Introduction

The World Karst Aquifer Map (WOKAM) project has found that karstified rocks and aquifer systems cover approximately 14.7% of the Earth’s ice-free land (Chen et al. 2017). In karst areas, many surface streams are fed by karst groundwaters, which discharge from aquifer at springs. Karst groundwater is often rich in dissolved inorganic carbon (DIC) because of carbonate dissolution by carbonic acid generated during the hydration of CO2. The elevated DIC concentrations may lead karst groundwater-fed streams to play an important role in regulating regional and global carbon cycles through carbon transfer, CO2 degassing, carbon burial, and carbon assimilation (Atekwana and Krishnamurthy 1998; Covington 2016; Gombert 2002; Jiang and Yuan 1999; Kurz et al. 2013; Liu et al. 2010, 2015; Liu and Zhao 2000; Martin 2017; Martin et al. 2013; Pu et al. 2017; Yuan 1997). In particular, these streams may represent an important carbon sink, as the clear water allows light penetration to stream benthic environments and DIC is assimilated by submerged aquatic phototrophs (de Montety et al. 2011; Jiang and Yuan 1999; Liu and Dreybrodt 2015; Liu et al. 2010; Pu et al. 2017; Zhang et al. 2017a). The assimilated DIC is then buried and sequestered as organic carbon (OC). The net atmospheric CO2 uptake by interactions between water, carbonate minerals, dissolved CO2 (carbonic acid), and aquatic phototrophs on land has been estimated to be as large as 0.477 Pg C/a (Liu and Dreybrodt 2015), and thus a large fraction of the net terrestrial residual sink (i.e., 0.8~1.2 Pg C/year) (Ciais et al. 2013). Therefore, revealing DIC dynamics in karst streams is critical to understanding carbon cycling and its role in regional and global carbon budgets.

Karst aquifers are very sensitive to external environmental changes at a range of timescales, including diel, event, seasonal, annual, and multi-annual (Liu et al. 2015, 2007; Martin et al. 2016; Nimick et al. 2011; Shuster and White 1971; Vesper and White 2004; Zhang et al. 2012). Thus, DIC in karst groundwater-fed streams can display high temporal and spatial variability in coherence with aquifer hydrological variations (Zeng et al. 2016). Diel variations in DIC concentrations is a common phenomenon in a karst stream that results from alternating carbonate dissolution and aquatic metabolism (de Montety et al. 2011; Demars et al. 2015; Khadka et al. 2014; Lynch et al. 2010; Spiro and Pentecost 1991; Tobias and Boehlke 2011). These diel variations occur in all seasons, and they can be as large as seasonal variations (Nimick et al. 2011). Variations in stream flow may also play an important role in regulating karst stream DIC transport and transformation at diel, seasonal, and annual timescales (Atkins et al. 2017; Peter et al. 2014; Zeng et al. 2016). Dry seasons often slow water motion in channels and increase the time available for DIC transformations through mineral precipitation and plant assimilation (Mann et al. 2014; Shin et al. 2011). By contrast, rainfall events during wet season can export significant quantities of DIC from karst aquifer to surface streams, thereby altering the DIC cycling through changes to CO2 gas exchange, turbidity, water temperature, and flow velocity (Looman et al. 2016; Peter et al. 2014). The control of changing flow conditions was poorly constrained, however, because most earlier work occurred during lower-water-level or stable flow periods (de Montety et al. 2011; Jiang et al. 2013; Liu et al. 2015; Parker et al. 2014; Pu et al. 2017; Yang et al. 2015). Only a few studies have reported on DIC or the CO2 diel cycle under the impact of rainfall events in surface streams (Dinsmore et al. 2013; Johnson et al. 2009; Looman et al. 2016; Peter et al. 2014), but they mostly focused on peatland or silicate rock streams. Little is known of how flow variations may alter diel DIC cycling in a karst stream, even though this question is critical for the accurate estimation of what may be a major terrestrial carbon sink.

