Continuous monitoring of CO2 soil diffuse degassing at Phlegraean Fields (Italy): influence of environmental and volcanic parameters

Continuous monitoring of CO2 soil diffuse degassing at Phlegraean Fields (Italy): influence of environmental and volcanic parameters

Available online at www.sciencedirect.com R Earth and Planetary Science Letters 212 (2003) 167^179 www.elsevier.com/locate/epsl Continuous monitorin...

1MB Sizes 0 Downloads 19 Views

Available online at www.sciencedirect.com R

Earth and Planetary Science Letters 212 (2003) 167^179 www.elsevier.com/locate/epsl

Continuous monitoring of CO2 soil di¡use degassing at Phlegraean Fields (Italy): in£uence of environmental and volcanic parameters D. Granieri  , G. Chiodini, W. Marzocchi, R. Avino INGV-Osservatorio Vesuviano, Via Diocleziano 328, 80124 Napoli, Italy Received 3 September 2002; received in revised form 16 April 2003; accepted 17 April 2003

Abstract Carbon dioxide soil flux was continuously measured during 4 years (1998^2002) inside the crater of Solfatara by using the ‘time 0, depth 0’ accumulation chamber method. The CO2 soil flux (BCO2 ) is strongly influenced by external factors, such as the barometric pressure, the air and soil temperature and humidity, the wind speed, the amount of rain, and so on. Here, we apply a two-step filtering technique to remove the contribution of these external factors from the raw data and to highlight variations in gas flow from depth. In the first step we apply multiple regression and a best-subset search procedure to determine the minimal number of parameters to insert in the regression model. In the second step we apply time filtering on the residuals of the previous analysis through an ARIMA (integrated autoregressive moving average) model which allows us to quantify long-term trends and short-term periodicities. The statistical analysis showed that (1) the highest frequency fluctuations are due to variations of environmental parameters (particularly soil humidity and air temperature) and (2) the long-term trend of the filtered data is correlated with the ground deformation. This correlation is enhanced by back-shifting the CO2 flux signal by 3 months. These observations, along with the likelihood that the ground deformation at Phlegraean Fields is controlled by fluid pressure within the hydrothermal system, indicate that the long-term trend in soil CO2 flux is related to fluid pressure changes at depth. The delay between the soil CO2 flux and the ground deformation is most probably due to the inertia of the gas moving in the subsoil. = 2003 Elsevier Science B.V. All rights reserved. Keywords: carbon dioxide soil £ux; Solfatara; accumulation chamber method; monitoring

1. Introduction A useful tool in the surveillance of an active volcanic area is continuous monitoring of geo-

* Corresponding author. Tel.: +39-081-6108449; Fax: +39-081-6108466. E-mail address: [email protected] (D. Granieri).

chemical parameters. Changes in degassing of CO2 or trace gases for which CO2 acts as a carrier (i.e. Rn and He) have been observed before and/ or during volcanic unrest in many volcanoes (i.e. Etna, Stromboli, Vulcano, Phlegraean Fields [1^8]). Previous studies have been focussed on degassing fumarolic features or volcanic plumes [1, 9,10]. Nevertheless important amounts of CO2 ,

0012-821X / 03 / $ ^ see front matter = 2003 Elsevier Science B.V. All rights reserved. doi:10.1016/S0012-821X(03)00232-2

EPSL 6666 23-6-03

168

D. Granieri et al. / Earth and Planetary Science Letters 212 (2003) 167^179

even higher than the gas discharged by active fumarolic systems, are often released through diffuse soil degassing [5,11,12]. In some cases, the di¡use soil emissions of CO2 are comparable to the rates of gas emitted by plumes at erupting volcanoes [1,13]. This paper describes the continuous monitoring of di¡use CO2 degassing from the Solfatara crater (Phlegraean Fields, Italy). At Solfatara CO2 degassing at anomalously high rates occurs over an area of V0.5 km2 (Fig. 1). This area, named the di¡use degassing structure (DDS), releases V1500 t day31 of CO2 , an amount even higher of that released by many erupting volcanoes [13]. Assuming that the original £uids that feed the soil

Fig. 1. Map of log soil BCO2 of the Solfatara DDS. The survey of the soil BCO2 was carried out from December 16 to 18, 1998, and consists of 406 measurements [13].

