A new method for real-time monitoring of soil CO2 efflux

Authors


Correspondence author. E-mail: martin21skifond@gmail.com

Summary

1. A better understanding of temporal and spatial variability of soil CO2 fluxes is essential to improve model predictions of soil effluxes. To accomplish that goal, high-frequency and long-term data sets for model development and validation are needed. However, the cost and high maintenance associated with the current technology make high-frequency measurements for small or large spatially distributed grids difficult to achieve. Here, we describe a new observational infrastructure for monitoring soil CO2 efflux, which is attractive because of its low cost and low power consumption compared to traditional methods.

2. Three observational stations equipped with forced diffusion (FD) chambers were deployed in the summer of 2010 across a 1000-km transect in Atlantic Canada. At half-hourly resolution, each observational station recorded soil carbon dioxide (CO2) efflux from two flux chambers and from a suite of meteorological sensors and peripherals. Each station was equipped with telemetry, and data were continuously downloaded for c. 1 year.

3. The average power consumption for each station was roughly a third of a LI-COR LI-8100 system. The FD chambers were approximately four times more affordable than conventional equipment and were also reliable with <1% of the data lost because of power failure.

4. High-frequency observations from the three sites showed that the systems were extremely dynamic, with CO2 efflux dependency to temperature and moisture on many time-scales. For instance, the data showed pronounced increases in soil CO2 efflux after major rain events. The results from the FD chambers also highlighted the role of other biological and physical factors on soil CO2 efflux.

5. Overall, this new method was very successful in key areas including survivability, management intensity and cost. The high-frequency data hold many interesting features that are not captured in synoptic data sets and which will be useful for tuning our understanding of soil carbon dynamics.

Introduction

Terrestrial soils represent a large carbon (C) reservoir that contains an estimated 1500 Pg/C, double that of the atmospheric reservoir (750 Pg/C) (Mielnick & Dugas 2000). Contributions of CO2 to the atmosphere from soil respiration are globally significant with an estimated 68 PgC/year of CO2 entering the atmosphere (Raich & Schlesinger 1992). It is expected that variations in temperature and precipitation rates resulting from global climate change will alter local and regional soil CO2 effluxes (Solomon et al. 2008). Derived relationships often lack universal applicability, while the relationships between soil CO2 fluxes, and temperature and precipitation are well explored (Raich & Schlesinger 1992; Hashimoto, Satoru & Ishizuka 2009), because characterization of the relationships is only site specific, with predictive models consisting of regressive equations which fit CO2 concentrations to the environmental variables (Suarez & Simunek 1993). Progress has been slow leaving forecasts of soil C efflux under suspected climate change scenarios inconclusive. Because of this uncertainty, it is essential to have a better understanding of spatio-temporal variability of soil respiration to improve local, regional and global model predictions of soil CO2 efflux (Martin & Boldstad 2009; Falloon et al. 2011; Zheng et al. 2010).

Soil CO2 effluxes are spatially and temporally variable in response to temperature, precipitation, soil and vegetation. For instance, temporal variability of soil efflux is especially dynamic, appearing at subdaily scales in response to precipitation events (Lee et al. 2002; Tang, Baldocchi & Xu 2005; Niinisto, Kellomaki & Silvola 2011; Vargas et al. 2010; Barron-Gafford et al. 2011; Deng et al. 2011; Wu & Lee 2011), daily scales related to diel cycles (Vargas & Allen 2008), seasonal and yearly scales following the temperature sinusoids (Rayment & Jarvis 2000; Savage & Davidson 2003; Parkin & Kaspar 2004) and also responding to root activity (Irvine et al. 2008; Moyano, Kutsch & Rebmann 2008). Spatial variability of soil CO2 effluxes can also be very difficult to define as CO2 responds to spatial patterning of physical and chemical properties of soils (Schwendemann et al. 2003; Scott-Denton, Sparks & Monson 2003; Epron et al. 2006; Ngao et al. 2012) and spatial distribution of vegetation (Fang et al. 1998; Stoyan et al. 2000; Epron et al. 2004, 2006; Baldocchi, Tang & Xu 2006; Brechet et al. 2009; Chatterjee & Jenerette 2011; Mendonca et al. 2011). Characterization of spatio-temporal variability of soil respiration lags, while the environmental factors that influence variability of soil CO2 from the landscape are well explored because conventional soil monitoring lacks the flexibility and endurance to monitor accurately and continuously. As a result, data sets that capture the essence of spatio-temporal variability are rare.

