CARBON DIOXIDE IN THE SOILS AND ADJACENT CAVES OF THE MORAVIAN KARST OGLJIKOVI DIOKSID V PRSTI IN JAMAH NA MORAVSKEM KRASU Jirl FAIMON1 & Monika LICBINSKÄ1-2 Abstract UDC 546.26:551.44(437.2) Jift Faimon & Monika Ličbinskd: Carbon dioxide in the soils and adjacent caves of the Moravian Karst Variations of soil/cave CO2 concentrations and further variables such as temperature, humidity, and cave visitor attendance were studied in two sites of the Moravian Karst (Czech Republic). All the variables showed the same seasonality; they were strongly correlated with each other. The dependence of soil CO2 levels on soil air temperature and absolute humidity was confirmed. Individual effects could not be distinguished because of multicollinearity. The effect of vegetation on soil CO2 production was not recognized. Cave attendance was identified as the most significant predictor of cave CO2 levels. Other variables, soil CO2 and temperature gradients, were less significant. A spurious relationship was alternatively considered, in which external temperature was the universal predictor of cave CO2 levels. Keywords: carbon dioxide, cave, correlation, multiple regression analysis, soil, spurious relationship, Chech Republic. Izvleček UDK 546.26:551.44(437.2) Jift Faimon & Monika Ličbinskd: Ogljikovi dioksid v prsti in jamah na Moravskem krasu Raziskovali smo spremembe koncentracije CO2 in drugih spremenljivk, kot so temperatura, vlaga in prisotnost turistov v jamah Moravskega krasa (Republika Češka). Vse spremenljivke kažejo podobne letne trende in so med seboj korelirane. Dokazali smo povezavo med koncentracijo CO2 ter temperaturo in vlago v prsti. Posamezne vplive zaradi multikolinearnosti nismo mogli izločiti. Vpliva vegetacije na produkcijo CO2 v prsti nismo zaznali. Prisotnost ljudi v jami se je izkazal za najpomembnejši prediktor vrednosti CO2. Druge spremenljivke, kot so CO2 v prsti in temperaturni gradienti so se izkazale za manj pomembne. Raziskovali smo tudi neprave povezave, pri čemer smo vzeli zunanjo temperaturo kot prediktor koncentracij CO2 v jamah. Ključne besede: ogljikov dioksid, jama, korelacija, regresijska analiza, prst, lažne povezave, Češka republika. INTRODUCTION Carbon dioxide plays a key role in karst processes such as limestone dissolution and calcite speleothem growth (Dreybrodt 1999). In general, CO2 levels correspond to a steady state, where CO2 fluxes into the system are balanced by fluxes out of the system. Soil CO2 concentrations vary between 0.1 and 10% vol. (Miotke 1974; Tro- ester & White 1984). Soil input flux results from organic matter decomposition and root exhalation (Brovkin et al. 2008; Kuzyakov 2006). Output flux is composed from the flux into the outdoor atmosphere by diffusion (Longdoz et al. 2008) and the flux into percolating waters via dissolution (Kaufmann & Dreybrodt 2007). Soil CO2 shows 1 Department of Geological Sciences, Faculty of Sciences, Masaryk University, Kotlarska 2, 611 37 Brno, Czech Republic, email: faimon@sci.muni.cz 2 Institute of Geological Engineering, Faculty of Mining and Geology, VŠB - Technical University of Ostrava, 17.listopadu 15, 708 33 Ostrava - Poruba, Czech Republic, email: monika.licbinska@vsb.cz Received/Prejeto: 16.02.2010 strong seasonal fluctuations (Spötl et al. 2005). Epikarst CO2 as an alternative source seems to be relatively invariant (Fairchild et al. 2006). Cave CO2 shows seasonal variations similarly to soil (Troester & White 1984; Bourges et al. 2001; Spötl et al. 2005). Common cave CO2 concentrations vary between 0.1 and 1.0% vol. (Tatar et al. 2004; Baldini et al. 2006). However, higher levels were also monitored in some caves (Atkinson 1977; Ek & Gewelt 1985). Cave input flux includes (1) natural fluxes, i.e. the fluxes derived from direct diffusion from soil/epikarst or drip-water degassing (Holland et al. 1964) and (2) anthropo- genic flux, i.e. the flux stemming from a person exhaling (Faimon et al. 2006). Output flux is controlled by ventilation, which is given by the cave geometry and pressure/temperature gradients between the cave and the exterior (Spötl et al. 2005; Faimon et al. 2006). When input fluxes increase, cave PCO2 increases and the driving force of speleothem growth reduces. In contrast, increasing output flux induces a decrease in cave PCO2 and, thus, an increase in the driving forces. The main goal of the study was to test (1) CO2 production in karst soil under different vegetation and (2) its impact on cave CO2. SITE OF STUDY The Moravian Karst is the most extensive karstic area of the Czech Republic (Balak 1999). It covers an area of 94 km2 as a belt 3-5 km wide and 25 km long. The alti- tude of the karst plateau varies between 250 m and 600 m asl. The granitoid rocks of the Brno Crystalline Massif (Proterozoic) form a crystalline basement. Limestones of the Macocha Formation of the Middle/Upper Devonian period are typical karst rocks (calcite content varies from 95 to 99% wt). Total rock thickness is 500-1000 m. Annual precipitation and temperatures are about 650 mm and 10°C, respectively. A sketch map of the monitoring sites is shown in Fig. 1. SOILS Grey rendzic Leptosols are typical for coniferous forests on the Macocha Plateau above the Punkevni Caves (S1-P) and the