Hydrogeochemical Dynamics and Response of Karst Catchment to Rainstorms in a Critical Zone Observatory (CZO), Southwest China

Karst water is vital for local drinking and irrigation but is susceptible to contamination. Hydrochemistry, which is highly related to carbonate weathering in karst catchments, can affect water quality and respond rapidly to climate change. In order to explore hydrogeochemical sources, dynamics, and their responses to rainstorms, rainwater, throughfall, hillslope runoff, surface water, and groundwater were sampled synchronously during rainstorms at a karst Critical Zone Observatory (CZO), Southwest China. Results showed that the total dissolved solids (TDS) concentration in throughfall increased by 30.1 ± 8.0% relative to rainwater, but both throughfall and rainwater contributed little to TDS in surface water and groundwater compared with terrestrial sources. Hydrochemistry in surface water and groundwater was diluted by rainstorms but displayed chemostatic responses with different intensities to increasing discharge. This is possibly regulated by hydrogeological conditions, available sources of various solutes, and the difference between solute concentrations before and after rainstorms. Ca2+ and Mg2+ dynamics were mainly regulated by carbonate weathering, gypsum dissolution, and gypsum-induced dedolomitization (geological sources), which also affect Ca2+, Mg2+, and SO42- in deep confined groundwater draining a gypsum stratum. For HCO3-, CO2 from respiration and microbiologic activities is one dominant contributor, especially for spring. The chemostatic behaviors of NO3-, Cl−, and K+ were related to agricultural activities, especially in surface water. These controls on hydrochemistry may already exist as hillslope runoff occurs, which has be further demonstrated by principle component analysis (PCA). The heterogeneous permeability of epikarst can affect the mixture of groundwater from different sources and flowing pathways, enabling hydrochemistry at different hydrogeological conditions to display discrepant responses to rainstorms. The epikarst aquifer with high permeability is susceptible to changes in external environment, such as rainstorms and agricultural activities, increasing the potential risk of water environment problems (chronic pollution of nitrogen and high hardness of water) during a certain period. Drinking water safety thus deserves consideration in the agricultural karst catchment.

such as rainstorms and agricultural activities, increasing the potential risk of water environment problems (chronic pollution of nitrogen and high hardness of water) during a certain period. Drinking water safety thus deserves consideration in the agricultural karst catchment.

INTRODUCTION
Karst landforms cover 2.2 × 10 7 km 2 and are scattered around the world, especially in Southeast Asia, South America, and the Mediterranean coasts (Yuan and Cai, 1988;Ford and Williams, 2007). Southeast Asia is the largest continuous karst region, centered on Southwest China (Zhao and Seip, 1991). Comprised of chemically soluble rocks with large passages or network of conduits and caves inside, karst aquifers are very permeable and can store and transport large amounts of water, supplying drinking and irrigation water to ∼25% of the world's population (Ford and Williams, 2007). However, the increasing water demand (residential, industrial, and agricultural use) can cause water shortage and water quality degradation. Additionally, the special hydrogeological conditions (high permeable soil/rock systems with caves and fractures inside) of karst systems benefit water and solutes migration, making the CO 2 -H 2 O-CaCO 3 system sensitive to hydrological changes (Yuan and Zhang, 2008;Beaulieu et al., 2012), and making karst aquifers vulnerable to contamination (Kačaroglu, 1999). Researching karstic hydrochemical dynamics during rainfall storms is conducive to understanding the transfer of contaminants and solutes and their responses to climate change.
