Patterns of NPP, GPP, respiration, and NEP during boreal forest succession

We combined year‐round eddy covariance with biometry and biomass harvests along a chronosequence of boreal forest stands that were 1, 6, 15, 23, 40, ∼74, and ∼154 years old to understand how ecosystem production and carbon stocks change during recovery from stand‐replacing crown fire. Live biomass (Clive) was low in the 1‐ and 6‐year‐old stands, and increased following a logistic pattern to high levels in the 74‐ and 154‐year‐old stands. Carbon stocks in the forest floor (Cforest floor) and coarse woody debris (CCWD) were comparatively high in the 1‐year‐old stand, reduced in the 6‐ through 40‐year‐old stands, and highest in the 74‐ and 154‐year‐old stands. Total net primary production (TNPP) was reduced in the 1‐ and 6‐year‐old stands, highest in the 23‐ through 74‐year‐old stands and somewhat reduced in the 154‐year‐old stand. The NPP decline at the 154‐year‐old stand was related to increased autotrophic respiration rather than decreased gross primary production (GPP). Net ecosystem production (NEP), calculated by integrated eddy covariance, indicated the 1‐ and 6‐year‐old stands were losing carbon, the 15‐year‐old stand was gaining a small amount of carbon, the 23‐ and 74‐year‐old stands were gaining considerable carbon, and the 40‐ and 154‐year‐old stands were gaining modest amounts of carbon. The recovery from fire was rapid; a linear fit through the NEP observations at the 6‐ and 15‐year‐old stands indicated the transition from carbon source to sink occurred within 11–12 years. The NEP decline at the 154‐year‐old stand appears related to increased losses from Clive by tree mortality and possibly from Cforest floor by decomposition. Our findings support the idea that NPP, carbon production efficiency (NPP/GPP), NEP, and carbon storage efficiency (NEP/TNPP) all decrease in old boreal stands.


Introduction
The patterns and controls on plant succession have interested ecologists for more than a century (Clements, 1916), and extensive bodies of work have accumulated on both the biological controls on recovery (e.g. Connell & Slatyer, 1977) and the implications of recovery for ecosystem function (e.g. Odum, 1969;Bormann & Likens, 1979). More recently, research has focused on the effects of disturbance and recovery on carbon (e.g. Houghton et al., 1983) and energy (e.g. Randerson et al., 2006) exchange, with the goals of understanding the contemporary terrestrial carbon cycle and predicting the climate forcing caused by possible future changes in disturbance regime (e.g. Hurtt et al., 2002).
Several conceptual models have been proposed to describe the recovery of forest production and biomass from disturbance (Kira & Shidei, 1967;Odum, 1969;Bormann & Likens, 1979;Sprugel, 1985;Waring & Schlesinger, 1985;Ryan et al., 1997;Chapin et al., 2002). Sprugel (1985) provides an overview of the expected changes in carbon stocks during regrowth (Fig. 1a), which draws on the findings at and around Hubbard Brook (Vitousek & Reiners, 1975;Bormann & Likens, 1979;Covington, 1981). Disturbance reduces live biomass and increases the amount of carbon in the coarse woody debris (C CWD ) and the forest floor (C forest floor ). This detritus subsequently decomposes, resulting in declining detrital stocks, while plant production recovers, resulting in gradually increasing live biomass (C live ). The loss of detritus, especially from the forest floor, is thought to proceed more rapidly than the accumulation of live biomass, resulting in an initial loss of total carbon (C total ) and driving a net efflux of CO 2 to the atmosphere (a negative Net ecosystem production or NEP; see Fig. 1a and b). Over time, plant growth and biomass accumulation increase, and the input of new Fig. 1 Hypothesized trends in biomass and production during secondary succession. (a) Live biomass (C live ), forest floor (C forest floor ), coarse woody debris (C CWD ), and total live biomass (C total ; all digitized from Sprugel, 1985's Fig. 2). (b) Net ecosystem production (NEP; digitized from Chapin et al., 2002's Fig. 13.11). (c) Gross primary production (GPP) and net primary production (NPP; all digitized from Chapin et al., 2002's Fig. 13.9). (d) Autotrophic respiration (R auto ), heterotrophic respiration (R hetero ), and total respiration (R total ; all digitized or calculated from Chapin et al., 2002's Figs 13.9 and 13.11). (e) Plant production efficiency (NPP/GPP) and ecosystem carbon storage efficiency (NEP/NPP; calculated from Chapin et al., 2002's Figs 13.9 and 13.11). The time of disturbance is indicated by the dashed vertical line.
litter causes the detrital pools to build up, resulting in an extended period of carbon accumulation and a positive NEP. Finally, the buildup of C live , C forest floor , and C CWD cause mortality and decomposition losses to accelerate, which moves the pools toward steady state and causes NEP to approach zero in the oldest stands (Odum, 1969). Chapin et al. (2002) expanded this model to consider the successional changes in gross primary production (GPP; the annual carbon fixation by autotrophs), net primary production (NPP; the annual growth by autotrophs), autotrophic respiration (R auto ; the annual respiration by autotrophs), and heterotrophic respiration (R hetero ; the annual respiration by heterotrophs). Chapin and colleagues hypothesized NPP varies markedly during recovery, reaching a peak in middle age, and declining by as much as 50% in old stands (Fig. 1c). Chapin and colleagues further hypothesized the changes in NPP with age are driven exclusively by changes in GPP (Fig. 1c), and that the NPP to GPP ratio remains constant, with R auto tracking NPP and GPP ( Fig. 1d and e). The observation that forest NPP peaks in mid-succession is well established (Kira & Shidei, 1967;Ryan et al., 1997;Bond-Lamberty et al., 2004;Mack et al., 2008), whereas the hypothesis that the NPP to GPP ratio remains constant is more recent (Ryan et al., 1997;Waring et al., 1998) and marks a departure from early work that attributed NPP trends almost exclusively to increasing R auto (Waring & Schlesinger, 1985) or a combination of decreasing GPP and increasing R auto (Kira & Shidei, 1967).
The changes in biomass and production with stand age are most effectively investigated using a chronosequence, or space for time, approach Law et al., 2003;Litvak et al., 2003;Bond-Lamberty et al., 2004;Clark et al., 2004;Amiro et al., 2006;Goulden et al., 2006;Humphreys et al., 2006;Randerson et al., 2006;Noormets et al., 2007;Mkhabela et al., 2009;Zha et al., 2009). The boreal forest provides an ideal system to investigate the changes in production and carbon stocks with recovery. Lightning caused crown fires occur at individual locations about once every 100 years, resulting in a mosaic of large patches at different stages of succession (Van Cleve et al., 1983;Bonan & Shugart, 1989). We combined year-round eddy covariance measurements with in situ observations of carbon stocks and production along a fire-recovery chronosequence in Manitoba, Canada to investigate the patterns of NPP, GPP, respiration, and NEP during boreal forest succession. Our goals were to better develop the methods and strategy to investigate the changes in carbon pools and fluxes during succession and to explicitly test the trends predicted by Sprugel (1985) and Chapin et al. (2002).