Sloup sites above the Sloup-Šošuvka Caves (S1-S). Brown rendzic Leptosols make up the decid- Fig. 1: Sketch map of the monitoring sites. a) Details of Sloup-Šošuvka Caves and b) Punkevni Caves. For explanation of the abbreviations, see Tab. 1 and Tab. 2. Tab. 1: Soil monitoring sites. code site dt i d soil type de^'^'ie'd (luss Working Group WRB 2006) PD(a) [m] spatially associated with S1-P Macocha Plateau coniferous forest soil grey rendzic Leptosol 0.8 C3-P, C4-P S2-P Macocha Plateau deciduous forest soil mull rendzic Leptosol 0.3 C1-P, C2-P, C3-P S1-S Sloup-Šošuvka coniferous forest soil grey rendzic Leptosol 0.6 C3-S, C2-S S2-S Sloup-Šošuvka deciduous forest soil brown rendzic Leptosol 0.5 C1-S, C2-S (a) soil profile mean depth Tab. 2: Cave monitoring sites. code Cave detailed site projection area [m2] volume [m3] TO(a) [m] spatially associated with C1-P Punkevni C. Tunnel Corridor 545 3815 136 S2-P C2-P Punkevni C. Andel Chamber 140 1400 134 S2-P C3-P Punkevni C. Punkva Sail 2640 10560 140 S1-P, S2-P C4-P Punkevni C. Masaryk Hall 340 6120 140 S1-P C1-S Sloup-Šošuvka C. Eliška Hall 915 18300 72 S2-S C2-S Sloup-Šošuvka C. Chamber above Stupnovita Abyss 3430 102900 48020* 51 S1-S, S2-S C3-S Sloup-Šošuvka C. Chamber above Černa Abyss 550 33000 6600* 50 S1-S thickness of overburden ^chamber volume without abyss uous forest soils above the Sloup-Šošuvka Caves (S2-S). Mull rendzic Leptosols are located in deciduous forest on the Macocha Plateau above the Punkevni Caves (S2-P). A summary of the soil monitoring sites is given in Tab. 1. CAVES The Punkevni Caves are open to tourists and consist of a complex of chambers, corridors, the Macocha Abyss, and the underground Punkva River. The sites for CO2 monitoring were the Tunnel Corridor (C1-P), the Andel Speleothem Chamber (C2-P), the Punkva Sail (C3-P) and the Masaryk Hall (C4-P). The Sloup-Šošuvka Caves are open to tourists and form a two-level complex of chambers, corridors and deep abysses. The monitoring sites were the Eliška Hall (C1-S), the Stupnovita Abyss Chamber (C2-S) and the Černa Abyss Chamber (C3-S). A summary of the cave sites is given in Tab. 2. METHODS MONITORING CO2 concentrations, temperature and humidity were monitored at two-week intervals during the years 20062007. Soil monitoring was carried out in probe holes drilled into the soil A-horizon by a steel bar (cca 25 cm, 5 cm in diameter). The wall of each probe hole was reinforced with a cylinder of polyethylene netting and sealed with a plastic cover. Cave monitoring was accomplished in free atmosphere at a 1-m height above the cave floor. CO2 concentrations were measured with a handheld device (2-channel A600-CO2H IR-detector FT linked with an ALMEMO 2290-4 V5, Ahlborn, Germany). All the measurements were performed between 10:00 and 16:00, close to the daily maximum. Relative humidity and temperature were monitored by a digital GFTH 200 hydro/thermometer from Greisinger electronic GmbH, Germany. External temperature data comes from two weather stations in Lhota u Rapotina and Protivanov. Along a straight line, the stations are about 16 and 18 km away from the study area. The presented data are mean values from both the stations (standard deviation ~ 0.8°C; 3.4% relative deviation). STATISTICAL ANALYSIS All statistical calculations were performed in the Statis-tica code, Stat Soft. Inc. (Statistica 2010). Variables The monitored/derived variables are distinguished as UVW-Z abbreviations, where U stands for the physical entity/property (O for carbon dioxide, T for temperature, dT for temperature gradient, RH for relative humidity, AH for absolute humidity, and AT for attendance). The rest of the abbreviation, VW-Z, is consistent with Tabs. 1 and 2. The symbols VW are ignored for attendance, as they are associated with all cave sites. The temperature gradient was assumed either as an absolute value (e.g. \dTCl-P\ = \TC1-P - T(ext)\) or as a logical value marked with index L (e.g. dTC1-PL) defined as follows: when T(ext) < T(cave), then dTCi-jL = T(cave) - T(ext); when T(ext) > T(cave), then dTCi-jL = 0. Outliers To detect outliers, Grubbs' test of raw data was conducted at the a = 0.05 significance level. Only a few outliers were identified, always singly in individual populations (RHS2-P, RHCl-P, RHC2-S, OC3-P, and TC2-S). The outliers were not rejected, as they did not change the results of the data analysis significantly. Correlation Analysis Correlation between the raw data allowed appropriate variables to be selected for subsequent analysis. Based on cross-correlation, the selected variables were tested for a time lag. The weekly data were transformed by linear interpolation into equidistant data with a 15-day step. Data on cave attendance, available as monthly integral attendance, were recalculated into mean daily data and then transformed by linear interpolation into equidistant data consistent with the former data (with a 15-day step). Based on the found lag, the relevant data were transformed into new data without a lag. Multicollinearity A strong correlation between predictors (multicolline-arity) produces redundancy of independent variables in regression analysis. Multicollinearity was assessed using the Variance Inflation Factor (VIF). VIF>5 