Hydrogeochemical Controls on δ¹³C-DIC in a Dolomite Cave: Insights from 7-Year Monitoring at Mahuang Cave, Southwest China

Abstract

Differences in the dissolution kinetics between dolomite and limestone constrain both the understanding of regional carbon cycling and the precision of speleothem-based paleoclimate reconstructions. To elucidate the mechanisms by which hydrogeochemical processes in a dolomite cave regulate dissolved inorganic carbon isotopes (δ¹³C-DIC), this study analyzed hydrochemical parameters and δ¹³C-DIC of five hydrological flow types, coupled with soil and cave CO₂ and their isotopic compositions, based on monitoring data from 2019 to 2025 at Mahuang Cave (dolomite) in southwestern China, and compared them with six limestone caves. The results show that the hydrochemistry of Mahuang Cave is of the Ca·Mg-HCO₃ type, with major ions exhibiting higher concentrations in the warm season and lower in the cold season, while δ¹³C-DIC is lighter in the rainy season and heavier in the dry season. On the interannual scale, δ¹³C-DIC shows only a weak correlation with ENSO phase variability, suggesting that the direct modulation of ENSO signals on the cave system is relatively limited. Water–rock–gas interactions and hydrological flow paths regulate the spatial gradient of δ¹³C-DIC (approximately 2.7‰): rapid fracture flow transmits light carbon signals of soil CO₂ in an open system, slow matrix flow enriches heavy isotopes through enhanced prior calcite precipitation (PCP) in a near-closed system, and the underground river exhibits decoupling between Sr/Ca and δ¹³C-DIC due to multi-source mixing. Compared with limestone caves, Mahuang Cave is characterized by high soil CO₂, low cave CO₂, and heavier δ¹³C-DIC, which is attributed to the slow dissolution of dolomite, inhibited CO₂ transport through thick overlying strata, and limited degassing due to low cave temperature. This study provides empirical evidence for understanding carbon cycling mechanisms in dolomite regions and emphasizes the need to account for lithological constraints.

1. Introduction

Chemical weathering of carbonate rocks constitutes a major carbon sink in the global carbon cycle, with karst regions contributing significantly [1,2]. Cave drip water links surface climate with subsurface sedimentary records [3], and its DIC concentration and δ¹³C-DIC are key proxies for tracing carbon sources and reconstructing paleoenvironments [4,5,6]. The hydrogeochemical processes and driving mechanisms of carbon isotope fractionation in limestone karst systems have been relatively well elucidated [7,8,9,10]. However, due to the presence of Mg²⁺ and slower dissolution kinetics, dolomite exhibits water–rock interaction (WRI) pathways, ion release patterns, and saturation state evolution that differ from those of limestone [11,12], rendering the pathways of DIC production and consumption more complex [4,8,13,14] and introducing additional uncertainties into carbon flux estimation and speleothem-based paleoclimate interpretation [2,6]. Therefore, investigating the mechanisms controlling δ¹³C-DIC in dolomite caves is an essential step toward understanding and refining karst carbon cycle models.

In recent years, significant progress has been made in understanding the hydrogeochemistry and carbon transport in karst caves. Long-term monitoring has confirmed that surface vegetation restoration can increase ecosystem carbon flux, leading to higher DIC concentrations and more negative δ¹³C-DIC values in drip water [15]. Soil CO₂, as the initial carbon source for drip water δ¹³C-DIC, has its supply flux modulated by regional surface environmental conditions: at the seasonal scale, elevated soil CO₂ concentrations in summer enhance carbonate dissolution and transmit more negative δ¹³C-DIC signals to drip water [16,17]. At the interannual scale, it is sensitive to regional climate variability (e.g., large-scale climate models such as ENSO [18,19]). Different hydrological pathways and flow rates not only affect WRI time, degassing efficiency, and the magnitude of isotope fractionation [20], but also give rise to differences in paleoclimate signals preserved in speleothems [21]. Fast drip flow responds rapidly but undergoes incomplete dissolution, whereas slow drip flow experiences prolonged WRI, yet mixing effects in the vadose zone can induce signal lags [7]. Conventional studies generally hold that fast-drip water has more negative δ¹³C values and slow-drip water more positive δ¹³C values. However, monitoring at Furong Cave has revealed that when “old carbon” from bedrock dissolution dominates, fast drip water can exhibit elevated δ¹³C-DIC values, while slow drip water may instead show more negative values due to mixing with soil organic carbon [22]. Before reaching the cave floor, karst water may undergo PCP when it becomes supersaturated with respect to calcite, particularly if pCO₂ changes along its flow path or in the cave environment. Cave ventilation enhances CO₂ degassing and PCP intensity by lowering cave-air pCO₂, thereby collectively modulating the δ¹³C-DIC signals of drip water [23]. Under low pCO₂ conditions in winter, rapid degassing leads to significantly more positive δ¹³C-DIC; consequently, speleothems primarily record winter information during ventilation periods and summer information during stagnation periods [18,24].

The Mg/Ca and Sr/Ca ratios are sensitive to hydrological changes and serve as effective proxies for identifying PCP, and indicating wet and dry conditions [25]. WRI jointly controls these ratios [26], PCP processes [14], and overlying lithology [27]. During PCP, Ca²⁺ preferentially partitions into the solid phase, leaving the residual water enriched in Mg/Ca and Sr/Ca. Long-term observations from Furong Cave [28,29], Shihua Cave [30], Jiguan Cave [31], Shawan Cave [8], and Shijiangjun Cave [32] consistently show that drip water Mg/Ca increases during the dry season and decreases during the rainy season. In addition, the involvement of exogenous acids (H₂SO₄, HNO₃) in carbonate dissolution introduces non-atmospheric inorganic carbon, thereby affecting the δ¹³C-DIC signal [33,34].

In summary, modern cave monitoring integrated with isotopic tracing has advanced our understanding of the carbon cycle. Nevertheless, most studies have focused on single-site monitoring in limestone caves, and comparative research on the complex hydrological flow paths in dolomite caves remains scarce. In particular, the mechanisms regulating drip water δ¹³C-DIC under different flow paths are still poorly understood. Mahuang Cave, a branch of the Shuanghe Cave system (the world’s longest dolomite cave) in southwestern China, encompasses five hydrological flow types (vadose zone flow, percolation flow, shaft-fissure mixed flow, seasonally transitional subcutaneous flow, and underground river flow), making it an ideal site for addressing the above research gaps. Therefore, the core scientific question addressed in this study is the following: What are the hydrogeochemical mechanisms regulating drip water δ¹³C-DIC in dolomite caves? More specifically, how do soil CO₂ input, WRI, degassing, and PCP collectively shape the spatiotemporal variability of δ¹³C-DIC under different hydrological flow paths? Drawing on continuous multi-parameter monitoring data from 2019 to 2025, this study overcomes the limitations of previous research and reveals the distinct characteristics of the mechanisms controlling drip water δ¹³C-DIC in dolomite caves.

To address these core scientific questions, this study was organized into three components: (1) identifying cave hydrological flow path types and analyzing the spatiotemporal variations in water chemistry; (2) resolving the mechanisms regulating δ¹³C-DIC under different hydrological flow paths, with emphasis on the roles of soil CO₂ input, WRI, degassing, and PCP, while also evaluating the potential influence of interannual regional climate variability (e.g., in the context of ENSO); and (3) comparing with six limestone caves to highlight the distinctiveness of the drip water δ¹³C-DIC signal in dolomite caves. These findings will provide a modern-process foundation for understanding the mechanisms controlling drip water δ¹³C-DIC in dolomite caves and for interpreting stalagmite paleoclimate records.

2. Study Area Site Description

The field observation site is located at the Shuanghe Cave National Geopark in Wenquan Town, Suiyang County, Guizhou Province, southwestern China (107°02′30″ E–107°25′00″ E, 28°08′00″ N–28°20′00″ N; Figure 1a). Shuanghe Cave is currently the longest dolomite cave in Asia and the world, with a surveyed length of 439.7 km as of October 2025. The exposed strata in this region are dominated by the Middle–Upper Cambrian Loushanguan Formation (Є_2–3ls) and the Lower Ordovician Tongzi Formation (O₁t), with the main lithologies consisting of platform-evaporite dolomite and dolomitic limestone, commonly interbedded with gypsum layers and abundant celestite [35].

Figure 1. (a) Location of Guizhou Province in southwestern China. (b) Hydrogeology of the Shuanghe Cave area and the location of Mahuang Cave; Stratigraphic codes: Cambrian (Є): Є_2–3—Middle–Upper Series; Є_2–3ls—Loushanguan Formation; Є_2–3ls²—Second Member of the Loushanguan Formation; Є_2–3ls³—Third Member of the Loushanguan Formation; the lithology is mainly platform-facies evaporitic thick-bedded dolomite and dolomitic limestone, with gypsum-bearing strata. Ordovician (O): O₁m—Lower Series Meitan Formation; O₁t+h—Lower Series Tongzi Formation and Honghuayuan Formation; O₂₊₃—Middle–Upper Series; mainly dark gray to grayish-black medium- to thick-bedded dolomitic limestone and dolomite, intercalated with limestone and bioclastic rocks. Silurian (S): S₁—Lower Series; S₂sh—Middle Series Shiniulan Group; predominantly grayish-black carbonaceous shale and silty shale. (c) Overlying soil monitoring sites. (d) Cave passage of Mahuang Cave and monitoring locations. Symbols indicate monitoring parameters and sampling points, respectively. Abbreviations: T, temperature; RH, relative humidity; Pa, atmospheric pressure; Ws, wind speed; El, elevation.

This study focuses on Mahuang Cave (Figure 1b), the most important first-order branch of the Shuanghe system. It is 1188 m long, with an average width of 13 m and height of 9 m (Figure 1d), and its entrance is at an elevation of 720 m a.s.l. Mahuang Cave is a typical horizontal, single-entrance natural cave; its passage characteristics have been described in detail by Xiong et al. [36,37]. The region has a typical subtropical monsoon climate, with a mean annual temperature of 15.5 °C and mean annual precipitation of 1210 mm, more than 80% of which falls between April and September. The vegetation above the cave is predominantly C3 natural vegetation, with relatively thick soil cover (>100 cm), while C4 crops are mainly corn and sorghum.