Sites
We focused on a chronosequence in central Manitoba that included seven even-aged boreal forest stands at various stages of succession (Table 1; Bond-Lamberty et al., 2004;Goulden et al., 2006). The ages of the oldest two burns (1850 and 1930) were determined by coring black spruce; the ages of the younger stands were determined from fire records. For simplicity, we refer to the age of each site in 2004, even though some of the observations were made in other years and the Figures reflect the actual age at observation. Most of the observations were made at the University of California, Irvine (UCI) sites, using methods described here or in Goulden et al. (2006). A few of the observations were made at the University of Wisconsin (UW) sites, using methods described in Bond-Lamberty et al. (2004) and Wang et al. (2003). The UCI and UW sites were located in the same burn scars and were no more than 3 km apart. The methods used at NOBS-1850 are described in Goulden et al. (1997) and Dunn et al. (2007). The study was conducted in the area of the BOReal Ecosystem-Atmosphere Study (BOREAS) Northern Study Area (NSA; Sellers et al., 1997); many additional data sets from the BOR-EAS project are available (http://daac.ornl.gov/).

Carbon stocks, NPP and leaf area index (LAI)
The aboveground biomass and LAI (Table 2) of herbaceous plants (principally grasses and forbs and not including mosses) was determined at each UCI site (Table 1) in September 2004 by harvest. The harvests occurred shortly before the end of the growing season and were assumed to reflect peak biomass, peak LAI, and annual aboveground NPP (ANPP). The herbaceous material in 30 to 40 0.25 m 2 plots at 10 m intervals along 100 m transects radiating in the cardinal directions from the eddy covariance towers was clipped, returned to the lab, dried at 65 1C, and weighed. Subsamples were sorted into deciduous leaves, new evergreen leaves, and old evergreen leaves. LAI was calculated as the product of this leaf biomass and the specific leaf area (SLA) for individual species, which was measured at the UW sites (Bond-Lamberty et al., 2002b).
Wood production of saplings and shrubs was calculated for individual plants based on basal diameter, ring width, and species-specific allometric equations. The basal stem diameter (at the soil surface) of the understory shrubs and tree saplings at the UCI sites was measured with vernier calipers in 13-40 1 m 2 plots along the transects.  (2002b). The biomass and LAI of alder (Alnus spp.), which Bond-Lamberty and colleagues did not investigate, were calculated with allometric equations that we developed between basal diameter, biomass, and the mass of leaves by harvesting 103 alder shrubs across the UCI sites. The annual basal diameter increment of the understory shrubs and tree saplings was determined on cross sectional discs cut from stems in the 1 m 2 plots (266 discs at UCI-1998, 212 discs at UCI-1989, 132 discs at UCI-1981, 94 discs at UCI-1964, 44 discs at UCI-1930, and 33 discs at UCI-1850. Overstory tree diameters at breast height (DBH; 1.3 m) were measured in three or four 25-50 m 2 plots in the cardinal directions from the eddy covariance tower at each of the UCI sites (100 m 2 total at UCI-1989, 75 m 2 total at UCI-1981, 100 m 2 total at UCI-1964, 150 m 2 total at UCI-1930, and 150 m 2 total at UCI-1850). The overstory biomass and LAI of each species at each site was calculated with species-specific allometric equations from Bond-Lamberty et al. (2002a) and SLAs from Bond-Lamberty et al. (2002b). Wood production was determined from ring width. Trees of the dominant species at each site were cored twice (on the north and west aspect of the tree) at breast height (DBH; 1.3 m) with an increment borer 35 trees cored at UCI-1989, 101 trees at UCI-1981, 88 trees at UCI-1964, 117 trees at UCI-1930, and 96 trees at UCI-1850. The cores were sanded and scanned at 1200 dpi, and the ring widths determined with LIGNOVISION software (Rinntech, Heidelberg, Germany). Wood and leaf production were calculated based on measured DBH, ring width in 2004, and species-specific allometric equations (Bond-Lamberty et al., 2002a). The rates of individual tree production were summed for the plot to calculate the annual leaf and wood production.
Belowground fine, medium, and coarse root production (BNPP; Table 2) was determined by sequential coring at each of the UW sites (Bond-Lamberty et al., 2004). Twenty 16.6 cm 2 Â 50 cm deep cores were taken and sorted at each site in July and October 2001. Moss NPP at the UCI sites was  (Harden et al., 2009), natural markers (Callaghan et al., 1978), and the crank wire technique (Clymo, 1970). Site specific new growth by species, as measured by one of the three methods, was multiplied by species cover at each site (Harden et al., 2009). Moss growth was not measured at the 23-year-old stand, and the average species growth rate across all of the other sites was used.
Coarse woody debris stock (C CWD ; Table 2) was measured at each of the UCI sites by the line intercept method using 18-20 20 m long transects (Manies et al., 2005). Forest floor carbon stock (C forest floor ; Table 2; dead moss, leaf litter, fine debris, and partially decomposed organic material above the mineral soil) and live moss biomass were measured at each of the UCI sites. We measured the thickness of organic soil layers every 3-20 m along two or more linear transects at each site. We collected samples from the L (live moss), D (dead moss), F (fibric), M (mesic), and H (humic) horizons at a series of soil pits along the transects (Manies et al., 2006). Cumulative carbon storage for all layers above the mineral soil was calculated as bulk density multiplied by %C and by the thickness of the horizon.
Total NPP (TNPP; Table 2) was calculated as the sum of aboveground vascular and moss NPP at the UCI sites and root production at the UW sites (Table 2). ANPP (Table 2) was calculated as the sum of vascular and moss NPP at the UCI sites. Live biomass (C live ; Table 2) was calculated as the sum of vascular and moss biomass at the UCI sites. Total carbon (C total ; Table 2) was calculated as the sum of C live , C forest floor , and C CWD at the UCI sites.
NEP by change in stocks (NEP stocks ; Table 2) was calculated as the sequential difference in C total between the 154-, 74-, 40-, 23-, and 6-year-old UCI stands divided by the age difference between stands. NEP by the difference between NPP and decomposition (NEP NPP ; Table 2) was calculated as the difference between NPP at the UW sites and decomposition at the UW sites (Bond-Lamberty et al., 2004). Decomposition at the UW sites was determined by combining chamber-based measurements of soil respiration for plots that had been trenched to exclude live roots with field-and laboratory-based measurements of woody debris heterotrophic respiration (Bond-Lamberty et al., 2004).

