Comparisons to existing estimates
Direct comparisons to existing work are somewhat challenging because reports of forest carbon sequestration are often presented for the total forest area or total area of the region under consideration (recalling that our objective is to address the need for current forest carbon information, downscaled and in a format easily useable by managers and policymakers). Time is an additional consideration in the use of such summary data; forest inventories change as do the approaches to estimating and reporting forest carbon. Many published estimates [1, 7, 8] are based on data collected 10–15 years previously, and partially under the periodic inventory system. In the intervening time, not only is newer data available but methods have changed , sometimes with appreciable effects. For example, the change from reporting live tree carbon according to Jenkins et al.  to the current FIA component ratio method  resulted in a substantial decrease in estimated tree carbon stock values ; similarly, Hoover and Smith  note that three different approaches to estimating carbon in live tree biomass do not always produce equivalent estimates.
Using FIA data and remote sensing to examine post-disturbance growth, Williams et al.  determined net ecosystem productivity for regional and forest type group summaries (with several classifications identical to those presented here). The net change in live tree carbon we present has some similarity to net ecosystem productivity, but net ecosystem productivity also includes changes in non-live carbon pools, as well as belowground live carbon. Although slightly different quantities, carbon accumulation rates by type group and region were generally comparable (see Table 2 of Williams et al.  in comparison with Table 3 here). Smith et al.  also examined FIA-based forest carbon stock and change with a focus on federal forest lands (although all ownerships were summarized) using the FIA regions North, South, Rocky Mountain, and Pacific Coast. Average aboveground live tree carbon density on United States’ federal forest lands across all ownerships was highest in the Pacific Coast region, which includes Pacific Northwest West, Pacific Northwest East, and Pacific Southwest; this agrees with our results (recalling that Pacific Northwest West and Pacific Southwest had the highest carbon densities). They also found the lowest carbon densities in the Rocky Mountain region, which agrees with our findings. Earlier estimates of forest carbon stock in the National Forest System  summarized by National Forest region generally agree, with highest aboveground live tree carbon densities in forests of southeast coastal Alaska (not considered here), and the Pacific Northwest and Pacific Southwest regions of the National Forest System. Heath et al.  also report the lowest carbon densities in the Southwestern and Intermountain regions (generally corresponding to Rocky Mountain South although not an exact match). Harris et al.  used a combination of FIA data and remote sensing methods to examine forest carbon stock and change across the US with a focus on disturbance; mapped carbon densities agree with values reported here. Similarly, FIA based state level forest inventory reports that summarize carbon stocks [25, 26] present estimates consistent with those provided here, which is expected because of the common data sources.
State-level estimates of carbon density and change are available in several reports; as mentioned previously, these numbers require additional conversions in order to be compared to the present study. While recent US EPA estimates include state-level estimates of total forest carbon [3, 4], they do not include forest areas, which are necessary to develop these summaries. While state level forest area and carbon summaries are included in some previous EPA reports , it does not separately report live tree carbon; USDA  is based on the same data as US EPA  and does report comparable live tree carbon and forest area per state. However, stock change reported is based on the older stock-change approach, which does not separately identify or exclude effects of land-use change. Comparing five states from different regions (Maine, Wisconsin, Alabama, Colorado, and Washington) we find generally good agreement in carbon density between Additional file 1: Table S1 here and , which includes belowground live tree carbon. The largest differences (16.3 and 12.8 tC/ha) are in Washington and Colorado, respectively, while the smallest differences are in Alabama and Wisconsin (2.9 and 5.5 tC/ha).
Note that any estimates of forest carbon stocks or rates of change developed at a state or regional level are average values and should not be expected to represent conditions on any specific parcel, since many factors influence stand growth. However, in the absence of site specific data, reference values serve the function of providing reasonable values of carbon density and rate of change (in this case, for the aboveground portion of live trees) that may be expected in a given forest type group or region. Rates of average annual change presented here represent an approximate period from 2007 to 2016, plus or minus a few years, depending on the state. This interval should provide reasonable values of forest change over the near term, although rates are expected to change over time as influenced by many factors including weather, disturbance, and changes in age class distribution.
Forest remaining forest
Carbon accumulation rates in the literature are rarely reported on a per area basis and are commonly presented as total carbon mass for stock (or change) for large geographic regions [3, 4], which is not clearly linked to area or area change, leading to challenges when attempting to use the estimates as reference values. The principal reason for this is that these reports are oriented toward whole-country reporting of greenhouse gas inventories, and change in carbon density on forestland does not sum to total stock change for the country (the reporting focus). This is because the interval generally includes a portion of non-forestland becoming forest and forest becoming non-forest, see the discussion of accounting for greenhouse gas emissions and land use change in Chap. 6 of US EPA . The recent-year accumulation of remeasurement data on the permanent FIA annualized inventory plots as well as increased information on land use change as it affects forests and forest carbon is changing approaches to reporting [3, 4]; our approach to determining annual change is consistent with this direction in reporting.
