Geoscientific Model Development
Geoscientific Model Development
GMD
Articles | Volume 8, issue 7
Articles | Volume 8, issue 7
https://doi.org/10.5194/gmd-8-1899-2015
© Author(s) 2015. This work is distributed under
the Creative Commons Attribution 3.0 License.
https://doi.org/10.5194/gmd-8-1899-2015
© Author(s) 2015. This work is distributed under
the Creative Commons Attribution 3.0 License.
Articles | Volume 8, issue 7
Methods for assessment of models
 | 
01 Jul 2015
Methods for assessment of models |  | 01 Jul 2015

Global sensitivity analysis, probabilistic calibration, and predictive assessment for the data assimilation linked ecosystem carbon model

C. Safta, D. M. Ricciuto, K. Sargsyan, B. Debusschere, H. N. Najm, M. Williams, and P. E. Thornton

Abstract. In this paper we propose a probabilistic framework for an uncertainty quantification (UQ) study of a carbon cycle model and focus on the comparison between steady-state and transient simulation setups. A global sensitivity analysis (GSA) study indicates the parameters and parameter couplings that are important at different times of the year for quantities of interest (QoIs) obtained with the data assimilation linked ecosystem carbon (DALEC) model. We then employ a Bayesian approach and a statistical model error term to calibrate the parameters of DALEC using net ecosystem exchange (NEE) observations at the Harvard Forest site. The calibration results are employed in the second part of the paper to assess the predictive skill of the model via posterior predictive checks.

How to cite. Safta, C., Ricciuto, D. M., Sargsyan, K., Debusschere, B., Najm, H. N., Williams, M., and Thornton, P. E.: Global sensitivity analysis, probabilistic calibration, and predictive assessment for the data assimilation linked ecosystem carbon model, Geosci. Model Dev., 8, 1899–1918, https://doi.org/10.5194/gmd-8-1899-2015, 2015.
Received: 01 Sep 2014 – Discussion started: 15 Oct 2014 – Revised: 08 Jun 2015 – Accepted: 09 Jun 2015 – Published: 01 Jul 2015
Download
Short summary
In this paper we propose a probabilistic framework for an uncertainty quantification study of a carbon cycle model and focus on the comparison between steady-state and transient simulation setups. We study model parameters via global sensitivity analysis and employ a Bayesian approach to calibrate these parameters using NEE observations at the Harvard Forest site. The calibration results are then used to assess the predictive skill of the model via posterior predictive checks.
Read more