This study assesses DIC concentration variations during changing flow conditions and how the variations may alter the carbon cycle in a karst system. This assessment is based on data collected from a subtropical karst groundwater-fed stream (Guancun stream (GS)) located in southwest China. The study is based on a 3-day record of high-resolution measurements of diel variations within a stream of water for DIC concentrations, δ13CDIC values, and physicochemical parameters. These results will help to improve evaluations of the amount of carbon that could be sequestered in karst streams and improve the understanding of carbon transportation and transfer with varying hydrological processes in inland water bodies.

Study area

The Guancun surface stream (GSS) is a tributary of the Shimen River and a part of the Rongjiang River drainage system, which is located in Daliang town in Rong’an county of Guangxi Zhuang Autonomous Region, China (Fig. 1). GSS is a typical subtropical headwater stream and almost entirely fed by karst groundwater from an upper Devonian (D3r) limestone aquifer (Pu et al. 2017). The outlet of the Guancun underground stream (GUS) is the head of the GSS with no surface tributaries flowing into it (Fig. 1). The GSS channel is underlain by Lower Carboniferous (C1y) limestone interbedded with dolomite of the Yingtang Formation. The length and average width of the GSS are 1.32 km and 3.5 m, respectively. At base level, the water depth is shallow (0.2~1.2 m). The study area is characterized by a cold–dry winter from late November through March and a hot–rainy summer from April through October and has an annual average temperature of 19.7 °C. The area is dominated by the East Asian Monsoon with an annual average precipitation of 1726 mm, 72% of which occurs in the wet season from late April to early September. As a typical monsoon region, air temperature and precipitation in the GSS catchment co-vary, both of which being high in the wet season and low in the dry season.

Fig. 1
figure 1

Maps of Guancun surface stream and sampling sites. a Photograph is the location of Guancun stream in SW China. b Photographs are the scenes in Guancun stream from groundwater outlet to stream mouth. c Map shows the surface stream flow route and sampling sites in study area (modified from Google Earth 2015) (Pu et al. 2017)

Full size image

This study focuses on monitoring sites for detailed diel monitoring and sampling (Fig. 1). The upstream CK (24° 52′ 10″, 109° 20′ 07″) site is located at a GUS outlet, which receives typical karst groundwater. The downstream LY (24° 51′ 32″, 109° 20′ 01″) site is approximately 1.30 km downstream from CK near the stream mouth and is the site of a gauging station.

Methods

Data for this study was acquired August 17–19, 2013. Two multi-parameter meters (WTW 3430, WTW GmbH, Weilheim, Germany) were set at both the CK and LY sites. Hydrochemical variables including water temperature, pH, dissolved oxygen (DO), and specific conductivity (SpC) were continuously measured in situ at 15-min time intervals at both sites. Resolutions for water temperature, pH, dissolved oxygen (DO), and specific conductivity (SpC) are 0.1 °C, 0.004 pH units, 0.01 mg/L, and 1 μS/cm, respectively. Specific conductivity (SpC) refers to the electrical conductivity corrected from ambient temperature to 25 °C by a nonlinear correction function. Turbidity was measured in situ at 15-min time intervals at the LY site using a YSI 6600 datasonde (Yellow Springs, OH, USA). All monitoring probes for the instruments were calibrated according to manufacturer’s specifications prior to deployment. Weir water stages were also continuously monitored at 15-min time intervals at the LY site using the same YSI 6600 datasonde. These water stage measurements were converted to discharge units using the rectangular-weir-discharge formula (Pu et al. 2017).

From August 17 to 19, 2013, water samples were collected every 2 h at both sites from the mid-channel of the stream using the ISCO 6712 autosampler (Teledyne ISCO, Inc., USA). Water was pumped from 0.1 m below the water surface. Every water sample was temporarily stored in 2-L pre-rinsed high-density polyethylene (HDPE) bottles in the autosampler base after samplings. To keep water samples chilled, the autosampler base was filled with ice for entire sampling time. Unfiltered water samplers were titrated for alkalinity immediately in the field with an accuracy of 0.05 mmol/L using a portable testing kit by Merck KGaA Co. (Germany). Because HCO3− constitutes ~ 90% of DIC over the pH range of 7.3–8.5 (Rice et al. 2012), we used alkalinity measurements as an approximation of DIC concentration for this study. Water from the autosampler bottles was immediately filtered (0.45-μm cellulose acetate membrane) into other small clean bottles for later analysis of major cation (Ca2+, Mg2+, K+, and Na+), major anions (Cl−, SO42−), and δ13CDIC. Samples used to detect cations were acidified with trace metal grade nitric acid (7 M HNO3) to a pH of < 2.0. Samples for δ13CDIC were collected in 25-mL acid-washed dry HDPE bottles, and three drops of HgCl2 were added in order to prevent microbial activity. A portable cooler was used to store all samples in the field. Samples were delivered to a laboratory where they were stored and chilled in a refrigerator at 4 °C until analysis.