Meteorological and tower fluxes
Eddy covariance towers were installed at the UCI sites in summer and operated until September 2005. Matched instruments and programs were used, which maximized site-to-site precision and allowed us to operate the tower network in an efficient manner . Aboveground net primary production Calculated as the sum of vascular and moss NPP at the UCI sites BNPP Belowground net primary production Measured at the UW sites by sequential coring TNPP Total net primary production Calculated as the sum of aboveground vascular and moss NPP at the UCI sites and root production at the UW sites C live Carbon stocks in live plants Calculated as the sum of vascular and moss biomass at the UCI sites C forest floor Carbon stocks in the forest floor Measured at the UCI sites by harvest C CWD Carbon stocks in coarse woody debris Measured at the UCI sites by line intercept C total Total carbon stocks excluding the soil Calculated as the sum of C live , C forest floor , and C CWD at the UCI sites R total Total ecosystem respiration Calculated by integrated eddy covariance at the UCI sites using a u* filter of 0.3 ms À1 and daytime extrapolation based on air temperature GPP Gross primary production Calculated by integrated daytime eddy covariance at the UCI sites after subtracting R total NEP EC Net ecosystem production determined by Eddy covariance Calculated at the UCI sites using a u* filter of 0.3 ms À1 NEP stocks Net ecosystem production determined by change in C total between sites Calculated as the difference in C total between sequentially aged stands divided by the age difference NEP NPP Net ecosystem production determined by the difference between NPP and decomposition Calculated as the difference between TNPP and decomposition at the UW sites (Bond-Lamberty et al., 2004). Decomposition at the UW sites was determined with measurements of soil respiration in trenched plots R auto Annual plant respiration Calculated as the difference between GPP and TNPP R hetero Annual heterotrophic respiration Calculated as the difference between TNPP and NEP EC TNPP/GPP Total primary production efficiency Calculated as TNPP divided by GPP ANPP/GPP Aboveground primary production efficiency The CO 2 fluxes were calculated as the 30 min covariance of vertical wind velocity and CO 2 mixing ratio. Wind and temperature were measured with three-axis sonic anemometers (CSAT3, Campbell Scientific, Logan UT, USA); the molar concentrations of CO 2 and H 2 O were measured with closed path IRGAs (LI7000, LiCor, Lincoln, NE, USA). The IRGAs were calibrated automatically for CO 2 once a week by sequentially sampling CO 2 and H 2 O scrubbed air and CO 2 standard in air ( AE 1%, Scott Marin, Riverside, CA, USA). Net radiation (REBS Q*7.1, Seattle, WA, USA), incoming and reflected shortwave radiation (CM3, Kipp and Zonen, Delft, the Netherlands), incoming and reflected photosynthetically active photon flux density (PPFD) LiCor), and aspirated air temperature and relative humidity (HMP45C, Vaisala, Helsinki, Finland) were measured at the tower tops. The tower infrastructure at UCI-2003 differed somewhat from the other sites; UCI-2003 was installed after the main study had started, and budget constraints restricted measurements there to the growing season. The annual tower fluxes at the UCI sites were calculated from October 1 to September 30 of each year by summing continuous, gap-filled time series that accounted for data loss from instrument failure and inadequate turbulence. Nocturnal respiration and NEP by integrated eddy covariance (NEP EC ; Table 2) were calculated using a u* filter of 0.3 ms À1 . Total respiration (R total ) was calculated by extrapolating the nocturnal observations of respiration to daytime based on air temperature. GPP was calculated by integrating the daytime eddy covariance observations after subtracting the extrapolated rates of respiration. The measurements at UCI-2003 site were restricted to the summer, and the annual fluxes there were estimated based on the winter observations at the other young sites. The annual tower fluxes at NOBS-1850 were calculated from January 1 to December 31 of each year as described by Dunn et al. (2007).
The use of u* filters (Goulden et al., 1996) remains somewhat controversial, though there is consensus some type of correction on calm nights is needed (Finnigan, 2008). We investigated the sensitivity of NEP EC to u* threshold for the six sites with year-round fluxes by sequentially calculating NEP EC at u*s of 0, 0.1, 0.15, 0.2, 0.225, 0.25, 0.3, and 0.4 ms À1 (Fig. 2). NEP EC initially decreased with an increasing u* filter at all sites. NEP EC reached a plateau at a filter of $ 0.2 ms À1 at some sites (1998,1981) and a filter of $ 0.3 ms À1 at other sites (1930,1850). The degree of site-to-site variation, as well as the general relationship, is typical of reports in the literature. There was no systematic trend in behavior; sites that required a higher filter included UCI-1989, which had a relatively open and short canopy, and UCI-1850, which had one of the densest and tallest canopies. The slope of u* vs. NEP EC was consistent between sites, making the site-to-site patterns of NEP EC largely independent of the threshold selected. We used a u* filter of 0.3 ms À1 for all data processing; all of the sites reached a plateau by this point, decreasing by no more than 14 g C m À2 yr À1 in response to an incremental increase from 0.3 to 0.4 ms À1 (Fig. 2).
We created a modeled respiration data set based on the relationship between air temperature and nighttime efflux. We recalculated the temperature sensitivity term in the model at 5-day intervals (Reichstein et al., 2005). Gaps at night were filled with the modeled respiration data set. Gaps during the day were filled using modeled daytime net uptake calculated from a three parameter hyperbolic fit between CO 2 uptake and incoming PPFD (Falge et al., 2001). Two-month time windows were used for the CO 2 uptake regressions. Gaps during periods when regression statistics indicated a poor fit (r 2 o0.1 at night or r 2 o0.4 during daytime) were filled by diel look-up tables and/or linear interpolation. The gap filling for daytime uptake was done over a series of temperature ranges to account for the effect of temperature on photosynthetic light response.
The flux tower analysis emphasizes data from October 2002 to September 2004, since these years had the most continuous observations  and since the site-to-site patterns of exchange during these years were similar to those observed in the other years . The final data sets are available from the authors (http://www.ess. uci.edu/ $ boreal_canada/index.html) and the AmeriFlux Data Archive (http://public.ornl.gov/ameriflux/data-accessselect.shtml). The raw data are available from the authors.