Intense carbonate weathering and mixing of "new" and "old" water can alter solute concentrations in karst water (Gabet et al., 2006;Basu et al., 2010). Solute concentrations also normally show distinct seasonal patterns due to hydrological variations and some biogeochemical processes (Han and Liu, 2004;Lang et al., 2006;Zhong et al., 2020). Therefore, hydrogeochemical behaviors in karst catchments need to be demonstrated by high-resolution information that can capture key processes. High-frequency field campaigns during rainstorms are thus necessary to analyze the dynamic response of chemical compositions to hydrological variations and its potential effects on water quality as well as public health. The dynamic behaviors and corresponding controls of hydrochemistry in river and/or groundwater have been reported (Koger et al., 2018;Piazza et al., 2018;Correa et al., 2019). However, there are few studies conducting high-frequency sampling to analyze hydrochemistry in both surface water and groundwater in karst regions of Southwest China. Additionally, rainwater, one important source of surface and ground water, also plays a significant role in hydrochemical dynamics (Polkowska et al., 2005;Ma et al., 2017). But most research only pays attention to sources of solutes in rainwater Lu et al., 2011;Wu et al., 2012;Zhang et al., 2012;Rao et al., 2017;Zhou et al., 2019); the dynamic variation of hydrochemistry among different waters (rainwater, throughfall, hillslope runoff, surface water, and groundwater) are rarely explored during rainstorms.
Rainwater may play a more important role in karst catchments than non-karst systems, since the acidic rainwater may also dissolve carbonate minerals and then alter the hydrochemistry as well. Analyzing hydrochemistry of different water bodies in spatial profile can offer comprehensive information to understand hydrogeochemical dynamics in Earth's Critical Zone. This is necessary for modeling the migration processes of hydrochemistry and better implementing cost-effective water quality management in karst catchment.
Against this background, high-frequency sampling for surface water and spring was conducted during rainstorms at a karst catchment in Southwest China. Rainwater, throughfall, hillslope runoff, and well water were also synchronously collected, to analyze (1) the dynamic response of surface water and groundwater to rainstorms and corresponding controls and (2) the dynamic variations of hydrochemistry among different water bodies at a catchment scale and corresponding controls.
The elevation of this catchment decreases from east of 1,520 m to west of 1,320 m (outlet). Carbonate rocks dominate the lithology, over which Quaternary soils are unevenly distributed. Limestone is the dominant geological strata in the higher elevation area with 150-200 m thickness, and the underlying layer is impermeable marlite strata (Zhang et al., 2019). Soil is thin (mean < 50 cm) in hills but thicker (40-100 cm) in valley depressions. Deciduous broad-leaved forest and scrubgrassland, occupying 83% of total land use, grow mainly on the mountains. Agricultural paddy and dry land are spread in valley depressions, accounting for about 3 and 14% of the catchment area, respectively (Qin et al., 2020). The catchment outlet is in the valley depression surrounded by scattered hills. There are three FIGURE 1 | Location of the study area and sampling sites. W1, W2, and W3 refer to three observation wells.
observation wells with different depth distributed in depression (Figure 1). The depth (below the ground surface) of well W1 and well W3 are 20 and 12 m, respectively. The permeability of aquifer around W1 is low (most resistivity > 1,000 m) but that around W3 is high (most resistivity < 100 m) (Chen et al., 2018). The well W2 (depth > 20 m) is fed by deep confined spring draining gypsum stratum and scarcely suffers outside disturbance. The 3wells can thus reflect the circumstances at different depth in the karst structure.

Sampling and Measurement
High-frequency sampling was conducted at the outlet of spring (SP) and surface water (SU) during rainstorms from June 11 to 17, 2018 (Figure 1). Well water samples of W2 were synchronously collected at the outlet through a drain line. Well water at different depths of W1 and W3 were sampled before (B), during (D), after (A) and on the fourth days (F) after the rainstorms. Meanwhile, rainwater (RA), throughfall (TH), and hillslope runoff (HI) were also collected on surrounding hills.
At the sampling time, water temperature (T), dissolved oxygen (DO), electrical conductivity (EC), and pH were measured in the field by a portable multiple parameter sensor (WTW, Multi 3630 IDS). All water samples were collected in duplicate in dry and clean bottles that were rinsed twice beforehand with in-situ water and stored in darkness at 4 • C until analysis. One part was used for measuring HCO − 3 concentration through acid titration with 0.02 mol L −1 HCl. The other was filtered through 0.22 µm filters to measure major ions (Ca 2+ , Mg 2+ , Na + , K + , SO 2− 4 , Cl − , NO − 3 ) by the Ion Chromatography (Thermo Scientific, Dionex Aquion). Blank test and sample duplicates were conducted during analytical processes to perform quality assurance and control. The blank was lower than the method detection limits and the relative standard deviation (RSD) of above analyses were all within 5%.
