the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Multiscale assessment of North American terrestrial carbon balance
Kelsey T. Foster
Yoichi P. Shiga
Jiafu Mao
Anna M. Michalak
Abstract. Comparisons of carbon uptake estimates from bottom-up terrestrial biosphere models (TBMs) to top-down atmospheric inversions help assess how well we understand carbon dioxide (CO2) exchange between the atmosphere and terrestrial biosphere. Previous comparisons have shown varying levels of agreement between bottom-up and top-down approaches, but they have almost exclusively focused on large, aggregated scales, providing limited insights into reasons for the mismatches. Here we explore how consistency, defined as the spread in net ecosystem exchange (NEE) estimates within an ensemble of TBMs or inversions, varies with spatial scale. We also evaluate how well consistency informs accuracy in overall NEE estimates by filtering models based on their agreement with the variability, magnitude, and seasonality in observed atmospheric CO2 drawdowns or enhancements. We find that TBMs produce more consistent estimates of NEE for most regions and at most scales compared to inversions. Filtering models using atmospheric CO2 metrics causes ensemble spread to decrease substantially for TBMs, but not for inversions. This suggests that ensemble spread is likely not a reliable measure of the uncertainty associated with the North American carbon balance. Promisingly, applying atmospheric CO2 metrics leads to a set of models with converging flux estimates across TBMs and inversions. Overall, we show that multiscale assessment of the agreement between bottom-up and top-down NEE estimates, aided by regional-scale observational constraints, illuminates a promising path towards identifying fine-scale sources of uncertainty and improving both ensemble consistency and accuracy. These findings help refine our understanding of biospheric carbon balance, particularly at scales relevant for informing regional carbon-climate feedbacks.
- Preprint
(1116 KB) - Metadata XML
-
Supplement
(928 KB) - BibTeX
- EndNote
Kelsey T. Foster et al.
Status: final response (author comments only)
-
RC1: 'Comment on bg-2023-111 by Guillermo Murray-Tortarolo', Guillermo Murray-Tortarolo, 07 Sep 2023
Dear authors,
I had the pleasure of reviewing your manuscript, which I found very interesting and well written. Here you employed different measurements of variability to filter NEE from different models, showing that employing basic atmospheric constraints greatly improved consistency and reduces uncertainty. The approach is certainly promising and interesting, but also comes with several limitations that are not necessarily presented as they should. I believe the manuscript will be a strong contribution to the benchmarking field after a few things are addressed (see comments below).Major comments
I have only one major comment, which is the need to include a result sub-section for limitations of the approach.
Other comments (mostly minor).
Mostly suggestions for improving tables and general readability for the reader.
Abstract
Is very well written, but the spatial scale is defined almost at the end of the abstract. To aid the reader, a mention of your spatial scales is needed in line 11 (e.g. Previous comparisons “in North America”), line 14 (e.g. varies with spatial scale “(pixel, country, continent)”), line 19 (North American carbon balance “at any spatial scale”).
Introduction
Line 27.- change “beyond the plot scale” for “greater than a plot (1km2)”
Line 28.- change “larger” for “broader”
Second paragraph: incomplete ideas. At least two things need to be included here: 1) a line that helps the reader connecting the ideas with the next paragraph (i.e. “the issue arises from the particular characteristics of TBMs and Inversions”) and 2) a stronger argument for the need to reduce uncertainties beyond an academic interest (e.g. reducing uncertainty also helps creating better mitigation policies).
Line 48.- actually the highest discrepancies across models are likely how they incorporate land use change, vegetation dynamics and fire (I believe far more than nitrogen or permafrost). Perhaps is worth mentioning a list (e.g. however other large discrepancies also arise from the different approaches on how to model land use change, vegetation dynamics and fire).
Line 48.- is not only differences in parametrization, but also on the driving data. TRENDY models sorted this issue by employing the same forcing and protocol, otherwise this is a major issue of variability. In summary, there are three elements that create model discrepancies: 1) structure, 2) parametrization and 3) driving data.