Uncertainty
We provide our best estimates of uncertainty in Table 3, which draws heavily on our previous work. Sources of uncertainty are divided into: (1) Measurement and aggregation error, such as that caused by poorly calibrated instruments or by within-site spatial variability or by day-to-night biases in the calculation of eddy covariance sums (Goulden et al., 1996;Bond-Lamberty et al., 2004).
(2) Site-to-site and year-to-year imprecision, such as that caused by the inconsistent application of a technique from one site to another or by measurement drift over time (Goulden et al., 1996. (3) Across landscape sampling error, such as that caused by stand-to-stand or burn-to-burn heterogeneity (Bond-Lamberty et al., 2004;Goulden et al., 2006). (4) Temporal sampling error, such as that caused by interannual or decadal variability that results in an anomalous measurement period (Dunn et al., 2007).
Ninety percent confidence intervals are expressed in absolute physical units. In some cases, the uncertainties are highly subjective (e.g., the accuracy of R auto and R hetero ). In other cases, especially involving spatial sampling, there is insufficient information to even allow an estimate. Relatively simple measures of vegetation density, such as LAI, ANPP, and C live , have low measurement and aggregation errors and good precision. The greatest uncertainty for these measures is caused by spatial heterogeneity across the landscape. More complex measures, especially those involving nocturnal eddy covariance fluxes (e.g., R total ), have large measurement and aggregation uncertainty and somewhat more favorable precision.
Site-to-site precision is most important for comparing between sites (e.g., determining whether one site is different from another). Site-to-site precision and spatial sampling error are most important for assessing how ecosystem properties change during succession (e.g., determining how NEP changes during succession). Measurement and aggregation errors, spatial sampling errors, and temporal sampling errors are most important when comparing the observations with independent information (e.g., using the observations to test a model or develop a regional carbon budget). Readers are urged to keep these issues in mind as they evaluate and use the data.

Sign convention
Annual production (i.e. NPP, NEP, GPP) is presented in the ecological sign convention, where a positive flux indicates the net transfer of carbon from the atmosphere to the ecosystem. Half-hour eddy covariance fluxes are presented in the atmospheric sign convention, where a positive flux indicates the net transfer of carbon from the ecosystem to the atmosphere. Studies of ecosystem production during succession have 1 m 2 m À2 * 1 m 2 m À2 * 2 m 2 m À2w,z 1 m 2 m À2 § C live 500 g C m À2 * 500 g C m À2 * 1000 g C m À2w,z 100 g C m À2} C forest floor 500 g C m À2 * 500 g C m À2 * 1000 g C m À2z Unknown C CWD 300 g C m À2 * 300 g C m À2 * 2000 g C m À2z 100 g C m À2} C total 800 g C m À2 * ,k 800 g C m À2 * ,k 2500 g C m À2w,z,k 150 g C m À2},k ANPP 30 g C m À2 yr À1 * 30 g C m À2 yr À1 * 100 g C m À2 yr À1w,z 50 g C m À2 yr À1 §,} BNPP 75 g C m À2 yr À1 * 75 g C m À2 yr À1 * Unknown Unknown TNPP 80 g C m À2 yr À1 * ,k 80 g C m À2 yr À1 * ,k 150 g C m À2 yr À1w,z,k 75 g C m À2 yr À1 §,},k R total 200 g C m À2 yr À1 * , ** ,ww 50 g C m À2 yr À1 * , ** ,ww Unknown 100 g C m À2 yr À1zz GPP 100 g C m À2 yr À1 * , ** 50 g C m À2 yr À1 * , ** 200 g C m À2 yr À1w 100 g C m À2 yr À1 §,zz NEP EC 50 g C m À2 yr À1 * , ** ,ww 30 g C m À2 yr À1 * , ** ,ww Unknown 50 g C m À2 yr À1zz NEP stocks 20 g C m À2 yr À1 * ,k 20 g C m À2 yr À1 * ,k 150 g C m À2 yr À1w,z 5 g C m À2 yr À1},k NEP NPP 50 g C m À2 yr À1 * , ** ,ww 20 g C m À2 yr À1 * ,k 100 g C m À2 yr À1w,z Unknown R auto 125 g C m À2 yr À1 * , ** ,k 95 g C m À2 yr À1 * , ** ,k Unknown 50 g C m À2 yr À1zz,k R hetero 95 g C m À2 yr À1 * , ** ,k 85 g C m À2 yr À1 * , ** ,k Unknown 50 g C m À2 yr À1zz,k TNPP/GPP 0.2k 0.  Goulden et al., 2006). }Based on tree ring chronologies or multiyear sampling (Bond-Lamberty et al., 2004;Rocha et al., 2006). kBased on propagation of component uncertainties. **Based on sensitivity analysis during aggregation process (e.g., Goulden et al., 1996). wwBased on comparisons with independent measures. zzBased on measured interannual and decadal variability at all UCI sites and NOBS-1850Dunn et al., 2007). LAI, leaf area index; C live , carbon stocks in live plants; C forest floor , carbon stocks in the forest floor; C CWD , carbon stocks in coarse woody debris; ANPP, aboveground net primary production; BNPP, belowground net primary production; TNPP, total net primary production; GPP, gross primary production; NEP EC , net ecosystem production determined by Eddy covariance; NEP stocks , net ecosystem production determined by change in C total between sites; NEP NPP , net ecosystem production determined by the difference between NPP and decomposition; R auto , annual plant respiration; R hetero , annual heterotrophic respiration.
typically used the ecological sign convention (e.g., Chapin et al., 2002), whereas most recent studies of half-hour wholeecosystem flux have used the atmospheric sign convention (e.g., Baldocchi, 2008).

Species composition and LAI
The 1-to 15-year-old stands were dominated by small resprouting shrubs and herbaceous ruderals, with tree species present only as seedlings (Table 1). The 23-and 40-year-old stands were dominated by jack pine, aspen, and black spruce saplings, with small patches of sphagnum (Sphagnum spp.) and feathermoss (Pleurozium or Hylocomium spp.). The 74-and 154-year-old stands were closed-canopy black spruce forest with a thick, continuous layer of sphagnum and feathermoss. The LAI at the UCI sites increased rapidly from 0.3 m 2 m À2 1 year after the burn to 2.1 m 2 m À2 6 years after burn to 5.5 m 2 m À2 23 years after burn (Fig. 3). LAI plateaued at 5.3-5.5 m 2 m À2 from 23 to 154 years, except for a peak of 7.2 m 2 m À2 at the 74-year-old site. The relative importance of deciduous and evergreen plants shifted during succession, with deciduous herbs (grass and fireweed) and shrubs (alder and willow) assuming initial importance, and evergreens (black spruce and moss) dominating older sites.

