the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 17 Jun 2025
Ensemble-based observation impact of surface CO2 concentration observations on analysis and forecast of atmospheric CO2 concentrations over East Asia
Abstract. The effect of assimilated surface CO2 concentration observations on the analysis and forecast errors of model CO2 concentrations was evaluated using the data assimilation (DA)-forecast system combining the Weather Research and Forecasting model coupled with Chemistry and modified Data Assimilation Research Testbed. To investigate the impact of surface CO2 observations, four observing system simulation experiments (OSSEs) were conducted in July 2019. The impact of CO2 observations on the CO2 concentration analysis was calculated using self-sensitivity. Average self-sensitivity of four OSSEs was 21.0 %, implying that surface CO2 observations provided average 21.0 % information to the CO2 concentration analysis. Self-sensitivity was highly correlated with the root mean square error of the analysis and hourly variability in the surface CO2 observations at each observation site. The impact of CO2 observations on reducing forecast errors, calculated using nonlinear forecast error reduction (NER), showed that NER with DA was reduced by average 17.0 % compared with that without DA. Linear forecast error reduction, calculated using the ensemble forecast sensitivity to observation (EFSO) impact, showed that the EFSO impact was greater at surface CO2 observation sites with higher self-sensitivity and active vegetation types. Average fraction of beneficial observations for all experiments was 68.9 % (66.3 %) for 6 h (12 h) forecasts, implying that more than half of the assimilated CO2 observations contributed to reducing forecast errors. The assessment of observation impact on the CO2 concentration analysis and forecast can be useful for monitoring and estimating atmospheric CO2 concentrations, optimizing surface CO2 fluxes, and designing atmospheric CO2 observation networks.
- Preprint
(4615 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
- RC1: 'Comment on egusphere-2025-2367', Anonymous Referee #1, 08 Jul 2025
-
RC2: 'Comment on egusphere-2025-2367', Anonymous Referee #2, 25 Jul 2025
The manuscript "Ensemble-based observation impact of surface CO2 concentration observations on analysis and forecast of atmospheric co2 concentrations over East Asia" of Min-Gyung Seo and Hyun Mee Kim analyze the impact of CO2 observations in OSSEs for the regional analysis (via data assimilation) and forecast of atmospheric CO2 concentrations in East Asia.
The introduction does not really state a clear objective for such an analysis of the observation impact for existing and potential CO2 observation sites (in particular, something is missing in lines 73-74) and lines 211-214 narrow down the potential scope of the study. However, in principle, it should support the extension of surface networks "to better analyze and forecast atmospheric CO2 concentrations in East Asia" (l74).
The authors deploy complex experimental, data assimilation and analysis theoretical and practical frameworks to conduct this study. However, as explained below, many major aspects of the study and of the manuscript raise concerns, from the rationale and objective of the study to the implementation of the experiments and interpretation of the results, including the quality of the writing. These concerns are such that they can hardly be addressed if limiting the request of the journal to major revisions, which is why I suggest to reject this manuscript and to encourage the authors to reconsider and rebuild their study before resubmitting a new manuscript based on their valuable tools.
From my point of view, the regional analysis and forecast of the atmospheric CO2 concentrations is not a sensible target for the deployment of surface CO2 networks. There is no major scientific or societal need for accurate forecasts of the CO2 concentrations over short timescales. Global CO2 forecasts at relatively high spatial resolution are often used to constrain the boundary conditions of regional and local CO2 atmospheric inversion systems (solving for the surface CO2 fluxes), but in many cases, such regional and local inversion systems are coupled to global CO2 inversion systems (solving for the surface fluxes). One could still argue that regional forecasts could also be used to constrain the boundary conditions of local atmopsheric inversion systems. However, given the current lack of CO2 surface stations, it would not make sense to optimize the design of relatively dense continental monitoring networks for such an objective. In any case, the analysis and discussions in this study do not provide indications regarding the potential for such a coupling.
