The work to produce these projections represents a mammoth effort with a large interdisciplinary team. This project/data set will have a lasting legacy with a large number of potential applications, and I hope the potential benefits of using this data set are realised by the community so as to reward the authors on their hard work. The projections have been well described and evaluated extensively. The choice of GCMs and downscaling methods has been done with care and are appropriate to the outcomes of the project. Finally, the projections themselves are of great interest to the hydrologic (among others) community. Please see my comments below which at first may seem extensive but are relatively minor and can often be treated as suggestions rather than being prescriptive.

General comments:

# Some of the figure legends/axes were a bit small and hard to read.

# The abstract describes succinctly the product development and how it was performed. I wonder if a sentence at the start of the abstract on the need for the product might help with context.

# I know it was mentioned, but it just wasn’t quite clear to me how the 9am-9am data from Australia was matched to the GCM data (which I assume is 12am-12am).

# Page 5, Line 15: Clarification for me please - Is it usual to only use SSTs as the forcing from the GCM in CCAM? I understand CCAM doesn’t have lateral boundary conditions making it quite unique – is my understanding correct?

# Section 2.2: I don’t think the authors should change their text, but as a comment, it felt the GCM selection was given 1-2 lines of attention on Page 4 and then two pages of attention was given to how the GCM projections fit within the ensemble of GCMS. This felt a little unbalanced to me. I understand it is important to show the spread of possible futures and how this ensemble covers it, but some text sounds like the authors justifying that ‘only’ four GCMs are sufficient. In particular on Page 7, Lines 9-15 almost seem to defensive to me, and I don’t see a reason why the authors need to defend ‘only’ four GCMs when they do in fact represent a good range plausible future. Moreover, I think the authors analysis is superior for the fact that they considered the best GCMs for Australia (rather than using all GCMs blindly).

# Page 5, Line 21: It feels odd to state the method not used was spectral nudging when the method that was used wasn’t stated?

Page 7, Line 9: “uncertainties are underestimated” – which uncertainties? Should this be “uncertainties in the GCM choice are underestimated”?

# Page 7, Line 25: “calibrate the GCM output” I would prefer the word calibrate to not be used, also calibrating data doesn’t quite make sense. Can this be reworded please?

# Page 8, Line 20. Up to the authors if they want to keep this sentence, but AWAP is gridded and will by definition underestimate point data. It is true also if one looks at catchment averages for extremes AWAP is slightly biased down but (to me anyway) the differences aren’t great. See Figure 5 in Nathan, R., Jordan, P., Scorah, M., Lang, S., Kuczera, G., Schaefer, M., Weinmann, E., 2016. Estimating the exceedance probability of extreme rainfalls up to the probable maximum precipitation. J. Hydrol. 543, 706–720. https://doi.org/10.1016/j.jhydrol.2016.10.044

# Page 9, Line 25: The description sounds more like downscaling “modify coast-scale GCM projections at a finer scale” rather than bias correction. Maybe some rewording in this paragraph would be appropriate?

# Page 12, Line 24: When you say decreases the warming signal it sounds like it has decreased the trend, but to me to the trend before and after bias correction of the CCAM data (brown and blue lines) is identical? So maybe some rewording here is necessary. See the comment below.

# Page 12, Line 27: Maybe I am taking exception with the word “signal”. That implies to me some sort of temporal trend, but here you just talking about the GCM being wetter, which isn’t a signal, it’s just bias. Apologies about the long comment – maybe just changing the word from signal to bias would be beneficial?

# Figure 8: Because you summarise seasonal results in Figure 9 and Figure 10, Figure 8 could just have the annual results only to make the figure more manageable? I know Vogel et al., 2022 has an extensive evaluation of the bias correction, but I think one figure just for one variable (say runoff) with all the bias correction methods would really be beneficial (can just be for one GCM) – given the amount of time spent outlining the bias correction methods (and their potential impact on the results).

# Page 14, Line 32 (and elsewhere): Sometimes precipitation is pr, and sometimes it is Pr (in italics).

# Figure 11: Bottom panel missing units on the y-axis?

# Page 16, Line 19: Did you use the wet and dry season? I think Figures 13 and 14 just use the regular DJF etc seasons?

# Page 16, Line 9: Not sure, but I know of work that found that the bias correction method was the greater contributor to the ensemble spread. Not sure if the authors have comments on why the different results? See Wasko, C., Guo, D., Ho, M., Nathan, R., Vogel, E., 2023. Diverging projections for flood and rainfall frequency curves. J. Hydrol. 620, 129403. https://doi.org/10.1016/j.jhydrol.2023.129403

# Figure 15: Given that AWRA-L is a water balance model, has it been evaluated for extremes and if not can a comment be made on its applicability for this purpose. The above manuscript and the following found a possible underestimation of extremes or the change signal in changes for extreme events. Ho, M., Nathan, R., Wasko, C., Vogel, E., Sharma, A., 2022. Projecting changes in flood event runoff coefficients under climate change. J. Hydrol. 615, 128689. https://doi.org/10.1016/j.jhydrol.2022.128689

# Section 7.3: I wonder if the “maps” came first (Section 7.3 was Section 7.1), then it would make an easier transition to Section 7.1 and Section 7.2. Looking at the maps, you see the strongest change in SSWF and then you can drill down on the results for that region. My other concern with just focussing on the JJA season. Most rainfall occurs in the summer in the tropics so the results presented here aren’t as meaningful as they could be – I guess I would prefer these maps to be annual – and to be the first item displayed in Section 7. This would also follow better as again Section 7.4 focuses on SSWF.