NPP
Vascular plant ANPP at the UCI sites increased rapidly, rising from 12 g C m À2 yr À1 at the 1-year-old stand to 70 g C m À2 yr À1 at the 6-year-old stand (Fig. 4a). Vascu-lar ANPP peaked at 243 g C m À2 yr À1 at the 23-year-old stand and declined to 126 g C m À2 yr À1 at the 154-yearold stand. Vascular ANPP was similar between the UCI and UW sites, with the exception of the 40-year-old burn, where the UW stand was much more productive (Fig. 4a).
The distribution of production by plant functional type (Fig. 4b) paralleled the changes in LAI and species composition ( Fig. 3; Table 1). Production at the younger stands was dominated by shrubs and herbaceous plants; production at the older stands was dominated by trees.
BNPP was greatest at the 74-year-old stand and somewhat reduced at the 23-, 40-, and 154-year-old stands (Total roots in Fig. 4b). BNPP was always less than ANPP. Some of the BNPP differences between stands may reflect sampling variability (Table 3; Bond-Lamberty et al., 2004). BNPP measurements were unavailable for the 1-year-old stand and were considered unreliable at the 6-year-old stand. Many of the plants at the 6-year-old stand were geophytes or cryptophytes, which store carbohydrates in their roots or rhizomes  during winter. These belowground stores were not well sampled by the max-min coring approach, which assumes root production and stocks are greatest in mid-summer.
TNPP peaked at the 23-and 74-year-old stands, declined at the 154-year-old stand, and was somewhat reduced at the 40-year-old stand (Fig. 4b). The dip at the 40-year-old stand may have been due in part to uncertainty and variability in the measurements of belowground production, and also to possible mismatches in site selection (see 'Discussion').

Biomass and carbon stocks
Live biomass (vascular plants and bryophytes) increased from 12 g C m À2 at the 1-year-old stand to 5680 g C m À2 in the 154-year-old stand (Fig. 5). The stand-to-stand biomass increment was comparatively low among the youngest and among the oldest stands and high between the 40-and 74-year-old stands. The initial acceleration in biomass accumulation (Fig. 5) paralleled the observed increases in ANPP and allocation to long-lived tissues such as wood (Fig. 4). The net biomass increment between the youngest two sites (17 g C m À2 yr À1 ) was similar to the ANPP of long-lived tissues at the 6-year-old stand (17 g C m À2 yr À1 ). The net increment between the 15-and 23-year-old stands (75 g C m À2 yr À1 ) was less than the ANPP of long-lived tissues at the 23-year-old stand (166 g C m À2 yr À1 ).
Carbon stocks in the forest floor (C forest floor ) were comparatively high in the 1-year-old stand, low in the 6-, 15-, 23-, and 40-year-old stands, and very high in the 74-and 154-year-old stands. C CWD were broadly constant across sites, with somewhat greater stocks at the 6-, 23-, and 40-year-old stands and lower stocks at the 74-year-old stand. C forest floor was more dynamic than C CWD ; C forest floor ranged by 3200 g C m À2 between stands, while C CWD ranged by 800 g C m À2 . C forest floor and C CWD followed nearly opposite trajectories, with C forest floor increasing markedly and C CWD decreasing between the 40-and 74-year-old stands.
The sum of live biomass, forest floor, and CWD provides a measure of total carbon (C total ), which can be used to calculate NEP based on changes in stocks (NEP stocks ; Table 2) if we assume the amount of carbon in the soil beneath the forest floor is static Wang et al., 2003). Total carbon was comparatively constant among the 15-, 23-, and 40-year-old stands and also among the 74-and 154year-old stands. Total carbon increased considerably between the 40-and 74-year-old stands (Fig. 5), reflecting the accumulation of carbon in live plants and the forest floor.

GPP and respiration
The peak rate of mid-day CO 2 uptake was remarkably consistent at all sites except the 1-year-old burn (Fig. 6). Peak CO 2 uptake was not related solely to either LAI or leaf longevity; the 6-year-old stand, which had a deciduous canopy with an LAI of 2.1 m 2 m À2 , and the 74year-old stand, which had an evergreen canopy with an LAI of 7.2 m 2 m À2 , both had a maximum CO 2 uptake of 10 mmol m À2 s À1 (see also . The growing season length and seasonal patterns of NEE mirrored the shifts in leaf phenology (Fig. 3). The 2004 growing season length, calculated as the period when the maximum daily uptake was at least 35% of the maximum summer NEE , lasted 68 days at the 6-year-old stand and 88 days at the 15-year-old stand. The 2004 growing season was 126 days at the 23-year-old stand, 138 days at the 40-yearold stand, 137 days at the 74-year-old stand, and 146 days at the 154-year-old stand. The youngest two stands were dominated by deciduous plants and had comparatively short growing seasons; the oldest four stands were dominated by evergreens and had long growing seasons.
The trends in GPP with age followed the shifts in growing season length (Fig. 6). Average GPP increased from 380 g C m À2 yr À1 at the 6-year-old stand to 450 g C m À2 yr À1 at the 15-year-old stand to 710 g C m À2 yr À1 at the 23-year-old stand (Fig. 7). GPP ranged from 630 to 720 g C m À2 yr À1 at the three oldest stands. The increase in GPP from the 15-to 23-year-old Live plant carbon stocks (C live ; filled circles, g C m À2 ) calculated as the sum of vascular biomass and moss biomass at the UCI sites. Forest floor carbon stocks (C forest floor ; open circles, g C m À2 ) at the UCI sites. Coarse woody debris carbon stocks (C CWD ; filled triangles, g C m À2 ) at the UCI sites (data from Manies et al., 2005). Total carbon stocks (C total , g C m À2 ) calculated as the sum of C live , C forest floor , and C CWD . C CWD and C total were unavailable for the 1-and 15-year-old stands. stand coincided with the transition from mainly deciduous to mainly evergreen vegetation (Fig. 3) and the associated increase in growing season length.
Whole ecosystem respiration (R total ) was comparatively high at the 23-, 45-, and 154-year-old stands (Fig. 7). The site-to-site patterns of R total broadly paralleled the patterns of GPP. The higher rate of R total at the 23-and 45-year-old stands coincided with comparatively high rates of GPP. The maximum rates of both R total and GPP were observed at the 154-year-old stand. R total was more consistent between stands than was GPP; R total ranged by 230 g C m À2 yr À1 between sites whereas GPP ranged by 340 g C m À2 yr À1 .

