a. Study site

In situ CO₂ measurements began in October 2001 in the village of Berezorechka (56°08′N, 84°19′E, ∼150 m above mean sea level), Tomsk state, in western Siberia. The location of the observation site is shown in Fig. 1. The site is located in an extensive boreal forest dominated by Picea sylvestris and Betula pendula, with a canopy height typically at about 15 m. The land is generally covered with snow from October to May. The population of the village is on the order of dozens, and no large-scale agriculture or industry occurs here. The largest city near Berezorechka is Tomsk, which is located about 55 km northeast of the village.

b. Measurement system

Atmospheric air is taken from air sample inlets situated at heights of 5, 20, 40, and 80 m on a 90-m radio relay tower located on the outskirts of the village, and polyethylene tubes connect the inlets to the CO₂ measurement system housed in a laboratory at the base of the tower. Each inlet is placed about 3 m away from the tower on an extension arm. The laboratory is a converted International Organization for Standardization (ISO) 20-ft freight container lined with insulation to reduce heat loss during the cold Siberian winters.

The developed CO₂ measurement system employs a triple dehumidification system to dry the sample air and also has a standard gas saving system to allow long-term observation in remote areas without the frequent replacement of standard gas tanks. A schematic diagram of the CO₂ measurement system is shown in Fig. 2. Sample air is continuously drawn in through the inlet by a diaphragm pump (N86KNE, KNF, Germany); the pressure is maintained at 0.15 MPa by a back-pressure valve. The pressurized air is dried by 1) adiabatic expansion in a glass water trap, 2) a semipermeable membrane dryer (PD-625–24SS, Permapure, United States), and 3) a magnesium perchlorate trap. The first stage of the drying system applies water condensation associated with adiabatic expansion, which decreases the temperature of the sample air that flows from the narrow tube into a spherical glass trap. The condensed water in the water trap is periodically exhausted through the valve attached underneath. The semipermeable membrane dryer used in the second drying stage removes water vapor from the pressurized inner tube to an outer tube that serves as a conduit for the flow of purged gas. The material of the inner tube also assists in absorbing water vapor. In this system, sample air flowing through the inner tube is divided into two flows. One is introduced into a nondispersive infrared analyzer (NDIR) (LI-7000, LI-COR, United States) at a constant flow rate of 35 cm³ min⁻¹ using a mass flow controller (SEC-4400, STEC, Japan) via a multiport valve and magnesium perchlorate. The other enters the outer tube of the semipermeable dryer, which dries the sample air at a rate over 100 cm³ min⁻¹. The third stage of the drying system employs the magnesium perchlorate trap and is set after the multiposition valve. Sample air and working standard gases flow into this trap and retain their dewpoint at around −50°C. A stainless steel tube with dimensions of 2 cm for the inner diameter and 10 cm in length is filled with magnesium perchlorate, which is replaced every month. The CO₂ concentration is defined as the mole fraction in dried air, and water vapor correction has not been adopted. The sample air is pumped in for 6 min, and the average value for the last minute is retained. The flow rate of the reference gas used for the NDIR is set at 3 cm³ min⁻¹ by another mass flow controller (SEC-400 MARK3, STEC, Japan).

A small computer (CPM-2C, OMRON, Japan) managed the analysis operation, and the signals were recorded to a datalogger (NR-1000, KEYENCE, Japan) every 2 s from October 2001 to April 2003. Since April 2003, both the analysis operation and data logging are performed by a measurement and control system (CR10X, CAMPBELL, United States). Stored data are retrieved once a month, at which time both a system check and replacement of the magnesium perchlorate take place.

c. Standard gas saving system

Because of the logistical problems associated with performing measurements at remote sites in Siberia, it is advantageous to reduce the consumption rate of the standard gases. To achieve this, the measurement system compresses the air on site. The on-site compressed air is used as the reference gas for the NDIR and also used as a subworking standard gas to track the baseline drift of the NDIR. The subworking standard gas assumes the role of standard gas apart from the three working standard gases.