was taken to indicate multicollinearity (Neter et al. 1989; Mayers 1990). Multiple Linear Regression Analysis Multiple Linear Regression Analysis (MLRA) was chosen to find the most significant predictors of the soil/cave CO2-levels. Stepwise Ridge Regression with Backward Elimination was applied (Schmidt & Muller 1978; Roze-boom 1979). RESULTS SOIL DATA The progress of carbon dioxide, humidity, and temperature of the soil atmosphere over one year of monitoring is given in Fig. 2. All the variables were seasonally dependent; the trends in evolution of CO2 and temperature are mutually similar; the trend in relative humidity evolution is opposite (Fig. 2b). Temperature Soil atmosphere temperatures roughly copied outdoor temperatures. They exceeded 30°C in some sites in July 2006 and approached 30°C in June 2007. The temperature drops below zero at the end of January 2007 (Fig. 2a). Humidity The relative humidity of the soil atmosphere varied between 40 and 85%. Minima were registered in the summer months (July 2006 and August 2007). An extensive maximum is obvious during the monitoring period, from August 2006 to May 2007. A shallow local minimum is presented in January 2007 (Fig. 2b). Carbon dioxide Maxima of carbon dioxide concentrations (between 0.4 and 0.5% vol.) were registered during the late summer/ early fall months (September and October). The highest carbon dioxide concentrations were systematically monitored during summer/early fall (June to September). Minima (about 0.1 to 0.2% vol.) were recorded during the winter/early spring months (December to March). The Fig. 2: Soil atmosphere: the progress of a) temperature, b) humidity, and c) carbon dioxide concentration during one year of monitoring. For explanation of the abbreviations, see Tab. 1 and Tab. 2. lowest carbon dioxide concentrations were registered in coniferous forest soils (S1-P) during winter (Fig. 2c). CAVE DATA Cave CO2 data are highly seasonally dependent. In contrast, cave humidity is less dependent, and temperature is almost conserved in most of the caves (Fig. 3). Fig. 3: Cave atmosphere: the progress of a) temperature, b) humidity, and c) carbon dioxide concentration during one year of monitoring. For explanation of the abbreviations, see Tab. 1 and Tab. 2. Temperature Cave temperatures remained almost constant during the year. Depending on locality, temperatures were between 8 and 14°C. Only the Punkva Sail site (C3-P) showed larger seasonal variations, from 5 to 13°C (Fig. 3a). Humidity Cave humidity shows similar seasonal trends as soil humidity, however, less obvious. Minima were registered in the summer months (July), maxima are in the winter/spring months (February 2007 to May 2007). A local minimum is visible in January 2007 similarly to soils (Fig. 3b). Carbon dioxide Maxima of carbon dioxide concentrations (between 0.3 and 0.4% vol.) were recorded during late summer/early fall (August to September). Minima (about 0.1% vol.) were recorded during winter/early spring (December to April). During the period, somewhat enhanced concentrations (up to 0.19% vol.) were achieved in the Černa Abyss (C3-S). The largest seasonal variations were registered in the Masaryk Dom Chamber (C4-P). In contrast, only slight variations were found in the Punkva Sail (C3-P), Andel Dom Chamber (C2-P), Stupnovita Abyss (C2-S), and the Eliška Dom Chamber (C1-S) (Fig. 3c). DATA ANALYSIS RAW DATA CORRELATIONS Soils Positive correlations were found between all the soil variables except for relative humidity. For individual soils, strong correlations are found between absolute humidity and temperature (r > 0.9), CO2 concentrations and temperature, and CO2 concentrations and absolute humidity (r ~ 0.74 to 0.83). (r ~ 0.66 to 0.86). Important correlations are given in Tab. 3. All correlations are significant at a < 0.05. Punkevni Caves In the C1-P, C2-P, and C4-P sites, CO2 levels are positively correlated with the soil CO2 concentrations (r ~ 0.74 to 0.85), attendance (r ~ 0.74 to 0.77), and external temperature (r ~ 0.68 to 0.72). tte correlations with absolute value of temperature gradient are insignificant (r ~ 0.22 Tab. 3: Correlation matrix: Macocha Plateau and Sloup-Šošuvka soils. Cl o CL CL CL O CL CL op O op op op "■N 00 O op l-N op l-N 1- OS1-P 1.00 TS1-P 0.74 1.00 AHS1-P 0.74 0.95 1.00 OS2-P 0.95 0.76 0.74 1.00 TS2-P 0.74 1.00 0.96 0.76 1.00 AHS2-P 0.71 0.95 0.99 0.74 0.96 1.00 OS1-S 0.81 0.82 0.78 0.87 0.83 0.79 1.00 TS1-S 0.74 1.00 0.95 0.77 1.00 0.95 0.83 1.00 AHS1-S 0.73 0.95 0.98 0.76 0.96 0.99 0.79 0.96 1.00 OS2-S 0.74 0.75 0.71 0.87 0.75 0.74 0.96 0.77 0.74 1.00 TS2-S 0.74 1.00 0.95 0.77 1.00 0.96 0.84 1.00 0.96 0.78 1.00 AHS2-S 0.72 0.96 0.98 0.76 0.97 0.98 0.81 0.97 0.99 0.77 0.97 1.00 T(ext) 0.66 0.86 0.82 0.70 0.86 0.83 0.81 0.86 0.83 0.76 0.86 0.85 1.00 In addition, strong correlations are found between the same quantities in different soils and even different sites (the Macocha Plateau and Sloup-Šošuvka sites). ttis is the case for CO2 concentrations (r ~ 0.74 to 0.95), temperature (r ~ 1), and absolute humidity (r ~ 0.98 to 0.99). All variables correlate with external temperature to 0.31). In turn, the correlations with logical temperature gradients are stronger and negative (r —0.59 to -0.67). The