Abstract

A statistical framework we call CQUESST (Carbon Quantification and Uncertainty from Evolutionary Soil STochastics), which models carbon sequestration and cycling in soils, is applied to a long-running agricultural experiment that controls for crop type, tillage, and season. CQUESST embeds a dynamic stochastic model of soil carbon, motivated by the deterministic RothC soil-carbon model, within a Bayesian hierarchical statistical model. CQUESST has a coherent framework that acknowledges uncertainties in soil-carbon dynamics, in physical parameters, and in observations. The long-running experiment ran from 2000-2010 and is called the Millenium Tillage Trial; here CQUESST is used to model soil-carbon in six pools, across 42 agricultural plots, and on a monthly time-step for a decade. It is implemented efficiently in the probabilistic programming language Stan using its MapReduce parallelization. We infer the effectiveness of different experimental treatments for soil-carbon sequestration; and we show how CQUESST can be used for the analysis of designed experiments to draw statistically defensible conclusions about the dependence of soil-carbon decay rates on crop rotations and tillage treatments. These results take into account the uncertainties in the model, resulting in inferences that could inform soil-carbon sequestration decisions and policies.

Coauthors: Noel Cressie, Jeff Baldock, David Clifford, Ryan Farquharson, Lawrence Murray, Mike Beare, Denis Curtin