Efficient ensemble data assimilation for coupled models with the Parallel Data Assimilation Framework: example of AWI-CM (AWI-CM-PDAF 1.0)

Nerger, Lars; Tang, Qi; Mu, Longjiang

doi:https://doi.org/10.5194/gmd-13-4305-2020

Articles | Volume 13, issue 9

https://doi.org/10.5194/gmd-13-4305-2020

© Author(s) 2020. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/gmd-13-4305-2020

© Author(s) 2020. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 13, issue 9

Development and technical paper

|

16 Sep 2020

Development and technical paper |

| 16 Sep 2020

Efficient ensemble data assimilation for coupled models with the Parallel Data Assimilation Framework: example of AWI-CM (AWI-CM-PDAF 1.0)

Lars Nerger, Qi Tang, and Longjiang Mu

Download

Final revised paper (published on 16 Sep 2020)
Preprint (discussion started on 01 Aug 2019)

Interactive discussion

Status: closed

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

- Printer-friendly version

- Supplement

SC1: 'Executive Editor Comment on gmd-2019-167', Astrid Kerkweg, 07 Aug 2019
- AC1: 'Response to Executive Editor, Astrid Kerkweg', Lars Nerger, 26 Nov 2019
RC1: 'Review', Anonymous Referee #1, 16 Oct 2019
- AC2: 'Response to Reviewer \#1', Lars Nerger, 26 Nov 2019
SC2: 'Several concerns regarding this PDAF article', Li Liu, 19 Oct 2019
- AC3: 'Response to interactive comment by Li Liu', Lars Nerger, 26 Nov 2019
RC2: 'Review of gmd-2019-167', Anonymous Referee #2, 21 Oct 2019
- AC4: 'Response to Reviewer \#2', Lars Nerger, 26 Nov 2019

Peer-review completion

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

AR by Lars Nerger on behalf of the Authors (26 Nov 2019) Author's response Manuscript

ED: Referee Nomination & Report Request started (11 Feb 2020) by James Annan

RR by Anonymous Referee #1 (04 Mar 2020)

Suggestions for revision or reasons for rejection

For reference, these are the 3 first main points of my previous review (point 4 has been addressed):

1. No actual results of the assimilation system are presented. Only the execution time for different settings. It is unclear to me what the role of a reviewer can be in this case. I rather think that the paper should also include the results of such model (see also the following point).

2. The manuscript mentions different approaches to implement the assimilation in a coupled system: in a combined state vector spanning the atmospheric and ocean model or separately. The question about which approach is better is still open and it should not be too difficult for the authors to check both approaches. This would help also in addressing the previous point and add substantially to the scientific value of this paper.

3. The different time scales of ocean and atmosphere are not discussed and the assimilation is done only in the ocean. To really appreciate the effectiveness of the coupling, data should be assimilated in both the atmosphere and the ocean and the question regarding the assimilation frequency should be addressed. As usual, the models should be validated against independent observations.

The arguments from the authors, are essentially the following:

A: the authors think that it is enough for GMD to just present run times and presenting actual results is out-of-scope for GMD
B: only a short experiment (10 days) was conducted
C: it is unclear if the strong assimilation does actually work better than weak-assimilation (with our present understanding)

Please refer to the full response available at https://www.geosci-model-dev-discuss.net/gmd-2019-167/gmd-2019-167-AC2.pdf

For me, my 3 major comments are still relevant and within the scope of the journal. They have not been addressed in this review.

I have checked the 5 papers at the front page (as of 27 November, when I first read the authors reply) and all present the actual results of the model (as opposed to just run-times).
Most of them also do actually compare the model with observations.
Also the paper by Kurtz et al, 2016 published by GMD mentioned by the author and using PDAF, displays the results of a twin experiment (Figure 6 and Figure 9-12). I am surprised that the authors thinks that this is out of scope for the journal.

It also unclear why the authors limit their simulation to just 10 days which is very short for an ocean model at this resolution (from 160 km to 30 km). Most of the fields would remain fairly constant over that time scale (as also noted by the authors). The authors mentioned computing resources as a problem, but at the same time, they mentioned that a full year simulation should be completed in 6.5 hours (requiring 79,000 core-hours). I am left with the question why the authors tackle this problem before securing the necessary computing resources to make a meaningful simulation.

Also, to me it is quite natural that when a new method is introduced or implemented one needs to show its benefit compared to other techniques.

The other points raised in my previous review have been addressed in an adequate manner.

My understanding of the scope of GMD might be wrong (and presenting just run times as opposed of actual results and not presenting improvements relative to other approaches is acceptable). If this is the case, then the Editor is invited to correct me.

Hide

RR by Anonymous Referee #2 (08 Mar 2020)

ED: Reconsider after major revisions (30 Apr 2020) by James Annan

AR by Lars Nerger on behalf of the Authors (13 May 2020) Author's response Manuscript

ED: Referee Nomination & Report Request started (04 Jun 2020) by James Annan

RR by Anonymous Referee #1 (23 Jun 2020)

ED: Publish as is (04 Aug 2020) by James Annan

AR by Lars Nerger on behalf of the Authors (05 Aug 2020) Manuscript

Short summary

Data assimilation combines observations with numerical models to get an improved estimate of the model state. This work discusses the technical aspects of how a coupled model that simulates the ocean and the atmosphere can be augmented by data assimilation functionality provided in generic form by the open-source software PDAF (Parallel Data Assimilation Framework). A very efficient program is obtained that can be executed on high-performance computers.