Dynamic baselines with blended BAU scenarios ensure conservative mitigation estimates for improved forest management accounting

Improved forest management (IFM) is vital for climate change mitigation. Yet, precise carbon accounting to ensure credible attribution is a challenge. Current accounting protocols vary in their analytical approaches, leading to substantial differences in mitigation outcomes. We evaluated four common decision points in baseline design—modeled versus measured, static versus dynamic, matched versus random, and single versus blended business-as-usual (BAU) scenarios— across diverse regions and forest types in the eastern United States (US) using Forest Inventory and Analysis (FIA) and the Forest Vegetation Simulator (FVS). Dynamic baselines that integrate blended BAUs produced, on average, 63% more conservative estimates of net climate mitigation benefits than static, single BAU models, highlighting how baseline design choices can substantially influence results, with estimated mitigation varying by more than fivefold across accounting approaches, regions, and forest types. Stand characteristics, such as stand density, also affected outcomes. Since the true BAU counterfactual cannot be directly observed, adopting dynamic baselines with blended BAU scenarios can reduce the risk of overestimation, strengthen the credibility of carbon accounting, support more informed investment in IFM projects, and reinforce the integrity of carbon accounting for natural climate solutions.

To meet global climate goals, annual climate finance must increase fivefold across sectors (Buchner et al., 2023). In the forest sector alone, investments must rise by approximately $400 billion annually by midcentury to reach the approximately 6 GtCO 2 e mitigation target aligned with the Paris Agreement (Austin et al., 2020). Realizing this scale of finance, in the form of public funding, private investment, or carbon market transactions, hinges on credible measurements of mitigation outcomes. Improving the rigor and transparency of how mitigation is estimated and reported is therefore critical to unlocking investment, including in forest-based natural climate solutions (NCS). Accurate estimates of net greenhouse gas benefits—hereafter referred to as “mitigation”—are essential for decision-making, capital allocation, financial risk assessment, and ensuring integrity (Trencher et al., 2024).

Among NCS strategies, improved forest management (IFM) could contribute up to 28% of 2030 land-based climate commitments in the US and Canada while supporting biodiversity, resilience, and timber economies, especially through increased funding opportunities such as the Inflation Reduction Act and the Infrastructure Investment and Jobs Act (Roe et al., 2021; Fargione et al., 2018; Mo et al., 2023; Shanley et al., 2024; Favero et al., 2025). Unsustainable forestry practices, which deplete timber and biomass faster than they can regrow, often provide strong short-term economic incentives to land managers (Sass et al., 2022). However, emerging voluntary carbon markets offer land managers compensation for IFM practices that emphasize increased carbon storage and sequestration and confer timber sustainability benefits (Sass et al., 2022). Expanding financing mechanisms, such as carbon markets, corporate insetting, and public funding, has the potential to further incentivize and scale IFM (Seddon et al., 2020; Waring et al., 2023; Favero et al., 2024).

In mitigation accounting, a baseline represents the counterfactual trajectory of emissions or removals that would have occurred in the absence of an intervention. Because implementing NCS can be costly and risky, especially under increasing natural disturbance (Wu et al., 2026), strengthening baseline rigor and consistency in mitigation estimates is critical to ensure integrity and unlock investment (Ellis et al., 2024). Approaches to estimating baselines have faced sustained criticism, particularly when they fail to represent realistic business-as-usual (BAU) scenarios (Badgley et al., 2022; Coffield et al., 2022; Haya et al., 2023). Further, variations in how baselines are developed can lead to widely diverging estimates of mitigation, particularly in the IFM context. Drawing from existing IFM methodologies, we identify four key design choices in the development of a baseline BAU scenario: (1) whether carbon stocks are empirically measured for BAU, or inferred solely through models; (2) whether the BAU scenario is defined as a static, ex-ante projection or updated dynamically using ex-post information; (3) whether the BAU scenario assumes a single deterministic harvest outcome (e.g., 100% harvest) or incorporates multiple plausible harvest scenarios (hereafter, “blended” baselines); and (4) whether the BAU scenario is constructed through matching and weighting techniques, or by randomly selecting the BAU scenario from an eligible donor pool.

