Carbon stocks on forestland of the United States, with emphasis on USDA Forest Service ownership

The U.S. Department of Agriculture Forest Service (USFS) manages one-fifth of the area of forestland in the United States. The Forest Service Roadmap for responding to climate change identified assessing and managing carbon stocks and change as a major element of its plan. This study presents methods and results of estimating current forest carbon stocks and change in the United States for public and private owners, consistent with the official 2010 U.S. greenhouse gas inventory, but with improved data sources for three states. Results are presented by National Forest System region, a major organizational management unit within the Forest Service, and by individual national forest. USFS forestland in the United States is estimated to contain an average of 192 Mg C/ha (mega grams carbon per hectare) on 60.4 million ha, for a total of 1 1,604 Tg C (teragrams C) in the year 2005. Privately-owned forestland averages 150 Mg C/ha on 173.8 million ha, with forestland of other public owners averaging 169 Mg C/ha on 43.1 million ha. In terms of change, private and USFS ownerships each sequester about a net 150 Tg C O2/yr, but an additional 92 Tg C O2/yr is stored in products from private harvests compared to about 3 Tg C O2/yr from harvest on USFS land. Emissions from other disturbances such as fires, as well as corresponding area estimates of disturbance are also important, but the needed datasets are not yet available. Recommendations are given for improving the estimates.

Older state-level estimates are available (e.g.,

Definitions and units
Forestland as defined here is "Land at least 36.6 meters wide and 0.405 hectare in size with at least 10% cover (or equivalent stocking) by live trees of any size, including land that formerly had such tree cover and that wiIi be naturally or artificially regenerated" (Smith et a1. 2009

Forest Inventory and Analysis survey
The FIA program is the primary source for information about the extent, condition, status and trends of forest resources across all owner ships in the United States (Smith 2002). FIA applies a nationally consistent sampling protocol which began implementation in the late 1990s covering all forestland in the nation following an annualized design (Bechtold and Patterson 2005). An annualized design means a statistically valid subset of plots is measured every year in a state. Several years of data may be required to include all measurements on all forested plots within a state. The complete set of plot data provides for a greater level of precision geographically, but the aggregated data lose temporal specificity. On each permanent inventory plot, field crews collect data on more than 300 variables, including land ownership, forest type, tree species, tree size, tree condition, and other site attributes (e.g., slope, aspect, disturbance, land use) (Smith 2002;Woudenberg et al., in press). Plot intensity for measurements is approximately one plot for every 2,400 ha of land (1 30,000 forested plots nationally). These data are compiled, and are publicly available via the Internet (USDA FS 2010c).
Th e FIA data are collected on all ownerships in the 48 conterminous states, coastal Alaska, and territories. This stu dy does not include forestland in interior Alaska and Hawaii because FIA plot data have either not been collected or are not yet available. Puerto Rico data were not available at the time of this analysis. FIA plot data available before the annualized implementation were surveyed periodically, and may only be available at the plot level rather than the tree level. To calculate change, an approach must include a way to use these older data such th at they are comparable to the newer data.