3. Materials and Methods

3.1. Field Monitoring

From the cave near section to the deep zone, five drip water monitoring sites (#1, #2, #3, #4, and #6) and one underground river site (#5) were established (Figure 1d, Table 1), along with three overlying soil monitoring sites (S1, Cultivated land; S2, shrubland; S3, shrub-grassland) (Figure 1c). Cave ceiling thicknesses are given as reference values [38].

Table 1. Summary of cave water monitoring sites in Mahuang Cave.

Site	Cave Water Type	Drip Rate	Distance from Entrance/m	Elevation/m	Drop Height/m	Ceiling Thickness/m
#1	Stalactite drip	Perennially slow	109	720	6.71	99.9
#2	Stalactite drip	Perennially slow	175	724	4.34	107.8
#3	Fissure water	Fast	419	720	1.5	123.5
#4	Drip/Wall seepage	Varies with dry/wet season	508	718	0.5	96.7
#5	Underground river	—	682	708	—	86.9
#6	Stalactite drip	Perennially slow	1188	717	14.44	107.6

From January 2019 to December 2025, monthly field monitoring and water sampling were conducted. Water temperature (Tw), electrical conductivity (EC), pH, and dissolved oxygen (DO) were measured in situ using a Multi-3430 multiparameter water quality meter (model 2FD470, WTW, Munich, Germany), with measurement ranges of −5.0 to 105.0 °C, 0 to 2000 μS·cm⁻¹, and −2 to 20, respectively. The probes were calibrated with standard solutions before measurement, and each reading took approximately 5 min. HCO₃⁻ concentration was determined on site by titration using an Aquamerck alkalinity kit (Merck, Darmstadt, Germany), with a precision of 0.1 mmol/L. Drip rates at the five drip sites were measured three consecutive times using the volumetric cylinder–stopwatch method and averaged (mL/min). Water samples were filtered through 0.22 μm membrane filters and promptly transferred into 50 mL polyethylene bottles. For cation analysis, samples were acidified with 1:1 HNO₃ to pH < 2 to maintain ion activity; for δ¹³C-DIC analysis, two drops of HgCl₂ were added. All sample bottles were filled without headspace, sealed with Parafilm, and stored at 4 °C until laboratory analysis. Methods for measuring cave microclimatic parameters (temperature and humidity), soil/cave air CO₂ concentrations, and their carbon isotopic compositions have been detailed in Xiong et al. [37].

Due to the COVID-19 pandemic, monitoring was interrupted for a total of six months (February and March 2020; February and October 2021; March and September 2022). In addition, δ¹³C data for certain months were missing owing to instrumental failures. Meteorological data outside the cave were obtained from the Suiyang County Meteorological Station and from a weather service website (https://www.qweather.com/, accessed on 8 January 2026).

3.2. Laboratory Analyses

Water sample analyses and carbon isotope measurements were completed at the State Key Laboratory of Environmental Geochemistry, Institute of Geochemistry, Chinese Academy of Sciences.

(1) Water chemistry analysis: One duplicate sample was analyzed for every ten samples. Cations (Ca²⁺, Mg²⁺, Sr²⁺) were determined using a VISTA MPX inductively coupled plasma optical emission spectrometer (ICP-OES, Varian, Walnut Creek, CA, USA), with a spectral coverage of 175–785 nm, a resolution of 0.009 nm, and a detection limit of 0.001 mg/L. The relative standard deviation (RSD) was better than 2%.

(2) δ¹³C-CO₂ (soil/cave air) measurement: Air samples in sampling bags were connected to a vacuum extraction system (pressure < 10⁻³ mbar), and CO₂ was cryogenically collected using a liquid nitrogen cold trap to remove non-condensable gases. The purified CO₂ was then transferred into sample tubes and measured using a MAT 252 gas isotope ratio mass spectrometer (Finnigan MAT, Bremen, Germany). A dual-inlet mode was employed, with at least 8 comparison measurements per analysis. The instrument has a resolution of 200, a mass range of 1–150 amu, and a precision better than 0.15‰. Results are reported relative to the Vienna Pee Dee Belemnite (VPDB) standard.

(3) δ¹³C-DIC (cave water) measurement: A 50 mL water sample was placed in a reaction flask (vacuum < 10⁻² mbar), to which 2–3 mL of anhydrous phosphoric acid (concentration ≥ 99%) was added. The mixture was allowed to react at room temperature for 2 h to convert DIC into CO₂. The evolved gas was passed through an −80 °C cold trap to remove water vapor, and CO₂ was then cryogenically trapped in a −196 °C liquid-nitrogen cold trap. The vaporized CO₂ was introduced into the MAT 252 mass spectrometer for measurement, with a precision of ≤0.15‰, and results are reported relative to the VPDB standard.

3.3. Characterization of DIC

DIC typically exists in three inorganic species: carbonic acid (H₂CO₃, including dissolved CO₂(aq)), bicarbonate (HCO₃⁻), and carbonate (CO₃²⁻), with their speciation mainly controlled by pH. In natural waters with pH values between 7 and 9, HCO₃⁻ generally accounts for more than 90% of total DIC [2]. Given that the mean pH of the water samples in this study is 8.39, HCO₃⁻ is the dominant DIC species. Accordingly, we use the measured HCO₃⁻ concentration as a proxy for DIC, a practice adopted in previous studies [33,39].

This study identified six samples with pH > 10 (June 2021). Under such high pH conditions, the contribution of CO₃²⁻ to total DIC is non-negligible. To address this, we used the PHREEQC program to calculate total DIC and the major carbon-bearing species (HCO₃⁻, CO₃²⁻, CaCO₃, MgCO₃, and CO₂(aq)) for these six samples. The results showed that HCO₃⁻ accounted for 29.7% of total DIC on average, whereas CO₃²⁻ accounted for 38.2% on average (Table S1, Supplementary Materials). However, these six high-pH data points represent only 1.3% of all valid water chemistry samples (n = 466) and occurred in a single month (June 2021), with no recurrence before or after the monitoring period. As such, they represent an isolated hydrological event that does not affect the seasonal patterns or statistical results derived from the long-term monitoring. Accordingly, measured HCO₃⁻ concentrations were retained as a proxy for DIC for these six samples.

3.4. Calculation of Precipitation Anomaly in Percentage

The precipitation anomaly in percentage (PA) is the deviation of precipitation during a given period from the climatic mean for the same period. For analytical convenience, following the Grades of meteorological drought (GB/T 20481-2017), a PA ≤ −40% at the annual scale is classified as “severe drought” and is designated as an extreme drought year in this study; when −40% < PA ≤ −15%, it is classified as “mild to moderate drought” and is designated as a drought year. Correspondingly, wetter conditions are defined as 15% < PA < 40% for a wet year, and PA ≥ 40% for an extreme wet year. The PA is calculated as:

$$\mathrm{PA}={\displaystyle \frac{\mathrm{P}-\overline{\mathrm{P}}}{\overline{\mathrm{P}}}}\times 100\%$$

(1)

where PA is the precipitation anomaly percentage (%), P is the precipitation during a given period, and $\overline{\mathrm{P}}$ is the climatic mean precipitation for the same period (mm).

3.5. Calculation of the Carbonate Equilibrium Model

The calcite saturation index (SIc) is a key indicator of the saturation state of ions in water and reflects whether the water is saturated with respect to calcite, thereby indicating the tendency for calcite dissolution or precipitation: SIc > 0 indicates a saturated state, whereas SIc < 0 indicates an undersaturated (dissolving) state. The partial pressure of CO₂ in drip water (pCO₂(w)) reflects the degree of degassing at the water-air interface. The SIc and pCO₂(w) were calculated using the PHREEQC program (version 3.4.0; U.S. Geological Survey), with the following equations:

$${\mathit{SI}}_C=\lg {\displaystyle \frac{\left[{\mathrm{Ca}}^{2+}\right]\left[{\mathit{CO}}_3^{2-}\right]}{K_c}}$$

(2)

where lg denotes the base-10 common logarithm, [Ca²⁺] and [CO₃²⁻] are the ion activities in water, and K_c is the equilibrium constant for calcite dissolution in water [40].

$$P^{{\mathit{CO}}_{2\left(W\right)}}={\displaystyle \frac{\left[{\mathrm{H}}^{+}\right]\left[{\mathit{HCO}}_3^{-}\right]}{{k_{1}k}_{\mathrm{h}}}}$$

(3)

where [H⁺] and [HCO₃⁻] are ion activities in water (mol/L), and k₁ and k_h are the first dissociation constant of H₂CO₃ and Henry’s law constant, respectively.

3.6. Selection of PCP Proxy Model

Two main types of models are commonly used to identify PCP: linear models based on single-element ratios (Mg/Ca or Sr/Ca), and slope-threshold methods based on the logarithmic ratio of two elements (ln(Sr/Ca) vs. ln(Mg/Ca)). The slope-threshold method requires that the partition coefficients of Mg and Sr remain constant over time. However, in cave environments, these partition coefficients are influenced not only by temperature but also by multiple factors, including calcite growth rate, initial fluid composition, and crystal structure [14,41], making the constant-coefficient assumption difficult to satisfy in complex karst systems.

In Mahuang Cave, the mean annual temperature variation from 2019 to 2025 is only about 1.5 °C, suggesting that the effect of temperature on partition coefficients is limited [42]. Nevertheless, the slope-threshold method is more sensitive to variations in initial water chemistry and flow paths, which limits its applicability in this study area. In contrast, the linear relationship between Mg/Ca and Sr/Ca is clearly responsive to WRI and PCP processes and has been widely validated in cave studies across diverse climatic settings worldwide [43,44]. Therefore, this study adopts the linear Mg/Ca–Sr/Ca relationship as the proxy for identifying PCP.