The three working standard gases shipped from our laboratory are CO₂-in-air standard gases; the concentrations were selected to be around 340, 365, and 390 μmol mol⁻¹ and determined to the second decimal place in our laboratory using the National Institute for Environmental Studies (NIES) standard gas scale. The differences between the NIES and the National Oceanic and Atmospheric Administration Earth System Research Laboratory Global Monitoring Division (NOAA/ESRL/GMD) scales are within 0.12 μmol mol⁻¹ in a range of 343–372 μmol mol⁻¹ (Peterson et al. 1997). Errors associated with extrapolating calibrations over 390 μmol mol⁻¹ are evaluated in our laboratory. The difference between the extrapolated value and our standard gas scale value is generally within 0.1 μmol mol⁻¹ of 450 μmol mol⁻¹.

The subworking standard gas is introduced to reduce the consumption of the working standard gases when estimating the baseline drift of the analyzer. This method utilizes the fact that the NDIR signal is mainly due to baseline drift, not span drift. Figure 3 shows the test result obtained by analyzing three working standard gases and a subworking standard gas every 12 h and 30 min, respectively, by the system in our laboratory. The signal difference in each gas is predominantly constant relative to baseline drift.

On-site compressed air is dehumidified ambient air taken from the 80-m-height inlet and pressurized to about 0.5 MPa in a 0.048 m³ aluminum cylinder, and it is used as the subworking standard gas in this method. The taken air is first evacuated from the system by a two-way valve for 1 min to stabilize the dehumidifying conditions of the semipermeable membrane dryer. The air is then compressed into a cylinder with a flow rate of about 650 cm³ min⁻¹ for about 5 h by a pump (LOA-P103-NL, GAST, United States) through 1) a water trap with restricted vent, 2) a semipermeable membrane dryer (SWF-M06–400, AGC, Japan) with purge gas supplied by the NDIR exhaust, and 3) a magnesium perchlorate trap that is the same type of trap used for sample air drying. The air is kept for about 10 days before using. The dewpoint differs depending on the weather condition and almost always ranges between −10° and 0°C. Two cylinders are available for on-site compressed air: when one cylinder is in use, the other is being prepared. The cylinder is automatically replaced when its inner pressure decreases to 0.1 MPa. The pressure of the on-site compressed air flowing from the cylinder is maintained at 0.07 MPa, which is the same for the three working standard gases.

Sample air from the tower and the subworking standard gas are analyzed every 30 min. Three CO₂-in-air working standard gases are analyzed twice a day, and the signal drift of the working standard gases over 12 h are estimated using the temporal variation of the subworking standard gas signals. Using the subworking standard gas conserves the working standard gas by a volume of about 0.15 m³ in atmospheric pressure per cylinder each year.

d. Assessment of accuracy

To estimate the CO₂ concentration stability of subworking standard gas, the CO₂ concentration in the on-site compressed air was measured. The concentration was determined from four standard gases every 30 min using the measurement system in our laboratory, and its variation is shown in Fig. 4. This analysis frequency is the same as the field measurements in Berezorechka. The CO₂ concentration of the first analysis was 380.93 μmol mol⁻¹, which is relatively lower than the concentration results for the following analyses; this shows that the replacement of air in the NDIR reference cell is insufficient when one subworking standard gas is changed to another one. Consequently, the signal for the first analysis of a subworking standard gas was excluded from data processing. During the 2nd and 39th analyses, corresponding to 19.5 h, the subworking standard gas concentration was stable with a standard deviation of 0.07 μmol mol⁻¹.

The total precision of this system also depends on estimating the signal for working standard gas from the subworking standard gas variation. To quantitatively examine the baseline and span drifts, three gases with known CO₂ concentrations of 344.53, 358.67, and 368.45 μmol mol⁻¹ were measured using the NDIR separated from the system in our laboratory. These standard gases were analyzed every 30 min, and the variations of the signals are shown in Fig. 5 along with differences from each first analysis value. Systematic differences in their concentrations were negligible; most of the variation was due to baseline drift. The standard deviation of delta CO₂ signals for the three standard gases was 0.29 mV on average. In this system, the subworking standard gas was calibrated by working standard gases every 12 h, and the signals of the working standard gases could thus be estimated by the subworking standard gas to within 0.29 mV, which corresponds to a CO₂ concentration of 0.03 μmol mol⁻¹. This value was sufficient for analyzing the on-site atmospheric concentration.

