.86	−1.91	<2	20,380 ± 280	7.22
11	DCAS	Aquifer III	−46.66	−6.48	5.18	<2	24,880 ± 260	4.19
12	DCAS	Aquifer III	−49.12	−6.38	1.92	<2	22,150 ± 250	5.83
13	DCAS	Aquifer IV	−50.33	−6.34	0.39	<2	18,870 ± 230	8.67
14	DCAS	Aquifer IV	−48.8	−6.37	2.16	6.94	22,310 ± 280	5.72
15	DCAS	Aquifer IV	−55.31	−7.22	2.45	5.31	16,900 ± 230	11.01
16	DCAS	Aquifer IV	−48.05	−6.4	3.15	8.7	19,540 ± 180	8.00
17	DCAS	Aquifer IV	−52.76	−6.97	3	<2	25,890 ± 300	3.71
18	DCAS	Aquifer III	−41.64	−4.54	−5.32	13.3	22,490 ± 170	5.60
19	DCAS	Aquifer III	−51.42	−5.96	−3.74	12	18,600 ± 170	8.96
20	DCAS	Aquifer III	−28.2	−3.67	1.16	<2	7410 ± 800	34.69
21	DCAS	Aquifer III	−28.15	−3.25	−2.15	<2	9370 ± 80	27.36
22	None	River	−50.45	−7.18	6.99		-	-
23	None	River	−50.43	−6.98	5.41		-	-
24	None	River	−28.35	−2.96	−4.67		-	-
25	None	Sea	−6.32	−0.56	−1.84		-	-
26	None	Sea	−14.52	−1.73	−0.68		-	-
27	PAS	Phreatic Aquifer	−38.67	−5.71	7.01		-	-
28	PAS	Phreatic Aquifer	−31.13	−4.4	4.07		-	-
29	PAS	Phreatic Aquifer	−30.69	−3.91	0.59		-	-
30	ICAS	Aquifer I	−33.37	−4.04	−1.05		-	-
31	PAS	Phreatic Aquifer	−42.83	−6.39	8.29		-	-
32	PAS	Phreatic Aquifer	−37.9	−5.28	4.34		-	-
33	PAS	Phreatic Aquifer	−38.63	−5.46	5.05		-	-

. In the latter, we also incorporated the deuterium excess (D_ex) value for each sample, calculated using the equation mentioned by Dansgaard [22]: D_ex = δ²H − 8 × δ¹⁸O. This addition provides further insight into the isotopic composition and potential evaporation effects influencing groundwater evolution.

Mao et al. [23] presented new data (including TDS, pH, Cl⁻, SO₄²⁻, HCO₃⁻, Na⁺, Ca²⁺, Mg²⁺, K⁺, Br, Sr, B, δ²H, δ¹⁸O, ⁸⁷Sr/⁸⁶Sr, and δ¹¹B) for samples in a broader area of the Jiangsu Coastal Plain. From these, only 12 samples were located within our study area, 2 of them in the DCAS, 9 in the PAS/ICAS, and 1 river/estuary sample. All the samples including ⁸⁷Sr/⁸⁶Sr, and δ¹¹B were located outside our study area. As our assessment was focused in the DCAS only, we use Mao et al.’s [23] results for comparison in later sections but did not include them in our dataset. We conducted a QA/QC assessment of the data used in this study, originally presented by Xu et al. [10]. As a first step, we calculated the Charge Balance Error (CBE) for the 21 samples with available hydrochemical data, as shown in Table 1. The CBE for 20 out of the 21 samples was below 5%, while one sample exhibited an error between 5% and 10%. Based on these CB error values, all samples were deemed to be of sufficient quality for use in this assessment, following the criteria established by Güler et al. [7] and Guggenmos [8].

Additionally, we compared laboratory-measured total dissolved solids (TDS) with calculated TDS (TDScalc; Table 1), prompted by the detection of unusually high Fe and Mn concentrations in certain samples. In natural waters, TDS can be estimated if the concentrations of major ions are known (e.g., Hem [5]). To derive TDScalc, we applied the methodology outlined by Hounslow [6].