cave CO2 levels are strongly correlated with each other between different sites (r ~ 0.90 to 0.97), except for site 3. In site 3, the correlations of all variables are quite insignificant (r —0.24 to 0.15). Important correlations Tab. 4: Correlation matrix: Punkevni Caves. CL O CL O CL O CL ■Q CL O CL ^^ ■Q CL C3 o CL Či ■Q CL o CL či t? ■Q 1- CL OS1-P 1.00 OS2-P 0.95 1.00 OC1-P 0.83 0.76 1.00 \dTC1-P\ 0.41 0.48 0.31 1.00 dTC1-PL -0.52 -0.51 -0.60 0.00 1.00 OC2-P 0.85 0.76 0.90 0.29 -0.67 1.00 \dTC2-P\ 0.40 0.47 0.31 1.00 0.01 0.29 1.00 dTC2-PL -0.52 -0.51 -0.60 0.00 1.00 -0.67 0.01 1.00 OC3-P -0.15 -0.24 0.10 -0.20 -0.14 0.15 -0.21 -0.15 1.00 \dTC3-P\ 0.31 0.40 0.24 0.92 0.05 0.23 0.92 0.04 -0.15 1.00 dTC3-PL -0.42 -0.41 -0.45 -0.08 0.92 -0.51 -0.08 0.91 0.00 0.07 1.00 OC4-P 0.78 0.74 0.97 0.34 -0.58 0.90 0.34 -0.58 0.10 0.27 -0.42 1.00 \dTC4-P\ 0.31 0.39 0.20 0.98 0.14 0.17 0.98 0.14 -0.17 0.94 0.07 0.22 1.00 dTC4-PL -0.54 -0.54 -0.62 -0.08 0.99 -0.68 -0.07 0.99 -0.08 0.00 0.94 -0.59 0.07 1.00 T(ext) 0.66 0.70 0.68 0.59 -0.80 0.72 0.59 -0.80 0.01 0.51 -0.78 0.68 0.47 -0.84 1.00 AT-P 0.78 0.84 0.76 0.59 -0.66 0.77 0.58 -0.66 -0.09 0.52 -0.61 0.74 0.48 -0.72 0.89 1.00 are summarized in Tab. 4. tte correlations significant at a < 0.05 are highlighted. Sloup-Šošuvka Caves tte CO2 concentrations in the Sloup-Šošuvka Cave sites are positively correlated with the soil concentrations (r ~ 0.66 to 0.85), external temperature (r ~ 0.69 to 0.77), and attendance (r ~ 0.76 to 0.91). Insignificant or weak correlations are found between CO2 levels and absolute temperature gradients (r ~ 0.33 to 0.59). Negative correlations are found between the CO2 levels and logical tem- perature gradients (r ~ -0.49 to -0.66). Similarly to the Punkevni Caves, CO2 concentrations themselves strongly correlate between adjacent parts of the cave system (r ~ 0.80 to 0.86), but less strongly between non-adjacent sites (r ~ 0.59). Important correlations are given in Tab. 5. tte correlations significant at a < 0.05 are highlighted. CROSS-CORRELATION OF EQUIDISTANT DATA tte equidistant data on soil CO2 levels were cross-correlated with those on soil temperature (T), relative/absolute humidity (RH/AH), and external temperature (T(ext)). Tab. 5: Correlation matrix: Sloup-Šošuvka Caves. op ^^ op op O o^ 00 O o^ o^ C3 o o^ op ■Q 1- OS1-S 1.00 OS2-S 0.96 1.00 OC1-S 0.82 0.66 1.00 \dTC1-S\ 0.49 0.52 0.33 1.00 dTC1-SL -0.60 -0.53 -0.56 0.08 1.00 OC2-S 0.85 0.78 0.86 0.30 -0.67 1.00 \dTC2-S\ 0.47 0.47 0.37 0.98 0.05 0.34 1.00 dTC2-SL -0.58 -0.52 -0.55 0.11 1.00 -0.66 0.09 1.00 OC3-S 0.77 0.80 0.59 0.57 -0.49 0.80 0.56 -0.46 1.00 \dTC3-S\ 0.50 0.53 0.37 1.00 0.06 0.34 0.99 0.10 0.59 1.00 dTC3-SL -0.60 -0.53 -0.55 0.07 1.00 -0.67 0.04 1.00 -0.49 0.05 1.00 T(ext) 0.81 0.76 0.69 0.48 -0.83 0.77 0.50 -0.81 0.74 0.49 -0.84 1.00 AT-S 0.91 0.89 0.76 0.51 -0.63 0.91 0.52 -0.61 0.91 0.54 -0.63 0.84 1.00 ^e results are presented in Tab. 6. All time lags are zero, except for OS1-P, which lags after soil absolute humidity and external temperature (both lags ~ 2). The cave CO2 concentrations were cross-correlated with attendance, logical temperature gradients, and soil CO2 levels. tte results are given in Tab. 7. Time lags vary from -1 (where the lagging variable follows the first variable) to an extreme of 5 (where the lagged variables precede the first variable). Whereas cave CO2 levels do not significantly lag behind soil levels (except for the pair OC1-P/OS2-P), the logical temperature gradient precedes the cave CO2 levels (except for the pair OC3-S/dTC3-S). tte CO2 levels in the Punkevni Cave sites lag after attendance by lag ~ 2, except for the extreme lag ~ 5 at site 3. In the Sloup-Šošuvka Cave sites, the attendance is without any lag. Tab. 6: Time lag of selected variables against soil CO^ concentrations. first (dependent) variable lagged independent variable OS1-P OS2-P OS1-S OS2-S soil temperature 0 0 0 0 soil relative humidity 0 0 0 0 soil absolute humidity 2 0 0 0 external temperature 2 0 0 0 j stands for relevant environment P or S; i stands for relevant sites 1 to 2 lag ~ 1 corresponds to 15-day step Tab. 7: Time lag of selected variables against cave CO2 concentrations. Soils Both soil air temperature (sites OS1-P, OS2-P, and OS1-S) and absolute humidity (sites OS1-P, OS2-S) appear to be the best predictors of soil CO2 concentrations. For site OS1-P, the effect of both lag-transformed predictors were distinguished. In this case, temperature and humidity explain the soil CO2 by 38 and 60%, respectively. Alternatively, linear models with external temperature as an alternative predictor were derived (Tab. 9). All models are statistically significant. Caves Almost all models indicate visitor attendance as the most significant predictor of cave CO2 levels. This is the case for the Punkevni Caves except for site C2-P, where the untransformed soil CO2 and temperature gradient are predictors. For site C1-P, soil CO2 is an additional predictor to attendance. The attendance is the sole predictor at sites C3-P and C4-P, although the former model is less significant. In the case of the Sloup-Šošuvka Caves, attendance is the sole predictor in all the models in which untrans-formed data were used. In the case of lag-transformed