Meteorological parameters, including rainfall, air temperature, wind speed, barometric pressure, and solar radiation, were measured using an on-site Vantage Pro 2 weather station (Davis Instruments Corp., USA) from August 16 to 20, 2013. Resolutions of rainfall, air temperature, wind speed, barometric pressure, and solar radiation were 0.2 mm, 0.1 °C, 5%, 0.1 hPa, and 1 W/m2, respectively.

Major anions and cations were measured by an automated Dionex ICS-900 ion chromatograph and an ICP-OES (IRIS Intrepid II XSP, Thermo Fisher Scientific, USA), respectively, using procedures based on APHA 2012 methods (Rice et al. 2012). The calculated errors of charge balance were within ± 5%. The δ13CDIC values of water samples were analyzed using a MAT-253 mass spectrometer coupled with a Gas Bench II automated device. The results are expressed as δ13CDIC (‰) with respect to the Vienna Pee Dee Belemnite (V-PDB) standard with an analytical precision of ± 0.15‰. All lab analyses were carried at the Environmental and Geochemical Analysis Laboratory of the Institute of Karst Geology, Chinese Academy of Geological Science (Pu et al. 2017; Zhang et al. 2017a).

The partial pressure of CO2 (pCO2) and saturation index of calcite (SIc) in the stream waters were calculated through the program WATSPEC (Wigley 1977) according to the hydrochemical data sets, including pH, water temperature, and concentrations of K+, Na+, Ca2+, Mg2+, Cl−, SO42−, and HCO3−.

CO2 fluxes across water–air interface were calculated using a molecular diffusion model (Raymond et al. 2012):

$$ F=k\times \left({\left[{\mathrm{CO}}_2\right]}_{\mathrm{water}}-{\left[{\mathrm{CO}}_2\right]}_{\mathrm{air}}\right) $$
(1)

where F is the CO2 evasion flux (mg/m2/h) between water and atmosphere, k is the gas transfer velocity (cm/h), and [CO2]water − [CO2]air is the CO2 concentration gradient between the water and air. Previous studies measured atmospheric CO2 concentrations 1.5 m above the water surface that is to be 445 ppmv (Mo 2015). k is a key parameter for accurately calculating CO2 evasion flux. We calculated k using the temperature-dependent Schmidt number (ScT) for freshwater:

$$ k={k}_{600}\times {\left(S{c}_{\mathrm{T}}/600\right)}^{-0.5} $$
(2)

with

$$ {Sc}_{\mathrm{T}}=1911.1\hbox{--} 118.11T+3.4527{T}^2\hbox{--} 0.04132{T}^3 $$
(3)

where k600 is the k for CO2 at 20 °C in freshwater, that is, k at a Schmidt number of 600 (Raymond et al. 2012), and T is the in situ water temperature (°C). k600 was derived using an equation described by Raymond et al. (2012):

$$ {k}_{600}=4725\times {\left(v\times S\right)}^{0.86}\times {Q}^{-0.14}\times {D}^{0.66} $$
(4)

where v is the velocity (m/s), S is the channel slope (m/km), Q is the stream discharge (m3/s), and D is the water depth (m).

Results

Hydrological and meteorological variations

There were some sporadic rain events with a total rainfall of 0.8 mm on the afternoon of August 17, 2013, which did not raise the GSS water level. During the study period, a concentrated rainfall of ~ 22.8 mm fell from 0700 to 1345 h on August 18, 2013 at an average intensity of 3.38 mm/h and increased the discharge from the base flow value of 20.34 to 83.91 L/s (Figs. 2 and 3). Discharge in GSS is usually controlled by groundwater discharge at GUS outlet (Pu et al. 2017) and had a mean discharge of 46.13 L/s and median of 55.28 L/s during this study (Figs. 2 and 3). Discharge rose quickly in response to the rain event approximately 2.0 h after the rain started. The discharge of GSS was elevated following the peak discharge for the remainder of the study. Turbidity increased after rainfall coincident with the increased discharge, showing two peaks, but a 2-h lagging discharge (Figs. 2 and 3).

