Bayesian Calibration of a Process-Based Ecosystem Model to Simulate Soil Organic Carbon Dynamics and Mitigation Potentials
Abstract
Benefits of carbon sequestration in agricultural soils are well recognized, and process-based models have been developed to better understand sequestration potential. However, most studies ignore the uncertainty arising during model prediction—a critical requirement for scientific understanding, policy implementation and carbon emission trading. Furthermore, parameterization is challenging due to the dependencies created in process-based models due to many parameters and a relatively small set of empirical data. We have implemented a Bayesian approach using sampling importance resampling (SIR) to calibrate the DayCent ecosystem model for estimating soil organic carbon (SOC) stocks, stock differences between management practices and their uncertainties. A SOC dataset compiled from 19 long-term field experiments, representing 117 combinations of management treatments, with 506 observations, was split into independent datasets for model calibration and model evaluation. The most important DayCent model parameters were identified through global sensitivity analysis (GSA) for parameterization. We summarize the DayCent model parameterization with marginal probability density functions (PDFs) and make predictions associated with posterior predictive distributions and potential carbon sequestration estimates using 95% Bayesian credible intervals.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2019
- Bibcode:
- 2019AGUFM.B31B..02G
- Keywords:
-
- 0428 Carbon cycling;
- BIOGEOSCIENCES;
- 0439 Ecosystems;
- structure and dynamics;
- BIOGEOSCIENCES;
- 1630 Impacts of global change;
- GLOBAL CHANGE;
- 1631 Land/atmosphere interactions;
- GLOBAL CHANGE