di¡use degassing have the same composition of the high temperature fumaroles of Solfatara (i.e. 160‡C, H2 O/CO2 V2.2 by weight), an amount of V3300 t day31 of steam condenses during the process [13] releasing thermal energy at V0.8 1013 J day31 . In the present period of quiescence, this amount represents the main energy release of the Phlegraean Fields volcanic system. It is several times higher than the energy released through conduction over the entire caldera, the elastic energy associated with seismic events, and the energy dissipated by ground deformation [13]. The monitoring of the di¡use degassing at Solfatara is done through a combination of an automatic continuously operating station at a selected site and periodic measurements of £ux over an array of sites. The continuous monitoring station measures soil CO2 £ux and various environmental parameters which can potentially a¡ect the soil gas £ux. Measurements of soil CO2 £ux and soil temperature (10 cm depth) at 30 ¢xed stations homogeneously distributed in the £oor of the Solfatara crater are made one to two times per month (Fig. 2). In addition, more detailed measurements (200 points regularly distributed in a 20 m2 grid) are repeated several times per year. Here, we report the CO2 £ux data collected from continuous monitoring at the Solfatara crater during the period 1998^2002. In order to separate the part of the signal linked to variations in the deep source of the CO2 £ux, we must remove the in£uence of environmental external factors from the raw data. To accomplish this goal, we investigate the relations between external factors (i.e. rain, barometric pressure, soil and air temperature and humidity, wind speed) and the CO2 £ux. Moreover, on the base of the results obtained during the 4 years of continuous and periodic monitoring, we delineate the representativeness of the soil BCO2 measured in the continuous monitoring site (few dm2 ) to the degassing over the entire Solfatara DDS (107 dm2 ). In Section 4 we will consider also the results of another automatic station (FLXOV4) which is installed inside the crater of Mt. Vesuvio.

EPSL 6666 23-6-03

D. Granieri et al. / Earth and Planetary Science Letters 212 (2003) 167^179

Fig. 2. The crater of Solfatara and the location of the 30point network. The position of FLXOV1 SCAS coincides with a point of the network.

169

The SCAS prototype was jointly developed by IGGI (Istituto di Geocronologia e Geochimica Isotopica) of Pisa, University of Perugia and West System srl, and tested at Santorini (Greece) in the framework of the European project ‘Santorini Volcano Laboratory’ in 1995. Since 1998, 25 SCAS units have been installed in the following volcanic areas : Masaya Volcano (Nicaragua); Poa's Volcano (Costa Rica); La Palma (Canary Islands, Spain); San Salvador, San Miguel, San Vincente and Santa Ana volcanoes (El Salvador); Usu Volcano (Japan); Mammoth Mountain (California, USA); Etna (Italy) ; Stromboli (Italy). The Osservatorio Vesuviano, Neapolitan Department of Istituto Nazionale di Geo¢sica e Vulcanologia (Italy), has installed one SCAS at the Solfatara crater (FLXOV1) and one SCAS at Vesuvio (FLXOV4). The data measured from the SCAS installation are summarized in Table 1. Both £ux stations were installed in November 1998, and measurements with portable equipment were made to select suitable sites. Since that time, the Solfatara station has been collecting data every 2 h while Vesuvio station every 4 h.

2. Soil CO2 automatic station (SCAS) The SCAS acquires, at selectable time intervals, values of CO2 £ux (BCO2 ) and various parameters that can potentially in£uence the CO2 £ux, i.e. barometric pressure, air and soil temperature, air and soil humidity, wind speed and rainfall. The BCO2 is measured by the accumulation chamber method [14,15,16]. The accuracy of individual measurements of gas £ux is better than 10% for CO2 £uxes over a range of 100^10 000 g m32 day31 [12,16]. The use of the accumulation chamber method with the SCAS allows a direct measurement of the CO2 transferred from the ground to the atmosphere, independent of the transport regime in the system (di¡usive and/or e¡usive) and the physical properties of the soil. Each measurement is made over a time ranging from a few seconds to 1^2 min. The SCAS is equipped with speci¢c sensors for the measurement of environmental parameters during the same time interval as the CO2 £ux measurement. Solar panels and batteries provide the required electric power.

3. Filtering the CO2 £ux data The ¢ltering of CO2 £ux data is applied only to the data of Solfatara where an almost continuous sampling in time, from November 1998 to June 2002, is available. In the case of Vesuvio, a complete and detailed analysis of the signals was not possible because of data gaps due to battery failures from absence of solar radiation during winter time, landslides which occasionally damaged the station, and short periods of failure of the automatic accumulation chamber device. The SCAS FLXOV1 is located inside the DDS of Solfatara, in a zone where the degassed CO2 is derived from a hydrothermal system [17]. The deep origin of the gas is consistent with the high measured £ux values (300^5000 g m32 day31 in Table 1), orders of magnitude higher than the values of the local biological background (30 g m32 day31 [13]). Variations in CO2 £ux could be caused by variations in both the up£ow of

EPSL 6666 23-6-03

170

D. Granieri et al. / Earth and Planetary Science Letters 212 (2003) 167^179

Table 1 Summary of results from two sites

CO2 £ux (g m32 day31 )

Air relative humidity (%)

Air temperature (‡C)

Bar pressure (mbar)

Soil relative humidity (a.u.)