3.7. Identification of ENSO Event Conditions

This study identifies El Niño–Southern Oscillation (ENSO) events using the Niño 3.4 region (120° W–170° W, 5° N–5° S) sea surface temperature (SST) anomalies and the Southern Oscillation Index (SOI), obtained from the U.S. National Centers for Environmental Prediction (NCEP) via the data access website (https://www.cpc.ncep.noaa.gov/data/indices/, accessed on 23 January 2026), with the goal of providing a climatic context for the interannual variability of δ¹³C-DIC during the monitoring period. Following the Identification method for El Niño/La Niña events (GB/T 33666-2017), an ENSO event is identified when the SST anomalies in the Niño 3.4 region persist for at least five consecutive months at or above ±0.5 °C (≥0.5 °C for El Niño, ≤−0.5 °C for La Niña), and is further validated by concurrent SOI anomalies (negative for El Niño, positive for La Niña).

3.8. Statistical Analyses

Pearson correlation analysis was used to assess linear correlations between δ¹³C-DIC and environmental parameters, with 95% confidence intervals reported for each correlation coefficient. To account for serial dependence in monthly time-series data, autocorrelation function (ACF) analysis (max lag = 24 months) was performed to extract the first-order autocorrelation coefficient (r₁) for each variable, and cross-correlation function (CCF) analysis (max lag = ±12 months) was applied to identify optimal lag times between δ¹³C-DIC and each environmental parameter. Effective sample sizes (N_eff) were calculated as N_eff = N × (1 − r₁,X × r₁,Y)/(1 + r₁,X × r₁,Y) to adjust significance tests for serial autocorrelation. Bonferroni correction was applied for multiple comparisons (m = 7, α = 0.05/7 ≈ 0.0071). Regression diagnostics included the Durbin–Watson statistic, studentized residuals, and variance inflation factors (VIF). Relevant data analysis was performed using Origin 2021 and IBM SPSS Statistics 25.

4. Results

4.1. Environmental Conditions Outside Mahuang Cave

Statistical analysis of meteorological parameters in the study area from 2019 to 2025 (Figure 2c) shows that monthly mean air temperatures ranged from 3.7 to 27.4 °C, with an overall mean of 15.9 °C. Annual mean temperatures were 15.1, 15.6, 16.0, 15.9, 16.1, 16.4, and 16.1 °C, respectively, indicating a warming trend. Temperatures began to rise markedly in March each year, with the highest values occurring in August and the lowest in February of the following year. Monthly precipitation ranged from 0 to 412.8 mm, with a monthly mean of 73.6 mm. Annual total precipitation amounts were 1069.8, 1421.4, 752.2, 224.4, 664.0, 1113.7, and 938.0 mm, respectively, with a mean annual total of 883 mm, of which 79.1% fell between April and September, reflecting the typical mid-subtropical climate pattern of concurrent high temperature and rainfall.

Figure 2. Drip sites in Mahuang Cave and meteorological variations. (a) Slow drip sites #1, #2, and #6. (b) Fast drip sites #3 and #4; (c) Air temperature and precipitation (cyan bars indicate months with concentrated rainfall). ED, D, N, W, and EW denote dry–wet year categories, corresponding to extreme drought, drought, normal, wet, and extreme wet years, respectively.

Based on the PA (Equation (1)), the years were classified as follows: 2022 (−75%) was an extreme drought year; 2023 (−25%) was a drought year; 2021 (−14.8%) and 2025 (6%) were normal years; 2019 (21%) and 2024 (26%) were wet years; and 2020 (61%) was an extreme wet year (with a single-month precipitation of 412.8 mm in September, approximately twice the total annual precipitation of 2022).

4.2. Hydrological Dynamics and Classification of Drip Water Types

During the monitoring period, the five drip sites exhibited pronounced spatial differences in discharge rates (Figure 2a,b). Sites #1, #2, and #6 showed low and stable discharge rates, ranging from 0.01 to 0.29 mL/s, with mean values of 0.07–0.09 mL/s. Site #3 exhibited considerable variability (0.2–20 mL/s, mean 4.05 mL/s), with a trickle flow during the dry season and pulsed-surge flow during the rainy season. Site #4 showed the most extreme variability (0.03–90 mL/s, mean 7.92 mL/s), even ceasing flow during the dry season (e.g., September–October 2019), while producing extreme peaks driven by rainstorms during the rainy season (e.g., 76.67 mL/s in October 2020; 90 mL/s in April 2024), reflecting a typical seasonal transition response.

Based on the response characteristics of drip water to rainfall [45], the five drip sites were classified into four hydrological types (Figure 3): Vadose zone flow type (#1, #2)—low discharge with moderate coefficient of variation, modulated by storage in the thick vadose zone through pore and fracture media. Percolation flow type (#6)—the most stable discharge (lowest CV), reflecting slow percolation through fine pores and micro-fractures. Shaft-fracture mixed flow type (#3)—a multiple-flow-path system composed of fast shaft conduits and slow fracture networks, with conduit flow during the rainy season and fracture seepage during the dry season. Seasonally transitional subcutaneous flow type (#4)—extreme discharge variability, characterized by wall seepage or even cessation during the dry season and storm pulse responses during the rainy season, serving as a typical indicator of rapid transmission and transient storage in the epikarst zone. Overall, discharge at all sites was higher in the rainy season than in the dry season, but the amplitude of the response and the lag time varied with the hydrological pathway.

Figure 3. Hydrological types of drip water in Mahuang Cave are classified following the method of Smart and Friedrich (1987) [45].

4.3. Spatiotemporal Characteristics and Types of Hydrochemical Ions

Hydrochemical ion concentrations in Mahuang Cave exhibited pronounced seasonal fluctuations (Figure 4). Ca²⁺ and HCO₃⁻ showed a pattern of higher concentrations in the warm season and lower in the cold season (Ca²⁺ summer–autumn mean: 46.19–47.44 mg/L; winter–spring mean: 43.80–45.08 mg/L; HCO₃⁻ peak of 317.73 mg/L in summer and trough of 122.04 mg/L in winter), reflecting that the warm season, with higher temperatures and increased rainfall, enhances soil CO₂ input and intensifies carbonate dissolution, whereas the cold season shows the opposite trend. Mg²⁺ exhibited a seasonal increase, with a peak occurring in winter (15.83 mg/L). Na⁺ concentrations are anomalously high in spring (up to 2.66 mg/L, approximately twice those of the other seasons), while the spring mean of SO₄²⁻ is the highest of the four seasons (20.12 mg/L) but with limited amplitude, indicating that the spring “initial flushing effect” leached soluble surface salts and sulfides accumulated during the preceding dry season [46,47]. In contrast, K⁺, Cl⁻, and NO₃⁻ do not exhibit spring enrichment comparable to that of Na⁺, which mainly arises from differences in their storage locations and transport pathways within the epikarst: K⁺ is subject to double buffering by growing-season vegetation uptake and clay mineral adsorption, Cl⁻ has diffuse sources and limited net accumulation, and NO₃⁻ is regulated by multiple processes including source mixing (chemical fertilizer, soil nitrogen, manure, and atmospheric deposition), nitrification–denitrification, and plant uptake.

Figure 4. Interannual variations in hydrochemical ions from 2019 to 2025. (a) Rainfall; (b) Ca²⁺; (c) Mg²⁺; (d) Sr²⁺; (e) Na⁺; (f) K⁺; (g) HCO₃⁻; (h) Cl⁻; (i) SO₄²⁻; (j) NO₃⁻.

On the interannual scale, concentrations remained generally stable, except for an abrupt change from March to June 2024, characterized by a sharp rise followed by a steep decline. In March–April, Na⁺ concentrations first showed anomalously high values (reaching 16.28 and 15.86 mg/L), followed by a transient peak in HCO₃⁻ at sites #5 and #6 in April (258.79 and 317.73 mg/L). In May–June, Ca²⁺ and Mg²⁺ dropped precipitously: Ca²⁺ fell sharply from 53.96 mg/L in April to 13.22 mg/L (a 75.7% decrease), while Mg²⁺ synchronously dropped to a low range of 0.34–3.77 mg/L.

This abrupt change can be attributed to the cumulative effects of soil CO₂ and solute storage in the vadose zone resulting from two consecutive dry years (224.4 mm in 2022, 664 mm in 2023) [48]. The precipitation events in March–April 2024 triggered layered leaching. In the initial stage, infiltrating water first flushed readily soluble salts (e.g., NaCl and nitrates) that had accumulated on the rock and soil surfaces, driving a sharp increase in Na⁺ [49]. Subsequently, the water continued to percolate downward, dissolving high-concentration CO₂ stored in the vadose zone and forming highly corrosive carbonic acid, which caused a transient HCO₃⁻ peak at sites #5 and #6. By May–June, sustained heavy rainfall triggered a flow-regime shift and piston-flow flushing, whereby newly infiltrated low-mineralization rainwater displaced the previously stored high-mineralization “old water” [10,50], resulting in an abrupt decline in Ca²⁺ and Mg²⁺. Notably, the precipitation amounts during this period (20.2–198.4 mm) were not the highest recorded in the study period, suggesting that the temporal distribution of rainfall and the antecedent hydrological state (vadose zone moisture conditions and solute storage) exert greater control over hydrochemical responses than the total precipitation amount alone [51,52].

Ion concentrations showed pronounced spatial heterogeneity (Table 2). Among cations, Ca²⁺ ranged from 40.39 to 48.48 mg/L, decreasing from #1 to #5 with a slight rebound at #6. Mg²⁺ was highest at #6 (22.66 mg/L, approximately twice that of #3), with the largest standard deviation (6.18), indicating a significant contribution from dolomite dissolution. Sr²⁺ was lowest at #1 and #3 and highest at #6 (1.36 mg/L), suggesting PCP occurrence. Na⁺ showed little spatial variation and remained at relatively low concentrations across all sites (1.53–1.79 mg/L), while K⁺ was highest at #6 (2.28 mg/L) and lowest at #1 (0.83 mg/L).

Table 2. Spatial characteristics of hydrochemical parameters at monitoring sites (2019–2025).