According to the first design choice, defining baselines based on models forecasting decades into the future is inherently constrained by historical data, current knowledge, and existing modeling capabilities. These forecasting models may therefore fail to capture all factors influencing future mitigation outcomes. These limitations can be particularly problematic in the context of family forest owners, whose management decisions are often heterogeneous and difficult to predict (Silver et al., 2015). This limitation also applies to protocols based on well-established growth and yield models such as the Forest Vegetation Simulator (FVS; Cook-Patton et al., 2021a; Anderson-Teixeira and Belair, 2022). By contrast, ex-post measurements benefit from incorporating realized outcomes but introduce their own set of challenges. Measurement-based approaches must carefully define and select comparison areas that robustly represent a BAU counterfactual. Failure to do so can introduce bias due to unobserved confounding factors that are not adequately represented in measured controls (Badgley et al., 2022).

According to the second design choice, dynamic baselines incorporate new observations as conditions change, allowing BAU estimates to adjust as conditions evolve and capturing factors such as timber market fluctuations, mill closures, and changing natural disturbance regimes (Haya et al., 2023) that affect outcomes but are unrelated to IFM interventions. In contrast, static ex ante baselines are more susceptible to adverse selection because project proponents can anticipate baseline trajectories and preferentially select project sites that will outperform BAU even in the absence of IFM interventions (Badgley et al., 2022). These incentives can lead to systematic over-crediting (Ayers et al., 2025). At the same time, static baselines enable greater certainty in crediting outcomes and thus greater participation by financially risk-averse landowners and funders (Ayers et al., 2025).

According to the third design choice, blended baseline scenarios that represent a distribution of plausible BAU outcomes across a region and forest type reduce bias from relying on a single deterministic BAU scenario, particularly one that assumes a 100% harvest likelihood (Andam et al., 2008). Even where harvesting plans are formally approved and documented, realized harvest outcomes often diverge from expectations (Silver et al., 2015). Baselines that assume a given harvest risk can overstate counterfactual emissions. In contrast, incorporating probabilistic or multi-scenario approaches better reflects uncertainty in future harvest behavior and may reduce bias associated with relying on a single deterministic BAU scenario.

According to the fourth design choice, BAU scenarios can be constructed by statistically matching project sites to similar, non-project sites rather than randomly selecting comparison areas. The goal is to estimate what would have happened in the absence of project implementation by comparing the project area to forests with similar biophysical characteristics, management history, and growing conditions. This approach is commonly referred to as nearest-neighbor matching and is widely used in quasi-experimental analyses to approximate a counterfactual outcome when a true experimental control is unavailable (Andam et al., 2008). By selecting comparison areas that are most similar to the project site, matching-based BAU approaches can better estimate the effect of an intervention on climate mitigation outcomes (Stuart, 2010; Shoch et al., 2025). Furthermore, objectivity can be enhanced through statistical matching. Statistical matching allows practitioners to select key variables for comparison while using data-driven estimates of carbon stock and flux values for the counterfactual (Ayers et al., 2025; Terra et al., 2023; Pan et al., 2026). Matching also reduces discretionary choices that could bias BAU estimates. In addition, finding a single “perfect” comparison to serve as a counterfactual is unlikely because harvesting events are rare. A blended, composite control that estimates change at the landscape scale will better represent the average BAU. Matching methods rely on underlying assumptions that there are no unobserved confounders that cause the matched controls to systematically differ in forest carbon outcomes from those who enroll.

Among all methodological choices, ensuring independent reproducibility of BAU over the project’s lifetime—particularly in terms of data sourcing and analytical choices—is critical. Reproducibility supports adaptive IFM management, enables third-party validation, strengthens funder due diligence, and fosters constructive criticism and continuous improvement in carbon markets and other outcomes-based transactions (Griscom et al., 2025; Swinfield et al., 2024).