Total dissolved solids (TDS) were calculated by the sum of eight ions measured above. The bracket "[]" was used to symbolize the concentration of solutes, e.g., Ca 2+ concentration is simply described as [Ca 2+ ]. To more accurately compare solute concentrations within this study or between different researches, discharge-weighted concentration (DWC) of solutes was calculated as follows: where Q i is the discharge at timestep i and C i is the synchronous concentration of solutes.

RESULTS
During the sampling period, two considerable rainstorms occurred continuously, with amounts of 30.4 mm on June 11 and 24 mm on June 12 (Figure 2). The pH value of rainwater here is 6.27 ± 0.03, which increases by 0.44 in throughfall (Supplementary Table 1 Table 1). The first four ions dominate chemical compositions, accounting for 94.1%. After passing through the vegetation canopy, [NO − 3 ] decreases by 9.13 ± 6.13 µmol L −1 but other ions increase with different degrees. The TDS concentration in TH is slightly higher than that in RA (24.2 vs. 18.6 mg L −1 ). Both RA and TH show lower total cation charge During rainstorms, discharge varied from 0.4 × 10 −3 to 179.5 × 10 −3 m 3 s −1 at SP outlet and from 0 to 378.1 × 10 −3 m 3 s −1 at SU outlet, and high discharge normally accompanied high DO and low EC (Figure 2). SU samples have higher DO and pH but lower EC compared with SP samples, and all of them have small coefficient of variation (CV < 10%) (Supplementary Table 1). TDS in SU and SP range from 331 to 397 mg L −1 and from 345 to 475 mg L −1 , respectively. The DWC of ions in SU follow the order of [ Table 1). In SP, the DWC of HCO − 3 , Ca 2+ , and Mg 2+ are higher than that in SU, at 3,812, 2,053, and 436 µmol L −1 , respectively. While the DWC of NO − 3 , Cl − , and K + in SP are lower than that in SU, at 217, 61.0, and 21.8 µmol L −1 , respectively. Except NO − 3 , parameters have higher CV in SP than in SU (Supplementary Table 1). The NICB of both SU and SP samples are within ±10% (Figure 3).
The water levels in W1 and W3 fluctuated by 1.6 and 3.7 m, respectively (Supplementary Figure 1). All ion concentrations in W1 have relatively small spatial variations in vertical profiles and temporal variations among the four sampling campaigns [before (B), during (D), after (A), and on the fourth days (F) after the rainstorms], as reflected by their CV in Supplementary Table 1 and box plots in Figure 4. In vertical profiles of W3, however, ion concentrations normally fluctuate more widely, except the third sampling (A) when almost all ions have the minimum CV. The mean ion concentrations in W1 follow the order of [HCO . The light gray zone stands for that the net inorganic charge balance (NICB = [TZ + -TZ − ] * 100/ [TZ + + TZ − ]) is within ±10. RA, TH, HI, SU, and SP refer to rainwater, throughfall, hillslope runoff, surface water, and spring water, respectively. W1, W2, and W3 refer to three observation wells.
respectively. The water velocity of W2 flowing from the drain pipeline was almost unchanged. All ion concentrations in W2 are relatively stable at different depths or moments throughout the whole rainstorms, following the order of [SO 2− 4 ] (11,560 µmol Table 1).

DISCUSSIONS Dynamic Responses of Surface Water and Groundwater to Rainstorms
Solute concentrations normally vary inversely with discharge, which can be well reflected by power-law function (Godsey et al., 2009;Musolff et al., 2015;Zimmer et al., 2019;Ackerer et al., 2020). To demonstrate relationships between concentrations and discharge, power-law function was adopted as follows: where C i and Q i are instantaneous solute concentrations (C) and discharge (Q) at timestep i. The exponent "b" is an indicator reflecting the sensitivity of solute concentrations to discharge variation (Godsey et al., 2009). When b = −1, solute fluxes (the product of C and Q) equal the constant "a" and the dilution effect occurs; when b is near-zero, chemostatic behavior occurs; when b > 0, the flushing effect arises. In this study, the C-Q fitting curves generate "b" closer to 0 for HCO − 3 , Ca 2+ , and Mg 2+ than other ions (Figure 5), indicating that there are more supplementary inputs alleviating dilution of increasing discharge, resulting in stronger chemostatic response (Clow and Mast, 2010). Except NO − 3 and Na + , other ions all exhibit weaker chemostatic behavior in SP than SU (Figure 5). This observation may be largely related to their initial concentrations at or before the beginning of rainstorms (low discharge). In this catchment, there is no flow in ground surface at no-rain days, but the discharge at the spring outlet occurs almost all the time.
