Bayesian inference and predictive performance of soil respiration models in the presence of model discrepancy

Elshall, Ahmed S.; Ye, Ming; Niu, Guo-Yue; Barron-Gafford, Greg A.

doi:https://doi.org/10.5194/gmd-12-2009-2019

Articles | Volume 12, issue 5

https://doi.org/10.5194/gmd-12-2009-2019

© Author(s) 2019. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/gmd-12-2009-2019

© Author(s) 2019. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 12, issue 5

Methods for assessment of models

|

23 May 2019

Methods for assessment of models |

| 23 May 2019

Bayesian inference and predictive performance of soil respiration models in the presence of model discrepancy

Ahmed S. Elshall, Ming Ye, Guo-Yue Niu, and Greg A. Barron-Gafford

Download

Final revised paper (published on 23 May 2019)
Supplement to the final revised paper
Preprint (discussion started on 26 Nov 2018)

Interactive discussion

Status: closed

AC: Author comment | RC: Referee comment | SC: Short comment | EC: Editor comment

- Printer-friendly version

- Supplement

RC1: 'review', Anonymous Referee #1, 17 Jan 2019
- AC1: 'Author Response to RC1', Ming Ye, 14 Mar 2019
RC2: 'Review', Anonymous Referee #2, 30 Jan 2019
- AC2: 'Author Response to RC2', Ming Ye, 14 Mar 2019

Peer-review completion

AR: Author's response | RR: Referee report | ED: Editor decision

AR by Ming Ye on behalf of the Authors (14 Mar 2019) Author's response Manuscript

ED: Referee Nomination & Report Request started (15 Mar 2019) by Christoph Müller

RR by Anonymous Referee #2 (27 Mar 2019)

RR by Anonymous Referee #1 (02 Apr 2019)

Suggestions for revision or reasons for rejection

I revisit some of my previous comments, considering response from the authors in the discussion forum.

1. Contribution: the authors have now better articulated their contribution. The paper is somewhat incremental, since the methodology is not new and the findings are similar to applications in other disciplines. Nevertheless it is still useful to have an explicit evaluation and comparison of the impacts of statistical assumptions in soil respiration models.

2. Problems with residual autocorrelation: I agree that the authors do not need to solve this issue in this paper. However, the alternative suggested approach (Evin et al) could be easily tested (simply swap the order in which correlation and heteroscedasticity are applied in the likelihood function). In their extended response in the discussion forum the authors seem to indicate that they have tested this. Why then not add it to the paper? If not in the results then in the discussion part of the paper ("preliminary results show..."). My suggestion is to include the entire extended response posted in the discussion forum (under comment 2 of RC1) in the paper itself, as it provides more information that is useful for readers.

3. Grammatical/spelling errors: the authors state that they have "corrected several other
grammatical errors", but it's not clear what was corrected exactly.

Additional comment:
- the abstract states that "not accounting for heteroscedasticity ... will definitely underestimate uncertainty". It is not clear why this would be the case. Perhaps you mean to say that not accounting for any residual error beyond parameter uncertainty will underestimate predictive uncertainty (as in section 4.2)? Although that is not really a significant finding.

Hide

ED: Publish subject to minor revisions (review by editor) (08 Apr 2019) by Christoph Müller

AR by Ming Ye on behalf of the Authors (16 Apr 2019) Author's response Manuscript

ED: Publish subject to technical corrections (23 Apr 2019) by Christoph Müller

AR by Ming Ye on behalf of the Authors (24 Apr 2019) Author's response Manuscript

Short summary

The assumptions that the residuals are independent, identically distributed, and have constant variance tend to simplify the underlying mathematics of data models for Bayesian inference. We relax these three assumptions step-wise, resulting in eight data models. Using three mechanistic soil respiration models with different levels of model discrepancy, we discuss the impacts of data models on parameter estimation and predictive performance, and provide recommendations for data model selection.