Many of the past reports (as reviewed above) were unable to separate effects of land use. Therefore, prior to FIA implementing an annual inventory system and the availability of data from remeasured plots, changes in aboveground live tree carbon stocks were generally computed as the difference in carbon stock measurements at two points in time, which necessarily includes the effects of land transitioning into and out of forest. Effects of changing area of forest land on carbon change estimates can be low, but the difficulty is that they were unknown. For example, a comparison of mostly similar change summaries per state (Additional file 2: Table S2) relative to state summaries in USDA  shows generally good agreement for many states, with the largest difference approximately 0.25 tC/ha/yr, for both Washington and Colorado. No pattern of consistently higher or lower estimates of average annual change were noted between the two approaches. As an extension of this comparison between the remeasured-plot (here) versus stock-change (e.g., Smith et al. ) approaches, we calculated change following both methods for our dataset. The results are in Additional file 5: Table S5, which also shows variable levels of agreement with no pattern of extreme differences. The current estimates of carbon accumulation rates reported here are based on a subset of all available plots; those that have been remeasured and are majority one forest condition class. As such, change is a slightly different quantity than is required with whole-state or national-level reporting—here, change in live tree carbon on forested lands is the sum of survivor growth, in-growth, mortality, and removals.
While land-use change is an important component of forest carbon estimates, it is difficult to link land use change to quantities of current and expected future carbon stocks that are needed for applications such as state climate action plans or forest carbon project feasibility assessments. For this reason, when assessing change we focus on remeasured plots on forest land remaining in forest. One of our main objectives is to provide current estimates of aboveground live tree carbon density and average annual change in a format and at a level of aggregation useful for managers and policymakers seeking reference values to aid in understanding the current state of forest carbon or developing forest carbon plans.
Role of harvest and natural disturbance
While forest carbon density is increasing in most regions, in both Rocky Mountain regions carbon in live tree aboveground biomass is decreasing, and carbon densities in the Southeast and South Central regions are lower than those in the Northeast (Table 1). A possible explanation is harvesting and natural disturbance; several studies [22, 24, 30, 31] estimate the effects of harvest and natural disturbance on United States forest carbon stocks. Effects vary by region, with the largest impacts generally in the South and West. When plots with removals or disturbance activity during our study period are excluded, mean aboveground live tree carbon accumulation rates of the remaining records can increase considerably, with the smaller effect from disturbance. Williams et al.  identified this result as reflecting harvesting practices and regional climates with the greatest effects in the Southeast, South Central, and Pacific Northwest regions. This is evident in our results for the East, in harvest effects in South Central and Southeast regions (Fig. 3a); natural disturbance has a small effect. In the West, harvesting has a substantial impact in the Pacific Northwest West region (and is noticeable in the Great Plains), but natural disturbance appears to have decreased live tree carbon accumulation rates in all regions, most clearly in the Rocky Mountain North, Rocky Mountain South, and Great Plains regions (Fig. 3b). Note that while harvesting generally results in higher growth rates when a stand is regenerated, this effect is not illustrated in Fig. 3, which simply reflects average effect on all forest at all stages, without disturbance or removal effects.
Scope of this analysis
The amount of live tree biomass on the landscape, and the change in that quantity, are a function of many variables, including site productivity, species mix, age class distribution, and ownership. In this work, our primary objective is to provide estimates in a format easily used by those requiring current information on the state of aboveground live tree carbon. Past work  has examined the role of site productivity using the productivity classification in the FIA database, and future work will focus on the effect of age class distribution, which has pronounced effects on estimates of live tree carbon stock and change. In order to maintain a sufficient number of plots in each classification, especially at the state level, estimates for forest type by age class are not feasible using the current dataset.
Note that we are not making region-to-region comparisons at the scale of the summaries provided here because forest biomes, climate, and land use vary across the country. Quantifying or testing differences among regions is of limited use without reducing an analysis to a relatively limited number of factors. Even reducing summaries to forest type group has limited value as a basis for comparing across regions because most groups summarize several forest types, which can be regionally specific. For example, the white/red/jack pine groups in Northeast and Northern Lake States are predominantly different types; in the Northeast 82% of stands are Eastern white pine or Eastern hemlock stands, while in Northern Lake States these pines are 72% jack or red pine. So, largely different pine forests are represented in the two regions. Regional differences in type groups such as seen in Tables 2 and 3 have been reported by Smith et al.  and Williams et al. .
These estimates are developed from current FIA data and do not include a modeling component but reflect current live tree carbon densities and average annual change in forests including the effects of management, natural disturbance, and harvesting. Our choices to limit forest carbon estimates to aboveground live tree carbon, include only forest-remaining-forest (factoring out land use change), and set political bounds rather than ecological divisions of the land base are all aimed at providing a more easily used reference for state level analysis/planning. The data summarized here are essentially the same as what goes into the whole-U.S. forest carbon reporting in US EPA  except that scale and the carbon pool focus are narrowed considerably. It is useful to note that while the summarized data are derived from the same source, the values here are not readily summed back to the whole-country reported values; this is primarily because the net change we present is strictly limited to forest-remaining-forest change over generally longer intervals.