Line 50.- you can take the argument further (since you are filtering models in this work), argue that you by addressing the sources of uncertainty you can benchmark model results, which can quickly lead to overall model improvement (e.g. if you know which model structure yields more realistic results, you can push other models to incorporate said structure).
Line 52.- the first sentence is too long. You could perhaps start with something simpler (e.g. “For inversions, uncertainties arise from several measurement and processing aspects”).
Line 57.- why? Can you explain a bit why we find large spread despite having high data availability?
Line 58.- change “Another challenge” for “Despite their usefulness, a key limitation”
Line 62.- change “how well we understand” for “our understanding of”
Line 64.- is not that the confidence in both types of model increases, I would say that they become more reliable as sources of “realistic” information.
Lines 66- 70. Please expand on the examples, as you are comparing different spatial scales and regions. Particularly, please provide detailed examples for previous results in North America.
Line 70 (sixth paragraph). The paragraph provides incomplete reasoning and needs to be improved grammatically. You could start with an opening line such as: “There are different approaches to compare TBMs with inversions. On the first hand, there are direct comparisons of the means, which is usually referred to as “agreement” (citation); on the other hand, there are approaches centered on the variability, which we defined as “consistency” (citation). The first provides XXX type of information, while the second XXX”.
Line 75.- add “Previous” to the beginning of the sentence. Change “reveal” to “have revealed”. Add- the agreement “between estimates”.
Line 76.- remove “do”
Line 77.- move “,however,” to the beginning of the sentence.
Line 78.- remove the “,” before the “and”.
Line 87. Opening sentence is too long. Perhaps start with: “A key step forward is to look at agreement and comparison across scales”.
Line 101.- This argument is always complicated. In theory, you would expect that models who provide better estimates in historical runs, would be better suit for future projections; however, this is not always the case. For example, models that include a N cycle usually perform worst than C-only models in present-day conditions, however they are likely better at recreating future scenarios where N becomes limiting for NPP.
I believe you can leave the sentence as is in the introduction, but the arguments presented need to be included in the discussion.
Lines 101-103. Repetitive, you have already defined the terms. Not needed.
Line 116. Change “North American NEE” for “NEE in North America”
Methods
One aspect that is complicated is how very little information there is to benchmark models based on seasonality. There are only 4 towers employed in the seasonality analysis, all of which are located in croplands; even in the larger Tower compendium, there is little representation across drylands which have been shown to drive most of the IAV of the CO2 cycle. This needs to be addressed in the discussion into detail.
Lines 147-153. One key issue with TRENDY data is the land-mask employed to remove ocean fractions. This needs to be clearly stated. If the data was crop first at the original resolution (0.5°) then regridded, this is not an issue; but if you regridded first (to a 1x1 grid), the estimates for NEE become much larger. Please specify how you perform post-processing of the data.
Line 175. Why not using MODIS GPP?
Acknowledgements
Please notice that the TRENDY data policy states that you need to clearly acknowledge them for using their data.
ResultsLine 276.- Please change the “however” to the beginning of the paragraph, and merge this paragraph with the previous one.
Line 371-379.- I believe these values should be presented as a table. Particularly show the how the mean and deviation for the region (and land categories) changes with model filtering.
Please add a section on limitations of the study and the approach. I belive this should be clearer. While the results are really interesting and promising, several data-limitation issues are presented (mentioned above).
Tables.
Table 1.- please organize the table by type of ensemble instead of model name.
Tables 1 & 2.- I strongly suggest to merge both tables. A simple solution is to add four columns to table one (one for each metric and the total). This way the reader can quickly see which models meet which metrics.
Figures
Figure 2.- I belive this figure should go into the supplementary
Figure 5.- Why is there no comparison for grasslands and drylands? They represent a major proportion of land across NA!