data, both temperature gradient and soil CO2 are significant variables for site C1-S. tte soil CO2 is an additional predictor together with attendance for site C2-S. LINEAR REGRESSION Linear models of soil/cave CO2 levels with the external temperature as a unique predictor were derived (Tab. 9). first (dependent) variable lagged independent variable OC1-P OC2-P OC3-P OC4-P OC1-S OC2-S OC3-S cave attendance 2 2 5 2 0 0 0 temperature gradient (logical) 2 2 5 2 2 1 -1 soil CO2 (coniferous) 0 0 0 0 0 0 0 soil CO2 (deciduous) 1 0 0 0 0 0 0 i stands for relevant sites 1 to 4 j stands for relevant environment P or S lag ~ 1 corresponds to 15-day step REGRESSION ANALYSIS The Multiple Linear Regression Analysis (MLRA) was conducted separately for the data for which the time lag was accepted (transformed data) versus unaccepted (raw data without any transformation). All significant models are presented in Tab. 8. tte terms in regression equations with p-values exceeding 0.05 are mentioned in the notes. tte models that were physically inappropriate, e.g. those including a term with an illogical sign, were rejected. Except for OC3-P, all models are significant at a < 0.05 and show that external temperature explains the CO2 levels by 68 to 77%. ESTIMATION OF ANTHROPOGENIC CO^ CONTENT IN CAVE CO^ Based on (1) monthly attendance, (2) visiting period at individual sites, (3) cave site volumes, and (4) exhaled CO2 (15 L of exhaled air per minute per person; 5% vol. Tab. 8: Multiple linear regression analysis (stepwise ridge regression). model beta coefficients dependent . ., equation variable ^ df F-value vaiUe I. II. p va ue variable variable notes cave CO, OC1-P OCl-P = 0.0681 + 0.000036 AT-P(a^ + 0.2294 OS2-P(b^ 2/23 16.8 0.59 <0.001 0.41 0.38 nLa OC1-P = 0.0841 + 0.000086 AT-P 1/22 137.9 0.86 <0.001 0.89 n La OC2-P OC2-P = 0.0727 + 0.1929 OS2-P - 0.00348 dTC2-PL 2/23 20.7 0.64 <0.001 0.52 -0.36 nLa OC2-P = 0.0758 + 0.000048 AT-P 1/22 74.3 0.77 <0.001 0.84 n La OC3-P No model n n n n n n nLa OC3-P = 0.0838 + 0.000008 AT-P 1/19 11.9 0.38 0.003 0.59 n La OC4-P OC4-P = 0.0953 + 0.000162 AT-P 1/24 23.9 0.50 <0.001 0.67 n nLa OC4-P = 0.0561 + 0.000219 AT-P 1/22 110.1 0.83 <0.001 0.87 n La OC1-S OC1-S = 0.0820 + 0.000029 AT-S 1/24 26.8 0.53 <0.001 0.69 n nLa OC1-S = 0.0701 - 0.000548 dTC1-SL + 0.0890 OS2-S 2/21 39.1 0.79 <0.001 -0.25 0.69 La OC2-S OC2-S = 0.0837 + 0.000068 AT-S 1/24 75.7 0.75 <0.001 0.83 n nLa OC2-S = 0.0723 + 0.000044 AT-S + 0.0827 OS1-S 2/22 52.8 0.83 <0.001 0.54 0.36 La OC3-S OC3-S = 0.1581 + 0.002750 AT-S 1/24 74.9 0.75 <0.001 O.83 n nLa OC3-S = 0.1574 + 0.000275 AT-S 1/23 70.8 0.75 <0.001 0.83 n La soil CO2 OS1-P OS1-P = 0.1003 + 0.01203 AHS1-P 1/24 23,6 0,50 <0.001 0.67 n nLa OS1-P = 0.0831 + 0.00279 TS1-P + 0.00943 AHS1-P 2/21 59.1 0.85 <0.001 0.38 0.60 La OS2-P OS2-P = 0.1180 + 0.00611 TS2-P 1/24 27.0 0.53 <0.001 0.69 n nL OS1-S OS1-S = 0.1017 + 0.00627 TS1-S 1/24 40.3 0.63 <0.001 0.75 n nL OS2-S OS2-S = 0.0733 + 0.01656 AHS2-S 1/24 28.1 0.54 <0.001 0.70 n nL (a)p = 0.051; (b)p = 0.069 df - degree of freedom; n - not relevant Beta-coefficient indicates relative weight of single independent variable for prediction of dependent variable notes: nL - no lag; nLa - no lag accepted; La - lag accepted Tab. 9: Linear regression analysis: soil/cave CO2 vs. external temperature. model regression coefficient df F-value R2 p-value bo p-value b, p-value beta OS1-P 1/24 19.0 0.44 <0.001 0.1392 <0.001 0.00142 <0.001 0.66 OS2-P 1/24 23.0 0.49 <0.001 0.1224 <0.001 0.00722 <0.001 0.70 OS1-S 1/24 45.1 0.65 <0.001 0.0989 <0.001 0.00799 <0.001 0.81 OS2-S 1/24 35.6 0.60 <0.001 0.0776 <0.001 0.00160 <0.001 0.77 OC1-P 1/24 20.6 0.46 <0.001 0.0900 <0.001 0.00427 <0.001 0.68 OC2-P 1/24 25.5 0.52 <0.001 0.0752 <0.001 0.00272 <0.001 0.72 OC3-P 1/24 0.0 0.00 0.980 0.0870 <0.001 0.00007 0.980 0.01 OC4-P 1/24 20.2 0.46 <0.001 0.0711 0.048 0.01140 <0.001 0.68 OC1-S 1/24 22.4 0.48 <0.001 0.0021 <0.001 0.00074 <0.001 0.69 OC2-S 1/24 35.6 0.60 <0.001 0.0776 <0.001 0.00160 <0.001 0.77 OC3-S 1/24 29.4 0.55 <0.001 0.1360 <0.001 0.00626 <0.001 0.74 of CO2), contents of anthropogenic CO2 were estimated for individual cave sites under the assumption that the sites were not ventilated. The results are presented in Fig. 4. as the ratio of hypothetical anthropogenic CO2 concentrations to the actual CO2 concentration. In the Punkevni Caves, the levels of exhaled CO2 should exceed by many times the actual CO2 levels. In contrast, the anthropogenic CO2 levels in the Sloup-Šošuvka Caves show a much lower proportion relative to the actual CO2 concentrations: at sites C2-S and C3-S, the anthropogenic CO2 would not cover the actual levels. Fig. 4: The ratio of hypothetical anthropogenic CO2 concentrations to the actual CO^ concentration in a) the Punkevni Caves and b) Sloup-Šošuvka Caves. For explanation of the abbreviations, see Tab. 1 and Tab. 2. external temperature could explain the soil CO2 levels by 66 to 81%. The strong correlations of the CO2 concentrations found between different soil types and even between different sites did not confirm the influence of vegetation on soil CO2 production and did moderate the concern about the impact of vegetation on karst processes (e.g. Balak et al. 1999; Barany-Kevei 