Parameter	Values	Sampling Site
		#1	#2	#3	#4	#5	#6
Drip rate	mean (mL/s)	0.09	0.07	4.05	7.92	—	0.09
	std (cv~%)	0.06 (67)	0.04 (57)	4.45 (110)	16.23 (205)	—	0.04 (44)
	range (mL/s)	0.02–0.29	0.01–0.22	0.2–20	0.03–90	—	0.03–0.23
Ca2+	mean (mg/L)	48.48	47.41	46.34	44.39	40.39	46.77
	std (cv~%)	10.05 (21)	7.51 (16)	8.35 (18)	10.83 (24)	8.55 (21)	8.37 (18)
	range (mg/L)	9.6–64.15	12.74–61.81	13.91–61.32	13.86–66.51	8.14–62.10	12.08–58.67
Mg2+	mean (mg/L)	13.12	17.17	11.43	11.66	13.09	22.66
	std (cv~%)	3.45 (26)	4.06 (24)	2.74 (24)	2.89 (25)	3.85 (29)	6.18 (27)
	range (mg/L)	0.38–16.36	0.21–22.73	0.3–14.94	0.28–16.13	0.01–22.46	0.02–28.81
Sr2+	mean (mg/L)	0.44	0.69	0.44	0.59	1.19	1.36
	std (cv~%)	0.29 (66)	0.30 (43)	0.29 (65)	0.33 (56)	0.46 (38)	0.38 (28)
	range (mg/L)	0.01–1.30	0.01–1.69	0.01–1.54	0.01–1.65	0.01–2.48	0.01–2.32
Na+	mean (mg/L)	1.54	1.53	1.56	1.58	1.79	1.53
	std (cv~%)	2.24 (146)	2.97 (194)	2.01 (129)	1.97 (124)	2.10 (117)	3.85 (252)
	range (mg/L)	0.51–14.37	0.25–18.85	0.43–13.07	0.37–12.83	0.31–14.59	0.04–24.71
K+	mean (mg/L)	0.83	1.32	1.07	1.00	1.42	2.28
	std (cv~%)	0.57 (68)	0.70 (53)	0.50 (47)	0.53 (53)	0.52 (37)	1.29 (56)
	range (mg/L)	0.15–3.52	0.75–4.61	0.6–3.5	0.52–3.46	0.62–3.28	0.69–8.26
HCO3−	mean (mg/L)	172.91	196.27	175.17	169.23	171.17	210.28
	std (cv~%)	16.44 (10)	14.39 (7)	19.32 (11)	25.68 (15)	20.30 (12)	21.70 (10)
	range (mg/L)	122.04–216.62	158.65–234.93	143.40–216.62	125.09–228.83	134.24–258.79	146.45–317.73
Cl−	mean (mg/L)	4.90	2.76	2.72	2.62	2.17	1.07
	std (cv~%)	1.85 (38)	1.49 (54)	1.13 (42)	0.77 (29)	0.62 (28)	1.71 (160)
	range (mg/L)	0.24–9.05	1.29–9.47	1.42–8.52	0.99–4.98	0.65–3.72	0.18–8.86
SO42−	mean (mg/L)	15.83	17.42	11.86	13.68	13.60	42.96
	std (cv~%)	5.43 (34)	1.39 (8)	3.49 (29)	2.40 (18)	1.58 (12)	10.20 (24)
	range (mg/L)	12.23–48.17	13.2–21.67	4.47–18.04	5.92–19.44	8.03–17.42	7.04–52.83
NO3−	mean (mg/L)	22.39	15.00	12.41	10.81	9.97	1.24
	std (cv~%)	6.64 (30)	3.66 (24)	4.39 (35)	3.42 (32)	2.82 (28)	2.75 (221)
	range (mg/L)	0.19–40.45	4.93–28.32	3.93–31.6	4.07–19.7	3.2–17.93	0.01–12.87
Tw	mean (°C)	13.02	13.19	13.21	13.35	13.49	14.81
	std (cv~%)	3.03 (23)	2.45 (19)	1.85 (14)	2.13 (16)	1.63 (12)	1.54 (10)
	range (°C)	5.2–17.2	6.5–18.9	8.2–17.8	7.6–15.9	9.5–15.8	9.3–19.9
pH	mean	8.42	8.45	8.27	8.38	8.44	8.38
	std (cv~%)	0.33 (4)	0.37 (4)	0.36 (4)	1.02 (12)	0.39 (5)	0.37 (4)
	range	7.43–10.31	7.45–10.75	7.15–10.33	7.36–10.40	7.33–10.40	7.52–10.61
EC	mean (μS/cm)	337.52	345.22	306.35	293.69	291.66	389.12
	std (cv~%)	16.32 (5)	30.2 (9)	33.99 (11)	38.11 (13)	16.2 (6)	14.42 (4)
	range (μS/cm)	288–370	222–480	259–509	231–388	264–332	323–432
DO	mean (mg/L)	9.41	9.43	9.44	9.49	9.52	9.14
	std (cv~%)	0.74 (8)	0.59 (6)	0.53 (6)	0.66 (7)	0.57 (6)	0.72 (8)
	range (mg/L)	8.07–11.5	8.42–12	8.01–12.1	8.55–12.6	8.48–12.7	8–14.1
SIc	mean	0.68	0.76	0.54	0.61	0.63	0.71
	std (cv~%)	0.26 (38)	0.28 (37)	0.30 (55)	0.31 (51)	0.34 (55)	0.29 (41)
	range	−0.16~1.76	−0.10~1.82	−0.42~1.77	−0.26~1.79	−0.33~1.76	−0.06~1.73
pCO2(w)	mean	−3.29	−3.28	−3.13	−3.26	−3.31	−3.17
	std (cv~%)	0.40 (12)	0.47 (14)	0.45 (14)	0.44 (13)	0.47 (14)	0.49 (15)
	range	−5.85~−2.23	−6.64~−2.18	−5.88~−1.89	−5.99~−2.17	−5.96~−2.13	−6.48~−2.24
Mg/Ca molar ratio	mean (mol/mol)	0.45	0.60	0.41	0.45	0.54	0.80
	std (cv~%)	0.09 (19)	0.11 (19)	0.08 (21)	0.13 (28)	0.13 (24)	0.16 (20)
	range (mol/mol)	0.01–0.70	0.01–0.81	0.01–0.58	0.01–0.69	0.00–0.75	0.00–1.03
100 × Sr/Ca molar ratio	mean (mol/mol)	0.47	0.73	0.50	0.70	1.42	1.40
	std (cv~%)	0.59 (126)	0.64 (87)	0.60 (120)	0.66 (94)	0.81 (57)	0.72 (51)
	range (mol/mol)	0.01–3.85	0.01–4.35	0.01–3.76	0.01–3.86	0.01–4.9	0.01–5.15
δ13C-DIC	mean (‰)	−9.66	−8.53	−10.63	−8.69	−8.43	−7.92
	std (cv~%)	1.03 (11)	1.05 (12)	1.45 (14)	3.01 (35)	2.89 (34)	0.69 (9)
	range (‰)	−11.71~−6.01	−10.53~−6.35	−13.66~−7.46	−13.44~−2.14	−13.51~−2.76	−9.61~−6.5

Among anions, HCO₃⁻ ranged from 2.77 to 3.45 mmol/L, with the highest value at #6 and the lowest at #4. Sites #1 and #2 exhibited elevated Cl⁻ (4.90 mg/L) and NO₃⁻ (22.39 mg/L), likely indicating the strongest influence from agricultural activities. Site #6 had the highest SO₄²⁻ concentration (42.96 mg/L), which is 2.5 to 3.6 times that of the other sites, reflecting sulfuric-acid-driven dissolution. NO₃⁻ showed a marked decreasing trend from #1 to #6, likely constrained by agricultural nitrogen inputs.

Piper diagram analysis (Figure 5) indicates that the hydrochemical type of Mahuang Cave water is Ca·Mg-HCO₃. Cations are dominated by Ca²⁺ (61.3%) and Mg²⁺ (32.7%), while among anions, HCO₃⁻ is the overwhelmingly dominant species, accounting for 91.57% of total anion equivalents.

Figure 5. Piper diagram of hydrochemical types in Mahuang Cave.

4.4. Spatiotemporal Characteristics of Hydrochemical Parameters and Elemental Ratios

The main physicochemical parameters of Mahuang Cave water are shown in Table 2 and Figure 6b–e. Tw ranged from 5.2 to 19.9 °C (mean 13.5 °C), with higher values in summer and lower in winter, and increased progressively from #1 to #6 (13.02–14.81 °C), with #6 showing the smallest variation (CV = 10%), indicating the most thermally stable environment. The pH ranged from 7.15 to 10.75 (mean 8.39), indicating weakly alkaline conditions, with slightly higher values in winter–spring than in summer–autumn and relatively small spatial variation. EC ranged from 222 to 509 μS/cm, with values in the order of summer > spring ≈ autumn > winter, highest at #6 (389.12 μS/cm) and lowest at #5 (291.66 μS/cm). DO averaged 9.14–9.52 mg/L, with higher values in winter and lower in summer, showing no significant differences among sites. At the interannual scale, these parameters remained generally stable; however, several anomalies warrant attention, including sharp increases in EC in spring and summer 2019 (509 μS/cm at site #3 in April and 480 μS/cm at site #2 in August), a pH < 7.5 in September 2025, and pH values > 10 in June 2021. Notably, the six monitoring sites recorded exceptionally high pH values (10.31–10.75) in June 2021, with precursor pH elevations (9.24–9.74) already observed in May of the same year, suggesting a regional hydrological event. This anomaly was most likely caused by the “piston-flow effect” triggered by sustained rainfall during the Meiyu (plum rain) season, whereby high-pH “old water” was displaced en masse by infiltrating rainfall. The anomaly occurred only in the June 2021 monitoring event (n = 6, accounting for 1.3% of all valid samples) and did not recur thereafter, representing an isolated event that did not affect the overall seasonal pattern.

Figure 6. Variations in major physicochemical parameters from 2019 to 2025. (a) Rainfall and air temperature; (b) cave water temperature; (c) pH; (d) electrical conductivity (EC); (e) dissolved oxygen (DO); (f) saturation index of calcite (SIc); (g) partial pressure of CO₂ in drip water (pCO₂(w)); (h) Mg/Ca molar ratio; (i) 100 × Sr/Ca molar ratio.

SIc and pCO₂(w) exhibited a typical inverse relationship (Figure 6f,g). SIc ranged from 0.54 to 0.76 (all >0, n = 462), indicating that the water was generally supersaturated with respect to calcite; negative values were also observed (minimum SIc = −0.42, n = 6), suggesting localized episodes of undersaturation. SIc varied in phase with pH (Figure 6c). The negative logarithm of pCO₂(w) ranged from −3.13 to −3.31, with small differences among sites, indicating relatively uniform CO₂ partial pressure in the water.

