Review of A regional hindcast model simulating ecosystem dynamics, inorganic carbon chemistry and ocean acidiﬁcation in the Gulf of Alaska

Abstract. The coastal ecosystem of the Gulf of Alaska (GOA) is especially vulnerable to the effects of ocean acidification and climate change. Detection of these long-term trends requires a good understanding of the system’s natural state. The GOA is a highly dynamic system that exhibits large inorganic carbon variability on subseasonal to interannual timescales. This variability is poorly understood due to the lack of observations in this expansive and remote region. We developed a new model setup for the GOA that couples the three-dimensional Regional Oceanic Model System (ROMS) and the Carbon, Ocean Biogeochemistry and Lower Trophic (COBALT) ecosystem model. To improve our conceptual understanding of the system, we conducted a hindcast simulation from 1980 to 2013. The model was explicitly forced with temporally and spatially varying coastal freshwater discharges from a high-resolution terrestrial hydrological model, thereby affecting salinity, alkalinity, dissolved inorganic carbon, and nutrient concentrations. This represents a substantial improvement over previous GOA modeling attempts. Here, we evaluate the model on seasonal to interannual timescales using the best available inorganic carbon observations. The model was particularly successful in reproducing observed aragonite oversaturation and undersaturation of near-bottom water in May and September, respectively. The largest deficiency in the model is its inability to adequately simulate springtime surface inorganic carbon chemistry, as it overestimates surface dissolved inorganic carbon, which translates into an underestimation of the surface aragonite saturation state at this time. We also use the model to describe the seasonal cycle and drivers of inorganic carbon parameters along the Seward Line transect in under-sampled months. Model output suggests that the majority of the near-bottom water along the Seward Line is seasonally undersaturated with respect to aragonite between June and January, as a result of upwelling and remineralization. Such an extensive period of reoccurring aragonite undersaturation may be harmful to ocean acidification-sensitive organisms. Furthermore, the influence of freshwater not only decreases the aragonite saturation state in coastal surface waters in summer and fall, but it simultaneously decreases the surface partial pressure of carbon dioxide (pCO2), thereby decoupling the aragonite saturation state from pCO2. The full seasonal cycle and geographic extent of the GOA region is under-sampled, and our model results give new and important insights for months of the year and areas that lack in situ inorganic carbon observations.


1. I would recommend adding an additional sentence to the abstract to highlight the addition of variable freshwater forcing to this model. I know it's mentioned in L7-8, but it seems to be a substantial addition to regional modeling of the GOA and should be mentioned as such.
2. It would be nice to include the model mesh in Fig. 1 to make clear the horizontal resolution of GOA-COBALT relative to the features of the GOA, especially given this paper is associated with the public release of the model. Also to make clear it's not e.g., telescopic. If 4.5 km is too fine to see in Fig. 1, you could have an inset where you zoom in on a sub-region and show the mesh.
3. L19-23, p4: Can you give some approximation of what this vertical resolution is? E.g., N meters per grid cell in the nearshore and offshore. [2016] provides a good overview of colormap selection in "How to select an honest, effective, colormap." Every colormap used here is a red-to-blue diverging map, whereas most of this data is sequential and would be much more honestly portrayed on a sequential colormap. Some great colormaps available in python, Matlab, etc. are cmocean and Fabio Crameri's color maps. The main issue is that the red-to-blue colormaps diverge at an arbitrary value in this manuscript, causing a large visual distinction between red and blue regions that is not meaningful physically. The red-to-blue can be used for Ω and in the case of anomalies, but should be centered around 1 for Ω, since that is a critical threshold for that variable, and around zero for the anomalies. I imagine this sounds tedious, but I think it will drastically improve the visual presentation and interpretability of the output. More meaningful features will be apparent in the cross sections, e.g. in Figure 4, which will help the reader compare the model to observations. Below I compared a CESM hindcast run to ERSST observations as a demonstration. In the first example I use a red-to-blue diverging colormap. In the second, I use a perceptually uniform sequential colormap. I think the advantages will be more clear in the cross-section maps, but this still shows the differences and draws the eye away from the arbitrary divergent point at 15C.
9. On a similar note, I am surprised that there is no third column in Figs. 2, 3, 4, 5, etc. showing the difference between the model and observations. Effort was made to interpolate the model and observations to the same grid, so it should be relatively straightforward to display the bias in the model by subtracting the two. I think showing this is crucial for the reader to see the regional expression and the magnitude of the bias. Currently, the reader relies on the author's highlighting of certain subregions of these biases in the text. For Figure 2 in particular, it's very hard to compare these by eye. On another note, in Figure 4, it looks like nearshore surface pH could be 0.2 units too acidic in the model, which would represent a 60% bias in the hydrogen ion concentration. Many of the quantitative arguments in the text about "overestimation" and "underestimation" will be made significantly more clear with the addition of a difference column (either raw or in percent bias). 13. L13-15, p15: Perhaps I am misreading this, but both photosynthesis and freshwater dilution should reduce DIC, and thus raise pH to more basic levels. So why is it surprising that pH is high, "despite the freshwater influence and its diluting character"? Or did you mean acidic by "high" here? Please clarify.
14. I found Section 5 very hard to interpret, and suggest that it is re-written and the methodologies here made more clear. Firstly, the end member analysis methodology should be made more clear. Admittedly I do not have a background in end member analysis, so perhaps it is clear to the informed reader what is happening here. As someone with a modeling background, I first thought "end member" implied that multiple simulations were run and one at the edge of the distribution of riverine boundary conditions was selected for analysis. It's unclear what "non-zero" DIC means when all of the DIC range in Table 2 is non-zero. In general, it needs to be spelled out that this is an end member mixing analysis (I assume), and more care should be taken explaining the methodology here. Secondly, the linear Taylor decomposition should also be spelled out. I don't think every step of Rheuban et al. (2019) needs to be replicated here, but it would be helpful to the reader to have some of the key equations and assumptions. Particularly that the sensitivity terms aren't explicitly calculated, how anomalies are generated, etc. I would suggest an additional section to the methods summarizing the end member analysis and linear Taylor decomposition.
3 Technical Comments 1. I find the first sentence of the abstract awkward: "The coastal ecosystem of the Gulf of Alaska (GOA) is especially vulnerable to the effects of ocean acidification and climate change that can only be understood within the context of the natural variability of physical and chemical conditions." Is it the coastal ecosystem or the effects of OA/climate change that can only be understood within the context of natural variability? I wouldn't say that natural variability is a key topic addressed in this paper either. I would suggest revising this sentence to change its content or to make it more clear.
6. Table 1: Is "alpha" supposed to be α? The formatting of this table is a bit difficult to interpret. E.g. the italicized sub-header, and it's not clear immediately that the a and b below explain the table values. Maybe this will be fixed on typesetting.
7. L7-8, pg2: I think this would be cleaner with something like "... high physical, biological, and chemical spatiotemporal variability across the GOA continental shelf." 8. L11, pg2: What is "its" referring to here? Grammatically it could relate to "this region" or "climate change and ocean acidification," among other interpretations. I would break L9-12 up into 2-3 sentences for clarity.
9. L14, pg2: It might be helpful to spell out why a seasonal increase in vertical mixing would lead to reduced carbonate concentrations here, since this is early in the introduction. Also what does seasonal "increase" refer to? Does it increase from winter to summer, summer to winter? Is the seasonality of mixing increasing with time?
10. L18, pg2: I find the use of "endowed" awkward here and elsewhere. I think it would be more simple to just say "contains low TA" or "is characterized by low TA" for example.