Western Projection Model

In the western regions (Pacific Coast and Rocky Mountain) where forest sampling is less intense and repeated measures of forest plots are not available, we modeled carbon flux using a stage class model of forest population growth. We quantified and predicted changes in forest carbon at the state level except in California, Oregon, and Washington where we separated the states into areas on the western and eastern sides of the Cascade mountain divide because of vast differences in forest productivity. For each state/substate, we queried the FIA inventory for all plot records and stratified them by stage class i and forest type j. Each stage was defined by a 5-year age class (from age 1–5 to greater than age 200) for each of 27 forest types. From the plot records, we estimated the initial area (ha) F _ij0 and carbon density (mg/ha) D _ij of each stage and forest type. The estimate of current forest carbon for state/substates is the sum of the products of forest area and carbon density across stages and forest types:

(1)$$C_0 = \mathop \sum \limits_{i\comma j} F_{ij0}D_{ij}$$

We projected forest carbon in each state/substate using a stage-class model of forest dynamics. Given parameters p _ij for the proportion of stage i forest type j that moves up one stage; d _ij for the proportion of stage i and type j that is subject to stand-replacing disturbance by fire, insects, weather, or harvest; and L _ijt for the area of land of stage i forest type j in period t that conversions to or from forest land use during projection interval t, the stage class model for each forest type is as follows:

(2)$$F_{1jt + 1} = F_{1jt}-F_{1jt}p_{1j} + \mathop \sum \limits_{i = 1}^n F_{ijt}d_{ij} + L_{1jt}$$

(3)$$F_{ijt + 1} = F_{ijt}-F_{ijt}\lpar {\,p_{ij} + d_{ij}} \rpar + F_{i-1jt}p_{i-1j} + L_{ijt}\; \; \; \; \; \; \; \; i = 2\comma \; \ldots\comma \; n-1$$

(4)$$F_{njt + 1} = F_{njt}-F_{njt}d_{ij} + F_{n-1jt}p_{n-1j} + L_{njt}$$

for periods t = 0,…,T – 1. Equation (1) updates the area in of the youngest stage by subtracting the proportion of area that moves up one stage, adding all of the disturbed area across the forest, and adding the net change in forest land use (positive for net additions, negative for net subtractions). Equation (2) subtracts the areas that either move up one stage or are subject to stand-replacing disturbance, adds the area that moves up from the younger stage, and adds net change in forest land use. Equation (3) accumulates area in the oldest stage while accounting for changes because of disturbance and land use of the oldest stage. Given these dynamics, the time-sequence of total forest carbon for the state/substate is the following:

(5)$$C_{t + 1} = \mathop \sum \limits_{ij} F_{ijt + 1}D_{ij}\; \; \; \; \; \; \; t = 0\comma \; \ldots\comma \; T-1$$

Simulations proceeded by applying equations (2) to (5) over the T-period horizon. Then, change in forest carbon during period t is as follows:

(6)$$\Delta C_{t + 1} = C_{t + 1}-C_t\; \; \; \; \; \; \; t = 0\comma \; \ldots\comma \; T-1$$

To apply the model, we made several assumptions about its parameter values. The average historical stand-replacing disturbance rate d _ij was obtained by dividing the area of forest of type j currently in the youngest stage, F _1j0, by total forest area, $<![CDATA[{\mathop \sum \nolimits_{ij}} F_{ij0}$. Assuming that the recent disturbance pattern leading to forest replacement carries into the future, we divided this average disturbance rate equally across stages. Note that d _ij includes all events that reset the forest to the youngest stage including fire, weather, insects, and harvesting. We also estimated the average annual disturbance rate (percent of forest area) for each type of disturbance (Table 3). For example 8.2% of the forest area in the Pacific Coast region had some amount of forest cutting. Then, the harvested forest carbon was transferred to a durable forest product carbon pool, and the change in forest carbon associated with the harvested portion of disturbance was adjusted to account for carbon storage in wood products. From a subset of plots with repeated observations, we observed that not all of the plots moved up one stage every 5 years. Therefore, we defined a stage transition rate p _ij of 0.85 for all stages. The values of L _ijt, t = 1, …, T are defined for a set of projection scenarios (described subsequently). For net gains in forest area, we assumed that new forest is added to the youngest stage; for losses, we removed forest area proportionately across stages. We assumed that carbon density associated with each land-use conversion L _ijtis limited to the soil organic carbon pool. Separate simulations were constructed for each of the 18 state/substate units, and results were summarized for Pacific Coast and Rocky Mountain regions.