# Section 8: Am I right in saying that temperature projections are not available as part of the Australian Water Outlook Service but are available on NCI? I feel temperature is an important variable for example when calculating fire risk, and one that many other users would be interested in.

# Section 8.3: Line 27 confused me a bit – would it be better to have a link to the reports here (instead of the end of Section 8.3)?

# Section 9: The first paragraph could almost be removed, and the section relabelled “Limitations”.

Page 22, Line 7: I wonder if “due to time and personnel constraints” could be rephrased with “due to the large spatial domain…” it is clear (to me anyway) that you couldn’t be expected to use more GCMs than you already have due to the large domain and sheer scale of the project.

Editorial:

Page 2, Line 11: “…south-east with changes in streamflow typically…” might read better.

Page 3, Line 5: missing a space after the reference.

Page 5, Line 24: Doesn’t have to be bold and can be part of the paragraph.

Page 7, Line 9: extra new line.

Page 19, Line 22: Change from 3^rd person to 1^st person with “we”. Could revert to be consistent with the rest of manuscript.

Page 20, Line 23: I think there is a track changes mark under the apostrophe in “model’s”.

Page 21, Line 12: CO2 (subscript the 2)

The authors document the development of a national set of hydrological projections for Australia. They provide good motivation for the development of the product/service and provide extensive description and evaluation of the output. My main comment is around the use of the 3 bias correction methods to expand the ensemble. It’s not clear to me why they looked at 3 methods and what it adds. E.g. at one point (for temperature) the authors describe how “the three bias corrections methods are almost indistinguishable…” (pg 12, line 9, Figure 6). Also at pg 16, line 9 (re Figure 12), the authors comment: “The first observation implies that in general, the choice of GCM contributes more to the ensemble spread than the application of the BC techniques”. I think some more discussion and justification for the inclusion of all 3 bias correction methods is required.

Other points:

Page 5, line 1-2: I respect that pragmatic choices need to be made but I think the authors should make some comment around the validity of RCP8.5 for future risk assessment e.g. https://www.nature.com/articles/d41586-020-00177-3 

Page 5, line 24: Why is this sentence in bold?

Figure 2: Perhaps it’s obvious, but I think somewhere in the caption the authors should write that the darker shaded bars indicate selected models

Page 6, line 1: The use of 1976-2005 historical period is a departure from CCiA. Can the authors comment why they chose that period?  

Page 7, line 10: Why not include the CCAM simulations to get a better picture of the spread relative to CCiA?

Page 7, line 21: GCM/RCM not just GCM

Page 7, line 23: This sentence doesn’t make sense. I think the “are” before precipitation should be replaced with a colon or dash.

Page 7, line 26: Does this need to be updated to AGCD?

Page 9, line 15: This sentence doesn’t make sense and needs to be rewritten.

Page 12, line 26: As I understand it, the purpose of an ensemble member is to add new information. How are the 3 bias correction methods adding new information? Can the authors comment on this? (see my main comment)

Figure 6, 7: Perhaps I missed it but why are CCAM-MRNBC and CCA-QME not included?

Fig 8: Although it’s mentioned in the text, I think it would be helpful to add a sentence in the caption about why there are data gaps in the maps.

Page 16, line 5: NRMs (or NRM regions) not NRMS

Figure 12: It seems pointless to label each of the 4 plots with “Southern and South Western Flatlands” – this could just be written in the caption. I think having the variable (e.g. precipitation, soil moisture etc.) clearly visible at the top of each plot would be helpful.

Figure 13: It’s very difficult to interpret these plots, can the resolution be sharpened?

Page 17, line 13/14: You’ve written “antecedent conditions soil moisture conditions”. I assume you mean “antecedent soil moisture conditions”.

Page 22, line 8: It seems strange that ‘personnel issues’ is listed here but not earlier in the manuscript.

This article introduces the development of a national hydrological projections (NHP) service for Australia, including the choice of GCMs and RCM, application and evaluation of three bias correction methods, and driving the Bureau's landscape water balance hydrological model (AWRA-L) to produce hydrological projections. This national hydrological service provides valuable information on the impact of climate change on hydrological cycles over Australia to end users. The overall structure of the manuscript is coherent while wordsmithing is necessary, especially in the first half of the article. Besides, I have a few comments and suggestions for authors to consider.

# Specific comments

Page 2 Line 22-26: It is mentioned here that Australian states may prefer to use their own downscaled projection products. Key issues are that data are too heterogenous for use across intersect jurisdictional boundaries, and clear instructions are not provided. These issues are addressed in the NHP service, but are users in these states now tend to use your products rather than use state operated ones? Could you give some insights into this point?

