Carbon exchange of a mature, naturally regenerated pine forest in north Florida

Study sites

The study was conducted from July 2000 to June 2002 and January 2004 to December 2005 in a 41 ha pine flatwoods ecosystem in the Austin Cary Memorial Forest (ACMF). The ACMF is managed by the School of Forest Resources and Conservation of the University of Florida and located 15 km northeast of Gainesville, Alachua County, Florida, USA (29°44′17″N, 82°13′8″W; elevation 50 m). Prior to state purchase in 1936, the site had been selectively harvested for timber and used for low-intensity cattle grazing. In 1936, the area was allowed to regenerate naturally from remnant seed trees. The resulting stand was thinned in 1991 by removing 27% of the basal area to open the canopy. The current management objective is to restore an uneven-aged, mixed slash and longleaf pine stand by encouraging natural regeneration through prescribed burns every 3–5 years.

A 30-m walkup scaffolding tower was erected in the forest for this study. The fetch from the tower was greater than 1 km in the north and south directions, and approximately 0.5 km in the east and west directions. The fetch encompassed two similar administrative compartments; before this study, the understory of one was burned in the winter of 1997 and the other in the winter of 1998. The overstory was composed of longleaf pine and slash pine (72% and 28% of tree basal area, respectively), with tree ages ranging from 20 to 80 years in 2001. The top of the tree canopy averaged 22 m. The tree canopy was vertically separated by 15 m over a 1.5-m-tall, dense understory. In 2000, stand basal area was 18.6 m² ha⁻¹, and stem density was 363 trees ha⁻¹. The understory consisted of native species dominated by saw palmetto [Serenoa repens (Bartr.) Small], gallberry [Ilex glabra (L.) Gray], wax myrtle (Myrica cerifera, L.), and wiregrass (Aristrida stricta Michx). The soils were poorly drained ultic alaquods (sandy, siliceous, thermic) with a discontinuous spodic horizon 30–60 cm deep and argillic horizon approximately 1.25 m deep.

Measurements covering the same periods were also made at the Donaldson Tract (DT) pine plantation located 5.75 km to the northeast of the ACMF. Slash pine trees were planted to harvest density (∼2000 trees ha⁻¹) following a clearcut in 1990. Dominant canopy height ranged from 11.4 to 14.9 m during the study. The sonic anemometer height was raised from 15 to 21 m over the course of the study. The fetch was >800 m in all directions from the tower. Soil properties at this site were similar to the ACMF, but the water table was generally deeper. Further details about this site, experimental methods and previous flux measurements are reported in Gholz & Clark (2002) and Clark et al. (2004).

Meteorological and soil measurements

Photosynthetically active photon flux density (LI-190, LI-COR Inc., Lincoln, NE, USA), net radiation (Q7, Radiation and Energy Balance Systems Inc., Seattle, WA, USA), wind speed and direction (No. 3001-5, R. M. Young Company, Traverse City, MI, USA), temperature and relative humidity (HMP 23 UT, Vaisala Inc., Helsinki, Finland), and precipitation (tipping bucket, Sierra Misco Inc., Berkeley, CA, USA) were measured continuously from the top of each site's tower. At each site, three soil heat flux plates (HFT-3.1, Radiation and Energy Balance Systems Inc.) were buried 10 cm below the soil surface in separate locations within 8 m of the tower. All sensors were measured every minute and 30-min averages were collected on an EasyLogger EL824-GP data logger (Omnidata International, Ogden, UT, USA). Water table depth was measured adjacent to each site's tower using a Stevens water depth gauge (F-68, Leupold and Stevens, Beaverton, OR, USA).

Ecological measurements

The following description of the biomass sampling methods applies to the ACMF. DT biomass was sampled using similar procedures that are described in Gholz & Clark (2002) and Clark et al. (2004). Four inventory plots (50 m × 50 m) were established in random directions and distances between 50 and 200 m around the tower (directions constrained to place one plot in each compass quadrant). Tree density, height, and stem diameter at breast height (DBH; 1.37 m height) were measured in each plot in February 2000, 2001, and 2002. Allometric equations were used to determine standing wood, bark, and foliage biomass, and annual increment for each tissue (Taras & Clark, 1977; Taras & Phillips, 1978). Coarse root biomass was estimated as 13% of aboveground wood (plus bark) biomass (Gholz & Fisher, 1982). Litterfall was collected monthly from 10 1 m² traps located within the inventory plots, and then dried for 72 h at 70 °C and separated into needles or other material. Needle accretion in the canopy for each year was estimated by applying needlefall from the following year to a normalized logistic accretion curve, assuming 18 months for needle turnover (Gholz et al., 1991; Dougherty et al., 1995; Jokela & Martin, 2000; Martin & Jokela, 2004). LAI was then calculated from needlefall and needle accretion estimates.