The Mg/Ca ratio (Figure 6h) was higher in winter (0.584) and lower in summer (0.536), with higher values in dry years and lower in wet years. Spatially, it increased progressively from #1 (0.45) to #6 (0.8), with #2 (0.6) and #5 (0.54) at intermediate levels, and #3 (0.41) and #4 (0.45) relatively lower; #4 exhibited the largest variation (CV = 28%).

The 100 × Sr/Ca ratio (Figure 6i) also showed higher values in winter (0.973) and lower values in summer (0.768), but its peaks and troughs did not correspond to dry/wet year classifications, and its interannual variation was irregular. Spatially, #5 (1.42) and #6 (1.4) had significantly higher values than the other sites, while #1 (0.47) and #3 (0.5) had the lowest values. #5 and #6 showed the smallest variation (CV = 51–57%), whereas the remaining sites exhibited larger variability (CV = 87–126%).

Overall, the concomitant increase in Mg/Ca and Sr/Ca indicates enhanced water–rock interaction, while their decoupling suggests contributions from different mineral end-members (e.g., preferential dissolution of dolomite vs. calcite).

4.5. Spatiotemporal Characteristics of Soil–Cave–Water Carbon Isotopes Under Interannual Climate Variability

During the monitoring period, the regional climate exhibited pronounced interannual variability. As a background reference, ENSO underwent a phase transition from a weak El Niño (2019) to a triple La Niña (summer 2020, spring 2021, autumn 2022), then to a moderate El Niño (May 2023–April 2024), and subsequently returned to La Niña conditions (Figure 7b), accompanied by interannual fluctuations in precipitation and temperature (Figure 7a).

Figure 7. (a) Air temperature and precipitation; (b) ENSO events (El Niño and La Niña phases are indicated by light yellow and light blue shading, respectively); (c–g) Soil and cave CO₂, DIC concentrations, and their δ¹³C values.

Soil air CO₂ concentrations (Figure 7c) were highest at S3 (16,089 ppmv), followed by S1 (10,932 ppmv) and S2 (10,531 ppmv), with overall higher values in the rainy season and lower in the dry season. Soil δ¹³C-CO₂ exhibited an inverse pattern (Figure 7d), with ranges of −26.57‰ to −13.17‰ at S1, −26.89‰ to −17.23‰ at S2, and −26.57‰ to −15.71‰ at S3, and mean values of −21.21‰, −23.79‰, and −23.25‰, respectively.

Cave air CO₂ (318–974 ppmv) followed a similar seasonal pattern to soil CO₂ but at significantly lower absolute concentrations, with seasonal means of 710 ppmv in summer, 539 ppmv in autumn, 489 ppmv in spring, and 392 ppmv in winter (Figure 7e), and showed a slight increase from the entrance to the interior (near cave section, 526 ppmv; transition zone, 542 ppmv; deep zone, 547 ppmv). Cave δ¹³C-CO₂ (Figure 7f) was lightest in summer (−16.57‰) and heaviest in winter (−10.82‰). Spatially, it became progressively lighter from the entrance toward the interior: #2 (−13.51‰) > #3 (−13.54‰) > #6 (−13.88‰).

Cave water δ¹³C-DIC ranged from −13.66‰ to −2.14‰, with a mean of −9.1‰, showing a pattern of lighter values in the rainy season and heavier values in the dry season, with interannual fluctuations following precipitation variations (Figure 7g): mean values were lighter in the wet year of 2019 (−9.60‰) and the extreme wet year of 2020 (−10.08‰), increased to −8.97‰ in the normal year of 2021, reached their heaviest in the extreme drought year of 2022 (−8.21‰) and remained heavy in the drought year of 2023 (−8.24‰), then rebounded to −9.12‰ in the wet year of 2024 and stabilized at −8.70‰ in the normal year of 2025, showing an overall negative correlation with precipitation. Spatially, δ¹³C-DIC was lightest at #3 (−10.63‰), followed by #1 (−9.66‰), then #4 (−8.69‰), #2 (−8.53‰), and the underground river #5 (−8.43‰), with the heaviest value at #6 (−7.92‰). The mean δ¹³C-DIC varied by approximately 2.7‰ across sites due to differences in hydrological pathways and WRI, with the largest variations observed at #4 and #5 (CV = 34–35%).

4.6. Time-Series Characteristics, Cross-Correlation, and Regression Analysis Results

ACF analysis revealed significant first-order autocorrelation for all variables (Table S2, Supplementary Materials): δ¹³C-DIC (r₁ = 0.687), temperature (0.807), rainfall (0.453), soil CO₂ (0.626), cave CO₂ (0.684), and Niño 3.4 SST (0.956) (all p < 0.001). Based on these values, N_eff ranged from 16.1 (SST–δ¹³C-DIC) to 43.1 (soil δ¹³C-CO₂–δ¹³C-DIC) (Table S3, Supplementary Materials).

CCF analysis indicated that temperature, rainfall, soil CO₂, cave CO₂, soil δ¹³C-CO₂, and cave δ¹³C-CO₂ all achieved their maximum correlation with δ¹³C-DIC at lag 0 months (Table S3; Table 3), indicating a response time of less than one month and reflecting a sub-monthly geochemical response of the cave system. Niño 3.4 SST peaked at lag +5 months (r = −0.282), but this value fell below the 95% confidence interval (±2SE ≈ 0.234) and was not statistically significant. This result suggests that the ENSO signal is substantially attenuated and buffered during transmission through the atmosphere-hydrology-karst system, exerting only a weak direct influence on dripwater cave δ¹³C.

Table 3. Summary of Pearson correlation coefficients, cross-correlation lags, and regression diagnostics between δ¹³C-DIC and environmental parameters.

Parameter	N	Pearson r	95% CI	CCF Optimal Lag (Months)	R2	Durbin–Watson	Max Studentized Residual
Temperature	78	−0.747	[−0.83, −0.63]	0	0.557	0.891	2.171
Rainfall	78	−0.649	[−0.74, −0.51]	0	0.421	1.149	2.21
Soil CO2 concentration	78	−0.742	[−0.84, −0.61]	0	0.551	1.033	2.214
Soil δ13C-CO2	77	0.396	[+0.20, +0.57]	0	0.156	0.693	2.713
Cave CO2 concentration	78	−0.823	[−0.89, −0.72]	0	0.678	1.312	2.154
Cave δ13C-CO2	77	0.794	[+0.69, +0.86]	0	0.63	1.046	2.024
Niño 3.4 SST	78	0.002	[−0.22, +0.23]	5	0	0.521	2.161

Note: N = number of valid observations after listwise deletion; CI = 95% confidence interval; CCF optimal lag indicates the lag (in months) at which the maximum cross-correlation occurs (positive values indicate the environmental variable leads δ¹³C-DIC); R² = coefficient of determination; DW = Durbin–Watson statistic. All correlations except Niño 3.4 SST are significant at p < 0.001 (Bonferroni-corrected threshold α = 0.05/7 = 0.0071). N_eff values and ACF results are provided in Tables S2–S4 (Supplementary Materials). DW values < 1.8 indicate positive residual autocorrelation, consistent with monthly time-series data. Max studentized residuals are all within ±3, indicating no significant outliers; VIF = 1.000 for all bivariate models (no multicollinearity).

Pearson correlation analysis (Table 3) showed that δ¹³C-DIC was significantly negatively correlated with temperature (r = −0.747, n = 78), rainfall (r = −0.649, n = 78), soil CO₂ (r = −0.742, n = 78), and cave CO₂ (r = −0.823, n = 78) (all p < 0.001, Bonferroni-corrected α = 0.05/7 = 0.0071), and significantly positively correlated with soil δ¹³C-CO₂ (r = 0.396, n = 77) and cave δ¹³C-CO₂ (r = 0.794, n = 77). Niño 3.4 SST showed no significant correlation with δ¹³C-DIC (r = 0.002, p = 0.984).

Regression diagnostics (Table S4, Supplementary Materials; Table 3) showed Durbin–Watson statistics ranging from 0.521 to 1.312, maximum studentized residuals < 3, and VIF = 1.000. Cave CO₂ was the strongest predictor (R² = 0.678), whereas Niño 3.4 SST had near-zero explanatory power (R² ≈ 0).

5. Discussion

5.1. Influence of Interannual Hydroclimatic Variability on δ¹³C in Soil–Cave–Drip Water

The δ¹³C-DIC is a sensitive geochemical indicator of carbon cycling in karst systems, with its initial value primarily inherited from soil CO₂ after dissolution fractionation [53]. Because overlying vegetation uses different photosynthetic pathways, the theoretical δ¹³C-CO₂ values of soil CO₂ derived from decomposing plant residues differ significantly (C3 plants: ~−27‰; C4 plants: ~−13‰), and this signal can be transmitted to the cave via infiltrating water [54].

The δ¹³C-DIC variations measured at Mahuang Cave reveal a sensitivity to interannual hydroclimatic variability in the region. While ENSO was included in the analysis as a background framework for interannual climate variability, this study places greater emphasis on how precipitation/temperature changes driven by ENSO phases propagate to the drip water δ¹³C-DIC signal via soil CO₂ input, infiltration recharge, and PCP processes. During El Niño phases, regional moisture increased; enhanced plant productivity and soil microbial activity elevated soil CO₂ production, while abundant precipitation inhibited the diffusion of soil CO₂ into the atmosphere, thereby increasing the concentration of CO₂ dissolved in infiltrating water. This, in turn, promoted rapid carbonate dissolution and transported DIC enriched in light carbon (¹²C) into cave water, resulting in more negative δ¹³C-DIC values [55,56]. During the maximum precipitation event in September 2020 (412.8 mm), the δ¹³C-DIC value at site #3 reached its lowest value over the seven-year monitoring period (−13.66‰). During La Niña phases, reduced precipitation not only lowered soil CO₂ production but also prolonged the residence time of drip water in the vadose zone, driving PCP, during which light carbon (¹²C) was preferentially partitioned into the solid phase, leaving the residual water enriched in ¹³C [10,57]. During the triple La Niña period, δ¹³C-DIC values at sites #4 and #5 were significantly enriched (ranging from −3.3‰ to −2.19‰), approximately 4.5‰ heavier than the multi-year mean.