Page 3 Line 1: I could not fully understand what transient projection means. Please give a clearer definition and/or example. Also, repeated word ‘applied’ in footnote 1.

Page 3 Line 24: According to the context, you simply interpolate GCM and RCM to 5km spatial resolution before applying the bias corrections. I reckon using bias correction techniques for downscaling should be the key point here.

Another point I am interested in is whether you have tried only applying simple mean (additive or multiplicative) correction to the GCM outputs to drive hydrological model, and using sophisticated methods to correct hydrological outputs. What is your rationale of bias correcting climate outputs prior to driving the hydrological model? Even though the multivariate bias correction accounts for the inter-variable, temporal and spatial structure of the model outputs, the bias adjustment process may have changed temporal features of the model series.

Page 4 Line 8: This is the first time AWRA-L model is mentioned in the introduction. I think more descriptions of AWRA-L are needed in this section because the choice/development of hydrological model is definitely an important part of the NHP project.

Page 4 Line 21-26: In line 24, what does ‘variation between CMIP5 models’ mean here? The temporal variance and climatological mean? Regarding the GCM selection, I would like the authors to explain more about how you narrow down the selection from 8 to 4 GCMs. You mentioned that all required variable data are available among 47 CMIP5 models, and CCiA recommended 8 models. What are your criteria to choose these four CMIP5 models out of eight. In addition, I am curious why you include an RCM to increase the ensemble range, and four RCM simulations are only corrected using one bias correction technique. Why not simply include other four GCMs recommended by CCiA?

Figures 2 and 3 are too small. Please consider redo them into a 4 rows × 4 cols plot. Figure 8 is also too small. Please consider split it into two or more plots.

Page 6 Line 26: What is your rationale of calling these four GCMs a ‘reasonable’ subsample of the CCiA ensemble? Please specify.

Page 7 Line 25: The bias correction methods, ISIMIP2b and MRNBC, are trained over 1976-2005. Is it because the wind speed observations start from 1975? The QME method is trained over 1975-2017, which is 13 years longer. Please clarify and comment on to what extent the use of different training period would affect the bias corrected climate variables, and further the hydrological projections.

Page 8 Line 25: Before 1990, daily climatological averages (for each day of the year) are used for solar radiation. How did this affect the training of the bias correction models as the ‘true’ values are not recorded?

Page 12 Line 18-19: This statement could be moved to before Section 4.1, where the data required for the bias correction is introduced.

Page 13 Line 27: Do you train bias correction model using the period from 1976, and apply the trained model to correct climate model simulations from 1960 to 2099? Please clarify this in the text.

In Figure 8c, despite small absolute biases, the relative biases for root zone soil moisture over all four seasons are very large compared to other variables. Will the large relative biases of the soil moisture lead to inaccurate information for the community?

Page 14 Line 17: Figures 9 and 10 are plotted without any interpretations. Please give some comments/explanations on these results. From my understanding, part of Figure 9 shows area-averaged relative biases presented in Figure 8. However, the results of MRNBC-ACCESS1-0 in Figure 9c contradicts those in Figure 8c, where the averaged relative biases should be at least <-10%. Moreover, in these figures, these relative biases or bias values may not be representative because the negative and positive values may be cancelled out in the area averages. I suggest plotting averaged absolute biases without signs across Australia and NRM regions.

Page 14: It would be better to show the bar charts for NRM regions (similar to Figure 9 and 10), at least SSWF, in the manuscript or in the supplementary material.

Page 15 Line 23-24: Why not showing the ensemble statistics using the yearly averaged data instead of 30-year running mean? I suspect the range of 10th and 90th percentiles over time will not be too messy.

Page 16 Line 9-10: Do the results imply that perhaps only one best performing bias correction technique is needed for your application? Perhaps more GCMs and/or RCMs should be included to better gauge the uncertainty of the future projections.

Figure 14: Blockings are apparently seen in the precipitation (Figure 14a) and root-zone soil moisture (Figure 14c) maps compared to runoff results (Figure 14b). Is it because low-res GCMs are spatially interpolated into fine resolutions followed by statistical bias corrections? Why runoff results are smoother? Why not much spatial variability is seen in the PET plot (Figure 14d)? Furthermore, in previous spatial maps, the data-sparse regions are masked out. I suggest doing the same masking for this figure.

Page 17 Line 26: You may want to show the same analysis for the historical period to be confident about the performance of GCM-driven hydrological projections in simulating the extreme events.

# Editorial

Page 1 Line 23: It is hard to understand this sentence without reading the main content. I suggest replacing ‘one to output from a regional climate model forced by…’ with ‘one regional climate model (RCM) that is forced by …’

Careful proofreading is required throughout the article. Typical issues are:

• Missing commas: For example, in Page 3 Line 11-13, there should be a comma after ‘To address these deficits in hydrological projections’.
• Sentences too long to read: For example, in Page 6 Line 1-4, it is better to break this sentence into two or more before the words ‘hence’ and ‘nevertheless’.
• Duplicate descriptions: For example, the sentence in Page 10 Line 19-21 is a duplicate description of Page 10 Line 25. The first sentence in Section 6 is also mentioned before.