Ten randomly distributed subplots situated in each of the four inventory plots were established, and then measurements of saw palmetto and gallberry – the two dominant understory species – were applied to allometric equations to determine their biomass (Gholz et al., 1999; Powell et al., 2005). Biomass of grass and herbs was estimated from nine 1 m² clip plots located around the tower at the time of peak biomass in 2000 and 2002. Forest floor mass was sampled in the winter of 2002 from five 100 cm² subplots randomly located within each inventory plot. Samples were dried for 72 h at 70 °C and then weighed.

Needle litter decomposition was estimated using a simulation model for slash pine forest floor dynamics (Gholz et al., 1985), adjusted to reflect dynamics in longleaf pine forest. The model consisted of three coupled ordinary differential equations representing fresh needle litter, a labile litter fraction, and a lignin and cutin (refractory) fraction. The measured rate of fresh needle decomposition (0.15 year⁻¹) in slash pine plantations does not accurately represent decomposition in longleaf pine-dominated ecosystems; Hendricks et al. (2002) reported a decomposition rate for longleaf pine needles on the forest floor surface which was 64% of the slash pine plantation estimate. Therefore, the parameters for fresh litter decomposition and transition rate from fresh to labile compartments were reduced to 0.096 and 0.319 year⁻¹ to reflect this difference. To estimate the relative sizes of the three litter pools, the model was simulated from a starting point of no litter mass (postfire conditions). The spatial heterogeneity of the standing litter mass required separate simulation of each plot. Measured litterfall was used as annual inputs for the model.

Meteorological data

Environmental variables measured at the ACMF during the study period are shown in Fig. 1. The sum of precipitation for each measurement year is also reported in Table 5. Long-term (1971–2000) mean annual precipitation in the vicinity was 1228 mm, and mean maximum and minimum temperatures for the months of January and July were 19.0 and 5.8 °C, and 32.7 and 21.6 °C, respectively (Fig. 1a; NCDC, 2002). During the study, the general climatic pattern was dry mild winters (November–March), warm dry springs, (April and May), and warm humid summers (June–October). However, there was a great deal of variation within this pattern. Moreover, three major environmental anomalies occurred during this study. The first was a severe ‘100-year drought’ that was in progress at the commencement of the study and was not alleviated until the summer of 2002. A water table depth below 2 m at both the ACMF and DT was common from June 2000 to July 2002 and reflected the severity of the drought (Fig. 1b). The water table was generally near or at the surface for both ecosystems during 2004 and 2005 when precipitation was higher. The second anomaly was the winter between 2000 and 2001, when 32 days had minimum temperatures below 0 °C – 16 more freeze days than the long-term average (NCDC, 2002). The third anomaly was that three tropical storms hit the sites in August and September 2004. Heavy rain from these storms left both sites inundated in September and October, and high winds impacted tree canopy leaf area.

Ecological measurements

In January 2001, aboveground standing biomass was estimated as 5911 g C m⁻² and belowground biomass as 721 g C m⁻² (Table 1). In December 2002, total forest floor biomass was estimated as 1452 g C m⁻². Annual increment for trees, understory, litterfall, and forest floor was 136 and 77 g C m⁻² yr⁻¹ for 2000 and 2001, respectively (Table 1).

During the 2 years with average precipitation (i.e. 2004 and 2005), one large pulse of needlefall occurred in the fall (Fig. 1e). In contrast, the drought caused the canopy to adjust by prematurely dropping 50% of annual needlefall in May and June of 2000 to 2002 (Fig. 1e). The unusually high pulse of needlefall in September 2004 was caused by high winds from the tropical storms. LAI was seasonal and the effects of the premature needle fall and high winds are reflected in the maximum summertime values in 2000, 2001, and 2005 (Fig. 1f). In comparison, DT LAI was approximately 30% greater than ACMF LAI during the drought years and very similar in 2004 before the wind damage (Fig. 1f). The drought appeared to have less of an effect on summertime LAI in the DT, while the tropical force winds also reduced LAI.

