Estimating regional agricultural supply of greenhouse gas abatements by land-based biological carbon sequestration: a Bayesian sampling-based simulation approach

Abstract

In this study, we develop a sampling-based simulation approach for estimating regional agricultural supply of greenhouse gas (GHG) emission abatements by land-based biological carbon sequestration. We explicitly consider producer behaviour in a market setting that would pay for carbon sequestration depending on current land use and management, target practice to be adopted and spatial location. We construct a behaviour model in the benefit-cost framework to characterise producer decision in relation to preferences and production attributes. We combine the Markov Chain Monte Carlo technique and choice modelling in a Bayesian setting to develop an empirical procedure that may be calibrated by observed producer behaviour and agricultural census data and that can simulate regional agricultural carbon sequestration by sampling individual preferences and production attributes. An empirical application of our approach depicts potential agricultural supply of GHG abatements by carbon sequestration in a production region in the USA. This approach is flexible to be applied to different regions with minimum information requirement while accounting for spatial heterogeneity of both preferences and production.

Keywords:

• greenhouse gas
• carbon sequestration
• economics
• carbon offsets
• carbon market
• preference
• climate change mitigation

JEL classification:

• Q15
• Q24
• Q52
• Q54
• Q57

Acknowledgments

The work was conducted at the Center for Agricultural Policy and Trade Studies, Department of Agribusiness and Applied Economics, North Dakota State University

Notes

^aAll programmes required at least a 5-year commitment.

^bIncluding planting methods commonly referred to as: no till, strip till, direct seed, zero till, slot till and zone till.

^cA contract longer than 5 years might be required.

^dCredits depend on tree age and species; at least a 20 acres enrolment was required.

^eA carbon price of $25 per metric ton of CO₂ equivalent was assumed in this example.

Data source: USDA (2010).

^aThe results are based on MCMC simulation with 500 random draws from the posterior distribution of producer preference parameters.

^aIn Bayesian statistics, the estimated coefficients and standard errors are actually the means and standard deviations of the posterior distribution of coefficient parameters. However, we interpret the result in classic statistics for ease of understanding based on the Bernstein–von Mises theorem that establishes that: (1) the posterior mean considered as a classic estimator is asymptotically equivalent to the maximum likelihood estimator; and (2) the posterior variance is asymptotically equivalent to the sampling variance of the maximum likelihood estimator. A fixed effect independent of producer production attributes was identified for carbon revenue across programmes, which was 0.0779** (0.0391) and omitted. The standard errors of the estimated coefficients are in parenthesis. *** denotes significance at the 0.01 level, ** denotes significance at the 0.05 level and * denotes significance at the 0.1 level.

^bIn our case study, the Gibbs sampling combined with a generic single-component MH algorithm took independent random draws, with a size of burn-in of 40,000 (i.e., number of iterations to make prior to retaining draws). The number of draws retained after burn-in is 500, and the number of iterations made between retained draws is 100, which imply a Markov chain with a length of 50,000. Our determination of convergence was based on an empirical approach specifying a sufficient large size of burn-in and on observation of trending draws during burn-in.

^cProducer production attributes were measured by deviations from their sample averages, which were interacted with carbon programme attributes to create independent variables incorporated in the mixed logit model. The coefficients for the Constant variable represent the sample average utility for carbon programmes. The coefficients for other independent variables represent the marginal utility for one unit deviation in these variables from their sample averages.

^a Const represents the Constant variable in producer production attributes. Dtillage represents the dummy variable for conservation tillage, and so on for other programmes. ConstXDtillage represents the interacted independent variables.

1. As a matter of fact, the respondents of the mail survey used in this study listed requirement for capital investment and wet and cold soil conditions as the major factors prohibiting adoption of conservation tillage on their farmland.