There are a number of methods by which in situ soil CO2 fluxes are measured. The most common techniques employ the use of chambers that are placed on the soil surface and measure the rate of CO2 accumulation (closed system) or the instantaneous CO2 flux (open system) (Le Dantec, Epron & Dufrene 1999; Pumpanen et al. 2004; Rochette & Hutchinson 2005). Soil respiration with chambers can be determined with either alkali absorbents or infrared gas analysis, which are generally characterized by the absence (static) or presence (dynamic) of airflow in the chamber (Pongracic, Kirschbaum & Raison 1997). Power requirements of these chambers vary with type and measurement frequency. For example, a LI-COR LI-8100 can draw upwards of 15 W a day during continuous sampling. A simple static chamber that is equipped with a small Vasiala GMP 343 will have significantly lower power requirements. There are trade-offs associated with different chambers, specifically in terms of measurement frequency; the LI-COR LI-8100 can measure at higher frequency but requires constant power, whereas the measurement frequency for the static chambers is lower, as is the power requirements. Field campaigns that employ chambers are also time-consuming because often data collection is not automated and many chambers cannot operate autonomously. Eddy covariance (EC) towers are also a common method for measuring total terrestrial CO2 fluxes at the ecosystem level, but they do not provide specific information on soil C effluxes (Baldocchi 2003; Goulden et al. 1996). In addition, the high power requirements of EC instrumentation can make off-grid applications challenging. At last, the vertical gradient measurement method is also growing in popularity because it allows to continuously and automatically measure soil CO2 flux at different temporal scales with minimal disturbances to the natural soil structure during installation (Pingintha et al. 2010). However, direct measurement of CO2 flux is highly desirable and has advantages over indirect flux estimation methods such as the gradient method (Burton & Beauchamp 1994; Risk, Kellman & Beltrami 2008; Lee, Schuur & Vogel 2010). Previous work has shown that the accuracy of calculated fluxes is degraded significantly without excellent estimates of soil gas diffusivity (Risk, Kellman & Beltrami 2008), while the gradient method seems a reasonable and simple alternative for continuous monitoring, which is seldom measured or well constrained especially across time.

Further considerations with regard to selecting a method for monitoring soil CO2 efflux are the cost associated with purchasing and maintaining the instrumentation (hardware). There are costs associated with either working the instrument during monitoring campaigns, as is the case with static chambers, or deploying personnel to install and maintain the instruments, as with EC towers, for example. Similar to power requirements, there are trade-offs associated with the cost of soil CO2 method. The cost of purchasing a LI-COR LI-8100 far exceeds the costs associated with a simple static chamber equipped with an IRGA, which cannot provide the same sampling density as a LI-COR LI-8100, but is a much more affordable option. In short, high power requirement, high maintenance and cost of the chambers and EC towers make difficult long-term monitoring of CO2 fluxes of large-spatial grid. However, a new technology described by Risk et al. (2011) could achieve this goal more easily. This new forced diffusion (FD) chamber technology is characterized by a gas-permeable membrane that passively regulates mixing of atmosphere and soil air in the chamber, in place of the active pumping system inside a regular dynamic efflux chamber system. In addition, this passive regulation of gas flow allows that internal concentration sensors to be switched off between measurements, thereby achieving very low power consumption. Despite promising results, this technology has yet to be tested under various field conditions and for extended period of time.