To check the CO₂ concentration accuracy and stability for the atmospheric air flowing through the triple dehumidification system, air with a known CO₂ concentration and compressed into a 0.048 m³ aluminum cylinder was analyzed over the whole air intake line for the system in our laboratory. First, the air in the cylinder was bubbled in deionized water to increase the water vapor content and equilibrate the CO₂ concentration. After ensuring that the CO₂ concentration was stable, the air was forced from the inlet through the entire sample line of the system using a diaphragm pump. The CO₂ concentration of the air was analyzed 14 times, and the measured value was 348.96 μmol mol⁻¹ with standard deviation of 0.33 μmol mol⁻¹, which is 0.13 μmol mol⁻¹ lower than the known value. The magnesium perchlorate used to dehumidify the sample air was found to absorb CO₂ in our experiment (Watai et al. 2006) and is a potential cause for some of the discrepancy in CO₂ values. Laboratory experiments on both the subworking standard gas and air intake line show that the total precision of this system is 0.3 μmol mol⁻¹.

e. Other measurements

Air temperature and relative humidity are measured by the sensors (HMP45D, Vaisala, Finland) at heights of 5, 40, and 80 m on the tower; the sensors are installed in radiation shields that actively aspirated, and the wind direction and speed (model 81000, R. M. Young, United States) are taken at 40 m. Solar radiation (CM3, Kipp and Zonen, Holland) and precipitation (Model 52202, R. M. Young, United States) are measured on top of the container laboratory where the measurement system is housed. The air temperature exhibits clear seasonal variations: the daytime maximum exceeds 30°C and occurs between June and July, and the wintertime temperature is generally below 0°C from November to February, with the minimum value occurring between December and February. Extremely low temperatures with values below −20°C and lasting for several days are sometimes observed in the winter. Annual precipitation is about 600 mm, with slightly more during the warm season than in the cold season. The wind speed is relatively weak (up to 2–5 m s⁻¹) from June to August and relatively stronger (up to 2–8 m s⁻¹) in other seasons. The wind is usually from the south during December–March but is less distinct at other times.

Ozone (O₃) concentrations are measured at heights of 5 and 40 m. Air is introduced into the analyzer (model 1150, Dylec, Japan) through Teflon tubes, and data are saved every 30 min. The O₃ data are useful in analyzing the CO₂ vertical distribution and variations. Radon-222 is also measured at 5- and 40-m heights every 30 min using an electronic radon monitor (Iida et al. 1991).

a. Diurnal variation

Figure 6 shows typical diurnal and day-to-day variations for CO₂ observed at Berezorechka during February (proxy for winter) and July (proxy for summer) 2002. The time shown is the local time at the observation site. The diurnal variation was small or nonexistent during the winter but had large amplitudes in the summer; the highest values were just prior to sunrise, and the lowest values were in the afternoon. These characteristics are consistent with other CO₂ observations made at various continental forested sites (Bakwin et al. 1995; Haszpra et al. 2001; Higuchi et al. 2003; Inoue and Matsueda 2001).

Figure 6 also shows the day-to-day variations in temperature, solar radiation, and wind speed that are associated with changes in CO₂. A notable feature was the rise in CO₂ that was observed from the evening of 24 February to the following morning. The increase was first observed at the lowest level on the tower and then successively at higher levels, with a gradual decrease after sunrise at all levels. The meteorological conditions associated with the CO₂ phenomenon showed a vertical temperature gradient between the 5- and 80-m levels of about 10°C under relatively calm conditions (wind speeds of less than 3–4 m s⁻¹) and indicated the formation of surface stratification. Small CO₂ increases followed by decreases with amplitudes of 3–7 μmol mol⁻¹ were observed from the evening to the next morning for both 25–26 and 26–27 February. Temperature variations suggest that the surface stratification also occurred, although the gradients were weak relative to the night of 24–25 February. The winter soil respiration under the snowpack has been reported to provide a substantial contribution to the boreal CO₂ efflux (Zimov et al. 1996; Winston et al. 1997). The winter inversion layer that was observed for these days can trap CO₂ emitted from the surface. The observed radon-222 distribution also showed that the soil was able to respire CO₂ through the snow and into the atmosphere (Moriizumi et al. 2008, personal communication).