Measurement error

Similar to other estimates of eddy covariance C flux measurement error (Richardson et al., 2006, 2008), the distribution of NEE random measurement error (δ) in this study was better described by a double-exponential distribution than by a normal distribution (Fig. 2). The distributions of δ were similar between ACMF and DT, with approximately equivalent medians, standard deviations, and skewness between the sites (Table 2). Kurtosis for DT was higher than for ACMF, however, meaning DT δ clustered nearer the median and had slightly lower density in the tails than did ACMF (Fig. 2). Random error tended to be lower at night than during the day, especially at DT, and error was highest at higher radiation levels at both sites (Table 2). Growing season (March–October) error was slightly higher than during the winter (November–February) for both sites.

Net CO₂ exchange

The ACMF was photosynthetically active during all parts of the year (Fig. 3). Light response curve parameters for selected months of 2001 and 2005, two hydrologically contrasting years, are given in Table 3. The drought and the seasonality of LAI and soil moisture cause little change in the magnitude of maximum NEE_day (i.e. the dotted lines that mark the lower boundary of points in Fig. 3) during the growing season (March–October). For example, maximum NEE_day reached approximately −16 to −18 μmol CO₂ m⁻² s⁻¹ from May to October when soil moisture and LAI were highly variable in both 2001 and 2005.

The wide scatter of data points within the light response curves indicated that there were important secondary controls regulating NEE_day , particularly during more droughty months, such as May and July 2001 – two months with relatively low R² values (Table 3). Therefore, the midday (11:00–15:00 hours) residuals of each monthly light response regression (NEE_res) were plotted against VPD and T_a to elucidate when these two variables became important constraints on NEE_day (Fig. 4). These relationships were not affected by differences in soil moisture or season (data not shown) and therefore the data were pooled over each year. The ACMF reached optimum C uptake near VPD=1.0 kPa, after which C uptake declined linearly as VPD increased (Fig. 4a). Interestingly, the function of this relationship was not altered by drought. Air temperature below 10 °C severely reduced net C uptake (Fig. 4b), whereas the temperature range of 10–30 °C had a little effect on NEE_day and above 30 °C caused net C uptake to decline somewhat.

At ACMF, WUE declined linearly with increasing VPD, and was approximately constant at VPD>1.5 kPa; this relationship was similar in both droughty 2001 and wet 2005 (Fig. 5a). In contrast, the relationship between WUE and VPD at DT varied between wet and dry years (Fig. 5b). In 2005, the relationship at DT was similar to that at ACMF, but in 2001, the relationship was more complex, showing an increase in WUE at VPD above 2.5 kPa.

Mean nightly (22:00–4:00 hours) C exchange, NEE_night, was a significant function of mean night-time T_a during well-coupled () conditions (Table 4). Although SWC was highly variable over the year in this ecosystem, the addition of a SWC function to Eqn (4) was not statistically significant. However, separating the data into two different SWC groups (<5.5% and >5.5%) produced two significant functions, the former showing a 33% reduction in respiration rates at 25 °C (Table 4, Fig. 6). It should be noted that values for b (Table 4), which indicate how sensitive respiration was to changes in temperature, were significantly different (P=0.001), while a values were not significantly different (P=0.557) for these two conditions (determined statistically with an indicator variable; Powell et al., 2005). When SWC was >5.5%, mean NEE_night was 6.8 μmol CO₂ m⁻² s⁻¹ at 25 °C, while under dry conditions, (SWC<5.5%) it was 4.6 μmol CO₂ m⁻² s⁻¹ at 25 °C. Q₁₀ for moist soil conditions was 2.0; while under dry conditions, Q₁₀ was reduced to 1.4.