In this study, we compared how these four design choices influenced mitigation estimates across a consistent set of IFM project cases spanning four regions across the eastern US. We compared five carbon accounting approaches: (I) a modeled data scenario with a static, single-instance BAU estimate; (II) a modeled data scenario with a static, blended BAU estimate; (III) a modeled data scenario with a dynamic, blended BAU estimate; (IV) a measured data scenario with a dynamic, blended BAU estimate; and (V) a measured data scenario with a dynamic, matched, blended BAU estimate (Figure 1, Supplementary Table S1). We systematically report on the impacts of four key methodological design choices—measured versus modeled data, static versus dynamic baselines, single versus blended scenarios, and matched versus randomly selected controls—to identify best practices for predicting potential mitigation prior to IFM investment, and for estimating the real impacts of IFM projects once they have been established.

Study area

Our study explores four distinct US regions where forest inventory data are spatially and temporally robust, there is a large concentration of private landowners, and there is a history of unsustainable forestry practices (Supplementary Table S6). Across the four regions of the Northeast, Central Appalachians, Southern Appalachians, and the Northwoods region in the Lake States, we selected U.S. Forest Service Forest Inventory and Analysis (FIA) Phase 2 sample plots (USDA, 2024) that were typed either as the Maple/Beech/Birch or Oak/Hickory forest group, because these are the most populous and widespread across the region. The FIA program is a nationwide forest inventory that consists of roughly one 0.07-ha plot per 2,400 ha (Westfall et al., 2022).

Developing a simulated project scenario

We used FIA plot remeasurement data to characterize the historical harvesting behaviors across the four regions over a historical lookback period spanning up to four measurement cycles across a 15-year period. We selected and cleaned the FIA data to ensure standardization across regions and forest type groups. To establish the project scenario of an extended rotation, we selected FIA plots that (1) were unharvested for the entire period; (2) had initial conditions collected in the last 10 years; (3) contained a single condition class; (4) were classified as a natural stand [i.e., the tree’s status code (live/dead) at previous measurement (PREV_STATUS_CD) is equal to 1, the tree’s status code (live/dead) at current measurement (STATUSCD) is equal to 2, and the agent of mortality is attributed to harvesting or other silvicultural activity (AGENTCD) is equal to 80 in the data table containing measurements of trees (i.e., the TREE table)]; (5) were of the private ownership class (Shoch et al., 2025); and (6) were located in the same ecoregion as the project scenarios (Cleland et al., 2007). Each project scenario across the four regions and two forest types was based on 20 randomly selected FIA plots. For the Southern Appalachians Maple/Beech/Birch group, we randomly selected only 10 FIA plots since the sample size was small (Supplementary Table S2). We calculated the total Mg CO 2 e for each project scenario.

Establishing the BAU scenarios for each approach

We identified harvests for the modeled scenarios (Approaches I–III) using the FIA treatment code (TRTCD1) from the condition-level inventory table (COND), which records management activities and silvicultural treatments observed on a plot between measurement periods. Harvests were also identified if there was no known disturbance on the plot and the total stand basal area of harvested trees (i.e., STATUSCD equal to 3 in the TREE table) decreased by more than 25% between measurement cycles (Luppold and Bumgardner, 2018; Russell et al., 2025). We calculated the total Mg CO 2 e for each approach and BAU scenario. Approaches I and II used fixed harvest likelihoods (i.e., the number of plots in a donor pool that were harvested divided by the total number of plots) and harvest intensities (i.e., the average removal rates of basal area per acre in plots that had a harvest), whereas Approaches III–IV used dynamic harvest likelihoods and intensities for each time step (Supplementary Figures S7, S8).