Fig. 2
figure 2

Variation in rainfall, air temperature, solar radiation, stream discharge, hydrochemical parameters, δ13CDIC, and CO2 evasion flux at CK over 2 days from August 17 to 19, 2013. The spike of δ13CDIC in the afternoon of August 18, 2013 may be caused by erroneous sampling or analyzing procedures

Full size image
Fig. 3
figure 3

Variation in rainfall, air temperature, solar radiation, stream discharge, turbidity, hydrochemical parameters, δ13CDIC, and CO2 evasion flux at LY over 2 days from August 17 to 19, 2013

Full size image

Hydrochemical variations

At the CK site, the coefficient of variation (CV) for all measured parameters was very small during the study period. The CV of pH, water temperature, and DO are 0.23%, 0.28%, and 0.16%, respectively, indicating a minor variation during the study period. However, SpC exhibited relatively larger variations, although still small, with a range of 435.0 to 470 μs/cm with a CV of 2.4% (Fig. 2, Table S1). DIC and Ca2+ varied from 278.8 to 287.2 mg/L and 81.0 to 86.1 mg/L with CV values of 0.87% and of 1.7%, respectively (Table S1). Minor variations also occurred in pCO2 with a CV of 3.9%. Water at CK was supersaturated with respect to calcite, with SIc values ranging from 0.08 to 0.14, with an average value of 0.10. The observed small CV for all the measured parameters at CK suggests no distinct diel variations during the study period (Fig. 2). Although the discharge of GSS quickly rose in response to rainfall, hydrochemical parameters at CK (GUS outlet) show a relative steady status with little storm effect, piston effect, or dilution as can commonly occur in some karst springs (Hess and White 1988; Liu et al. 2004, 2007; Pu et al. 2014; Vesper and White 2003a, b). The small variations appear to reflect chemostatic behavior in the GUS during the rain event (Clow and Mast 2010; Godsey et al. 2009; Karis et al. 2016).

At the LY site, water temperature, SpC, pH, DO, Ca2+, DIC, pCO2, and SIc values showed pronounced temporal variations during the study period (Fig. 3). The mean stream water temperature was 23.1 °C, with a range from 22.3 °C just prior to dawn to 24.0 °C in the afternoon, consistent with, but slightly lagging with, solar radiation and air temperature variations (Fig. 3). Changes in pH were temporally coincident with changes in DO, with peaks occurring in the afternoon (14:00–16:00) (Fig. 3). pH ranged from a high of 8.11 during the day to a low of 7.99 at night with a mean of 8.05. DO varied from 7.51 mg/L during the day to 6.56 mg/L at night with a mean of 7.18 mg/L. Notably, the pH and DO maxima roughly coincided with the peak of discharge but had an approximate 2-h lag. The temporal pattern of SpC showed an inverse correlation with pH and DO with minima occurring in the afternoon and maxima at night.

There was a significant decrease in the concentrations of DIC and Ca2+ between site CK and LY (Table S1, t test, p < 0.001). DIC and Ca2+ concentrations have diel patterns that are 180° out of phase with those of pH and DO at LY (Fig. 3). DIC and Ca2+ concentrations increased to maximum values of 287.9 mg/L and 84.3 mg/L, respectively, at 0800 h on August 18. The peaks were followed by fast decline to afternoon lows of 221.6 mg/L and 69.9 mg/L, respectively at 1400 h (Fig. 3). The peak of discharge was consistent with the troughs of DIC and the Ca2+ time series. The water SIc values at LY were more than 0.7, reflecting greater supersaturation than at CK. The changing diel pattern of SIc at LY was similar with the curve of Ca2+ and DIC concentrations, with low values in the afternoon and high values occurring at night. The derived pCO2 also showed a pronounced temporal change and was in phase with Ca2+, DIC, and SIc (Fig. 3). The pCO2 values at LY were significantly lower than at CK (Table S1, t test, p < 0.01). These results indicated that the diel cycles of hydrochemical parameters at LY still occurred during rain and increasing discharge events.