Notably high CO₂ concentrations near the surface exceeding 400 μmol mol⁻¹ were sometimes observed in winter. Figure 7 shows the temporal CO₂ variation observed in January 2004 along with the temperature, wind speed, atmospheric pressure, and O₃ concentration. The CO₂ concentration generally increased from 11 to 23 January, with repeated fluctuations in concentration. The maximum CO₂ concentration observed during this period was about 440 μmol mol⁻¹; the concentration then decreased drastically on 24 January. The temperature was low (under −10°C) for most of this period, and the variation at both 5- and 80-m heights showed that stratification occurred every night under the 80-m height. The wind speed was generally lower compared to other periods, with a value under 4 m s⁻¹, and the atmospheric pressure was over 1010 hPa, which suggests that a Siberian anticyclone was probably situated over the region during this period. The CO₂ accumulation over several days in winter was probably due to the creation of an inversion layer at night with a boundary layer cap under relatively calm and cold conditions in a Siberian anticyclone. The O₃ concentration was low during this period and generally decreased with time; its variations were like a mirror image of the CO₂ variation. The O₃ variation verified that stable stratification occurred near the surface, because O₃ is destroyed near the ground by dry deposition.

During the warm season, the observed CO₂ level showed a diurnal variation with large amplitudes caused by biospheric activity; the nighttime high and daytime low values of CO₂ were typical of those for a boreal forest site, as shown in Fig. 6. However, there were times (e.g., 29–30 July) when this diurnal pattern broke down to show large diurnal day-to-day variations. These variations reflect a complicated interaction between the atmospheric conditions and biospheric activity. On each of the calm nights for 27–28 and 28–29 July, a surface inversion layer developed, causing a CO₂ buildup exceeding 440 μmol mol⁻¹ at 5 m and producing a large vertical CO₂ gradient on the order of 100 μmol mol⁻¹ between the ground surface and 80-m level. The night of 29–30 July, however, was cloudy and windy. These meteorological conditions persisted into the daytime of 30 July, which suppressed photosynthesis (because of lower radiation) and produced a well-mixed boundary layer. The vertical gradient of CO₂ therefore decreased to less than 5 μmol mol⁻¹, and relatively little variation in CO₂ for 29–30 July resulted. The daytime CO₂ on 30 July was higher than the levels observed for 27 and 28 July because of the increased influence of respiration.

Figure 8a shows seasonal changes in the monthly average of diurnal variations for the 5-, 20-, 40-, and 80-m height levels on the tower for 2003 and 2004, with a large diurnal amplitude in July caused by respiration during the night and photosynthesis during the day that became smaller as winter approached. This diurnal variation was also observed in April 2004; however, very little diurnal variation was observed at all height levels in December 2003. During the photosynthesis active season, the lowest height level of 5 m showed the largest amplitude of about 62 μmol mol⁻¹ (July 2003), whereas 3–4 μmol mol⁻¹ was observed in winter (December 2003). The diurnal amplitudes at higher heights (e.g., 80 m) were smaller (about 17 and 1.5 μmol mol⁻¹ in July and December 2003, respectively) and affected less by local sources and sinks, because the CO₂ concentrations at these heights incorporated advective influences.