NEP
The annually integrated net CO 2 exchange (NEP EC ; Fig. 8) indicated the 1-and 6-year-old stands were losing carbon, the 15-year-old stand was gaining a small amount of carbon, the 23-and 74-year-old stands were gaining considerable carbon, and the 40-and 154-yearold stands were gaining modest amounts of carbon. A linear fit through all the NEP EC observations at the 6-and 15-year-old stands indicated the transition from carbon source to sink occurred at an age of 11-12 years. NEP EC was variable among the four oldest sites, with consistently high rates of carbon storage at the 23-and Fig. 6 (a-g) Net CO 2 exchange (F CO2 ; 30 min covariances, mmol m À2 s À1 ) during 2004 at seven sites that differ in age since last burn. (hm) Gross primary production (GPP, g C m À2 yr À1 ), ecosystem respiration (R total , g C m À2 yr À1 ), and net ecosystem production (NEP EC , g C m À2 yr À1 ) integrated from October 1, 2003 to September 30, 2004 at the six sites with year-round observations. F CO2 is presented in the atmospheric sign convention, with a positive flux indicating the net transfer of carbon from the ecosystem to the atmosphere. GPP, R total , and NEP EC are presented in the ecological sign convention, with a positive flux indicating the net transfer of carbon from the atmosphere to the ecosystem.
74-year-old stands and low rates at the 40-and 154year-old stands (see also . The comparatively high rates of NEP EC at the 23-year-old stand (Fig. 8) coincided with high rates of GPP (Fig. 7), whereas the high rates of NEP EC at the 74-year-old stand coincided with reduced R total .
UCIs 154-year-old stand was 400 m from the BOREAS NSA NOBS eddy covariance tower (Table 1; NOBS-1850), which began operation in 1995 (Goulden et al., 1997;Dunn et al., 2007). The NEP EC s at NOBS-1850 and UCI-1850 were comparable, with similar patterns of interannual variability (Table 4). The long-term record at NOBS-1850 indicates a marked increase in NEP around [1998][1999][2000]. NEP EC at NOBS-1850 averaged 40 g C m À2 yr À1 in 2003-04 and 2 g C m À2 yr À1 in 1995-2004 (Table 4). The cause of this trend is not fully understood, though it may be related to water table depth (Dunn et al., 2007).

Comparison of integrated eddy covariance with biomass accumulation
The suite of measurements allowed us to calculate NEP along the chronosequence using three fully independent approaches ( Table 2): (1) integrated eddy covariance (NEP EC ; Fig. 8), (2) site-to-site differences in total carbon stocks (NEP stocks ; Fig. 5), and (3) NPP minus decomposition estimated from soil and woody debris respiration (NEP NPP ; Bond-Lamberty et al., 2004). The three approaches gave broadly similar results (Fig. 9), with a loss of carbon that lasted $ 10 years and peaked at 50-150 g C m À2 yr À1 , followed by an extended period of net carbon uptake of 50-150 g C m À2 yr À1 from $ 10 to 100 years, and a reduced net carbon uptake of 0-50 g C m À2 yr À1 after $ 100 years. The NEP stocks approach was comparatively variable over time, which likely reflects spatial heterogeneity and landscape scale sampling uncertainty (Table 3). Unfortunately, C CWD observations were unavailable for the 1-year-old site, and we were unable to determine the changes in C total during the first few years after fire.
A more quantitative comparison of methods is possible by considering the long-term rates of carbon sequestration. The difference in C total between the 6and 74-year-old stands was 6500 g C m À2 (Fig. 5), whereas the integrated NEP EC between these stands was 5400 g C m À2 (Fig. 8). The difference in C total between the 6-and 154-year-old stands was 8400 g C m À2 , whereas the integrated NEP EC between these stands was 13 000 g C m À2 . The calculation of integrated NEP EC between the 6-and 154-year-old stands may be confounded by the wide age difference between the oldest stands and uncertainty over how long the high NEP EC observed at the 74-year-old stand continues. Additionally, integrated NEP EC may have overestimated longterm carbon storage as a result of temporal sampling error (Table 3). The observations at NOBS-1850 indicate the carbon uptake during 2003-2004 was $ 40 g C m À2 yr À1 greater than during 1995-2004 (Table  4). Tree rings indicate this trend may have occurred at the 74-year-old stand as well , suggesting our study period was anomalous. This possibility is consistent with the observation that NEP EC exceeded NEP NPP and NEP stocks in the oldest two stands (Fig. 9). The integrated NEP EC between the 6-and 154-year-old stands after subtracting Fig. 7 Gross primary production (GPP; unfilled circles, g C m À2 yr À1 ) and total respiration (R total ; filled circles, g C m À2 yr À1 ) averaged across 2002-03 and 2003-04 as a function of time since last burn. 40 g C m À2 yr À1 at the 74-and 154-year-old stands was 9200 g C m À2 . Given the variety and range of uncertainties (Table 3), the agreement between the measures of NEP is quite reasonable.

Sampling challenges
Landscape heterogeneity, both between and within fires, is a poorly characterized and potentially large source of uncertainty in chronosequence studies (referred to as across landscape sampling accuracy in Table 3). The UCI chronosequence appeared well matched with the notable exception of the 40-year-old stand, where a variety of measures indicate anomalously low productivity. TNPP, C live , and NEP EC were less at the 40-year-old stand than would be expected based on the 23-and 74-year-old stands (Figs 4, 5 and 8). Likewise, vascular plant ANPP at the UCI sites was generally similar to that at the UW sites except for UCI-1964, which was much less productive than UW-D1964 (Fig. 4a). A 6-week record of eddy covariance measurements for an area near UW-D1964 (Litvak et al., 2003) suggests a much higher NEP EC than at UCI-1964. A possible discrepancy at UCIs 40-year-old stand was also identified in remote sensing analyses, which indicated lower enhanced vegetation indices (EVI) at UCI-1964 than other comparably aged stands . The EVI analysis also indicated UCI's 23-yearold stand may be somewhat more productive than is typical for the area. The possibility of site-to-site mismatches, especially at the 40-year-old stand, should be considered when interpreting the figures.
Our selection of eddy covariance sites was conducted in 2001 and included visits to several candidates in each burn. In retrospect, the selection process would have benefited from an extensive survey of the spatial variation within each burn using satellite imagery. Satellite analyses during the early phases of site selection and identification would have been particularly useful for objectively identifying representative stands of given ages. However, we were unable to carry out this analysis until late 2005 , when advances in the availability of MODIS and Landsat imagery, image analysis software, and computing power facilitated the investigation. Vegetation indices and products are available for albedo, LAI, and CO 2 uptake, allowing consideration of several biophysical attributes . Incorporation of remote sensing analyses at an early planning stage could have improved the match between sites and should be considered in future chronosequence studies.