Variation in carbon isotope and CO2 evasion flux

The range of measured δ13CDIC values in CK was − 14.03 to − 13.24‰, with a mean of − 13.74‰; although they a large range, the values show no regular diel pattern. The δ13CDIC values at LY were significantly higher than at CK (Table S1, p < 0.003) and vary from − 12.93 to − 11.80‰. In contrast to CK, LY showed diel variations of δ13CDIC values with lower values at night and higher values in the afternoon. The δ13CDIC maxima at LY corresponded to DIC, Ca2+, SpC, SIc, and pCO2 minima and pH, DO, and water temperature maxima (Fig. 3). The δ13CDIC peaks also coincide roughly with the discharge and turbidity peaks but lagged about 2 h after discharge.

The difference of CO2 evasion flux between CK and LY is also significant (Table S1, t test, p = 0.000). CO2 evasion at CK varied from 1138.8 to 1505.3 mg/m2/h with a mean value of 1302.8 mg/m2/h, which was about eight times higher than at LY, which ranged from 103.5 to 201.9 mg/m2/h with a mean value of 153.1 mg/m2/h (Table S1, Fig. 3).

CO2 evasion did not show diel variation at the CK site over the study period, but it gradually decreased after the discharge and turbidity peaks (Y = − 259.58X + 107, R2 = 0.70, p < 0.01). The curve of CO2 evasion at LY was smoother, and a diel pattern occurred with maxima at night that were about 1.3 times higher than during the daytime minima (Fig. 3). After the peaks of discharge and turbidity, CO2 evasion decreased linearly at LY (Y = − 28.562X + 106, R2 = 0.19, p < 0.01). CO2 evasion at LY was positively correlated with DIC, Ca2+, and pCO2 and inversely correlated with pH, DO, water temperature, and δ13CDIC (Fig. 3). These results also suggest that the diel cycles of carbon isotope and CO2 evasion flux at LY still occur in the periods of rain and rising discharge.

Discussions

As diel frequency is an important timescale in this study, we focus on the data from 0600 h on August 18 to 0600 h on August 19 to evaluate factors controlling DIC and to estimate the carbon budget relative to increased discharge related to a rain event.

Controls on carbon cycling

Rainwater can directly fall into the GSS channel and contribute to increased streamflow. The mean width of the water surface of the GSS was about 4.5 m during the study period. Although such a measurement has a large degree of uncertainty due to the limited information on channel width between two sites, it offered first-hand data for discussing the influence of rainwater. Regardless of the uncertainty in width, the measured regional rainfall into the GSS open channel contributed only about 1.9% of total discharge during the study period. Stream evapotranspiration will decrease water discharge (de Montety et al. 2011; Nimick et al. 2011). However, GUS discharge continuously increased after rain events during the study period, indicating that evapotranspiration was limited for influencing discharge variation. Therefore, in the following calculations and discussions, we neglect the influence of precipitation directly recharging into the GSS open channel and evapotranspiration on the chemical compositions of stream water (Nimick et al. 2011).

Because CK is located at the outlet of the GUS, its water source is groundwater and its hydrochemistry is controlled by the GUS karst system. The composition of this water was chemostatic throughout the study, yielding low CV values (Clow and Mast 2010; Karis et al. 2016). The lack of compositional variation resulted in weak correlations of DIC vs. δ13CDIC, DIC vs. pCO2, DIC vs. DO, DIC vs. CO2 evasion flux, and DIC vs. discharge at CK at diel timescale (24 h) (Fig. 4a–i). Karst groundwater typically shows that the DIC and δ13CDIC originate from soil CO2 produced by the degradation of organic matter and plant root respiration and the dissolution of carbonate minerals (Marx et al. 2017; Nimick et al. 2011; Tobias and Böhlke 2010). The chemostatic characteristics of DIC and δ13CDIC suggest that inorganic carbon at CK also originates from multiple invariant processes (carbonate rock dissolution and soil CO2 hydration) in the subsurface as suggested during a seasonal (dry versus wet) study at this location (Zhang et al. 2017b).