The introduction is quite revealing regarding this concern: the part dedicated to the rationale of the study up to line 36 mainly discusses the need to estimate the CO2 fluxes. However, line 37 jumps into the analysis and forecast of CO2 concentrations without explanations, starting with a "Therefore", which artificially connects the two parts.
Even if assuming that optimizing the design of continental networks as a function of their skill for supporting the analysis and forecast of CO2 concentrations could make sense, the study keeps on raising concerns.
The analysis for each assimilation window focuses on the correction of the CO2 initial conditions, without correcting the surface fluxes. However, even over few hours, CO2 concentrations are highly impacted by such surface fluxes, and in particular (when considering networks such as thoses tested here) by the surface land ecosystem fluxes. In the real world, the system that is tested here would thus project large biases in the analysis of the CO2 atmospheric fields to compensate for the large uncertainties arising from the surface fluxes during the assimilation window, and the forecasting skill of such a system would be strongly limited by ignoring these fluxes in the analysis. This point is missed by the experiments here, because the authors implicitly assume that the biogenic surface fluxes are perfectly known (the true and
"perturbed" estimate of these fluxes are identical = the outputs from VPRM). This issue may not be so significant if there had not been a large number of regional scale atmopsheric inversion systems and studies in the past decades.Furthermore, even though they use complex diagnostics to read the results from their experiments, the authors face difficulties to interpret them:
- a challenge associated to such a study is the mix between the observation impact of a given station within a given network, which can be very different within another network, and the impact of using different networks. The experimental framework and the analysis here do not fully ensure the distinction between these two impacts which limits the ability to draw robust conclusions regarding specific types of stations or of networks.
- the lack of account for the atmospheric transport conditions over East Asia in July 2019 when evaluating the impact of the different stations or networks is an issue, e.g., since the positioning of the stations with respect to the study domain, to each other or to the domain boundaries when following the wind probably plays an important role in the forecasting skills, and since the transport conditions in July 2019 could be specific.
- the lack of clear information on and characterization of the initial and/or sequential derivation of the uncertainties in the background state / initial conditions (the Pb matrix and the spread of the (xb)_i) of the assimilation windows (what are the spatial scales of the correlations associated to these uncertainties ?) further limits the ability to analyze properly the observation impacts (page 4 is confusing; the discussions on min distances between stations = 600 km or 300 km on page 9 is highly questioning).
- The conclusion that stations located in areas with strong biogenic fluxes have a larger impact could be questioning since there is no perturbation of these fluxes in the experiments. The authors do not propose a mechanism to explain this. It could actually be linked to the transport, e.g. the PBL (a strong driver of the CO2 diurnal variations that is ignored at lines 451-452), whose modeling scheme is perturbed in the OSSEs. In areas with high biogenic fluxes, such a transport uncertainty would propagate into higher CO2 errors.
The authors do not discuss any potential problem associated with the assimilation of CO2 observations at night while most of the global to local inverse modelling systems keep on avoiding to assimilate nighttime CO2 observations from plain or low altitude stations due to the large CO2 transport modelling biases at night.
The writing of the manuscript is not satisfactory, many sections are unclear, and they do not introduce the main goals, concepts and ideas in a clean way, which does not encourage the reader to delve into the mathematical framework. As an example, the lines 105-117 which contain key information are confusing: here, the authors do not really try to describe or explain things rigorously but rather to list the values or options for their various input parameters. Part of the basic information on the study domain, period, modelling framework (content of the state vector x, spatial resolutions, duration of the assimilation windows and forecasts...) etc. is delivered too late, i.e., after considerations that should be driven by such information, or is simply skipped (see the discussion on Pb above). The abstract provides a first good illustration of this general problem. In the introduction and result sections, the authors often lose the reader with a high number of statements and statistics which do not systematically follow a logical flow or which do not seem to be the most relevant.
Going into more details in this manuscript raise further questions and concerns, but I limit this review to this list of general issues.
Citation: https://doi.org/10.5194/egusphere-2025-2367-RC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 293 | 44 | 9 | 346 | 7 | 18 |
- HTML: 293
- PDF: 44
- XML: 9
- Total: 346
- BibTeX: 7
- EndNote: 18
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1