Using Eddy Covariance Sensors to Quantify Carbon Metabolism of Peatlands: A Case Study in Turkey

2.1. Description of Study Site

The Yenicaga peatland is located about 38 km east of the city of Bolu (40°47N′, 32°1′E) in the western Black Sea region of Turkey (Figure 1).

An EC flux tower site was installed about 1 km north of Lake Yenicaga (18 km²) at the elevation of 988 m above seal level on July 12, 2010. The climate in the Yenicaga region is classified as cool temperate, with a mean annual temperature and precipitation of 10.2 °C and 538 mm, respectively [5]. The Yenicaga peatland is reported to contain Devonian and cretaceous limestone, basaltic tuff, lava, and olistolites, with the uppermost layer consisting of tertiary and quaternary formations [5]. The natural vegetation types of the Yenicaga peatland consist of the following dominant communities: (1) Phragmites australis and Typha domingensis (41.2 ha); and (2) Ranunculus lingua, Acorus calamus, Najas marina, Pedicularis palustris, Senecio paludosus (107.8 ha) [6]. The mean vegetation height around the flux tower is about 0.5 m, and the terrain observed around the flux tower exhibits flat and uniform grasslands.

2.2. Eddy Covariance Flux and Ancillary Measurements

Net ecosystem exchange (NEE) rates of carbon dioxide (F_c, mg m⁻² s⁻¹) in Yenicaga peatland were estimated using an eddy covariance (EC) system consisting of an open-path CO₂/H₂O gas analyzer (LI-7500, Licor Inc., Lincoln, NB, USA), a 3-D sonic anemometer/thermometer (CSAT3, Campbell Scientific Inc., Logan, UT, USA), a data logger (CR3000, Campbell Scientific Inc.), and a 3-m tower on which EC flux sensors were mounted. The distance between the LI-7500 and CSAT3 sensors was 0.15 m, with CSAT3 oriented towards the prevailing wind direction (an azimuth angle of 30° from true north) and LI-7500 vertically rotated 15° towards the footprint.

Eddy fluxes and associated signals were recorded at 10 Hz, block averaged over one hour (h) and corrected for the effects of fluctuations in air density on CO₂/H₂O fluxes (F_{c_wpl} and LE_wpl with WPL correction) through the online flux computation. EC data were collected swapping two 2-GB Compact Flash cards at 14-to-18-day intervals. The net radiation (R_n), downwelling and upwelling longwave (4 to 50 μm) and shortwave radiation (0.2 to 4 μm) (R_{l_dn}, R_{l_up}, R_{s_dn}, and R_{s_up}, W m⁻²) were measured using Kipp & Zonen CNR-4 net radiometers (Kipp & Zonen USA Inc., Bohemia, NY, USA). Air temperature (T_a, °C), and relative humidity (RH, %) were sampled using HMP45C probe (Vaisala, Finland). Precipitation (PPT, mm), evapotranspiration (ET, mm), soil water content (SWC, %), and mean, maximum and minimum soil temperature (ST, ST_max, and ST_min, °C) were measured on a hourly basis using ET107 weather monitoring station (Campbell Scientific Inc.). Photosynthetically active radiation (PAR) was estimated from net shortwave radiation (R_{s_n}) using a conversion factor of PAR: R_{s_n} = 0.5 [7–9]. The values of CO₂ fluxes in unit of mg m⁻² s⁻¹ were also converted to unit of kg C ha⁻¹day⁻¹, based on the following conversion ratios of mg:kg = 10⁶; s:day = 86,400; m²:ha = 10,000; and CO₂:C = 44/12.

2.3. Data Processing and Analyses

As with the conventional meteorological sign notation, downward and upward fluxes are considered negative (−) and positive (±), respectively. The first step of data processing involved the removal of erroneous spikes and their associated CO₂ fluxes when latent heat flux (LE_wpl) < −100 or > 800 W m⁻²; sensible heat flux (H_s) < −150 or > 500 W m⁻²; and precipitation events occurred [10]. For both hourly daytime and nighttime CO₂ fluxes, descriptive statistics were given, and best-fit cumulative distribution function (CDF) was selected. Tukey’s multiple comparison was performed after one-way analysis of variance (ANOVA) for the entire dataset in order to test significant differences in hourly, nighttime versus daytime and monthly means. All the statistical analyses were performed with Minitab 15.1 (Minitab Inc. 2006).