Soil temperature (‡C)

Wind speed (m s31 )

Rain (mm)

Range Mean No. of Range Mean No. of Range Mean No. of Range Mean No. of Range Mean No. of Range Mean No. of Range Mean No. of Range Mean No. of

data

data

data

data

data

data

data

data

FLXOV1 November 1998^June 2002

FLXOV4 November 1998^March 2002

315^4 724 746 15 655 17.3^98.2 78.5 15 655 32^38.2 16.3 15 655 979^1 023 1 004 15 655 37^86.2 45.5 15 655 37.6^68.6 52 14 754 0^9.56 0.81 15 655 0^36.6 0.17 10 265

34^19 952 2 479 3 849

38.3^17.8 5.7 4 423 879^919 905 4 475

51.7^91.5 78.3 4 110

0^58.6 0.386 3 111

a.u. = arbitrary unit proportional to soil water content.

hydrothermal £uids, called the endogenous signal in the following, and in environmental parameters. In order to better determine the e¡ects of variations in the hydrothermal system on di¡use CO2 £ux, we used multivariate regression analysis (MRA) to delineate the relations between £ux and external factors and then used these relations to ¢lter out the e¡ects of these factors on the measured £ux history. Before discussing the details of the analysis, some cautionary remarks are warranted. In particular, when we do regression calculations on unplanned data (that is, data arising from continuous operations and not from a designed experiment), some potentially erroneous results can arise. For example, a false e¡ect (a bias) on a visible variable may be caused by an unmeasured latent variable. Another undesired effect arises when a correlated variable undergoes a variation whose range is small compared to its average value, leading to an apparent lack of sig-

ni¢cance. A third problem is that the use of unplanned data often causes large correlations between predictors ; this makes it impossible to attribute a causal e¡ect to one speci¢c predictor. This technical problem can be avoided by using orthogonal dummy variables obtained, for example, through the PCA (principal component analysis) procedure [18,19]. However, because these dummy variables have no physical meaning, we chose not to use this technique. From a qualitative graphical analysis we built a linear model in which the dependent variable is the natural logarithm of the CO2 £ux, and the independent variables are the external factors that have been measured at the station. In order to verify the quality of the model, the dataset was divided into three subsets, one learning dataset (LD) and two control datasets (CDs). LD, which has been used to build the model, represents the data from the central period (November 1999^November 2000, 1-year interval), while CDs are the data used to

EPSL 6666 23-6-03

D. Granieri et al. / Earth and Planetary Science Letters 212 (2003) 167^179

check the model (CD #1, November 1998^November 1999, the year preceding LD; and CD #2, November 2000^June 2002, 19 months following LD). The ¢rst important step in the regression analysis consists of selecting the relevant parameters to explain the variability of the CO2 £ux in LD. Here, we select the most relevant variables by using the procedure called the ‘best-subset search’, described by Garside [20]. The procedure ¢rst requires the ¢tting of every possible regression equation which involves any combination of independent variables, and then selecting the case which is the best respect one criterion. The criterion adopted here is based on the R2 coe⁄cient. In other words, we report the case with the highest R2 coe⁄cient obtained by using a single independent variable, two independent variables, three independent variables, and so on (Table 2). The ‘best’ regression is the one for which, adding more variables, further gain in the R2 coe⁄cient is only minor. Table 2 shows that the ‘best’ regression is the second one, with Hs (humidity of soil) and Ta (temperature of air) as independent variables. The linear ¢ltering model obtained for LD is: lnð B CO2 Þ ¼ ð5:78  0:03Þ3ð1:36  0:05ÞU1032 T a þð2:07  0:05ÞU1032 H s

The quality of the model is further veri¢ed by using this equation on the other two independent datasets (CD #1 and CD #2). For each dataset we calculate the Pearson correlation coe⁄cient r that can be used as an indicator of the degree of similarity between the observed and predicted time series. The high values of the correlation coe⁄cient r obtained for LD (0.774) and CD #1 (0.825) are a clear indication of the applicability of the model (Fig. 3a). Actually, the r value for CD #2 dataset (0.402), is signi¢cantly lower. However, by reducing the temporal length of the whole CD #2 (19 months) to the same size as LD and CD #1 (1 year, following LD) the value of r is 0.802, i.e. quite comparable to the values for LD and CD #1. In other words, the correlation coe⁄cient r is lowered by the inclusion of the additional 7 months of CD #2. This worsening