The discrepancy between laboratory-measured TDS and TDScalc was within 15% for 15 out of the 21 samples, which was considered acceptable. However, for the remaining six samples, the difference ranged from 74% to 457%, indicating anomalous concentrations within the dataset. A closer examination of the laboratory-reported values from Xu et al. [10] confirmed that these six samples exhibited unusually high Fe and Mn concentrations. To assess whether Fe and Mn contributed to the observed discrepancies in TDS, we recalculated TDScalc, this time excluding Fe and Mn from the computation. The revised calculations yielded a discrepancy of less than 15% across all samples, strongly suggesting errors in the reported Fe and Mn concentrations for these six samples.

Given this inconsistency, we excluded Fe and Mn concentrations from the statistical analysis, as well as from further data processing and interpretation.

This inconsistency between TDS and TDScalc leading to Fe and Mn data exclusion was not acknowledged by CBE, highlighting the need to use this tool in combination with other QA/QC techniques. This was also due to the fact that CBE was calculated using major element concentrations only, which is the common practice, but it would be recommended to use all ions concentration per CBE calculation instead.

Before conducting HCA and PCA, all parameters (except for pH) were log-transformed to approximate a normal distribution, following the results of a normality test, which indicated that only pH was normally distributed.

Hierarchical cluster analysis (HCA) and principal component analysis (PCA) were employed to identify patterns within the hydrochemical dataset. These multivariate techniques only consider samples where each input parameter has a numerical value. To ensure the inclusion of all 21 samples in both analyses, the selected input parameters were pH, TDS, Na⁺, Ca²⁺, Mg²⁺, K⁺, HCO₃⁻, Cl⁻, and SO₄²⁻. Fe and Mn were excluded, as their concentrations in six samples were deemed to be erroneous based on the QA/QC assessment, as previously discussed.

HCA was performed using Ward’s linkage method, a widely applied algorithm in hydrochemical studies (e.g., Güler et al. [7]). This method relies on an analysis of variance to separate clusters based on data homogeneity. The measure of similarity employed at both clustering stages was the Euclidean distance, ensuring the equal weighting of each variable (e.g., Davis, [24]). This approach to HCA has been utilized in several studies, including [8,9,25,26].

PCA, a mathematical technique used for data reduction and pattern recognition [27,28,29], was applied in this study to generate scatter plots representing the clusters identified in HCA.

The carbon-14 data presented by Xu et al. [10] was only available in the form of corrected years before present (BP), which, according to the authors, was derived using the Vogel correction method. However, no “raw” data was provided. For enhancing transparency and providing a less ambiguous ¹⁴C measure, we calculated the ¹⁴C activity back to a percentage of modern carbon (pmc) using Equation (1):

$$Carbon-14\left(pmc\right)=85e^{-{\displaystyle \frac{t}{8267}}}$$

(1)

where t corresponds to the time (y BP) calculated for ¹⁴C data by Xu et al. [10]. The methodological differences between Xu et al. [10] and this study are summarized in

Table 3. Methodological differences between Xu et al. [10] and this study.

Methodology Steps and Results	This Study	Xu et al. [10]
Data QA/QC before statistical analysis	Yes	No
Dataset log-transformation and/or normality test	Yes	No
Used anomalous Fe and Mn data in statistical process	No	Yes
HCA linkage method	Ward
HCA measure of similarity	Euclidean distance
Number of identified clusters	3	4
Number of significant components identified (Eigenvalue > 1)	2	3

.

4. Results

The hydrochemical data allowed for the identification of two primary water types: Na⁺–HCO₃⁻ water and Na⁺–Cl⁻ water. The first type is associated with freshwater, exhibiting TDS values ranging from 462 mg/L to 723 mg/L, while the second type shows TDS values varying from 852 mg/L to 19,718 mg/L. This high variability is primarily influenced by two highly anomalous samples (Samples 20 and 21). When these outliers are excluded, the maximum TDS value for Na⁺–Cl⁻ waters is 2479 mg/L. Mao et al. [23] presents two groundwater sample in the DCAS within the study area that present TDS values in agreement with the data presented here.

There is no clear distinction in water type between Aquifer III and Aquifer IV, as both aquifers contain samples representing each water type within the DCAS. This suggests that similar hydrochemical processes are influencing both aquifers. However, this observation may not be applicable to the two anomalous samples, as both belong exclusively to Aquifer III.