Approach
The current FIA survey was not designed nor was it funded as a carbon inventory. Our approach is based on data taken from FIA surveys (Bechtold and Patterson 2005), but augmented by a set of basic models which are either ecologically process-based or statistical carbon conversion models (USEPA 2010;Heath et al., in press).  describes the methods used for estimating the density of carbon component pools, as well as the approach for calculating carbon change. In general, our approach is to calculate carbon stocks derived from the augmented FIA plot data by multiply ing area estimates by estimates of carbon density for that area. For example, estimates of carbon per hectare for the permanent inventory plots labeled as NFS ownership are multiplied by the appropriate expansion factors, and then summed over the total area of interest, such as national forest. Privately-owned land occasionally occurs within national forest boundaries; an FIA plot on private lands is labeled as privately-owned and is summed in the private ownership. Change in carbon (also called net sequestration) is calculat ed as the difference between consecutive stocks (each from a specifi c inventory), which is then divided by the number of years in the period between the stocks. This approach provides a net annual difference and is known as the stock change approach.
We used procedures from the computer application of , although we duplicated the code in SAS (SAS Institute 2003) to produce consistent estimates by ownership for NFS regions and for individual national forests. An additional step was included to review the data for consistency in terms of ownership and national forest deSignation. About 0.1% of the USFS field plots did not include a valid national forest design ation, but these were aSSigned based on state or count y codes. Methods and data sources are the same as those in USEPA (2010) with one exception. Data from the Integrated Database (IDB, Waddell and Hiserote 2005) were used for the older forest inventories for Califor nia, Oregon, and Washington in place of the corresponding data used for those states as identified in USEPA (2010). Previously, we had focused on using national-level datasets, but we recently recognized that the older data in the IDB were more consistent with the current annual ized data for these states, which is a crucial consideration for the inventory-based methods used for change (Smith et a1. 2010). Recent GHG reporting (USEPA 2010, and similar previous reports) included notable differences in forest land between past and current inventories for Californ ia, although analyses could not attribute the differences to any specific cause. Incorporat ing data from the IDB into th e GHG inventory removed this apparent discontinuity. We applied additional updates to th e publicly available IDB on parts of 63 plots in eastern Oregon that were predominantly the juniper forest type because guidelines for classifying these plots had changed over the last 12 years. The modification made the older data more comparable with current inventories in terms of the basis for determining forestland.
We do not include the soil pool when presenting carbon change because changes in the land base can result in transfers of large amounts of soil carbon to other land use which will appear to be losses to or gains from th e atmosphere. Thus, we use and report the term nonsoil carbon which includes all pools (live tree and standing dead tree, down dead wood, understory, and forest floor) except soil. We recognize that soil carbon on forestland remain ing forestland may be emitting or sequestering GHGs, but this stu dy assumes no change in that pool. We emphasize that both forestland area change and carbon denSity (carbon per area) change can affect total carbon . That is, an increase in forestland area will result in increased carbon sequestration if the average carbon density is not declining. An increase in carbon density will result in increased carbon sequestration even if area of forestland is constant. A decrease in forestland area with an increase in carbon density can result in an increase or decrease in carbon sequestration, dependin g on the amount of change in each factor.

Carbon in harvested wood products
Carbon removed from forests as harvested wood can also remain stored rather than retu rn ing to the atmosphere for a long time, depending on the mix of wood products produced or burned as a substitute for fossil fuels. Carbon in ECOSPHERE .:. wv\'w.esajournals.org 5 HEATH ET AL.
HWP continues to provide carbon benefits, which can be an appreciable part of the overall forest carbon budget (Heath et al., in press). The net annual contribution to the total forest carbon budget depends on harvest allocation to prod uct, life-span, and methods of disposal (Skog 2008). Analyses can also be performed to determine the carbon value chain including accounting for emissions in manufacturing  to be independent for purposes of combining the simulated uncertainties. The same process was followed for other ownerships or regional totals. These quantities do not account for all uncer tainties. For example, the U.S. GHG inventories require a base year of 1990; inventory data prior to about the year 2000 were collected under a periodic inventory system, and in some states may have not included the entire forestland base now being surveyed. Although we have made comparisons and adjustments between these datasets to reduce error (e.g., such as for the state of Oregon with the change to and adjust ments to the lOB), there may be other area-based mismatches, as well as additional uncertainties.