171

might be due to either technical or physical factors. By reducing the length of the CD #2 interval, we may have eliminated a bias introduced by a di¡erent performance of the model as a function of the seasons. We may also have eliminated the possible in£uence of physical changes in the soil or in the hydrothermal system that may have changed the response of the site to the external parameters. This is an intrinsic factor that must be considered in modelling non-stationary processes. In such cases, the extrapolation of any model tends unavoidably to worsen, moving away from the time interval used for the learning phase. The time evolution of the residuals of MRA ¢ltering is reported in Fig. 3b. The time dependence in the residuals of the regression can at least partially be attributed to an endogenous signal. However, some changes in non-measured variables might explain the variation of BCO2 , or there might be some non-linear e¡ects in operation among the external variables. The time dependence of the residuals of the regression (Zt ) can be expressed as: Zt ¼ X t þ St þ O t

where Xt is a low-order ARIMA (integrated autoregressive moving average; see [21]) process which can be used to characterize the long-term trend of the signal, St is a periodic component which can characterize signals with higher frequencies, and Ot is a random noise. The component Xt can be estimated by ¢tting an ARIMA (1, 1, 1) model to the £ux data: BX t ¼ a þ a1 X t31 þ R t 3b1 R t31

where a is a calibration constant, a1 is the paramTable 2 The best-subset search process for FLXOV1 data No. of variables

Variables in the equation

Best R2 (%)

1 2 3 4 5 6

ln(BCO2 ) = f(Hs ) ln(BCO2 ) = f(Hs ,Ta ) ln(BCO2 ) = f(Hs,Ta ,Vw ) ln(BCO2 ) = f(Hs,Ta ,Vw ,R) ln(BCO2 ) = f(Hs ,Ta ,Vw ,R,Pa ) ln(BCO2 ) = f(Hs ,Ta ,Vw ,R,Pa ,Ha )

57.8 67.9 68.4 68.8 69.0 69.0

EPSL 6666 23-6-03

172

D. Granieri et al. / Earth and Planetary Science Letters 212 (2003) 167^179

Fig. 3. (a) Observed (black line) and predicted (gray line) log soil BCO2 for the data series of Solfatara. Dashed vertical lines mark the intervals used to construct the linear model (LD) and to check the model (CD #1 and CD #2). (b) Residuals of the MRA for the same data series.

eter of the autoregressive process, b1 is the parameter of the moving average process, Rt is the residual process, and B is the backward shift operator de¢ned as BXt = Xt 3Xt31 [21,22]. Here, we do not ¢t a general ARIMA (k, p, l) process, but we apply only p = 1, and k, l 9 1. This choice usually guarantees a satisfactory ¢t of the non-periodic and low-frequency (long-term) time dependence of the time series. The parameters of the model a, a1 , b1 and the variance of the residual process Rt are estimated through a least-square technique [21,22]. The correct values of k and l (0 or 1) are selected through the AIC criterion (Akaike information criterion; [23]). By analyzing the residuals relative to LD, we obtain: BX t ¼ ð0:44  0:03ÞX t31 þ R t 3ð0:77  0:02Þ R t31

In the above equation, Rt represents the time series of the ‘new’ residuals further ¢ltered by the non-periodic time dependence. In other words, while the multivariate regression, described above, removes the e¡ects of the external

parameters from the CO2 £ux variations, the ARIMA ¢ltering removes (if it exists) the longterm time dependence. Fig. 4 reports the residuals of the ARIMA ¢ltering for LD and CDs of the FLXOV1 station. In the case without a periodic component (St = 0), the results of the ¢ltering, namely the Rt process, should be white, or at least uncorrelated, noise. Instead, the spectral analysis [22,24] on these residuals shows a peak at 24 h and some related harmonics (Fig. 5). These cycles represent the main feature of the St component. We argue this residual periodicity is still due to external processes, rather than endogenous processes which would likely be characterized by longer periods. The presence of an exact 24-h cycle rules out a predominant e¡ect of the tides that act at di¡erent periods; the strongest luni^solar e¡ect, for instance, is at 12.42 h. We suggest this periodicity could be due to one or more external factors that we did not consider, or, more likely, due to the fact that the link between the CO2 £ux and the air temperature might be non-linear; the air