The correlation and cross-correlation analyses (Table 3; Figure 8; Tables S2–S4, Supplementary Materials) indicate that the seasonal variations in δ¹³C-DIC are primarily regulated by the interplay of soil CO₂ input, rainfall recharge, and cave CO₂ degassing inhibition. Rising temperatures and increased precipitation promote soil microbial activity, leading to elevated soil CO₂ concentrations; the ¹²C-enriched soil CO₂ then enters karst water, thereby reducing drip water δ¹³C-DIC [9,55]. The high cave CO₂ concentration in summer (710 ppmv) inhibits degassing, preserving light isotopes and resulting in more negative δ¹³C-DIC; the opposite occurs in winter, with enhanced degassing leading to more positive values. This interpretation is further supported by regression analyses (Table 3): cave CO₂ concentration (R² = 0.678) and cave δ¹³C-CO₂ (R² = 0.63) were the strongest predictors of δ¹³C-DIC, confirming the central role of the “degassing inhibition” effect in controlling the carbon isotopic composition of cave water. The moderate explanatory power of temperature (R² = 0.557) and rainfall (R² = 0.421) reflects the transmission of surface climate signals to the cave system through soil CO₂ production and infiltration processes. In contrast, the negligible explanatory power of Niño 3.4 SST (R² ≈ 0) further indicates that the ENSO signal is substantially attenuated during passage through the atmosphere-hydrology-karst system, with a direct influence on cave δ¹³C-DIC that can be considered negligible. These findings are consistent with observations from Ernesto Cave in Italy [3] and Qianfo Cave in China [58].

Figure 8. Correlations between δ¹³C-DIC and monitoring parameters. Sample sizes (N) are identical to those reported in Table 3.

It should be noted, however, that the observation window (2019–2025) spans only seven years, which is relatively short for robust ENSO attribution. The 2022–2023 dry years and the 2024 piston-flow event may have introduced hydrological memory effects and extreme anomalies that could obscure the pure ENSO signal. The significant correlations observed at lag 0 months do not necessarily imply causation; they may partly reflect seasonal cycles. Therefore, the associations documented here should be interpreted as correlative evidence rather than definitive causal attribution. Our results suggest—rather than confirm—that cave systems have the potential to record interannual climate variability; longer time-series data and process-based modeling are needed to disentangle the relative contributions of ENSO forcing, local hydrological processes, and cave-specific internal dynamics.

5.2. Control of Water–Rock–Gas Interactions on δ¹³C-DIC

Water–rock–gas interactions constitute the core mechanism regulating cave drip water δ¹³C-DIC, with SIc and pCO₂(w) serving as the key coupled variables that control mineral phase equilibrium and the driving force for CO₂ exchange at the gas–liquid interface, jointly determining the direction and magnitude of spatiotemporal variations in δ¹³C-DIC. High pCO₂(w) in water indicates that more CO₂ is dissolved, promoting the formation of H₂CO₃ and driving the reaction CO₂(g) + H₂O(l) + CaCO₃(s) ⇌ Ca²⁺(l) + 2HCO₃⁻(l) to the right, leading to increased Ca²⁺ and HCO₃⁻ concentrations and decreased SIc. Conversely, when CO₂-rich drip water enters the cave, if the cave air CO₂ partial pressure is lower than that of the water, CO₂ degassing occurs, pCO₂(w) decreases, pH increases, and the reaction shifts to the left, driving the water toward calcite supersaturation and increasing SIc [24].

In Mahuang Cave, SIc ranged from 0.54 to 0.76 (all >0, n = 462), indicating supersaturation with respect to calcite and favoring calcite precipitation [50]; negative SIc values (n = 6) were attributed to dilution effects from heavy rainfall [47]. Seasonal mean SIc decreased in the order: spring (0.76) > winter (0.67) > summer (0.64) > autumn (0.62), suggesting that summer and autumn were jointly modulated by precipitation dilution and degassing competition [59]. Spatially, slow drip sites (#1, #2, #6) exhibited higher and more stable SIc (mean 0.68–0.71, CV = 37–41%), whereas fast drip sites (#3, #4) and the underground river (#5) showed lower and more variable SIc (mean 0.54–0.63, CV = 51–55%). This pattern is consistent with observations from Golgotha Cave [60] and further reveals the heterogeneity of hydrological pathways in the vadose zone [10]. pCO₂(w) was higher and more variable in summer (range −6.64 to −2.78, CV = 21%), attributed to vigorous soil respiration and rapid infiltration that inhibited degassing [9], while it was lower and more stable in winter (range −2.95 to −3.99, CV = 6%) due to weaker biological activity, longer residence time, and enhanced ventilation that facilitated CO₂ evasion [50].

The intensity and mode of water–rock–gas synergy, as reflected by the above SIc and pCO₂(w) dynamics, are ultimately manifested in the systematic coupling between the [Ca²⁺+Mg²⁺]/[HCO₃⁻] equivalent ratio and δ¹³C-DIC. As shown in Figure 9, the evolutionary trajectories of these two variables exhibit a continuous distribution from the lower left to the upper right, clearly indicating a gradient transition from open to near-closed conditions along vadose-zone flow paths.

Figure 9. Relationship between the [Ca²⁺+Mg²⁺]/[HCO₃⁻] equivalent ratio and δ¹³C-DIC. Circular scatter points of different colors (#1–#6) represent different monitoring water sites, and the background color bands represent different isotopic fractionation endmembers.

(I) In the open system, soil CO₂ (δ¹³C ≈ −22.8‰) is continuously supplied to infiltrating water, pCO₂(w) remains relatively stable, and DIC is primarily controlled by equilibrium fractionation between CO₂(aq) and HCO₃⁻ (+9‰, [61]), yielding a theoretical lower limit of drip water δ¹³C-DIC of approximately −13.8‰. The measured δ¹³C-DIC values at fast drip sites during the rainy season in Mahuang Cave (e.g., #3 = −13.66‰, #4 = −13.44‰) are close to this theoretical minimum, indicating that their DIC is predominantly inherited from the soil end-member and that fractionation is primarily controlled by dissolution kinetics.

(II) In the near-closed system, the longer flow paths and extended residence time of drip water in the vadose zone restrict CO₂ supply, allowing significant PCP to occur before the water reaches the cave or at the early stage of entry. This process reduces Ca²⁺ and Mg²⁺ concentrations, while preferentially incorporating light ¹²C into the calcite lattice, leaving the residual DIC enriched in heavy ¹³C [10]. Once the drip water enters the cave, if pCO₂(air) < pCO₂(w), CO₂ degassing occurs, and ¹²C-CO₂, with its faster diffusion rate, preferentially escapes, further increasing the δ¹³C-DIC of the remaining water. Therefore, CO₂ degassing is the primary driver of heavy δ¹³C-DIC in the closed system, whereas PCP indirectly facilitates this process by prolonging water–air contact time and increasing calcite supersaturation. This interpretation has been corroborated by monitoring studies in other caves [47,62].

Theoretical calculations further validate this inference. Soil CO₂ (δ¹³C ≈ −22.8‰), after Rayleigh diffusion fractionation (+4.4‰, [54]), yields a δ¹³C value of –18.4‰; following water–rock equilibrium fractionation (+9‰, [61]), the theoretical δ¹³C-DIC is calculated as −9.4‰, which is in excellent agreement with the measured mean δ¹³C-DIC in Mahuang Cave (−9.1‰). Moreover, the difference between the measured mean (−9.1‰) and the theoretical lower limit of direct dissolution (−13.8‰) is +4.7‰, which is consistent with the diffusion fractionation factor (+4.4‰). Furthermore, the mean cave air δ¹³C-CO₂ in Mahuang Cave is −13.64‰, and that of atmospheric CO₂ is −10.88‰ [37]. If the water were to reach equilibrium with these two CO₂ sources, the theoretical δ¹³C-DIC values would be −4.64‰ and −1.88‰, respectively. Thus, the measured δ¹³C-DIC values are in close agreement with the theoretical value calculated from the diffusion-fractionated soil CO₂ end-member, and, together with the fact that they are significantly higher than the theoretical lower limit of the open system, this confirms that the combined effects of CO₂ degassing and PCP in the closed system are the dominant mechanisms driving the marked enrichment of heavy δ¹³C-DIC. In addition, Figure 9 shows that the mean [Ca²⁺+Mg²⁺]/[HCO₃⁻] equivalent ratio is 1.19 (> 1.0), providing clear evidence for significant PCP occurrence in Mahuang Cave.

5.3. Element Ratios as Indicators of PCP and Their Relationship with δ¹³C Fractionation

The Mg/Ca and Sr/Ca ratios are effective proxies for tracing PCP intensity. During PCP, Ca²⁺ preferentially enters the calcite lattice, while Mg²⁺ and Sr²⁺ remain in the liquid phase because their partition coefficients are far smaller than that of Ca²⁺, leading to synchronous increases in Mg/Ca and Sr/Ca in the residual water. Specifically, Sr²⁺, owing to its larger ionic radius relative to Ca²⁺, causes lattice distortion upon incorporation and thus has difficulty replacing Ca²⁺; Mg²⁺, although smaller than Ca²⁺ in ionic radius, has a high hydration energy, dehydrates with difficulty, and has a coordination preference that does not fully match the calcite lattice, so it is likewise not readily incorporated into the lattice [14].

Data from Mahuang Cave (Figure 10) show that (a) Ca²⁺ shows no significant correlation with Mg/Ca (R = −0.06, p > 0.05), whereas (c) Ca²⁺ shows a significant negative correlation with 100 × Sr/Ca (R = −0.52, p < 0.05), indicating that Ca²⁺ input from wall-rock dissolution and consumption by PCP are approximately balanced, while Sr²⁺ accumulates in the liquid phase due to its much lower partition coefficient compared to Ca²⁺ [14]. (b) Mg²⁺ shows a significant positive correlation with Mg/Ca (R = 0.86), and (d) Sr²⁺ also shows a significant positive correlation with 100 × Sr/Ca (R = 0.77), confirming that PCP is the common mechanism driving the increase in both ratios [8,14]. Notably, the correlation strength for Sr (R = 0.77) is slightly lower than that for Mg (R = 0.86), suggesting that PCP is not the sole controlling factor for Sr/Ca [63]. This may be related to external Sr inputs (e.g., dust deposition, silicate weathering) or incongruent dissolution of dolomite [44].