1999). Cave CO2 The monitored cave CO2 levels are consistent with the values up to 1% vol. found by many researchers (Baldini et al. 2006, 2008). In comparison to soils, the cave CO2 levels showed greater variability. One problem with cave CO2 modelling is the time lag of variables. It is obvious that soil CO2 requires a certain period of time in order to reach a given cave. Similarly, cave ventilation associated with the temperature gradient needs some period to exchange the cave atmosphere. Although anthropogenic CO2 appears in the cave immediately, a certain period is needed for CO2 levels to return to their natural state. Faimon et al. (2006) showed that the relaxation time of a well-ventilated cave is about 24 hours. However, this period could be much higher in the case of poorly ventilated caves. The lag ~ 2 (corresponding to 30 days) of the attendance in the Punkevni Cave sites C1-P, C2-P, C4-P against cave CO2 is long but perhaps acceptable. In contrast, the lag ~ 5 at site C3-P is clearly inconceivable. A data transformation into new data without the lag is a possible approach to identifying the driving variable. Because the resulting regression equations with differently lagged variables are hardly applicable for a convenient cave CO2 level prediction, alternative models based on the original data were derived. DISCUSSION Soil CO2 The observed soil CO2 levels up to 1% vol. are in the range found by others (Zhang et al. 2005). The data analysis confirmed that soil CO2 concentrations are controlled by soil temperature and humidity. This is consistent with the findings of other authors (Jassal et al. 2004; Iqbal et al. 2008). Both quantities are strongly interrelated, which makes it difficult to separate individual effects (Li et al. 2008). MLRA allowed the distinguishing of lag-transformed soil temperature and absolute humidity (the site S1-P), but this distinguishing is based purely on the significance of individual variables. For a convenient prediction of soil CO2 concentrations, linear models with external temperature as the predictor were designed. Beta coefficients showed that Cave CO2 sources Data analysis suggests that the generally accepted belief that soils are the main source of cave CO2 could be questioned. MLRA showed that the soil CO2 levels appeared as predictors in only four models (of thirteen in total) and always combined with another predictor. In these models, the share of soil CO2 in cave CO2 levels varied between 38 and 69%. Doubts about the dominant role of soils in cave CO2 resonate with some authors (Miotke 1974; Barany-Kevei 1999; Tatar et al. 2004; Baldini et al. 2005). Even if the soil CO2 effect was superimposed by anthropogenic CO2 in this study, alternative sources (e.g. epikarstic sediments) should be considered in future studies. Attendance was identified as a main predictor of cave CO2 levels in both the caves, which indicates a broad anthropogenic impact. An exception is site C3-P, where no model was found for untransformed data and the model for lagged data is physically unacceptable. In this site, the CO2 values are probably controlled by distinct factors despite the MLRA results (see the discussion later). The attendance impact is most obvious in the Sloup-Šošuvka Caves, especially in sites C2-S and C3-S, where the lag of variables is near zero. Paradoxically, based on the estimations of exhaled CO2, the contributions of anthropogenic CO2 levels in these sites should be lowest. The reason for this contradiction may be an overestimation of cave site volumes. Both the sites are linked to abysses lying below the visitor route with a disputable contribution to total site volumes. If the abyss volumes are omitted, the share of anthropogenic CO2 rises to 87% (C2-S) or above 100% (C3-S) of actual cave CO2. Despite the clear influence of anthropogenic CO2 on the cave environment, long-term monitoring of dripwaters (in the Punkevni Caves especially) shows permanent water supersaturation (Faimon & Ličbinska, unpublished data), which indicates that the impact is not destructive. This conclusion is consistent with the study of the anthropogenic CO2 impact in the Cisarska Cave (Faimon et al. 2006). Factors suppressing cave CO2 levels It is well known that cave air circulation depends on temperature gradients between the interior and exterior (de Freitas et al. 1982; Russell & McLean 2008). Dynamic caves (see Geiger et al. 2003; Spötl et al. 2005; Linan et al. 2008) are ventilated year-round, although the ventilation is more intensive at external temperatures below the cave temperature (Faimon, unpublished work). In static/ semi-dynamic caves, such effects are emphasized under the same conditions. MLRA only sporadically identified the temperature gradient as a significant predictor of cave CO2 levels (only at sites C2-P and C1-S). ttis indicates the minor role of cave ventilation. However, this is contradictory to the estimations of the anthropogenic CO2 share of actual CO2 levels at individual cave sites. Therefore, we guess that the ventilation effect is undervalued. This is especially the case at site C3-P, with its extremely low CO2 levels at low variance. Because the site is unique due to its large free water table surface, the