Second, EC data were separated into daytime (R_n and/or R_{s_dn} > 10 W m⁻²) and nighttime (R_n and/or R_{s_dn} ≤ 10 W m⁻²) periods since EC data are more reliable during daytime hours than during nighttime hours, and nighttime EC data ( $F_{\mathrm{c}\_\text{wpl}}^{\mathit{night}}$) can be used to estimate daytime as well as nighttime ecosystem respiration (both plant and soil respiration) ( $R_E=F_{\mathrm{c}\_\text{wpl}}^{\mathit{night}}$). Negative night CO₂ flux data were deleted as no gross primary productivity (GPP = 0) occurs during the nighttime. The site-specific threshold value of friction velocity (u*) was determined as 0.03 m s⁻¹ below which low vertical wind velocity led to underestimation of the nighttime CO₂ fluxes [11]. Likewise, nighttime periods where horizontal wind velocity was less than 1 m s⁻¹ were removed from the dataset. Finally, night fluxes where CO₂ density data had the standard deviation of > 14 mg m⁻³ (0.6 μmole m⁻³) were eliminated from further analyses [12]. Multiple (non-)linear regression models were fitted to the resultant night CO₂ dataset, and best-fit M(N)LR models with and without the inclusion of temporal variables (hour, and month) were chosen using best subsets procedure (low Mallows’ C_p, high adjusted R², and low SE). The following soil respiration equations of RothC [13] and CENTURY [14] models were also used to model daytime R_E:

Hourly GPP values were estimated as the sum of NEE and daytime R_E. Daytime R_E was obtained extrapolating nighttime R_E fluxes to the remaining daytime fluxes, based on M(N)LR models.

3.1. Diurnal CO₂ Fluxes

Descriptive statistics of despiked and averaged EC data for the period of July 12 to October 17, 2010 indicated that mean hourly CO₂ fluxes above the canopy had a larger temporal variability (CV = 498) than the rest of the measured variables and varied between −1.5 and 1.5 mg m⁻² s⁻¹ (Table 1). During the study period, the study site cumulatively received 92 mm PPT and lost 305 mm water vapor via ET. The mean Bowen ratio of H_s:LE_wpl quantifying the evaporative demand of an environment was estimated at 0.37, thus characterizing the Yenicaga peatland as a mesic environment. A logistic CDF with a location of 0.05197 and a scale of 0.1596 appeared to fit our daytime and nighttime F_{c_wpl} data fairly well as follows (r = 0.99; n = 1884; P < 0.001) (Figure 2):

Quartic functions (an equation of fourth degree) fitted to both entire and monthly F_{c_wpl} datasets as a function of the explanatory temporal variable (h) provided a meaningful representation of CO₂ fluxes, with a R²_adj value range of 70.9% in July to 31.3% in October (Figures 3 to 7).

3.2. Multitemporal Comparisons of CO₂ Fluxes

The entire dataset of daytime and nighttime CO₂ fluxes (n = 1,884) was used for a multiple comparison according to hours, nighttime versus daytime, and months, based on Tukey’s test following one-way ANOVA (Table 2). A comparison of hourly mean F_{c_wpl} values showed that the Yenicaga peatland acted as a CO₂ sink for the daytime periods between 9:00 AM and 17:00 PM and acted as a CO₂ source for the periods between 18:00 PM and 8:00 AM. On average, the Yenicaga peatland had minimum and maximum CO₂ fluxes ranging from −0.05 ± 0.21 mg CO₂ m⁻² s⁻¹ at 9:00 AM to −0.28 ± 0.18 mg CO₂ m⁻² s⁻¹ at 12:00 PM as a CO₂ sink and from 0.05 ± 0.12 mg CO₂ m⁻² s⁻¹ at 18:00 PM to 0.38 ± 0.29 mg CO₂ m⁻² s⁻¹ at 5:00 AM as a CO₂ source, respectively (Table 2).