Fig. 4
figure 4

Cross-plots between a δ13CDIC and DIC, bpCO2 and DIC, c DO and DIC, d CO2 evasion flux and DIC, e DO and δ13CDIC, f DO and pCO2, g DO and CO2 evasion flux, h δ13CDIC and CO2 evasion flux, i discharge and DIC, and j DO and discharge at extractive 24-h timescale from 0600 h on August 18 to 0600 h on August 19

Full size image

In contrast to CK, LY showed significant correlations of DIC with δ13CDIC, pCO2, DO, CO2 evasion flux, and discharge over the study period (Fig. 4a–i). These correlations suggest that other processes, in addition to soil-CO2 production and carbonate rock dissolution, could affect water chemistry in the GSS. Commonly, varying temperature could decrease pH by increasing CO2 evasion as water warms during the day (de Montety et al. 2011; Nimick et al. 2011; Spiro and Pentecost 1991). In Fig. 3, pCO2 and CO2 evasion were lower and DO concentration was higher during the day than at night at the LY site. Water temperature at LY showed weak positive correlations with DO concentration (R2 = 0.15, p < 0.01; Fig. S1) and a slightly stronger negative correlation with CO2 evasion (R2 = 0.34, p < 0.01; Fig. S1). Moreover, DO concentrations showed a significant and strong negative correlation with both pCO2 (R2 = 0.77, p < 0.01; Fig. 4f) and CO2 evasion (R2 = 0.86, p < 0.01; Fig. 4g). Consequently, temperature does not appear to be the primary controlling factor on pCO2 in the GSS. CO2 evasion at LY should decrease DIC concentration, but a significant positive correlation occurred between water CO2 evasion and DIC concentration (R2 = 0.73, p < 0.01, Fig. 4d). Previous studies have demonstrated that CO2 outgassing can increase δ13CDIC values in the residual DIC in water (Deirmendjian and Abril 2018; Doctor et al. 2008; Spiro and Pentecost 1991). However, LY shows a significant negative correlation between CO2 evasion and δ13CDIC values (R2 = 0.57, p < 0.01, Fig. 4h), suggesting that CO2 outgassing is not the primary control of δ13CDIC values.

The diel cycle of DIC and DO concentrations is affected by in-stream metabolism in the GSS, including photosynthesis and respiration during sunny days with relatively stable and higher water level (Pu et al. 2017). These processes control DIC because photosynthesis releases O2 into water and consumes CO2 during the day, while respiration consumes O2 and releases CO2 into water during the night, making DO concentrations and pCO2 ideal for evaluating aquatic metabolic process (de Montety et al. 2011; Demars et al. 2015; Kurz et al. 2013). The inverse variation of DIC with DO at LY (Fig. 4c) thus reflects in-stream metabolism even during overcast periods, changing discharge, decreased temperature, and increased turbidity during precipitation. Even with these limiting environmental condition, the rate of photosynthesis in the stream exceeded respiration during the day, causing a net release of O2 and consumption of CO2, thereby increasing pH (Fig. 3). In contrast, respiration consumed O2 and released CO2 during the night thereby decreased the pH (Fig. 3). Therefore, metabolic processes are more important controls than temperature, turbidity, and discharge effects for DIC, pCO2, and DO diel cycle in the GSS.

The δ13CDIC values in stream water are controlled by CO2 sources (Marx et al. 2017). At LY, δ13CDIC has a positive correlation with DO concentrations (R2 = 0.35, p < 0.01; Fig. 4e) and a significantly negative correlation with pCO2 (R2 = 0.74, p < 0.01; Fig. S1), reflecting in-stream metabolic controls on δ13CDIC values. This control results from a 13C enrichment of residual pools of DIC during the day as DIC is consumed by photosynthesis and release of isotopically light biogenic CO2 during nighttime due to respiration of organic matter (Cavalli et al. 2012; Hasler et al. 2016; Pedersen et al. 2013; Zhao and Su 2014).