Thus, the main objectives of this study are to: (i) build a pilot-scale observatory using FD chambers that can continuously monitor soil CO2 efflux, along with environmental and micrometeorological peripherals such as temperature, soil moisture (as a proxy for precipitation), humidity and wind speed, (ii) view and analyse data from three different sites in real time, and evaluate the FD chamber survivability for the field measurements through evaluating the power and (iii) evaluate how well the temporal variability of CO2 effluxes measured by the FD chambers reflects the processes happening on a larger scale.

Materials and methods

Sites

Three observational stations were deployed across a 1000-km transect in Atlantic Canada during the summer of 2010. Each of these sites is environmentally and climatically distinct, and the sites are strategically placed to represent different climatologically and landscape types allowing for observation of soil C dynamics in different systems.

The first station is located in the Cape Breton Highlands National Park, Nova Scotia, Canada, on the plateau of North Mountain of the Cape Breton Highlands, at an elevation of 370 m above sea level (46°49′4·85″N, 60°40′19·86″W). This site was selected for the heavy snow accumulation and high winds experienced there (Neily et al. 2008). The rough mountain terrain is characterized by a stony sandy loam till parent material, with variable drainage (Neily et al. 2008). The vegetation stand is primarily conifers (balsam fir, black spruce, white spruce and eastern larch) (Neily et al. 2008), and the station is situated within the remnants of a past spruce bud worm infestation that has drastically altered the ecosystem (Neily et al. 2008). Moose grazing has prevented recovery of the vegetation. Average daily temperature at the Highlands site is 5·5 °C with an average daily maximum of 9·1 °C and an average daily minimum of 1·7 °C, with the lowest daily mean temperature during February (−6·7 °C) and the greatest mean daily temperature occurring during July (17·7 °C). Total average precipitation at the site is 1309·4 mm, with July seeing the lowest average precipitation (74·3 mm) and December seeing the highest (175·6 mm), mainly as snow (Environment Canada 2011).

The Woods Harbour site (43°31′36·93″N, 65°43′47·46″W) is located in Shelburne County, Nova Scotia, Canada, and sits on Lydgate series soils (Cann, MacDougal & Hilchey 2008) that are moderately coarse and derived from glacial drift (Neily et al. 2008). This site was selected because it is characterized by wet soils and experiences mild winters that allow for multiple freeze–thaw events. The soil covers undulating topography with slow, variable drainage and both the landscape and subsurface are littered with boulders (Cann, MacDougal & Hilchey 2008). Climatically, the area is characterized by mild winters, with frost-free periods for over half the year, and cool, foggy summers (Neily et al. 2008). Black spruce dominate the vegetation regime along with white spruce and balsam fir (Neily et al. 2008). The station is positioned at the grass–tree ecotone with trees to the east and south-east and is located c. 1 km east from the coast. The Woods Harbour station has an annual average temperature of 7 °C, with the greatest average daily temperature maximum occurring in August at 19·1 °C and the lowest daily minimum occurring in January −5·7 °C. Average total precipitation for the area is 1263·8 mm, with an average of 92·9 mm of snow and an average of 1170·8 mm of rain, with the greatest average precipitation occurring in January (123·7 mm), November (123·5 mm) and March (121·3 mm) as a mix of rain and snow, which is indicative of the local winters in the area (Environment Canada 2011).

The Gros Morne station (49°55′56·11″N, 57°46′37·93″W) is located in Shallow Bay at the north end of the park on a sandy morne deposition that is imperfectly drained and stratified. Physically, the area is characterized by undulating marine terrances at low elevations (Kirby, Guthrie & Hender 1992). This site was selected for its high winds and minimal snow accumulation. The climate in Gros Morne is cool and wet at sea level and is influenced by the ocean by the strong prevailing south-westerly winds from the Gulf of St Lawrence. Mean annual air temperature is 3 °C with a mean maximum of 15 °C in July and minimum of −8·4 °C in February. Mean annual precipitation for the area includes 1397 mm of rain and 3281 mm of snow, with 10–30 days of fog (World Heritage Sites, 2005).