EPSL 6666 23-6-03

D. Granieri et al. / Earth and Planetary Science Letters 212 (2003) 167^179

173

Fig. 4. Results of the ARIMA ¢ltering for LD and CDs.

temperature, in fact, has an exact 24-h periodicity. It is worthwhile to note that the same 24-h periodicity was also found in the seismic activity of the Phlegraean Fields (e.g. [25]). Although some e¡ects of external factors were not removed by the ¢ltering, the ¢ltering process was successful because of the great di¡erence between the period observed in the residuals and the expected characteristic transient of the internal processes. We argue, therefore, that endogenous processes are responsible for the long-term trend and that the computed variable Xt is the best representation of the endogenous ‘signal’ contained on the original BCO2 data. The ¢ltered variation in the average daily Xt values (Fig. 6) shows a general decreasing trend passing from values of +0.8 in 1998 to 30.6 at the end of the sampling period (June 2002). A period of general decline of the signal was evident in the period November 1998^February 2000, followed by, from March 2000 to March 2002, a period in which the signal remained almost constant, oscillating around a value of 0 and showing two positive peaks in October 2000^April 2001 and October 2001^March 2002. During the last 3 months of 2002 a new phase of decrease in the Xt values was observed.

midity of the soil, i.e. rain, and from the air temperature. These two parameters alone can explain most of the observed variance. However, the analysis of the residuals, performed with an ARIMA process, showed further, minor, correlation with environmental parameters characterized by variations of 24-h frequency, which could be hypothesized to be due either to wind or to air temperature. This result applied exclusively to the FLXOV1 site and it should not be considered to apply everywhere. Di¡use soil degassing is in fact a¡ected by a number of environmental factors, i.e. meteorological parameters, soil properties (porosity, hydraulic conductivity, soil water conditions, soil temperature) and topographic e¡ects. Localized e¡ects on BCO2 at FLXOV1 are indicated by comparing the results obtained at Solfatara FLXOV1 with the data of other two SCAS: one located at Vesuvio (FLXOV4) and the other

4. Discussion and conclusions The variance of BCO2 data at FLXOV1 during 1998^2002 is linked to environmental parameters and deep sources. Regarding the environmental parameters, the multivariate regression analysis of raw data showed a main control from the hu-

Fig. 5. Spectral analysis of the ARIMA residuals showing a peak at 1 cycle day31 (24 h). The peaks at 2,3,4 cycles day31 represent the harmonics of the diurnal signal.

EPSL 6666 23-6-03

174

D. Granieri et al. / Earth and Planetary Science Letters 212 (2003) 167^179

Fig. 6. General trend of the soil BCO2 after MRA and ARIMA ¢ltering. Xt is the best representation of the endogenous ‘signal’ contained in the original BCO2 data.

at Mammoth Mt., California. At all three sites the CO2 £ux originates from deep sources [26]. Details of the SCAS installed at Mammoth are reported in [27] while the main characteristics of the FLXOV4 station are shown in Table 1. At the Solfatara FLXOV1 site, we found that the main controlling factor is the humidity of the soil which alone explains about 58% of the £ux variance. A graphical example of the strong correlation between soil BCO2 , soil humidity and rain episodes is seen in the £ux data for the period May^August 1999 (Fig. 7). Each of the three rain episodes was accompanied by concurrent increments of soil humidity and BCO2 . This positive correlation could appear surprising because the in¢ltrating waters should decrease the e¡ective

porosity and possibly dissolve some of the £owing gas, leading to a decrease in the £ux. Instead, there is a topographic e¡ect of rainfall on gas £ux related to the fact that FLXOV1 is situated on ground that is higher than the surrounding degassing area. Rainfall preferentially drains into the low-lying areas and gas £ow is diverted to the high ground. In contrast, a more normal inverse correlation between rain and £ux was observed at the Vesuvio station FLXOV4. The chronogram (time history) of Fig. 8, relative in the autumn 2000, clearly shows that the rain events caused a sharp decrease in the soil BCO2 . During the dry summer months, another di¡erence was observed between the £ux stations at Solfatara and Vesuvio in terms of the relation

Fig. 7. Soil relative humidity, soil BCO2 and rainfall chronograms at the FLXOV1 site from May to August, 1999. The positive anomalies of the soil humidity, due to rainfall episodes, produce a growth in the soil BCO2 .