CO2 evasion fluxes at both CK and LY decreased with the rising discharge (Figs. 2 and 3), although LY was characterized by a diel pattern. According to Eq. 4, increasing discharge will decrease the k600 value, causing a decrease in CO2 evasion (Eq. 1). This decrease differs from CO2 evasion from the Santa Fe River, FL, USA (Khadka et al. 2014), and the Madeira River, the largest tributary of the Amazon River, Brazil (Almeida et al. 2017), where flooding was found to enhance CO2 evasion from water. However, CO2 evasion increased at LY after maximum discharge. Daytime CO2 evasion (mean = 129 mg/m2/h) is about 15.1% lower than nighttime CO2 evasion (mean = 152 mg/m2/h) because of differences in the production and consumption of CO2 during photosynthesis and respiration. Therefore, in-stream metabolism controls CO2 evasion at LY. Our results show that diel variations in CO2 evasion to atmosphere from streams and rivers should be taken account for a more accurate estimation of carbon budgets (Pu et al. 2017; Zhang et al. 2017a).

Although variations in stream discharge showed no relationship with DIC and DO concentrations (Fig. 4i, j) at CK, discharge had a negative relationship with DIC (R2 = 0.23, p < 0.01, Fig. 4i) and a significantly positive relationship with DO (R2 = 0.57, p < 0.01, Fig. 4j) at LY. We used a partial least-squares regression (PLS) model to evaluate to what extent discharge influences DIC diel variation at LY (Li et al. 2014; Paranaíba et al. 2018; Peter et al. 2014). PLS analyses included discharge, pH, SpC, DO, water temperature, DIC, Ca2+, δ13CDIC, pCO2, and CO2 evasion flux at LY only, using SIMCA 14.1 software (32-bit, Umetrics, Sweden). Due to a lack of any linear relationship between DIC, DO, and discharge, this PLS model did not analyze the data at CK. The PLS model performance is expressed in the terms R2Y and Q2. R2Y is comparable to R2 in linear regression and expresses how much of the variance in Y is explained by the X variables. Q2 is a measure of the predictive power of the PLS model. The model is more robust as Q2 approaches R2Y (Paranaíba et al. 2018; Peter et al. 2014; Sobek et al. 2003). Variable importance in projection (VIP) describes how much a variable contributes to explaining the Y variable (DIC). Highly important variables have VIP > 1.0, moderately important variables have 0.8 < VIP < 1.0, and unimportant variables have VIP < 0.8. In this study, the PLS model explained the variability in DIC well, with a R2Y of 0.99 and a Q2 of 0.98 (Table S2). The PLS models show that SIc, Ca2+, pCO2, δ13CDIC, CO2 evasion flux, and DO were important variables (VIP > 0.8; Table S2) to explain DIC concentrations. However, discharge was less important (VIP < 0.8; Table S2), reflecting limited control by discharge for the DIC diel cycle at LY. Consequently, in-stream metabolism remained an important driver of the DIC diel cycle, even during rising discharge following precipitation and the overcast nature of the study period.

Carbon sink produced by in-stream metabolism

Lower DIC concentrations at LY compared to CK (Figs. 2 and 3, Table S1) reflect carbon losses. The loss represents the carbon sink produced by in-stream metabolism, which could be buried in benthic sediments similar to the oceanic biological pump (McElroy 1983) and would represent a net loss of atmospheric CO2 to the sediments (Jiang et al. 2013; Liu et al. 2010, 2015; Liu and Dreybrodt 2015; Pu et al. 2017; Yang et al. 2015). A usual mass balance method was used to estimate the mass of carbon sink due to in-stream metabolism (de Montety et al. 2011; Liu et al. 2015; Liu and Dreybrodt 2015; Pu et al. 2017; Yang et al. 2015). The popular balance equation can be found below (Liu et al. 2015; Yang et al. 2015; Pu et al. 2017):

$$ {F}_{\mathrm{OC}}={\sum}_{t_1}^{t_2}{\left(\frac{\triangle \mathrm{DIC}}{5.08}\right)}_t\times {Q}_t-{\sum}_{t_1}^{t_2}{\left(\frac{\triangle \mathrm{Ca}}{3.337}\right)}_t\times {Q}_t-{\sum}_{t_1}^{t_2}{\left(\frac{F}{3.664}\right)}_t\times A $$
(5)