For Approach I, we used the Forest Vegetation Simulator (FVS), a widely used growth and yield model administered by the US Forest Service, with the FIA data selected above to model both the BAU and an extended rotation IFM practice, which had no allowable harvest over 15 years. FVS is a spatially non-explicit model of individual tree growth that can implement a wide variety of silvicultural prescriptions (Dixon, 2022). The FVS model uses tree lists from different stands to forecast different aspects of structure and composition (Crookston and Dixon, 2005) and has been shown to be a robust modeling platform for forest carbon modeling (Daigneault et al., 2022). We compared the model predictions for BAU growth and yield against the IFM scenarios to derive mitigation outcomes. Under this approach, the BAU scenario is defined using FVS with a fully deterministic harvest assumption. A harvest is assumed to occur with 100% likelihood over the 15 years. Harvest intensity was fixed using the maximum observed removal rate for each plot, averaged across the BAU, and applied uniformly whenever a harvest is triggered. Because a single, fixed harvest outcome defines the BAU—with no probabilistic variation, no distribution of possible harvest behaviors, and no updating over time—this approach produces a static, single-scenario baseline. Harvests in the BAU scenario were triggered in FVS once the stand reached 90 square feet of basal area per acre or higher, which is a typical basal area per acre for stands across these regions and forest type groups (USDA, 2024). We used the Northeast variant for the Northeast region, Central Appalachian variants for the Central Appalachian regions, the Lake States variant for the Northwoods, and the Southern variant for the Southern Appalachians region in FVS.

For Approach II, the methods were the same as those in Approach I except that we specified a harvest likelihood over the 15-year study period. To calculate this harvest likelihood, we used the most recent measurement cycle to identify how many plots were harvested and divided that number by the total number of plots, then multiplied the result by 15 years. Harvest likelihoods were less than 100%, ranging from 20 to 90% (Supplementary Tables S7, S8).

For Approach III, the methods were the same as those in Approaches I and II except that the harvest likelihood and removal rates were measured for each time step. For the harvest likelihood, we identified the number of plots that were harvested in each cycle (i.e., 0–5, 6–10, and 11–15) and divided that number by the total number of plots. For the removal rates, we took the average harvest intensity per cycle, meaning individual plots could have removals each measurement period. Removal rates ranged from 5 to 62% (Supplementary Tables S7, S8).

For Approach IV, we randomly selected 10 FIA plots from the eligible FIA plots determined in Establishing the project scenario. We then compared the Mg CO 2 e/ha/year from each project scenario with the 10 randomly selected comparison plots, with equally weighted contributions to the BAU. BAU plots were classified with the same ecoprovince and forest type group as the project plots but were not matched on any other characteristic.

For Approach V, we matched each project plot to the 10 most similar plots in the donor pool, using a k -nearest neighbor optimal matching approach with Mahalanobis distances. We then derived relative weights proportional to the inverse of the Mahalanobis distance that sum to 1. BAU sites had to be in the same ecoprovince and forest type group. Matching was based on site productivity (SITECLCD), elevation, stand age, relative density, and distance to the nearest road (RDDISTCD; Shoch et al., 2025). Matches were evaluated using the standardized mean difference and had to be less than 0.3. The SITECLCD and RDDISTCD variables were obtained directly from FIA plot information, while relative density was calculated from tree measurements on each FIA plot. Calculations for relative density were obtained for both the overstory (live trees greater than or equal to 5 in DBH) and understory (live trees between 1 and 5 in DBH). Overstory and understory relative densities were calculated using the method outlined in Ducey and Knapp (2010), using DBH and specific gravity for the species of interest.

We then compared the stock change for each project scenario to each BAU approach and scenario. Under each approach, the project scenario and the output metrics of Mg CO 2 e were the same. We incorporated harvested wood products into our carbon stock estimates across all approaches (Shoch et al., 2025). Estimates reflect gross additional emission reductions and removals to adequately capture the effect of decision-making on baseline design prior to accounting for leakage (i.e., the unintended increase in emissions in another region due to restrictions from an intervention), durability (i.e., long-term maintenance of mitigation benefits), and uncertainty (i.e., lack of predictability of supply and regulatory frameworks; Wells et al., 2023).

Data analysis

Determining key parameters of variations in carbon accounting approaches

We used a Bayesian linear model to estimate the mitigation resulting from the four carbon accounting approaches. We ran two models: (1) one assessing the four major design choices tested across the carbon accounting methodologies (i.e., single versus multiple scenarios, static versus dynamic, measured versus modeled, and unmatched versus matched); and (2) one conducting covariate analysis to disentangle the strongest stand characteristics determining mitigation (Mg CO 2 e/ha/year) differences.