Figure 10. Correlations of Ca²⁺, Mg²⁺, and Sr²⁺ concentrations with Mg/Ca and 100 × Sr/Ca. (a) Ca²⁺ vs. Mg/Ca; (b) Mg²⁺ vs. Mg/Ca; (c) Ca²⁺ vs. 100 × Sr/Ca; (d) Sr²⁺ vs. 100 × Sr/Ca. The black solid lines represent linear regression fits.

Climatic differences between dry and wet years significantly modulated PCP intensity and its covariance characteristics (Figure 11): (a) The overall data fit shows a weak positive correlation between Mg/Ca and Sr/Ca (slope = 0.04, R = 0.26, p < 0.05), reflecting that PCP acts as a common mechanism driving the concomitant increase in both ratios [14,64]. (b) In wet years (2019–2020, 2024), the Mg/Ca–Sr/Ca correlation was weak (R = 0.2), indicating that rapid runoff shortened residence time and limited PCP development, resulting in reduced covariance between the two ratios and clustering of data points in the low-ratio range [65]. (c) In dry years (2022–2023), the correlation was significantly enhanced (R = 0.65), with the slope nearly doubling (slope = 0.07), indicating that precipitation deficit prolonged WRI through slow baseflow, allowing PCP to develop fully and resulting in highly synchronous enrichment of Mg²⁺ and Sr²⁺ with a strong linear covariance [8].

Figure 11. Relationship between Mg/Ca and Sr/Ca ratios: (a) all data; (b) wet years (2019–2020, 2024); (c) dry years (2022–2023). Regression lines and correlation coefficients are shown for each panel. The black solid lines represent linear regression fits.

The fractionation effect of PCP on δ¹³C-DIC was further modulated by the spatial differentiation of hydrological pathways (Figure 12): (a, b) Overall data show weak positive correlations among the three variables, but pronounced spatial differentiation exists among sites. Site #6 (percolation flow) exhibited high Mg/Ca (0.81), high Sr/Ca (1.4), and heavy δ¹³C-DIC (−7.92‰), indicating well-developed PCP under closed-system conditions. Site #3 (shaft flow) showed low ratios (0.42, 0.49) and light δ¹³C-DIC (−10.64‰), reflecting weak PCP and dominant inheritance of soil CO₂ signals. Sites #1 and #2 (matrix flow) displayed intermediate characteristics. Site #4 (subcutaneous flow) exhibited the greatest variability, reflecting the influence of seasonal flow-regime transitions. Notably, site #5 (underground river), despite its high Sr/Ca (1.43), had relatively low δ¹³C-DIC (−8.45‰), showing decoupling between the two proxies. This suggests that its elevated Sr/Ca originated from dissolution of deep Sr-bearing minerals rather than from PCP dominance, indicating that Sr/Ca alone is not a reliable proxy for PCP. Comparison between winter and summer further revealed that (c, e) in summer, the data points for each site were more dispersed, with reduced covariance and lighter δ¹³C, whereas (d, f) in winter, covariance was enhanced and δ¹³C was heavier, highlighting the control of hydrological pathway differences on isotope fractionation.

Figure 12. Relationships between δ¹³C-DIC and Mg/Ca and Sr/Ca. (a,b) All data; (c,e) summer; (d,f) winter.

5.4. Spatial Differentiation of δ¹³C-DIC Under Different Hydrological Pathways

The karst critical zone is governed by lithology, fractures, and hydrological heterogeneity [66], with percolation generally exhibiting mixing behavior [67,68] and forming spatial differentiation between high- and low-flow subsystems [43,69]. As shown in Figure 13, the surface environment dominates the initial solute input, while the aquifer modulates PCP intensity through differences in flow paths and residence times, ultimately manifesting as three types of δ¹³C-DIC spatial differentiation in Mahuang Cave: fast, slow, and mixed flow.

Figure 13. Conceptual diagram of the karst critical zone in Mahuang Cave. Units in the table: discharge (mL/s); Ca²⁺, Mg²⁺and HCO₃⁻ (mmol/L); Mg/Ca and 100 × Sr/Ca ratios (mol/mol); δ¹³C-DIC (‰). Orange arrows represent warm air, dark blue arrows represent cold air, and light blue arrows represent the flow direction of the subterranean river. S1–S3 and #1–#6 denote soil and cave monitoring sites, respectively. (a) Wet season and (b) dry season schematic diagrams of hydrological pathways at drip sites, redrawn based on the conceptual framework of refs. [19,66]. The letter “w” (water phase) denotes the water-filled portion of the aquifer; “g” (gas phase) represents the ventilated air space; the rectangular box indicates the assumed reservoir boundary; the horizontal line represents the water table; blue arrows indicate flow direction, with arrow thickness representing relative flow velocity; horizontal black arrows represent lateral flow or lateral recharge/discharge; vertical double-headed black arrows denote water level fluctuations or water mixing; curved (rotational) arrows indicate the mixing and circulation between new and old water within the reservoir. Note: Site #5, which is an underground river, is not included in this drip site conceptual model. The table summarizes the response characteristics of hydrochemical processes.

(I) Fast flow paths efficiently transmit soil CO₂ signals. Site #3, located in the high-flow subsystem and dominated by shaft–fracture preferential flow, exhibited the highest drip rate (4.05 mL/s) and the lowest Ca²⁺ (1.18 mmol/L), HCO₃⁻ (2.81 mmol/L), and Mg/Ca (0.41), along with the lightest δ¹³C-DIC (−10.63‰), indicating incomplete carbonate dissolution and typical open-system characteristics. Site #4, a seasonal subcutaneous flow, exhibited a high drip rate (7.92 mL/s) with extreme variability (CV = 205%) and was the most sensitive to external climate forcing: during the rainy season, it responded rapidly to precipitation with lighter δ¹³C-DIC, whereas during the dry season, reduced recharge—or even cessation of flow—enhanced PCP and shifted isotopes toward heavier values.

(II) Slow flow paths, combined with PCP effects, drive isotope enrichment. Sites #1, #2, and #6, located in the low-flow subsystem and dominated by matrix seepage, exhibited drip rates of only 0.07–0.09 mL/s, with prolonged residence time in the vadose zone and sufficient WRI. Among them, site #6 showed the highest Ca²⁺, Mg²⁺, and HCO₃⁻ concentrations, as well as the highest Mg/Ca (0.8), 100 × Sr/Ca (1.4), and δ¹³C-DIC (−7.92‰), indicating that the seepage path involved dolomite dissolution, and the continuous accumulation of PCP effects along the upstream path resulted in heavy δ¹³C-DIC [41,60,64].

(III) Mixed flow did not exhibit isotope enrichment effects. Site #5, an underground river that receives recharge from #4 and a deep waterfall, is characterized by multi-source mixing [68]. Although it had the highest 100 × Sr/Ca (1.43), its δ¹³C-DIC (−8.43‰) was heavier than most drip sites but still lighter than that of #6. The progressive enrichment effect observed in Furong Cave, where “pool water δ¹³C-DIC is approximately 2‰ heavier than drip water” [17], was not observed here. This is because the waterfall, carrying a lighter isotopic signal (estimated −10‰ to −11‰), mixed with the heavier signal from #4 (−8.69‰), lowering the overall mean. In addition, the gentle flow regime of the underground river reduced CO₂ degassing efficiency, suppressing δ¹³C enrichment [56], and preventing its δ¹³C from reaching the heavy values observed at #6, where intense PCP occurred. Furthermore, the underground river at site #5 receives inputs of exogenous acids (H₂SO4, HNO₃) from overlying agricultural activities, which further complicates the isotopic signals [33] and leads to decoupling between Sr/Ca and δ¹³C-DIC.

In summary, the spatial differentiation of δ¹³C-DIC in Mahuang Cave indicates that, under the heterogeneous conditions of the karst critical zone, fast, slow, and mixed flow modify δ¹³C signals through distinct pathways. Therefore, the interpretation of this proxy should be site-specific.

5.5. Differences in Carbon Cycling Between Dolomite and Limestone Caves and Their Implications

To further reveal the differences in carbon cycling between dolomite and limestone caves, we selected six limestone caves (Shawan Cave, Liangfeng Cave, Furong Cave, Xueyu Cave, Qianfo Cave, and Jiguan Cave) from the monsoon region of China for comparison, following the principles of cave intercomparison [70]. The comparison parameters include cave temperature and humidity, soil and cave CO₂ concentrations and their δ¹³C values, and drip water DIC-δ¹³C (Table 4).

Table 4. Comparison of cave characteristics and carbon-related parameters between Mahuang Cave (dolomite) and six typical limestone caves.

Cave Name

City/Province, CN

Latitude (°N)

Longitude (°E)

Dominant Lithology (Formation)

Cave Type

Entrance Elevation (m a.s.l.)