4.1. HCA

This procedure led to the identification of three distinct clusters, represented in the dendrogram (Figure 2), with their mean values summarized in

Table 4. Mean values of hydrochemical data for clusters/subclusters identified by HCA.

Cluster	pH	TDS	Cl−	HCO3−	SO42−	Ca2+	Mg2+	Na+	K+	H2SiO3	Water Type
		mg/L	mg/L	mg/L	mg/L	mg/L	mg/L	mg/L	mg/L	mg/L
C1	7.81	698	136.3	480.3	14.1	55.1	25.2	165.1	2.1	25.3	Na-HCO3
sC1a	7.72	545	51.8	486.8	3.8	64.2	27.4	98.3	1.6	32.3	Na-HCO3
sC1b	7.88	784	169.9	487.2	22.4	48.3	23.6	202.0	2.3	19.3	Na-HCO3
C2	7.9	1479	656.2	332.3	62.3	113.8	49.2	339.4	5.2	17.1	Na-Cl
C3	7.95	19840	9674	391.5	1196.5	710	610.5	4946	41.1	23.5	Na-Cl

.

Cluster 1 (C1), consisting of 12 samples, is dominated by Na⁺–HCO₃⁻ water and exhibits the lowest mineralization among all groups, with a mean TDS of 698 mg/L, classifying it as freshwater. Within C1, two subclusters (sC1a and sC1b) were identified: sC1a (five samples) had a lower mean TDS of 545 mg/L, while sC1b (seven samples) presented a slightly higher mean TDS of 784 mg/L. The main differences between these subclusters lie in their ion composition, with sC1a exhibiting higher Ca²⁺ concentrations and lower Cl⁻, SO₄²⁻, and Na⁺ levels compared to sC1b, while the other chemical species remain relatively similar.

Seven groundwater samples form Cluster 2 (C2), which is characterized by the Na⁺–Cl⁻ water type with a mean TDS of 1479 mg/L, placing it in the brackish water category. The two samples with the highest mineralization belong to Cluster 3 (C3), with an exceptionally high mean TDS of 19,840 mg/L, classifying them as saline water. The TDS levels in C3 are at least one order of magnitude higher than in C2, highlighting their anomalous nature within the dataset.

A spatial trend is observed only in sC1a, where four out of the five samples originate from Aquifer III and are located near the Rutai Canal river mouth. However, samples from sC1b, C2, and C3 do not exhibit clear spatial relationships, with roughly half of them distributed between Aquifer III and Aquifer IV. This indicates that aquifer membership and location are not significant factors influencing hydrochemical processes for these clusters. Both C3 samples belong to Aquifer III and are spatially close to each other, positioned between Matang and Yoco Port. Their location coincides with a zone where groundwater flow shifts eastward, attributed to a depression cone detected through water table measurements in Aquifer III.

4.2. PCA

Only two principal components (PCs) with Eigenvalues greater than 1 were considered for further interpretation, following Kaiser’s criterion [9,30]. PC1 accounts for 65.8% of the total variance in the dataset (

Table 5. Table of component weightings, principal component Eigenvalues, and variance of principal components.

Variable	PCA1	PCA2
pH	0.12	−0.41
Log TDS	0.38	0.10
Log Cl−	0.37	−0.11
Log HCO3−	−0.16	0.52
Log SO42−	0.36	−0.02
Log Ca2+	0.35	0.26
Log Mg2+	0.36	0.21
Log Na+	0.38	0.02
Log K+	0.38	0.01
Log H2SiO3	−0.06	0.65
Eigenvalue	6.58	1.76
Percentage of explained variance	65.8	17.6
Cumulative percentage of variance	65.8	83.4

) and exhibits positive weightings for TDS, Cl⁻, SO₄²⁻, Ca²⁺, Mg²⁺, Na⁺, and K⁺, which are associated with mineralization and are likely linked to Na⁺–Cl⁻ waters. In contrast, pH, HCO₃⁻, and H₂SiO₃ have weightings close to zero, indicating a minimal influence on this component.