To best compare the effects of each stand characteristic, we scaled all predictors to a z -score following the binary predictor approach (Gelman and Hill, 2007). The predictors included in the model were variables that may impact the FVS model run outputs and also the variables included in the matching protocol for the measured, blended, dynamic approach: average basal area removed during a harvest (sqft/acre), change in quadratic mean diameter after harvest (in), species diversity across the stand (Shannon’s Index), site class code, relative density of the overstory, relative density of the understory, and distance to road. Overstory and understory relative density were included separately to reflect the stand density of larger, established trees versus regeneration and smaller trees. We modeled the carbon accounting approach hierarchically, grouping factors on the intercept to capture model differences in methods for the stand characteristics model.

For both models, we fit a linear regression model to predict the differences in estimated mitigation using the brms package (Bürkner, 2017), version 2.3.1, in R (2017), version 3.3.1. We fit the models described below, where y i is the estimated mitigation (Mg CO 2 e/ha/year), α represents the intercept, β terms represent regression coefficients associated with each predictor variable, ϵ i is the residual error term, μ α is the overall mean intercept across methods, σ α is the random error unattributable to method, and σ 2 y is the residual variance:

Model equation: carbon accounting approaches

Model equation: stand characteristics

The α was modeled for the carbon accounting method, as follows:

We ran two chains of 3,000 iterations each. The first 2,000 iterations in each chain were treated as warm-up samples and discarded to allow the chains to converge to the posterior distribution. The remaining iterations from each chain were retained, resulting in a total of 2,000 posterior samples for estimating model parameters. Weakly informative priors were used throughout the analysis. Increasing the priors fivefold did not affect the model results. We evaluated our model performance based on R ̂ values that were close to one and by the effective sample size estimates. We additionally assessed chain convergence visually and with posterior predictive checks. We present the mean effect in natural units alongside a ± uncertainty estimate calculated from the 90% uncertainty interval of the posterior (taking the average of the 5 and 95% posterior estimates; the posteriors were generally symmetrical).

Evaluating mitigation across forest types and regions

Each carbon accounting approach produced different estimates of mitigation in Mg CO 2 e/ha/year for a 15-year period across the four regions and two forest types (Figure 2, Supplementary Figure S1). Approach I consistently estimated the highest mitigation, while the dynamic approaches (III–V) generally produced lower and more conservative estimates. The magnitude of these differences varied by region and forest type. Across all regions, mitigation estimates ranged from −0.03 Mg CO 2 e/ha/year in Oak/Hickory stands in the Northwoods under Approach IV to 5.07 Mg CO 2 e/ha/year in Maple/Beech/Birch stands in the Central Appalachians under Approach I. This represents more than a fivefold difference in estimated mitigation attributable to methodological choices alone. In general, modeled static approaches (Approaches I and II) produced greater estimates than dynamic approaches, whereas the measured dynamic approaches (Approaches IV and V) tended to converge on similar values.

Factors in carbon accounting approaches that contribute to variation in estimated mitigation

Bayesian hierarchical model results showed clear differences among approaches (Figure 3 and Supplementary Table S3). Approach I produced the highest estimates of mitigation using Bayesian hierarchical models (3.32 ± 0.95), followed by Approach II (1.74 ± 0.95), which were two to three times higher than estimates from all dynamic approaches. Approaches III–V produced similar estimates despite differences in their underlying baseline design methods, suggesting that dynamic models have a larger influence on estimated mitigation than whether BAU scenarios are measured or modeled.

Among the methodological design choices, the blended versus single BAU scenario emerged as the strongest driver of differences in mitigation in Mg CO 2 e/ha/year, with blended BAU scenarios having the largest impact (−1.68 ± 0.45). Dynamic models also influenced mitigation in Mg CO 2 e/ha/year (−0.78 ± 0.43), while there was no significant difference between modeled and measured approaches when both were dynamic and blended (0.33 ± 0.52) or between matched versus unmatched approaches when both were measured, dynamic, and blended (0.09 ± 0.42; Figure 4, Supplementary Table S4).