Table 3.

Disturbance Levels by Region and Disturbance Type

Eastern Projection Model

The projection models for the two eastern regions (North and South in Figure 2) are based on estimates of land use and forest stage transitions obtained from remeasured, permanent inventory plots in those regions. For each region, remeasured plots were grouped according to seven land-use change and disturbance classes: undisturbed forest remaining as undisturbed forest, nonforest to forest conversion, forest to nonforest conversion, cut forest remaining as forest, forest remaining as forest disturbed by fire, forest remaining as forest disturbed by weather, and forest remaining as forest disturbed by insects and diseases. Plots were not grouped by forest type, so we drop the index j. For each land-use and disturbance class (indexed by k), the observed transitions of plot conditions were used to estimate forest stage transitions p _ik, forest disturbance proportions d _ik, carbon density change by age class δD _ik, and land conversions L _ik0.

The primary disturbance in the east was cutting, where 12.7% and 7.0% of the forest area in the South and North, respectively, experiences some level of forest cutting over the remeasurement period (Table 3). Given assumptions about future land conversions (see the next section), L _ikt, t = 1, …, T − 1, these parameters were used to update the forest age distribution, F _iktt = 0, …, T − 1, for each disturbance class k using equations (2)–(4), and then compute the carbon flux for each region:

(7)$$\Delta C_{t + 1} = \mathop \sum \limits_{ik} F_{ikt + 1}\delta D_{ik}\; \; \; \; \; \; \; t = 0\comma \; \ldots\comma \; T-1$$

Projection Scenarios

We projected carbon sequestration in each region under a baseline scenario and three alternative policy scenarios that are structured in an additive fashion (Table 4). The baseline scenario anticipates the elimination of net gains in forest area by 2025, followed by a slight decline in forest area through 2050, and represents assumptions developed for the 2016 U.S. biennial report (U.S. Department of Agriculture 2016).

Table 4.

Definition of the Baseline and Policy Scenarios Used to Estimate the Present Value of the Increase in Forest Carbon Sequestration

^a Land-use scenarios are from the 2016 U.S. biennial report by the U.S. Department of Agriculture (2016).

Note: See text for explanation of scenario components.

The first policy scenario—reduced development—assumes that land-use policies are implemented to reduce development intensities in future years, resulting in no net loss of forest area beginning in 2025. Current USDA projections anticipate increasing land development in response to a growing U.S. population and economy resulting in a net decline in forest area over the coming decades (e.g., USDA 2016), and so this policy alternative anticipates countering that trend.

The second policy scenario—afforestation and reforestation—assumes that landowner incentives can be implemented to encourage afforestation of private land in the eastern United States and funding is provided for reforestation of understocked federal forest land in the western United States, along with implementation of land-use policies to reduce development as described in the first policy scenario. In the eastern United States, the Conservation Reserve Program (CRP) currently funds the retirement of private marginal cropland to support conservation efforts, and the cap on program size has reached as high as 14.9 million hectares. In this second policy scenario, we assume in the eastern United States that a CRP-like program compensates landowners for establishing trees on 12.1 million hectares of marginal cropland between 2015 and 2020. Currently, in the western United States, 5.3 million hectares of forest land remains persistently understocked, of which 3.7 million hectares is federal forest land. We assumed that funding is provided for reforestation of 80% (3.0 million hectares) of the understocked federal forest land between 2015 and 2020.