possibility of CO2 dissolution was considered. Based on the analyses of 13 water samples, however, partial pressures of CO2 in the water (logPco2 = -2.20±0.26) exceeded those in the air (logPCO2 = -2.98±0.35). tterefore, degassing must be expected instead of dissolution. Based on these facts, the hypothesis about CO2 dissolution was rejected and ventilation remained the sole factor explaining the cave CO2 levels. This is consistent with enhanced temperature variations (Fig. 3). Temperature gradients seem to be an unsuitable proxy for ventilation in case the cave atmosphere is totally exchanged with the external atmosphere, and CO2 levels are nearly constant. The strong correlation of the CO2 concentrations between different sites (except for site C3-P) indicates the strong mutual dependency of cave sites. The dependence diminishes with site distance. Spurious relationship problem It is well known that statistically related variables (correlated) need not show a causal connection and that the correlation can be the result of a spurious relationship (see, e.g., Ben-Zeev & Star 2001; Pearl 2009). tterefore, we considered the possibility that between cave CO2 concentrations and other tested variables there is no causal interrelation and that all correlations are the result of external temperature as a confounding factor. A set of linear models was derived, in which external temperature is a unique cave CO2 level predictor. All the models are significant at a = 0.05 and valid for all the cave sites except for C3-P. ttese models explain cave CO2 levels by 68 to 77%. We believe that further studying of more sophisticated data (equidistant data with a short distance in the range of hours or minutes) could contribute to a better understanding of the problem. CONCLUSIONS Spatial and temporal variations of carbon dioxide were studied in two sites of the Moravian Karst: (1) soils in the Macocha Plateau with the adjacent Punkevni Caves, and (2) soils in the Sloup-Šošuvka field with the adjacent Sloup-Šošuvka Caves. tte soil air CO2 levels, cave air CO2 levels, cave attendance, and external temperatures showed similar seasonality. It was confirmed that soil CO2 production is controlled by temperature/humidity. Both effects are indistinguishable because of multicol-linearity. tte impact of vegetation was not proven. Based on multiple linear regression analyses, cave attendance seems to be the most significant variable controlling cave CO2 levels and, subsequently, calcite deposition in the given sites. Temperature gradients and soil CO2 levels were identified as further controlling variables. Because statistical analysis is not able to reveal a causal relation- ship, a spurious relationship was considered with external temperature as a lurking variable. To demonstrate such possibility, alternative linear models were derived in which external temperature operates as unique universal predictor of cave CO2 levels. Two general conclusions may be derived from the study: (1) soils need not necessarily control cave CO2 levels, and (2) the anthropogenic impact may easily superimpose upon natural processes in caves. From the former conclusion, it follows that the speleothem growth rate does not need to be directly related to soil/surface conditions. ttis is important for palaeoenvironmental reconstructions based on the study of terrestrial speleo-thems. tte latter conclusion is important for karst/cave environment conservation and protection: it shows that the anthropogenic impact in caves is not negligible even if its consequences are not evident or destructive. ACKNOWLEDGEMENTS We would like to thank Dr. Petr Štepanek from the Czech Hydrometeorological Institute, Regional Office Brno, and Jiri Hebelka, director of the Administration of the Moravian Karst Caves, Blansko, for providing the data on regional temperatures and attendance of open caves, respectively. In addition, we would like to thank two anonymous reviewers for their valuable comments. tte work was supported by the MSM0021622412 grant from the Ministry of Education, Youth and Sports of the Czech Republic and by the IGS 2101/541 grant from the VŠB - Technical University of Ostrava, Faculty of Mining and Geology." REFERENCES Atkinson, T.C., 1977: Carbon dioxide in the atmosphere of the unsaturated zone: an important control of groundwater hardness in limestones.- Journal of Hydrology, 35, 111-123. Balak, I., Jančo, J., Štefka, L. & P. Bosak, 1999: Agriculture and nature conservation in the Moravian Karst (Czech Republic).- International Journal of Speleology, 28B, 71-88. Baldini, J.U.L., McDermott, F., Baker, A., Railsback, L.B., Baldini, L.M. & D.P. Mattey, 2005: Vegetation effects on stalagmite growth rate and isotopic content.- Earth and Planetary Science Letters, 240, 2, 486-494. Baldini, J.U.L., Baldini, L.M., McDermott, F. & N. Clip-son, 2006: Carbon dioxide sources, sinks, and spatial variability in shallow temperate zone caves: Evidence from Ballynamintra Cave, Ireland.- Journal of Cave and Karst Studies, 68, 4-11. Baldini, J.U.L., McDermott, F., Hoffmann, D.L., Richards, D.A. & N. Clipson, 2008: Very high-frequency and seasonal cave atmosphere PCO2 variability: Implications for stalagmite growth and oxygen isotope-based paleoclimate records.