An ordinary least squares regression analysis showed that over all times of the day, understory net ecosystem C exchange, NEE_u, was linearly related to above-canopy NEE and accounted for approximately 37% of total ecosystem C fluxes (Fig. 7). This relationship was insensitive to changes in above-canopy VPD and soil moisture. It should be noted that random error associated with measurements of the independent variable, NEE, violates error assumptions of an ordinary least squares regression. Therefore, we also calculated the half-hourly ratio of NEE_u to NEE and then examined the distribution of this ratio for 24-h, midday, and night-time periods. Median values of this ratio were 0.28 for 24-h period, 0.27 for midday, and 0.85 for night. Season did not have any apparent effect on these ratios. The frequency distribution of NEE_u : NEE during both the night and midday was not normally distributed (Fig. 7b and c; Shapiro–Wilke, P<0.05). The night-time frequency distribution was broadly centered around the median value with 66% falling within the range of 0.65–1.08 (skewness of 0.66 and kurtosis of 2.0). In contrast, the midday frequency distribution was strongly centered around the median value with 66% falling within the range of 0.22–0.31 (skewness of 0.43 and kurtosis of 28.8).

ACMF NEP was remarkably similar (range: 154–192 g C m⁻² yr⁻¹) among the four measurement years of this study regardless of the considerable differences in precipitation (Table 5). The seasonality of C assimilation in the ACMF was also similar among the 4 years, with the greatest monthly C sequestration occurring in the early spring and then being nearly C neutral from late summer through the winter (Fig. 8a). April almost universally experienced the highest NEP each year (Fig. 8b). In comparison, NEP for DT was three to four times greater than that of the ACMF for each simultaneous measurement year (Table 5). DT seasonal NEP patterns also contrasted ACMF in that DT generally experienced high rates of C uptake during all parts of the year (Fig. 8a).

The gap-filling analysis indicated that there was no major error or bias introduced by the distribution of gaps in our dataset. The annual NEP sum for the original, complete 2005 dataset was 154 g C m⁻² yr⁻¹. By comparison, NEP sums for the 50 gap-filling scenarios ranged from 135 to 170 g C m⁻² yr⁻¹, with a mean sum of 154 g C m⁻² yr⁻¹and standard deviation of 7 g C m⁻² yr⁻¹. The distribution of NEP sums was normal (Shapiro–Wilke, P=0.6818), with skewness of 0.25 and kurtosis of 0.09.

Annual R_e and GEP are given for both forests in Table 5. The tropical storms in 2004 caused a reduction in R_e and GEP in both ecosystems. Interestingly, the compensatory reduction of these fluxes resulted in very little difference in NEP with respect to the other years. In the ACMF, R_e and GEP had an asynchronous seasonal pattern where GEP generally outpaced R_e in the spring and early summer (Fig. 8c), which helps to explain why C gain was greatest during this period. The asynchronous pattern was also present in DT, but it was considerably less pronounced (Fig. 8c).

Error analyses

Recent studies have reported the distribution of random flux errors at diverse flux sites (Hollinger & Richardson, 2005; Richardson et al., 2006, 2008). The σ(δ) reported for forested sites in these studies ranged from 2.4 to 4.3; both ACMF and Donaldson Tract σ(δ) was 2.6. Most sites also report higher σ(δ) under high radiation conditions compared with that under low radiation, and during the growing season vs. winter, as we also observed in our data (Table 2). These patterns are consistent with a scaling of random error with flux magnitude, with δ being lowest near fluxes of 0.0, and increasing as fluxes become more negative or more positive (Richardson et al., 2006).

Equations (2) and (4) have been demonstrated to perform well as gap-filling models for a wide range of forests (Falge et al., 2001; Moffat et al., 2007). In our case, the gap analysis for 2005 data established uncertainty boundaries within 12% for annual NEP and indicated that no major bias was introduced by the distribution of gaps. Therefore, we assumed that estimates of annual NEP for the other measurement years likely have a gap-filling error similar in magnitude.

