The study site is located within the Oak Openings Region of northwest Ohio, USA (N41°33′16.98″, W83°50′36.76″) and comprises a mosaic of oak (Quercus spp.) forests, maple (Acer spp.) floodplains, and remnants of oak savanna, barrens and prairie (Brewer & Vankat, 2004). The region has a perched water table at the depth of 2 m, developed on a series of sandy glacial beach ridges and swales over fine textured till. The 30 yr mean annual temperature is 9.2°C and precipitation is 840 mm. Our study area is a 100 ha woodland within the 1500 ha Metroparks of Toledo Area Oak Openings Preserve. The topography is flat with an elevation range of 200–205 m. Eighty per cent of the area is upland forest, dominated by oaks on a sandy mixed, mesic, Spodic Udipsamments. The remaining 20% of the forest is a lowland area near a stream where red maple (Acer rubrum) dominates on a sandy, mixed, mesic Typic Endoaquolls. Before European settlement, this area supported oak savanna in the uplands and wet prairie in the lowland areas. After settlement, portions of this area were grazed and farmed, but there is no evidence of widespread tree harvesting. At present, the main canopy consists primarily of 45-yr-old oaks, and the understory of 15-yr-old red maples.
Plant species composition and stand biometric properties were measured on 12 FIA-style plots within the 100 ha area using the USDA Forest Service Forest Inventory and Analysis (FIA) plot design (http://www.fia.fs.fed.us/library/). Each plot consisted of four circular subplots, 14 m in diameter and 36.5 m apart in a triangle formation. All trees with diameter at breast height (dbh; 1.5 m above ground) > 3 cm were tagged, identified to species and measured annually. The predominant species, by biomass, were Quercus rubra (32%), Q. alba (27%), Q. velutina (14%) and Acer rubra (20%). The remaining 7% were Prunus serotina (black cherry) and Sassafras albidum (sassafras). The woody ground cover was primarily Vaccinium spp. with very few tree seedlings. Standing biomass estimates were calculated from allometric relations established for the Great Lakes Region (Perala & Alban, 1994; Ter-Mikaelian & Korzukhin, 1997). Total biomass at the site was 20 345 ± 2180 (mean ± SD) g C m−2, of which 9600 ± 650 g C m−2 was in above-ground biomass, 1360 ± 1280 g C m−2 in below-ground biomass, 1140 ± 655 g C m−2 in litter, and 8225 ± 2405 g C m−2 in soil. Leaf area index (LAI) was estimated with LAI-2000 plant canopy analyzer (Li-Cor Inc., Lincoln, NE, USA) in the vegetation plots. Measurements were made monthly in 2004, and once during peak canopy cover, in July, in 2005. In July, LAI was 4.6 ± 0.52 m2 m−2 (mean ± SD) in 2004, and 4.0 ± 0.49 m2 m−2 in 2005. In 2004, when additional measurements were available, peak LAI was observed in August, reaching 5.0 ± 0.39 m2 m−2. The cumulative leaf litterfall was 388 ± 19 g DW m−2 in 2004 and 333 ± 35 g DW m−2 in 2005, lending support to the observed difference in optical LAI estimates between the years. Phenological stages were defined by bud-break (start of pre-growth phase), 95% full leaf expansion (start of growth phase), start of leaf discoloration (start of pre-dormancy phase) and 95% leaf fall (start of dormancy phase), following DeForest et al. (2006).
Micrometeorology and turbulent fluxes
The following micrometeorological parameters and turbulent exchange of carbon, water and energy were measured from a 32 m instrument tower, located in the center of the stand. There was homogeneous fetch for a distance of at least 600 m in all directions. We measured 30-min means of air temperature (Ta (°C); HMP45AC, Vaisala, Finland) and relative humidity (RH (%); HMP45AC) above and below canopy, soil temperature at 5 cm (Ts (°C); CS107, Campbell Scientific Inc. (CSI), Logan, UT, USA), soil water content (SWC (%); CS616, CSI) in the top 20 cm, photosynthetically active radiation above and below canopy (PAR (µmol quanta m−2 s−1); LI-190SB, Li-Cor), and precipitation (P (mm); TE525, Texas Electronics, Dallas, TX, USA).
Turbulent exchange of CO2 (Fc) was measured using the eddy-covariance method (Lee et al., 2004) 7 m above the forest canopy (canopy height 24 m) throughout 2004 and 2005. The eddy-covariance system consisted of a LI-7500 open-path infrared gas analyzer (IRGA, Li-Cor), a CSAT3 3-dimensional sonic anemometer (CSI), and a CR5000 data logger (CSI). The LI-7500 was calibrated every 3–4 months in the laboratory using chemically scrubbed nitrogen (in 2004) or zero-grade nitrogen (in 2005) for zeroing the CO2 and H2O, and a dew-point generator (LI-610, Li-Cor) and NOAA/CMDL-traceable primary CO2 standards for setting the span. The 30-min mean flux of CO2 was computed as the covariance of vertical wind speed and the concentration of CO2, after removing spikes in raw data (> 6 STD), correcting sonic temperatures for humidity and pressure (Schotanus et al., 1983), and rotating wind coordinates to the mean streamline plane (Wilczak et al., 2001), using the EC_Processor software package (http://research.eeescience.utoledo.edu/lees/ECP/ECP.html). The 30-min mean fluxes were corrected for fluctuations in air density using the Webb–Pearman–Leuning expression (Webb et al., 1980; Paw U et al., 2000; Massman & Lee, 2002), including the term for the warming of IRGA above air temperature (Burba et al., 2006; Grelle & Burba, 2007). This temperature difference leads to additional sensible heat flux that is proportional to ambient temperature and wind speed. The effect of Burba's correction on annual NEE, ER and GEP was +68, +61 and −7 g C m−2, differing by 1–4 g between the two years. In both years, the correction affected ER more than GEP.
