Comparison of carbon-stock changes, eddy-covariance carbon fluxes and model estimates in coastal Douglas-fir stands in British Columbia

Forests are a large component of the global carbon (C) stocks, containing an estimated 1146 Pg C (Dixon et al. 1994). Forest processes, which may be influenced by forest management, can therefore have a large impact on the global C budget, either by storing or releasing C. It is therefore critical to understand forest C dynamics, including forest C stock components and transfer mechanisms in order to develop accurate forest C models such as the Carbon Budget Model of the Canadian Forest Sector (CBM-CFS3) (Kurz et al. 2009), 3PG (Landsberg and Waring 1997), and Ecosys (Grant et al. 2007), amongst others, as well as to inform forest management policy and for national and international reporting (Kurz et al. 2002). Factors affecting forest C dynamics include natural (e.g. fire, insect outbreaks) and anthropogenic disturbances from land use management and change (e.g. harvest, reforestation, deforestation) (Kurz et al. 2002) as well as weather (Morgenstern et al. 2004), fertilization (Jassal et al. 2010), and stand age and species composition, which can be related to disturbance history, site edaphic characteristics, or management practices (e.g. silvicuture) (Humphreys et al. 2006; Krishnan et al. 2009).

To measure the net exchange of C between land ecosystems and the atmosphere (net ecosystem exchange, NEE, with -NEE referred to as net ecosystem productivity, NEP), a global network of over 400 eddy-covariance (EC) flux stations has been established across a range of ecosystems, building an extensive data record of NEP, in some cases spanning up to two decades. The majority of these towers are located on sites not undergoing major disturbances so the fluxes measured reflect the interaction of weather, vegetation composition, stand age, and seasonal phenology. EC flux-towers use micrometeorological equipment to take measurements at the canopy scale of the exchanges of gasses including CO₂, water vapor, sensible heat, and some are equipped to measure other trace gases (Baldocchi 2008). The source area contributing to measurements made by the instruments mounted on the EC flux-tower, the flux-tower footprint, is variable in size and shape depending on height of measurement, surface roughness length, wind speed and direction, and atmospheric stability (Leclerc and Thurtell 1990; Schmid 2002), and proper interpretation of EC flux-tower based measurements depends on the flux-footprint over which the fluxes are sampled (Chen et al. 2008). Early estimates modelled flux-footprints as simple ovals. Recently, estimates of flux-footprint climatology may demonstrate more complex geometries and continuous probability density surfaces to quantify the upwind distribution of weighting factors over long time periods (Chen et al. 2009).

Forest measurements can be taken to quantify forest C stocks (Dixon et al. 1994; Clark and Brown 2001; NFI 2008), and by measuring forest C stocks at two times with a sufficient interval between measurements, the C stock change (ΔC) can be determined and used to estimate the cumulative net uptake of C from the atmosphere to the forest for the period of measurement (ΣNEP) (Clark and Brown 2001). Measurements are typically made of individual plants, for example, all trees within a plot are measured for diameter, species, and height, then allometric relationships are used to determine tree mass (Ter-Mikaelian 1997). Similarly, woody debris pieces are measured along transects using line intersect sampling and allometric relationships are applied to find biomass (Brown 1971). Other components are made by direct measurements, for example, understory vegetation can be sampled, dried, and weighed (Bailey et al. 1998). To measure soil C, field samples are typically dried and weighed for calculation of bulk density, and sent for laboratory analysis of C using dry combustion (Janzen 2005). Soil C measurements are important because the underground processes driving soil respiration form a large component of ecosystem gas exchanges, but these are less well-understood and measured compared to above ground stocks and processes (Ehman et al. 2002). Changes in soil C storage over time have implications as a source or sink for the global C cycle, and as an indicator of environmental function and health (Janzen 2005).

One application of forest measurement data to provide accounting of forest management actions and subsequently inform forest management decisions is as an input to C budget models such as CBM-CFS3. CBM-CFS3 is a forest C budget model that utilizes growth and yield information for biomass and generates explicit simulation of dead organic matter dynamics (Kurz et al. 2009). In contrast, forest process models use approaches based on the understanding of processes such as photosynthesis; these process-based models have more potential to model changing conditions because they do not rely on historic growth and yield and therefore are better suited to simulate forest conditions under global change situations (Landsberg and Waring 1997). CBM-CFS3 was designed to meet reporting requirements for forest management, provide policy support, and a function as a C budgeting tool for operating foresters, thus functioning at a range of spatial scales, and estimating records of C stocks, transfers between pools, and emissions (Kurz et al. 2002; Kurz et al. 2009).