Field Infrastructure and Methods

The FD chamber is a new soil CO2 monitoring instrument developed by Risk et al. (2011). The FD chambers are conceptually similar to dynamic chambers. Once deployed, the FD chambers can measure continuously (i.e. 30-min interval), requiring only minimal downtime for maintenance and recalibration. The durability and design of the FD chambers provide it with an ability to measure CO2 under snow packs, and if deployed during snow-free periods, the chambers maintain their contact with the soil surface and do not disturb the snow pack. Each of the stations is equipped with two flux chambers and an accompanying atmospheric reference chamber. The soil FD chambers were <2 m apart, with FD atmospheric reference sensor intermediate between the FD chambers. The moisture and temperature sensors were installed also within 1 m, closest to the FD atmospheric reference sensor. The FD chambers were deployed on 7 July 2010 for the Highlands and on 9 July 2010 for the Woods Harbour sites, and starting on 8 August 2010 for the Gros Morne site.

In addition to the CO2 measurements made by the FD chambers, the stations also measure and record soil temperature at the soil surface, at 10 and 30 cm below the surface, air temperature using 107B temperature sensors (Campbell Scientific, Edmonton, AB, Canada), volumetric water content (VWC) in the soil profile at 10 and 30 cm deep using CS 616 VWC reflectometers (Campbell Scientific), soil oxygen using a SO-200 (Apogee Scientific, Englewood, CO, USA) and relative humidity using the TRH-100 sensor (Pace Scientific, Mooresville, NC, USA).

Each observational station is equipped with a 12-V battery, a solar power charge converter (Morningstar Corporation, Newtown, PA, USA) and an 80 W solar panel (Sharp Electronics, Huntington Beach, CA, USA). A solid-state relay (Crydom, San Diego, CA, USA) is used to toggle power to the FD chambers that rather uniquely can be left unpowered between measurements. A DC ammeter (Pace Scientific) monitors amperage draw. A CR1000 datalogger (Campbell Scientific) controlled instrumentation timing (i.e. turning on and off the FD chambers, controlling timing of temperature and soil moisture sampling) and the datalogger, and together with batteries, cellular modem and additional parts were housed within a 14″ × 16″ fibreglass-reinforced enclosure. The enclosure was mounted to a small custom-fabricated tower, which was identical at the Woods Harbour and Gros Morne observational stations. At the Highlands site, the tower construction consists of a rugged 4-m-high ‘tee-pee’ style tower secured with deep rebar stakes to maintain solar panel clearance above the deep typical snow pack and also to withstand the high winds historically observed at North Mountain. The network of observational sites is managed remotely using cellular connections to each of the stations.

To compare the data collected by the FD chambers, we also monitored soil CO2 efflux using a LI-COR LI-8100 during field visits in August 2011. At each site, we installed 20 collars (6 cm diameter, 2·5 cm height) that were inserted to a depth of 2·5 cm into the soil. The 20 collars were distributed in a cross-shape design (N–S and E–W directions) with five collars in each direction. Starting at the centre, collars were installed 5 m apart. The measurements were taken within a 2-h time period. The purpose of this comparison was not to conduct an instrument intercomparison test, as the measurement technique itself has been previously proven in head–head tests in more controlled environments (Risk et al. 2011). Instead, the objective was to determine whether our chosen measurement locations were representative of the landscape, or whether they were high or low in terms of landscape soil CO2 efflux.