PC2 explains 17.6% of the total variance (Table 5) and is primarily associated with positively correlated weightings of HCO₃⁻ and H₂SiO₃ while showing a negative correlation with pH. This suggests a relationship with carbonate speciation and/or water–rock interactions, making it more relevant for Na⁺–HCO₃⁻ water types. The first two principal components together account for approximately 83.4% of the total variability in the dataset, indicating that they capture most of the hydrochemical variance.

Figure 3 shows the distribution of values on PC1 and PC2 for all the groundwater samples, grouped by cluster. Only clusters C2 and C3 have positive weightings on PC1, representing the most mineralized groundwater in the dataset. Although both clusters are highly mineralized, they are clearly separated along the PC1 axis.

In contrast, the samples in C1 (including its two subclusters) exhibit negative weightings in PC1, indicating their association with the freshest groundwater. Regarding PC2, sC1a and C3 show exclusively positive weightings, while sC1b and C2 display both positive and negative weightings. Notably, a subset of C2 samples exhibits the most negative weightings in PC2, which corresponds to the highest pH values recorded in the dataset and represents Na⁺–HCO₃⁻ water types.

4.3. Environmental Isotopes

In addition to the 21 groundwater samples from DCAS, the environmental isotope data also include samples from the PAS (6 samples), the ICAS (1 sample), and surface waters (5 samples). The δ²H and δ¹⁸O compositions in the DCAS range from −55.31 ‰ to −28.15 ‰ and from −7.22 ‰ to −3.25 ‰, respectively. Similarly, the δ²H and δ¹⁸O compositions in the PAS range from −42.83 ‰ to −30.69 ‰ and from −6.39 ‰ to −3.91 ‰, respectively.

Figure 4 presents a δ²H versus δ¹⁸O diagram, including all groundwater and river water samples. It also displays the Global Meteoric Water Line (GMWL; δ²H = 8 × δ¹⁸O + 10; Craig [31]) and the Local Meteoric Water Line (LMWL; δ²H = 8.49 × δ¹⁸O + 17.71; [32]). The higher intercept of the LMWL compared to the GMWL suggests a more humid and rainy climate in the coastal area [10]. All samples lie to the right of both the GMWL and LMWL, indicating that the groundwater has undergone some degree of evaporation before recharge [33].

Additionally, Figure 4 introduces local regression lines (LELs) for the PAS and the DCAS, with equations of δ²H = 5.2 × δ¹⁸O − 9.6 and δ²H = 6.2 × δ¹⁸O − 9.5, respectively. Both lines exhibit a lower slope compared to the GMWL and LMWL, while their intercepts remain similar.

D_ex has been widely used to assess waters affected by evaporation [33,34,35]. In this study, the D_ex values (Table 2) for PAS waters range from 0.59‰ to 8.29‰, while those for the DCAS show greater variability, from −5.32‰ to 6.63‰, mainly due to Aquifer III waters, as the values for Aquifer IV range from 0.39‰ to 3.15‰. The single ICAS sample has a D_ex of −1.05‰. All these values are lower than the D_ex for global (10‰; Craig [31]) and local precipitation (17.71‰; Xu et al. [32]), further confirming that evaporation has influenced groundwater, as D_ex decreases during this process [30].

4.4. Radioactive Isotopes

Tritium data is presented in Table 2. Most samples (14 out of 21) have ³H values below the detection limit (<2 TU), suggesting groundwater residence times predating the 1960s. It is important to note that this detection limit is relatively high, as many laboratories worldwide now report ³H detection limits lower than those used in this study. Some laboratories, such as GNS New Zealand, register detection limits as low as 0.02 TU [36]. The difference between this lower detection limit and that reported by Xu et al. [10] spans several half-life cycles, meaning that only very young water can be reliably traced with the available data.

Among the seven samples with ³H values above the detection limit, three (samples 3, 18, and 19, all from Aquifer III) show values above 10 TU. Three others (samples 14, 15, and 16, all from Aquifer IV) register between 5 TU and 10 TU, while one sample (sample 9, from Aquifer III) records 2.47 TU. These results indicate that these seven samples have undergone active recharge within the last 60 years, with higher ³H values corresponding to younger groundwater.