Forest carbon stocks and uncertainties
Relevant u.s. carbon statistics include average year of measurement, forestland areas, average carbon stock per hectare (carbon density), and total carbon and uncertainties estimated from the most recen t PIA inventory for each ownership ( Table 1). The carbon stocks (and their corre sponding forestland areas) are based on data f rom different survey years, but the mean survey year is 2005. That is, 2005 represents the overall average year of data collected by field crews over the large number of permanent inventory plots maintained by PIA. USFS forestland features greater carbon density, on average 28% more per forested hectare, than that of private land. Results further indicate t hat the range of carbon density is also notably greater: 51 4 Mg C/ha compared to 326 Mg C/ha on private land at their respective 97.5 percentile values, with the value for other public ownership in the middle (434 Mg C/ha). The values on the low end of this interval (2.5 percentile) are about the same for all ECOSPHERE . : . www.esajoumals.org 6 ownerships, about 55 Mg C/ha. Within each region, Forest Service forestland features greater carbon density than other own erships ( Fig. 2) with the exception of other public ownership being greater in the Pacific Northwest region, and other public and USFS carbon densities being similar in the Southern region. In spite of differences in magnitude, the pattern of carbon density arranged by largest to smallest by region within each ownership is quite similar. Carbon densities in the Alaska and Pacific Northwest regions rank highly, with the largest depending on owner, followed by carbon densi ties of the Eastern and Pacific Southwest regions. The Southwestern and Intermountain regions exhibit the least carbon stock density in all ownerships, respectively. The order varies in th e remaining regions of intermediate values, but these regions have similar magnitu des.
Th ese similar patterns across regions indicate the importance of regional effects such as soil, forest type, and underlying climatic drivers, on carbon stocks. Land use history can also affect broad regions. In the Eastern region, for instance, national forests were established on cutover land, whereas in the West, many areas were inacces sible and the forests relatively unused when they were designated as national forests. Thus, al though the land use history of both areas is quite different, intra-regional differences are minor. Carbon stocks of the coastal Pacific Northwest and coastal Alaska regions occur in areas of mostly publicly owned land, with tree species large at maturity, low decay and disturbance rates, and a history of limited deforestation and active management so large carbon stock densi ties are expected. In the Southwest region, the less productive growing conditions with greater likelihood of disturbance will generally feature lower forest carbon stocks on average.  In contrast to carbon densities, total forest carbon is 2.2 times greater (26,058 Tg C com pared to 11,604 Tg C; Table 1) for privately owned land, largely because of the alm ost three fold difference in forestland area (173.8 Mha (million hectares) private compared to 60.4 Mha USFS; Table 1). At the national level, about 63%, 22%, and 15% of forestland area ( Table 1) is in private, USFS, and other public ownership. There are large regional differences in ownership patterns, with notably more area of forestland in private ownership in the Eastern and Southern regions (Fig. 3), and least in Alaska (this is a survey of only coastal Alaska), and in the Intermountain region. If all forestland in Alaska were surveyed, there would be substantially more forestland area in private and other public ownerships.
Within USFS forestland only (Table 2), the Pacific Northwest region has the largest area of forestland (9. 1 Mha), followed closely by the Intermountain and Northern region, with Alaska the least (4.4 M h a). The carbon stocks (and th eir corresponding forestland areas) from different states are likely based on data from different survey years, but the mean survey year of most regions is similar to the mean for all USFS land, 2004.8, which we round up for this discussion to year 2005. That is, 2005 represents the overall average of data collected by field crews over the large number of permanent inventory plots maintained by PIA The exception to similar year of data collection is the Southwestern region with mean survey year of 2001 (rounded up from 2000.8). ConSidering the ecological conditions in the Southwest, the difference in results due to t he four-year average lag time is likely minor. In ter m s of uncertainties, the percent uncertainty ranges from :::: 1% for all USFS forestland up to 6% for USFS forestland in one region only.    carbon density for all pools averaging 317.1 Mg C/ha, whereas the Southwest and Intermountain regions have the least carbon densities, at 100.6 Mg Ciha and 135.3 Mg C/ha, respectively (Fig. 4). The greatest percentage of aboveground live biomass carbon is in the Southern region (47%), and lowest (30�J{)) in the Eastern region. The Eastern region has the highest relative soil carbon (51 %), followed by the Southern region (309(.), with a number of regions in the western United States in the 20-30% range. The relatively high proportion of forest carbon in forest floor in the Southwest region is thought to be due to the use of regional models for dead wood and forest fIoor pools for hardwood woodland forest types.
Within most regions (Fig. 5), forest carbon stock densities from individual national forests are relatively similar (e.g., Southern), with distinct patterns emerging in others. (See Appen dix B for carbon stock statistics including uncertainties for USFS forestland by individual national forest.) For instance, as might be expected, the carbon densities on the west side of the Cascades in Oregon and Washington are ECOSPHERE .:. www.esajouma!s.org 9 large due to the forest types, older forests, and relatively lush groV\Ting conditions, but on the eastern side with less favorable growing condi tions, carbon densities are relatively smaller. ll1e Pacific SouthV'lest region appears to show the greatest distinctions between forests within a region. The highest carbon densities per national forest are in the Pacific Northwest and Pacific Southwest regions, and the least in the semi-arid areas in the Intermountain and Soutln vest regions. Some forested plots fall within national grasslands or other USFS administered lands, and these are induded (as additional USFS areas in Appendix B.) National forest units are not randomly located across the landscape (Fig. 5). For example, the forests are bunched together in much of the West, in mountainous terrain where forests are more likely to occur or where land had not yet been settled upon before establishment of the national forests. In the Southern region, only 5% of the forestland is in USFS ownership, with 88% in private ownership varying from highly produc tive forestland intenSively managed for timber  production, to areas of woodlands in west Texas managed predominantly for grazing. Given this diversity of forest ecosystems, climate, produc tivity, ownership patterns and local preferences, effective, preferred management activities to increase carbon benefits will likely need to differ regionally if not by individual forest.