Change in the canopy air space CO2 storage was estimated from the change in the mean CO2 concentration by sampling air from four different heights (1.5, 5, 16 and 22 m above ground). Air from all four inlets was mixed in a 5 l PVC chamber before sampling with a temperature-controlled LI-800 CO2 analyzer (Li-Cor). The flow of air (1 l min−1 total, 0.25 l min−1 per inlet) was regulated with a flowmeter (model 4112K35, McMaster-Carr Supply Company, Atlanta, GA, USA) and driven by a continuously operating microdiaphragm pump (model UNMP50KNDC BLDC, KNF Neuberger, Trenton, NJ, USA). The IRGA was connected after the flowmeter and before the pump to avoid excess pressure in the IRGA. The NEE is the sum of the corrected turbulent flux and the storage flux. We use the sign convention by which positive NEE indicates flux away from the surface. Daytime ER was estimated from night-time measurements, assuming consistency of temperature sensitivity between night- and daytime exchange. GEP was calculated as the difference between NEE and ER (GEP = ER − NEE).
Quality checking and gap-filling
Eddy-covariance data were screened for quality, by flagging periods of highly stable and highly unstable atmosphere (Hollinger et al., 2004), nonstationary turbulent fluxes (Foken et al., 2004; Göckede et al., 2004), rain, dew or ice on sensors, low turbulence, out-of-range fluxes and power failures. The threshold of friction velocity (u*) below which flux loss occurred was determined from the seasonal binned relationship between turbulent flux of CO2 and friction velocity (u*) (Schmid et al., 2003). The threshold was consistent between different seasons, but differed slightly between years: 0.18 m s−1 in 2004 and 0.21 m s−1 in 2005. No directional effects were observed on fluxes during the growing season, but in winter the sectors with conifers closer than 1 km (at directions 150–160°, 180–190° and 210–270°) exhibited different exchange rates than other wind sectors and were excluded from developing temporal integrals (treated as gaps).
The data gaps were more frequent during the night than during the day. Overall, 54% of daytime and 76% of night-time fluxes had to be gap-filled in 2004, and 43 and 58% in 2005, which is higher than the network-wide average of 35% (Falge et al., 2001), but typical of sites with open-path IRGAs that are sensitive to precipitation and condensation. The gaps were filled with dynamic parameter mechanistic models as described by Noormets et al. (2007), simultaneously fitting both day- and nighttime data. The gap-filling models were chosen from among 32 evaluated model variants based on the magnitude and bias of residuals, and the stability of model parameter estimates. While the complete details of the model selection process go beyond the intended scope of this paper, the residuals of ER were smaller with dynamic temperature-sensitivity models than with Q10 models, and smaller with Ta than with Ts. Residuals of annual (fixed-parameter) models showed temporal bias, which was eliminated with a seasonally fit (i.e. dynamic) parameters. A comparison of our gap-filling algorithm against 16 other widely used approaches showed that the bias was about average for short-term gaps, and among the lowest of all models for medium and long-term gaps (Moffat et al., 2007). The uncertainty in annual NEE, caused by gaps and gap-filling, was estimated according to Aurela et al. (2002) and Flanagan & Johnson (2005). The structure of the gap-filling model was:
- (Eqn 1)
where α is apparent quantum yield (µmol CO2µmol−1 PAR), φ is PAR (µmol quanta m−2 s−1), Pmax (µmol CO2 m−2 s−1) is the maximum apparent photosynthetic capacity of the canopy (µmol CO2 m−2 s−1), and ER is the eddy covariance-based ecosystem respiration which, in turn, was expressed as a function of air temperature (Ta, K) using a version of the Lloyd & Taylor (1994) model:
- (Eqn 2)
where Ea is the activation energy (kJ mol−1 K−1), R is the universal gas constant (8.3134 J mol−1 K−1), and R10 is the reference respiration, normalized to a common temperature (Tref = 283.15 K = 10°C). R10 was estimated as a constant in spring, fall and winter, when respiration was insensitive to variations in soil moisture content (data not shown), or as a function of soil volumetric water content (SWC, %) during summer months when significant (P < 0.1) residual trends in ER were observed after accounting for temperature dependence. Thus, the model became:
- (Eqn 3)
where a0 is equal to R10 at moisture saturation, or in the absence of moisture sensitivity (a1 = 0). Parameter a1 indicates unit change in R10 per unit change in SWC. Including moisture sensitivity terms in the gap-filling model improved the fit only during the growing season, whereas during pre-growth, pre-dormancy and dormancy only the temperature-based model was used.
- (Eqn 4)
where P is atmospheric pressure (kPa), E is evapotranspiration (kg m−2 s−1), ρ is air density (kg m−3), VPD is vapor pressure deficit (kPa) and Rd and Rw are universal gas constants for dry air and water vapor, respectively.