Carbon-14 data is also presented in Table 2. The data are reported in years before present (y BP) rather than as a percentage of modern carbon (pmc), making a more detailed evaluation of the original dataset from Xu et al. [10] unfeasible. The authors applied the Vogel correction model to estimate average groundwater residence times, reporting ages within the method’s applicable range, from 7410 ± 800 y BP to 26,140 ± 300 y BP, suggesting ancient recharge. However, this interpretation contradicts the findings for the seven samples with detectable ³H, which indicate recent recharge.

5. Discussion: The Processes Affecting Groundwater Evolution in the Jiangsu Coastal Plain

5.1. Evaporation Processes

As shown in Figure 4, all the samples are shifted to the right relative to both the GMWL and LMWL, indicating that evaporation occurred before recharge. Additionally, the D_ex values for all samples are lower than those characteristic of global and local precipitation, further supporting the likelihood of an evaporation process. Two Local Evaporation Lines (LELs) are presented in Figure 4, one for PAS waters and another for DCAS waters, with slopes of 5.2 and 6.2, respectively. These slopes are lower than those of the GMWL and LMWL, which is indicative of evaporation; however, they remain higher than the threshold assigned for intense evaporation conditions (<5; Clark [37]).

Xu et al. [10] also calculated LELs for these two aquifer systems. While we obtained a similar equation and slope for DCAS waters, a notable difference was observed for PAS waters, where Xu et al. [10] reported an LEL equation of δ²H = 4.69 × δ¹⁸O − 12.1. This discrepancy arises from their inclusion of one sample from Aquifer I (ICAS) alongside Phreatic Aquifer (PAS) samples to define their LEL for the shallow system. According to our findings, this approach is not appropriate, as groundwater from Aquifer I, being confined, likely represents distinct recharge and hydrochemical conditions. Consequently, grouping these samples for LEL calculations is not advisable. Given the lack of hydrochemical data for the PAS samples, further discussion of evaporation will focus exclusively on DCAS waters.

Stable isotopes of oxygen and hydrogen can help differentiate between evaporation, which fractionates these isotopes, and other non-fractionating processes [35,38,39]. Figure 5a presents a Dex versus TDS diagram of DCAS waters, categorizing samples by their dominant water type. Samples 20 and 21 (C3) are excluded due to their exceptionally high TDS values, which would obscure other observable trends, and because the main factor controlling C3 hydrochemistry is different. Na⁺–HCO₃⁻ waters exhibit minimal mineralization variability and no clear relationship with Dex (r 0.45), suggesting that evaporation is not a significant factor in their mineralization. In contrast, most Na⁺–Cl⁻ waters follow a pattern where more negative Dex values correspond to higher TDS (r 0.74), indicating an evaporation-related influence on mineralization.

A similar trend is evident in Figure 5b (δ¹⁸O vs. Cl⁻), which presents data for DCAS samples, where chloride concentrations increase alongside δ¹⁸O fractionation in Na⁺–Cl⁻ waters (r 0.85), whereas no clear association is observed for Na⁺–HCO₃⁻ waters (r 0.17). This confirms that, unlike Na⁺–HCO₃⁻ waters, evaporation is responsible for the observed mineralization increase in Na⁺–Cl⁻ waters, with the exception of C3 samples. This pattern applies to all C2 samples and nearly half of the sC1b samples (three out of seven).

Another study examined aquitard porewater in the same Jiangsu Coastal Plain area studied by Xu et al. [10], analyzing shallow porewater (PAS equivalent), intermediate porewater (ICAS equivalent), and deep porewater (DCAS equivalent). Their findings also indicated that evaporation affected most porewaters, as reflected in the δ²H and δ¹⁸O data. Also, Mao et al. [23] identified evapo-concentration as the main trigger controlling higher TDS concentrations, which agrees with our findings.

5.2. Water–Rock Interaction in the DCAS

PCA allowed the identification of two principal components, highlighting distinct hydrochemical groups among the DCAS water types. Figure 6 presents the projection of scores in PC1 and PC2, categorized by water type. While Na⁺–Cl⁻ waters are primarily associated with PC1, Na⁺–HCO₃⁻ waters are more closely linked to PC2. Notably, seven of the nine Na⁺–HCO₃⁻ waters that exhibit positive weightings in PC2 also register the highest H₂SiO₃ (silicic acid) concentrations in the dataset.

