The manuscript investigates the processes driving near-bed oxygen changes on the Northwest European Continental Shelf under a high-emissions climate change scenario, with a focus on the intermodel uncertainties in these processes and their effects on oxygen. This work extends and qualifies the results of a previous study (Wakelin et al., 2020) by adding two additional sets of regionally downscaled model projections within the high-emissions forcing scenario (RCP 8.5).

Ocean deoxygenation and coastal hypoxia under climate change pose a serious threat to marine ecosystems. Robust understanding and projection of these processes is important for effective adaptation of ecosystem services. Given the lack of skill of coarse resolution global ESMs in coastal regions, regional downscaling of ESM projections will likely play a critical role in exploring this topic. 

Although these additional model simulations provide valuable new insights into the fate of the oxygen in the region, some of the main conclusions reached by the authors are not well supported by the evidence presented. The scope of the study is not well defined and the manuscript overall lacks focus and rigor. While the scientific premise of the study is valuable, major revisions are required for this work to be fit for publication in Biogeosciences.

Specific comments:

1. A major result of the paper is the attribution of the deoxygenation hotspot in the Norwegian Trench to a relaxation/reversal of the Norwegian Trench Current; but this interpretation is not well supported or well argued. The authors argue that (1) a relaxation of the advective current causes a freshening of the shelf region causing increased stratification, and (2) correlation suggests that the increased stratification is responsible for deoxygenation. Holt et al (2018) argue that changes in stratification are responsible for the relaxation of the current, opposite to the authors’ explanation. In most cases, an increase in stratification would come from surface warming and precipitation changes; this null hypothesis should be disproved before seeking alternative explanations. It is also not clear in the results whether vertical mixing or horizontal advective transport is dominating oxygen supply to the Norwegian Trench region, which should guide the conclusions made. Note that Wakelin et al (2020) do link reduced current to a recoupling of export with near-bed respiration; perhaps this is connected to the change in sign of correlation between SS and stratification (320). Lastly, ‘tight coupling’ in Figure 11 is not necessarily convincing by eye. A stronger link has to be made. 
2. The title of the manuscript suggests that the focus of the paper is on ‘intra-scenario variability’; however, it is unclear what the scope of this is and how effectively it can actually be investigated with available tools. Uncertainty in ESM climate projections (and by extension, downscaled projections) fall broadly into three categories, regarding (1) internal model variability, (2) intermodel uncertainty, (3) and scenario uncertainty. The term ‘intra-scenario’ would suggest that you look at both internal variability and intermodel uncertainty, which is not really the case. Due to the small sample size (three models) and inconsistencies in the model and methods used for downscaling in the older HADGEM run versus the IPSL and GFDL simulations, neither internal variability nor intermodel uncertainty is well sampled nor well isolated. Perhaps the term ‘multi-model comparison’ used in the abstract is more appropriate here. This is already addressed somewhat in the introduction (125-135), but should be clarified and given more thought. Claims like “we added an intra-scenario variability dimension (375)” are unclear and misleading, and should be changed.
3. Throughout the study, the authors claim that oxygen changes in the study region across the three simulations scale with the climate sensitivity of the parent ESMs. If quantifications of these sensitivities are available, they should be presented here. Additionally, an issue with this claim is that the differences in downscaling model/methods for HADGEM vs the IPSL and GFDL simulations provide uncontrolled degrees of freedom. The authors should provide an argument whether the differences in downscaling techniques should significantly impact the magnitude of oxygen changes. If possible, the authors could run some short sensitivity experiments using the new (used for IPSL, GFDL) setup to test sensitivity to e.g. vertical resolution.
4. In the model used by Wakelin et al (2020), oxygen is not included in open boundary conditions of the regional model so that changes in open ocean oxygen is not included. Is this the case here? This is very important for how the results may be interpreted and should be documented carefully. 
5. The authors need to be careful when interpreting correlation as causation. Correlations are only meaningful when there is a process that can explain the relationship. Please be thorough about when a physical/biogeochemical mechanism can explain a correlation and when a correlation cannot be explained. For example, why would you have a positive correlation between SS and stratification in some regions (Fig. 7)? If strong but erroneous correlations are prevalent, why can we still trust the results? The authors should also provide a discussion of any covariances that may influence the results (e.g. between temperature, stratification, respiration, NPP)
6. In calculating correlations, the long-term trend is removed. I see how this avoids false positives, but how can you assess the drivers of forced changes after removing the long term trends? In this case, it seems that correlations just classify the drivers of short-term variability, which is not what you purport to be investigating. Please explain/clarify.
7. What is gained by decomposition in section 2.3 as opposed to a traditional O2sat, AOU decomposition? Why is O2phys-ch (O2sat scaled by the initial saturation state) a more meaningful metric than O2sat? The authors end up using O2sat and SS (a.k.a 1-AOU) anyway, so this section can be removed entirely.
8. How are there negative values in the root-mean-square distance calculation (Fig. 2)? Need to provide formulae here for nbias and nurmsd.
9. Bias-correction for hypoxia measurements should be included in methods

Technical Corrections:

1. Grammatical errors and inconsistent capitalization throughout. Please proofread carefully.
2. In all figures, panel labels need to be included
3. Use consistent terminology for region names. Is the Danish strait the same as Skagerrak? Eastern North Sea is referenced throughout but not delineated on the map in Fig 1.
4. Nearly all instances of ‘in fact’ can be removed

Find attached my comments