Voluntary markets for agricultural carbon credits are expanding, promoting climate-smart practices purported to increase soil carbon and reduce greenhouse gas emissions. To contribute effectively to climate mitigation, markets must deliver credits that meet international standards guaranteeing credits are additional, conservative, and equivalent to at least one tonne of CO₂. Yet protocols for quantifying credits make different assumptions, raising questions about whether protocols meet the “equivalency standard”. We test for equivalency using a common dataset of 4,988 US Midwestern corn-soybean fields, representing a carbon market project, to estimate credit issuances for adoption of no-till plus cover cropping practices. We find issuances, across the three major protocols being used for US croplands, differing for this common project by up to ~130,000 credits per year. Our “Protocol Intercomparison Project” reveals how quantitative evaluations can identify assumptions generating marked differences in crediting, thereby guiding research that informs protocol revisions to build confidence in mitigation and demonstrating paths forward for protocol harmonization efforts.

Soil Carbon Crediting Protocols

We included two versions of three prominent soil carbon protocols, Nori’s Cropland Methodology (v1.3, December 2021 and v1.6, January 2024), Climate Action Reserve’s [the Reserve] Soil Enrichment Protocol (https://carbonplan.org/research/offsets-db; v1.0, September 2020 and v.1.1, May 2022), and Verra’s VM0042: Methodology for Improved Agricultural Land Management (https://registry.verra.org/, v.1.0, October 2020 and v.2.0, May 2023). In 2024, after our analysis concluded, the Nori Registry and website platform shut down and stopped issuing new credits (https://en.wikipedia.org/wiki/Nori_(company)], and therefore their protocol versions that were once freely available for download on their website are no longer available online. Please contact authors directly for more information about the Nori protocols as they pertain to the decisions made in this research.

We evaluated ‘earlier’ and ‘current’ versions of each protocol to compare how prior methods used to issue credits compare to current methods. All three protocols issue soil carbon credits for croplands in the United States if growers convert from a current ‘business as usual’ cropping practices to a list of eligible crediting practices enumerated in each protocol. We estimated carbon credits arising from a synthetic dataset of fields for two 10-year project scenarios: 1) adopting of no-till management and 2) “stacked” practice adoption of no-till and winter cover crops. We calculated credits on the following protocol criteria: 1) determining practice additionality; 2) calculating raw credits using methods and greenhouse gas accounting boundaries; and 3) calculating credit deductions for uncertainty and buffer pool contributions. These features vary widely among protocols according to qualitative analysis. For all protocols, we assumed a practice switch and project start date of January 1, 2021.

Compiling a synthetic dataset

We initially sought empirically collected soil carbon data, field boundaries, and management history for this analysis, but such data are not publicly available at the scale for which we conducted our analysis. While these data streams exist with federal agencies and private entities, we were unable to access them. The absence of publicly available, high-resolution datasets limits researchers’ capacity to evaluate biogeochemical models (which, in turn, limits confidence in model reliability), hinders proper evaluation of effect sizes of management practices at scale, and leads to a lack of transparency in the methods underpinning soil carbon crediting.

Field Selection

In lieu of empirical data, we relied on public data sources to create a synthetic dataset that contained the required detailed input data for common biogeochemical soil models (DayCent, DNDC) including field-specific management, soil and weather data and was representative in size (number of fields) of larger soil carbon crediting projects in the US (see CarbonPlan OffsetDB, https://carbonplan.org/research/offsets-db). Our final synthetic dataset included 4,988 hypothetical fields in six states in the midwestern United States (Illinois, Indiana, Iowa, Missouri, Kansas, and Nebraska; Fig. 2). We extracted field boundaries using the U.S. Department of Agriculture (USDA) Cropland Data Layer, which is a crop-specific annual land cover map with 30-m spatial resolution. We extracted all corn-soybean fields in a six state area from the 2020 Cropland Data Layer in GeoTIFF format and converted them to a shape file using ESRI ArcGIS® software. To select fields within each state, we first excluded all fields less than the median field area for each state, since many polygons appeared to be fragments of fields or incorrectly identified. We then overlaid county boundaries from the R Statistical software maps package and randomly selected fields from each county. The final dataset contained 4,988 fields that represented ~0.65 % of fields within each county.