Effects of stand characteristics

Model results indicate that stand stocking metrics were the strongest predictors of mitigation differences among project scenarios (Supplementary Figure S2, Supplementary Table S5). Relative density of both the overstory and understory exhibited the largest positive effects among all stand-level covariates considered. Higher relative overstory and understory densities were associated with increased mitigation (0.83 ± 0.13 and 0.69 ± 0.16, respectively; Supplementary Figure S2, Supplementary Table S5). The remaining covariates exhibited relatively small effects with uncertainty intervals overlapping zero, suggesting they contributed less consistently to variation in the estimated mitigation.

Our results revealed substantial variation in mitigation estimates across the five baseline development approaches, with important implications for decision-making and impact evaluation in IFM implementation across the US. For example, static, single baselines estimated around 2.7 times higher additional Mg CO 2 e/ha/year than the dynamic, blended approaches on average (using averages for Approaches III–V from Supplementary Table S3 against Approach I) and nearly twice as high as the static, blended baseline (i.e., Approach I compared to Approach II in Supplementary Table S3). This magnitude of variability underscores the importance of selecting appropriate baseline-setting methods, as the choice of approach can significantly influence estimated outcomes and perceived project effectiveness.

These differences are not merely technical, as they also have direct consequences for policy and investment decisions. If carbon credit prices remained uniform across IFM methodologies (Forest Trends’ Ecosystem Marketplace, 2025), project proponents may lack incentives to adopt more robust methodologies. Notably, we found that estimates from Approach I yielded a much higher mitigation potential compared to the other approaches. Applied at scale, such bias could distort reported outcomes. For example, if the Inflation Reduction Act investments in climate-smart forestry were measured using Approach I, mitigation could be overstated by 22.5–27.5 million Mg CO 2 e through 2030, relative to dynamic, blended baselines. This vast discrepancy could have cascading effects on mitigation policy choices, allocation of capital, and overconfidence in the role of forest-based natural climate solutions.

A central insight from our analysis is the importance of blended BAU scenarios. The likelihood of a given parcel or hectare being harvested over a 15-year project period varies greatly with stand condition, region, and forest type. Timber harvests are rare events, with annual harvest rates generally around 2% in eastern US forests (Wear and Coulston, 2024) and lower on lands owned by non-industrial forest owners (Thompson et al., 2017). Growth rates also vary across management, often being greater in industrially managed forests (Joo et al., 2025; Huang et al., 2016). Single-scenario baselines fail to capture this heterogeneity and rarity, implicitly assuming uniform behavior across landscapes. By contrast, blended approaches incorporate the distribution of plausible management outcomes across different forest conditions and ownership contexts, thus representing a wider range of potential BAU trajectories. As a result, blended baselines reduce bias introduced by oversimplified assumptions and produce estimates that may be less sensitive to assumptions about a single harvest outcome.

Another main takeaway from our assessment is the reduced bias associated with dynamic baselines, which account for change over time rather than assuming fixed management behavior, climate, and market drivers. Dynamic baselines reflect fluctuations in timber prices, pests and diseases, climate, and mill capacity, updating expectations for BAU over time. Because IFM projects operate over decades, failing to account for these changes can dramatically inflate or suppress estimated additional mitigation. Our results reflect this logic: dynamic baselines produced more conservative estimates and may be less vulnerable to biases associated with fixed assumptions about future conditions.

Dynamic baselines also reduce the potential for Type I error and adverse selection in carbon accounting. Because baseline values are not known prior to the project start, project proponents cannot preferentially initiate a project in a location guaranteed to outperform the baseline. This reduces asymmetric information and incentivizes genuine mitigation rather than strategic enrollment. This additionally creates an environment for continuous improvement of the IFM activities, where projects can change course over time in response to observed performance relative to baseline expectations, improving overall effectiveness. Dynamic baselines not only improve accounting accuracy and reduce bias, but also permit adaptability in the project scenario, fostering continuous learning and better management outcomes.