Comparing EC flux-tower cumulative NEP (ΣNEP) and inventory ground plot measurements of C stock changes (ΔC) over the same period can help inform forest C processes for several reasons. First, ground plot measurements can serve as an independent validation of ΣNEP measurements made at EC flux-towers. Second, inventory plot data can provide more detailed information about the stand structural changes that may be driving variation stand-level EC C exchange. Third, broad stand inventory measurement datasets are available for a greater spatial extent than the global EC flux-tower network. For example, the Canadian National Forest Inventory (NFI) samples approximately 22,000 photo plots locations across Canada with detailed ground plots measured at over 1000 locations (Gillis et al. 2005). However, comparing inventory plot changes and EC flux-tower measurements requires comparisons to be made across different spatial and temporal scales. Therefore, such comparisons pose methodological challenges beyond individual inventory plot and EC flux-tower measurements.

Comparisons between EC tower measurements of ΣNEP and inventory plot measurements of ΔC stocks have been completed at a number of sites globally (Schulze et al. 2000; Granier et al. 2000; Law et al. 2001; Barford et al. 2001; Ehman et al. 2002; Curtis et al. 2002; Miller et al. 2004; Black et al. 2005; Gough et al. 2008; Kominami et al. 2008; Yashiro et al. 2010; Gielen et al. 2013; Babst et al. 2014), and at two sites of the Canadian Carbon Program (CCP), Boreal Ecosystems Research and Monitoring Sites (BERMS) station, located in the boreal forest of northern Saskatchewan (Theede 2007). Over a ten-year interval (1994 – 2004), Theede (2007) found a convergence of 15.6 ± 4.0 MgC ha⁻¹ 10⁻¹ years (inventory plots) and 18.2 ± 8.09 MgC ha⁻¹ 10⁻¹ years (EC tower) at the Old Aspen site and 5.8 ± 2.0 MgC ha⁻¹ 10⁻¹ years (inventory plots) and 6.9 ± 1.6 MgC ha⁻¹ 10⁻¹ years (EC tower) at the Old Jack Pine site. However, the homogeneous stand structure contributing to canopy-level EC flux measurements and large fetch in the boreal forest reduced the need for consideration of spatial vegetation structure and footprint distributions.

These previous studies have contributed considerably to the understanding of forest C dynamics; however, at sites with more complex vegetation structure and topography, such as the forests found in coastal British Columbia, the spatial distribution of vegetation structure and the footprint distribution are important considerations for accurate comparisons of EC flux-tower and inventory plot measurements (Schmid and Lloyd 1999). More complex ecosystems, therefore, require more complex models and methods developed for accurate comparisons of EC flux-tower measurements and inventory-based measurements. For example, a limited number of studies, including Ehman et al. (2002), Chen et al. (2009), Ferster et al. (2011), and Gielen et al. (2013) applied weighting by EC flux-tower footprint climatology to inventory-based measurements, where areas of the footprint that contribute more to tower-measured flux are weighted more heavily than other areas, resulting in an overall improvement when compared to simple flux footprint geometries.

In this paper, the difference between 2002 and 2006 inventory plot measurements of ecosystem C stocks was determined and the ΔC stocks compared to CBM-CFS3 modelled and EC-flux-tower estimates of ΣNEP at the CCP coastal British Columbia Flux Station. Through a combination of variable topography and a complex disturbance history (Trofymow et al. 2008), the EC flux-tower sites possess fine-scale spatial variation in forest structure, thus presenting a challenge to the interpretation of EC flux data (Schmid and Lloyd 1999; Göckede et al. 2004). To allow the comparison between measurements from the inventory plots, C budget models, and EC flux-towers, our approach consisted of four steps: first, we found the change in C based on forest inventory ground plot measurements made in 2002 and 2006 (four year period); second, we developed an additional approach that utilized litterfall data and published decay equations, for a second inventory-based estimate of C stocks change; third, we defined stand attributes at a fine spatial resolution, stratified the site based on stand attributes, and weighted the two inventory plot estimates; and fourth, calculated ΔC. These were compared with CBM-CFS3 model-based estimates of ΣNEP and EC flux-tower-measured ΣNEP for four years (2003, 2004, 2005 and 2006) in the same period. Finally, we evaluated how the forest age and conditions at the different sites accounted for convergence or lack of convergence in estimates for the different methods, and discussed the relative advantages and constraints of each measurement method.