Field Management History

The DayCent model requires cropland management history to simulate soil organic carbon stock dynamics and nitrous oxide emissions. We associated a 10-year management history (2011-2020) with each of the 4,988 fields using multiple publicly-available datasets. To assign crop type, planting and harvest dates, and irrigation management to each field, we used the Land-use and Agricultural Management Practices web-Service, a web service which estimates management history by overlaying the Cropland Data Layer, U.S. Geological Survey irrigation maps, and the USDA-NRCS Land Management and Operation Database. As this analysis focuses on corn-soybean rotation systems, we omitted fields that had > 2 years of the 10 year rotation planted to crops other than corn, soybeans, sorghum, or fallow. To assign fertilizer rates, tillage type, and cover crop use, we used the USDA National Agricultural Statistics Service (NASS) survey data. We assigned fertilizer rates (kg ha ^-1) based on 2018-2020 NASS data for each crop (corn, soybean, sorghum) and state combination (exact year of data dependent on data availability). Because the date of fertilization application is not available, we assumed fertilizer application occurred once at planting. We randomly assigned cover crop adoption (winter cover crop or no cover crop) and tillage intensity [no-till, reduced-tillage, or full-tillage] to each field based on reported adoption rates for each county from the 2017 NASS Census of Agriculture data (https://www.nass.usda.gov/AgCensus/). We assumed that cover crop adoption and tillage intensity were consistent over the 10-year management history, while we allowed for annual variation in crop type, planting and harvest dates, and irrigation and fertilization rates.

Once fields were assigned a tillage regime, we had to specify the method and timing for each tillage event. We assigned tillage methods and dates for reduced- and full-tillage fields using general tillage regimes identified by LAMPS and published corn-soybean management in the U.S. Midwest (Dangl et al., 2022). We assumed that fields with full tillage used a moldboard plow 20 days before planting, a seed drill at planting, and a field cultivator five days before planting and five days after harvest. In fallow years, we assumed a moldboard plow was used on day-of-year (DOY) 130 and a field cultivator every 60 days after (DOY 190 and 250) for weed control. In reduced tillage fields, we assumed a field cultivator was used five days before planting, a seed drill was used for planting, and a tandem disk was used five days after harvest. In fallow years, we assumed a field cultivator was used at DOY 130 and disking was performed on DOY 190 and 250. In no-till fields, we assumed growers applied chemical weed control 10 days before planting and two days after harvest. We assumed crops were planted with a no-till drill, and, in fallow years, herbicides were used on DOY 130, 190 and 250.

For those fields assigned to cover cropping management history, we had to assign a cover crop type and planting and termination dates. We selected cereal rye (Secale cereale L.) as the cover crop and assumed rye was planted 7 days after cash crop harvest and terminated with herbicides 20 days before planting the following cash crop.

Finally, the DayCent model requires daily weather data and soil characteristics for each sample point. To assign climate data to each field, we used the 4-km gridded PRISM climate dataset to extract daily maximum and minimum temperatures and precipitation from 1981 to 2020 (https://prism.oregonstate.edu/). To assign field-specific soil texture, soil depth, soil bulk density, and soil pH values to each field, we used the USDA-NRCS Soil Survey Geographic Database (SSURGO; https://www.nrcs.usda.gov/resources/data-and-reports/soil-survey-geographic-database-ssurgo).

Determining Practice Additionality & Baseline Activity

We determined the total number of fields that met additionality criteria from each protocol based on protocol-specific Timing Tests and Common Practice Tests. All protocols required that enrolled fields adopt the project activity on or after a specific project start date (Timing Test). For our analysis, we assumed a project start date of January 1, 2021, and therefore we excluded any fields that were assigned no-till or cover crop in the synthetic dataset from the no-till or no-till and cover crop project scenarios, respectively.

To assess the impact of structural differences in Common Practice Test, we calculated practice adoption rate for each county and state included in our six state region. We used the 2017 Census of Agriculture data (https://www.nass.usda.gov/AgCensus/) and estimated single practice adoption rates for no-till and cover cropping as the proportion of acres reported using each practice, respectively, divided by the total cropland acres for that region. For the no-till practice scenario, we excluded fields as “non-additional” if the calculated state or county no-till adoption rate was higher than the specified protocol threshold. For the no-till and cover crop practice scenario, we followed protocol guidelines to estimate the weighted average adoption rate of both practices, and then excluded any fields where the weighted average adoption rate was higher than the protocol specified adoption rate for additionality.

After determining whether a project activity is additional, credit producers must then determine the most likely activity that would have taken place in the absence of the crediting payment. This is referred to as the ‘baseline’ or ‘business as usual activity’. All protocols in our analysis either explicitly or implicitly assume that the most likely ‘baseline’ activity would be a continuation of the pre-project management activities, which are demonstrated through management records for each field enrolled in a project. For our analysis, we used the 10-year management history for each field as the baseline activity in models to estimate baseline emissions.