Net CO2 change and carbon in harvested wood products
Over the period 2000-2008, private and USFS forests sequester about 30% of total average annual nonsoil net CO:u with other public forestland accOlmting for 38% (Table 3). Most of the statistically significant net sequestration on NFS land is occurring in the Pacific Northwest and Southern regions, with net sequestration on other public and privately owned forestland higher in the Eastern and Southern regions (Fig.  6). Error bars of 95% confidence indicate relative large uncertainties with estimates for a number of the regions not significantly different from zero. Change is not calculated for forestland units smaller than regions because the carbon changes on smaller areas will likely not be signifi cantly different from zero.
The increase on other public forestland is due in large part to the estimated increase in forestland (0.45 million ha/yr) over this period. Additi onal data exploration (results not shown) did not identify specific regions of th e United States or unusual circumstances for this increase. USFS forest area also increased although the rate of increase was almost one-quarter of that of Januar y 2011 .:. Volume 2(1) .:. Article 6 HEATH E1' AL.   other public land. Some of this increase may be available for analysis of trends. due to definitional changes in the FIA survey Net nonsoi! change over 2000-2008 of -481 Tg over this period, or an artifact of the change from C02/yr (Table 3) is about 8S·;: lower than the the periodic to the annualized survey emphasizcorresponding USEPA (2010) 9-year average of ing the need to have reconciled FIA datasets -522 Tg CO2/yr. A minor part of this difference is the effect of dis aggregating the stock-change calculations beyond the structure defined in  to include the three owner ships. However, most of the difference from the results of the USEPA (2010) report is the effect of using data from the IDB (Waddell and Hiserote 2005) for the Pacific Coast states.
Beyond the forest boundary, additional carbon .continues to be stored in HVVP, with notable amounts attributed to harvest on privately owned land (Table 3; total carbon sequestered in forests and stored in HWP on average is estimated by summing columns 1 and 3). Products from harvests on private land continue to store an additional 62% of the net carbon sequestration on private forestland, whereas the increase is 3% at most on publicly owned land. Including the continued storage in HWP results in private forestland (and their harvested wood products) contributing to 41% of total forest sector carbon sequestration, with USPS at 26% and other public dropping to 33%. Although carbon in HWP from USFS land is minor, considering this pool is important in the context of landscape-scale management because ceasing harvests in one large area often results in increasing harvests elsewhere, if demand for products remains the same.
Other fates of forest carbon can also be substantial. 'iNe do not present change estimates from carbon benefits from harvested carbon that was burned for energy as a substitute for fossil fuel which can be notable for some mvnerships. That is, trees harvested for this purpose have been subtracted from the amount in the forest, but we have not recognized that this loss may have positive benefits of substituting for fossil fuel emissions. Emissions of CO2 from forest wildfires and prescribed burning on average rival those from emissions from wood burned for energy, but we currently do not have these emissions partitioned by ownership, or by land cover (e.g., forestland or rangeland).

Uncertainty
The relative uncertainties for total forest carbon stocks are much larger for the individual national forests (usually in the range 8-25%) as compared to the regional uncertainties (2-6% ), especially those with a smaller area of forestland or small total carbon. The larger uncertainties are ECOSPHERE .:. www.esajournals.org 12 HEATH ET AL.
mainly due to the smaller sample size on smaller areas, but may also be due to lllcertainty of data sources. For ease of comparisons, we report tabular summaries of uncertainties ( Table 2; Appendix B tables) as though the bounds are symmetric, which would be unlikely. However, asymmetry is small, less than 2;:1.:, off of the mean for the largest percentages ( Table 2) and asym metry averages under 0.5% off of the mean for individual national forests in the Appendix B tables. By comparison, the percent uncertainty about estimates of net sequestration are relatively large. One aspect of this uncertainty is the sensitivity of small change between relatively large stocks. For example, an additional annual increment of stock equivalent to only 0.1% of current nonsoil carbon stock in the Pacific Southwest region (data not shown) would produce a response of a 33% increase in calculated stock-change (Fig. 6). A contributing factor to the large percentage difference in this example is that the change is relatively dose to zero, which further emphasizes the importance of consistent forest and carbon stock representation between successive invento ries when eXaminLl1g inventory trends.