EPSL 6666 23-6-03

D. Granieri et al. / Earth and Planetary Science Letters 212 (2003) 167^179

175

Fig. 8. Rain and soil BCO2 at the FLXOV4 site in September^November 2000. The rainfall episodes induce a decrease in the soil BCO2 , in contrast to the results found at Solfatara.

between £ux and barometric pressure. The £ux data for FLXOV4 (Fig. 9) show a signi¢cant inverse correlation, whereas the data for FLXOV1 show little in the way of variations with barometric pressure. An inverse correlation between £ux and barometric pressure has been reported for a SCAS at Mammoth Mountain (California). Rogie et al. [27] attribute this relation to the rising air pressure that retards viscous gas £ux (pressure gradient driven) while falling atmospheric pressure enhances viscous gas £ux from the soil explaining the inverse correlation. The same interpretation can be valid for Vesuvio, while the absence of a clear correlation between barometric pressure and £ux at the Solfatara FLXOV1 site is most probably due to the fact that the £ux is mainly di¡usive (gradient concentration driven) instead of viscous. The regression analysis and the subsequent ARIMA process on FLXOV1 data permit us to highlight the part of the signal most probably linked to the variation of the deep source. In order to interpret this link we consider the seismic activity and the ground deformation which are regularly monitored at Phlegraean Fields in the framework of volcanic surveillance. A subsidence period characterized the area of Pozzuoli in the ¢rst part of the last century (V0.8 m from 1905 to 1968 [28]). The subsidence was interrupted by two main periods of uplift: 1969^1972, when a maximum vertical displacement of 174 cm was measured at the benchmark 25 sited near the

town of Pozzuoli, and 1982^1984, when the maximum vertical displacement was 179 cm at the same benchmark 25 ([29]; Fig. 10). Since the beginning of 1985, the area has been a¡ected by a general subsidence (Fig. 10) which has been interrupted by three minor uplifting phases in 1989, 1995 and 2000. Recent geochemical and geophysical studies suggest that the ground deformation at Phlegraean Fields is controlled by £uid pressure within the hydrothermal system [13,30,31]. In agreement with this interpretation, both the general subsidence since 1985 and the general decrease of the soil CO2 £ux at site FLXOV1 could be related to £uid pressure declines at depth. In contrast, a £uid pressure increase at depth could

Fig. 9. Barometric pressure and soil BCO2 at the FLXOV4 site in dry periods. Note their inverse correlation.

EPSL 6666 23-6-03

176

D. Granieri et al. / Earth and Planetary Science Letters 212 (2003) 167^179

Fig. 10. Ground deformations at Phlegraean Fields based on the periodical measurements at the benchmark 25 located at the Pozzuoli port. Two main uplift episodes in 1969^1972 and 1982^1984 and three minor uplifting phases in 1989, in 1995 and in 2000 are evident. Inset: soil deformation during the activity of FLXOV1 SCAS.

have caused both the minor uplift episode that occurred in 2000 (Fig. 10, inset), and the temporary rise in the CO2 £ux decrease observed at FLXOV1 during February 2000^March 2002 (Fig. 6). The correlation between the two signals increases if we back-shift the CO2 £ux signal by 3 months (Fig. 11), which could suggest a delay due to the inertia of the gas di¡usion in the subsoil. A better correlation between the soil CO2 £ux and ground deformation is obtained considering the monthly mean of di¡use CO2 £ux mea-

sured in a network of 30 selected measuring points (Fig. 12). In this case the correlation between the two independent parameters is evident. The period of uplift in the year 2000 was accompanied by a doubling of the total out£ow of CO2 , which from 2000 g m32 day31 at the end of 1999 passed to 4000 g m32 day31 in the summer^autumn 2000. During the period of continuous £ux measurements at Solfatara no correlation was observed between seismic activity and changes in gas £ux,

Fig. 11. Xt and soil deformation chronograms at the Solfatara site. The best correlation is obtained by back-shifting the Xt series of 3 months.

EPSL 6666 23-6-03

D. Granieri et al. / Earth and Planetary Science Letters 212 (2003) 167^179

177

Fig. 12. Monthly average of the soil BCO2 at 30 selected points located on the £oor of the Solfatara crater together with the soil deformation. From January 1998 to June 2002, 110 surveys of di¡use soil CO2 £ux have been carried out on a network of 30 measurement points. The soil BCO2 values range from 150 g m32 day31 to 22 000 g m32 day31 , with an average of 4840 g m32 day31 .