Calculating Raw Credits

For all protocols, the number of raw credits is the difference between project activity emissions and baseline emissions. For our analysis, we measured baseline emissions arising from the continuation of the most recent management practices (2011-2020) in the synthetic dataset. We measured activity emissions from two activity scenarios: (1) the no-till project scenario where tillage events were substituted with herbicide applications and the seed drill was substituted with a no-till drill, and (2) the no-till by cover crop project scenario where the fallow winter between cash crops was substituted with a cover crop, following rules described to compile the synthetic dataset. We used management, weather, and soil data associated with each field in the synthetic dataset as inputs to simulate each field for each scenario.

For consistency, we opted to model credits using the DayCent biogeochemical model version used in the 2019 U.S. Greenhouse Gas Inventory and is available from the Soil Carbon Solutions Center Ecosystem modeling and data Consortium at Colorado State University (https://www.soilcarbonsolutionscenter.com/). We followed common model spin-up procedures to derive initial soil organic carbon and mineral soil nitrogen values for the year 2021. We simulated carbon dynamics of native grassland ecosystems for at least 1,800 years to achieve stable soil carbon equilibrium conditions before the approximate date of conversion to cropland. Simulations assumed low input agricultural practices until 1950. After 1950, simulations assumed the introduction of synthetic fertilizers with increased fertilizer rates along with more productive cropping varieties. We sourced feasible management histories prior to 2001 for each state for both irrigated and non-irrigated cropland that were informed by data from Major Land Resource Regions (MLRA) for each field location. From 2001-2020, management data compiled in the synthetic dataset (described above) were looped in 10-year periods. Daily weather data sourced from PRISM from years 1981-2020 were repeated throughout these historical simulations.

Protocols require that models are unbiased and account for sampling and model uncertainty. We used a Monte Carlo uncertainty approach following methods outlined in the Reserve’s and Verra’s protocols to adjust DayCent model outputs for bias and variance. This approach entails first constructing a mixed-effect statistical model to compare DayCent model outputs of soil carbon stocks and N₂O fluxes for fields with empirical data from long-term experiments (52,56). This approach assumes that the ‘true’ soil organic carbon stocks and N₂O fluxes are represented by the empirical field measurement data (56). The soil organic carbon statistical model included DayCent soil organic carbon predictions as a fixed effect. The N₂O statistical model included DayCent N₂O predictions, crop type (corn or other), and freeze-thaw cycling as fixed effects. Both statistical models included region, site, and year as random effects.

We used a Monte Carlo analysis to propagate error associated with the statistical models described above, adjusting for bias and deriving the variance in soil organic carbon stocks and N₂O emissions estimated by DayCent. We ran 500 iterations for each project and baseline scenario. Each iteration included a random selection of parameter values for the fixed and random variables in the statistical models described above. Each iteration estimated greenhouse gas flux (soil organic carbon stocks and N₂O emissions) for each field location from 2021 to 2030, providing 500 estimates that are used to derive confidence intervals in the estimated flux.

To calculate the raw credits for each protocol we either used the bias-adjusted DayCent model values (Nori) or the average of the Monte Carlo simulations (Verra and the Reserve) for each scenario. We converted soil organic carbon and N₂O estimates to carbon dioxide equivalent units following protocol guidelines and summed credits to calculate the net greenhouse gas flux.

Calculating Deductions

After calculating raw credits, protocols require credit producers to deduct credits to account for two types of uncertainty: uncertainty in the credit calculation approach and uncertainty in the future ‘durability’ of the credits. Protocols varied in their approach to both types of deduction calculation.

Uncertainty Deductions

Protocols require uncertainty deductions from raw credit calculations to be as conservative as possible in their credit issuances. However, uncertainty deductions varied between crediting programs and between earlier and current protocol versions. Nori’s protocols require credit producers to estimate an annual change in soil organic carbon stocks for each field in each of the 10 years of the crediting period as well as the annualized average change for each field across the 10 year period. To be conservative, Nori then requires that the credits issued for each field is the lesser of the annual or the 10-year average carbon gains.