MAP (mm)

MAT (°C)

Vegetation

Mahuang Cave

Suiyang, Guizhou

28°14′

107°17′

Dolomite, Dolomite limestone (Cambrian–Ordovician)

N

720

1210

15.5

Evergreen and deciduous broad-leaved mixed forest

Shawan Cave

Puding, Guizhou

26°15′

105°43′

Limestone (Triassic)

N

1158

1367

15.1

Mosaic of trees, shrubs, and grasses

Liangfeng Cave

Guilin, Guangxi

25°11′

110°31′

Limestone (Devonian)

N

183

1868

18.5

Evergreen broad-leaved forest, shrubland

Furong Cave

Wulong, Chongqing

29°13′

107°54′

Limestone, Dolomite (Cambrian–Ordovician)

S

480

1086

18.3

Evergreen broad-leaved forest, shrubland

Xueyu Cave

Fengdu, Chongqing

29°46′

107°47′

Limestone (Triassic)

S

233

1072

17.5

Evergreen broad-leaved forest, shrubland

Qianfo Cave

Ningxiang, Hunan

28°13′

112°07′

Limestone (Devonian)

N/S

219

1458

18.1

Evergreen broad-leaved forest

Jiguan Cave

Luanchuan, Henan

33°46′

111°34′

Chlorite marble (protolith: limestone; Sinian)

S

900

819

13.0

Secondary deciduous broad-leaved forest

Cave Name

Soil Thickness (cm)

Ceil. Thickness (m)

T_cave (°C)

RH_cave (%)

Soil CO2 (ppmv)

Soil δ13C-CO2 (‰)

Cave CO2 (ppmv)

Cave δ13C-CO2 (‰)

δ13C-DIC (‰)

Reference

Mahuang Cave

>100

80–130

10.9–15.9 (14.2)

85–96

753–50,000 (12,517)

–26.89 to –13.17 (–22.75)

318–974 (538)

–20.05 to –17.96 (–13.64)

–13.66 to –2.14 (–8.98)

This study

Shawan Cave

20–50

<30

16.06–18.24

—

1056–41,497

–24.36 to –17.93 (–21.34)

646–25,181

–23.68 to –18.59 (–21.78)

(–12.83)

[18,71,72]

Liangfeng Cave

0–100

20–30

7.8–22.2 (18.7)

100

3288–8384 (5575)

–23.21 to –20.09 (–22.42)

417–707 (533)

–18.9 to –9.9 (–12.87)

–15.79 to –4.07 (–10.18)

[73, 74,75,76]

Furong Cave

20–100

300–500

15.5–19.1 (17.5)

90–100

592–16,644 (4233)

–18 to –17

1045–1537 (1284)

–21.3 to –15.0 (–18.5)

–14.7 to –1.2 (–11.1)

[9,17,22,29,77]

Xueyu Cave

20–80

150–250

17.2

76.7

4000–17,000 (8890)

–23.9 to –19.3

2000–10,000

–23.6 to –18.2 (–21.05)

–14.68 to –3.82 (–10.83)

[78,79,80]

Qianfo Cave

20–60

—

15.7–18.1

99–100

473–16,691

(–19.35)

372–2492

–23 to –8.6 (–16.7)

–13.6 to –2.7 (–9.4)

[58]

Jiguan Cave

10–60

30–40

14–19.5 (16.4)

>90

885–5047 (2499)

–24.68 to –15.19 (–20.14)

335–4980 (1269)

–22.70 to –14.39 (–18.57)

–15.99 to –0.15 (–10.76)

[81,82]

Note: Values in parentheses denote means; — indicates no data; MAP and MAT denote mean annual precipitation and mean annual temperature, respectively; N = natural cave, S = show cave, N/S = natural cave with partial tourism development.

Comparative analysis shows that the soil δ¹³C-CO₂ in Mahuang Cave (−22.75‰) is close to that of Shawan Cave (−21.34‰) and Liangfeng Cave (−22.42‰), all falling within the C3 plant range, indicating similar sources of soil organic matter. However, carbon isotopic fractionation during transport from soil to cave is evident: the soil CO₂ concentration in Mahuang Cave (12,517 ppmv) is substantially higher than that in Liangfeng Cave (5575 ppmv), Furong Cave (4233 ppmv), and Jiguan Cave (2499 ppmv), primarily due to the accumulation effect of the thick soil layer (>100 cm). Nevertheless, this high soil CO₂ does not translate into high cave CO₂, with an annual mean of only 538 ppmv, which is far lower than those of Furong Cave (1284 ppmv), Jiguan Cave (1269 ppmv), and Xueyu Cave (≥2000 ppmv). The cave δ¹³C-CO₂ (−13.64‰) is significantly more positive than that of Furong Cave (−18.5‰), Xueyu Cave (−21.05‰), and Shawan Cave (−21.78‰), indicating that downward transport of biogenic soil CO₂ is inhibited and atmospheric CO₂ contribution is dominant [37]. More notably, the mean δ¹³C-DIC in Mahuang Cave (−8.98‰) is approximately 2‰ heavier than that in Liangfeng Cave (−10.18‰), Xueyu Cave (−10.83‰), Jiguan Cave (−10.76‰), and Furong Cave (−11.1‰), suggesting that lithology, dissolution kinetics, and cave microenvironments exert control over carbon isotope evolution.

Differences in mineral dissolution kinetics are the primary constraint. Under experimental conditions, the initial dissolution rate of dolomite is 3 to 60 times slower than that of limestone [11]. Limestone dissolution is concentrated along fractures and enlarged along flow paths, resulting in uneven pore distribution; dolomite, by contrast, undergoes diffusive leaching throughout the rock mass, with dissolution morphologies tending toward dispersed crystalline pores and solution vugs, and relatively uniform water-bearing properties [83,84]. Incongruent dissolution of dolomite and calcite further influences hydrochemical characteristics. At the early stage of infiltration, ion release ratios are close to stoichiometric proportions; as WRI proceeds, PCP preferentially removes Ca²⁺, while Mg²⁺ continues to accumulate due to its higher lattice energy and slower dissolution rate, leading to an increase in Mg/Ca with prolonged residence time [41,64]. In relatively closed systems, the early stage of incongruent dolomite dissolution may result in relatively lower Sr/Ca, contrasting with the PCP-driven increase in Mg/Ca [85]; however, in open systems or environments with significant PCP, sustained CO₂ degassing drives preferential calcite precipitation, with Ca²⁺ being largely removed, allowing Mg/Ca and Sr/Ca to increase synchronously [19]. In Mahuang Cave, site #3 (fast fracture flow) exhibits low Mg/Ca and Sr/Ca, whereas site #6 shows high Mg/Ca (0.81) and 100 × Sr/Ca (1.4), indicating the superposition of dolomite dissolution and significant PCP along the slow matrix flow path.

Cave structure, microclimate, and anthropogenic activities further modulate the carbon cycle. The thick soil layer (>100 cm) and thick ceiling (80–130 m) of Mahuang Cave, combined with the dense dolomite structure, significantly limit vertical CO₂ diffusion and convective exchange, in sharp contrast to Shawan Cave (thin soil, thin ceiling) [72]. The cave temperature in Mahuang Cave (14.2 °C) is lower than the surface temperature; this lower temperature increases CO₂ solubility in water, thereby inhibiting drip-water degassing and retaining more CO₂ as DIC in the water. In contrast, the higher cave temperatures at Furong Cave (17.5 °C) and Xueyu Cave (17.2 °C) promote stronger degassing, resulting in higher cave CO₂ concentrations and more complex isotopic fractionation. In addition, anthropogenic disturbances in show caves such as Furong Cave and Xueyu Cave may increase CO₂ fluxes, a distinction that should be made from lithological effects.

Karst carbon cycling is characterized by pronounced spatial heterogeneity. The comparison reveals that Mahuang Cave exhibits a pattern of “high soil CO₂—low cave CO₂—heavier δ¹³C-DIC.” Extrapolating the carbon isotope and hydrochemical evolution models established in limestone areas directly to dolomite areas may obscure the control of lithology on carbon cycling pathways. These findings suggest that future research on karst carbon cycling should pay attention to the following controlling factors: (1) lithological differences and their corresponding dissolution kinetic mechanisms; (2) the barrier effect of soil–ceiling structure on vertical CO₂ transport; (3) the modulation of degassing–dissolution equilibrium by cave microclimate; and (4) potential disturbances of cave carbon cycling by tourism activities. Only by examining carbon cycling processes under the multi-dimensional constraints of lithology, structure, and microclimate can the regional representativeness and mechanistic transferability of karst carbon cycling studies be enhanced. Owing to the scarcity of long-term monitoring data for dolomite caves and the lack of consistency in monitoring parameters, this study could not perform a cross-regional comparison of dolomite caves. Future research should therefore select representative dolomite caves for systematic comparison.

6. Conclusions

(1) Temperature and precipitation are the primary factors controlling the seasonal and interannual variations in cave water chemistry and δ¹³C-DIC. As a background framework for interannual climate variability, ENSO regulates regional hydrothermal conditions, thereby indirectly influencing soil CO₂ input efficiency and drip water residence time, and in turn modulating cave water chemical composition, cave air CO₂, and the carbon components of DIC in water. Major ion concentrations show higher values in the warm season and lower in the cold season, while δ¹³C-DIC is significantly lighter during El Niño and markedly heavier during La Niña, exhibiting a certain interannual correspondence with ENSO phase changes. This reflects potential linkages among surface ecological processes, the soil carbon pool, and drip water responses. These results indicate that the groundwater system is highly sensitive to climate change, with antecedent hydrological conditions exerting significant control over subsequent hydrochemical response patterns, and that cave systems have the potential to record interannual-scale climate variability.

(2) The δ¹³C-DIC in Mahuang Cave is jointly regulated by water–rock–gas interactions and PCP and exhibits spatial differentiation under different hydrological pathways. Fast fracture flow, with short residence times, efficiently transmits light carbon signals from soil CO₂ in an open system. In contrast, slow matrix flow, characterized by sufficient WRI and enhanced PCP, exhibits significantly heavier isotopic signatures resulting from the combined effects of CO₂ degassing and PCP in a closed system, forming an isotopic spatial gradient of “fast–open to slow–closed.” The multi-source mixing in the underground river leads to decoupling between Sr/Ca and δ¹³C, cautioning that high Sr/Ca alone is not a definitive indicator of PCP. These hydrological findings provide a mechanistic constraint framework for the paleoclimatic interpretation of speleothem δ¹³C.

(3) In contrast to limestone caves, the carbon isotope responses in Mahuang Cave exhibit distinct differences, characterized by a “high soil CO₂—low cave CO₂—heavier δ¹³C-DIC” pattern. These differences arise from the slow dissolution kinetics of the dolomite mineral phase, the inhibition of vertical CO₂ transport by the thick ceiling, and the limited degassing in the low-temperature cave environment. Therefore, when using cave carbon isotopes and hydrochemical parameters to interpret karst carbon cycling processes, lithological differences should be fully considered, and observation models established in limestone areas should not be directly extrapolated to dolomite areas. The spatial generality of this study’s findings still needs to be validated with additional monitoring data from dolomite caves.