Citation: https://doi.org/10.5194/bg-2023-111-RC1 -
RC2: 'Comment on bg-2023-111', Anonymous Referee #2, 15 Sep 2023
The manuscript is generally well written and delves into the discrepancies between NEE as estimated by terrestrial biosphere models (TBMs) and those derived from top-down atmospheric inversions across different scales. Addressing this topic is important in developing reliable carbon budgets. However, the manuscript misinterprets the NEEs estimated by these TBMs. Furthermore, the discussion section seems to lack comprehensive analysis, leaving out essential arguments that could better support the authors' conclusions. These sections would benefit significantly from a detailed review and subsequent refinement.
Comments:
1) In Table 1, results from both the BG1 and SG3 scenarios of MsTMIP are presented. However, in the methods section, only the usage of BG1 results is detailed. Could you please provide an explanation for this discrepancy?
2) In the MsTMIP project, the participating TBMs do not estimate the NEE directly; instead, they utilize the stock change approach. It's important to note that while some models incorporate factors like fire and harvesting into their simulations, others do not. This distinction should be clearly addressed in the manuscript. Additionally, while certain models use fire and harvesting data in calculating the NEE, Line 175 of the manuscript only acknowledges the photosynthesis and ecosystem respiration factors. This discrepancy needs addressing.
3) In your study, I noticed that a singular value was presented for all the TBMs and AIMs. Could you clarify how you combined the results from these models on Line 150? If you simply averaged them, I suggest referencing the 'integration approach' detailed by Schwalm et al. (2015), Toward 'optimal' integration of terrestrial biosphere models, Geophysical Research Letters. Given the significance of spatial patterns in this study, relying solely on a simple average might introduce notable uncertainties. It's worth mentioning that the NEEs estimated by different TBMs can vary significantly. Hence, it would be beneficial to analyze each model's estimates separately and delve into the biogeochemical processes that might account for the discrepancies observed between top-down and bottom-up approaches.
4) To explain the discrepancy between top-down and bottom-up estimates of NEE, numerous studies have been conducted, including notable publications by Peter A. Raymond and David E. Butman. The discrepancy is attributed to the lateral carbon flux of dissolved organic carbon, particulate organic carbon, and carbon in inorganic formats. Given that these TBMs have been calibrated and validated using field measurements, such as soil organic carbon, it is possible to incorporate the lateral carbon fluxes when estimating each carbon pool. Unfortunately, this process seems to have been overlooked in the current discussion.
Line 75: The lateral carbon flux should be discussed see Casas-Ruiz et al. (2023), Integrating terrestrial and aquatic ecosystems to constrain estimates of land-atmosphere carbon exchange, Nature Communications.
Line 101: To adequately convey the concept of "consistency" in the article, it's crucial to delve into the differences among the models. These models vary significantly in their simulation processes, leading to considerable variations in NEE. It's important to note that the models participating in the MsTMIP project use the same input data. I am not familiar with the TRENDY project. Ensure that both projects utilize the same input data; if not, the differing input data could be a significant source of variance.
Line 204: From where did you obtain the biome map? Additionally, did both MsTMIP and TRENDY projects utilize the same biome map?
Line 311: Could you please list the two-thirds of TBMs and analyze the potential reasons for their behavior? Specifically, do these models incorporate certain key processes?
Line 351: TBMS – TBMs
Line 380: The current discussion is insufficient. It's essential to address the role of lateral carbon flux in causing discrepancies between bottom-up and top-down estimates. Furthermore, the method of NEE calculation across different models should be discussed.
NEE estimation approaches for these TBMs:
NEE=-GPP+TR
NEE=-GPP+TR+Fire
NEE=-GPP+TR+ Harvesting
NEE=-GPP+TR+Fire+Harvesting
Citation: https://doi.org/10.5194/bg-2023-111-RC2
Kelsey T. Foster et al.
Kelsey T. Foster et al.
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 195 | 79 | 13 | 287 | 22 | 4 | 5 |
- HTML: 195
- PDF: 79
- XML: 13
- Total: 287
- Supplement: 22
- BibTeX: 4
- EndNote: 5
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1