The earlier versions of the Reserve and Verra protocols calculated uncertainty deductions by first combining soil organic carbon stock change and N₂O flux for each Monte Carlo iteration (as described above). The uncertainty deduction is the extent to which the margin of error of the average emissions reduction (er) exceeds 15 % of the estimated average emissions reduction. This earlier version “half-width” uncertainty deduction method is a function of the standard distribution of the data, the sample size, and the average emissions reductions. Half-widths will be relatively larger when the standard deviation is large or the sample size is small. Because this method also relativizes the deduction to the average emissions reduction, as the average decreases the uncertainty deduction increases.

The methods to calculate uncertainty deductions changed significantly between the earlier and current versions for the Reserve and Verra protocols. First, current protocols require separate estimates of uncertainty deductions for soil organic carbon stock changes and N₂O flux. Second, the current versions now use a “probability of exceedance” method based on the standard deviation of Monte Carlo iterations for each greenhouse gas source. Notably, while the current “probability of exceedance” method is still sensitive to the standard deviation of the data, it is no longer sensitive to the magnitude of the average emissions reduction.

Since methods based on Monte Carlo procedures produce more precise results across a greater area, we calculated raw credits and associated uncertainty deductions for various project sizes to test how project size influenced uncertainty deductions. For this analysis, we created four different ‘projects’ by randomly selecting 500, 1000, 3000 of the 4,988 total fields and estimating the uncertainty deductions.

Buffer pool contributions

Soil organic carbon pools can lose carbon to unavoidable disturbances such as erosion from wind storms and flooding, wildfires, or droughts, which are considered ‘unavoidable reversals’ by the Reserve and Verra and must be insured against with credits moved to 'buffer pools'. For both Verra protocols, buffer pool contributions were calculated from AFOLU non-permanence risk tool that assessed 26 elements that contribute to Internal Risk (Project Management, Financial Viability, Opportunity Cost, and Project Longevity Risks), 11 elements that contribute to External Related Risks (Land Resource Tenure, Community Engagement, and Political Risks), and 5 elements that contribute to natural risk. Each element is weighted based on the impact of the risk, and the total project risk is the sum of project-specific element scores but is bounded with a minimum risk of 10 %. Projects that score higher than 60 % risk are not allowed to generate credits. The Reserve’s approach to estimating buffer pool contributions were much simpler than Verra’s approach, ranged from 5 %-16.75 %, and is based on only two risk factors: a default risk of carbon loss based on whether the project fields were geographically concentrated (defined as more than 50 % of project area concentrated in a single U.S. Department of Agriculture Land Resource Region) and a risk of financial failure based on whether the project developer is a private individual or entity without any financial mechanism to reduce risk of failure.

To implement each risk assessment, we assumed the following about our project scenarios. Because our project scenarios encompassed thousands of fields across six states, the project would likely be run by a private third-party entity that had the technical expertise and staffing capacity to implement such a large project. We assumed that this third-party entity would take all contractual, financial, and legal steps possible to reduce project related risks, and for risks that would be outside the control of the third-party entity (such as natural risks or geographic location of fields), we choose conservative and realistic estimates of risk for the region. Using the Reserve’s simpler equation, we set Risk_default equal to 0.075 because > 50 % of our hypothetical project acreage for no-till and stacked no-till and cover cropping was concentrated in one Land Resource Region (Region M, 54). We determined Risk_FF as 0.1 because our company was likely to be a private entity and thus at risk of financial failure. Based on these assumptions, we calculated that Verra’s buffer pool deduction was 20 % and the Reserve's deduction was 16.75 %.

Total issued credits

We combined the analysis of additionality criteria, raw credits, and uncertainty deductions for both soil organic carbon stocks and net N₂O emission reductions, and buffer pool contributions to calculate total issued credits for each protocol.

Synthetic dataset of corn/soybean fields in Midwest US for carbon crediting analysis

Data files

Abstract

Description of the data and file structure

Sharing/Access information

Code/Software

Soil Carbon Crediting Protocols

Compiling a synthetic dataset

Field Selection

Field Management History

Determining Practice Additionality & Baseline Activity

Calculating Raw Credits

Calculating Deductions

Uncertainty Deductions

Buffer pool contributions

Total issued credits

Synthetic dataset of corn/soybean fields in Midwest US for carbon crediting analysis

Data files

Abstract

README: Synthetic dataset of corn/soybean fields in Midwest US for biogeochemical model analysis

Description of the data and file structure

Sharing/Access information

Code/Software

Methods

Soil Carbon Crediting Protocols

Compiling a synthetic dataset

Field Selection

Field Management History

Determining Practice Additionality & Baseline Activity

Calculating Raw Credits

Calculating Deductions

Uncertainty Deductions

Buffer pool contributions

Total issued credits