Net CO₂ exchange

NEE_day was a strong curvilinear function of PPFD throughout each year due to the relatively mild winters and year-round physiological activity of pines in northern Florida (Martin, 2000; McGarvey et al., 2004). The magnitude of maximum summertime NEE_day of the ACMF (−16 to −18 μmol CO₂ m⁻² s⁻¹, dotted lines in Fig. 3) was ∼10 μmol CO₂ m⁻² s⁻¹ lower than the adjacent DT plantation (Clark et al., 2004) and ∼5 μmol CO₂ m⁻² s⁻¹ lower than an adjacent rotation-aged plantation (Clark et al., 1999). The less negative NEE_day of the older ACMF stand likely was a result of higher maintenance respiration rates relative to net canopy assimilation and lower ecosystem LAI and stocking density relative to the plantations (Clark et al., 1999). However, ACMF maximum NEE_day was similar to the global value (ca. −18.0 μmol CO₂ m⁻² s⁻¹) reported for a wide range of temperate coniferous forests (Ruimy et al., 1995; Falge et al., 2002).

On a half-hourly time scale, PPFD was the dominant control over NEE_day explaining 30–80% of its variation. However, there were important secondary controls when both ACMF and DT were under water stress. After VPD exceeded 2.0 kPa, the magnitude of NEE_day in the ACMF decreased by as much as 5 μmol CO₂ m⁻² s⁻¹. We have previously observed stomatal closure at VPD of 1.5–2.0 kPa across a range of soil water contents which is consistent with this pattern (Powell et al., 2005). A similar type of analysis conducted for DT also showed a decrease in mean NEE_day of 5 μmol CO₂ m⁻² s⁻¹ at PPFD of 1500 μmol m⁻² s⁻¹ when VPD exceeded 2.0 kPa for both drought and nondrought conditions (Clark et al., 2004). These results agree with findings of an old-growth Ponderosa pine forest in Oregon, which also experiences periods of extreme water stress and high VPD, where NEE_day became increasingly suppressed with increasing VPD (Anthoni et al., 2002).

There was an important contrast in how the ACMF and plantations responded to the severe drought. Although the relative reduction in NEE_day (i.e. 5 μmol CO₂ m⁻² s⁻¹) was the same for both ecosystems during periods of high VPD, there was also a reduction in maximum NEE_day during drought compared with nondrought periods at DT (e.g. Fig. 2 in Clark et al., 2004). Maximum NEE_day at the ACMF, however, was similar between drought periods in 2001 compared with 2005 (e.g. dotted lines for My01 and Jy01 compared with My05 and Jy05 in Fig. 3); these contrasting responses were also reflected in differing patterns of WUE response to VPD in the two stands in dry and wet years (Fig. 5).

Residuals plotted against air temperature showed a broad optimum between 10 and 30 °C. This response can be explained by integrating the temperature responses of photosynthesis and ecosystem respiration. At the leaf level, Teskey et al. (1994) showed that slash pine net photosynthesis responded linearly and positively to temperature across a range of air temperature from 10 to 35 °C. We see similar responses in GPP, the sum of ecosystem photosynthesis. At a half-hourly time scale in this study, GPP declined linearly with air temperature (GPP=0.72–0.44T_a, n=18 186, R²=0.23, where GPP is the gross primary production in μmol CO₂ m⁻² s⁻¹ and T_a is the air temperature in  °C). When this linear response is combined with the weakly nonlinear response of NEE_night to temperature over the range of 10–25 °C (Fig. 5b), the result is a broad temperature optimum for NEE_res.

In coniferous forests, global values for F_sat and α are −32.4 μmol CO₂ m⁻² s⁻¹ and −0.024 μmol CO₂ μmol photon⁻¹, respectively (Ruimy et al., 1995). Although seasonality was weak in this ecosystem, F_sat was greatest in magnitude during growing season months when precipitation was not limiting (i.e. October 2001 and May–October 2005; Table 3) and similar to values reported for the adjacent rotation-aged plantation (−26.5 mol CO₂ m⁻² s⁻¹; Clark et al., 1999). The variation in ACMF's quantum efficiency (−0.021 to −0.047) was less than the range reported for adjacent pine plantations [−0.025 to −0.075 (excluding clearcut values); Clark et al., 2004] and similar to that reported for a mixed-temperate maritime forest (−0.029 to −0.044; Aubinet et al., 2001) and a Scots pine forest (−0.011 to −0.047; Zha et al., 2004).