Discussion of methodology and possible improvements
Forest carbon estimates based on augmented PIA data have long been considered the standard for landscape level and larger forestland (e.g., Pacala et al. 2001, Smith et al. 2006, USEPA GEG inventories, Climate Action Reserve 2010). Ad vantages of using FIA data are: it has a national level statistically sound design, the data are publicly available (with some exceptions related to precise location and specific owner), the data are collected in partnership with state forestry agencies and aU major forest components which relate to carbon are measured or sampled. However, the survey was not designed specifi cally for carbon estimation, so additional work is needed to ensure an efficient framework for carbon stocks and GHG changes. Furthel� sample precision !,vas designated for state-level report ing, thus, using these data to represent smaller areas such as individual national forests results in higher uncertainties. Consequently, even mod erate increases in carbon benefits from manage ment activities may not differ statistically from zero.
A number of near-term improvements could be made to the existing framework for use in future U.s. GHG inventories to reduce uncer tainties and align estimates more closely with measured data. These include: using recent measurements from a subset of the plots of non-live tree pools such as standing dead trees, down dead wood, as well as samples of forest Hoor carbon and soil organic carbon; using a more recent tree biomass equation approach based on regional net volume estimates (Heath et a1. 2008) for trees that was recently adopted in FIA's national publicly available database (USDA 2010[;Woudenberg et a1., in press); accounting for results from FIA field data recently available for the national forest in Puerto Rico; and delivering the information produced by the computer application CCT (Smith et a1. 2010) used in the U.s. GHG inventories via an online tool. The resulting well-documented online site could then automatically produce forest carbon stock and change estimates for areas chosen by users. One challenge in these improvements is that the carbon changes for the U.S. GHG inventories are required to begin with 1990 carbon change, and older surveys generally do not include non tree measurements. It is crucial that carbon estimates for these older surveys be derived to be consistent with newer data. Furthermore, some of the older data are only available at the plot-level, so biomass carbon estimates for the older surveys are also needed that are compara ble with the newer tree-level data.
In the longer-term, as FIA plots continue to be remeasured, change estimates for most national forests in the future should become available at a precision that aHows for change to be detected with increased precision. Remeasured plots 'will allow for gross growth sequestration to be calculated, which is information that will revo lutionalize the use of FIA field plots in analysis. However, these data will still be limited tempo rally with remeasurements occurring 5 or 10 years apart, such that grow-th cannot be attribut able to a specific year. Coupling these growth measures with the use of geospatially-specific datasets (which are under development) will be especially powerful for explicitly accounting for disturbances. One annual dataset under devel opment by the Monitoring Long-Term Bum Although this stu dy focused on forestland, management activities on all lands are capable of emitting or sequestering GHGs, including non CO2 gases. For instance, wetlands or peatlands in particular can feature much higher carbon densities than forests. Monitoring all land covers and uses with activities that cause significant GHG emissions or sequestration should be considered. We have not discussed livestock emissions, but USFS land (and land under other ownerships) can include grazing. Significant livestock activity should be considered for base GHG emissions. Finally, because land manage ment can produce multiple environmental bene fits on the same land area, the process for making any inventory and monitoring improvements for carbon should also consider other important benefits.