Computing and Processing Infrastructure

Telemetry was conducted using Campbell Scientific Raven CDMA modems, with scheduled calls to the field sites from an instance of Loggernet (Campbell Scientific) installed on a campus computer. The cellular modems at each of the observational stations were polled at 2-h scheduled intervals from the laboratory, and new data were collected. Following that, a sequence of automated scripts, also operating at 2-h intervals, conditioned the data files and made them available for viewing over the web. The scripts remove NULL values and replaced them with an interpolation between the two nearest neighbours. Any such replacements were automatically logged to an error file. The scripts then plotted all recorded values, across several time horizons (daily, weekly, monthly). These plots were then made available via javascript menu on a custom-designed CGI (Common Gateway Interface) web page. Some of the pages had interactive capabilities giving the user control over which variables to plot and over which time horizons. Graphical interfaces are available for currently operational sites can be seen at http://www.fluxlab.ca (Fig. 1).

Figure 1.

 Snapshot of data viewers available online. The panel is divided into three viewer modules including carbon accounting by Julian Day (JD), time-series plots accessible by Javascript menus (fluxes, temperatures and moisture) and lastly a custom plotting feature. Wavelet Coherence plots are also sometimes available for some sites.

As Loggernet has no facility for data conditioning or display, we built a custom solution based on a package of perl and cgi scripts. The scripts are packaged such that a script package, or folder, exists for each observational site. As these folders are identical to one another except for file input and output names, new sites can be brought online very easily by duplicating the folder and changing addresses. File FTP transfer, backup, processing and interpolation, error logging, plotting and web page creation were performed using this script package running on a Mac OSX server, executed hourly by the OSX-native chrontab scheduler, and using the operating systems native apache web server for cgi web display. The raw and conditioned data files, in addition to error logs specific to each site, are stored in the folders specific to each site. This structure readily supports a possible next step of database integration, email notification to operators or other add-ons, which would be very straightforward to integrate.

Statistical analysis

We used a one-way mixed analysis of variance (anova) model with repeated measurements (i.e. month) to test the effect of site on C effluxes. We also examined correlation between the probes for each site and between soil CO2 effluxes and soil temperature and soil moisture at the 30-min and daily frequency. Homogeneity of variance was investigated with residual plots, and data were log transformed when necessary. Finally, we also wanted to determine what temporal scale demonstrated the greatest variability. For this, we used Fourier transformations, which decompose time-series signals into frequency component each having an amplitude and phase. For this analysis, we used a subset of 2 months (June and July 2011) for each site. All statistical analyses were computed using sas.9.2 (SAS Institute Inc. 2008) except for Fourier transformations, which were performed with r 2.14.1 statistical software (R Development Core Team, 2011).

Results

Station Survivability, Performance and Power Consumption

These stations have shown that soil flux infrastructure can be maintained long term without grid power. After deployment, management of the stations has been minimal. The stations have survived under relatively harsh conditions: in temperatures from +35 °C to −30 °C (all sites); in repeated freeze–thaw conditions (Woods Harbour); under snow packs deeper than 2 m (Gros Morne, Highlands); in up to 5 consecutive days of thick fog where solar panels would not charge batteries; and in sustained winds above 100 km h−1 (all sites). After initial deployment, there were follow-up visits to one of the sites to address technical issues, but cumulatively, these equated to only a few days fieldwork.

Over the first year, the total data loss to power failure was <1% (total of 14 562 points) for all of the instrumentation at the Highlands (an average for the two chambers at the sites is 0·09%), and 0% (total of 16 346 points) reported data loss from Woods Harbour. Since its deployment on 18 August 2010, the station at Gros Morne has also reported no power failures (total of 14 051 points). Flux data that are less than zero occurs when the atmospheric reference records greater concentrations of CO2 than the FD concentration chambers. This occurred on occasion and was typically associated with heavy and persistent snow cover. Snow pack heterogeneity resulted in a high degree of spatial variation of snow pack properties which subsequently influenced the rates of diffusion of CO2 through the pack. This resulted in occurrences where the atmospheric reference and the soil chambers, though only metres apart, were temporally situated within different microenvironments (Risk et al. 2011). These data account for an average 16% of the Highlands data, 6·7% of the Woods Harbour data and 19·8% of the Gros Morne data. This issue is being addressed with the development of new prototype FD chambers that incorporate an atmospheric reference sensor into each soil chamber, making each FD chamber self-referencing. For the purposes of this manuscript, these under-snow data are excluded from this analysis. After the disappearance of snow, however, the sensors immediately returned to reliable values, and furthermore, re-calibration showed that minimal drift had occurred through the annual cycle in membrane diffusivities and Vaisala sensor calibration.