During low-flow periods, most groundwater is retained in underground networks (Zhang et al., 2019) and has enough time to interact with carbonates, leading to the accumulation of solutes and products. At the early stage of rainstorms, rainwater infiltrates into epikarst initially and is intercepted in the aquifer (Williams, 2008), mixing with "old" water in soil or matrix. This process usually accompanies the dissolution of soil CO 2 . As rainstorms go on, rainfall is enough to generate hillslope runoff and surface runoff, and "old" water is extruded and gradually replaced by the percolating "new" water. At this stage, the dominant pathway of underground flow is likely to change from matrix to conduit after saturating epikarst aquifer; the supplementary sources of solutes in spring can be quickly mobilized by the increasing discharge along pathways (Qin et al., 2020); the excess infiltration water would quickly pass through conduits and flow out at the outlet (Trček, 2007). This process can shorten the available time for both spring transportation and fluid-rock interaction (Tipper et al., 2006;Chen et al., 2018), generating differences on the transportation of both fluid and solutes (Bakalowicz, 2005;Bowes et al., 2005). Therefore, solute concentrations in SP decrease with increasing discharge and are gradually close to that in SU, and the effect of increasing discharge on solute concentrations is smaller in SU than SP. At the end of rainstorms, most available proximal sources near to or even within the flow pathways have been flushed. Hillslope runoff and surface runoff gradually decrease until they stop; spring runoff also decreases but normally maintains low discharge with slow velocity due to the recharge of matrix flow. Throughout the complete rainstorm, flow velocity, pathways, and the time for transportation and water-rock interaction all have been changing, leading to variations of sources and its contributions to solutes.
As highlighted by Thompson et al. (2011), the interpretation of b ≈ 0 might be incomplete when the concentration variability is small. The ratio of CV of concentrations to CV of discharge (CV C /CV Q ) is an alternative metric to statistically assess chemostatic behavior (Thompson et al., 2011;Musolff et al., 2015;Duncan et al., 2017;Zimmer et al., 2019), which could be calculated as follows: where µ and σ represent the mean and the standard deviation of concentrations (µ C and σ C ) and discharge (µ Q and σ Q ), respectively. The condition of −0.2 < b < 0.2 and CV C /CV Q < 1 is deemed a criterion of chemostatic behavior (Thompson et al., 2011;Zimmer et al., 2019). A CV C /CV Q > 1 indicates a relatively bigger variability in concentration than discharge, Frontiers in Water | www.frontiersin.org which is usually referred to as "chemodynamic" (Musolff et al., 2015). According to the export regime classification system proposed by Musolff et al. (2015), plot of b vs. CV C /CV Q can be used to intuitively categorize and compare various solutes in catchment or between catchments (Figure 6). HCO − 3 , Ca 2+ , and Mg 2+ in both SU and SP have near-zero b values and low CV C /CV Q ratios (<< 1), indicating chemostatic behaviors. The three ions in Yu River  and Xijiang River (Zhong et al., 2018) display similar export regime with this study, but their chemostatic behaviors are relatively weaker. Conversely, there are scenes of b > 0 for the three ions in the Bode River of central Germany (Musolff et al., 2015) and Ca 2+ in SU of Los Alamos (Koger et al., 2018), showing an enrichment trend. This phenomenon indicates their large potential supplements with increasing discharge. Other ions in this study display weaker chemostatic behavior, especially Na + and NO − 3 in SU and Na + , SO 2− 4 and NO − 3 in SP, implying that the pool of these ions is more susceptible to dilution, which is mainly due to their slower release or lower quantities along flowing pathways (Qin et al., 2020). SO 2− 4 in SP is more sensitive to hydrological variation than that in SU. This is similar to the results from Los Alamos (Koger et al., 2018) and signifies that the potential sources of SO 2− 4 in SP could be more quickly mobilized by increasing discharge. Overall, the dynamic response of hydrochemistry to rainstorms or climate change is largely controlled by regional hydrogeological properties (e.g., porosity, thickness, and hydraulic conductivity). It may differ between different conditions and needs to be analyzed using detailed local information. In this study, chemostatic is the predominant export regime of solutes in SU and SP.