Changes in stocks during succession
The hypotheses in Fig. 1 were originally sketched by hand, and were intended as qualitative and relative predictions. Judged against this goal, the models did a good job of capturing the dominant patterns during succession. The observed changes in biomass and detrital pools matched the predictions particularly well (Figs 1a and  10a), probably because biomass is comparatively easy to measure and these predictions could be based on direct observation (Bormann & Likens, 1979). The observed recovery of C live (Fig. 5) followed the expected logistic pattern, with slow initial accumulation followed by rapid accumulation in mid-succession and reduced accumulation in the oldest stand. The detrital pools also matched the predictions, with both C forest floor and C CWD showing comparatively high levels following fire, subsequent declines in early or mid-succession, and increases in late succession. C forest floor was more dynamic than C CWD , with larger and more rapid changes in pool size. Unfortunately, measurements of C CWD were unavailable for the 1-and 15-year-old stands, which prevented complete comparison of observed and predicted C total . Nonetheless, it is likely the changes in C total matched the predictions, with initial loss, followed by rapid accumulation, and reduced accumulation in the oldest stand. C forest floor was markedly different between the 1-and 6-year-old stands, corresponding to a loss of 430 g C m À2 yr À1 . Wang et al. (2003) observed a similar large drop in C forest floor between the youngest UW stands. It is likely C total declined considerably during the first 5 years following fire, with most of this loss coming from C forest floor .

Patterns of NPP, GPP, respiration, and NEP during succession
The observed changes in NEP, NPP, and GPP ( Fig. 10b and c) broadly matched the predictions (Fig. 1b and c). NEP EC was initially negative, transitioned to an extended period of carbon accumulation, and showed reduced accumulation in the oldest stand. NPP and GPP were initially low after fire, recovered to a peak in early and mid-succession, and either declined (NPP) or remained constant (GPP) at the oldest stand. The initial suppression of NPP and GPP following disturbance was brief, and the peak rate of NPP was observed at the 23-year-old stand. The TNPP decline at the oldest stand was not as large as expected. The lack of a dramatic decline may simply reflect the range of ages sampled; the rates of production in stands older than 154 years may be lower, though most forest in the area burns before it reaches this age.
The observed patterns of respiration (Fig. 10d) did not match the predictions as well (Fig. 1d). R auto was initially low and varied considerably during succession The time of fire is indicated by the dashed vertical line. C CWD , C total , GPP, TNPP, R total , R hetero , R auto , TNPP/GPP, and ANPP/GPP were unavailable for the 1-year-old stand. TNPP, R hetero , R auto , and TNPP/ GPP were unavailable for the 6-year-old stand. C CWD and C total were unavailable for the 15-year-old stand. in a way that paralleled GPP, as predicted. Variation in R total was most strongly driven by variation in R auto , and, as a result, R total peaked in mid-succession and was reduced in the youngest stands, also as predicted. But R auto increased in the oldest stand, at a time when a decrease was predicted. And the changes in R hetero in the youngest or oldest stands were less than predicted. The difficulty of predicting respiration mirrors the general observation that respiration in terrestrial ecosystems is poorly understood relative to photosynthesis (Trumbore, 2006).
Why is the loss of carbon in the decade following disturbance less than expected?
The initial loss of carbon following disturbance (Fig.  10b) was neither as protracted nor large as predicted (Fig. 1b). A similar pattern has been observed in other eddy covariance chronosequence studies of boreal fire recovery, which have reported only modest carbon losses in the decade following disturbance and a transition to annual net uptake within 10-15 years (Litvak et al., 2003;Amiro et al., 2006;Randerson et al., 2006).
This muted response may reflect the rapid regrowth of vegetation following fire. Boreal forests have an extensive community of ruderal and resprouting herbs and shrubs that rapidly reestablish LAI, GPP, and NPP (Figs 3, 4, 6, and 7), and allow a quick transition from carbon source to sink. Other ecosystems may be less well adapted to crown fire. The rapid recovery of midday NEE and annual NEP we observed contrasts with that observed after a crown fire in a southwestern ponderosa pine stand (Dore et al., 2008). We found the mid-season daytime rates of CO 2 uptake had recovered to the rates observed in older stands, and the annual NEP approached zero, within 10 years after fire (Fig. 6). In contrast, Dore et al. (2008) found the mid-season daytime rates of CO 2 uptake were less than half those observed in an older stand, and the site was still a large source of CO 2 , 10 years after fire. Southwestern US pine forests were historically exposed to frequent lowintensity ground fires, and, as a result, lack a plant community that is adapted for swift tree establishment and regrowth following crown fire (Savage & Mast, 2005).
Boreal fires consume most of the fine fuel, while killing the trees and leaving coarse woody debris standing. Standing debris takes a long time to decompose (Yatskov et al., 2003;Bond-Lamberty & Gower, 2008), and the relative lack of fine detritus may also help explain the muted NEP response. The annual carbon loss in the first years after disturbance normalized by the maximum uptake during regrowth may differ between forests recovering from clear cut and those recovering from crown fire. The annual carbon loss observed 2-3 years after clear cutting in a boreal pine forest (Zha et al., 2009) and also in a Douglas fir forest (Humphreys et al., 2006) was roughly twice the annual peak uptake observed for older but otherwise comparable stands. In contrast, we estimated a peak efflux 2-3 years after fire that was o70% of the peak uptake observed for older stands (Fig. 8). This pattern is consistent with a difference in the input of fine and coarse debris during disturbance. Clear cutting leaves large amounts of fine slash, which would be expected to decompose rapidly, resulting in a high peak efflux. Crown fire leaves mostly coarse and standing debris, which would be expected to decompose slowly, resulting in a muted efflux.
Why does NPP decline in older stands?
The cause of reduced NPP in old stands has been the subject of considerable interest (Ryan et al., 1997), with a variety of mechanisms proposed, including ones that posit NPP/GPP is constant and attribute declining NPP to declining GPP, and ones that posit NPP/GPP varies and attribute declining NPP to increasing R auto . Our findings at the 154-year-old stand contradict the idea that NPP/GPP is constant (Fig. 1e), and are most consistent with mechanisms involving increasing R auto . GPP increased slightly from the 23-to 154-year-old stand, while TNPP decreased by 35% and ANPP decreased by 41%. These trends resulted in a large decrease in biomass production efficiency (Fig. 10e); TNPP/GPP decreased from 0.47 at the 23-year-old stand to 0.31 at the 154-year-old stand, and ANPP/ GPP decreased from 0.36 at the 23-year-old stand to 0.21 at the 154-year-old stand. Our observations support previous reports that old growth forest often has a low NPP/GPP (e.g. Kira, 1975;Amthor & Baldocchi, 2001;Chambers et al., 2004;Delucia et al., 2007;Figueira et al., 2008).
We were unable to directly identify the physiological basis for increasing R auto in the 154-year-old stand. The increase in R auto did not appear to be caused by accelerated maintenance respiration with increased biomass (e.g., Waring & Schlesinger, 1985). R auto did not vary with age in direct proportion to C live ; the biggest change in C live occurred from the 23-and 40-year-old stands to the 74-year-old stand (Fig. 5), a period when R auto decreased or remained constant (Fig. 10). Likewise, the increase in R auto at the 154-year-old stand does not appear to be caused by increased growth respiration, since this trend coincided with declining ANPP and TNPP (Fig. 10).
A more plausible explanation involves increasing allocation to fine root turnover, root metabolism, alternative forms of respiration, mycorrhizal relationships, or root exudation, possibly caused by progressive nutrient limitation. Boreal forests are thought to experience nutrient limitation with age, as a result of sequestration in detritus and soil cooling by moss and the evergreen canopy (Van Cleve et al., 1983;Bonan & Shugart, 1989). Unfortunately, we lack detailed direct observations of the patterns of belowground and respiratory allocation during succession. The observations of root production are somewhat erratic from site-to-site as a result of sampling challenges ( Fig. 4; Table 3), and more information on belowground allocation and production is needed (Gower et al., 2001). Nonetheless, the existing data are consistent with increasing allocation belowground over time. The fraction of GPP allocated to root NPP (Figs 4b and 7) was lowest in the 15-year-old stand (4%), moderate in the 23-, 40and 154-year-old stands (9-11%) and greatest in the 74year-old stand (27%).