There was a statistical difference in the sensitivity of NEE_night to T_a when SWC of 5.5% was used as a threshold for dividing between wet and dry conditions (Table 4, Fig. 5). However, there is probably not any biological significance of this SWC value alone. This relationship more likely operates along a nonlinear soil moisture continuum as has been found in other studies (Reichstein et al., 2002). Unfortunately, our soil moisture sensor was damaged for 2004 and 2005, and thus our relatively small number of data points precluded us from further partitioning the data to better define this relationship for the ACMF. Under wet soil conditions, summertime NEE_night at this site (6.8 μmol CO₂ m⁻² s⁻¹ at 25 °C) was similar to DT and rotation-aged (24-year-old) pine plantation NEE_night (range: 6.4–6.7 μmol CO₂ m⁻² s⁻¹ at 25 °C; Clark et al., 1999, 2004). Similarly, under dry soil conditions (SWC<5.5% in the present study), summertime NEE_night (4.6 μmol CO₂ m⁻² s⁻¹) was similar to that of DT (Clark et al., 2004).

Understory C fluxes

Night-time measurements of understory fluxes have been found to be less reliable than daytime fluxes in open canopy forests due to a build up of a strong inversion layer (Misson et al., 2007). To overcome this problem, understory data were screened using the same threshold as the canopy (0.2 m s⁻¹) to ensure turbulent conditions. Nevertheless, the broader frequency distribution of night-time NEE_u : NEE and large number of values greater than 1 (Fig. 7b) was evidence that the canopy and understory were not always well coupled even after the filter was applied.

During the summertime, midday NEE_u at the ACMF accounted for 27% of NEE, which was on the higher end of values reported by Misson et al. (2007) for a broad range of evergreen and deciduous forests (range: −54% to 36%, negative indicates that respiration dominates). During the summertime, night-time NEE_u comprised 85% of NEE, which was also on the upper end of values reported for other pine forests (range: 35–65%; Misson et al., 2007). The proportion of 24-h mean half-hour NEE attributed to understory fluxes in the current study (∼25–35%; Fig. 6) was a bit higher than the estimate of 21% for a 23-year-old pine flatwoods plantation simulated by Golkin & Ewel (1984). Periodic ceptometer measurements indicated that 30–60% of incident PPFD penetrated the canopy of this stand (data not shown), suggesting that understory CO₂ assimilation capacity was relatively high. In contrast, only 18–42% of incident PPFD penetrates the canopy in adjacent pine plantations (Gholz et al., 1991), indicating that the CO₂ assimilation capacity of plantation understories is lower. Our results were similar to those of a boreal pine forest with a similar understory radiation environment, which accounted for 20–30% of NEE (Baldocchi & Vogel, 1996). However, our results do not support the hypothesis that NEE_u would become a greater part of NEE as stresses imposed by VPD and SWC increased, as has been found in other studies of coniferous forests (e.g. Black & Kelliher, 1989). Rather, the relationship between NEE and NEE_u was insensitive to changes in above-canopy VPD and soil moisture. This may be the result of the more open canopy resulting in the understory being subjected to similar stresses as the canopy trees.

Annual ecosystem C balance and partitioning

Annual NEP based on eddy covariance was greater than mensurational measurements in our study by 56 and 82 g C m⁻² yr⁻¹ for 2000–2001 and 2001–2002, respectively. The differences may be accounted for by unmeasured C translocation to coarse root carbohydrate storage pools. In addition to eddy covariance measurement errors, assumptions associated with the biomass measurements should be considered. First, our assumption that fine roots and soil organic matter were in steady state may be invalid. Second, although the years overlapped for the mensurational and eddy covariance measurements, they were not the same exact intervals. Third, designating June as the anniversary for eddy covariance during the first 2 years separated two halves of a growing cycle and may be problematic because labile C storage and use can differ between years (Sampson et al., 2001).

Spring to early summer appears to be the most intense period of C accumulation across a range of ecosystems within the Florida landscape [i.e. natural and intensively managed pine forests, cypress wetlands and scrub oak (Fig. 7a; Clark et al., 1999, 2004; Powell et al., 2006)], keeping in mind that maximum C gain may be decreased or shift to later in the summer in extremely dry springs. Nevertheless, this suggests that the asynchronous patterns of R_e and GEP are similar across species and management practices in this region. A major contrast between the natural and managed stands was that the ACMF was essentially C neutral from late summer until the following spring (Fig. 8a), while the plantations were strong C sinks throughout the year (8, 6 in Clark et al., 2004).