When carbonate species and dissolved silicic acid are present, they contribute to rock weathering and mineral dissolution. This interaction is described by the reaction in Equation (2):

Rock + CO₂(aq) + 2 × H₂O = cations + clay + HCO₃⁻ + H₂SiO₃(aq)

(2)

According to carbonate speciation hydrochemistry, dissolved CO₂ remains in solution when the pH is at or below approximately 8 [9,40]. For instance, in sample no. 13 (pH = 8.3), dissolved CO₂ is estimated at about 5 mg/L, whereas in sample no. 3 (pH = 7.4), it is around 66 mg/L. As per carbonate equilibrium principles, decreasing pH leads to increased dissolved CO₂ concentrations. However, as noted by Taulis and Milke, the actual dissolved CO₂ may exceed these estimates, as samples often experience CO₂ degassing during collection [40]. Given that all these samples contain dissolved CO₂, their interaction with rocks will lead to the release of HCO₃⁻ and H₂SiO₃, as described in Equation (2). This indicates that the dominant hydrochemical process affecting Na⁺–HCO₃⁻ waters in the DCAS is water–rock interaction, which accounts for all sC1a samples and four out of seven sC1b samples.

Li et al. [41] demonstrated that water–rock interaction plays a significant role in deep porewater hydrochemistry (DCAS equivalent) in the Jiangsu Coastal Plain, based on ⁸⁷Sr/⁸⁶Sr ratios and ionic composition. They identified ⁸⁷Sr/⁸⁶Sr values ranging from 0.7094 to 0.7112, a range higher than the average for seawater, supporting the silicate weathering of clays. These results are also in agreement with Mao et al. [23], who acknowledged water–rock interaction as one of the factors controlling TDS in the DCAS.

5.3. Seawater Intrusion in the DCAS

HCA revealed one cluster (C3) characterized by saline waters, with a mean TDS of 19,840 mg/L. This significantly higher TDS compared to other clusters in the DCAS suggests a distinct mineralization process, even though the water type remains the same as in C2. Given that seawater typically has a TDS of approximately 35,000 mg/L, the salinity of C3 falls between freshwater and seawater. If considering a simple mixing scenario between freshwater and seawater, the contribution of seawater to C3 would be slightly above 50%. PCA further confirmed that the high mineralization of C3 waters could not be explained by the processes affecting the rest of the dataset, such as evaporation or water–rock interaction.

The spatial position of the two wells (20 and 21) associated with C3 provides additional context. These wells are located in an area where a localized shift in groundwater flow occurs in Aquifer III (DCAS) due to a depression cone identified near Matang (Figure 1a). In this area, groundwater in Aquifer III flows inland from the shoreline toward Matang. This hydrodynamic condition creates a plausible scenario for modern or ongoing seawater intrusion.

Figure 7 presents an δ²H versus δ¹⁸O diagram for DCAS samples, categorized by aquifer membership, along with river and seawater samples. The samples can be grouped into three categories. Group 1 includes seawater samples, with δ²H values ranging from −14.52‰ to −6.32‰ and δ¹⁸O values from −1.73‰ to −0.56‰, which are consistent with seawater isotopic compositions recorded in water from the Yellow Sea by Li et al. [41]. Group 3 consists of DCAS and river water samples, with δ²H values between −55.31‰ and −36.58‰ and δ¹⁸O values between −7.22‰ and −3.97‰. Group 2 represents an intermediate composition, suggesting a mixture of Groups 1 and 3, with δ²H values from −28.35‰ to −28.15‰ and δ¹⁸O values from −3.67‰ to −2.96‰. Notably, two of the three samples within Group 2 correspond to C3 waters. In addition, the third sample (24; Figure 1) in this group corresponds to a river sample, located closed to the shoreline, indicating a likely seawater tidal influence and emphasizing that the isotopic character of this group may be affected by seawater. All of this further supports the hypothesis that seawater intrusion is the primary process influencing this group’s salinity. In groundwater, this process appears to be restricted to Aquifer III.

