Review of M. Chatting et al. : “Improving Marine Sediment Carbon Stock Estimates: The Role of Dry Bulk Density and Predictor Adjustments”, egusphere-2025-661

Summary

The present manuscript by Chatting et al. describes an improved modelling approach for calculating OC stock for marine shelf areas by upscaling point observations on OC content. Two approaches are compared: (1) The traditional one uses available globally resolved datasets for predictor variables, including dry bulk density as calculated from sediment porosity and combines it with point observations of OC content in the Irish Sea to derive the local OC stock using a random forest approach. (2) The improved one first performs a bias adjustment, which transforms global data to better represent local point observations. Dry bulk density is then extrapolated using a random forest model based on these transformed predictor data. A second random forest model is applied for the OC content and lastly dry bulk density and OC content are combined into a new OC stock estimate. The new approach shows improved performance (i.e. agreement with in situ data) in predicting dry bulk density and lowers the OC stock estimate for the region by around one third.

General Comments

The approach is methodologically sound. Bias adjustment has so far been mostly applied in climate models and incorporating it into marine OC stock estimation is a novel but timely application. The other modelling approach, random forest models, is well established in geospatial modelling and is an appropriate choice for the present case. The manuscript convincingly shows that bias adjustment as well as including refined DBD estimates can substantially improve our OC stock assessments.

However, the MS is in some section hard to follow, especially in the methods section. A more clear description of the modelling workflow (supported by Fig 1), a plain language summary of what bias adjustment (a central method of this study) entails, and how the success of these measures is actually measured should appear in the introduction or early in the methods sections. This way the reader can be more effectively guided through the novel data handling approach.

Another point of concern which needs to be clarified is the choice of study area. It does not become clear what the advantages of choosing the Irish Sea are, although certainly there were some. The authors mention that only 3% of DBD data (the most important predictor variable!) are available for the study area. Either expanding the scope of the modelling to the entire NW European shelf, which is the source for this DBD data, or making a good case for limiting it to the Irish Sea are needed.

Despite these points the present MS provides a useful and novel blueprint study which can help improving OC stock modelling globally, a topic with general relevance for climate science as well as biogeochemistry.

Specific Comments

1. Clarify methods: The manuscript would benefit from some more plain language step-by-step guidance throughout the methods section. E.g. in L66-72 or later, a comprehensible description of what QQ mapping entails could help familiarizing the reader with the approach.
2. Choice of study area: The authors should better justify their focus on the Irish Sea (e.g. in L22-23), especially as DBD seems to be sparsely available here. The reader gets the impression the entire shelf area might have been a better focus (600+ data points for DBD). Certainly there are arguments for this tighter spatial focus, which could be presented here.
3. Generalizability: The authors should briefly state how the findings are transferable to other geographical regions, and what it would mean for existing OC stock assessments. Is this only applicable in areas with dense available data and therefore limited to well studied zones, or can we improve global estimates? How, e.g., would the results of Atwood et al. (2020) which are cited in the text change considering the findings of the MS? Adding a global relevance section to the discussion can help the reader better grasp the implications of the novel modeling framework.

Line comments

L15:     the study addresses the short-term C cycle, not the geological one, consider rephrasing

L19:     “Data gaps” may be misleading here. The study does not collect new data but rather improves the interpretation of existing data, rephrase

L22-23: State why Irish Sea was chosen; maybe due to sufficient data availability?

L28:     “emphasize” instead of “emphasizes” as it refers to “findings”

L29:     ensure consistent formatting of in situ (throughout the MS)

L30:     consider removing “for policy makers”, as many more stakeholders are interested in improved OC stock assessment

L49-50: harmonize phrasing “uppermost 10 cm”, “surficial 0-10 cm” and “top 0.1 m” (in methods section) all refer to the same and should be consistent

L56:     possibly use “OC stock” instead of “OC content”; the sentence is hard to read

L58:     “sediment density” should be “soil density”

L59:     consider removing “however” to improve flow

L66-72: The repeated mention of “climate models” and references about them can be drastically shortened. Instead it could include a one sentence, plain-language summary of what bias adjustment entails

L78:     If OC and mud content measurements also use similar instrumentation maybe mention this (instead)

L87:     Be more clear how model improvement is assessed

L97:     Does this mosaic of sediment types help modelling here?

L104-105: Briefly mention how inshore area is defined in this data

L147: Add a sentence as plain-language summary of QQ mapping

L153-165: Consider moving to Supplementary, log ratio transformation is a standard procedure in compositional data and not crucial to the presented modelling approach

L192:   Is this the mud content from spatial averaging? Clarify

L206: refer to Fig 4 here (may be Fig 3 then)

L214:   could this k fold CV be replaced with the NNDM LOO CV, which is introduced later and said to perform better?

L236:   “by a grid cell” instead of “by grid cell”

L241:   be more explicit than “all possible combinations”, there are 4 combinations; OC adj/unadj with DBD adj/unadj correct?

L273:   Sort plots by predictor importance in Figure 4

L277:   How would this approach perform in regions with even fewer DBD observations?

L318-320: Clarify how the presented findings would influence their estimates

L326:   move the definition of mud to its first mention in the MS

L331-335: The sentence is long and unclear; what is the “topography” of a mineral grain? Also the references need sorting

L346:   Space missing between “(2024)” and “estimated”

L356:   “resuspension” instead of “suspension”

L362-63: “needs” instead of “need”

L388: “carries” instead of “carry”

L398: rephrase “increased in situ data” to “increased availability of in situ data”

Figure 2: It seems the yellow shade is not reached/used, maybe adjust the colormap. Also the thick outline is not very visually appealing and might obscure data points. A dashed line should be tested.

Figure 4: The plot is not very visually appealing, it is not clear what the letters refer to. Maybe the distance between the a and b row can be slightly increases. Plots should be sorted in the order of relative parameter importance. Also the figure caption for 4 c is not correctly describing the presented plots.

Figure 5, 6 and 7 all show the same region, but different parameters. Maybe combining them all into one large, page filling figure would allow the reader to better appreciate all present trends at once. Another side note is, that the text mentions the “Isle of Man”, which is not labelled in a map and might be unfamiliar to many readers.