Global climate change is expected to manifest in this region with increased tropical storm activity and more dramatic swings between very dry and very wet conditions (Twilley et al., 2001). Until this study, it was poorly understood how mature, less intensively managed pine flatwoods would respond to future weather scenarios. Our results demonstrate that both photosynthesis and respiration will be affected in different ways. For example, water stress caused both premature needlefall (Fig. 1e) and stomatal closure (Fig. 5), which resulted in reduced canopy photosynthesis (i.e. F_sat in Table 3). At the same time, soil respiration was reduced by dry soil conditions and presumably reduced substrate as a result of lowered photosynthesis (Hogberg et al., 2001; Bhupinderpal-Singh et al., 2003; Johnsen et al., 2007). Or in terms of the hurricane, GEP was reduced by needle loss, while R_e was thought to be reduced by anoxic conditions in the inundated soils coupled with a reduction in substrate due to needle loss. The compensatory effect of both reduced GEP and R_e in the face of either drought or tropical storms results in NEP values that are very similar to ‘average’ years. Our results are supported by Clark et al. (2004) and Li et al. (2007) who found similar compensatory effects of GEP and R_e on NEP for a pine plantation exposed to drought and a Florida scrub-oak exposed to a hurricane, respectively.

The ACMF sequestered ∼700 g C m⁻² over the four measurement years of the study. In comparison, the highly productive DT pine plantation sequestered ∼2500 g C m⁻² over the same 4-year period. Additionally, a nearby rotation-aged slash (24-year-old) pine plantation and a loblolly pine plantation in the North Carolina piedmont both had annual NEP estimates of the same order of magnitude as DT annual NEP (Clark et al., 1999, 2004; Juang et al., 2006). The lower R_e to GEP ratio for the plantations compared with the ACMF (Table 5) suggests that the higher GEP in the plantations was driven by higher LAI and a smaller fraction of photosynthate allocated to maintenance respiration. After the rotation-aged plantation was harvested, a large amount of C was released (e.g. −1269 and −882 g C m⁻² yr⁻¹) over the subsequent 2 years (Clark et al., 2004). These large differences highlight the importance of stand age and management regime for stand-level C balance in this region.

Implications of these management regimes on C sequestration in the long term can be assessed using simple calculations of cumulative NEP for the ACMF stand and for a generic slash pine plantation in the region over a 24-year period, which is a typical plantation rotation length. We used plantation calculations of NEP for regenerating, mid-rotation, and rotation-aged stands published by Clark et al. (1999, 2004) and Binford et al. (2006) that started after the harvest of a previous stand. Using the average NEP for the ACMF over the 4 years yields 4140 g C m⁻² after 24 years vs. 9900 g C m⁻² for the plantation. This calculation leaves two important uncertainties unaddressed. Timber harvesting removes most of the accumulated C from the plantations at the end of each rotation, and the fate of harvested C must be tracked with life cycle analysis (Skog & Nicholson, 1998; Ross & Evans, 2004). For the ACMF stand, periodic prescribed fire will return some proportion of understory and forest floor C to the atmosphere at each fire cycle; research on this topic is ongoing at this site.

The dynamics of landscape-level C sequestration is a function of interactions among environmental conditions, site-specific edaphic characteristics, and land management (e.g. evergreen pine vs. deciduous cypress, age or time since disturbance, soil type and topography, tree age demographics, season and year; Gholz et al., 1991; Liu, 1998; Binford et al., 2006). Intensively managed plantations change from a net C source to a net C sink as early as 3 years following harvest (Gholz & Fisher, 1982; Thornton et al., 2002; Clark et al., 2004), and rapidly develop high rates of C sequestration (Clark et al., 2004) that are offset by periodic timber harvesting. These large changes do not occur in older, less intensively managed pine stands, such as the ACMF, because canopy cover, stand structure, and biomass are largely maintained over time. Therefore, arguably one of the largest uncertainties in forecasting regional C budgets lies with our ability to predict how present management practices might change with future ownership. Finally, this study demonstrated that prolonged severe drought had little effect on NEP, which highlights that the effects of projected changes in climate on coastal plain pine forest NEP are likely to be small relative to those of land use and management change.