During rainstorms, hydrochemistry in 3-wells respond differently to rainstorms, which is quicker in W3. This is mainly attributed to the higher permeability of epikarst around W3 (Chen et al., 2018). In the epikarst with better permeability, the relative proportion of vertical flow from infiltration replenishment increases and the lateral flow from "old" water FIGURE 5 | Power-law relationships (C = a Q b ) between concentration (C) of major ions and discharge (Q). Rainwater dilution curves denote ideal conditions that ions are purely diluted by local rainwater. SU and SP refer to surface water and spring water, respectively. FIGURE 6 | Plot of b (the power-law exponent of C=aQ b ) vs. CV C /CV Q (the ratio of CV of concentrations to CV of discharge) for ions in different catchments. SU and SP refer to surface water and spring water, respectively. The shadow areas classify corresponding zones of ions in Bode River catchment (Musolff et al., 2015).
is gradually extruded out with the ongoing rainstorms (Zhang et al., 2019). At the second sampling (D), large inputting water in W3 diluted solute concentrations, excluding , which may be affected by fertilizer application and some biological processes . The highest CV of  in W3 at this time is possibly caused by the variation of dissolved soil CO 2 and soil respiration during infiltration processes (Qin et al., 2020). At the third sampling (A), the relative ratio of vertical flow to lateral flow might remain constant or infiltration water might dominate discharge, leading to small profile variation for solute concentrations in W3. At the fourth sampling (F), water level and chemistries in W3 nearly restored to the situation before the rainstorms. The permeability of epikarst normally weakens with increasing depth (Ford and Williams, 2007). W2 is a deep confined well fed by groundwater draining a gypsum stratum and not susceptible to changes in external environment (e.g., occurrences of rainstorms and agricultural activities) . Therefore, its hydrological environment is stable with smaller [NO − 3 ] and lower CV of most ions.  in W2 is approximately an order of magnitude higher than that in other 2wells; [Ca 2+ ] and [Mg 2+ ] are also higher. It is thus reasonable to deduce the existence of gypsum dissolution and gypsum-induced dedolomitization around W2. Overall, the heterogeneous permeability of epikarst can affect the mixture of groundwater from different sources and flowing pathways, eventually enabling hydrochemistry in groundwater at different hydrogeological  conditions to display discrepant responses to rainstorms or discharge variations.

Dynamic Variations of Hydrochemistry Among Different Spatial Levels
The pH value of RA in this study is higher than that of other karst or forest regions in Southwest China, e.g., Guiyang of 4.5  and Chongqing of 4.75 (Ma et al., 2017), but it is lower than that of some areas in Northwestern China, e.g., Alxa Desert Plateau of 7.6 (Rao et al., 2017) and Urumqi River Valley of 7.0 (Zhao et al., 2008). This is most likely attributed to active carbon circulation and intense carbonate weathering within this karst depression (Qin et al., 2020). These processes can release some alkaline products (e.g., CaCO 3 and MgCO 3 ) into the atmosphere, neutralizing rainwater acidity (Tang and Han, 2019).  ] and [Ca 2+ ] in RA are much lower than that a decade ago (Wu et al., 2012), indicating that acid deposition has decreased after implementing environmental protection policies. This may also account for the higher pH in RA in this study than before. But [SO 2− 4 ] and [NO − 3 ] in RA of this study are higher than that in Central Tibetan Plateau (Li et al., 2007), Eastern Tien Shan (Zhao et al., 2008), and Xisha Islands of South China Sea (Xiao et al., 2016), so there is still the possibility of anthropogenic emissions (NO x , NH 3 , and SO x ).