The covariance between the atmospheric mixing in the PBL and biospheric CO₂ flux (rectifier effect) influenced the CO₂ gradient near the surface. Figure 8b shows the diurnal variation of the CO₂ difference between 5 and 80 m that was observed at Berezorechka. The nighttime CO₂ gradient reflects the interplay between the low-level stratification and vegetative respiration. Respiration rates depend on the temperature and soil moisture and are thus largest during the summer. The daytime atmosphere is relatively well mixed, but the CO₂ has a small vertical gradient that varies depending on the month. The CO₂ concentration was lowest at the lower heights in July and September 2003 and April 2004 because of vegetative uptake at the surface; however, during December 2003, the vertical gradient was smaller and sometimes reversed, with higher concentrations at low levels. The observed daytime CO₂ concentration gradient was generally constant for 0800–1900 LT, with a value of about 3 μmol mol⁻¹ in July 2003. The CO₂ difference during the day was equally stable in other months, but the number of hours for the stable period was seasonally dependent, which corresponded to the changing periods of daylight.

b. Comparison with aircraft measurements

Aircraft measurements provide a vertical context for interpreting the tower data and understanding the atmosphere–biosphere exchange. The vertical profiles for the CO₂ concentration over Berezorechka were measured using the small aircraft Antonov-2. In the aircraft, sample air was dried with magnesium perchlorate and introduced into the NDIR cell (LI-820, LI-COR, United States). The CO₂ concentration of the samples was determined relative to high and low standard gases pressurized to 15 MPa in aluminum cylinders with 0.010 m³ volume. The system is very similar to the one used in our small unmanned aerial vehicle (Watai et al. 2006). Routine flights took place mainly in the afternoon once every 2–3 weeks throughout the year, and intensive flights took place several times in one day for a few days in the summer.

To ascertain how well the measurements on the tower represented the CO₂ variation in the PBL, the CO₂ vertical profiles obtained by the aircraft were compared with the CO₂ variation obtained from the tower. The diurnal variations of the CO₂ vertical profiles for 15 and 8 July 2003 are shown in Figs. 9a,b, and their temporal variations are shown in Figs. 9c,d, respectively. The PBL heights shown in Figs. 9c,d were determined from the vertical profiles for the temperature, specific humidity, and CO₂. These two sets of profiles show distinctly different temporal characteristics. The vertical profiles for 15 July show an almost classical case of decreasing CO₂ concentration in the PBL, resulting from daytime photosynthesis, accompanied by a decreasing vertical CO₂ gradient as the turbulent mixing and PBL height increased. Around 1915 LT, the CO₂ concentration was uniformly distributed with the height, and the average CO₂ in the PBL (PBL mean value) was around 353 μmol mol⁻¹. The CO₂ concentration observed at the tower similarly decreased with time, and the PBL mean value (indicated by solid dots in the figure) was consistent with the gradual decrease in CO₂ observed at the tower. The vertical profiles for 8 July show a relatively complicated pattern of changes in CO₂ concentration with height and time. The PBL height increased in the morning, reversed and began to decrease around 1145 LT, and remained relatively low and uniform in the afternoon. Detailed meteorological analysis suggests that, on 8 July, the CO₂ field at Berezorechka was affected by an airmass exchange, which caused CO₂ inside the PBL to decrease quite rapidly during the morning hours. The CO₂ level reached a minimum of nearly 340 μmol mol⁻¹ shortly after noon and increased again thereafter. Good consistency between the CO₂ concentrations at the 80-m height and the PBL mean value was also observed with the rapid meteorological change. The 80-m value on the tower accurately represented the PBL mean value, with a difference of less than 2 μmol mol⁻¹.

The seasonal variation in the CO₂ difference between the 80-m and PBL mean values was also examined. We obtained the PBL average for 108 CO₂ daytime afternoon profiles taken between October 2001 and March 2005 by routine aircraft observations and compared them to in situ CO₂ measurements obtained at 80 m on the tower. Figure 10 shows the CO₂ differences between the 80-m values and the aircraft-derived PBL mean values. Large differences with a maximum of about 25 μmol mol⁻¹ were sometimes observed during the winter season, when an intense Siberian anticyclone was in the area for several days; this produced a very strong shallow inversion layer around the region (shown as circled dots in Fig. 10). Except for these data points, 99% of the data agreed to within ±4 μmol mol⁻¹, and 94% agreed to within ±3 μmol mol⁻¹. When examining the difference in values between those taken at 40 m on the tower and the aircraft-derived PBL mean values, 98% of the data were within ±4 μmol mol⁻¹ and 93% were within ±3 μmol mol⁻¹ (not shown). During the growing season (July and August), the 80-m values were often less than the aircraft-derived mean PBL because of the influence of strong surface photosynthesis. Overall, the measurements taken at 80 and 40 m were representative of the changes that took place inside the PBL.