The third policy scenario—wildfire mitigation—assumes a 10% reduction in the area of stand-replacing wildfires throughout the United States, along with the land-use and afforestation and reforestation policies. Wildfire causes significant releases of carbon and lateral transfer of carbon among pools followed by recapture of carbon by growing forests. Wildfire, therefore, is an important consideration in any overall carbon sequestration strategy.

Cobenefits

Our analysis estimated changes in carbon sequestration rates and their monetized values resulting from hypothetical USDA policies and did not address potential changes to other valued ecosystem services that would potentially result from changes in carbon sequestration or other alternative policy effects. These changes could involve either enhancement or impairment of water quality, habitat for terrestrial and aquatic species, and resource outputs, including timber and wood fiber, among many others. Regional analysis of forest management effects on carbon sequestration suggest that increased sequestration can be associated with both increases in some ecosystem services and decreases in others (e.g., Seidl et al. Reference Seidl, Rammer, Jager, Currie and Lexer2007; Schwenk et al. Reference Schwenk, Donovan, Keeton and Nunery2012; Kline et al. Reference Kline, Harmon, Spies, Morzillo, Pabst, McComb, Schnekenburger, Olsen, Csuti and Vogeler2016). In analysis from the United States, for example, Schwenk et al. (Reference Schwenk, Donovan, Keeton and Nunery2012) found in their Vermont study that forest management prescriptions that increased carbon sequestration also resulted in reduced timber harvest. Kline et al. (Reference Kline, Harmon, Spies, Morzillo, Pabst, McComb, Schnekenburger, Olsen, Csuti and Vogeler2016) found in western Oregon that forest management alternatives that increased carbon sequestration led to increases in species favoring late successional forest conditions, decreases in species favoring more open conditions, and highly variable responses for species that depend on particular spatial patterns of key ecological conditions, such as edge contrast involving tree heights.

Similarly, afforestation also can affect a range of other ecosystem services, both positively and negatively. For example, McKinley et al. (Reference McKinley, Ryan, Birdsey, Giardina, Harmon, Heath and Houghton2011) suggested that although afforestation of agricultural lands and grasslands generally improves water quality, it can reduce water quantity (e.g., streamflow) when trees uptake more water than crops or grass cover. Plantinga and Wu (Reference Plantinga and Wu2003) found that conversion of agricultural lands to forest via afforestation programs reduces negative externalities associated with agricultural land, such as soil erosion, and improves wildlife habitat roughly commensurate with the costs of administering such programs. Afforestation cobenefits, including species diversity, generally are enhanced where seedlings are established on lands that historically were tree covered, with the greatest improvements to wildlife habitat and biodiversity arising from plantings of native species (McKinley et al. Reference McKinley, Ryan, Birdsey, Giardina, Harmon, Heath and Houghton2011).

Although the potential cobenefits and costs associated with complementary and competitive ecosystem services undoubtedly should be an important consideration in evaluating the efficacy of proposed USDA efforts to increase carbon sequestration, we elected not to attempt to address such effects largely because of the complexity of doing so at a national scale. We feel that such effects are likely to be highly variable across regions and localities, as well as across the spatial and temporal scales of analysis. For these reasons, we suspect that evaluating potential cobenefits (and costs) associated with changes in other ecosystem services may be more feasible at the regional scales or at individual national forest levels. Alternatively, opportunities may exist to draw on existing national-level analysis of other ecosystem services to augment our analysis of carbon sequestration.

Policy Costs

Our analysis considered only rough estimates of the costs associated with the policies and program alternatives that define our scenarios. Ideally, a more detailed accounting of the costs would be included in any analysis of the efficacy of policy and program options, such as would be accomplished in a cost-benefit analysis (e.g., Mishan and Quah Reference Mishan and Quah2007). Such an analysis would be necessary to determine whether the net gains in carbon sequestration resulting from the policy and programmatic scenarios examined are worth the investment necessary to achieve those gains.