The third sample in Group 2 is a river sample, which is the closest to the shoreline, located only 4 km from the coast. This suggests potential tidal influence or seawater mixing at this location. However, unlike C3, this hypothesis cannot be confirmed through hydrochemistry, as Xu et al. [10] did not provide detailed chemical data for river waters. Despite the evidence, Xu et al. [10] ruled out modern seawater intrusion as a cause of salinization in Aquifer III (corresponding to C3) based on residence time estimates, and provided an ambiguous discussion regarding past seawater intrusion. They attributed the high salinity in these samples to either evaporation under high temperatures, based on δ²H and δ¹⁸O data, or terrestrial salinization under arid Holocene Hypsithermal conditions, based on Na/Cl and Mg/Ca ratios as well as residence time estimations.

While evaporation likely plays a role, it does not appear to be the dominant process affecting C3 samples. The D_ex values for these two samples are 1.16‰ and −2.15‰, which are close to the D_ex values of seawater samples (−1.84‰ and −0.68‰). In contrast, three other Aquifer III samples (samples 2, 18, and 19) exhibit lower D_ex values than the C3 samples, suggesting a higher degree of evaporation in those samples. Consequently, evaporation alone cannot explain the high mineralization observed in C3 waters, reinforcing the likelihood of seawater intrusion as the primary process (further discussed in Section 5.4, Residence Times in the DCAS).

Seawater intrusion has previously been identified in groundwater in the shallow PAS/ICAS systems of the Jiangsu Plain by Mao et al. [23], and it has been identified in porewater by Li et al. [41], but it has not been acknowledged in the DCAS. In addition, according to Mao et al. [23] the DCAS TDS is explained by a combination of water–rock interaction and inter-aquifer mixing between the DCAS and the more saline PAS/ICAS. We could not identify the latter, as our dataset did not include hydrochemical data in the PAS/ICAS. Nevertheless, as Mao et al. [23] confirm seawater intrusion into the PAS/DCAS, whether inter-aquifer mixing occurs may also explain how local seawater intrusion may partially reach the DCAS.

Additionally, seawater intrusion has been documented by Zhang et al. in groundwater along the coast of Shandong Province, located north of Jiangsu, and more extensively in the Laizhou Bay and Bohai Sea areas of northern China [17]. In the latter area, the aquifer architecture is similar to that in the Jiangsu Coastal Plain, where a deep unconsolidated sedimentary sequence, composed of an aquifer/aquitard intercalation occurs, recognizing four main aquifer systems in depth. The similarities in aquifer architecture also help to confirm that seawater intrusion is plausible if the already mentioned conditions occur.

Based on this analysis, we conclude that two samples (20 and 21, or C3) within the dataset exhibit distinct evidence of seawater intrusion. This process appears to be restricted to Aquifer III of the DCAS and has currently not affected Aquifer IV of the DCAS, based on the available data.

5.4. Residence Times in the DCAS

Groundwater residence time calculation is a highly sensitive process. Xu et al. [10] mentioned the use of the Vogel model for age correction but did not specify the rationale or criteria for selecting this method, which applies a 0.85 correction factor for age estimation. Furthermore, they did not report ¹⁴C activity in pmc or δ¹³CDIC data, which would be essential for constraining carbon sources and refining age estimates. As previously noted, HCO₃⁻ is the dominant anion in freshwater, and its concentration increases in groundwater primarily due to water–rock interaction. This indicates that additional carbon sources beyond soil-derived carbon contribute to the system, making the assumption of a uniform 0.85 correction factor, as in the Vogel model, questionable for this study area.

Other publications, such as Suckow [42], have emphasized the limitations of groundwater age determinations due to geological complexity and the inherent assumptions in age estimation models. Instead, they recommend reporting only ¹⁴C activities or concentrations without interpreting absolute ages. Additionally, emerging approaches suggest plotting measured ¹⁴C values in pmc against distance along an assumed flow path when age estimation is uncertain [42,43]. Given these considerations, the age estimates provided by Xu et al. [10] are likely overestimated. Because of this, providing groundwater age calculations using uncorrected laboratory data and making paleoclimatic interpretations based on them is not advisable.