Mean daytime CO₂ flux (−0.11 ± 0.22 mg CO₂ m⁻² s⁻¹) significantly differed from mean nighttime CO₂ flux (0.26 ± 0.23 mg CO₂ m⁻² s⁻¹) (P < 0.001). All the monthly mean CO₂ fluxes were positive (upward into the atmosphere) and varied between 0.03 ± 0.35 mg CO₂ m⁻² s⁻¹ in July and 0.07 ± 0.25 mg CO₂ m⁻² s⁻¹ in September. However, the monthly mean CO₂ fluxes did not appear to significantly differ from one another (Table 2). The best-fit MNLR model was derived from the entire dataset of nighttime and daytime F_{c_wpl} by exploring possible interaction effects among the EC and meteorological variables as follows:

3.3. Ecosystem Components of Carbon Metabolism

Nighttime R_E was modeled using (1) best-fit M(N)LR models with/without the forced inclusion of temporal variables (hour and month), and (2) ST-dependent R_soil equations [Equations (1) and (2)] of RothC and CENTURY models with/without the forced addition of SWC. All the R_E models were built with the intercept forced through zero as follows:

A comparison of daytime R_E versus ST_max resulted in R²_adj values of 39.3% and 99.3% based on Equations (6) and (11), respectively (Figures 10 and 11).

Soil water content was also determined to play a significant role in controlling daytime R_E from the Yenicaga peatland. As a function of SWC, quadratic regression models elucidated 39.2% and 68.7% of variation in daytime R_E according to Equations (6) and (11), respectively (Figures 12 and 13). A significant relationship between estimated GPP and PAR values was found, yielding R²_adj values of 80.3% and 55% from Equations (6) and (11), respectively (Figures 14 and 15). The rate of NEE in the Yenicaga peatland for the study period was on average estimated at 0.11 mg CO₂ m⁻² s⁻¹ (26.5 kg C ha⁻¹ day⁻¹) and ranged from −1.26 mg CO₂ m⁻² s⁻¹ (−297 kg C ha⁻¹ day⁻¹) to 1.06 mg CO₂ m⁻² s⁻¹ (249 kg C ha⁻¹ day⁻¹). This case points to the domination of total (positive daytime ± nighttime fluxes) R_E over GPP (negative flux) throughout most of the study period, especially, towards the end of the period during which GPP decreased at a faster rate than R_E (Figures 8 and 9). During the peak growing season in July, GPP peaked and decreased as the season progressed. Over the study period, GPP was on average −0.74 ± 0.43 mg CO₂ m⁻² s⁻¹ (−175 ± 102 kg C ha⁻¹ day⁻¹) and −0.35 ± 0.18 mg CO₂ m⁻² s⁻¹ (−2 ± 41 kg C ha⁻¹ day⁻¹) according to Equations (6) and (11), respectively. On average, nighttime R_E was estimated at 0.23 ± 0.09 mg CO₂ m⁻² s⁻¹ (53 ± 22 kg C ha⁻¹ day⁻¹) based on the EC sensors, while daytime R_E values were 0.64 ± 0.31 mg CO₂ m⁻² s⁻¹ (150 ± 74 kg C ha⁻¹ day⁻¹) and 0.24 ± 0.05 mg CO₂ m⁻² s⁻¹ (57 ± 11 kg C ha⁻¹ day⁻¹) based on Equations (6) and (11), respectively.

Peatlands (bog hummock, bog hallow, and poor fen) in a cool temperate region of eastern Canada were reported to have NEE range from −0.18 to 0.13 mg CO₂ m⁻² s⁻¹; nighttime R_E range from 0.07 to 0.36 mg CO₂ m⁻² s⁻¹; and GPP range from −0.20 to −0.33 mg CO₂ m⁻² s⁻¹ [15,16]. The reported ranges of NEE, R_E and GPP values are in a close agreement with our findings for the Yenicaga peatland [NEE = 0.11 mg CO₂ m⁻² s⁻¹ based on the EC data; nighttime R_E = 0.23 mg CO₂ m⁻² s⁻¹ based on the EC data; daytime R_E = 0.24 mg CO₂ m⁻² s⁻¹ based on Equation (11); daytime R_E = 0.64 mg CO₂ m⁻² s⁻¹ based on Equation (6); GPP = −0.35 mg CO₂ m⁻² s⁻¹ based on Equation (11); and GPP = −0.74 mg CO₂ m⁻² s⁻¹ based on Equation (6)], in particular, based on Equation (11). Similarly, EC measurements from 12 wetlands ranging from ombrotrophic and minerotrophic peatlands to wet tundra ecosystems across Europe and North America under temperate-to-arctic climate regimes showed that CO₂ fluxes in July varied considerably between −3 and 1 g C m⁻² day⁻¹ for NEE; 1 and 4 g C m⁻² day⁻¹ for R_E; and −1 and −6 g C m⁻² day⁻¹ for GPP [17], which were close to our results.