As noted in the methods, we interpolated across periods of power outage (i.e. Highlands site). Outages were very infrequent but typically long (>1 day) and associated with extended periods (>6 days) of fog where the station might consume all its battery reserve. As a result, the automated interpolation was, in practice, used only to conveniently reconnect the time series across one or two multi-day outages during the measurement period. There was almost no instances where automated interpolation was applied on shorter time-scales, because individual dropped measurements were almost non-existent.

During the course of the deployments, the average power consumption for each station was roughly 100 mA, or just under 30 W per day for two soil CO2 efflux measurement locations plus meteorological data. This is considerably more efficient than the LI-COR LI-8100, which can consume an equivalent quantity of power on an hourly basis.

Soil CO2 Efflux

The soil temperature pattern was similar for the three sites, while soil moisture content tended to decrease towards the summer for the Woods Harbour and Highlands sites (Fig. 2). For the three sites, soil CO2 efflux followed the seasonal pattern in soil temperature (average of soil surface, 10 and 30 cm depth is shown) with the highest daily average rates in July, August and September and the lowest rates in May, October and November (Fig. 2). Soil CO2 efflux was also highly responsive to major rain events with several peaks of soil efflux following sharp increases in soil moisture content (Fig. 3). During the study period, the three sites showed minimum values of soil CO2 efflux near zero, while two of the sites (Highlands and Woods Harbour) peak up to 40–50 μmol m−2 s−1 (Fig. 2). Our study also showed that despite warmer (average of 15 °C vs. 11·7 °C and 12 °C) and wetter (average of 0·37 v/v vs. 0·28 v/v and 0·11 v/v) soils at Woods Harbour, there was no significant difference for soil CO2 effluxes between the three sites (F-value = 2·24, P = 0·3086; Fig. 4).

Figure 2.

 Results of high-frequency measurements for CO2 fluxes, soil temperature and volumetric water content (moisture) for three Atlantic sites from May 2010 to August 2011. At each site, two FD chambers, one temperature sensor and one moisture sensor were monitored. Probe 2 of the Highlands site was relocated for a trenching experiment after the second week of August 2011.

Figure 3.

 Zoom in on the results of high-frequency measurements for CO2 fluxes, soil temperature and volumetric water content (moisture) for three Atlantic sites from 29 August 2010 to 8 September 2010. At each site, two FD chambers, one temperature sensor and one moisture sensor were monitored.

Figure 4.

 Box plots expressing monthly averages for CO2 fluxes for three Atlantic sites from May 2010 to August 2011. At each site, two FD chambers were monitored. The boundary of the box plot closest to zero indicates the 25th percentile, the line within the box marks the median, and the boundary of the box plot farthest from zero indicates the 75th percentile. Whiskers (error bars) above and below the box indicate the 90th and 10th percentiles.

The correlation between the two FD chambers at each site was generally good (Table 1). For half-hourly measurements, the results showed positive correlations of 0·70, 0·57 and 0·53 for the Highlands, Gros Morne and Woods Harbour sites, respectively. The correlation coefficients were slightly better on a daily basis with correlation coefficients of 0·79, 0·68 and 0·69 for each of the sites. Despite good responses to precipitation events, we detected overall weak negative correlations (i.e. r < −0·40) between soil CO2 efflux and soil moisture content (Table 1). In contrast, soil CO2 effluxes were positively correlated with soil temperature but varied highly between FD chambers and sites and tend to be stronger on a daily basis (Table 1).