Migration processes of water among different spatial levels of Earth's Critical Zone can alter ion concentrations to different degrees (Figure 7). The variations of chemical compositions after passing through the vegetation canopy are mainly ascribed to (1) the adsorption of NO − 3 by vegetation (Polkowska et al., 2005); (2) the volatilization of HNO 3 after increasing its exposed area in vegetation surface; (3) the dissolution of Ca 2+ and K + transported from roots to leaves through transpiration and respiration; and (4) the flushing of atmospheric particulates that contain ions (Ca 2+ , Mg 2+ , SO 2− 4 , and K + ) and accumulate in vegetation surface during dry deposition. In karst regions, some weathering products could be released to ambient atmosphere with dust and pedogenesis processes (Lü et al., 2017). This possibly also affects Ca 2+ and Mg 2+ in RA and TH in this karst catchment, as demonstrated by Supplementary Figure 2. However, NH + 4 , one dominant ion in rainwater originating mainly from NH 3 emission of agricultural soil (Zeng et al., 2019), was not measured or considered when calculating TZ + and NICB in this study. This should be responsible for the imbalance of net inorganic charge. The better NICB in TH than in RA indirectly means that the vegetation canopy may intercept some NH + 4 from RA. All these hydrochemical variations above are also the causes of the elevation of pH in TH relative to RA. Solutes in rivers are mostly derived from rock weathering, atmosphere, and anthropogenic inputs (Gaillardet et al., 1999;Qin et al., 2019). In this study, most solute concentrations in SU and SP are much higher than that in RA and TH (Figure 7), indicating little contributions of atmospheric sources compared with terrestrial sources. Carbonate weathering is one major source for this catchment (Supplementary Figure 2). [Ca 2+ ], [Mg 2+ ], and  in HI are far higher than that in RA and TH but closer to that in SU and SP (Figure 7), indicating that the contribution of carbonate weathering and CO 2 dissolution may already exist when hillslope runoff occurs. Their relative contribution ratios are different from that generated during surface and underground processes, as illustrated in Figure 8. Additionally, [Ca 2+ ], [Mg 2+ ], and  in SP are higher than that in SU (Figure 7). One possible explanation is that during the infiltration of flow through epikarst and the spring transportation within an underground conduit, there is dissolution of more soil CO 2 and intense carbonate weathering (Qin et al., 2020). In this agricultural karst depression, NO − 3 characterizes nitrification and owns large input from reductive nitrogen fertilizer and manure . Cl − is deemed to have similar source area or transport pathways with NO − 3 (Qin et al., 2020). K + can also accumulate in soil surface after the application of potash fertilizer due to cation exchange and relative high contents in biomass (Boy et al., 2008). Therefore, NO − 3 , Cl − , and K + all have potential agricultural sources and might be intercepted by epikarst during infiltration processes, leading to lower concentrations in SP than in SU (Figure 7). Overall, the difference of hydrochemistry among different water bodies is related to the double properties of agriculture and karst in this depression catchment. In order to acquire descriptive statistics and explore interactive effects among physic-chemical parameters, principle component analysis (PCA) was performed with varimax rotation. The Kaiser-Meyer-Olkin (KMO) Measure of Sampling Adequacy and Bartlett's Test of Sphericity were conducted and generated a value of 0.738 with the significance of <0.001, indicating the suitability of the dataset for PCA. As shown in Table 1, there are four components (PC, eigenvalues >1), totally explaining 92.095% of the variation. PC1 explains 42.394% of the total variance and has high positive loading values (>0.9) of Ca 2+ , Na + , SO 2− 4 , EC, and Mg 2+ . This indicates that these parameters may follow an alike trend that is possibly due to same controls or sources areas or transit pathways. According to the common properties of these parameters, PC1 could be attributed to geological origins. This also indirectly reflects that EC is mainly controlled by Ca 2+ , Mg 2+ , and SO 2− 4 . The main geological source of SO 2− 4 is gypsum dissolution. PC2 can explain 18.56% of the total variance and includes two positive loadings (K + and T) and one negative loading (DO), indicating that temperature and dissolved oxygen may exert positive and negative influences on the eluviation of K + , respectively. PC2 could be partially attributed to the influence of the mixing of soil and natural factors. NO − 3 and Cl − are two positive loadings of PC3 that can explain 14.551% of the variance and is likely to denote agricultural contributions. Additionally, PC4 can only explain 12.935% variance by pH and HCO − 3 , and HCO − 3 owns a loading of <0.8. Because HCO − 3 is mostly contributed by soil CO 2 that is controlled by respiration and microbiologic activities (Qin et al., 2020), we conclude that PC4 is mainly contributed by biological sources.