CON CLUSION S
Forestland under USFS ownership features the largest average carbon density among owner ships, approximately 192 Mg C/ha in the year 2005, which is about 28% greater than that of private forestland. All carbon component pools are included: live and dead standing trees, down wood, forest floor and soiL In terms of total carbon stocks, however, private forests contain more carbon: 58%, 26% and 16S{, of the total forest carbon is in private, USPS, and other public ownership, reflecting the fact at the national level the majority ownership of area of U.s. forestland is private, about 63% compared to 22% and 15% for USPS and other public.
However, over the period 2000-2008, USFS and private lands have similar total net carbon sequestration in forests (not induding soil carbon effects), sequestering about -148 Tg C02/yr each, with 40% uncertainty. If carbon in HWP is also accounted for, private lands contribute to an additional -92 Tg CO2/yr sequestered compared to an additional -3 Tg CO2/yr from USFS lands. Other public ownerships indicate a larger total net sequestration of -185 Tg COjyr, heavily i nfluenced by an estimated notable increase in forest area over the period. We could not pinpoint any specific reason or particular region for this estimated forest area increase, so we look to futu re studies for more inform ation about thi s unexpected increase.
In spite of differences between ownerships, the pattern of carbon denSity arranged by largest to smallest by region within each ownership is quite sim ilar. This shows the importance of regional effects such as soil and forest type, and under lying climatic drivers. However, the pattern of total average annual sequestration by ownership by region differs because totals are influenced greatly by amount of forest area. The largest net sequestration rates are in the Eastern and Southern regions for private and other public ownerships, whereas the largest net rates in the Pacific Northwest followed by the Southern and then Rocky Mountain region for USFS owner ship. Due to the large uncertainties in change calculations, change for most of the other regions is not statistically different from zero.
The greatest gains in mitigation effects mini mize net carbon dioxide emissions to the atmosphere. Because forest carbon has carbon benefit effects beyond forestland boundaries, managing simply to maximize forestland carbon density is not necessarily the same as minimizing forest emissions to the atmosphere (or maximiz ing net sequestration) during the time frame of interest. That is, a strategy focusing on only increasing forestland carbon density on a limited area over time may produce limited carbon benefits compared to a more comprehensive strategy.
These carbon densities and forest areas by NFS region and individual national forest (Appendix B) could be used as preliminary base estimates for planning adaptation and mitigation activities. To consider the effects of specific silvicultural regimes, a tool such as the Forest Vegetation Simulator (Crookston and Dixon 2005) could be used to project plots into the future; carbon in forests and harvested wood products is an ECOSPHERE .:. www.esajournals.org 14 HEATH ET AL.
output (Hoover and Rebain 2008). A variety of management activities will be needed to increase carbon benefits in USFS lands across the matrix of ecological, physical, and social conditions, especially when management needs for adapta tion are a primary concern. However, demands and management choices on other ownerships should be a consideration in enhancing carbon benefits. A national-level forest fu tu ring an alysis that includes carbon outputs such as Heath and Birdsey (1993) and USEPA (2005), as well as climate change effects (Joyce et al. 1995), and global trade (Ince et al. 2007) would help ensure the maj or effects of large-scale processes are included.

ACKNOWLEDGMEN TS
We  Live trees with diameter at breast height (d.b.h., 1.37 m) of at least 2.5 em , including carbon mass of coarse roots (greater than 0.2 to 0.5 cm, published distinctions between fine and coarse roots are not always clear), stems, branches, and foliage. Standing dead trees with d.b.h. of at least 2.5 em, including carbon mass of coarse roots, stems, and branches. Live vegetation that includes the roots, stems, branches, and foliage of seedlings (trees less than 2.5 em d.b.h.), shrubs, and bushes. Woody material that includes logging residue and other coarse dead wood on the ground and larger than 7.5 cm in diameter, and stumps and coarse roots of stumps. Organic material on t he floor of the forest that includes fine woody debris up to 7.5 em in diameter, tree litter, humus, and fine roots in the organic forest floor layer above mineral soil. Belowground carbon without coarse roots but including fine roots and all other organic carbon not included in ot her pools, to a depth of 1 meter.

ApPEN DIX A
ApPENDIX B   No te: Estimates calculated using FIA data and methods consistent with U.s. greenhouse gas inventory estimates .
t Forested area of the Desert Range Experiment Station.
HEAT H ET AL.    Note: Estimates calculated using PIA data and methods consistent with U.s. greenhouse gas inventory estimates (Smith et al.

2010)
ApPENDIX C Notes: The 2.5 and 97.5 columns are the respective percentile value for a 95% confidence interval of uncertainty about the regional mean stock-dunge estimates from carbon conversion factors and sampling error. Negative values indicate more CO2 is sequestered by forests than is being emitted to the atmosphere.