Table 1.   Correlations between soil CO2 fluxes (probe 1 and 2) and soil temperature and soil moisture for the 30-min and daily frequency
SiteFlux probe 2Soil temperatureSoil moisture
Daily Gros Morne
 Flux probe 10·680·66−0·18
 Flux Probe 2 0·54−0·14
 Temperature  −0·63
30-min Gros Morne
 Flux Probe 10·570·55−0·10
 Flux Probe 2 0·50−0·08
 Temperature  −0·61
Daily highlands
 Flux Probe 10·790·62−0·40
 Flux Probe 2 0·30−0·40
 Temperature  −0·61
30-min highlands
 Flux Probe 10·700·47−0·31
 Flux Probe 2 0·25−0·10
 Temperature  −0·58
Daily woods harbour
 Flux Probe 10·690·74−0·34
 Flux Probe 2 0·47−0·18
 Temperature  −0·60
30-min woods harbour
 Flux Probe 10·530·65−0·24
 Flux Probe 2 0·34−0·10
 Temperature  −0·58

To compare our data collected with the FD chambers, we also measured soil CO2 efflux using a LI-COR LI-8100 at 20 locations. Our results showed that the two FD chambers at the Woods Harbour and Gros Morne sites were similar to what was recorded with the LI-COR (Fig. 5). In contrast, the FD chambers at the Highlands site showed only half the value measured with the LI-COR, likely due to the high spatial variability of CO2 efflux observed at the Highlands. There were additional field visits to the Highlands site during summer 2011 (July, August and October), and during these visits, additional LI-COR measurements were taken. When normalized, the averages of the three dates were 0·75 (sd = 0·25), 1·53 (sd = 0·47) and 0·91 (sd = 0·63) μmol m−2 s−1, respectively, which illustrates that spatial variability of soil CO2 efflux was changing over time at the Highlands site.

Figure 5.

 Recorded fluxes with a LI-COR LI-8100 and FD chambers. The bar graph for the LI-COR showed the mean and standard deviation of 20 measurements spatially distributed at each site. The bar graph for the two FD chambers showed the mean and standard deviation of the same 2-h period the LI-COR measurements were taken. Probe 2 of the Highlands site was relocated for a trenching experiment after the second week of August 2011.

With the help of the Fourier transformations, we learned that more than 45% of the temporal variability was found at the short-term scale (i.e. 6 h) (Table 2), which makes it clear that the dynamics of soil CO2 efflux are not slow-moving but have appreciable subdaily variability.

Table 2.   Results of the fast Fourier transform for each of the chambers at each of the sites. Percentage (%) of temporal variability explained by each period of time
Site6 h12 h24 h1 week1 month
Highlands
 Probe 162·97·05·49·713·9
 Probe 257·17·66·913·214·3
Gros Morne
 Probe 159·810·88·411·98·0
 Probe 253·310·49·115·410·5
Woods harbour
 Probe 145·011·212·118·812·1
 Probe 252·612·210·714·110·0

Discussion

The reliability of this infrastructure combined with the low cost of FD chambers offers great possibilities for long-term monitoring of spatially distributed grids, and for modelling of empirical soil carbon data. The FD chambers were very successful in key areas including survivability, management intensity and cost. The FD chambers were approximately ten times more energy efficient and four times more affordable (i.e. $13 000 vs. $50 000) than a conventional method such as a LI-COR LI-8100. In addition, of the >10 000 measurements made by the FD chambers at each site, <1% of the data were lost because of power failure. In addition, with the FD chambers, there are further opportunities for reductions in power consumption, which will primarily be associated with logic togglings of optic heaters on the Vaisala sensors, or identification of IRGA sensors with decreased power requirements. As the existing Vaisalas are powered for 15 min of every half-hour so that full accuracy can be achieved, significant reductions in power consumption could come with sensors that have faster warm-up times, and correspondingly lighter duty cycles.