Several approaches can be used to compare forest inventory data with EC ΣNEP. For example, several authors have used measurements or estimates of soil and detritus respiration in combination with inventory measurements of overstory productivity (Law et al. 2001; Ehman et al. 2002; Kolari et al. 2004; Black et al. 2005). In the present study, comparisons were based on measurements of ∆C which were higher than EC ΣNEP. Black et al. (2005) also found a divergence between inventory measurements of ∆C (including ∆C_S measurements) and EC ΣNEP, with ∆C higher than EC ΣNEP; however, they also calculated ΣNEP using measurements of heterotrophic respiration combined with biometric estimates of NPP, and found that the bias was not observed. They concluded that ∆C systematically overestimated NEP due to unaccounted decomposition processes and uncertainties in ∆C_S. At DF49, Jassal et al. (2010) measured heterotrophic soil respiration in 2007, which was a major portion of total soil respiration. Including measurements of soil heterotrophic respiration with net primary production estimates from forest inventory may improve convergence with EC-flux tower measurements. However, a limitation of taking measurements of soil respiration is that the measurements require long-term collection using special equipment and are not typically collected in traditional forest inventory.

Kolari et al. (2004) found that management practices, such as clearcut harvesting had a large impact on the soil C balance, with clearcut sites soil C sources, while non-disturbed sites had more stable soil C balances. In this study, a trend in soil C stocks, with the highest soil C stocks at the oldest site, and the lowest at the recently disturbed site, was consistent with the trend by seral stage reported by Kolari et al. (2004). For the two similar-aged juvenile stands, since there was no flux tower installed at HDF90 and CBM-CFS3 model runs results were not available at HDF88, a direct comparison was not possible. In addition to the differences in measurement methods, the differences in ΣNEP between the two juvenile stands could also be partially attributed to site conditions. Notably, HDF88 had more detrital C stocks, likely leading to larger releases of C due to decomposition. This was reflected in the more negative ΔC_D2 at HDF88 compared to HDF90.

For the forest inventory measurement of ∆C and its components ∆C_L, ∆C_D, and ∆C_S, differences in procedure, measurement error by field and lab crews, and sample design may introduce error into inventory measurements that may be larger than the magnitude of change in C stocks over short measurement intervals, such as the 4 year period used in this study. Overall, ΔC_D1 and ΔC_S were much larger than reported in previous studies (e.g. Law et al. 2001), and was much larger than expected given the measured litterfall inputs and estimated decomposition rates. The NFI (2008) calls for 10-year remeasurement intervals, which may be appropriate for trees and detritus, but may be too short for ΔC_S. In addition, direct use of forest inventory data relies on subjective classification of decay class that can differ from one-field-crew to the next, so utilizing decay and transfer algorithms may reduce the potential effect of these classifications. Developing site-specific allometric equations is labour intensive, destructive, and costly, and therefore uncommon in most studies; however, use of existing relationships can lead to errors due to differences in tree architecture, and wood density. Further, the use of inappropriate allometric equations can be a significant source of error in forest productivity studies (Clark and Brown 2001). The NFI data compilation procedure provided regionally applicable allometric equations; however, the errors in this large pool of equations have not yet been estimated and therefore is a limitation.

The measured changes in ΔC_S were much larger than was expected for a four year period. For example, Law et al. (2001) estimated soil sequestration of 0.7 – 0.8 MgC ha⁻¹ year⁻¹ in a Douglas-fir stand and Gielen et al. (2013) found no significant change over an eight-year interval in a Scots pine forest. In other studies by Ehman et al. (2002), Kolari et al. (2004), Miller et al. (2004), Ohtsuka et al. (2007), and Granier et al. (2008) ΔC_S was assumed to be zero over the measurement interval. The large value for ΔC_S in this study may be explained by several factors related to measurement methodology. First, mineral soil bulk densities in the 0–15 cm layer were assumed not to change and thus the BD values measured in 2002 were also used in 2006. If the bulk density decreased then the C_S change would be overestimated. Second, at the regenerating clearcut and juvenile stands, an increase in fine root mass from 2002 to 2006 included in the <2 mm soil fraction could also have accounted for some of the increase in soil C. Third, samples in 2002 were taken by 10–12 cm diameter hole excavations, while samples in 2006 were collected using a 2 cm diameter soil corer, which, due to compression at the opening of the sample corer, may be biased to soil from shallower depths where C content is likely higher. Therefore, in this study, measurements of ∆C_S were deemed unreliable, due to the large positive changes in ∆C_S, and assumed to be negligible over the four year period and thus not used for subsequent calculations of site level ∆C and further comparisons. In future work, efforts to reduce any chances of variation in the way the samples are collected and analysed may result in more reliable measurements of C stock changes. For example, collecting and processing the samples using identical methods at nearly the same locations, or collecting a much larger number of samples to capture a wider range of spatial variation could reduce the effect of spatial variability on measurements of ∆C_S.