Market bodies such as the Integrity Council for the Voluntary Carbon Market have begun recognizing the benefits of dynamic IFM baselines, selecting them as the first IFM protocols to receive the Core Carbon Principle label, which denotes achievement of higher-quality standards and confers higher prices (ICVCM, 2025). There is increasing interest in adopting dynamic baseline approaches in carbon accounting protocols to address externalities like changing timber regulatory conditions, mill capacity, pests and disease, and the impacts of climate change on forest stands (Sanders-DeMott et al., 2025; Haya et al., 2023; Cook-Patton et al., 2021b). Our results support this perspective, as dynamic baselines are designed to incorporate changing conditions and therefore may reduce sensitivity to external factors that are difficult to anticipate in static baseline designs. Our findings may become increasingly relevant as timber market conditions and climate change fluctuate unpredictably over the coming decades (Pan et al., 2026), in addition to natural disturbances that impact forest carbon stocks (Wu et al., 2026).

We observed no significant difference in mitigation between modeled and measured BAU scenarios when both were blended and dynamic. Given this similarity, the choice between modeled and measured approaches may be best guided by contextual factors, such as landowner type. Family forest owners often exhibit unpredictable management behavior even when intentions are known (Silver et al., 2015), making measured, blended dynamic baselines likely more suitable. Conversely, industrial owners tend to behave more predictably and respond consistently to market signals (Caputo et al., 2017), making them better suited to modeled approaches. As future research incorporates additional factors such as economic and social indicators, we expect that both modeled and measured estimates of mitigation will reduce uncertainty in mitigation estimates. However, the accuracy of baselines using growth and yield models such as FVS are likely to decline under climate change unless models are regularly updated to incorporate disturbances and shifting growth and survival responses (Woodall et al., 2025). The current repository of FIA data that meets our criteria does not allow researchers to look beyond a 15-year project retrospective period. However, modeling performed over longer timeframes could reduce variability, revealing more obvious differences between approaches.

Additionally, we did not find a substantial difference between matched and randomly selected BAU scenarios when both were dynamic, blended, and measured. Future research should continue to refine dynamic baseline approaches by investigating the impact of the chosen unit of matching (e.g., pixels, plots, parcels), developing standardized methods for identifying matching covariates, and refining statistical approaches that assess match quality.

Stand stocking was the strongest driver of higher mitigation across all carbon accounting approaches, and subsequently of potential credit yield in a carbon market context. These results make sense and highlight the importance of targeting forest stands and landowners most likely to engage in harvesting activities associated with the highest emissions while supporting sustainable wood production that maintains a financially viable approach to forest management. Landowner demographics, accessibility, and stumpage values might also factor into differences in quantified mitigation (Dhungel et al., 2026).

At the same time, these findings underscore the risks of “gaming” in baseline construction. When BAU scenarios are based on landscape averages and then project proponents select the most advantageous project locations, this can result in inflated mitigation while reducing the ability of third-party auditors to detect over-crediting. In contrast, incorporating blended, algorithmic dynamic baselines can limit discretionary choices and promote projects in the most cost-effective and climate-effective portion of the landscape.

A key limitation of all baseline approaches is that the true BAU counterfactual cannot be directly observed. Consequently, our analysis cannot determine which approach is most accurate. Instead, we evaluated how alternative methodological choices influence mitigation estimates and assessed the extent to which different approaches may reduce reliance on deterministic assumptions, improve reproducibility, and limit the potential for overestimation.

Overall, our findings emphasize the importance of careful methodological choice in IFM carbon accounting. Conservative, reproducible blended and dynamic approaches could improve the reliability of feasibility assessments and reduce subjectivity in modeling decisions, especially when multiple plots or parcels are evaluated in a priority-setting context. Incorporating additional factors, such as potential climate change effects and timber market projections, would further strengthen mitigation estimates by reducing uncertainty. While no single protocol will be suitable for all IFM practices and geographies, aligning accounting methods with the latest available science is critical to preventing gaming and directing investment toward credible projects with lower risk of overestimation. By evaluating alternative approaches to estimating mitigation from IFM projects and investments and highlighting those that are potentially prone to bias, our results strengthen the rigor and consistency of baseline approaches. This can provide investors and others evaluating NCS projects with clearer guidance and reduced risk when pursuing mitigation estimates. The use of high-quality, objective methods should unlock broader, more confident investment in IFM nationwide and help ensure that IFM projects deliver their promised climate benefits.