- Earth and Planetary Science Letters, 272, 118-129. Barany-Kevei, I., 1999: Impact of agriculture land use on some Hungarian karst regions.- International Journal of Speleology, 28B, 89-98. Ben-Zeev T. & J.R. Star, 2001: Spurious Correlations in Mathematical fkinking.- Cognition and Instruction, 19, 3, 253-275. Bourges, F., Mangin, A. & D. d'Hulst, 2001: Le gaz carbo-nique dans la dynamique de l'atmosphere des cavites karstiques: l'exemple de l'Aven d'Orgnac (Ardeche).-Earth and Planetary Sciences, 333, 685-692. Brovkin, V., Cherkinsky, A. & S. Goryachkin, 2008: Estimating soil carbon turnover using radiocarbon data: A case-study for European Russia.- Ecological modelling, 216, 178-187. de Freitas, C.R., Littlejohn R.N., Clarkson T.S. & I.S. Kristament, 1982: Cave climate: Assessment of airflow and ventilation.- Journal of Climatology, 2, 383-397. Dreybrodt, W., 1999: Chemical kinetics, speleothem growth and climate.- Boreas, 28, 347-356. Ek C. & M. Gewelt, 1985: Carbon dioxide in cave atmospheres. New results in Belgium and comparison with some other countries. Earth Surface Processes & Landforms, 10, 2, 173-187. Faimon, J., Štelcl, J. & D. Sas, 2006: Anthropogenic CO2-flux into cave atmosphere and its environmental impact: A case study in the Cisarska Cave (Moravian Karst, Czech Republic).- Science of the Total Environment, 369, 231-245. Fairchild, I.J., Smith, C.L., Baker, A., Fuller, L, Spötl, C., Mattey, D., McDermott, F. & EIMF, 2006: Modification and preservation of environmental signals in speleothems.- Earth-Science Reviews, 75, 105-153. Geiger, R., Aron, R. H. & P. Totdhunter, 2003: The Climate Near the Ground.- Sixth edition, Rowman & Littlefield Publishers, Inc., pp. 600, Lanham. Holland, H.D., Kirsipu, T.V., Huebner, J.S. & U.M. Ox-burgh, 1964: On some aspects of the chemical evolution of cave water.- Journal of Geology, 72, 36-67. Iqbal, J., Ronggui, H., Lijun, D., Lan, L., Shan, L., Tao, C. & R. Leilei, 2008: Differences in soil CO2 flux between different land use types in mid-subtropical China.- Soil Biology & Biochemistry, 40, 23242333. Jassal, R.S., Black, T.A., Drewitt, G.B., Novak, M.D., Gau-mont-Guay, D. & Z. Nesic, 2004: A model of the production and transport of CO2 in soil: predic-tiong soil CO2 concentrations and CO2 efflux from forest floor.- Agricultural and Forest Meteorology, 124, 219-236. Kaufmann, G. & W. Dreybrodt, 2007: Calcite dissolution kinetics in the system CaCO3-H2O-CO2 at high undersaturation.- Geochimica et Cosmochimica Acta, 71, 1398-1410. Kuzyakov, Y., 2006: Sources of CO2 efflux from soil and review of partitioning methods.- Soil Biology & Biochemistry, 38, 425-448. Li, H.J., Yan, J.X., Yue, X.F. & M.B. Wang, 2008: Significance of soil temperature and moisture for soil respiration in a Chinese mountain area.- Agricultural and Forest Meteorology, 148, 490-503. Linan, C., Vadillo, I. & F. Carrasco, 2008: Carbon dioxide concentration in air within the Nerja Cave (Malaga, Andalusia, Spain).- International Journal of Speleology, 37, 99-106. Longdoz, B., Yernaux, M. & M. Aubinet, 2008: Soil CO2 efflux measurements in a mixed forest: impact of chamber disturbances, spatial variability and seasonal evolution.- Global Change Biology, 6, 907 - 917. Mayers, R.H., 1990: Classical and Modern regression with applications.- Second edition, PWS and Kent Publishing, pp. 488, Boston. Miotke, F-D., 1974: Carbon dioxide and the soil atmosphere.- Abhandlungen zur Karst-Und Höhlenkunde, Reihe A, Speläologie, 9, 1-49. Neter, J., Wasserman, W. & M.H. Kutner, 1989: Applied linear regression models.- Second edition, Richard D. Irwin, pp. 688, Homewood, IL. Pearl, J., 2009: Causal inference in statistics: An overview.- Statistics Surveys, 3, 96-146. Rozeboom, W.W., 1979: Ridge regression: Bonanza or beguilement?- Psychological Bulletin, 86, 242-249. Russell, M.J. & V. MacLean, 2008: Management issues in a Tasmanian tourist cave: Potential microclimatic impacts of cave modifications.- Journal of Environmental Management, 87, 474-483. Schmidt, P. & E.N. Muller, 1978: The problem of multi-collinearity in a multistage causal alienation model: A comparison of ordinary least squares, maximum-likelihood and ridge estimators.- Quality and Quantity, 12, 267-297. Spötl, C., Fairchild, I.J. & A.F. Tooth, 2005: Cave air control on dripwater geochemistry, Obir Caves (Austria): Implications for speleothem deposition in dynamically ventilated caves.- Geochimica et Cos-mochimica Acta, 69, 10, 2451-2468. Statistica, 2010: StatSoft.- [Online] Available from: http:// www.statsoft.com/ [Accessed 19th October 2010]. Tatar, E., Mihucz, V.G., Tompa, K., Pöppl, L., Zaray, G. & L. Zambo, 2004: Study of soil leachates in doline above the Beke Cave, Hungary.- Geoderma, 120, 155-164. Troester, J.W. & W.B. White, 1984: Seasonal fluctuations in the carbon dioxide partial pressure in a cave atmosphere.- Water Resources Research, 20, 153156. IUSS Working Group WRB, 2006: World reference base for soil resources 2006.- Second edition, World Soil Resources Report number: 103. Zhang, C., Yuan, D. & J. Cao, 2005: Analysis of the environmental sensitivities of a typical dynamic epikarst system at the Nongla monitoring site, Guang-xi, China.- Environmental Geology, 47, 615-619.