This infrastructure presents rather unique opportunities in soil CO2 flux research, particularly where temporal dynamics are of interest. Short-term events could be observed, notably those associated with precipitation, which are challenging to observe using conventional monitoring techniques. For example, in this study, we found that despite an overall weak negative relationship between moisture and CO2 fluxes, there was visual evidence that the short-term effect of precipitation on soil CO2 fluxes was particularly important for the Woods Harbour and Highland sites. This was supported by the Fourier transformations results, which also demonstrated the dominance of short-term temporal variability (<6 h), and illustrated the importance of high-frequency measurements for explaining the temporal variability of soil CO2 fluxes. Thus, with long-term, continuous data sets, we will be able to better assess the contribution of short-term events to the relative spectrum of temporal variability, with respect to (i) when these short-term events, or spikes in CO2, occur, (ii) their respective magnitude in relation to changes in presumed drivers, (iii) the duration of the events and (iv) any residual or indirect effects that these short-term events may have on subsequent fluxes.

Spatio-temporal variability of soil CO2 fluxes in small or large spatial grids can be more easily assessed with this infrastructure. This study showed high temporal variability for three cool temperate ecosystems, and high spatio-temporal variability at one site. Ideally, we would use additional FD chamber units with each station to better capture this spatio-temporal variability rather than using a LI-COR during field visits and include the observations in modelling predictions. As costs are reduced and the FD chambers become commercially available, we propose that deployments of at least 12–25 FD chambers per site should be sufficient to capture spatio-temporal variability of soil CO2 efflux, although the exact number of FD chambers will be dependent on a comprehensive spatial analysis experiment that will detect the range of autocorrelation. In fact, we will attempt to perform such experiments at several sites during the coming year. In addition, with these new FD chambers capable of working under most snowy conditions, we expect to have a better picture of what to expect from soil CO2 fluxes under climate warming. It is expected that within the coming year, the FD chambers will have the capacity to measure under a wider range of snowy conditions, which will further extend the capability of these stations.

With this study, and as other studies have shown before, we illustrated that moisture and temperature were important contributors to temporal variability of soil CO2 fluxes. But in contrast to previous studies, we emphasized, with this new technology, that high-frequency measurement is essential to improve our understanding of temporal variability of soil CO2. Our results imply that many biological (e.g. vegetation) and physical (e.g. diffusivity) factors could explain soil CO2 efflux spatio-temporal variability (Baldocchi, Tang & Xu 2006; Moyano, Kutsch & Rebmann 2008; Vargas & Allen 2008; Vargas et al. 2010; Casals et al. 2011; Taneva & Gonzalez-Meler 2011; Luan et al. 2012). For instance, Vargas & Allen (2008) showed that C effluxes were positively correlated with type of vegetation, and fine roots and rhizomorphs length. Similarly, Luan et al. (2012) also demonstrated that the spatial variation of C effluxes was affected by total porosity, water-filled pore space and bulk density. However, this new type of soil observational station presents opportunities in that regular and reliable data flow can finally be fused in real time with modelling. This fusion would drive faster learning about a potentially important climate feedback mechanism. In this way, this infrastructure has direct applications for research related to soil CO2 processes in natural systems, validation of GCM models, calibration of remote sensing data, greenhouse gas accounting and monitoring at carbon capture and storage (CCS) and soil carbon offset projects.

Acknowledgements

We extend kind thanks to Parks Canada for logistical assistance in the Cape Breton Highlands and Gros Morne parks, and Reid Nickerson for hosting the Woods Harbour station on his property. Thanks also to Lisa Kellman for occasional loan of her LI-COR LI-8100 and Nick Nickerson for programming expertise.

Ancillary