although one cluster of ML 9 2.2 events did occur in the area in July^August 2000 [32]. At Vesuvio, many periods of seismic events occurred between 1999 and 2002, with hypocenters clustered within the volcanic conduit. However, there is again no correlation between the timing of these events and the timing of periods of anomalous gas £ux at FLXOV4 (Fig. 13). The chronogram shows instead two very marked anomalies in CO2 £uxes in periods of low seismic activity. The ¢rst period occurred in May 2001, when the £uxes doubled in few days, and the second in October 2001 when the £uxes increased by a factor of 10, simulating an anomalous degassing of the volcano. The ab-

sence of any other anomaly in the monitored parameters (seismicity, deformation, fumarolic gas composition) and the results of a ¢eld survey suggested that the anomalies could be caused by landslides which a¡ected the degassing structure of Vesuvio, possibly changing the up£ow pattern of the gases. The comparison between the signal £ux histories at FLXOV1, the results of the periodic gas £ux campaigns, the ground deformation at Phlegraean Fields, and the ‘false’ anomalies at Vesuvio highlight some limitations of the monitoring of soil CO2 £ux performed with the SCAS. We think that the main limitation is the small dimen-

Fig. 13. July 1, 1999, to March 31, 2002, soil BCO2 and seismic energy at Vesuvio. The marked anomalies of the soil BCO2 in June 2001 and in October 2001 do not coincide with the periods of anomalous seismic activity, but with the episodes of landslides located inside the crater.

EPSL 6666 23-6-03

178

D. Granieri et al. / Earth and Planetary Science Letters 212 (2003) 167^179

sion of the monitored area (few dm2 ) with respect to the size of the DDSs (V0.5 km2 in the case of Solfatara). In the small monitored sites the £ux variations can depend on local factors (i.e. a permanent permeability change caused by alluvial episodes or landslides in the degassing area) more than on changes in the deep £uid up£ow. Such super¢cial and localized processes a¡ect in a lesser way the total output of gas from a DDS and the total gas £ow from a large array of periodically measured £ux sites. In conclusion, SCAS FLXOV1 furnished data useful for a better understanding of the in£uence of environmental parameters on the CO2 £ux. However, the strong site e¡ects which can change the £uxes in the small portion of the monitoring point can blur the detection of deep variations. A possible method to improve the monitoring of the di¡use degassing process is to couple automatic station data with the execution of periodic measurement campaigns.

Acknowledgements This work was supported by the National Volcanic Group of Italy (GNV) and by the European Community (Geowarn project). We thank Dr. Mike Sorey for helpful comments and suggestions.[SK]

References [1] P. Allard, J.M. Carbonelle, R. Faivre-Peirret, D. Martin, J.C. Sabroux, P. Zettwoog, Eruptive and di¡usive emissions of CO2 from Mt, Etna Nat. 351 (1991) 387^391. [2] B. Badalamenti, S. Gurrieri, S. Hauser, M. Valenza, Soil CO2 output in the Island of Vulcano during the period 1984^1988: surveillance of gas hazard and volcanic activity, S.I.M.P. Carapezza Meml. 43 (1988) 893^899. [3] B. Badalamenti, S. Gurrieri, S. Hauser, F. Parello, M. Valenza, Change in the soil CO2 output at Vulcano during the summer 1998, Acta Vulcanol. 1 (1991) 219^221. [4] B. Badalamenti, S. Gurrieri, P.M. Nuccio, M. Valenza, Gas hazard on Vulcano Island, Nature 350 (1991) 26^ 27. [5] J.C. Baubron, P. Allard, J.P. Toutain, Di¡use volcanic emission of carbon dioxide from Vulcano Island, Italy Nat. 344 (1990) 51^53.