where FOC is the net mass of organic carbon formed in a day (mg/day), Q is the discharge of GS (L/s), △DIC is the amount of DIC loss between the CK and LY in each time step, 5.08 is a conversion factor between the molar mass of DIC and carbon (61.02/12.01), △Ca is the amount of CaCO3 lost between CK and LY in each time step, 3.337 is the conversion factor between the molar mass of calcium and carbon (40.08/12.011), F is the CO2 evasion flux from unit water surface area (mg/m2/h), 3.664 is the conversion factor between the molar mass of CO2 and carbon (44.01/12.01), and A is the GS water surface area during the study period (m2). The CO2 evasion flux from upstream to downstream regions showed strong spatial heterogeneity; however, the measured DIC concentration at CK was stable, suggesting that CO2 evasion flux cannot affect DIC concentrations at CK. Consequently, we use the evasion at LY for modeling. In Eq. 5, the first term on the right hand side of the equation denotes the total carbon loss/gain in GS over the study periods, the second term is the amount of carbon formed as calcite in GS, and the third term is the carbon loss via CO2 evasion from the GS to the atmosphere.

These calculations show that the total carbon loss was 15.19 kg C/day and the amount of carbon formed as calcite was 4.34 kg C/day. The carbon loss via CO2 evasion from the GS to the atmosphere was 5.21 kg C/day. Therefore, the carbon sink in the GSS channel produced by in-stream metabolism was around 5.6 kg C/day during the study period, which is about 6.1 times higher than it was in July, 2013 (sunny day, 0.91 kg C/day) (Pu et al. 2017). The result suggests that karst streams contribute to the terrestrial atmospheric carbon sink by stabilizing inorganic carbon originating from carbonate weathering process through the burial of organic carbon produced during aquatic photosynthesis (Liu and Dreybrodt 2015; Martin 2017; Pu et al. 2017).

Practical implications

Much work on diel cycles of inorganic and organic carbon in rivers, streams, and creeks occurs during sunny weather and stable water level, and much of it reflects that in-stream metabolism (photosynthesis and respiration) is the primary control of diel cycle of carbon (de Montety et al. 2011; Jiang et al. 2013; Liu et al. 2015; Parker et al. 2005, 2007, 2014; Pu et al. 2017; Tobias and Boehlke 2011; Yang et al. 2015). Our results suggest that even with overcast conditions, precipitation, rising discharge, and elevated turbidity, in-stream metabolism results in diel cycles of DIC due to weak solar radiation. Consequently, the carbon sink produced by in-stream metabolism can occur in karst streams regardless of variations in environmental conditions. Karst terrains cover around 14.7% of earth’s surface (Chen et al. 2017), and most of headwater systems are sources of groundwater with elevated DIC concentrations. The stream outflows are also highly productive because of the generally clear water and solar radiation. The carbon sink produced by in-stream metabolism should be considered within the context of the global carbon cycle.

However, this study focuses only on a small subtropical stream fed by karst groundwater and a medium rainfall event (~ 22.8 mm) in summer. Additional similar studies in different climatic zones, land use types, rainfall events, and flow regimes could improve the understanding of the extent and causes of diel cycle of DIC.

Conclusions

This study demonstrates that overcast conditions, precipitation, and increased discharge and turbidity did not disrupt the diel cycle of DIC and carbon sink produced by in-stream metabolism (photosynthesis and respiration) in a karst headwater stream in southern China. Although diel conditions developed in ~ 1.3 km downstream of the chemostatic stream source, water chemistry showed significant diel variations. The variations resulted from metabolism of the submerged aquatic community and were not influenced by rainfall and rising discharge. Daytime photosynthesis caused a net release of O2 and consumption of DIC that when buried in the sediment was sequestered; at night, respiration consumed O2 and released CO2, resulting in an increase of DIC. The estimated organic carbon sink by in-stream metabolism is around 5.6 kg C/day during the study period, indicating that the submerged aquatic community in a karst stream can significantly stabilize carbon originating from karst weathering processes. Further study is needed to better understand the carbon sink produced by in-stream metabolism in different karst streams covering different climatic zone and land use types and to consider the influences derived from rainfall event and flow regimes.