Despite the differences of hydrochemical concentrations and controlling factors among different spatial levels, all waters show the dominance of HCO − 3 and Ca 2+ in ions (except W2 draining a gypsum stratum) (Supplementary Figure 3), reflecting the characteristics of karst water. It is worth noting that Chenqi catchment has continual agricultural activities (e.g., fertilization and herding), so topsoil can store or accumulate NO − 3 and Cl − . As a rainstorm occurs, water head of groundwater varies with hydrology, regulating nutrient exchange between matrix flow and conduit flow. Because the transportation from conduits to matrix is faster than that from matrix to conduits , there is potential chronic pollution on karst groundwater in this agricultural catchment. Additionally, the accumulation of Ca 2+ and Mg 2+ in karst groundwater can elevate hardness of water. Both can deteriorate regional water quality during a certain period.

CONCLUSIONS
Rainwater, throughfall, hillslope runoff, surface water, spring, and well water at different hydrogeological conditions were collected synchronously during rainstorms to analyze the dynamic responses of hydrochemistry and its variation at different spatial levels at a karst Critical Zone Observatory (CZO), Southwest China. In this karst depression, pH and TDS in rainwater are 6.71 ± 0.03 and 18.6 ± 1.17 mg L −1 , respectively. After passing through the vegetation canopy, pH increases by 0.4, and most ion concentrations also have increments with different degrees (except [NO − 3 ] which decreases by 4.1-54.0%). Chemical compositions in rainwater and throughfall could be affected by both anthropogenic emission (NO x , NH 3 , and SO x ) and some weathering products (Ca 2+ and Mg 2+ ) released to ambient atmosphere with dust. Wet deposition generally contributes little to hydrochemistry in surface water and groundwater. The influences of CO 2 dissolution and carbonate weathering on hydrochemistry already exist when hillslope runoff occurs, but their relative contribution ratios are different from that generated during surface and underground processes. Throughout the whole rainstorms, TDS in surface water and spring are 331-397 mg L −1 and 345-475 mg L −1 . Hydrochemistry in surface water and spring displays chemostatic responses with different intensities to discharge variations, possibly controlling by available sources and the difference between solute concentrations before and after rainstorms. Carbonate weathering contributes most to Ca 2+ and Mg 2+ , gypsum dissolution and gypsum-induced dedolomitization also regulate Ca 2+ , Mg 2+ , and SO 2− 4 in deep confined well water draining a gypsum stratum. Soil CO 2 from respiration and microbiologic activities is one biological source of HCO − 3 . Agricultural activities are responsible for the chemostatic behavior of NO − 3 , Cl − , and K + , especially in surface water. All these were further demonstrated by principle component analysis (PCA). Epikarst aquifer with low permeability is insensitive to changes in external environment (e.g., occurrences of rainstorms and agricultural activities). High permeability can enhance the response of epikarst aquifer to hydrological variations, deteriorating regional water quality (potential chronic pollution of nitrogen and high hardness of water) during a certain period. Optimal measures of water protection are necessary in agricultural karst catchment.

DATA AVAILABILITY STATEMENT
The original contributions presented in the study are included in the article/Supplementary Materials, further inquiries can be directed to the corresponding author/s.

AUTHOR CONTRIBUTIONS
CQ, S-LL, HD, and F-JY designed the research objectives and interpreted the data and prepared the manuscript. CQ, Z-JW, and JZ carried out the field and laboratory work. All authors discussed the results and commented on the manuscript.

FUNDING
This study was jointly funded by the National Key R&D Program of China (grant number 2016YFA0601002) and the National Natural Science Foundation of China (grant numbers 41571130072 and 41861144026).