One thing to consider when thinking about likely policy costs is that our baseline scenario itself is the result of an array of policies, programs, and market forces that have affected changes in land use and forest cover in the several decades leading up to the present. These factors have exerted influence at a variety of spatial scales and via various administrative or jurisdictional authorities. For example, land use—and thus the amount of forest land—can be influenced by federal, state, and local policies, which all carry their own costs borne by the entities that enforce those policies. Similarly, how forests are managed can be influenced by local, regional, and international market forces. Although a full accounting of the costs and benefits of any given policy change to increase carbon sequestration necessarily would focus on the costs of implementing the policy and expected incremental gains in sequestration, it is important to remember that current levels of sequestration are at least partly because of past investments in various policies and programs that have, for example, incentivized landowners to retain land in forest and to manage it in a particular way.

Forest Carbon Projections

There are several uncertainties associated with our forest carbon sequestration projections that are common to all similar studies, including sample error, measurement error, modeling error, and error in the future state (Coulston, Wear, and Vose Reference Coulston, Wear and Vose2015). Because FIA data are sampled based, each estimated component (e.g., forest carbon sequestration by age class) has a standard error. Westfall and Patterson (Reference Westfall and Patterson2007) found that measurement error of changes in tree volume was approximately 4% of sampling error. Tree volume change is highly correlated with carbon sequestration in the live tree carbon pool. The combined uncertainty of historical sequestration estimates, developed using Monte Carlo analysis, is about ±17% (US EPA 2017). However, uncertainty approaches ±40% using error propagation techniques, suggesting that uncertainty is somewhat dependent on the assessment method. Our projected change in sequestration encompasses the previously mentioned uncertainties but has additional modeling uncertainty and error in the future state. Error in the future state includes error arising from, for example, unknowable future land-use changes, potential atmospheric CO₂ and N fertilization effects on sequestration rate by age class, and changes in temperature and precipitation patterns. Further, our projection approach relies on an age transition matrix arising from field observations of disturbance, cutting, and normal mortality rates. There can be significant temporal variability in the amount and types of forest cutting (e.g., clearcutting, partial cutting) and in the amount and severity of disturbance, which suggests that there could be significant variability in the age transition matrix.

Social Cost of Carbon

Discount rates (typically annual) make present and future benefits (or costs) fully comparable, because they take account of (a) human (including policy maker) impatience and (b) the effect of output and consumption growth over time. If discount rates are positive, future benefits are worth less than current benefits, and higher discount rates imply lower future values relative to present benefits. The choice of discount rates is therefore very important for our analysis.

Discount rate choice affects our results through two channels. First, it is a particularly important SCC parameter, because the SCC is essentially an estimated PV of the future damages of climate change at the time a ton of CO₂ is emitted. Second, we estimate the effect of carbon sequestration over the coming several decades. The rates chosen to discount values from the time of sequestration back to 2016 also have very important effects on our estimates. Analysts differ in their estimates of the impatience and output growth that underpin discount rates, and therefore, discount rates vary across studies. Because there is no universally accepted set of discount rates, analysts (e.g., Greenstone, Kopits, and Wolverton Reference Greenstone, Kopits and Wolverton2013) often conduct sensitivity analysis to illuminate the influence of the discount rate.

As discussed by many authors (e.g., Arrow et al. Reference Arrow, Cropper, Gollier, Groom, Heal, Newell and Nordhaus2014), uncertainties about the future affect discount rates and make them uncertain. As shown by Weitzman (Reference Weitzman2001) and many others, when key aspects of the future (e.g., output and consumption) are uncertain, lower discount rates should be applied to benefits and costs that occur further into the future. For example, Weitzman (Reference Weitzman2001) suggests that the immediate future (1–5 years) be discounted at 4% per year; 6 to 25 years, at 3%; 26 to 75 years, at 2%; 76 to 300 years, at 1%; and more than 300 years, at 0%. These findings are not incorporated into the interagency working group analysis, but as discussed in Arrow et al. (Reference Arrow, Cropper, Gollier, Groom, Heal, Newell and Nordhaus2014), the French and British governments apply lower discount rates to benefits and costs that occur further in the future, and revisions in this vein are recommended by the National Academies of Sciences, Engineering, and Medicine (2017).