Carbon accumulation by old stands
We found high rates of carbon accumulation at the 74year-old stand and reduced carbon accumulation at the 154-year-old stand (Figs 9 and 10). The transition between the 40-and 74-year-old stands was marked by the closure of the black spruce canopy and the establishment of a continuous moss layer (Table 1). Black spruce trees are relatively long lived, with a life span of up to $ 200 years (Viereck & Johnston, 1990). Likewise, dead moss is slow to decompose, with a turnover time of $100 years . The high NEP and NEP/TNPP (Fig. 10e) at the 74-year-old stand was a consequence of increased allocation of NPP to black spruce and moss (Fig. 4), which are both relatively longlived pools. The production of wood caused a large input to and buildup of C live ; the production of moss and evergreen needles caused a large input to and buildup of C forest floor . The buildup of C live and C forest floor at the 74-year-old stand resulted in a comparatively high NEP EC .
Both NEP EC and NEP stocks were reduced at the 154year-old stand relative to the 74-year-old stand (Figs 9 and 10). Moreover, the long-term observations at NOBS-1850 indicate the 154-year-old stand was in approximate steady state with the atmosphere, with a decadal average NEP EC close to zero (Table 4; Dunn et al., 2007). The reduced carbon sequestration at the 154-year-old stand did not reflect a decrease in the production of black spruce or moss; these components of production were similar between the 74-and 154-year-old stands (Fig. 4b). Rather, the declines in NEP EC , NEP stocks , and NEP/TNPP imply an increase in the loss from C live by tree mortality and an increase in the loss from C forest floor by decomposition. The wood NPP at the 74-year-old stand was 87 g C m À2 yr À1 (Fig. 4b) whereas the net C live increment between the 40-and 74-year-old stands was 80 g C m À2 yr À1 (Fig. 5), implying a small loss of C live by mortality. In contrast, the wood NPP at the 154-year-old stand was 75 g C m À2 yr À1 whereas the net C live increment between the 74-and 154-year-old stands was only 14 g C m À2 yr À1 , implying greater mortality losses. Likewise, the moss and needle production at the 74-year-old stand was 65 g C m À2 yr À1 whereas the net C forest floor increment between the 40-and 74-year-old stands was 74 g C m À2 yr À1 , implying that C forest floor decomposition was quite low. In contrast, the moss and tree leaf production at the 154-year-old stand was 66 g C m À2 yr À1 whereas the net C forest floor increment between the 74-and 154-year-old stands was just 6 g C m À2 yr À1 , implying increased decomposition of C forest floor . This apparent increase in C forest floor decomposition was not evident in the patterns of R hetero (Fig.  10d), though this may simply reflect measurement uncertainties (Table 3).
The cause of increased C live mortality and C forest floor decomposition losses at the 154-year-old stand is unclear, though it does not appear to be driven exclusively by simple first-order kinetics, where an increase in the size of a homogenous pool accelerates the loss from that pool simply because there is more carbon to die or decay. The rates of mortality and decomposition increased several fold between the 74-and 154-year-old stands, whereas the amounts of C forest floor and C live increased by just 15-25% (Fig. 5). Nonetheless, the NEP and carbon storage efficiency (NEP/TNPP) observations support the hypothesis that old stands approach steady state ( Fig. 1b and e; Odum, 1969), and are inconsistent with the notion that older forest stands continue to sequester large amounts of carbon (Luyssaert et al., 2008).

Implications and conclusions
The strategy of using multiple methods, including biometry and micrometeorology, worked well; in particular, the three independent measures of NEP during succession gave similar results. Nonetheless, landscape heterogeneity, both between and within fires, remains a poorly characterized and potentially large source of uncertainty. Incorporation of remote sensing analyses may help in the selection of sites and the identification of spatial sampling errors. Additionally, a stratified and tiered approach to deploying eddy covariance systems that combines many lightweight and portable towers with a few permanent ones is likely to maximize the science return for a fixed investment. The existing conceptual models did a good job of capturing the dominant patterns of NPP, GPP, and NEP during succession. The NEP and carbon storage efficiency (NEP/TNPP) observations support the hypothesis that old stands approach steady state. The initial loss of carbon following disturbance was neither as protracted nor large as predicted. This muted response reflects both the rapid regrowth of vegetation following fire and the prevalence of standing coarse woody debris following the fire, which is thought to decay slowly. In general, the patterns of forest recovery from disturbance should be expected to vary as a function of climate, ecosystem type, disturbance intensity, and disturbance type. Future efforts should be directed at understanding how recovery differs between contrasting types of ecosystems, and between different intensities and types of disturbance.
The NPP decline at the older stands appears related to increased R auto rather than decreased GPP. The increase in R auto in the older stands does not appear to be caused by accelerated maintenance respiration with increased biomass, and more likely involves increased allocation to fine root turnover, root metabolism, alternative forms of respiration, mycorrhizal relationships, or root exudates, possibly associated with progressive nutrient limitation. The importance of R auto rather than GPP in controlling mature stand NPP implies that the production of older stands is not limited by carbohydrate availability. If true, this suggests production and carbon storage by older stands may not accelerate with elevated atmospheric CO 2 concentration.
Studies such as this are producing a wealth of information on the patterns of ecosystem function during succession. Process-based models of land-atmosphere exchange are needed to translate this information into a more quantitative understanding of the effects of past disturbance on the contemporary carbon cycle, and also into more reliable predictions of the future climate forcing associated with changes in disturbance regime. Some of the patterns we observed, and the processes we identified as especially important, are poorly represented in many models. We hope researchers who are developing and using process-based models will make use of these observations (see Supporting Information), both to test their models and to identify which processes require particular attention.