[6] M.L. Carapezza, S. Giammanco, S. Gurrieri, S. Hauser, P.M. Nuccio, F. Parello, M. Valenza, Soil gas geochemistry: CO2 . In: F. Barberi, A. Bertagnini, P. Landi (Eds.), Mt. Etna the 1989 eruption, Giardini Editori e Stampatori, Pisa, 1990, pp. 62^64. [7] S. Lombardi, M. DiFilippo, L. Zantederchi, Helium in Phlegraean Fields, Bull. Volcanol. 47 (1984) 259^265. [8] S. Lombardi, G. Nappi, Helium in soil gas at Lipari and Stromboli Volcanoes, Period. Mineral. 55 (1986) 165^ 176. [9] S.N. Williams, J.S. Schaefet, L.C. Stephen, M.L. Calvache, D. Lopez, Global carbon dioxide emissions to the atmosphere by volcanoes, Geochim. Cosmochim. Acta 56 (1992) 1765^1770. [10] S.L. Brantley, K.W. Koepenick, Measured carbon dioxide emissions from Oloinyo Lengai and skewed distribution of passive volcanic £uxes, Geology 23 (1995) 933^936. [11] J.C. Baubron, P. Allard, J.C. Sabroux, D. Tedesco, J.P. Toutain, Soil gas emanations as precursory indicators of volcanic eruptions, J. Geol. Soc. Lond. 148 (1991) 571^ 576. [12] G. Chiodini, F. Frondini, B. Raco, Di¡use emission of CO2 from the Fossa crater, Vulcano Island, Bull. Volcanol. 58 (1996) 41^50. [13] G. Chiodini, C. Cardellini, F. Frondini, D. Granieri, L. Marini, G. Ventura, CO2 degassing and energy release at Solfatara Volcano, Campi Flegrei, Italy, J. Geophys. Res. 106 (B8) (2001) 16213^16221. [14] K.J. Parkinson, An improved method for measuring soil respiration in the ¢eld, J. Appl. Ecol. 18 (1981) 221^ 228. [15] F. Tonani, G. Miele, Methods for measuring £ow of carbon dioxide through soils in the volcanic setting, Int. Conf. on Active Volcanoes and Risk Mitigation, Napoli, 27 August^1 September, 1991. [16] G. Chiodini, R. Cioni, M. Guidi, L. Marini, B. Raco, Soil CO2 £ux measurements in volcanics and geothermal areas, Appl. Geochem. 13 (1998) 543^552. [17] R. Cioni, E. Corazza, L. Marini, The gas/steam ratio as indicator of heat transfer at the Solfatara fumaroles, Phlegraean Fields (Italy), Bull. Volcanol. 47 (1984) 295^302. [18] N. Draper, H. Smith, Applied Regression Analysis, 2nd edn., Wiley, New York 1981. [19] G.E.P. Box, Use and abuse of regression, Technometrics 8 (1966) 625^629. [20] M.J. Garside, Best subset search, Appl. Stat. 20 (1971) 112^115. [21] G.E.P. Box, G.M. Jenkins, Time Series Analysis: Forecasting and Control, Holden-Day, Merri¢eld, VA, 1976. [22] M.B. Priestley, Spectral Analysis and Time Series, Vol. 1, Academic, San Diego, CA, 1981. [23] H. Akaike, A Bayesian analysis of the minimum AIC procedure, Ann. Inst. Stat. Math. 30A (1978) 9^14. [24] W.A. Fuller, Introduction to Statistical Time Series, Wiley, New York, 1976. [25] W. Marzocchi, G. Vilardo, D.P. Hill, G.P. Ricciardi, C.

EPSL 6666 23-6-03

D. Granieri et al. / Earth and Planetary Science Letters 212 (2003) 167^179 Ricco, Common features and peculiarities of the seismic activity at Campi Flegrei, Long Valley, and Vesuvius, Bull. Seismol. Soc. Am. 91 (2001) 191^205. [26] G. Chiodini, L. Marini, M. Russo, Geochemical evidence for the existence of high-temperature hydrothermal brines at Vesuvio volcano, Italy, Geochim. Cosmochim. Acta 65 14 (2001) 1^19. [27] J.D. Rogie, D.M. Kerrick, M.L. Sorey, Dynamics of Carbon Dioxide emission at Mammoth Mountain, California, Earth Planet. Sci. Lett. 188 (2001) 535^541. [28] G. Orsi, L. Civetta, C. DelGaudio, S. DeVita, M. DiVito, R. Isaia, S.M. Petrazzuoli, G.P. Ricciardi, C. Ricco, Short-term deformations and seismicity in the resurgent Campi Flegrei caldera (Italy): An example of active block-resurgence in a densely populeted area, J. Volcanol. Geotherm. Res. 91 (1999) 415^451.

179

[29] F. Barberi, G. Corrado, F. Innocenti, G. Luongo, Phlegraean Fields 1982^1984: Brief chronicle of a volcano emergency in a densely populated area, Bull. Volcanol. 47 (1984) 175^185. [30] M. Bonafede, Hot £uid migration, an e⁄cient source of ground deformation, application to the 1982^1985 crisis at Campi Flegrei, Italy, J. Volcanol. Geotherm. Res. 48 (1991) 187^198. [31] M. Bonafede, M. Mazzanti, Modeling gravity variations consistents with ground deformation in the Campi Flegrei Caldera (Italy), J. Volcanol. Geotherm. Res. 81 (1998) 137^157. [32] G. Saccorotti, F. Bianco, M. Castellano, E. DelPezzo, The July^August 2000 seismic swarms at Campi Flegrei volcanic complex, Italy, Geophys. Res. Lett. 28 (2001) 2525^2528.

EPSL 6666 23-6-03