Another inconsistency in the recharge age estimations made by Xu et al. [10] is the contradiction between the ³H and ¹⁴C data. Seven samples were identified as having recent recharge based on ³H values, yet their ¹⁴C age estimates range from 17,000 to 23,000 years BP. Several factors could explain such discrepancies, including contamination during sampling, improper well sealing leading to leakage and mixing within boreholes, or inter-aquifer mixing. The latter was identified by Mao et al. [23] as one of the causes that controls TDS in DCAS waters. This could be a cause that explains the ³H and ¹⁴C discrepancy, but it is not necessarily the only one. This inconsistency in the data further underscores the unreliability of the paleoclimatic recharge estimations presented by Xu et al. [10].

Due to the limitations in the radioactive isotope dataset, it is not possible to confidently determine whether the seawater intrusion identified in C3, wells 20 and 21, represents a modern or historical process. These samples have ³H values below the detection limit, although, as previously discussed, this detection limit is too high to effectively constrain young water. On the other hand, Xu et al. [10] reported apparent ages of 7410 ± 800 years BP for sample 20 and 9370 ± 80 years BP for sample 21, making them the youngest groundwater samples based on ¹⁴C alone. This implies that these samples have the highest pmc values among all the measured waters, which, if ¹⁴C is considered in isolation, would indicate relatively recent genesis.

The location of these two wells within an area where groundwater flow shifts due to a depression cone and moves from the ocean toward Matang suggests that seawater intrusion could be an ongoing process. However, due to the inconsistencies and uncertainties in the available ³H and ¹⁴C data, this interpretation remains inconclusive. This limitation may be overcome by using additional tracers like the Cl/Br ratio or the ⁸⁷Sr/⁸⁶Sr isotopic ratio.

6. Conclusions

The main conclusions from the assessment and reanalysis of the data for the DCAS in the Jiangsu Coastal Plain presented by Xu et al. [10] are as follows:

• A QA/QC analysis identified anomalous Fe and Mn concentrations in five samples, which were inconsistent with TDS values. As a result, the Fe and Mn concentrations were discarded for these samples. Consequently, these elements were not included as input data for HCA and PCA, as reliable information was not available for all samples. This underscores the importance of conducting thorough QA/QC procedures to ensure data integrity.
• Charge Balance Error (CBE) is the tool used most usually for laboratory data QA/QC. Even though it is an excellent first approach, it is advisable to use it in combination with additional QA/QC techniques. In the analyzed dataset in this study, CBE validated samples that were later discarded for anomalous Fe and Mn concentrations. In order to take better advantage of CBE it is more helpful to evaluate it using it all ions and not only major ones.
• Three key processes were identified as drivers of hydrochemical variations in the deep groundwater of the Jiangsu Coastal Plain: water–rock interaction, evaporation, and seawater intrusion. Water–rock interaction was found to be the dominant hydrochemical process influencing Na⁺–HCO₃⁻ waters (freshwater). Evaporation was determined to be the primary process responsible for mineralization in Na⁺–Cl⁻ waters (brackish water).
• The occurrence of seawater intrusion was demonstrated in two wells (20 and 21, classified as saline waters). This conclusion is based on a combination of a statistical assessment (HCA and PCA) of hydrochemical data and environmental isotopes (δ²H and δ¹⁸O plots, along with D_ex values).
• Seawater intrusion was identified as affecting only Aquifer III of the DCAS. Given the location of the two affected wells, where groundwater flows inland from the shoreline towards drawdown cones induced by groundwater extraction, this process may be considered ongoing. However, this finding remains inconclusive due to contradictory radiogenic isotope data. It is advisable to further study the DCAS, including a more comprehensive analytical suite (including most dissolved anions and cations) plus additional isotopes like ⁸⁷Sr/⁸⁶Sr that may help to better refine this process.
• Based on these observations, it is highly advisable to account for seawater intrusion in groundwater management strategies for the area. Step drawdowns are recorded in the Matang area. Hence, a detailed monitoring program for groundwater levels and water quality using telemetry for live data and probably reducing extractions are good first steps.
• Residence time estimations revealed contradictions in seven samples, representing 33% of the dataset. These samples contained measurable amounts of ³H (with TU values above the detection limit), suggesting a recharge age of less than 60 years. However, the same samples were assigned corrected ¹⁴C ages ranging from 17,000 to 23,000 years BP. Due to this inconsistency, paleoclimatic reconstructions based on these data cannot be deemed reliable.