c. Seasonal variation

To examine the seasonal CO₂ cycle observed at Berezorechka, it is necessary to determine a time interval during which the CO₂ concentration is relatively stable and vertically well mixed. For reference, Bakwin et al. (1995) and Higuchi et al. (2003) used data obtained during 1500–1700 LT as the daily mean afternoon concentration when describing the CO₂ long-term variation at a tall tower in North Carolina and at a boreal forest site in northern Ontario, Canada, respectively. In slight contrast, Inoue and Matsueda (2001) used data for 1300–1600 LT that were taken from a tower in Tsukuba, Japan. At Berezorechka, the daytime afternoon data were also relatively stable and represented the PBL mean value well, but the period differed from month to month, as shown in Fig. 8. The concentration was almost stable in 1200–1900 LT in July 2003, but the stable period shifted to 1200–1700 LT and 1300–1700 LT in September 2003 and April 2004, respectively. The CO₂ concentration slightly decreased with time in the warm season and reached its daily minimum at different times depending on the month (e.g., around 2100 LT in July 2003 but about 1700 LT in September 2003). The concentration observed in December 2003 increased slightly with time. To normalize the hours for discussing the seasonal CO₂ cycle, four datasets over 1500–1700 LT were averaged and used because the CO₂ concentration was relatively stable during this time throughout the year. Note that the obtained daytime afternoon CO₂ concentration was not always the minimum daily value. For example, the CO₂ concentration was about 2.5 μmol mol⁻¹ higher than the minimum value around 2030 LT in July 2003, and it was 1.5 μmol mol⁻¹ lower than around 1200 LT in December 2003. The daytime afternoon CO₂ values were not necessarily the daily minimum, but they reflected the periods of most stability.

After deleting some extremely high CO₂ anomalies (associated with major low-level inversion stratifications in the wintertime), the daily mean afternoon values were analyzed using the digital filtering technique developed by Nakazawa et al. (1997a). The resultant average seasonal variation at Berezorechka, along with those at Cold Bay (55°N, 162°W) and Mace Head (53°N, 9°W), is shown in Fig. 11. The seasonal cycle at Berezorechka was characterized by a minimum in late July and a maximum in mid-January, with a seasonal amplitude of 30.9 μmol mol⁻¹. In comparison, the seasonal CO₂ amplitudes observed at Cold Bay and Mace Head, which are located at a similar latitude as Berezorechka, showed values of 14–16 μmol mol⁻¹ (Tans and Conway 2005); these stations, located on the continental rim, represent the background CO₂ concentration field. The seasonal cycles observed at these background stations showed a time lag of about a month with respect to the seasonal cycle observed at Berezorechka. These results are consistent with existing knowledge regarding the difference in seasonal CO₂ variations between continental and background stations. However, the maximum value at Berezorechka was observed during the coldest season, when large-scale vertical mixing was suppressed, and occurred 3–4 months earlier than at the two background stations. In comparison to Syktyvkar, which is located in the eastern European taiga and shows a seasonal amplitude of 22.1 μmol mol⁻¹ (Sidorov et al. 2002), and Zotino, which is located in central Siberia and shows a seasonal amplitude of 24 μmol mol⁻¹ (Lloyd et al. 2002), the seasonal amplitude at Berezorechka was noticeably larger. This reflects the influence of the heterogeneity of the vegetation and the different atmospheric transport characteristics. We are currently conducting a detailed analysis of the seasonal cycle being observed at various monitoring stations in Siberia.