Several authors have applied footprint analysis in an effort to correct for non-homogenous environments, and horizontal advection, for example due to daily upslope or downslope winds, which may introduce bias into the eddy-covariance measurements (Baldocchi 2008). For example, Ehman et al. (2002) used footprints to calculate an estimate of sensor bias given using a vegetation index, and this gave confidence that the sensor measurements were representative of the inventory measurements. Gielen et al. (2013) used a more complex footprint model applied to the processing of NEP data, were fluxes originating outside of the area where inventory measurements were sampled were removed from the NEP estimates. In this study, footprint weighting by ecological attributes and footprint probability density was evaluated where it facilitated comparison between inventory-based measurements of ecosystem ΔC stocks and tower ΣNEP by accounting for fine scale spatial variability in forest structure. The effect of K-MSN classification and footprint weighting had an effect on the estimated site-level values; however, this effect was smaller than differences due to measurement methodology. Notably, applying footprint weighting to the estimation of CBM-CFS3 ΣNEP at the regenerating clearcut increased convergence with EC tower ΣNEP; however, footprint weighting did not have the same effect for ΔC₂. The largest differences observed among methods were seen at the regenerating clearcut, where the inventory approach indicated the site was a much lower source than the ΣNEP methods. A possible cause is that the inventory methods did not properly account for C losses from decomposition of coarse roots from stumps and large woody debris at recently harvested sites, which can be a substantial source of respiration (Janisch et al. 2005). Therefore, if an estimate of stump coarse root decomposition were included in ΔCD₂ it may have increased convergence with the ΣNEP estimates. Additionally, the original sample locations were designed to capture the range of conditions in near proximity of the tower (based on interpretation of aerial photographs), and the relationship was similar when scaled up to the site level. Future studies may seek to derive accurate footprint estimates prior to establishing sample plots to ensure they are representative of the broader site conditions. Another more costly alternative is to establish a much greater number of plots in a systematic basis around the tower, to ensure the spatial variation within the tower footprint is adequately captured.

This research highlights how different measurements and approaches for C accounting include different constraints, and as a result, computations using different data sources may not converge. Each type of measurement has “different strengths and weaknesses, but the combination of multiple measurements and modeling has the potential for refining estimates of C stocks and fluxes” (Law et al. 2004). For example, eddy-covariance data are temporally detailed, and provide information about daily and seasonal processes. However, the data are less spatially extensive and detailed than the forest inventory data, for example, only three sites were available in this study each covering an area with diverse vegetation within the tower footprint. Measurement error was estimated to be relatively low, on the order of 0.9 Mg C ha⁻¹ year ⁻¹ (Morgenstern et al. 2004). In contrast, forest inventory data are more spatially extensive and include detailed measurements of forest attributes. The main constraint for forest inventory data is that they are less temporally detailed (for example, measurements were available at only two points in time), and the error in these measurements may be greater than the expected rate of change over the measurement interval. The inventory plots in this study sampled a broad range of vegetation conditions and provided insight into the representativeness of flux-tower footprints. Estimates of error of 1.2 to 1.3 Mg C ha⁻¹ were identified relating to spatial representativeness of the inventory plots within the tower footprints. Finally, C budget models are important for understanding landscape scale C processes. Since these models depend on numerous coefficients, algorithms and assumptions, it is valuable to compare model outputs with EC flux and inventory measurements to assess their performance and evaluate their behaviour. Considering all of these approaches together, the relative strengths and limitation of each approach can be taken into account in evaluating the suitability of each type of measurement at varying spatial and temporal scales.

The comparison of ΔC from inventory measurements with ΣNEP from CBM-CFS3 and EC flux-tower measurements demonstrated an agreement among the methods in trends across stand seral stage; however, due to differences in the measurement approaches, there was some divergence in the results. Convergence amongst methods was closest for the juvenile sites (HDF90 and HDF88), which were in transition from C source- to sink. At the most recently harvested site (HDF00), the EC flux ΣNEP and CBM CFS3 ΣNEP indicated that the site was a greater source of C over the time period than ∆C from inventory methods. At the near-rotation site the inventory ∆C₂ method, unweighted CBM CFS3 ΣNEP and EC flux ΣNEP also converged.

Each of the measurement methods had advantages and limitations. For example, while EC flux towers provide temporally rich data, they had limited spatial coverage. Obtaining ΔC using forest inventory measurements included several challenges such as the consistency of data collected over long time intervals, small changes in ΔC compared to uncertainty in the measurements, and insufficient measurement of processes such as decomposition of all detrital pools. Forest inventory data are available over broad areas; therefore methods that combine measurements with estimates of decomposition are needed to provide information over a regional scale. Footprint analysis using the spatial predictors from remote sensing and topography can give confidence that sample locations represent broader area site conditions. Considering the advantages and constraints of each type of measurement is helpful in choosing appropriate measurement methods at varying spatial and temporal scales.