An additional and particularly important type of uncertainty related to climate change is uncertainty regarding climate-induced catastrophe. Climate change is expected to create many types of damages in the United States, but how large will they be? How likely are extreme damages that significantly affect future welfare? There is, of course, significant uncertainty regarding such extreme effects, but hedge investments like carbon sequestration can reduce the chance of catastrophes.

Healthy forests often enhance and support ecosystem services (e.g., water quality and biodiversity) that are endangered by climate change and help mitigate extreme downside risk. As Weitzman (Reference Weitzman2013) discusses, if investments like carbon sequestration mitigate large downside risks, this also reduces the discount rate(s) that should be applied and increases the SCC. As possible in the three integrated assessment models used, the interagency working group included aspects of extreme risk in its SCC estimates. Fully incorporating risk and especially risk of catastrophe in such models is very challenging, however, particularly when analyzing forests, which likely reduce those risks.

Voluntary Incentives and Adverse Selection

Two of the scenarios include afforestation of 12.1 million hectares in the eastern United States, which would largely be achieved by providing incentives to private landowners. The USDA has five voluntary incentive programs, which account for more than 95% of USDA conservation spending (USDA 2014). When estimating benefits, the possibility of incentive-related adverse selection would need to be taken into account (Claassen, Cattaneo, and Johnson Reference Claassen, Cattaneo and Johnson2008).

First, it is possible in practice that carbon estimated to be sequestered by voluntary incentive programs may not be fully additional. The USDA is unable to observe what would have happened had a given incentive program not been implemented, and it is possible that some of the resulting gains in carbon would have been sequestered without the program. For example, landowners concerned about climate change may enroll in USDA conservation programs to get credit for steps they would have taken without such programs (Duke, Dundas, and Messer Reference Duke, Dundas and Messer2013). Lubowski, Plantinga, and Stavins (Reference Lubowski, Plantinga and Stavins2006) estimate that about 10% of land enrolled in the CRP between 1982 and 1997 would have been taken out of agricultural production anyway, because of market factors.

Second, slippage (or leakage) can also occur. Although steps are taken to avoid landowners “gaming” USDA incentive programs (Claassen, Cattaneo, and Johnson Reference Claassen, Cattaneo and Johnson2008), landowners may sometimes bring land into production that was previously unfarmed to compensate for land enrolled in a USDA voluntary conservation incentive program. Murray, McCarl, and Lee (Reference Murray, McCarl and Lee2004), for example, estimated that leakage from program effects can range from less than 10% to more than 90% depending on sequestration activity and where it occurs. Wu (Reference Wu2000) found that in the central United States, such slippage offset between 9% and 14% of erosion control benefits resulting from the CRP, for example. Lubowski, Plantinga, and Stavins (Reference Lubowski, Plantinga and Stavins2006) estimated that the CRP reduced unenrolled forested area by about 200,000 acres or approximately 15% of the measured impact on forests. Despite leakage effects, carbon-positive externalities also can accrue from USDA conservation policies and programs targeted toward other ecosystem services. For example, landowners interested in retiring land may participate in the Wetlands Reserve Enhancement Program, which will likely generate carbon benefits. Incorporating carbon-specific criteria in project selection may help moderate leakage while improving multibenefit returns on program investments. Accounting for leakage, as well as potential complementary benefits, would enable further refinement of our estimates of carbon sequestration policy effects but was beyond the scope of this particular study.